User manual | ADSP-21160 SHARC DSP Hardware Reference (Rev. 4.1) PDF

ADSP-21160 SHARC® DSP
Hardware Reference
Revision 4.1, April 2013
Part Number
82-001966-01
Analog Devices, Inc.
One Technology Way
Norwood, Mass. 02062-9106
a
Copyright Information
© 2013 Analog Devices, Inc., ALL RIGHTS RESERVED. This document may not be reproduced in any form without prior, express written
consent from Analog Devices, Inc.
Printed in the USA.
Disclaimer
Analog Devices, Inc. reserves the right to change this product without
prior notice. Information furnished by Analog Devices is believed to be
accurate and reliable. However, no responsibility is assumed by Analog
Devices for its use; nor for any infringement of patents or other rights of
third parties which may result from its use. No license is granted by implication or otherwise under the patent rights of Analog Devices, Inc.
Trademark and Service Mark Notice
The Analog Devices logo, Blackfin, SHARC, TigerSHARC, CrossCore,
VisualDSP++, and EZ-KIT Lite are registered trademarks of Analog
Devices, Inc.
All other brand and product names are trademarks or service marks of
their respective owners.
CONTENTS
PREFACE
Purpose of This Manual .............................................................. xxvii
Intended Audience ...................................................................... xxvii
Manual Contents ....................................................................... xxviii
What’s New in This Manual .......................................................... xxx
Technical Support ......................................................................... xxx
Supported Processors .................................................................... xxxi
Product Information .................................................................... xxxi
Analog Devices Web Site ....................................................... xxxii
EngineerZone ........................................................................ xxxii
Notation Conventions ................................................................ xxxiii
Register Diagram Conventions ................................................... xxxiv
INTRODUCTION
Overview – Why Floating-Point DSP? ........................................... 1-1
ADSP-21160 DSP Design Advantages ........................................... 1-2
ADSP-21160 DSP Architecture Overview ..................................... 1-5
ADSP-21160 SHARC DSP Hardware Reference
iii
Processor Core ........................................................................ 1-6
Processing Elements ............................................................ 1-6
Program Sequence Control ................................................. 1-8
Processor Internal Buses .................................................... 1-10
Processor Peripherals ............................................................. 1-11
Dual-Ported Internal Memory (SRAM) ............................. 1-11
External Port .................................................................... 1-12
I/O Processor ................................................................... 1-13
JTAG Port ............................................................................ 1-15
Development Tools ..................................................................... 1-15
Differences From Previous SHARC DSPs .................................... 1-16
Processor Core Enhancements ............................................... 1-16
Processor Internal Bus Enhancements .................................... 1-17
Memory Organization Enhancements .................................... 1-18
External Port Enhancements .................................................. 1-18
Host Interface Enhancements ........................................... 1-18
Multiprocessor Interface Enhancements ............................ 1-19
IO Architecture Enhancements .............................................. 1-19
DMA Controller Enhancements ........................................ 1-19
Link Port Enhancements ................................................... 1-19
Instruction Set Enhancements ............................................... 1-20
iv
ADSP-21160 SHARC DSP Hardware Reference
PROCESSING ELEMENTS
Overview ...................................................................................... 2-1
Setting Computational Modes ....................................................... 2-2
32-bit (Normal Word) Floating-Point Format .......................... 2-3
40-bit Floating-Point Format ................................................... 2-4
16-bit (Short Word) Floating-Point Format .............................. 2-5
32-Bit Fixed-Point Format ....................................................... 2-5
Rounding Mode ...................................................................... 2-6
Using Computational Status .......................................................... 2-7
Arithmetic Logic Unit (ALU) ........................................................ 2-7
ALU Operation ....................................................................... 2-8
ALU Saturation ....................................................................... 2-9
ALU Status Flags ..................................................................... 2-9
ALU Instruction Summary .................................................... 2-10
Multiply—Accumulator (Multiplier) ........................................... 2-13
Multiplier Operation ............................................................. 2-14
Multiplier (Fixed-Point) Result Register ................................. 2-15
Multiplier Status Flags ........................................................... 2-17
Multiplier Instruction Summary ............................................ 2-18
Barrel-Shifter (Shifter) ................................................................. 2-21
Shifter Operation .................................................................. 2-21
Shifter Status Flags ................................................................ 2-25
Shifter Instruction Summary .................................................. 2-26
Data Register File ........................................................................ 2-28
ADSP-21160 SHARC DSP Hardware Reference
v
Alternate (Secondary) Data Registers ........................................... 2-30
Multifunction Computations ...................................................... 2-31
Secondary Processing Element (PEy) ........................................... 2-35
Dual Compute Units Sets ...................................................... 2-37
Dual Register Files ................................................................ 2-38
Dual Alternate Registers ....................................................... 2-39
SIMD (Computational) Operations ....................................... 2-39
SIMD and Status Flags .......................................................... 2-42
PROGRAM SEQUENCER
Overview ...................................................................................... 3-1
Instruction Pipeline ...................................................................... 3-8
Instruction Cache ......................................................................... 3-9
Using the Cache .................................................................... 3-12
Optimizing Cache Usage ....................................................... 3-12
Branches and Sequencing ............................................................ 3-14
Conditional Branches ............................................................ 3-16
Delayed Branches .................................................................. 3-16
Loops and Sequencing ................................................................ 3-20
Restrictions On Ending Loops ............................................... 3-22
Restrictions On Short Loops ................................................. 3-23
Loop Address Stack ............................................................... 3-27
Loop Counter Stack .............................................................. 3-28
vi
ADSP-21160 SHARC DSP Hardware Reference
Interrupts and Sequencing ........................................................... 3-32
Sensing Interrupts ................................................................. 3-38
Masking Interrupts ................................................................ 3-39
Latching Interrupts ................................................................ 3-40
Stacking Status During Interrupts .......................................... 3-42
Nesting Interrupts ................................................................. 3-43
Reusing Interrupts ................................................................. 3-45
Interrupting IDLE ................................................................. 3-46
Multiprocessing Interrupts ..................................................... 3-47
Timer and Sequencing ................................................................ 3-48
Stacks and Sequencing ................................................................ 3-50
Conditional Sequencing .............................................................. 3-52
SIMD Mode and Sequencing ...................................................... 3-55
Conditional Compute Operations .......................................... 3-56
Conditional Branches and Loops ........................................... 3-57
Conditional Data Moves ........................................................ 3-57
Case 1:
Complementary Register Pair Data Move ....................... 3-57
Case 2:
Uncomplemented to Complementary Register Move ....... 3-61
Case 3:
Complementary Register => Uncomplimentary Register . 3-62
Case 4:
Data Move Involves External Memory or IOP Memory Space
3-63
Conditional DAG Operations ................................................ 3-64
ADSP-21160 SHARC DSP Hardware Reference
vii
DATA ADDRESS GENERATORS
Overview ...................................................................................... 4-1
Setting DAG Modes ..................................................................... 4-3
Circular Buffering Mode ......................................................... 4-4
Broadcast Loading Mode ......................................................... 4-5
Alternate (Secondary) DAG Registers ...................................... 4-6
Bit-Reverse Addressing Mode .................................................. 4-8
Using DAG Status ........................................................................ 4-9
DAG Operations .......................................................................... 4-9
Addressing With DAGs ......................................................... 4-10
Addressing Circular Buffers ................................................... 4-12
Modifying DAG Registers ..................................................... 4-17
Addressing in SISD and SIMD Modes ................................... 4-18
DAGs, Registers, and Memory .................................................... 4-18
DAG Register-to-Bus Alignment ........................................... 4-19
DAG Register Transfer Restrictions ....................................... 4-21
DAG Instruction Summary ......................................................... 4-22
MEMORY
Overview ...................................................................................... 5-1
Internal Address and Data Buses .............................................. 5-6
Internal Data Bus Exchange .................................................... 5-7
viii
ADSP-21160 SHARC DSP Hardware Reference
ADSP-21160 DSP Memory Map ................................................ 5-12
Internal Memory ................................................................... 5-14
Multiprocessor Memory ......................................................... 5-17
External Memory .................................................................. 5-20
Shadow Write FIFO .............................................................. 5-21
Memory Organization and Word Size .................................... 5-22
Placing 32-Bit Words and 48-Bit Words ............................ 5-23
Mixing 32-Bit and 48-Bit Words ....................................... 5-23
Restrictions on Mixing 32-Bit and 48-Bit Words ............... 5-26
Setting Data Access Modes .......................................................... 5-28
Using Boot Memory .............................................................. 5-29
Reading from Boot Memory .............................................. 5-30
Writing to Boot Memory ................................................... 5-31
Internal Interrupt Vector Table .............................................. 5-31
Internal Memory Block Data Width ...................................... 5-32
Memory Bank Size ................................................................ 5-33
External Bus Priority ............................................................. 5-33
Secondary Processor Element (PEy) ........................................ 5-34
Broadcast Register Loads ....................................................... 5-34
Illegal I/O Processor Register Access ....................................... 5-35
Unaligned 64-bit Memory Access ........................................... 5-36
External Bank X Access Mode ................................................ 5-36
External Bank X Waitstates .................................................... 5-37
External (Bank 0) DRAM Page Size ....................................... 5-38
ADSP-21160 SHARC DSP Hardware Reference
ix
Using Memory Access Status ....................................................... 5-38
Accessing Memory ...................................................................... 5-39
Access Word Size ................................................................... 5-40
Long Word (64-Bit) Accesses ............................................ 5-41
Instruction Word (48-Bit) and Extended Precision Normal Word
(40-Bit) Accesses ............................................................ 5-43
Normal Word (32-Bit) Accesses ........................................ 5-43
Short Word (16-Bit) Accesses ............................................ 5-44
SISD, SIMD, and Broadcast Load Modes .............................. 5-44
Single-and Dual-Data Accesses .............................................. 5-45
Data Access Options ............................................................. 5-45
Short Word Addressing of Single Data in SISD Mode ........ 5-47
Short Word Addressing of Single Data in SIMD Mode ...... 5-48
Short Word Addressing of Dual-Data in SISD Mode ......... 5-51
Short Word Addressing of Dual-Data in SIMD Mode ....... 5-53
32-Bit Normal Word Addressing of Single Data in SISD Mode 5-55
32-Bit Normal Word Addressing of Single Data in SIMD Mode
5-57
32-Bit Normal Word Addressing of Dual Data in SISD Mode 5-59
32-Bit Normal Word Addressing of Dual Data in SIMD Mode 5-59
Extended Precision Normal Word Addressing of Single Data 5-62
Extended Precision Normal Word Addressing of Dual Data in SISD
Mode ............................................................................ 5-64
Extended Precision Normal Word Addressing of Dual Data in SIMD
Mode ............................................................................ 5-64
x
ADSP-21160 SHARC DSP Hardware Reference
Long Word Addressing of Single Data ............................... 5-67
Long Word Addressing of Dual Data in SISD Mode .......... 5-69
Long Word Addressing of Dual Data in SIMD Mode ......... 5-69
Mixed Word Width Addressing of Dual Data in SISD Mode 5-72
Mixed Word Width Addressing of Dual Data in SIMD Mode 5-74
Broadcast Load Access ....................................................... 5-74
Arranging Data in Memory ......................................................... 5-84
I/O PROCESSOR
Overview ...................................................................................... 6-1
Setting I/O Processor—EPort Modes ........................................... 6-14
Boot Memory DMA Mode .................................................... 6-17
External Port Buffer Modes .................................................... 6-17
External Port Channel Priority Modes .................................... 6-19
External Port Channel Transfer Modes ................................... 6-21
External Port Channel Handshake Modes .............................. 6-22
Master Mode .................................................................... 6-25
Paced Master Mode ........................................................... 6-30
Slave Mode ....................................................................... 6-31
Handshake Mode .............................................................. 6-34
External-Handshake Mode ................................................ 6-40
Setting I/O Processor—LPort Modes ........................................... 6-43
Link Port Buffer Modes ......................................................... 6-45
Link Port Channel Priority Modes ......................................... 6-45
Link Port Channel Transfer Modes ......................................... 6-48
ADSP-21160 SHARC DSP Hardware Reference
xi
Setting I/O Processor—SPort Modes ........................................... 6-49
Serial Port Buffer Modes ....................................................... 6-51
Serial Port Channel Priority Modes ........................................ 6-52
Serial Port Channel Transfer Modes ....................................... 6-52
Using I/O Processor Status .......................................................... 6-53
External Port Status ............................................................... 6-57
Link Port Status .................................................................... 6-60
Serial Port Status ................................................................... 6-63
DMA Controller Operation ........................................................ 6-65
Managing DMA Channel Priority ......................................... 6-67
Chaining DMA Processes ...................................................... 6-69
Transfer Control Block (TCB) Chain Loading ................... 6-71
Setting Up and Starting The Chain ................................... 6-72
Inserting a TCB in an Active Chain .................................. 6-73
External Port DMA .................................................................... 6-74
Setting up External Port DMA .............................................. 6-74
Bootloading Through The External Port ................................ 6-76
Link Port DMA .......................................................................... 6-80
Setting up Link Port DMA .................................................... 6-81
Using Two-Dimensional Link Port DMA ............................... 6-83
Bootloading Through The Link Port ..................................... 6-87
Serial Port DMA ......................................................................... 6-89
Setting up Serial Port DMA ................................................... 6-90
Using Two-Dimensional Serial Port DMA ............................. 6-91
xii
ADSP-21160 SHARC DSP Hardware Reference
Optimizing DMA Throughput .................................................... 6-92
Internal Memory DMA ......................................................... 6-92
External Memory DMA ......................................................... 6-93
System-Level Considerations .................................................. 6-96
EXTERNAL PORT
Overview ...................................................................................... 7-1
Setting External Port Modes .......................................................... 7-2
External Memory Interface ............................................................ 7-3
Banked External Memory ...................................................... 7-10
Unbanked External Memory .................................................. 7-10
Boot Memory ........................................................................ 7-11
Idle Cycle ......................................................................... 7-11
Data Hold Cycle ............................................................... 7-13
Multiprocessor Memory Space Waitstates and Acknowledge 7-14
DRAM Page Boundary Detection .......................................... 7-15
Timing External Memory Accesses ......................................... 7-18
Asynchronous Mode Interface Timing ............................... 7-19
Asynchronous Mode Read – Bus Master ........................ 7-21
Asynchronous Mode Write – Bus Master ....................... 7-22
Synchronous Mode Interface Timing ................................. 7-23
Synchronous Mode Read – Bus Master .......................... 7-25
Synchronous Write, Zero-Waitstate Mode ...................... 7-27
Synchronous Write, One Waitstate Mode ....................... 7-31
ADSP-21160 SHARC DSP Hardware Reference
xiii
Synchronous Burst Mode Interface Timing ....................... 7-33
Burst Length Determination ......................................... 7-34
Burst Stall Criteria ........................................................ 7-35
Synchronous Burst Reads .............................................. 7-37
Synchronous Burst Writes ............................................. 7-39
Using External SBSRAM ....................................................... 7-43
Executing Instructions From External Memory ...................... 7-48
Host Processor Interface ............................................................. 7-49
Acquiring the Bus ................................................................. 7-53
Asynchronous Transfers ......................................................... 7-57
Asynchronous Transfer Timing .............................................. 7-59
Synchronous Transfers ........................................................... 7-62
Synchronous Broadcast Writes .......................................... 7-63
Synchronous Burst Read Transfers ..................................... 7-65
Slave Direct Reads and Writes ............................................... 7-66
IOP Shadow Registers ....................................................... 7-66
Instruction Transfers ......................................................... 7-67
Host Direct Writes and Reads ................................................ 7-68
Direct Writes .................................................................... 7-68
Direct Write Latency ........................................................ 7-69
Direct Reads ..................................................................... 7-70
Broadcast Writes ................................................................... 7-71
Shadow Write FIFO .............................................................. 7-71
Data Transfers Through the EPBx Buffers .............................. 7-71
xiv
ADSP-21160 SHARC DSP Hardware Reference
DMA Transfers ...................................................................... 7-72
Host Data Packing ................................................................. 7-73
32-bit Data Packing .......................................................... 7-73
48-Bit Instruction Packing ................................................ 7-76
Host Interface Status ............................................................. 7-77
Interprocessor Messages and Vector Interrupts ........................ 7-77
Message Passing (MSGRx) ................................................ 7-78
Host Vector Interrupts (VIRPT) ........................................ 7-79
System Bus Interfacing .......................................................... 7-80
Access to the DSP Bus—Slave DSP ................................... 7-80
Access to the System Bus—Master DSP ............................. 7-82
Processor Core Access To System Bus ................................. 7-84
Deadlock Resolution ......................................................... 7-84
DSP DMA Access To System Bus ...................................... 7-88
Multiprocessing with Local Memory .................................. 7-89
DSP To Microprocessor Interface ...................................... 7-91
Multiprocessor (DSPs) Interface .................................................. 7-91
Multiprocessing System Architectures .................................... 7-94
Data Flow Multiprocessing ................................................ 7-94
Cluster Multiprocessing .................................................... 7-95
Multiprocessor Bus Arbitration .............................................. 7-98
Bus Arbitration Protocol ................................................. 7-100
Bus Arbitration Priority (RPBA) ...................................... 7-104
ADSP-21160 SHARC DSP Hardware Reference
xv
Mastership Timeout Bus ................................................. 7-106
Priority Access ................................................................ 7-107
Bus Synchronization After Reset .......................................... 7-110
Booting Another DSP ......................................................... 7-113
Multiprocessor Direct Writes and Reads .............................. 7-113
IOP Shadow Registers ..................................................... 7-114
Instruction Transfers ....................................................... 7-114
Direct Writes .................................................................. 7-114
Direct Write Latency ...................................................... 7-115
Direct Reads ................................................................... 7-115
Broadcast Writes ................................................................. 7-115
Shadow Write FIFO ............................................................ 7-117
Data Transfers Through the EPBx Buffers ............................ 7-118
Bus Lock and Semaphores ................................................... 7-118
Interprocessor Messages and Vector Interrupts ..................... 7-120
Message Passing (MSGRx) .............................................. 7-121
Vector Interrupts (VIRPT) .............................................. 7-121
Multiprocessor Interface Status ....................................... 7-122
LINK PORTS
Overview ...................................................................................... 8-1
Link Port To Link Buffer Assignment ...................................... 8-3
Link Port DMA Channels ....................................................... 8-5
Link Port Booting ................................................................... 8-5
xvi
ADSP-21160 SHARC DSP Hardware Reference
Setting Link Port Modes ............................................................... 8-5
Link Data Path (and Compatibility) Modes .............................. 8-7
Using Link Port Handshake Signals ............................................... 8-7
Using Link Buffers ...................................................................... 8-10
Core Processor Access To Link Buffers ................................... 8-11
Host Processor Access To Link Buffers ................................... 8-11
Using Link Port DMA ................................................................ 8-12
Using Link Port Interrupts .......................................................... 8-12
Link Port Interrupts With DMA Enabled ............................... 8-13
Link Port Interrupts With DMA Disabled .............................. 8-13
Link Port Service Request Interrupts (LSRQ) ......................... 8-14
Detecting Errors On Link Transmissions ..................................... 8-16
Using Token Passing With Link Ports .......................................... 8-17
Designing Link Port Systems ....................................................... 8-21
Terminations For Link Transmission Lines ............................. 8-21
Peripheral I/O Using Link Ports ............................................. 8-22
Data Flow Multiprocessing With Link Ports ........................... 8-22
SERIAL PORTS
Overview ...................................................................................... 9-1
SPORT Interrupts ................................................................... 9-5
SPORT Reset .......................................................................... 9-6
ADSP-21160 SHARC DSP Hardware Reference
xvii
Setting Serial Port Modes .............................................................. 9-7
Transmit and Receive Control Registers
(STCTL, SRCTL) ................................................................ 9-9
Register Writes and Effect Latency ......................................... 9-12
Transmit and Receive Data Buffers (TX, RX) ......................... 9-12
Clock and Frame Sync Frequencies (TDIV, RDIV) ................ 9-14
Data Word Formats ............................................................... 9-17
Word Length .................................................................... 9-17
Endian Format ................................................................. 9-17
Data Packing and Unpacking ............................................ 9-18
Data Type ........................................................................ 9-19
Companding .................................................................... 9-20
Clock Signal Options ............................................................ 9-21
Frame Sync Options .............................................................. 9-22
Framed Versus Unframed .................................................. 9-22
Internal Versus External Frame Syncs ................................ 9-24
Active Low Versus Active High Frame Syncs ...................... 9-24
Sampling Edge For Data and Frame Syncs ......................... 9-25
Early Versus Late Frame Syncs .......................................... 9-25
Data-Independent Transmit Frame Sync ........................... 9-27
SPORT Loopback ................................................................. 9-27
Multichannel Operation ........................................................ 9-28
Frame Syncs in Multichannel Mode .................................. 9-30
Multichannel Control Bits in STCTL, SRCTL .................. 9-30
xviii
ADSP-21160 SHARC DSP Hardware Reference
Channel Selection Registers ............................................... 9-32
SPORT Receive Comparison Registers .............................. 9-33
Moving Data Between SPORTS and Memory .............................. 9-36
DMA Block Transfers ............................................................ 9-36
Single-Word Transfers ............................................................ 9-37
SPORT Pin/Line Terminations .................................................... 9-38
JTAG TEST EMULATION PORT
JTAG Test Access Port ................................................................. 10-3
Instruction Register ..................................................................... 10-4
EMUPMD Shift Register ....................................................... 10-5
EMUPX Shift Register ........................................................... 10-7
EMU64PX Shift Register ....................................................... 10-7
EMUPC Shift Register .......................................................... 10-8
EMUCTL Shift Register ........................................................ 10-8
EMUSTAT Shift Register .................................................... 10-12
BRKSTAT Shift Register .................................................... 10-12
MEMTST Shift Register ...................................................... 10-13
PSx, DMx, IOx, and EPx (Breakpoint) Registers .................. 10-14
EMUN Register .................................................................. 10-16
EMUCLK and EMUCLK2 Registers ................................... 10-17
EMUIDLE Instruction ........................................................ 10-17
In-Circuit Signal Analyzer (ICSA) Function ......................... 10-17
Boundary Register ..................................................................... 10-17
Device Identification Register .................................................... 10-55
ADSP-21160 SHARC DSP Hardware Reference
xix
Built-in Self-test Operation (BIST) ........................................... 10-55
Private Instructions ................................................................... 10-55
References ................................................................................ 10-55
SYSTEM DESIGN
DSP Pin Descriptions ................................................................. 11-2
Pin States At Reset .............................................................. 11-12
Clock Derivation ................................................................ 11-15
RESET and CLKIN ............................................................ 11-16
Input Synchronization Delay .......................................... 11-17
Interrupt and Timer Pins .................................................... 11-18
Flag Pins ............................................................................. 11-18
Flag Inputs ..................................................................... 11-19
Flag Outputs .................................................................. 11-19
JTAG Interface Pins ............................................................ 11-20
Dual-Voltage Power-up Sequencing ........................................... 11-21
Designing for JTAG Emulation ................................................. 11-24
Target Board Connector ...................................................... 11-25
Layout Requirements ................................................................ 11-30
Power Sequence for Emulation .................................................. 11-30
Additional JTAG Emulator References ...................................... 11-31
Pod Specifications ..................................................................... 11-31
DSP JTAG Pod Connector .................................................. 11-31
DSP 3.3V Pod Logic ........................................................... 11-32
DSP 2.5V Pod Logic ........................................................... 11-33
xx
ADSP-21160 SHARC DSP Hardware Reference
Conditioning Input Signals ....................................................... 11-34
Link Port Input Filter Circuits ............................................. 11-35
RESET Input Hysteresis ...................................................... 11-36
Designing For High Frequency Operation ................................. 11-36
Clock Specifications and Jitter ............................................. 11-37
Clock Distribution .............................................................. 11-38
Point-To-Point Connections ................................................ 11-40
Signal Integrity .................................................................... 11-41
Other Recommendations and Suggestions ............................ 11-44
Decoupling Capacitors and Ground Planes .......................... 11-45
Oscilloscope Probes ............................................................. 11-47
Recommended Reading ....................................................... 11-47
Booting Single and Multiple Processors ..................................... 11-48
Multiprocessor Host Booting ............................................... 11-48
Multiprocessor EPROM Booting ......................................... 11-49
All DSPs Boot in Turn from a Single EPROM ................. 11-49
One DSP is Booted, which then Boots the Others ........... 11-50
Multiprocessor Link Port Booting ........................................ 11-51
Multiprocessor Booting From External Memory ................... 11-52
REGISTERS
Control and Status System Registers ............................................. A-2
Mode Control 1 Register (MODE1) ....................................... A-3
Mode Mask Register (MMASK) .............................................. A-5
Mode Control 2 Register (MODE2) ....................................... A-6
ADSP-21160 SHARC DSP Hardware Reference
xxi
Arithmetic Status Registers (ASTATx and ASTATy) ................. A-8
Sticky Status Registers (STKYx and STKYy) .......................... A-12
User-Defined Status Registers (USTATx) ............................... A-15
Processing Element Registers ....................................................... A-15
Data File Data Registers (Rx, Fx, Sx) ..................................... A-16
Multiplier Results Registers (MRxF, MRxB) ........................... A-16
Program Memory Bus Exchange Register (PX) ....................... A-17
Program Sequencer Registers ....................................................... A-17
Interrupt Latch Register (IRPTL) .......................................... A-18
Interrupt Mask Register (IMASK) ......................................... A-23
Interrupt Mask Pointer Register (IMASKP) ........................... A-23
Link Port Interrupt Register (LIRPTL) .................................. A-24
Flag Value Register (FLAGS) ................................................. A-27
Program Counter Register (PC) ............................................. A-28
Program Counter Stack Register (PCSTK) ............................. A-29
Program Counter Stack Pointer Register (PCSTKP) ............... A-29
Fetch Address Register (FADDR) .......................................... A-30
Decode Address Register (DADDR) ...................................... A-30
Loop Address Stack Register (LADDR) .................................. A-30
Current Loop Counter Register (CURLCNTR) ..................... A-31
Loop Counter Register (LCNTR) .......................................... A-31
Timer Period Register (TPERIOD) ....................................... A-31
Timer Count Register (TCOUNT) ....................................... A-31
xxii
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generator Registers ............................................... A-32
Index Registers (Ix) ............................................................... A-32
Modify Registers (Mx) .......................................................... A-32
Length and Base Register (Lx, Bx) ......................................... A-33
I/O Processor Registers ............................................................... A-33
System Configuration Register (SYSCON) ............................ A-45
Vector Interrupt Address Register (VIRPT) ........................... A-48
External Memory Waitstate and Access Mode Register (WAIT) A-48
System Status Register (SYSTAT) .......................................... A-51
External Port DMA Buffer Registers (EPBx) .......................... A-52
Message Registers (MSGRx) ................................................. A-53
PC Shadow Register (PC_SHDW) ........................................ A-53
MODE2 Shadow Register (MODE2_SHDW) ...................... A-53
Bus Timeout Maximum Register (BMAX) ............................. A-53
Bus (Timeout) Counter Register (BCNT) ............................. A-54
Address of Last DRAM Page Register (ELAST) ..................... A-54
External Port DMA Control Registers (DMACx) .................. A-54
Internal Memory DMA Index Registers (IIx) ......................... A-58
Internal Memory DMA Modifier Registers (IMx) .................. A-59
Internal Memory DMA Count Registers (Cx) ....................... A-59
Chain Pointer For Next DMA TCB Registers (CPx) .............. A-59
General Purpose DMA Registers (GPx, DBx, DAx) ............... A-60
DMA Channel Status Register (DMASTAT) ......................... A-60
External Memory DMA Index Registers (EIx) ....................... A-61
ADSP-21160 SHARC DSP Hardware Reference
xxiii
External Memory DMA Modifier Registers (EMx) ................. A-61
External Memory DMA Count Registers (ECx) ..................... A-62
Link Port Buffer Registers (LBUFx) ....................................... A-62
Link Port Buffer Control Registers (LCTLx) .......................... A-62
Link Port Common Control Register (LCOM) ...................... A-65
Link Port Assignment Register (LAR) .................................... A-67
Link Port Service Request and Mask Register (LSRQ) ............ A-68
Link Port Path Registers (LPATHx) ....................................... A-70
Link Port Path Counter Register (LPCNT) ............................ A-70
Link Port Constant Registers (CNSTx) .................................. A-71
SPORT Serial Transmit Control Registers (STCTLx) ............. A-71
SPORT Serial Receive Control Registers (SRCTLx) ............... A-73
SPORT Transmit Buffer Registers (TXx) ............................... A-73
SPORT Receive Buffer Registers (RXx) .................................. A-76
SPORT Transmit Divisor Registers (TDIVx) ......................... A-76
SPORT Transmit Count Registers (TCNTx) ......................... A-77
SPORT Receive Divisor Registers (RDIVx) ........................... A-77
SPORT Receive Count Registers (RCNTx) ............................ A-78
SPORT Transmit Select Registers (MTCSx) ........................... A-78
SPORT Receive Select Registers (MRCSx) ............................. A-78
SPORT Transmit Compand Registers (MTCCSx) .................. A-79
SPORT Receive Compand Register (MRCCSx) ..................... A-79
SPORT Receive Comparison and Mask Registers (KEYWDx and
KEYMASKx) ..................................................................... A-80
xxiv
ADSP-21160 SHARC DSP Hardware Reference
SPORT Serial Path Length Registers (SPATHx) .................... A-80
SPORT Serial Path Counter Registers (SPCNTx) .................. A-80
Register and Bit #Defines File (def21160.h) ................................ A-81
INTERRUPT VECTOR ADDRESSES
NUMERIC FORMATS
IEEE Single-Precision Floating-Point Data Format ........................ C-1
Extended-Precision Floating-Point Format .................................... C-3
Short Word Floating-Point Format ............................................... C-4
Packing for Floating-Point Data ................................................... C-4
Fixed-Point Formats ..................................................................... C-6
GLOSSARY
INDEX
ADSP-21160 SHARC DSP Hardware Reference
xxv
xxvi
ADSP-21160 SHARC DSP Hardware Reference
Preface
PREFACE
Thank you for purchasing and developing systems using SHARC® processors from Analog Devices.
Purpose of This Manual
ADSP-21160 SHARC DSP Hardware Reference provides architectural
information on the ADSP-21160 Super Harvard Architecture (SHARC)
Digital Signal Processor (DSP). The architectural descriptions cover functional blocks, buses, and ports, including all features and processes they
support. For programming information, see ADSP-21160 SHARC DSP
Instruction Set Reference.
Intended Audience
The primary audience for this manual is a programmer who is familiar
with Analog Devices processors. The manual assumes the audience has a
working knowledge of the appropriate processor architecture and instruction set. Programmers who are unfamiliar with Analog Devices processors
can use this manual, but should supplement it with other texts, such as
hardware and programming reference manuals that describe their target
architecture.
ADSP-21160 SHARC DSP Hardware Reference
xxvii
Manual Contents
Manual Contents
This manual provides detailed information about the ADSP-214xx processor peripherals in the following chapters:
• Chapter 1, “Introduction”
Provides an architectural overview of the SHARC processors.
• Chapter 2, “Processing Elements”
Describes the arithmetic/logic units (ALUs), multiplier/accumulator units, and shifter. The chapter also discusses data formats, data
types, and register files.
• Chapter 3, “Program Sequencer”
Describes the operation of the program sequencer, which controls
program flow by providing the address of the next instruction to be
executed. The chapter also discusses loops, subroutines, jumps,
interrupts, exceptions, and the IDLE instruction.
• Chapter 4, “Data Address Generators”
Describes the Data Address Generators (DAGs), addressing modes,
how to modify DAG and pointer registers, memory address alignment, and DAG instructions.
• Chapter 5, “Memory”
Describes all aspects of processor memory including internal memory, address and data bus structure, and memory accesses.
• Chapter 6, “I/O Processor”
Describes input/output processor architecture.
• Chapter 7, “External Port”
Describes how the processor connects to external memories.
• Chapter 8, “Link Ports”
Describes the two bidirectional 8-bit wide link ports, which can
connect to other processor or peripheral link ports.
xxviii
ADSP-21160 SHARC DSP Hardware Reference
Preface
• Chapter 9, “Serial Ports”
Describes the data line serial ports. Each SPORT contains a clock,
a frame sync, and two data lines that can be configured as either a
receiver or transmitter pair.
• Chapter 10, “JTAG Test Emulation Port”
Discusses the JTAG standard and how to use the ADSP-21160 in a
test environment. Includes boundary-scan architecture, instruction, and breakpoint registers.
• Chapter 11, “System Design”
Describes system features of the ADSP-21160 processor. These
include power, reset, clock, JTAG, and booting, as well as pin
descriptions and other system level information.
• Appendix A, “Registers”
Provides a graphical presentation of all registers and describes the
bit usage in each register.
• Appendix B, “Interrupt Vector Addresses”
Provides descriptions of all ADSP-21160 DSP interrupts.
• Appendix C, “Numeric Formats”
Provides descriptions of the supported data formats.
hardware reference is a companion document to SHARC Pro This
cessor Programming Reference.
ADSP-21160 SHARC DSP Hardware Reference
xxix
What’s New in This Manual
What’s New in This Manual
This manual is Revision 4.1 of ADSP-21160 SHARC DSP Hardware Reference. This revision corrects minor typographical errors and the following
issues:
• Globally replaced ADSP-21535 with ADSP-21160.
• Active low signals represented correctly in equations for ALU conditions in Chapter 3, “Program Sequencer”.
• Bit 0 descriptions for the STYKx and STYKy registers and link port
interrupt bits of the LIRPTL register in Appendix A, “Registers”.
Technical Support
You can reach Analog Devices processors and DSP technical support in
the following ways:
• Post your questions in the processors and DSP support community
at EngineerZone®:
http://ez.analog.com/community/dsp
• Submit your questions to technical support directly at:
http://www.analog.com/support
• E-mail your questions about processors, DSPs, and tools development software from CrossCore® Embedded Studio or
VisualDSP++®:
xxx
ADSP-21160 SHARC DSP Hardware Reference
Preface
Choose Help > Email Support. This creates an e-mail to
[email protected] and automatically attaches
your CrossCore Embedded Studio or VisualDSP++ version information and license.dat file.
• E-mail your questions about processors and processor applications
to:
[email protected] or
[email protected] (Greater China support)
• In the USA only, call 1-800-ANALOGD (1-800-262-5643)
• Contact your Analog Devices sales office or authorized distributor.
Locate one at:
www.analog.com/adi-sales
• Send questions by mail to:
Processors and DSP Technical Support
Analog Devices, Inc.
Three Technology Way
P.O. Box 9106
Norwood, MA 02062-9106
USA
Supported Processors
The name “SHARC” refers to a family of high-performance, floating-point
embedded processors. Refer to the CCES or VisualDSP++ online help for
a complete list of supported processors.
Product Information
Product information can be obtained from the Analog Devices Web site
and the CCES or VisualDSP++ online help.
ADSP-21160 SHARC DSP Hardware Reference
xxxi
Product Information
Analog Devices Web Site
The Analog Devices Web site, www.analog.com, provides information
about a broad range of products—analog integrated circuits, amplifiers,
converters, and digital signal processors.
To access a complete technical library for each processor family, go to
http://www.analog.com/processors/technical_library. The manuals
selection opens a list of current manuals related to the product as well as a
link to the previous revisions of the manuals. When locating your manual
title, note a possible errata check mark next to the title that leads to the
current correction report against the manual.
Also note, myAnalog is a free feature of the Analog Devices Web site that
allows customization of a Web page to display only the latest information
about products you are interested in. You can choose to receive weekly
e-mail notifications containing updates to the Web pages that meet your
interests, including documentation errata against all manuals. myAnalog
provides access to books, application notes, data sheets, code examples,
and more.
Visit myAnalog to sign up. If you are a registered user, just log on. Your
user name is your e-mail address.
EngineerZone
EngineerZone is a technical support forum from Analog Devices, Inc. It
allows you direct access to ADI technical support engineers. You can
search FAQs and technical information to get quick answers to your
embedded processing and DSP design questions.
Use EngineerZone to connect with other DSP developers who face similar
design challenges. You can also use this open forum to share knowledge
and collaborate with the ADI support team and your peers. Visit
http://ez.analog.com to sign up.
xxxii
ADSP-21160 SHARC DSP Hardware Reference
Preface
Notation Conventions
Text conventions in this manual are identified and described as follows.
Example
Description
File > Close
Titles in reference sections indicate the location of an item within the
IDE environment’s menu system (for example, the Close command
appears on the File menu).
{this | that}
Alternative required items in syntax descriptions appear within curly
brackets and separated by vertical bars; read the example as this or
that. One or the other is required.
[this | that]
Optional items in syntax descriptions appear within brackets and separated by vertical bars; read the example as an optional this or that.
[this,…]
Optional item lists in syntax descriptions appear within brackets delimited by commas and terminated with an ellipsis; read the example as an
optional comma-separated list of this.
.SECTION
Commands, directives, keywords, and feature names are in text with
letter gothic font.
filename
Non-keyword placeholders appear in text with italic style format.

Note: For correct operation, ...
A Note provides supplementary information on a related topic. In the
online version of this book, the word Note appears instead of this
symbol.

Caution: Incorrect device operation may result if ...
Caution: Device damage may result if ...
A Caution identifies conditions or inappropriate usage of the product
that could lead to undesirable results or product damage. In the online
version of this book, the word Caution appears instead of this symbol.

Warning: Injury to device users may result if ...
A Warning identifies conditions or inappropriate usage of the product
that could lead to conditions that are potentially hazardous for devices
users. In the online version of this book, the word Warning appears
instead of this symbol.
ADSP-21160 SHARC DSP Hardware Reference
xxxiii
Register Diagram Conventions
Register Diagram Conventions
Register diagrams use the following conventions:
• The descriptive name of the register appears at the top, followed by
the short form of the name in parentheses.
• If the register is read-only (RO), write-1-to-set (W1S), or
write-1-to-clear (W1C), this information appears under the name.
Read/write is the default and is not noted. Additional descriptive
text may follow.
• If any bits in the register do not follow the overall read/write convention, this is noted in the bit description after the bit name.
• If a bit has a short name, the short name appears first in the bit
description, followed by the long name in parentheses.
• The reset value appears in binary in the individual bits and in hexadecimal to the right of the register.
• Bits marked x have an unknown reset value. Consequently, the
reset value of registers that contain such bits is undefined or dependent on pin values at reset.
• Shaded bits are reserved.
upward compatibility with future implementations,
 Towriteensure
back the value that is read for reserved bits in a register,
unless otherwise specified.
xxxiv
ADSP-21160 SHARC DSP Hardware Reference
Preface
The following figure shows an example of these conventions.
Timer Configuration Registers (TIMERx_CONFIG)
15 14 13 12 11 10
9
8
7
6
5
4
3
2
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
ERR_TYP[1:0] (Error Type) - RO
00 - No error.
01 - Counter overflow error.
10 - Period register programming error.
11 - Pulse width register programming error.
EMU_RUN (Emulation Behavior Select)
0 - Timer counter stops during emulation.
1 - Timer counter runs during emulation.
TOGGLE_HI (PWM_OUT PULSE_HI Toggle Mode)
0 - The effective state of PULSE_HI
is the programmed state.
1 - The effective state of PULSE_HI
alternates each period.
CLK_SEL (Timer Clock Select)
This bit must be set to 1, when operating the PPI in GP Output modes.
0 - Use system clock SCLK for counter.
1 - Use PWM_CLK to clock counter.
OUT_DIS (Output Pad Disable)
0 - Enable pad in PWM_OUT mode.
1 - Disable pad in PWM_OUT mode.
Reset = 0x0000
TMODE[1:0] (Timer Mode)
00 - Reset state - unused.
01 - PWM_OUT mode.
10 - WDTH_CAP mode.
11 - EXT_CLK mode.
PULSE_HI
0 - Negative action pulse.
1 - Positive action pulse.
PERIOD_CNT (Period
Count)
0 - Count to end of width.
1 - Count to end of period.
IRQ_ENA (Interrupt
Request Enable)
0 - Interrupt request
disable.
1 - Interrupt request enable
TIN_SEL (Timer Input
Select)
0 - Sample TMRx pin or
PF1 pin.
1 - Sample UART RX pin
or PPI_CLK pin.
Figure 1. Register Diagram Example
ADSP-21160 SHARC DSP Hardware Reference
xxxv
Register Diagram Conventions
xxxvi
ADSP-21160 SHARC DSP Hardware Reference
Introduction
1 INTRODUCTION
Thank you for purchasing Analog Devices SHARC® digital signal processor (DSP).
Overview – Why Floating-Point DSP?
A digital signal processor’s data format determines its ability to handle signals of differing precision, dynamic range, and signal-to-noise ratios.
Because floating-point DSP math reduces the need for scaling and probability of overflow, using a floating-point DSP can ease algorithm and
software development. The extent to which this is true depends on the
floating-point processor’s architecture. Consistency with IEEE workstation simulations and the elimination of scaling are two clear ease-of-use
advantages. High-level language programmability, large address spaces,
and wide dynamic range allow system development time to be spent on
algorithms and signal processing concerns, rather than assembly language
coding, code paging, and error handling. The ADSP-21160 is a highly
integrated, 32-bit floating-point DSP that provides many of these design
advantages.
ADSP-21160 SHARC DSP Hardware Reference
1-1
ADSP-21160 DSP Design Advantages
ADSP-21160 DSP Design Advantages
The ADSP-21160 processor is a high-performance 32-bit DSP for medical
imaging, communications, military, audio, test equipment, 3D graphics,
speech recognition, motor control, imaging, and other applications. This
DSP builds on the ADSP-21000 DSP core to form a complete system-on-a-chip, adding a dual-ported on-chip SRAM, integrated I/O
peripherals, and an additional processing element for Single-Instruction-Multiple-Data (SIMD) support.
SHARC is an acronym for Super Harvard Architecture. This DSP architecture balances a high performance processor core with high performance
buses (PM, DM, IO). In the core, every instruction can execute in a single
cycle. The buses and instruction cache provide rapid, unimpeded data
flow to the core to maintain the execution rate.
Figure 1-1 shows a detailed block diagram of the processor, illustrating the
following architectural features:
• Two Processing Elements (PEx and PEy), each containing a 32-Bit
IEEE floating-point computation units—multiplier, ALU, Shifter,
and data register file
• Program sequencer with related instruction cache, interval timer,
and Data Address Generators (DAG1 and DAG2)
• Dual-ported SRAM
• External port for interfacing to off-chip memory, peripherals,
hosts, and multiprocessor systems
• Input/Output (IO) processor with integrated DMA controller,
serial ports, and link ports for point-to-point multiprocessor
communications
• JTAG Test Access Port for emulation
1-2
ADSP-21160 SHARC DSP Hardware Reference
Introduction
Figure 1-1 also shows the three on-chip buses of the ADSP-21160: the
Program Memory (PM) bus, Data Memory (DM) bus, and Input/Output
(IO) bus. The PM bus provides access to either instructions or data.
During a single cycle, these buses let the processor access two data operands (one from PM and one from DM), access an instruction (from the
cache), and perform a DMA transfer.
The buses connect to the ADSP-21160 DSP’s external port, which provides the processor’s interface to external memory, memory-mapped I/O,
a host processor, and additional multiprocessing ADSP-21160 DSPs. The
external port performs bus arbitration and supplies control signals to
shared, global memory and I/O devices.
INS T R U CT ION
CACH E
T W O I NDE P E ND E NT
DU AL -P OR T E D B L OCK S
32 x 4 8- B IT
P R OCE S S OR P OR T
A DD R
D A TA
AD D R
DAG1
D AG2
8x4 x32
8 x4x 32
BUS
CONNECT
(P X)
B LO C K 0
T IME R
I/O P OR T
D A TA
DA T A
ADDR
D AT A
B LO C K 1
DUAL-PORTE D SRAM
CORE PROCESSOR
JT AG
6
T EST &
E MU L AT ION
AD D R
P R OGR AM
S E QU E NCE R
P M AD DR E S S B U S
32
D M AD DR E S S B U S
32
P M DAT A B U S
48/64
D M DAT A B U S
32/40/64
IOD
64
IOA
32
EXTERNAL
PO RT
AD DR B U S
MU X
32
MUL T IPR OCE S S OR
INT E R F ACE
DAT A B U S
MU X
64
H OS T P OR T
MU L T
DAT A
REGISTER
FILE
(PEx)
16 x 40- BIT
B AR R E L
S H IF T E R
AL U
B AR R E L
S H IF T E R
DATA
R EG ISTER
F ILE
(PEy)
16 x 40- BIT
MU L T
AL U
IOP
R E GIS T E R S
(MEMORY MAPPED)
CONT R OL ,
S T AT U S , &
D AT A B U F F E R S
DMA
CONTROLLER
S E R IAL P OR T S
(2)
L INK P OR T S
(6)
4
6
6
60
I/O PROCESSOR
Figure 1-1. ADSP-21160 SHARC DSP Block Diagram
Figure 1-2 illustrates a typical single-processor system. The ADSP-21160
DSP includes extensive support for multiprocessor systems as well. For
more information, see “Multiprocessor (DSPs) Interface” on page 7-91.
ADSP-21160 SHARC DSP Hardware Reference
1-3
ADSP-21160 DSP Design Advantages
ADSP-2116X
CLKIN
4
FLAG3-0
TIMEXP
LINK
DEVICES
(6 MAX)
(OPTIONAL)
LXACK
SERIAL
SERIAL
DEVICE
DEVICE
(OPTIONAL)
(OPTIONAL)
TCLK0
RCLK0
TFS0
RSF0
DT0
DR0
SERIAL
SERIAL
DEVICE
DEVICE
(OPTIONAL)
(OPTIONAL)
TCLK1
RCLK1
TFS1
RSF1
DT1
DR1
LXCLK
LXDAT7-0
RPBA
ID2-0
RESET
ADDR
CIF
DATA
BRST
ADDR31-0
ADDR
DATA63-0
DATA
OE
WE
RDX
WRX
ACK
ACK
CS
MS3-0
PAGE
SBTS
CLKOUT
DMAR1-2
BOOT
EPROM
(OPTIONAL)
MEMORY
AND
PERIPHERALS
(OPTIONAL)
DATA
3
EBOOT
LBOOT
IRQ2-0
CS
BMS
CLK_CFG3-0
ADDRESS
4
CONTROL
CLOCK
DMA DEVICE
(OPTIONAL)
DATA
DMAG1-2
CS
HOST
PROCESSOR
INTERFACE
(OPTIONAL)
HBR
HBG
REDY
BR1-6
ADDR
PA
DATA
JTAG
6
Figure 1-2. ADSP-21160 Processor System
Further, the ADSP-21160 DSP addresses the five central requirements for
DSPs:
• Fast, flexible arithmetic computation units
• Unconstrained data flow to and from the computation units
• Extended precision and dynamic range in the computation units
• Dual address generators with circular buffering support
• Efficient program sequencing
1-4
ADSP-21160 SHARC DSP Hardware Reference
Introduction
Fast, Flexible Arithmetic. The ADSP-21000 Family processors execute all
instructions in a single cycle. They provide both fast cycle times and a
complete set of arithmetic operations. The DSP is IEEE floating-point
compatible and allows either interrupt on arithmetic exception or latched
status exception handling.
Unconstrained Data Flow. The ADSP-21160 DSP has a Super Harvard
Architecture combined with a 10-port data register file. In every cycle, the
DSP can write or read two operands to or from the register file, supply
two operands to the ALU, supply two operands to the multiplier, and
receive three results from the ALU and multiplier. The processor’s 48-bit
orthogonal instruction word supports parallel data transfers and arithmetic operations in the same instruction.
40-Bit Extended Precision. The DSP handles 32-bit IEEE floating-point
format, 32-bit integer and fractional formats (twos-complement and
unsigned), and extended-precision 40-bit floating-point format. The processors carry extended precision throughout their computation units,
limiting intermediate data truncation errors.
Dual Address Generators. The DSP has two data address generators
(DAGs) that provide immediate or indirect (pre- and post-modify)
addressing. Modulus, bit-reverse, and broadcast operations are supported
with no constraints on data buffer placement.
Efficient Program Sequencing. In addition to zero-overhead loops, the
DSP supports single-cycle setup and exit for loops. Loops are both
nestable (six levels in hardware) and interruptable. The processors support
both delayed and non-delayed branches
ADSP-21160 DSP Architecture Overview
The ADSP-21160 DSP forms a complete system-on-a-chip, integrating a
large, high-speed SRAM and I/O peripherals supported by a dedicated
ADSP-21160 SHARC DSP Hardware Reference
1-5
ADSP-21160 DSP Architecture Overview
I/O bus. The following sections summarize the features of each functional
block in the ADSP-21160 SHARC architecture, which appears in
Figure 1-1 on page 1-3. With each summary, a cross reference points to
the sections where the features are described in greater detail.
Processor Core
The processor core of the ADSP-21160 DSP consists of two processing
elements (each with three computation units and data register file), a program sequencer, two data address generators, a timer, and an instruction
cache. All digital signal processing occurs in the processor core.
Processing Elements
The processor core contains two Processing Elements (PEx and PEy).
Each element contains a data register file and three independent computation units: an ALU, a multiplier with a fixed-point accumulator, and a
shifter. For meeting a wide variety of processing needs, the computation
units process data in three formats: 32-bit fixed-point, 32-bit floating-point and 40-bit floating-point. The floating-point operations are
single-precision IEEE-compatible. The 32-bit floating-point format is the
standard IEEE format, whereas the 40-bit extended-precision format has
eight additional Least Significant Bits (LSBs) of mantissa for greater
accuracy.
The ALU performs a set of arithmetic and logic operations on both
fixed-point and floating-point formats. The multiplier performs floating-point or fixed-point multiplication and fixed-point multiply/add or
multiply/subtract operations. The shifter performs logical and arithmetic
shifts, bit manipulation, field deposit and extraction, and exponent derivation operations on 32-bit operands.
These computation units perform single-cycle operations; there is no computation pipeline. All units are connected in parallel, rather than serially.
The output of any unit may serve as the input of any unit on the next
1-6
ADSP-21160 SHARC DSP Hardware Reference
Introduction
cycle. In a multifunction computation, the ALU and multiplier perform
independent, simultaneous operations.
Each processing element has a general-purpose data register file that transfers data between the computation units and the data buses and stores
intermediate results. A register file has two sets (primary and alternate) of
sixteen registers each, for fast context switching. All of the registers are 40
bits wide. The register file, combined with the core processor’s Harvard
architecture, allows unconstrained data flow between computation units
and internal memory.
Primary Processing Element (PEx). PEx processes all computational
instructions whether the DSP is in Single-Instruction, Single-Data (SISD)
or Single-Instruction, Multiple-Data (SIMD) mode. This element corresponds to the computational units and register file in previous
ADSP-21000 DSPs.
Secondary Processing Element (PEy). PEy processes each computational
instruction in lock-step with PEx, but only processes these instructions
when the DSP is in SIMD mode. Because many operations are influenced
by this mode, more information on SIMD is available in multiple
locations:
• For information on PEy operations, see “Processing Elements”
• For information on data addressing in SIMD mode, see “Addressing in SISD and SIMD Modes” on page 4-18
• For information on data accesses in SIMD mode, see “SISD,
SIMD, and Broadcast Load Modes” on page 5-44
• For information on multiprocessing in SIMD mode, see “Multiprocessor (DSPs) Interface” on page 7-91.
• For information on SIMD programming, see ADSP-21160 SHARC
DSP Instruction Set Reference.
ADSP-21160 SHARC DSP Hardware Reference
1-7
ADSP-21160 DSP Architecture Overview
Program Sequence Control
Internal controls for ADSP-21160 DSP’s program execution come from
four functional blocks: program sequencer, data address generators, timer,
and instruction cache. Two dedicated address generators and a program
sequencer supply addresses for memory accesses. Together the sequencer
and data address generators allow computational operations to execute
with maximum efficiency since the computation units can be devoted
exclusively to processing data. With its instruction cache, the
ADSP-21160DSP can simultaneously fetch an instruction from the cache
and access two data operands from memory. The data address generators
implement circular data buffers in hardware.
Program Sequencer. The program sequencer supplies instruction
addresses to program memory. It controls loop iterations and evaluates
conditional instructions. With an internal loop counter and loop stack,
the ADSP-21160 DSP executes looped code with zero overhead. No
explicit jump instructions are required to loop or decrement and test the
counter.
The ADSP-21160 DSP achieves its fast execution rate by means of pipelined fetch, decode and execute cycles. If external memories are used, they
are allowed more time to complete an access than if there were no decode
cycle.
Data Address Generators. The data address generators (DAGs) provide
memory addresses when data is transferred between memory and registers.
Dual data address generators enable the processor to output simultaneous
addresses for two operand reads or writes. DAG1 supplies 32-bit addresses
to data memory. DAG2 supplies 32-bit addresses to program memory for
program memory data accesses.
Each DAG keeps track of up to eight address pointers, eight modifiers and
eight length values. A pointer used for indirect addressing can be modified
by a value in a specified register, either before (pre-modify) or after
(post-modify) the access. A length value may be associated with each
1-8
ADSP-21160 SHARC DSP Hardware Reference
Introduction
pointer to perform automatic modulo addressing for circular data buffers;
the circular buffers can be located at arbitrary boundaries in memory.
Each DAG register has an alternate register that can be activated for fast
context switching.
Circular buffers allow efficient implementation of delay lines and other
data structures required in digital signal processing, and are commonly
used in digital filters and Fourier transforms. The DAGs automatically
handle address pointer wraparound, reducing overhead, increasing performance, and simplifying implementation.
Interrupts. The ADSP-21160 DSP has four external hardware interrupts:
three general-purpose interrupts, IRQ2-0, and a special interrupt for reset.
The processor also has internally generated interrupts for the timer, DMA
controller operations, circular buffer overflow, stack overflows, arithmetic
exceptions, multiprocessor vector interrupts, and user-defined software
interrupts.
For the general-purpose external interrupts and the internal timer interrupt, the ADSP-21160 DSP automatically stacks the arithmetic status and
mode (MODE1) registers in parallel with the interrupt servicing, allowing fifteen nesting levels of very fast service for these interrupts.
Context Switch. Many of the processor’s registers have alternate registers
that can be activated during interrupt servicing for a fast context switch.
The data registers in the register file, the DAG registers, and the multiplier
result register all have alternates. The Primary Registers are active at reset,
while the Alternate (or Secondary) Registers are activated by control bits
in a mode control register.
Timer. The programmable interval timer provides periodic interrupt generation. When enabled, the timer decrements a 32-bit count register every
cycle. When this count register reaches zero, the ADSP-21160 DSP generates an interrupt and asserts its timer expired output. The count register is
automatically reloaded from a 32-bit period register and the count
resumes immediately.
ADSP-21160 SHARC DSP Hardware Reference
1-9
ADSP-21160 DSP Architecture Overview
Instruction Cache. The program sequencer includes a 32-word instruction cache that enables three-bus operation for fetching an instruction and
two data values. The cache is selective—only instructions whose fetches
conflict with program memory data accesses are cached. This caching
allows full-speed execution of core, looped operations such as digital filter
multiply-accumulates and FFT butterfly processing.
Processor Internal Buses
The processor core has six buses: PM address, PM data, DM address, DM
data, IO address, and IO data. Due to processor’s Harvard Architecture,
data memory stores data operands, while program memory can store both
instructions and data. This architecture allows dual data fetches, when the
instruction is supplied by the cache.
Bus Capacities. The PM address bus and DM address bus transfer the
addresses for instructions and data. The PM data bus and DM data bus
transfer the data or instructions from each type of memory. The PM
address bus is 32 bits wide allowing access of up to 4 Gwords of mixed
instructions and data. The PM data bus is 64 bits wide to accommodate
the 48-bit instructions and 64-bit data.
The DM address bus is 32 bits wide allowing direct access of up to 4G
words of data. The DM data bus is 64 bits wide. The DM data bus provides a path for the contents of any register in the processor to be
transferred to any other register or to any data memory location in a single
cycle. The data memory address comes from one of two sources: an absolute value specified in the instruction code (direct addressing) or the
output of a data address generator (indirect addressing).
The IO address and IO data buses let the IO processor access internal
memory for DMA without delaying the processor core. The IO address
bus is 32 bits wide, and the IO data bus is 64 bits wide.
Data Transfers. Nearly every register in the processor core is classified as a
Universal Register (UREG). Instructions allow transferring data between
1-10
ADSP-21160 SHARC DSP Hardware Reference
Introduction
any two universal registers or between a universal register and memory.
This support includes transfers between control registers, status registers,
and data registers in the register file. The PM bus connect (PX) registers
permit data to be passed between the 64-bit PM data bus and the 64-bit
DM data bus or between the 40-bit register file and the PM data bus.
These registers contain hardware to handle the data width difference. For
more information, see “Processing Element Registers” on page A-15.
Processor Peripherals
The term processor peripherals refers to everything outside the processor
core. The ADSP-21160 DSP’s peripherals include internal memory, external port, I/O processor, JTAG port, and any external devices that connect
to the DSP.
Dual-Ported Internal Memory (SRAM)
The ADSP-21160 DSP contains 4 megabits of on-chip SRAM, organized
as two blocks of 2 Mbits each, which can be configured for different combinations of code and data storage. Each memory block is dual-ported for
single-cycle, independent accesses by the core processor and I/O processor
or DMA controller. The dual-ported memory and separate on-chip buses
allow two data transfers from the core and one from I/O, all in a single
cycle.
All of the memory can be accessed as 16-, 32-, 48-, or 64-bit words. On
the ADSP-21160 DSP, the memory can be configured as a maximum of
128K words of 32-bit data, 256K words of 16-bit data, 80K words of
48-bit instructions (and 40-bit data), or combinations of different word
sizes up to 4 megabits.
The DSP supports a 16-bit floating-point storage format, which effectively doubles the amount of data that may be stored on chip. Conversion
between the 32-bit floating-point and 16-bit floating-point formats completes in a single instruction.
ADSP-21160 SHARC DSP Hardware Reference
1-11
ADSP-21160 DSP Architecture Overview
While each memory block can store combinations of code and data,
accesses are most efficient when one block stores data, using the DM bus
for transfers, and the other block stores instructions and data, using the
PM bus for transfers. Using the DM bus and PM bus in this way, with one
dedicated to each memory block, assures single-cycle execution with two
data transfers. In this case, the instruction must be available in the cache.
The DSP also maintains single-cycle execution when one of the data operands is transferred to or from off-chip, using the DSP’s external port.
External Port
The ADSP-21160 DSP’s external port provides the processor’s interface
to off-chip memory and peripherals. The 4-gigaword off-chip address
space is included in the ADSP-21160 DSP’s unified address space. The
separate on-chip buses—for PM address, PM data, DM address, DM data,
IO address, and IO data—multiplex at the external port to create an external system bus with a single 32-bit address bus and a single 64-bit data
bus. External SRAM can be 16, 32, 48, or 64 bits wide; the DSP’s on-chip
DMA controller automatically packs external data into the appropriate
word width during transfers.
On-chip decoding of high-order address lines generates memory bank
select signals for addressing external memory devices. Separate control
lines support simplified addressing of page-mode DRAM. The
ADSP-21160 DSP provides programmable memory waitstates and external memory acknowledge controls for interfacing to DRAM and
peripherals with variable access, hold, and disable time requirements.
Host Processor Interface. The ADSP-21160 DSP’s host interface allows
easy connection to standard microprocessor buses, both 16-bit and 32-bit,
with little additional hardware required. The interface supports asynchronous and synchronous transfers at speeds up to the half the internal clock
rate of the ADSP-21160 DSP. The host interface operates through the
DSP’s external port and maps into the unified address space. Four channels of DMA are available for the host interface; code and data transfers
1-12
ADSP-21160 SHARC DSP Hardware Reference
Introduction
occur with low software overhead. The host can directly read and write the
internal memory of the ADSP-21160 DSP and can access the DMA channel setup and mailbox registers. Vector interrupt support provides for
efficient execution of host commands.
Multiprocessor System Interface. The ADSP-21160 DSP offers powerful
features tailored to multiprocessing DSP systems. The unified address
space allows direct interprocessor accesses of each ADSP-21160 DSP’s
internal memory. Distributed bus arbitration logic on the DSP allows
simple, glueless connection of systems containing up to six ADSP-21160
DSPs and a host processor. Master processor changeover incurs only one
cycle of overhead. Bus arbitration handles either fixed or rotating priority.
Processor bus lock allows indivisible read-modify-write sequences for
semaphores. A vector interrupt capability is provided for interprocessor
commands. Broadcast writes allow simultaneous transmission of data to
all ADSP-21160 DSPs and can be used to implement reflective
semaphores.
I/O Processor
The ADSP-21160 DSP’s Input/Output Processor (IOP) includes two
serial ports, six link ports, and a DMA controller. One of the I/O processes that the IO processor automates is booting. The DSP can boot from
the external port (with data from an 8-bit EPROM or a host processor) or
a link port. Alternatively, a no-boot mode lets the DSP start by executing
instructions from external memory without booting.
Serial Ports. The ADSP-21160 DSP features two synchronous serial ports
that provide an inexpensive interface to a wide variety of digital and
mixed-signal peripheral devices. The serial ports can operate at up to half
the processor core clock rate. Independent transmit and receive functions
provide greater flexibility for serial communications. Serial port data can
automatically transfer to and from on-chip memory using DMA. Each of
the serial ports offers a TDM multichannel mode and supports m-law or
A-law companding.
ADSP-21160 SHARC DSP Hardware Reference
1-13
ADSP-21160 DSP Architecture Overview
The serial ports can operate with little-endian or big-endian transmission
formats, with word lengths selectable from 3 to 32 bits. They offer selectable synchronization and transmit modes. Serial port clocks and frame
syncs can be internally or externally generated.
Link Ports. The ADSP-21160 DSP features six 8-bit link ports that provide additional I/O capabilities. Link port I/O is especially useful for
point-to-point interprocessor communication in multiprocessing systems.
The link ports can operate independently and simultaneously. The data
packs into 32-bit or 48-bit words, which the processor core can directly
read or the IO processor can DMA-transfer to on-chip memory.
Clock/acknowledge handshaking controls link port transfers. Transfers are
programmable as either transmit or receive.
DMA Controller. The ADSP-21160 DSP’s on-chip DMA controller
allows zero-overhead data transfers without processor intervention. The
DMA controller operates independently and invisibly to the processor
core to enable DMA operations to occur while the core is simultaneously
executing its program. Both code and data can be downloaded to the
ADSP-21160 DSP using DMA transfers.
DMA transfers can occur between the ADSP-21160 DSP’s internal memory and external memory, external peripherals, or a host processor. DMA
transfers can also occur between the ADSP-21160 DSP’s internal memory
and its serial ports or link ports. DMA transfers between external memory
and external peripheral devices are another option. External bus packing
to 16-, 32-, 48-, or 64-bit words is automatically performed during DMA
transfers.
Fourteen channels of DMA are available on the ADSP-21160 DSP—six
over the link ports, four over the serial ports, and four over the processor’s
external port. The external port DMA channels serve for host processor,
other ADSP-21160 DSPs, memory, or I/O transfers.
1-14
ADSP-21160 SHARC DSP Hardware Reference
Introduction
JTAG Port
The JTAG port on the ADSP-21160 DSP supports the IEEE standard
1149.1 Joint Test Action Group (JTAG) standard for system test. This
standard defines a method for serially scanning the I/O status of each
component in a system. Emulators use the JTAG port to monitor and
control the DSP during emulation. Emulators using this port provide
full-speed emulation with access to inspect and modify memory, registers,
and processor stacks. JTAG-based emulation is non-intrusive and does not
effect target system loading or timing.
Development Tools
The processor is supported by a complete set of software and hardware
development tools, including Analog Devices’ emulators and the CrossCore Embedded Studio or VisualDSP++ development environment. (The
emulator hardware that supports other Analog Devices processors also
emulates the processor.)
The development environments support advanced application code development and debug with features such as:
• Create, compile, assemble, and link application programs written
in C++, C, and assembly
• Load, run, step, halt, and set breakpoints in application programs
• Read and write data and program memory
• Read and write core and peripheral registers
• Plot memory
Analog Devices DSP emulators use the IEEE 1149.1 JTAG test access
port to monitor and control the target board processor during emulation.
The emulator provides full speed emulation, allowing inspection and
ADSP-21160 SHARC DSP Hardware Reference
1-15
Differences From Previous SHARC DSPs
modification of memory, registers, and processor stacks. Nonintrusive
in-circuit emulation is assured by the use of the processor JTAG interface—the emulator does not affect target system loading or timing.
Software tools also include Board Support Packages (BSPs). Hardware
tools also include standalone evaluation systems (boards and extenders). In
addition to the software and hardware development tools available from
Analog Devices, third parties provide a wide range of tools supporting the
Blackfin processors. Third party software tools include DSP libraries,
real-time operating systems, and block diagram design tools.
Differences From Previous SHARC DSPs
This section identifies differences between the ADSP-21160 DSP and previous SHARC DSPs: ADSP-21060, ADSP-21061, and ADSP-21062
processors. The ADSP-21160 DSP preserves much of the ADSP-2106x
architecture, while extending performance and functionality. For background information on SHARC and the ADSP-2106x DSPs, see
ADSP-2106x SHARC User’s Manual.
Processor Core Enhancements
Computational bandwidth on the ADSP-21160 DSP is significantly
greater that on the ADSP-2106x DSPs. The increase comes from raising
the operational frequency and adding another processing element: ALU,
Shifter, Multiplier, and register file. The new processing element lets the
DSP process multiple data streams in parallel (SIMD mode).
The program sequencer on the ADSP-21160 DSP differs from the
ADSP-2106x DSP family, having several enhancements: new interrupt
vector table definitions, SIMD mode stack and conditional execution
model, and instruction decodes associated with new instructions. Changes
to interrupts include new interrupt vectors for detecting illegal memory
accesses and supporting new unshared DMA channels. Link port interrupt
1-16
ADSP-21160 SHARC DSP Hardware Reference
Introduction
control has moved to a new register to support the additional DMA channels. Also, new mode stack and mode mask support has been added to
improve context switch time.
Data address generators on the ADSP-21160 DSP differ from the
ADSP-2106x DSPs in that DAG2 (for the PM bus) has the same addressing capability as DAG1 (for the DM bus). The DAG registers are
read/writable in pairs, moving 64-bits/cycle. Additionally, the DAGs support the new memory map and Long Word transfer capability. Circular
buffering on the ADSP-21160 DSP can be quickly disabled on interrupts
and restored on the return. Data “broadcast”, from one memory location
to both data register files, is determined by appropriate index register
usage.
previous SHARCs, the ADSP-21160 DSP has a global cir Unlike
cular buffering enable (
) bit. Because at reset this bit defaults
CBUFEN
to disabled, programs that use circular buffering and are being
ported from previous SHARCs need to add a line of code to enable
circular buffering. For more information, see “Addressing Circular
Buffers” on page 4-12.
Processor Internal Bus Enhancements
The PM, DM, and IO data buses on the ADSP-21160 DSP are much
wider than on the ADSP-2106x DSPs, increasing to 64 bits. Additional
multiplexing and control logic on the ADSP-21160 DSP enables 16-, 32-,
or 64-bit wide moves between both register files and memory.
The ADSP-21160 DSP also has the capability of broadcasting a single
memory location to each of the register files in parallel. Also, the
ADSP-21160 DSP permits register contents to be exchanged between the
two processing elements’ register files in a single cycle.
ADSP-21160 SHARC DSP Hardware Reference
1-17
Differences From Previous SHARC DSPs
Memory Organization Enhancements
The ADSP-21160 memory map differs from the ADSP-2106x DSPs. The
system memory map on the ADSP-21160DSP supports double-word
transfers each cycle, reflects extended internal memory capacity for derivative designs, and works with updated control register for SIMD support.
External Port Enhancements
The ADSP-21160 DSP’s external port differs from the ADSP-2106x
DSPs, greatly extending the external interface. The data bus on the
ADSP-21160 DSP is 64 bits wide. The ADSP-21160 DSP has a new synchronous interface that improves local bus switching frequency. Also,
burst support on the ADSP-21160 DSP improves bus usage.
previous SHARC DSPs, the ADSP-21160DSP sets the buf Unlike
fer hang disable ( ) bit at reset. Because this bit prevents the
BHD
processor core from detecting a buffer-related stall condition, programs that use external port, link port, or serial port I/O and are
being ported from previous SHARC DSPs need to add a line of
code to disable BHD. For more information, see the BHD discussion
on page 6-14.
Host Interface Enhancements
The ADSP-21160’s host interface differs from the ADSP-2106x DSPs in
that this interface can take advantage of the 64-bit data bus width.
Though the ADSP-21160 DSP supports the ADSP-2106x’s asynchronous
host interface protocols, the ADSP-21160 DSP also provides new synchronous interface protocols for maximum throughput.
The host/local bus deadlock resolution function on the ADSP-21160 DSP
is extended to the DMA controller. The function allows the host (or
bridge) logic to force the local bus to back off and allow the host to complete it’s operation first.
1-18
ADSP-21160 SHARC DSP Hardware Reference
Introduction
Multiprocessor Interface Enhancements
The ADSP-21160’s multiprocessor system interface supports greater
throughput than the ADSP-2106x DSPs. The throughput between
ADSP-21160 DSPs in a multiprocessing application increases due to
shared data bus width increase to 64-bits, new shared bus transfer protocols, shared bus cycle time improvements due to synchronous interface,
and improvements in Link Port throughput. The external port supports
glueless multiprocessing, with distributed arbitration for up to six
ADSP-21160 DSPs.
IO Architecture Enhancements
The IO processor on the ADSP-21160 DSP provides much greater
throughput than the ADSP-2106x DSPs. The Link Ports and DMA controller differ on the ADSP-21160 DSP.
DMA Controller Enhancements
The ADSP-21160’s DMA controller supports 14 channels (versus 10 on
the ADSP-2106x DSPs), with no channel sharing. New packing modes
support the new 64-bit external/internal busing. To resolve potential
deadlock scenarios, the ADSP-21106’s DMA controller relinquishes the
local bus in a similar fashion to the processor core when host logic asserts
both HBR and SBTS.
Link Port Enhancements
The ADSP-21160’s Link ports provide greater throughput than the
ADSP-2106x DSPs. The link port data bus width on the ADSP-21160
DSP is 8 bits wide (versus 4 bits on the ADSP-2106x DSPs). Link port
clock control on the ADSP-21160 supports a wider frequency range.
ADSP-21160 SHARC DSP Hardware Reference
1-19
Differences From Previous SHARC DSPs
Instruction Set Enhancements
ADSP-21160 DSP provides source code compatibility with the previous
SHARC family members, to the application assembly source code level.
All instructions, control registers, and system resources available in t the
ADSP-2106x core programming model are available in ADSP-21160
DSP.
New instructions, control registers, or other facilities, required to support
the new feature set of ADSP-21160 processor core are:
• Supersets of the ADSP-2106x programming model
• Reserved facilities in the ADSP-2106x programming model
• Symbol name changes from the ADSP-2106x programming model
These name changes can be managed through re-assembly using the
ADSP-21160 DSP’s development tools to apply the ADSP-21160 symbol
definitions header file and linker description file. While these changes
have no direct impact on existing core applications, system and I/O processor initialization code and control code do require modifications.
This approach simplifies porting of source code written for the
ADSP-2106x DSPs to ADSP-21160 DSP. Code changes will be required
to take full advantage of the new ADSP-21160 DSP features. For more
information, see ADSP-21160 SHARC DSP Instruction Set Reference.
1-20
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
2 PROCESSING ELEMENTS
The DSP’s Processing Elements (PEx and PEy) perform numeric processing for DSP algorithms. Each processing element contains a data register
file and three computation units: an arithmetic/logic unit (ALU), a multiplier, and a shifter. Computational instructions for these elements include
both fixed-point and floating-point operations, and each computational
instruction can execute in a single cycle.
Overview
The computational units in a processing element handle different types of
operations. The ALU performs arithmetic and logic operations on
fixed-point and floating-point data. The multiplier does floating-point
and fixed-point multiplication and executes fixed-point multiply/add and
multiply/subtract operations. The shifter completes logical shifts, arithmetic shifts, bit manipulation, field deposit, and field extraction operations
on 32-bit operands. Also, the Shifter can derive exponents.
Data flow paths through the computational units are arranged in parallel,
as shown in Figure 2-1. The output of any computation unit may serve as
the input of any computation unit on the next instruction cycle. Data
moving in and out of the computational units goes through a 10-port register file, consisting of sixteen primary registers and sixteen alternate
registers. Two ports on the register file connect to the PM and DM data
ADSP-21160 SHARC DSP Hardware Reference
2-1
Setting Computational Modes
buses, allowing data transfer between the computational units and memory (and anything else) connected to these buses.
The DSP’s assembly language provides access to the data register files in
both processing elements. The syntax lets programs move data to and
from these registers and specify a computation’s data format at the same
time with naming conventions for the registers. For information on the
data register names, see “Data Register File” on page 2-28 provides a
graphical guide to the other topics in this chapter. First, a description of
the MODE1 register shows how to set rounding, data format, and other
modes for the processing elements. Next, an examination of each computational unit provides details on operation and a summary of
computational instructions. Outside the computational units, details on
register files and data buses identify how to flow data for computations.
Finally, details on the DSP’s advanced parallelism reveal how to take
advantage of multifunction instructions and SIMD mode.
Setting Computational Modes
The MODE1 register controls the operating mode of the processing elements. Table A-2 on page A-3 lists all the bits in MODE1. The following bits
in MODE1 control computational modes:
• Floating-point data format. Bit 16 (RND32) directs the computational units to round floating-point data to 32 bits (if 1) or round
to 40 bits (if 0)
• Rounding mode. Bit 15 (TRUNC) directs the computational units to
round results with round-to-zero (if 1) or round-to-nearest (if 0)
• ALU saturation. Bit 13 (ALUSAT) directs the computational units to
saturate results on positive or negative fixed-point overflows (if 1)
or return unsaturated results (if 0)
2-2
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
PM DATA BUS
MODE1
DM DATA BUS
REGISTER FILE
(16  40-BIT)
X
MULTIPLIER
Y
R0
R1
R2
R3
R8
R9
R10
R11
R4
R5
R6
R7
R12
R13
R14
R15
Z
Y
SHIFTER
X
Y
X
ALU
MR2F MR1F MR0F
ASTATx
STKYx
TO PROGRAM SEQUENCER
Figure 2-1. Computations Units
• Short word sign extension. Bit 14 (SSE) directs the computational
units to sign extend short-word, 16-bit data (if 1) or zero-fill the
upper 16 bits (if 0)
• Secondary processor element (PEy). Bit 21 (PEYEN) enables computations in PEy—SIMD mode—(if 1) or disables PEy—SISD
mode—(if 0)
32-bit (Normal Word) Floating-Point Format
In the default mode of the DSP (RND32 bit=1), the multiplier and ALU
support a single-precision floating-point format, which is specified in the
ADSP-21160 SHARC DSP Hardware Reference
2-3
Setting Computational Modes
IEEE 754/854 standard. For more information on this standard, see
“Numeric Formats”. This format is IEEE 754/854 compatible for single-precision floating-point operations in all respects except that:
• The DSP does not provide inexact flags.
• NAN (“Not-A-Number”) inputs generate an invalid exception and
return a quiet NAN (all 1s).
• Denormal operands flush to zero when input to a computation
unit and do not generate an underflow exception. Any denormal or
underflow result from an arithmetic operation flushes to zero and
generates an underflow exception.
• The DSP supports round to nearest and round toward zero modes,
but does not support round to +Infinity and round to -Infinity.
IEEE single-precision floating-point data uses a 23-bit mantissa with an
8-bit exponent plus sign bit. In this case, the computation unit sets the
eight LSBs of floating-point inputs to zeros before performing the operation. The mantissa of a result rounds to 23 bits (not including the hidden
bit), and the 8 LSBs of the 40-bit result clear to zeros to form a 32-bit
number, which is equivalent to the IEEE standard result.
In fixed-point to floating-point conversion, the rounding boundary is
always 40 bits even if the RND32 bit is set.
40-bit Floating-Point Format
When in extended precision mode (RND32 bit=0), the DSP supports a
40-bit extended precision floating-point mode, which has eight additional
LSBs of the mantissa and is compliant with the 754/854 standards; however, results in this format are more precise than the IEEE single-precision
standard specifies. Extended-precision floating-point data uses a 31-bit
mantissa with a 8-bit exponent plus sign bit.
2-4
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
16-bit (Short Word) Floating-Point Format
The DSP supports a 16-bit floating-point storage format and provides
instructions that convert the data for 40-bit computations. The 16-bit
floating-point format uses an 11-bit mantissa with a 4-bit exponent plus
sign bit. The 16-bit data goes into bits 23 through 8 of a data register.
Two shifter instructions, Fpack and Funpack, perform the packing and
unpacking conversions between 32-bit floating-point words and 16-bit
floating-point words. The Fpack instruction converts a 32-bit IEEE floating-point number in a data register into a 16-bit floating-point number.
Funpack converts a 16-bit floating-point number in a data register into a
32-bit IEEE floating-point number. Each instruction executes in a single
cycle.
When 16-bit data is written to bits 23 through 8 of a data register, the
DSP automatically extends the data into a 32-bit integer (bits 39 through
8). If the SSE bit in MODE1 is set (1), the DSP sign extends the upper 16
bits. If the SSE bit is cleared (0), the DSP zeros the upper 16 bits.
The 16-bit floating-point format supports gradual underflow. This
method sacrifices precision for dynamic range. When packing a number
that would have underflowed, the exponent clears to zero and the mantissa
(including “hidden” 1) right-shifts the appropriate amount. The packed
result is a denormal, which can be unpacked into a normal IEEE floating-point number.
32-Bit Fixed-Point Format
The DSP always represents fixed-point numbers in 32 bits, occupying the
32 MSBs in 40-bit data registers. Fixed-point data may be fractional or
integer numbers and unsigned or twos-complement. Each computational
unit has its own limitations on how these formats may be mixed for a
given operation. All computational units read the upper 32 bits of data
(inputs, operands) from the 40-bit registers (ignoring the 8 LSBs) and
write results to the upper 32 bits (zeroing the 8 LSBs).
ADSP-21160 SHARC DSP Hardware Reference
2-5
Setting Computational Modes
Rounding Mode
The TRUNC bit in the MODE1 register determines the rounding mode for all
ALU operations, all floating-point multiplies, and fixed-point multiplies
of fractional data. The DSP supports two modes of rounding:
round-toward-zero and round-toward-nearest. The rounding modes comply with the IEEE 754 standard and have the following definitions:
• Round-Toward-Zero (TRUNC bit=1). If the result before rounding is
not exactly representable in the destination format, the rounded
result is the number that is nearer to zero. This is equivalent to
truncation.
• Round-Toward-Nearest (TRUNC bit=0). If the result before rounding is not exactly representable in the destination format, the
rounded result is the number that is nearer to the result before
rounding. If the result before rounding is exactly halfway between
two numbers in the destination format (differing by an LSB), the
rounded result is the number that has an LSB equal to zero.
Statistically, rounding up occurs as often as rounding down, so there is no
large sample bias. Because the maximum floating-point value is one LSB
less than the value that represents Infinity, a result that is halfway between
the maximum floating-point value and Infinity rounds to Infinity in this
mode.
Though these rounding modes comply with standards set for floating-point data, they also apply for fixed-point multiplier operations on
fractional data. The same two rounding modes are supported, but only the
round-to-nearest operation is actually performed by the multiplier. Using
its local result register for fixed-point operations, the multiplier
rounds-to-zero by reading only the upper bits of the result and discarding
the lower bits.
2-6
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
Using Computational Status
The multiplier and ALU each provide exception information when executing floating-point operations. Each unit updates overflow, underflow,
and invalid operation flags in the processing element’s arithmetic status
(ASTATx and ASTATy) register and sticky status (STKYx and STKYy) register.
An underflow, overflow, or invalid operation from any unit also generates
a maskable interrupt. There are three ways to use floating-point exceptions from computations in program sequencing:
• Interrupts. Enable interrupts and use an interrupt service routine
to handle the exception condition immediately. This method is
appropriate if it is important to correct all exceptions as they occur.
•
ASTATx and ASTATy registers. Use conditional instructions to test
the exception flags in the ASTATx or ASTATy register after the
instruction executes. This method permits monitoring each
instruction’s outcome.
•
STKYx and STKYy registers. Use the Bit Tst instruction to examine
exception flags in the STKY register after a series of operations. If
any flags are set, some of the results are incorrect. This method is
useful when exception handling is not critical.
More information on ASTAT and STKY status appears in the sections that
describe the computational units. For summaries relating instructions and
status bits, see Table 2-1 on page 2-11, Table 2-2 on page 2-12, Table 2-4
on page 2-19,Table 2-6 on page 2-20,and Table 2-7 on page 2-26.
Arithmetic Logic Unit (ALU)
The ALU performs arithmetic operations on fixed-point or floating-point
data and logical operations on fixed-point data. ALU fixed-point instructions operate on 32-bit fixed-point operands and output 32-bit
fixed-point results. ALU floating-point instructions operate on 32-bit or
ADSP-21160 SHARC DSP Hardware Reference
2-7
Arithmetic Logic Unit (ALU)
40-bit floating-point operands and output 32-bit or 40-bit floating-point
results. ALU instructions include:
• Floating-point addition, subtraction, add/subtract, average
• Fixed-point addition, subtraction, add/subtract, average
• Floating-point manipulation: binary log, scale, mantissa
• Fixed-point add with carry, subtract with borrow, increment,
decrement
• Logical And, Or, Xor, Not
• Functions: Abs, pass, min, max, clip, compare
• Format conversion
• Reciprocal and reciprocal square root primitives
ALU Operation
ALU instructions take one or two inputs: X input and Y input. These
inputs (also known as operands) can be any data registers in the register
file. Most ALU operations return one result; in add/subtract operations,
the ALU operation returns two results, and in compare operations, the
ALU operation returns no result (only flags are updated). ALU results can
be returned to any location in the register file.
The DSP transfers input operands from the register file during the first
half of the cycle and transfers results to the register file during the second
half of the cycle. With this arrangement, the ALU can read and write the
same register file location in a single cycle. If the ALU operation is
fixed-point, the inputs are treated as 32-bit fixed-point operands. The
ALU transfers the upper 32 bits from the source location in the register
file. For fixed-point operations, the result(s) are always 32-bit fixed-point
2-8
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
values. Some floating-point operations (Logb, Mant and Fix) can also
yield fixed-point results.
The DSP transfers fixed-point results to the upper 32 bits of the data register and clears the lower eight bits of the register. The format of
fixed-point operands and results depends on the operation. In most arithmetic operations, there is no need to distinguish between integer and
fractional formats. Fixed-point inputs to operations such as scaling a floating-point value are treated as integers. For purposes of determining status
such as overflow, fixed-point arithmetic operands and results are treated as
twos-complement numbers.
ALU Saturation
When the ALUSAT bit is set (1) in the MODE1 register, the ALU is in saturation mode. In this mode, all positive fixed-point overflows return the
maximum positive fixed-point number (0x7FFF FFFF), and all negative
overflows return the maximum negative number (0x8000 0000).
When the ALUSAT bit is cleared (0) in the MODE1 register, fixed-point results
that overflow are not saturated; the upper 32 bits of the result are returned
unaltered.
The ALU overflow flag reflects the ALU result before saturation.
ALU Status Flags
ALU operations update seven status flags in the processing element’s
Arithmetic Status (ASTATx and ASTATy) register. Table A-4 on page A-9
lists all the bits in these registers. The following bits in ASTATx or ASTATy
flag ALU status (a 1 indicates the condition) for the most recent ALU
operation:
• ALU result zero or floating-point underflow. Bit 0 (AZ)
• ALU overflow. Bit 1 (AV)
ADSP-21160 SHARC DSP Hardware Reference
2-9
Arithmetic Logic Unit (ALU)
• ALU result negative. Bit 2 (AN)
• ALU fixed-point carry. Bit 3 (AC)
• ALU X input sign for Abs, Mant operations. Bit 4 (AS)
• ALU floating-point invalid operation. Bit 5 (AI)
• Last ALU operation was a floating-point operation. Bit 10 (AF)
• Compare Accumulation register results of last 8 compare operations. Bits 31-24 (CACC)
ALU operations also update four “sticky” status flags in the processing element’s Sticky status (STKYx and STKYy) register. Table A-5 on page A-13
lists all the bits in these registers. The following bits in STKYx or STKYy flag
ALU status (a 1 indicates the condition). Once set, a sticky flag remains
high until explicitly cleared:
• ALU floating-point underflow. Bit 0 (AUS)
• ALU floating-point overflow. Bit 1 (AVS)
• ALU fixed-point overflow. Bit 2 (AOS)
• ALU floating-point invalid operation. Bit 5 (AIS)
Flag update occurs at the end of the cycle in which the status is generated
and is available on the next cycle. If a program writes the arithmetic status
register or sticky status register explicitly in the same cycle that the ALU is
performing an operation, the explicit write to the status register supersedes
any flag update from the ALU operation.
ALU Instruction Summary
Table 2-1 and Table 2-2 list the ALU instructions and how they relate to
ASTATx,y and STKYx,y flags. For more information on assembly language
2-10
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
syntax, see ADSP-21160 SHARC DSP Instruction Set Reference. In these
tables, note the meaning of the following symbols:
• Rn, Rx, Ry indicate a register file location; treated as fixed-point
• Fn, Fx, Fy indicate a register file location; treated as floating-point
• * indicates that the flag may be set or cleared, depending on results
of instruction
• ** indicates that the flag may be set (but not cleared), depending
on results of instruction
• – indicates no effect
Table 2-1. Fixed-Point ALU Instruction Summary
Instruction
ASTATx,y Status Flags
Fixed-point:
A
Z
AV A
N
A
C
AS
AI
AF
C
A
C
C
A
AV A
US S
O
S
AI
S
Rn = Rx + Ry
*
*
*
*
0
0
0
–
–
–
**
–
Rn = Rx – Ry
*
*
*
*
0
0
0
–
–
–
**
–
Rn = Rx + Ry + CI
*
*
*
*
0
0
0
–
–
–
**
–
Rn = Rx – Ry + CI – 1
*
*
*
*
0
0
0
–
–
–
**
–
Rn = (Rx + Ry)/2
*
0
*
*
0
0
0
–
–
–
–
–
COMP(Rx, Ry)
*
0
*
0
0
0
0
*
–
–
–
–
COMPU(Rx,Ry)
*
0
*
0
0
0
0
*
--
--
--
--
Rn = Rx + CI
*
*
*
*
0
0
0
–
–
–
**
–
Rn = Rx + CI – 1
*
*
*
*
0
0
0
–
–
–
**
–
Rn = Rx + 1
*
*
*
*
0
0
0
–
–
–
**
–
Rn = Rx – 1
*
*
*
*
0
0
0
–
–
–
**
–
Rn = –Rx
*
*
*
*
0
0
0
–
–
–
**
–
Rn = ABS Rx
*
*
0
0
*
0
0
–
–
–
**
–
ADSP-21160 SHARC DSP Hardware Reference
STKYx,y Status Flags
2-11
Arithmetic Logic Unit (ALU)
Table 2-1. Fixed-Point ALU Instruction Summary (Cont’d)
Instruction
ASTATx,y Status Flags
STKYx,y Status Flags
Fixed-point:
A
Z
AV A
N
A
C
AS
AI
AF
C
A
C
C
A
AV A
US S
O
S
AI
S
Rn = PASS Rx
*
0
*
0
0
0
0
–
–
–
–
–
Rn = Rx AND Ry
*
0
*
0
0
0
0
–
–
–
–
–
Rn = Rx OR Ry
*
0
*
0
0
0
0
–
–
–
–
–
Rn = Rx XOR Ry
*
0
*
0
0
0
0
–
–
–
–
–
Rn = NOT Rx
*
0
*
0
0
0
0
–
–
–
–
–
Rn = MIN(Rx, Ry)
*
0
*
0
0
0
0
–
–
–
–
–
Rn = MAX(Rx, Ry)
*
0
*
0
0
0
0
–
–
–
–
–
Rn = CLIP Rx BY Ry
*
0
*
0
0
0
0
–
–
–
–
–
Table 2-2. Floating-Point ALU Instruction Summary
Instruction
ASTATx,y Status Flags
Floating–point:
AZ
AV
AN
AC
AS
AI
AF
CA
CC
AU
S
AV
S
AO
S
AIS
Fn = Fx + Fy
*
*
*
0
0
*
1
–
**
**
–
**
Fn = Fx – Fy
*
*
*
0
0
*
1
–
**
**
–
**
Fn = ABS (Fx + Fy)
*
*
0
0
0
*
1
–
**
**
–
**
Fn = ABS (Fx – Fy)
*
*
0
0
0
*
1
–
**
**
–
**
Fn = (Fx + Fy)/2
*
0
*
0
0
*
1
–
**
–
–
**
COMP(Fx, Fy)
*
0
*
0
0
*
1
*
–
–
–
**
Fn = –Fx
*
*
*
0
0
*
1
–
–
**
–
**
Fn = ABS Fx
*
*
0
0
*
*
1
–
–
**
–
**
Fn = PASS Fx
*
0
*
0
0
*
1
–
–
–
–
**
Fn = RND Fx
*
*
*
0
0
*
1
–
–
**
–
**
Fn = SCALB Fx BY Ry
*
*
*
0
0
*
1
–
**
**
–
**
Rn = MANT Fx
*
*
0
0
*
*
1
–
–
**
–
**
2-12
STKYx,y Status Flags
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
Table 2-2. Floating-Point ALU Instruction Summary (Cont’d)
Instruction
ASTATx,y Status Flags
STKYx,y Status Flags
Floating–point:
AZ
AV
AN
AC
AS
AI
AF
CA
CC
AU
S
AV
S
AO
S
AIS
Rn = LOGB Fx
*
*
*
0
0
*
1
–
–
**
–
**
Rn = FIX Fx BY Ry
*
*
*
0
0
*
1
–
**
**
–
**
Rn = FIX Fx
*
*
*
0
0
*
1
–
**
**
–
**
Fn = FLOAT Rx BY Ry
*
*
*
0
0
0
1
–
**
**
–
–
Fn = FLOAT Rx
*
0
*
0
0
0
1
–
–
–
–
–
Fn = RECIPS Fx
*
*
*
0
0
*
1
–
**
**
–
**
Fn = RSQRTS Fx
*
*
*
0
0
*
1
–
–
**
–
**
Fn = Fx COPYSIGN
Fy
*
0
*
0
0
*
1
–
–
–
–
**
Fn = MIN(Fx, Fy)
*
0
*
0
0
*
1
–
–
–
–
**
Fn = MAX(Fx, Fy)
*
0
*
0
0
*
1
–
–
–
–
**
Fn = CLIP Fx BY Fy
*
0
*
0
0
*
1
–
–
–
–
**
Multiply—Accumulator (Multiplier)
The multiplier performs fixed-point or floating-point multiplication and
fixed-point multiply/accumulate operations. Fixed-point multiply/accumulates are available with either cumulative addition or cumulative
subtraction. Multiplier floating-point instructions operate on 32-bit or
40-bit floating-point operands and output 32-bit or 40-bit floating-point
results. Multiplier fixed-point instructions operate on 32-bit fixed-point
data and produce 80-bit results. Inputs are treated as fractional or integer,
unsigned or twos-complement. Multiplier instructions include:
• Floating-point multiplication
• Fixed-point multiplication
• Fixed-point multiply/accumulate with addition, rounding optional
ADSP-21160 SHARC DSP Hardware Reference
2-13
Multiply—Accumulator (Multiplier)
• Fixed-point multiply/accumulate with subtraction, rounding
optional
• Rounding result register
• Saturating result register
• Clearing result register
Multiplier Operation
The multiplier takes two inputs: X input and Y input. These inputs (also
known as operands) can be any data registers in the register file. The
multiplier can accumulate fixed-point results in the local Multiplier Result
(MRF) registers or write results back to the register file. The results in MRF
can also be rounded or saturated in separate operations. Floating-point
multiplies yield floating-point results, which the multiplier always writes
directly to the register file.
The multiplier transfers input operands during the first half of the cycle
and transfers results during the second half of the cycle. With this arrangement, the multiplier can read and write the same register file location in a
single cycle.
For fixed-point multiplies, the multiplier reads the inputs from the upper
32 bits of the data registers. Fixed-point operands may be either both in
integer format or both in fractional format. The format of the result
matches the format of the inputs. Each fixed-point operand may be either
an unsigned or a twos-complement number. If both inputs are fractional
and signed, the multiplier automatically shifts the result left one bit to
remove the redundant sign bit. The register name(s) within the multiplier
instruction specify input data type(s)—Fx for floating-point and Rx for
fixed-point.
2-14
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
Multiplier (Fixed-Point) Result Register
Fixed-point operations place 80-bit results in the multiplier’s foreground
MRF register or background MRB register, depending on which is active. For
more information on selecting the result register, see “Alternate (Secondary) Data Registers” on page 2-30.
The location of a result in the MRF register’s 80-bit field depends on
whether the result is in fractional or integer format, as shown in Table 2-1
on page 2-11. If the result is sent directly to a data register, the 32-bit
result with the same format as the input data is transferred, using bits
63-32 for a fractional result or bits 31-0 for an integer result. The eight
LSBs of the 40-bit register file location are zero-filled.
Fractional results can be rounded-to-nearest before being sent to the register file. If rounding is not specified, discarding bits 31-0 effectively
truncates a fractional result (rounds to zero). For more information on
rounding, see “Rounding Mode” on page 2-6.
79
63
31
MR2F
MR1F
MR0F
OVERFLOW
FRACTIONAL RESULT
UNDERFLOW
OVERFLOW
OVERFLOW
INTEGER RESULT
Figure 2-2. Multiplier Fixed-Point Result Placement
The MRF register is divided into MR2F, MR1F, and MR0F registers, which can
be individually read from or written to the register file. Each of these registers has the same format. When data is read from MR2F, it is
sign-extended to 32 bits as shown in Figure 2-3. The DSP zero fills the
eight LSBs of the 40-bit register file location when data is read from MR2F,
MR1F, or MR0F to the register file. When the DSP writes data into MR2F,
ADSP-21160 SHARC DSP Hardware Reference
2-15
Multiply—Accumulator (Multiplier)
MR1F,
or MR0F from the 32 MSBs of a register file location, the eight LSBs
are ignored. Data written to MR1F is sign-extended to MR2F, repeating the
MSB of MR1F in the 16 bits of MR2F. Data written to MR0F is not
sign-extended.
16 BITS
SIGN EXTEND
16 BITS
16 BITS
MR2F
ZEROS
32 BITS
8 BITS
MR1F
ZEROS
32- BITS
8- BITS
MR0F
ZEROS
Figure 2-3. MR Transfer Formats
In addition to multiplication, fixed-point operations include accumulation, rounding and saturation of fixed-point data. There are three MRF
register operations: Clear, Round, and Saturate.
The clear operation—MRF=0—resets the specified MRF register to zero.
Often, it is best to perform this operation at the start of a multiply/accumulate operation to remove results left over from the previous operation.
The rounding operation—MRF=Rnd MRF—applies only to fractional
results, so integer results are not effected. This operation rounds the
80-bit MRF value to nearest at bit 32; for example, the MR1F-MR0F boundary.
Rounding of a fixed-point result occurs either as part of a multiply or multiply/accumulate operation or as an explicit operation on the MRF register.
The rounded result in MR1F can be sent either to the register file or back to
the same MRF register. To round a fractional result to zero (truncation)
instead of to nearest, a program would transfer the unrounded result from
MR1F, discarding the lower 32 bits in MR0F.
2-16
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
The saturate operation—MRF=Sat MRF—sets MRF to a maximum value if the
MRF value has overflowed. Overflow occurs when the MRF value is greater
than the maximum value for the data format—unsigned or twos-complement and integer or fractional—as specified in the saturate instruction.
The six possible maximum values appear in Table 2-3. The result from
MRF saturation can be sent either to the register file or back to the same MRF
register.
Table 2-3. Fixed-Point Format Maximum Values (for Saturation)
Maximum Number
(Hexadecimal)
MR2F
MR1F
MR0F
2’s complement fractional (positive)
0000
7FFF FFFF
FFFF FFFF
2’s complement fractional (negative)
FFFF
8000 0000
0000 0000
2’s complement integer (positive)
0000
0000 0000
7FFF FFFF
2’s complement integer (negative)
FFFF
FFFF FFFF
8000 0000
Unsigned fractional number
0000
FFFF FFFF
FFFF FFFF
Unsigned integer number
0000
0000 0000
FFFF FFFF
Multiplier Status Flags
Multiplier operations update four status flags in the processing element’s
arithmetic status register (ASTATx and ASTATy). Table A-4 on page A-9 lists
all the bits in these registers. The following bits in ASTATx or ASTATy flag
multiplier status (a 1 indicates the condition) for the most recent multiplier operation:
• Multiplier result negative. Bit 6 (MN)
• Multiplier overflow. Bit 7 (MV)
• Multiplier underflow. Bit 8 (MU)
• Multiplier floating-point invalid operation. Bit 9 (MI)
ADSP-21160 SHARC DSP Hardware Reference
2-17
Multiply—Accumulator (Multiplier)
Multiplier operations also update four “sticky” status flags in the processing element’s Sticky status ( STKYx and STKYy) register. Table A-5 on
page A-13 lists all the bits in these registers. The following bits in STKYx or
STKYy flag multiplier status (a 1 indicates the condition). Once set, a sticky
flag remains high until explicitly cleared:
• Multiplier fixed-point overflow. Bit 6 (MOS)
• Multiplier floating-point overflow. Bit 7 (MVS)
• Multiplier underflow. Bit 8 (MUS)
• Multiplier floating-point invalid operation. Bit 9 (MIS)
Flag update occurs at the end of the cycle in which the status is generated
and is available on the next cycle. If a program writes the arithmetic status
register or sticky register explicitly in the same cycle that the multiplier is
performing an operation, the explicit write to ASTAT or STKY supersedes
any flag update from the multiplier operation.
Multiplier Instruction Summary
Table 2-4 on page 2-19 and Table 2-6 on page 2-20 list the Multiplier
instructions and how they relate to ASTATx,y and STKYx,y flags. For more
information on assembly language syntax, see ADSP-21160 SHARC DSP
Instruction Set Reference. In these tables, note the meaning of the following
symbols:
• Rn, Rx, Ry indicate any register file location; treated as fixed-point
• Fn, Fx, Fy indicate any register file location; treated as
floating-point
• * indicates the flag may be set or cleared, depending on results of
instruction
2-18
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
• ** indicates the flag may be set (but not cleared), depending on
results of instruction
• – indicates no effect
• The Input Mods column indicates the types of optional modifiers
that you can apply to the instructions inputs. For a list of modifiers, see Table 2-5.
Table 2-4. Fixed-Point Multiplier Instruction Summary
Instruction
Input
Mods
ASTATx,y Flags
MU
MN
MV
MI
MU
S
MO
S
MVS MIS
Rn = Rx * Ry
1
*
*
*
0
–
**
–
–
MRF = Rx * Ry
1
*
*
*
0
–
**
–
–
MRB = Rx * Ry
1
*
*
*
0
–
**
–
–
Rn = MRF + Rx * Ry
1
*
*
*
0
–
**
–
–
Rn = MRB + Rx * Ry
1
*
*
*
0
–
**
–
–
MRF = MRF + Rx * Ry
1
*
*
*
0
–
**
–
–
MRB = MRB + Rx * Ry
1
*
*
*
0
–
**
–
–
Rn = MRF – Rx * Ry
1
*
*
*
0
–
**
–
–
Rn = MRB – Rx * Ry
1
*
*
*
0
–
**
–
–
MRF = MRF – Rx * Ry
1
*
*
*
0
–
**
–
–
MRB = MRB – Rx * Ry
1
*
*
*
0
–
**
–
–
Rn = SAT MRF
2
*
*
*
0
–
**
–
–
Rn = SAT MRB
2
*
*
*
0
–
**
–
–
MRF = SAT MRF
2
*
*
*
0
–
**
–
–
MRB = SAT MRB
2
*
*
*
0
–
**
–
–
Rn = RND MRF
3
*
*
*
0
–
**
–
–
Rn = RND MRB
3
*
*
*
0
–
**
–
–
MRF = RND MRF
3
*
*
*
0
–
**
–
–
Fixed-Point:
For Input Mods, see
Table 2-5
ADSP-21160 SHARC DSP Hardware Reference
STKYx,y Flags
2-19
Multiply—Accumulator (Multiplier)
Table 2-4. Fixed-Point Multiplier Instruction Summary (Cont’d)
Instruction
Input
Mods
ASTATx,y Flags
MU
MN
MV
MI
MU
S
MO
S
MVS MIS
MRB = RND MRB
3
*
*
*
0
–
**
–
–
MRF= 0
–
–
–
–
–
–
–
–
–
MRB= 0
–
–
–
–
–
–
–
–
–
MRxF = Rn
–
–
–
–
–
–
–
–
–
MRxB = Rn
–
–
–
–
–
–
–
–
–
Rn = MRxF
–
–
–
–
–
–
–
–
–
Rn = MRxB
–
–
–
–
–
–
–
–
–
Fixed-Point:
For Input Mods, see
Table 2-5
STKYx,y Flags
Table 2-5. Input Modifiers for Fixed-Point Multiplier Instruction
Input
Mods from
Table 2-2
Input Mods—Options For Fixed-point Multiplier Instructions
Note the meaning of the following symbols in this table:
SSigned input
UUnsigned input
IInteger input(s)
FFractional input(s)
FRFractional inputs, Rounded output
Note that (SF) is the default format for 1-input operations, and (SSF) is the
default format for 2-input operations
1
(SSF), (SSI), (SSFR), (SUF), (SUI), (SUFR), (USF), (USI), (USFR), (UUF),
(UUI), or (UUFR)
2
(SF), (SI), (UF), or (UI)
3
(SF) or (UF)
Table 2-6. Floating-Point Multiplier Instruction Summary
Instruction
ASTATx,y Flags
Floating-Point:
MU
MN
MV
MI
MUS MOS MVS
MIS
Fn = Fx * Fy
*
*
*
*
**
**
2-20
STKYx,y Flags
–
**
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
Barrel-Shifter (Shifter)
The shifter performs bit-wise operations on 32-bit fixed-point operands.
Shifter operations include:
• Shifts and rotates from off-scale left to off-scale right
• Bit manipulation operations, including bit set, clear, toggle, and
test
• Bit field manipulation operations, including extract and deposit
• Fixed-point/floating-point conversion operations, including exponent extract, number of leading 1s or 0s
Shifter Operation
The shifter takes from one to three inputs: X-input, Y-input, and Z-input.
The inputs (also known as operands) can be any register in the register
file. Within a shifter instruction, the inputs serve as follows:
• The X-input provides data that is operated on
• The Y-input specifies shift magnitudes, bit field lengths or bit
positions
• The Z-input provides data that is operated on and updated
In the following example, Rx is the X-input, Ry is the Y-input, and Rn is
the Z-input. The shifter returns one output (Rn) to the register file.
Rn = Rn OR LSHIFT Rx BY Ry;
As shown in Figure 2-4, the shifter fetches input operands from the upper
32 bits of a register file location (bits 39-8) or from an immediate value in
the instruction. The shifter transfers operands during the first half of the
cycle and transfers the result to the upper 32 bits of a register (with the
eight LSBs zero-filled) during the second half of the cycle. With this
ADSP-21160 SHARC DSP Hardware Reference
2-21
Barrel-Shifter (Shifter)
arrangement, the shifter can read and write the same register file location
in a single cycle.
The X-input and Z-input are always 32-bit fixed-point values. The
Y-input is a 32-bit fixed-point value or an 8-bit field (shf8), positioned in
the register file. These inputs appear in Figure 2-4.
Some shifter operations produce 8-bit or 6-bit results. As shown in
Figure 2-5, the shifter places these results in either the shf8 field or the
bit6 field and sign-extends the results to 32 bits. The shifter always returns
a 32-bit result.
39
7
0
7
0
32-BIT Y-INPUT OR RESULT
39
15
SHF8
8-BIT Y-INPUT OR RESULT
Figure 2-4. Register File Fields for Shifter Instructions
The shifter supports bit field deposit and bit field extract instructions for
manipulating groups of bits within an input. The Y-input for bit field
instructions specifies two 6-bit values: bit6 and len6, which are positioned
in the Ry register as shown in Figure 2-5. The shifter interprets bit6 and
len6 as positive integers. Bit6 is the starting bit position for the deposit or
extract, and len6 is the bit field length, which specifies how many bits are
deposited or extracted.
2-22
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
39
19
13
7
LEN6
0
BIT6
12-BIT Y-INPUT
Figure 2-5. Register File Fields for FDEP and FEXT Instructions
Field deposit (Fdep) instructions take a group of bits from the input register (starting at the LSB of the 32-bit integer field) and deposit the bits as
directed anywhere within the result register. The bit6 value specifies the
starting bit position for the deposit. Figure 2-6 shows bit placement for
the following field deposit instruction:
R0 = FDEP R1 BY R2;
39
R2
32
00000000
24
00000000
16
8
bit6
len6
39
R1
32
00000000
24
00000000
16
00000000
16
39
R0
32
00000000
8
16
0
0 0 0 0 0 0 0 0 0x0000 00FF 00
0
16
00000000
len6 = 8
bit6 = 16
11111111
8
24
11111111
0
0 0 0 0 0 0 0 0 0x0000 0210 00
00000010 0001 0000
8
8
Starting bit
position
for deposit
0
00000000
00000000
0x00FF 0000 00
0
Reference
point
Figure 2-6. Bit Field Deposit Example
ADSP-21160 SHARC DSP Hardware Reference
2-23
Barrel-Shifter (Shifter)
Figure 2-7 shows how the inputs, bit6 and len6, work in an field deposit
instruction (Rn=Fdep Rx By Ry). Field extract (Fext) instructions extract
a group of bits as directed from anywhere within the input register and
place them in the result register (aligned with the LSB of the 32-bit integer field). The bit6 value specifies the starting bit position for the extract.
Figure 2-8 shows bit placement for the following field extract instruction:
R3 = FEXT R4 BY R5;
39
R5
32
00000000
24
00000000
16
0 0 00 11 00 11 11 11
00000010
len6
39
R4
32
10000111
24
16
39
32
00000000
0
00000000
0x0000 0217 00
len6 = 8
bit6 = 23
8
00000000
8
Starting bit position
for deposit
R3
bit6
16
00000000
1 0000000
8
0
00000000
0x8788 0000 00
0
Reference point
24
00000000
16
00000000
16
8
8
00000111
0
00000000
0x0000 000F 00
0
Figure 2-7. Bit Field Extract Example
2-24
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
39
19
RY
13
LEN6
7
0
BIT6
RY DETERMINES LENGTH OF BIT FIELD TO TAKE FROM RX AND STARTING POSITION FOR DEPOSIT IN RN
39
7
0
7
0
RX
LEN6 = NUMBER OF BITS TO TAKE FROM RX, STARTING FROM LSB OF 32-BIT FIELD
39
RN
DEPOSIT FIELD
BIT6
REFERENCE POINT
BIT6 = STARTING BIT POSITION FOR DEPOSIT, REFERENCED FROM LSB OF 32-BIT FIELD
Figure 2-8. Bit Field Deposit Instruction
Shifter Status Flags
Shifter operations update three status flags in the processing element’s
arithmetic status register (ASTATx and ASTATy). Table A-4 on page A-9 lists
all the bits in these registers. The following bits in ASTATx or ASTATy flag
shifter status (a 1 indicates the condition) for the most recent ALU
operation:
• Shifter overflow of bits to left of MSB. Bit 11 (SV)
• Shifter result zero. Bit 12 (SZ)
• Shifter input sign for exponent extract only. Bit 13 (SS)
Flag update occurs at the end of the cycle in which the status is generated
and is available on the next cycle. If a program writes the arithmetic status
register explicitly in the same cycle that the shifter is performing an
ADSP-21160 SHARC DSP Hardware Reference
2-25
Barrel-Shifter (Shifter)
operation, the explicit write to ASTAT supersedes any flag update caused by
the shift operation.
Shifter Instruction Summary
Table 2-7 on page 2-26 lists the Shifter instructions and how they relate
to ASTATx,y flags. For more information on assembly language syntax, see
ADSP-21160 SHARC DSP Instruction Set Reference. In these tables, note
the meaning of the following symbols:
• Rn, Rx, Ry indicate any register file location; bit fields used depend
on instruction
• Fn, Fx indicate any register file location; floating-point word
• * indicates the flag may set or cleared, depending on data
Table 2-7. Shifter Instruction Summary
Instruction
2-26
ASTATx,y Flags
SZ
SV
SS
Rn = LSHIFT Rx BY Ry
*
*
0
Rn = LSHIFT Rx BY <data8>
*
*
0
Rn = Rn OR LSHIFT Rx BY Ry
*
*
0
Rn = Rn OR LSHIFT Rx BY <data8>
*
*
0
Rn = ASHIFT Rx BY Ry
*
*
0
Rn = ASHIFT Rx BY<data8>
*
*
0
Rn = Rn OR ASHIFT Rx BY Ry
*
*
0
Rn = Rn OR ASHIFT Rx BY <data8>
*
*
0
Rn = ROT Rx BY Ry
*
0
0
Rn = ROT Rx BY <data8>
*
0
0
Rn = BCLR Rx BY Ry
*
*
0
Rn = BCLR Rx BY <data8>
*
*
0
Rn = BSET Rx BY Ry
*
*
0
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
Table 2-7. Shifter Instruction Summary (Cont’d)
Instruction
ASTATx,y Flags
SZ
SV
SS
Rn = BSET Rx BY <data8>
*
*
0
Rn = BTGL Rx BY Ry
*
*
0
Rn = BTGL Rx BY <data8>
*
*
0
BTST Rx BY Ry
*
*
0
BTST Rx BY <data8>
*
*
0
Rn = FDEP Rx BY Ry
*
*
0
Rn = FDEP Rx BY <bit6>:<len6>
*
*
0
Rn = Rn OR FDEP Rx BY Ry
*
*
0
Rn = Rn OR FDEP Rx BY <bit6>:<len6>
*
*
0
Rn = FDEP Rx BY Ry (SE)
*
*
0
Rn = FDEP Rx BY <bit6>:<len6> (SE)
*
*
0
Rn = Rn OR FDEP Rx BY Ry (SE)
*
*
0
Rn = Rn OR FDEP Rx BY <bit6>:<len6> (SE)
*
*
0
Rn = FEXT Rx BY Ry
*
*
0
Rn = FEXT Rx BY <bit6>:<len6>
*
*
0
Rn = FEXT Rx BY Ry (SE)
*
*
0
Rn = FEXT Rx BY <bit6>:<len6> (SE)
*
*
0
Rn = EXP Rx (EX)
*
0
*
Rn = EXP Rx
*
0
*
Rn = LEFTZ Rx
*
*
0
Rn = LEFTO Rx
*
*
0
Rn = FPACK Fx
0
*
0
Fn = FUNPACK Rx
0
0
0
ADSP-21160 SHARC DSP Hardware Reference
2-27
Data Register File
Data Register File
Each of the DSP’s processing elements has a data register file: a set of data
registers that transfer data between the data buses and the computation
units. These registers also provide local storage for operands and results.
The two register files each consist of 16 primary registers and 16 alternate
(secondary) registers. All of the data registers are 40 bits wide. Within
these registers, 32-bit data is always left-justified. If an operation specifies
a 32-bit data transfer to these 40-bit registers, the eight LSBs are ignored
on register reads, and the eight LSBs are cleared to zeros on writes.
Program memory data accesses and data memory accesses to/from the register file(s) occur on the PM data bus and DM data bus, respectively. One
PM data bus access for each processing element and/or one DM data bus
access for each processing element can occur in one cycle. Transfers
between the register files and the DM or PM data buses can move up to
64-bits of valid data on each bus.
If an operation specifies the same register file location as both an input
and output, the read occurs in the first half of the cycle and the write in
the second half. With this arrangement, the DSP uses the old data as the
operand, before updating the location with the new result data. If writes
to the same location take place in the same cycle, only the write with
higher precedence actually occurs. The DSP determines precedence for
the write operation from the source of the data; from highest to lowest,
the precedence is:
1. Data memory or universal register
2. Program memory
3. PEx ALU
4. PEy ALU
5. PEx Multiplier
2-28
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
6. PEy Multiplier
7. PEx Shifter
8. PEy Shifter
The data register file in Figure 2-1 on page 2-3 lists register names of R0
through R15 within PEx’s register file. When a program refers to these registers as R0 through R15, the computational units treat the registers’
contents as fixed-point data. To perform floating point computations,
refer to these registers as F0 through F15. For example, the following
instructions refer to the same registers, but direct the computational units
to perform different operations:
F0=F1 * F2; floating-point multiply
R0=R1 * R2; fixed-point multiply
The F and R prefixes on register names do not effect the 32-bit or 40-bit
data transfer; the naming convention only determines how the ALU, multiplier, and shifter treat the data.
compatibility with code written for previous SHARC
 ToDSPs,maintain
the assembly syntax accommodates references to
data
PEx
registers and PEy data registers.
Code may only refer to the PEy data registers (S0 through S15) for data
move instructions. The rules for using register names are as follows:
•
R0 through R15 and F0 through F15 always refer to PEx registers for
data move and computational instructions, whether the DSP is in
SISD or SIMD mode
•
R0 through R15 and F0 through F15 refer to both PEx and PEy register for computational instructions in SIMD mode
•
S0
through S15 always refer to PEy registers for data move instructions, whether the DSP is in SISD or SIMD mode
ADSP-21160 SHARC DSP Hardware Reference
2-29
Alternate (Secondary) Data Registers
For more information on SISD and SIMD computational operations, see
“Secondary Processing Element (PEy)” on page 2-35. For more information on ADSP-21160 assembly language, see ADSP-21160 SHARC DSP
Instruction Set Reference.
Alternate (Secondary) Data Registers
Each register file has an alternate register set. To facilitate fast context
switching, the DSP includes alternate register sets for data, results, and
data address generator registers. Bits in the MODE1 register control when
alternate registers become accessible. While inaccessible, the contents of
alternate registers are not effected by DSP operations. Note that there is a
one cycle latency between writing to MODE1 and being able to access an
alternate register set. The alternate register sets for data and results are
described in this section. For more information on alternate data address
generator registers, see “Alternate (Secondary) Data Registers” on
page 2-30.
Bits in the MODE1 register can activate independent-alternate-data-register
sets: the lower half (R0-R7 and S0-S7) and the upper half (R8-R15 and
S8-S15). To share data between contexts, a program places the data to be
shared in one half of either the current processing element’s register file or
the opposite processing element’s register file and activates the alternate
register set of the other half. For information on how to activate alternate
data registers, see the description on page 2-30.
Each multiplier has a primary or foreground (MRF) register and alternate or
background (MRB) results register. A bit in the MODE1 register selects which
result register receives the result from the multiplier operation, swapping
which register is the current MRF or MRB. This swapping facilitates context
switching. Unlike other registers that have alternates, both MRF and MRB are
accessible at the same time. All fixed-point multiplies can accumulate
results in either MRF or MRB, without regard to the state of the MODE1 register. With this arrangement, code can use the result registers as primary
2-30
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
and alternate accumulators, or code can use these registers as two parallel
accumulators. This feature facilitates complex math.
The MODE1 register controls the access to alternate registers. Table A-2 on
page A-3 lists all the bits in MODE1. The following bits in MODE1 control
alternate registers (a 1 enables the alternate set):
• Secondary registers for computation unit results. Bit 2 (SRCU)
• Secondary registers for hi register file, R8-R15 and S8-15. Bit 7
(SRRFH)
• Secondary registers for lo register file, R0-R7 and S0-S7. Bit 10
(SRRFL)
The following example demonstrates how code should handle the one
cycle of latency from the instruction setting the bit in MODE1 to when the
alternate registers may be accessed.
BIT SET MODE1 SRRFL;/* activate alternate reg. file */
NOP;/* wait for access to alternates */
R0=7;
Multifunction Computations
Using the many parallel data paths within its computational units, the
DSP supports multiple-parallel (multifunction) computations. These
instructions complete in a single cycle, and they combine parallel operation of the multiplier and the ALU or dual ALU functions. The multiple
operations perform the same as if they were in corresponding single-function computations. Multifunction computations also handle flags in the
same way as the single-function computations, except that in the dual
add/subtract computation the ALU flags from the two operations are
ORed together.
ADSP-21160 SHARC DSP Hardware Reference
2-31
Multifunction Computations
To work with the available data paths, the computation units constrain
which data registers may hold the four input operands for multifunction
computations. These constraints limit which registers may hold the
X-input and Y-input for the ALU and multiplier.
Figure 2-9 shows a computational unit and indicates which registers may
serve as X-inputs and Y-inputs for the ALU and multiplier.
PM DATA BUS
MODE1
DM DATA BUS
NOTE THAT SHIFTER IS FADED
HERE, INDICATING THAT IT IS
NOT AVAILABLE FOR
MULTIFUNCTION INSTRUCTIONS.
REGISTER FILE
(16  40-BIT)
X
MULTIPLIER
Y
R0
R1
R2
R3
R8
R9
R10
R11
R4
R5
R6
R7
R12
R13
R14
R15
Z
Y
SHIFTER
X
Y
X
ALU
MR2F MR1F MR0F
ASTATx
STKYx
TO PROGRAM SEQUENCER
Figure 2-9. Input Registers for Multifunction Computations (ALU and
Multiplier)
For example, the X-input to the ALU can only be R8, R9, R10 or R11. Note
that the shifter is gray in Figure 2-9 to indicate that there are no shifter
multifunction operations.
2-32
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
Table 2-8, Table 2-9, Table 2-10, and Table 2-11 list the multifunction
computations. For more information on assembly language syntax, see
ADSP-21160 SHARC DSP Instruction Set Reference. In these tables, note
the meaning of the following symbols:
• Rm, Ra, Rs, Rx, Ry indicate any register file location; fixed-point
• Fm, Fa, Fs, Fx, Fy indicate any register file location; floating-point
• R3-0 indicates data file registers R3, R2, R1, or R0, and F3-0 indicates data file registers F3, F2, F1, or F0
• R7-4 indicates data file registers R7, R6, R5, or R4, and F7-4 indicates data file registers F7, F6, F5, or F4
• R11-8 indicates data file registers R11, R10, R9, or R8, and F11-8
indicates data file registers F11, F10, F9, or F8
• R15-12 indicates data file registers R15, R14, R13, or R12, and
F15-12 indicates data file registers F15, F14, F13, or F12
• SSFR indicates the X-input is signed, Y-input is signed, use Fractional inputs, and Rounded-to-nearest output
• SSF indicates the X-input is signed, Y-input is signed, use Fractional input
Table 2-8. Dual Add And Subtract
Ra = Rx + Ry, Rs = Rx – Ry
Fa = Fx + Fy, Fs = Fx – Fy
ADSP-21160 SHARC DSP Hardware Reference
2-33
Multifunction Computations
Table 2-9. Fixed-Point Multiply and Add, Subtract, or Average
(Any combination of left and right column)
Rm=R3-0 * R7-4 (SSFR),
Ra=R11-8 + R15-12
MRF=MRF + R3-0 * R7-4 (SSF),
Ra=R11-8 – R15-12
Rm=MRF + R3-0 * R7-4 (SSFR),
Ra=(R11-8 + R15-12)/2
MRF=MRF – R3-0 * R7-4 (SSF),
Rm=MRF – R3-0 * R7-4 (SSFR),
Table 2-10. Floating-Point Multiply And ALU Operation
Fm=F3-0 * F7-4, Fa=F11-8 + F15-12
Fm=F3-0 * F7-4, Fa=F11-8 – F15-12
Fm=F3-0 * F7-4, Fa=FLOAT R11-8 by R15-12
Fm=F3-0 * F7-4, Ra=FIX F11-8 by R15-12
Fm=F3-0 * F7-4, Fa=(F11-8 + F15-12)/2
Fm=F3-0 * F7-4, Fa=ABS F11-8
Fm=F3-0 * F7-4, Fa=MAX (F11-8, F15-12)
Fm=F3-0 * F7-4, Fa=MIN (F11-8, F15-12)
Table 2-11. Multiply With Dual Add and Subtract
Rm = R3-0 * R7-4 (SSFR), Ra = R11-8 + R15-12, Rs = R11-8 – R15-12
Fm = F3-0 * F7-4, Fa = F11-8 + F15-12, Fs = F11-8 – F15-12
Another type of multifunction operation is also available on the DSP,
combining transfers between the results and data registers and transfers
between memory and data registers. Like other multifunction instructions,
these parallel operations complete in a single cycle. For example, the DSP
can perform the following multiply and parallel read of data memory:
MRF=MRF-R5*R0, R6=DM(I1,M2);
2-34
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
Or, the DSP can perform the following result register transfer and parallel
read:
R5=MR1F, R6=DM(I1,M2);
Secondary Processing Element (PEy)
The ADSP-21160 DSP contains two sets of computation units and associated register files. As shown in Figure 2-10 on page 2-34, these two
Processing Elements (PEx and PEy) support Single Instruction, Multiple
Data (SIMD) operation.
DIFFERENT DATA GOES TO EACH ELEMENT
BUS
CO NNECT
(PX)
MULT
PM DATA BUS
16/32/40/64
DM DATA BUS
16/32/40/64
DATA
REGISTER
FILE
(PEx)
16 x 40-BIT
BARREL
SHIFTER
BARREL
SHIFTER
ALU
DATA
REGISTER
FILE
(PEy)
16 x 40-BIT
MULT
ALU
SAME INSTRUCTION GOES TO BOTH ELEMENTS
PROGRAM
SEQUENCER
Figure 2-10. Block Diagram Showing Secondary Execution
The MODE1 register controls the operating mode of the processing elements. Table A-2 on page A-3 lists all the bits in MODE1.
ADSP-21160 SHARC DSP Hardware Reference
2-35
Secondary Processing Element (PEy)
The PEYEN bit (bit 21) in the MODE1 register enables or disables the PEy
processing element. When PEYEN is cleared (0), the ADSP-21160 DSP
operates in Single-Instruction-Single-Data (SISD) mode, using only PEx;
this is the mode in which ADSP-2106x DSPs operate. When the PEYEN bit
is set (1), the ADSP-21160 DSP operates in SIMD mode, using the PEx
and PEy processing elements. There is a one cycle delay after PEYEN is set
or cleared, before the change to or from SIMD mode takes effect.
To support SIMD, the DSP performs the following parallel operations:
• Dispatches a single instruction to both processing element’s computation units
• Loads two sets of data from memory, one for each processing
element
• Executes the same instruction simultaneously in both processing
elements
• Stores data results from the dual executions to memory
the information here and in the ADSP-21160 SHARC DSP
 Using
Instruction Set Reference, it is possible through SIMD mode’s parallelism to double performance over similar algorithms running in
SISD (ADSP-2106x DSP compatible) mode.
The two processing elements are symmetrical, each containing the following functional blocks:
• ALU
• Multiplier primary and alternate result registers
• Shifter
• Data register file and alternate register file
2-36
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
Dual Compute Units Sets
The computation units (ALU, Multiplier, and Shifter) in PEx and PEy are
identical. The data bus connections for the dual computation units permit
asymmetric data moves to, from, and between the two processing elements. Identical instructions execute on the PEx and PEy computational
units; the difference is the data. The data registers for PEy operations are
identified (implicitly) from the PEx registers in the instruction.
This implicit relation between PEx and PEy data registers corresponds to
complementary register pairs in Table 2-12. Any universal registers that
do not appear in Table 2-12 have the same identities in both PEx and
PEy. When a computation in SIMD mode refers to a register in the PEx
column, the corresponding computation in PEy refers to the complimentary register in the PEy column.
Table 2-12. SIMD Mode Complementary Register Pairs
PEx
PEy
PEx
PEy
R0
S0
R11
S11
R1
S1
R12
S12
R2
S2
R13
S13
R3
S3
R14
S14
R4
S4
R15
S15
R5
S5
USTAT1
USTAT2
R6
S6
USTAT3
USTAT4
R7
S7
ASTATx
ASTATy
R8
S8
STKYx
STKYy
R9
S9
PX1
PX2
R10
S10
ADSP-21160 SHARC DSP Hardware Reference
2-37
Secondary Processing Element (PEy)
Dual Register Files
The two 16 entry data register files (one in each PE) and their operand
and result busing and porting are identical. The same is true for each 16
entry alternate register files. The transfer direction, source and destination
registers, and data bus usage depend on the following conditions:
• Computational mode:
• Is PEy enabled (PEYEN bit=1 in MODE1 register)?
• Is the data register file in PEx (R0-R15, F0-F15) or PEy
(S0-S15)?
• Is the instruction a data register swap between processing
elements?
• Data addressing mode:
• What is the state of the Internal Memory Data Width (IMDW) bits
in the System Configuration (SYSCON) register?
•
Is Broadcast write enabled (BDCST1,9 bits in MODE1 register)?
• What is the type of address (long, normal, or short word)?
• Is long-word override (LW) specified in the instruction?
• What are the states of instruction fields for DAG1 or DAG2?
• Program sequencing (conditional logic):
• What is the outcome of the instruction’s condition comparison on
each processing element?
2-38
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
For information on SIMD issues that relate to computational modes, see
“SIMD (Computational) Operations” on page 2-39. For information on
SIMD issues relating to data addressing, see “SIMD Mode and Sequencing” on page 3-55. For information on SIMD issues relating to program
sequencing, see “Addressing in SISD and SIMD Modes” on page 4-18.
Dual Alternate Registers
Both register files consist of a primary set of 16 by 40-bit registers and an
alternate set of 16 by 40-bit registers. Context switching between the two
sets of registers occur in parallel between the two processing elements. For
more information, see “Alternate (Secondary) Data Registers” on
page 2-30.
SIMD (Computational) Operations
In SIMD mode, the dual processing elements execute the same instruction, but operate on different data. To support SIMD operation, the
elements support a variety of dual data move features.
The DSP supports unidirectional and bidirectional register-to-register
transfers with the conditional compute and move instruction. All four
combinations of inter-register file and intra-register file transfers
(PEx <-> PEx, PEx <-> PEy, PEy <-> PEx, and PEy <-> PEy) are possible
in both SISD (unidirectional) and SIMD (bidirectional) modes.
In SISD mode (PEYEN bit=0), the register-to-register transfers are unidirectional, meaning that an operation performed on one processing element is
not duplicated on the other processing element. The SISD transfer uses a
source register and a destination register, and either register can be in
either element’s data register file. For a summary of unidirectional transfers, see the upper half of Table 2-12 on page 2-37. Note that in SISD
mode a condition for an instruction only tests in the PEx element and
applies to the entire instruction.
ADSP-21160 SHARC DSP Hardware Reference
2-39
Secondary Processing Element (PEy)
In SIMD mode (PEYEN bit=1), the register-to-register transfers are bidirectional, meaning that an operation performed on one element is duplicated
in parallel on the other element. The instruction uses two source registers
(one from each element’s register file) and two destination registers (one
from each element’s register file).
For a summary of bidirectional transfers, see the lower half of Table 2-12
on page 2-37. Note that in SIMD mode a conditional for an instruction
test in both the PEx and PEy elements, dividing control of the explicit and
implicit transfers as detailed in Table 2-12 on page 2-37.
Bidirectional register-to-register transfers in SIMD mode are allowed
between a data register and DAG, control, or status registers. When the
DAG, control, or status register is a source of the transfer, the destination
can be a data register. This SIMD transfer duplicates the contents of the
source register in a data register in both processing elements.
programming is required when a DAG, control, or status
 Careful
register is a destination of a transfer from a data register. If the destination register has a complement (for example ASTATx and
ASTATy), the SIMD transfer moves the contents of the explicit data
register into the explicit destination and moves the contents of the
implicit data register into the implicit destination (the complement). If the destination register has no complement (for example,
I0), only the explicit transfer occurs.
Even if the code uses a conditional operation to select whether the
transfer occurs, only the explicit transfer can take place if the destination register has no complement.
In the case where a DAG, control, or status register is both source and destination, the data move operation executes the same as if SIMD mode
were disabled.
2-40
ADSP-21160 SHARC DSP Hardware Reference
Processing Elements
In both SISD and SIMD modes, the DSP supports bidirectional register-to-register swaps. The swap always occurs between one register in each
processing element’s data register file.
Registers swaps use the special swap operator, <->. A register-to-register
swap occurs when registers in different processing elements exchange values; for example R0 <-> S1. Only single, 40-bit register to register swaps
are supported—no double register operations.
When they are unconditional, register-to-register swaps operate the same
in SISD mode and SIMD mode. If a condition is added to the instruction
in SISD mode, the condition tests only in the PEx element and controls
the entire operation. If a condition is added in SIMD mode, the condition
tests in both the PEx and PEy elements and controls the halves of the
operation as detailed in Table 2-12 on page 2-37.
Table 2-13. Register-to-Register Move Summary (SISD versus SIMD)
Mode
SISD
1
SIMD2
Instruction
Explicit Transfer
Implicit Transfer
IF condition compute, Rx = Ry;
Rx loaded from Ry
None
IF condition compute, Rx = Sy;
Rx loaded from Sy
None
IF condition compute, Sx = Ry;
Sx loaded from Ry
None
IF condition compute, Sx = Sy;
Sx loaded from Sy
None
IF condition compute, Rx <-> Sy;
Rx swaps to Sy
Sy swaps to Rx
None
IF condition compute, Rx = Ry;
Rx loaded from Ry
Sx loaded from Sy
IF condition compute, Rx = Sy;
Rx loaded from Sy
Sx loaded from Ry
IF condition compute, Sx = Ry;
Sx loaded from Ry
Rx loaded from Sy
IF condition compute, Sx = Sy;
Sx loaded from Sy
Rx loaded from Ry
IF condition compute, Rx <-> Sy; 3
Sy swaps to Rx
Rx swaps to Sy
1
In SISD mode, the conditional applies only to the entire operation and is only tested against
PEx’s flags. When the condition tests true, the entire operation occurs.
2 In SIMD mode, the conditional applies separately to the explicit and implicit transfers. Where
the condition tests true (PEx for the explicit and PEy for the implicit), the operation occurs in
that processing element.
ADSP-21160 SHARC DSP Hardware Reference
2-41
Secondary Processing Element (PEy)
3 Register to register transfers (R0=S0) and register swaps (R0<->S0) do not cause a PMD bus conflict. These operations use only the DMD bus and a hidden 16-bit bus to do the two register
moves.
conditional instructions with the same destination registers
 SIMD
do not produce predictable transfers. For example, the instruction
IF EQ R4 = R14 – R15, S4 = R6; may not work as expected. This
kind of usage is prohibited, as it is not logical to use it this way.
SIMD and Status Flags
When the DSP is in SIMD mode (PEYEN bit=1), computations on both
processing elements generate status flags, producing a logical OR’ing of
the exception status test on each processing element. If one of the four
fixed-point or floating-point exceptions is enabled, an exception condition
on either or both processing elements generates an exception interrupt.
Interrupt service routines must determine which of the processing elements encountered the exception. Note that returning from a floating
point interrupt does not automatically clear the STKY state. Code must
clear the STKY bits in both processing element’s sticky status (STKYx and
STKYy) registers as part of the exception service routine. For more information, see For more information, see “Interrupts and Sequencing” on
page 3-31.
2-42
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
3 PROGRAM SEQUENCER
The DSP’s program sequencer implements program flow, constantly providing the address of the next instruction to be executed by other parts of
the DSP.
Overview
Program flow in the DSP is mostly linear with the processor executing
program instructions sequentially. This linear flow varies occasionally
when the program uses non-sequential program structures, such as those
illustrated in Figure 3-1. Non-sequential structures direct the DSP to execute an instruction that is not at the next sequential address, following the
current instruction. These structures include:
Loops. One sequence of instructions executes several times with zero
overhead.
• Subroutines. The processor temporarily interrupts sequential flow
to execute instructions from another part of program memory.
• Jumps. Program flow transfers permanently to another part of program memory.
ADSP-21160 SHARC DSP Hardware Reference
3-1
Overview
• Interrupts. Subroutines in which a runtime event (not an instruction) triggers the execution of the routine.
• Idle. An instruction that causes the processor to cease operations,
holding its current state until an interrupt occurs. Then, the processor services the interrupt and continues normal execution.
L INE AR F L OW
ADDRESS:
L OOP
JUMP
DO UNTIL
JUMP
N+1 INSTRUCTION
INSTRUCTION
INSTRUCTION
N+2 INSTRUCTION
INSTRUCTION
INSTRUCTION
N+3 INSTRUCTION
INSTRUCTION N TIMES
INSTRUCTION
N+4 INSTRUCTION
INSTRUCTION
INSTRUCTION
N+5 INSTRUCTION
INSTRUCTION
INSTRUCTION
INT E R R UP T
IDL E
N INSTRUCTION
S UB R OUT INE
IRQ
CALL
INSTRUCTION
INSTRUCTION
INSTRUCTION
INSTRUCTION
INSTRUCTION
IDLE
INSTRUCTION WAITING
FOR IRQ
INSTRUCTION
VECTOR
INSTRUCTION
…
…
INSTRUCTION
INSTRUCTION
INSTRUCTION
INSTRUCTION
INSTRUCTION
INSTRUCTION
INSTRUCTION
INSTRUCTION
INSTRUCTION
RTS
RTI
INSTRUCTION
INSTRUCTION
Figure 3-1. Program Flow Variations
The sequencer manages execution of these program structures by selecting
the address of the next instruction to execute. As part of its process, the
sequencer handles the following tasks.
3-2
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
• Increments the fetch address
• Maintains stacks
• Evaluates conditions
• Decrements the loop counter
• Calculates new addresses
• Maintains an instruction cache
• Handles interrupts
To accomplish these tasks, the sequencer uses the blocks shown in
Figure 3-2. The sequencer’s address multiplexer selects the value of the
next fetch address from several possible sources. The fetched address
enters the instruction pipeline, made up of the fetch address register,
decode address register, and program counter (PC). These contain the
24-bit addresses of the instructions currently being fetched, decoded, and
executed. The PC couples with the PC stack, which stores return addresses
and top-of-loop addresses. All addresses generated by the sequencer are
24-bit program memory instruction addresses.
To manage events, the sequencer’s interrupt controller handles interrupt
processing, determines whether an interrupt is masked, and generates the
appropriate interrupt vector address.
With selective caching, the instruction cache lets the DSP access data in
program memory and fetch an instruction (from the cache) in the same
cycle. The DAG2 data address generator outputs program memory data
addresses.
The sequencer evaluates conditional instructions and loop termination
conditions using information from the status registers. The loop address
stack and loop counter stack support nested loops. The status stack stores
status registers for implementing nested interrupt routines.
ADSP-21160 SHARC DSP Hardware Reference
3-3
Overview
MODE1
MODE2
ST KYX
ASTATX
AST ATY
STKYY
USTAT1
USTAT2
USTAT4
INSTRUCTION
CACHE
TPERIOD
INPUT
FLAGS
INSTRUCTION
LATCH
LOOP ADDRESS STACK
(LADDR)
MULTIPLEXER
CONDITION
LOGIC
TCOUNT
USTAT3
LOOP COUNT STACK
(CURLCNTR, LCNTR)
DECREMENT
LOOP CO NTROL
BRANCH
CO NTROL
YES
ADDRESS
FROM DAG2
TCOUNT=0
NO
TIMEXP
OTHER
INTERRUPTS
I NST RUCTION PIPELI NE
INTERRUPT
CONTROLLER
PROGRAM
COUNTER STACK
INTERRUPT LAT CH
(IRPTL)
INTERRUPT MASK
(IMASK)
TOP OF PC
STACK (PCSTK)
INTERRUPT MASK
POINTER (IMASKP)
PC STACK
POI NTER (PCSTKP)
INTERRUPT
VECTO R
32
32
RETURN ADDRESS
OR TO P OF LO OP
FETCH
DECODE PROGRAM
ADDRESS ADDRESS COUNTER
(FADDR) (DADDR)
(PC)
PC-RELATIVE
ADDRESS
+1
NEXT
ADDRESS
(LINEAR
FLOW)
REPEATED
ADDRESS
(IDLE)
NEXT ADDRESS MULTIPLEXER
+
DIRECT
BRANCH
INDIRECT
BRANCH
48
24
DM DATA BUS
(64 BITS)
PM ADDRESS BUS
(32 BITS)
PM DATA BUS
(64-BITS)
Figure 3-2. Program Sequencer Block Diagram
3-4
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Table 3-1 and Table 3-2 list the registers within and related to the program sequencer. All registers in the program sequencer are universal
registers, so they are accessible to other universal registers and to data
memory. All the sequencer’s registers and the tops of stacks are readable,
and all these registers are writable, except for the fetch address, decode
address, and PC. Pushing or popping the PC stack is done with a write to
the PC stack pointer, which is readable and writable. Pushing or popping
the loop address stack requires explicit instructions.
A set of system control registers configures or provides input to the
sequencer. These registers appear across the top and within the interrupt
controller of Figure 3-2. A bit manipulation instruction permits setting,
clearing, toggling, or testing specific bits in the system registers. For information on this instruction (Bit), see ADSP-21160 SHARC DSP Instruction
Set Reference.
Writes to some of these registers do not take effect on the next cycle. For
example, after a write to the MODE1 register to enable ALU saturation
mode, the change does not take effect until two cycles after the write.
Also, some of these registers do not update on the cycle immediately following a write. It takes an extra cycle before a read of the register returns
the new value.
With the lists of sequencer and system registers, Table 3-1 and Table 3-2
summarize the number of extra cycles (latency) for a write to take effect
(effect latency) and for a new value to appear in the register (read latency).
A “0” indicates that the write takes effect or appears in the register on the
next cycle after the write instruction is executed, and a “1” indicates one
extra cycle.
ADSP-21160 SHARC DSP Hardware Reference
3-5
Overview
Table 3-1. Program Sequencer Registers Read and Effect Latencies
Register
Contents
Bits
Read Latency
Effect Latency
FADDR
fetch address
24
—
—
DADDR
decode address
24
—
—
PC
execute address
24
—
—
PCSTK
top of PC stack
24
0
0
PCSTKP
PC stack pointer
5
1
1
LADDER
top of loop address stack
32
0
0
CURLCNTR
top of loop count stack (current loop
count)
32
0
0
LCNTR
loop count for next DO UNTIL loop 32
0
0
Table 3-2. System Registers Read and Effect Latencies1
3-6
Register
Contents
Bits
Read Latency
Maximum
Effect Latency
MODE1
mode control bits
32
0
1
MODE2
mode control bits
32
0
1
IRPTL
interrupt latch
32
0
1
IMASK
interrupt mask
32
0
1
IMASKP
interrupt mask pointer (for nesting)
32
1
1
MMASK
mode mask
32
0
1
FLAGS
flag inputs
32
0
1
LIRPTL
link port interrupt latch/mask
32
0
1
ASTATX
arithmetic status flags
32
0
1
ASTATY
arithmetic status flags
32
0
1
STKYX
sticky status flags
32
0
1
STKYY
sticky status flags
32
0
1
USTAT1
user-defined status flags
32
0
0
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Table 3-2. System Registers Read and Effect Latencies1 (Cont’d)
Register
Contents
Bits
Read Latency
Maximum
Effect Latency
USTAT2
user-defined status
32
0
0
USTAT3
user-defined status
32
0
0
USTAT4
user-defined status
32
0
0
1
The number of cycles it takes for the effect latencies for different registers (for example,
MODE1) given above is just a maximum value. Different bits in these registers have different
effect latencies ranging from 0 to the maximum value listed. Hence users can write code
– That does not have any dependency on the above effect latencies
– Such that there are delays up to the number of cycles specified in the above Maximum Effect
Latency column
The following sections in this chapter explain how to use each of the functional blocks in Figure 3-2 on page 3-4.
• “Instruction Pipeline” on page 3-8
• “Instruction Cache” on page 3-9
• “Branches and Sequencing” on page 3-14
• “Loops and Sequencing” on page 3-20
• “Interrupts and Sequencing” on page 3-31
• “Timer and Sequencing” on page 3-48
• “Stacks and Sequencing” on page 3-50
• “Conditional Sequencing” on page 3-52
• “SIMD Mode and Sequencing” on page 3-55
ADSP-21160 SHARC DSP Hardware Reference
3-7
Instruction Pipeline
Instruction Pipeline
The program sequencer determines the next instruction address by examining both the current instruction being executed and the current state of
the processor. If no conditions require otherwise, the DSP executes
instructions from program memory in sequential order by incrementing
the fetch address. Using its instruction pipeline, the DSP processes
instructions in three clock cycles:
• Fetch cycle. The DSP reads the instruction from either the on-chip
instruction cache or from program memory.
• Decode cycle. The DSP decodes the instruction, generating conditions that control instruction execution.
• Execute cycle. The DSP executes the instruction; the operations
specified by the instruction complete in a single cycle.
These cycles overlap in the pipeline, as shown in Table 3-3. In sequential
program flow, when one instruction is being fetched, the instruction
fetched in the previous cycle is being decoded, and the instruction fetched
two cycles before is being executed. Sequential program flow always has a
throughput of one instruction per cycle.
Table 3-3. Pipelined Execution Cycles
3-8
Cycles
Fetch
Decode
Execute
1
0x08 
2
0x09 
0x08 
3
0x0A 
0x09 
0x08
4
0x0B 
0x0A 
0x09
5
0x0C
0x0B
0x0A
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Any non-sequential program flow can potentially decrease the DSP’s
instruction throughput. Non-sequential program operations include:
• Program memory data accesses that conflict with instruction
fetches
• Jumps
• Subroutine calls and returns
• Interrupts and return
• Loops
Instruction Cache
Usually, the sequencer fetches an instruction from memory on each cycle.
Occasionally, bus constraints prevent some of the data and instructions
from being fetched in a single cycle. To alleviate these data flow constraints, the DSP has an instruction cache, which appears in Figure 3-2 on
page 3-4. When the DSP executes an instruction that requires data access
over the PM data bus, there is a bus conflict because the sequencer uses
the PM data bus for fetching instructions. To avoid these conflicts, the
DSP caches these instructions, reducing delays. Except for enabling or disabling the cache, its operation requires no user intervention. For more
information, see “Using the Cache” on page 3-12.
The first time the DSP encounters a fetch conflict, the DSP must wait to
fetch the instruction on the following cycle, causing a delay. The DSP
automatically writes the fetched instruction to the cache to prevent the
same delay from happening again. The sequencer checks the instruction
cache on every program memory data access. If the instruction needed is
in the cache, the instruction fetch from the cache happens in parallel with
the program memory data access, without incurring a delay.
ADSP-21160 SHARC DSP Hardware Reference
3-9
Instruction Cache
Because of the three-stage instruction pipeline, as the DSP executes
instruction (at address n) that requires a program memory data access, this
execution creates a conflict with the instruction fetch (at address n+2),
assuming sequential execution. The cache stores the fetched instruction
(n+2), not the instruction requiring the program memory data access.
If the instruction needed to avoid a conflict is in the cache, the cache provides the instruction while the program memory data access is performed.
If the needed instruction is not in the cache, the instruction fetch from
memory takes place in the cycle following the program memory data
access, incurring one cycle of overhead. The fetched instruction is loaded
into the cache, if the cache is enabled and not frozen, so that it is available
the next time the same conflict occurs.
Figure 3-3 shows a block diagram of the instruction cache. The cache
holds 32 instruction-address pairs. These pairs (or cache entries) are
arranged into 16 (15–0) cache sets according to their address’ 4 least significant bits (3–0). The two entries in each set (entry 0 and entry 1) have a
valid bit, indicating whether the entry contains a valid instruction. The
least recently used (LRU) bit for each set indicates which entry was not used
last (0=entry 0 and 1=entry 1).
3-10
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
LRU VALID
BIT
BIT
SET 0
INSTRUCTIONS
ADDRESSES
BITS (23-4)
ADDRESSES
BITS (3-0)
ENTRY 0
0000
ENTRY 1
0000
ENTRY 0
0001
ENTRY 1
0001
ENTRY 0
0010
ENTRY 1
0010
SET 13 ENTRY 0
1101
ENTRY 1
1101
SET 14 ENTRY 0
1110
SET 1
SET 2
ENTRY 1
1110
SET 15 ENTRY 0
1111
ENTRY 1
1111
Figure 3-3. Instruction Cache Architecture
The cache places instructions in entries according to the 4 LSBs of the
instruction’s address. When the sequencer checks for an instruction to
fetch from the cache, it uses the 4 address LSBs as an index to a cache set.
Within that set, the sequencer checks the addresses of the two entries,
looking for the needed instruction. If the cache contains the instruction,
the sequencer uses the entry and updates the LRU bit (if necessary) to indicate the entry did not contain the needed instruction.
When the cache does not contain a needed instruction, the cache loads a
new instruction and its address, placing these in the least recently used
entry of the appropriate cache set and toggling the LRU bit (if necessary).
ADSP-21160 SHARC DSP Hardware Reference
3-11
Instruction Cache
Using the Cache
After a DSP reset, the cache starts cleared (containing no instructions),
unfrozen, and enabled. From then on, the MODE2 register controls the
operating mode of the instruction cache. Table A-3 on page A-7 lists all
the bits in MODE2. The following bits in MODE2 control cache modes:
• Cache Disable. Bit 4 (CADIS) directs the sequencer to disable the
cache (if 1) or enable the cache (if 0). Disabling the cache does not
mark the current content of the cache as invalid. If the cache is to
be enabled again, the existing content is used again. To clear the
cache, use the FLUSH CACHE instruction.
• Cache Freeze. Bit 19 (CAFRZ) directs the sequencer to freeze the
contents of the cache (if 1) or let new entries displace the entries in
the cache (if 0).
self-modifying code (e.g. software loader kernel) or software
 Ifoverlays
are used, execute a
instruction followed by a
FLUSH CACHE
before executing the new code. Otherwise old content from the
cache could still be used, although the code has changed.
NOP
When changing the cache’s mode, note that an instruction containing a
program memory data access must not be placed directly after a cache
enable or cache disable instruction, because the DSP must wait at least one
cycle before executing the PM data access. A program should have a NOP
inserted after the cache enable instruction if necessary.
Optimizing Cache Usage
Usually, cache operation is efficient and requires no intervention, but certain ordering of instructions can work against the cache’s architecture and
can degrade cache efficiency. When the order of PM data accesses and
instruction fetches continuously displaces cache entries and loads new
entries, the cache is not being efficient. Rearranging the order of these
instructions remedies this inefficiency.
3-12
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
An example of code that works against cache efficiency appears in
Table 3-4.
Table 3-4. Cache-Inefficient Code
Address
Instruction
0x0100
lcntr=1024, do Outer until lce;
0x0101
r0=dm(i0,m0), pm(i8,m8)=f3;
0x0102
r1=r0-r15;
0x0103
if eq call (Inner);
0x0104
f2=float r1;
0x0105
f3=f2*f2;
0x0106
Outer: f3=f3+f4;
0x0107
pm(i8,m8)=f3;
...
0x0200
Inner: r1=R13;
0x0201
r14=pm(i9,m9);
...
0x0211
pm(i9,m9)=r12;
...
0x021F
rts;
The program memory data access at address 0x101 in the loop, Outer,
causes the cache to load the instruction at 0x103 (into set 3). Each time
the program calls the subroutine, Inner, the program memory data
accesses at 0x201 and 0x211 displace the instruction at 0x103 by loading
the instructions at 0x203 and 0x213 (also into set 3). If the program only
calls the Inner subroutine rarely during the Outer loop execution, the
repeated cache loads do not greatly influence performance. If the program
frequently calls the subroutine while in the loop, the cache inefficiency has
a noticeable effect on performance.
ADSP-21160 SHARC DSP Hardware Reference
3-13
Branches and Sequencing
To improve cache efficiency on this code (for instance, if the execution
time of the Outer loop is critical), you should rearrange the order of some
instructions. Moving the subroutine call up one location (starting at
0x201) would work here, because with that order the two cached instructions end up in cache set 4 instead of set 3.
Branches and Sequencing
One of the types of non-sequential program flow that the sequencer supports is branching. A branch occurs when a Jump or Call/return
instruction begins execution at a new location, other than the next
sequential address. For descriptions on how to use the Jump and
Call/return instructions, see ADSP-21160 SHARC DSP Instruction Set
Reference.
Briefly, these instructions operate as follows.
• A Jump or a Call instruction transfers program flow to another
memory location. The difference between a Jump and a Call is that
a Call automatically pushes the return address (the next sequential
address after the Call instruction) onto the PC stack. This push
makes the address available for the Call instruction’s matching
return instruction to allow easy return from the subroutine.
• A return instruction causes the sequencer to fetch the instruction at
the return address, which is stored at the top of the PC stack. The
two types of return instructions are return from subroutine (RTS)
and return from interrupt (RTI). While the return from subroutine
(RTS) only pops the return address off the PC stack, the return
from interrupt (RTI) pops the return address and:
3-14
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
• Pops the status stack if the ASTATx,y and MODE1 status registers have been pushed for any of the following interrupts:
IRQ2-0, timer, or VIRPT.
• Clears the interrupt’s bit in the interrupt latch register
(IRPTL) and the interrupt mask pointer (IMASKP).
You can specify a number of parameters for branches:
• Jump and Call/return instructions can be conditional. The program sequencer can evaluate status conditions to decide whether to
execute a branch. If no condition is specified, the branch is always
taken. For more information on these conditions, see “Conditional
Sequencing” on page 3-52.
• Jump and Call/return instructions can be immediate or delayed.
Because of the instructions pipeline, an immediate branch incurs
two lost (overhead) cycles. A delayed branch has no overhead. For
more information, see “Delayed Branches” on page 3-16.
• Jump instructions that appear within a loop or within an interrupt
service routine have additional options. For information on the
loop abort (LA) option, see “Loops and Sequencing” on page 3-20.
For information on the loop re-entry (LR) option, see “Restrictions
On Ending Loops” on page 3-22. For information on the clear
interrupt (CI) option, see “Interrupts and Sequencing” on
page 3-31.
The sequencer block diagram in Figure 3-2 on page 3-4 shows that
branches can be direct or indirect. The difference is that the sequencer
generates the address for a direct branch, and the PM data address generator (DAG2) produces the address for an indirect branch.
Direct branches are Jump or Call/return instructions that use an absolute—not changing at runtime—address (such as a program label) or use a
PC-relative address. Some instruction examples that cause a direct branch
are:
ADSP-21160 SHARC DSP Hardware Reference
3-15
Branches and Sequencing
jump fft1024; {where fft1024 is an address label}
call (pc,10); {where (pc,10) a PC-relative address}
Indirect branches are Jump or Call/Return instructions that use a
dynamic—changes at runtime—address that comes from the PM data
address generator. For more information on the data address generator,
see “Data Address Generators”. Some instruction examples that cause an
indirect branch are:
jump (m8,i12); {where (m8,i12) are DAG2 registers}
call (m9,i13); {where (m9,i13) are DAG2 registers}
Conditional Branches
The sequencer supports conditional branches. These are Jump or
Call/return instructions whose execution is based on testing an If condition. For more information on condition types in If condition
instructions, see “Conditional Sequencing” on page 3-52. Note that the
DSP’s Single-Instruction, Multiple-Data mode influences the execution of
conditional branches. For more information, see “SIMD Mode and
Sequencing” on page 3-55.
Delayed Branches
The instruction pipeline influences how the sequencer handles branches.
For immediate branches—Jumps and Call/return instructions not specified as delayed branches (DB), two instruction cycles are lost (NOPs) as
the pipeline empties and refills with instructions from the new branch.
As shown in Table 3-5 on page 3-17 and Table 3-6 on page 3-17, the
DSP does not execute the two instructions after the branch, which are in
the fetch and decode stages. For a Call, the decode address (the address of
the instruction after the Call) is the return address. During the two lost
(no-operation) cycles, the pipeline fetches and decodes the first instruction at the branch address.
3-16
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Table 3-5. Pipelined Execution Cycles For Immediate Branch (Jump/Call)
Cycles
Fetch
Decode
Execute
1
n+2
n+1nop
2
j2
n+2nop3
NOP
3
j+1
j
NOP
4
j+2
j+1
j
1
n
Note that n is the branching instruction, and j is the instruction branch address
1. n+1 suppressed
2. For call, n+1 pushed on PC stack
3. n+2 suppressed
Table 3-6. Pipelined Execution Cycles For Immediate Branch (Return)
Cycles
Fetch
Decode
Execute
1
n+2
n+1nop1
n2
2
r
n+2nop3
NOP
3
r+1
r
NOP
4
r+2
r+1
r
Note that n is the branching instruction, and r is the instruction branch address
1. n+1 suppressed
2. r (n+1 in Table 3-5) popped from PC stack
3. n+2 suppressed
For delayed branches—Jumps and Call/return instructions with the
delayed branches (DB) modifier, no instruction cycles are lost in the pipeline, because the DSP executes the two instructions after the branch while
the pipeline fills with instructions from the new branch.
ADSP-21160 SHARC DSP Hardware Reference
3-17
Branches and Sequencing
As shown in Table 3-7 and Table 3-8, the DSP executes the two instructions after the branch, while the instruction at the branch address is
fetched and decoded. In the case of a Call, the return address is the third
address after the branch instruction. While delayed branches use the
instruction pipeline more efficiently than immediate branches, it is
important to note that delayed branch code can be harder to understand
because of the instructions between the branch instruction and the actual
branch.
Table 3-7. Pipelined Execution Cycles For Delayed Branch (Jump or Call)
Cycles
Fetch
Decode
Execute
1
n+2
n+1
n
2
j1
n+2
n+1
3
j+1
j
n+2
4
j+2
j+1
j
Note that n is the branching instruction, and j is the instruction branch address
1. For call, n+3 pushed on PC stack
Table 3-8. Pipelined Execution Cycles For Delayed Branch (return)
Cycles
Fetch
Decode
Execute
1
n+2
n+1
n1
2
r
n+2
n+1
3
r+1
r
n+2
4
r+2
r+1
r
Note that n is the branching instruction, and r is the instruction branch address
1. r (n+3 in Table 3-7) popped from PC stack
Besides being somewhat more challenging to code, there are also some
limitations on delayed branches that stem from the instruction pipeline
architecture. Because the delayed branch instruction and the two instructions that follow it must execute sequentially, the instructions in the two
3-18
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
locations that follow a delayed branch instruction may not be any of the
following:
• Other branches (no Jump, Call, or return instructions)
• Any manipulations of the PC stack, status stack or loop stacks
• Any writes to the PC stack pointer
• Any loops or other breaks in sequential operation (no Do/Until or
Idle instructions)
software for the DSP should always flag these types
 Development
of instructions in the two locations after a delayed branch instruction as code errors.
to follow a delayed branch instruction with a Jump,
 ItCall,is possible
or return instruction in one special case. If the sequential
branch instructions use mutually exclusive conditions, one branch
may follow another. The following example is valid.
if gt jump (PC, 7) (db); // if greater than...
if led jump (PC, 11) (db); // if less than or equal...
Interrupt processing is also influenced by delayed branches and the
instruction pipeline. Because the delayed branch instruction and the two
instructions that follow it must execute sequentially, the DSP does not
immediately process an interrupt that occurs in between a delayed branch
instruction and either of the two instructions that follow. Any interrupt
that occurs during these instructions is latched, but not processed until
the branch is complete.
a delayed branch, a program can read the PC stack or PC
 During
stack pointer immediately after a delayed call or return, but this
read shows that the return address on the PC stack has already been
pushed or popped, even though the branch has not occurred yet.
ADSP-21160 SHARC DSP Hardware Reference
3-19
Loops and Sequencing
Loops and Sequencing
Another type of non-sequential program flow that the sequencer supports
is looping. A loop occurs when a Do/Until instruction causes the DSP to
repeat a sequence of instructions until a condition tests true.
A special condition for terminating a loop is Loop Counter Expired
(LCE). This condition tests whether the loop has completed the number
of iterations in the LCNTR register. Loops that terminate with conditions
other than LCE have some additional restrictions. For more information,
see “Restrictions On Ending Loops” on page 3-22 and “Restrictions On
Short Loops” on page 3-23. For more information on condition types in
Do/Until instructions, see “Conditional Sequencing” on page 3-52.
DSP’s Single-Instruction, Multiple-Data mode influences the
 The
execution of loops. For more information, “SIMD Mode and
Sequencing” on page 3-55.
The Do/Until instruction uses the sequencer’s loop and condition features, which appear in Figure 3-2 on page 3-4. These features provide
efficient software loops, without the overhead of additional instructions to
branch, test a condition, or decrement a counter. The following code
example shows a Do/Until loop that contains three instructions and iterates 30 times.
LCNTR=30, DO the_end UNTIL LCE; {loop iterates 30 times}
R0=DM(I0,M0), F2=PM(I8,M8);
R1=R0-R15;
the_end: F4=F2+F3;
{last instruction in loop}
When executing a Do/Until instruction, the program sequencer pushes
the address of the loop’s last instruction and loop’s termination condition
onto the loop address stack. The sequencer also pushes the top-of-loop
address—address of the instruction following the Do/Until instruction—
onto the PC stack.
3-20
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
The sequencer’s instruction pipeline architecture influences loop termination. Because instructions are pipelined, the sequencer must test the
termination condition (and, if the loop is counter-based, decrement the
counter) before the end of the loop. Based on the test’s outcome, the next
fetch either exits the loop or returns to the top-of-loop.
The condition test occurs when the DSP is executing the instruction two
locations before the last instruction in the loop (at location e – 2, where e
is the end-of-loop address). If the condition tests false, the sequencer
repeats the loop, fetching the instruction from the top-of-loop address,
which is stored on the top of the PC stack. If the condition tests true, the
sequencer terminates the loop, fetching the next instruction after the end
of the loop and popping the loop and PC stacks.
A special case of loop termination is the loop abort instruction,
Jump (LA). This instruction causes an automatic loop abort when it
occurs inside a loop. When the loop aborts, the sequencer pops the PC
and loop address stacks once. If the aborted loop was nested, the single
pop of the stacks leaves the correct values in place for the outer loop.
Table 3-9 and Table 3-10 on page 3-22 show the pipeline states for loop
iteration and termination.
Table 3-9. Pipelined Execution Cycles For Loop Back (Iteration)
Cycles
Fetch
Decode
Execute
1
e
e–1
e – 21
2
b2
e
e–1
3
b+1
b
e
4
b+2
b+1
b
Note that e is the loop end instruction, and b is the loop start instruction
1. Termination condition tests false
2. Loop start address is top of PC stack
ADSP-21160 SHARC DSP Hardware Reference
3-21
Loops and Sequencing
Table 3-10. Pipelined Execution Cycles For Loop Termination
Cycles
Fetch
Decode
Execute
1
e
e–1
e – 21
2
e+12
e
e–1
3
e+2
e+1
e
4
e+3
e+2
e+1
Note that e is the loop end instruction
1. Termination condition tests true
2. Loop aborts and loop stacks pop
Restrictions On Ending Loops
The sequencer’s loop features (which optimize performance in many ways)
limit the types of instructions that may appear at or near the end of the
loop. Nested loops may not use the same end-of-loop instruction address.
• Nested loops with a non-counter-based loop as the outer loop must
place the end address of the outer loop at least two addresses after
the end address of the inner loop.
• Nested loops with a non-counter-based loop as the outer loop that
use the loop abort instruction, Jump (LA), to abort the inner loop
may not Jump (LA) to the last instruction of the outer loop.
• An instruction that writes to the loop counter from memory may
not be used as the third-to-last instruction of a counter-based loop
(at e – 2, where e is the end-of-loop address).
• An If Not LCE instruction may not be used as the instruction that
follows a write to CURLCNTR from memory.
3-22
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
• Branch (Jump or Call/return) instructions may not be used as any
of the last three instructions of a loop. This no end-of-loop
branches rule also applies to single-instruction and two-instruction
loops with only one iteration.
There is one exception to the no end-of-loop branches rule. The last three
instructions of a loop may contain an immediate Call—a Call without a
DB modifier—that is paired with a loop re-entry return—a return (RTS)
with loop re-entry modifier (LR). The immediate Call may be one of the
last three instructions of a loop, but not in a one-instruction loop or a
two-instruction, single-iteration loop.
Restrictions On Short Loops
The sequencer’s pipeline features (which optimize performance in many
ways) restrict how short loops iterate and terminate. Short loops (1- or
2-instruction loops) terminate in a special way because they are shorter
than the instruction pipeline. Counter-based loops (Do/Until LCE) of
one or two instructions are not long enough for the sequencer to check the
termination condition two instructions from the end of the loop. In these
short loops, the sequencer has already looped back when the termination
condition is tested. The sequencer provides special handling to avoid overhead (NOP) cycles if the loop is iterated a minimum number of times.
Table 3-11 on page 3-24 and Table 3-12 on page 3-24 show the pipeline
execution for counter-based single-instruction loops.
ADSP-21160 SHARC DSP Hardware Reference
3-23
Loops and Sequencing
Table 3-11. Pipelined Execution Cycles for Single Instruction
Counter-Based Loop With Three Iterations
Cycles
Fetch
Decode
Execute
1
n+2
n+1
n1
2
n+12
n+1
n+1 (pass 1)
3
n+23
n+1
n+1 (pass 2)
4
n+3
n+2
n+1 (pass 3)
5
n+4
n+3
n+2
Note: n is the loop start instruction, and n+2 is the instruction after the loop
1. Loop count (LCNTR) equals 3
2. No opcode latch or fetch address update; count expired tests true
3. Loop iteration aborts; PC and loop stacks pop
Table 3-12. Pipelined Execution Cycles for Single Instruction
Counter-Based Loop With Two Iterations (Two Overhead Cycles)
Cycles
Fetch
Decode
Execute
1
n+2
n+1
n1
2
n+12
n+1
n+1 (pass 1)
3
n+13
n+1nop4
n+1 (pass 2)
4
n+2
n+1nop5
NOP
5
n+3
n+2
NOP
6
n+4
n+3
n+2
Note: n is the loop start instruction, and n+2 is the instruction after the loop
1. Loop count (LCNTR) equals 2
2. No opcode latch or fetch address update
3. Count expired tests true
4. Loop iteration aborts; PC and loop stacks pop; n+1 suppressed
5. n+1 suppressed
3-24
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Table 3-13 and Table 3-14 on page 3-26 show the pipeline execution for
counter-based two-instruction loops. For no overhead, a loop of length
one must be executed at least three times and a loop of length two must be
executed at least twice. Loops of length one that iterate only once or twice
and loops of length two that iterate only once incur two cycles of overhead, because there are two aborted instructions after the last iteration to
clear the instruction pipeline.
Table 3-13. Pipelined Execution Cycles For Two Instruction
Counter-Based Loop With Two Iterations
Cycles
Fetch
Decode
Execute
1
n+2
n+1
n1
2
n+12
n+2
n+1 (pass 1)
3
n+23
n+1
n+2 (pass 1)
4
n+34
n+2
n+1 (pass 2)
5
n+4
n+3
n+2 (pass 2)
6
n+5
n+4
n+3
Note: n is the loop start instruction, and n+3 is the instruction after the loop
1. Loop count (LCNTR) equals 2
2. PC stack supplies loop start address
3. Count expired tests true
4. Loop iteration aborts; PC and loop stacks pop
ADSP-21160 SHARC DSP Hardware Reference
3-25
Loops and Sequencing
Table 3-14. Pipelined Execution Cycles For Two Instruction
Counter-Based Loop With One Iteration (Two Overhead Cycles)
Cycles
Fetch
Decode
Execute
1
n+2
n+1
n1
2
n+12
n+2
n+1 (pass 1)
3
n+23
n+1nop4
n+2 (pass 1)
4
n+3
n+2nop5
NOP
5
n+4
n+3
NOP
6
n+5
n+4
n+3
Note: n is the loop start instruction, and n+3 is the instruction after the loop
1. Loop count (LCNTR) equals 1
2. PC stack supplies loop start address
3. Count expired tests true
4. Loop iteration aborts; PC and loop stacks pop; n+1 suppressed
5. n+2 suppressed
Processing of an interrupt that occurs during the last iteration of a
one-instruction loop that executes once or twice, a two-instruction loop
that executes once, or the cycle following one of these loops (which is a
NOP) is delayed by one cycle. Similarly, in a one-instruction loop that
iterates at least three times, processing is delayed by one cycle if the interrupt occurs during the third-to-last iteration. For more information on
pipeline execution during interrupts, see “Interrupts and Sequencing” on
page 3-31.
3-26
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Short non-counter-based loops terminate differently from short
counter-based loops. These differences stem from the architecture of the
pipeline and conditional logic.
• In a three-instruction non-counter-based loop, the sequencer tests
the termination condition when the DSP executes the top of loop
instruction. When the condition tests true, the sequencer completes the iteration of the loop and terminates.
• In a two-instruction non-counter-based loop, the sequencer tests
the termination condition when the DSP executes the last (second)
instruction. If the condition becomes true when the first instruction is executed, the condition tests true during the second
instruction, and the sequencer completes one more iteration of the
loop before exiting. If the condition becomes true during the second instruction, the sequencer completes two more iterations of
the loop before exiting.
• In a one-instruction non-counter-based loop, the sequencer tests
the termination condition every cycle. After the cycle when the
condition becomes true, the sequencer completes three more iterations of the loop before exiting.
Loop Address Stack
The sequencer’s loop support, which appears in Figure 3-2 on page 3-4,
includes a loop address stack. The loop address stack is six levels deep by
32 bits wide.
The LADDR register contains the top entry on the loop address stack. This
register is readable and writable over the DM Data bus. Reading and writing LADDR does not move the loop address stack pointer; only a stack push
or pop, performed with explicit instructions, moves the stack pointer.
LADDR contains the value 0xFFFF FFFF when the loop address stack is
empty. Table A-13 on page A-30 lists all the bits in LADDR.
ADSP-21160 SHARC DSP Hardware Reference
3-27
Loops and Sequencing
The sequencer pushes an entry onto the loop address stack when executing
a Do/Until or Push Loop instruction. The stack entry pops off the stack
two instructions before the end of its loop’s last iteration or on a Pop
Loop instruction. A stack overflow occurs if a seventh entry (one more
than full) is pushed onto the loop stack. The stack is empty when no
entries are occupied.
The loop stacks’ overflow or empty status is available. Because the
sequencer keeps the loop stack and loop counter stack synchronized, the
same overflow and empty flags apply to both stacks. These flags are in the
sticky status register (STKYx). For more information on STKYx, see
Table A-5 on page A-13. For more information on how these flags work
with the loop stacks, see “Loop Counter Stack”. Note that a loop stack
overflow causes a maskable interrupt.
Because the sequencer tests the termination condition two instructions
before the end of the loop, the loop stack pops before the end of the loop’s
final iteration. If a program reads LADDR at either of these instructions, the
value is already the termination address for the next loop stack entry.
Loop Counter Stack
The sequencer’s loop support, which appears in Figure 3-2 on page 3-4,
includes a loop counter stack. The sequencer keeps the loop counter stack
synchronized with the loop address stack, with both stacks always having
the same number of locations occupied. Because these stacks are synchronized, the same empty and overflow status flags from the STKYx register
apply to both stacks.
The loop counter stack is six locations deep. The stack is full when all
entries are occupied, is empty when no entries are occupied, and is overflowed if a push occurs when the stack is already full. Bits in the STKYx
register indicate the loop counter stack full and empty states. Table A-5
on page A-13 lists the bits in the STYKx register. The STKYx bits that indicate loop counter stack status are:
3-28
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
• Loop stacks overflowed. Bit 25 (LSOV) indicates that the loop
counter stack and loop stack are overflowed (if 1) or not overflowed (if 0)—A sticky bit
• Loop stacks empty. Bit 26, (LSEM) indicates that the loop counter
stack and loop stack are empty (if 1) or not empty (if 0)—Not
sticky, cleared by a Push
Within the sequencer, the current loop counter (CURLCNTR) and loop
counter (LCNTR) registers allow access to the loop counter stack. CURLCNTR
tracks iterations for a loop being executed, and LCNTR holds the count
value before the loop is executed. The two counters let the DSP maintain
the count for an outer loop, while a program is setting up the count for an
inner loop.
The top entry in the loop counter stack (CURLCNTR) always contains the
current loop count. This register is readable and writable over the DM
Data bus. Reading CURLCNTR when the loop counter stack is empty returns
the value 0xFFFF FFFF.
The sequencer decrements the value of CURLCNTR for each loop iteration.
Because the sequencer tests the termination condition two instruction
cycles before the end of the loop, the loop counter also decrements before
the end of the loop. If a program reads CURLCNTR at either of the last two
loop instructions, the value is already the count for the next iteration.
The loop counter stack pops two instructions before the end of the last
loop iteration. When the loop counter stack pops, the new top entry of the
stack becomes the CURLCNTR value—the count in effect for the executing
loop. If there is no executing loop, the value of CURLCNTR is 0xFFFF FFFF
after the pop.
ADSP-21160 SHARC DSP Hardware Reference
3-29
Loops and Sequencing
Writing CURLCNTR does not cause a stack push. If a program writes a new
value to CURLCNTR, it changes the count value of the loop currently executing. When no Do/Until LCE loop is executing, writing to CURLCNTR has
no effect. Because the processor must use CURLCNTR to perform
counter-based loops, there are some restrictions on how a program can
write CURLCNTR. For more information, see “Restrictions On Ending
Loops” on page 3-22.
The next-to-top entry in the loop counter stack (LCNTR) is the location on
the stack that takes effect on the next loop stack push. To set up a count
value for a nested loop without changing the count for the currently executing loop, a program writes the count value to LCNTR.
 A value of zero in
LCNTR
causes a loop to execute 232 times.
A Do/Until LCE instruction pushes the value of LCNTR onto the loop
count stack, making that value the new CURLCNTR value. Figure 3-4 on
page 3-31 demonstrates this process for a set of nested loops. The previous
CURLCNTR value is preserved one location down in the stack. If a program
reads LCNTR when the loop counter stack is full, the stack returns invalid
data. When the loop counter stack is full, the stack discards any data written to LCNTR. If a program reads LCNTR during the last two instructions of a
terminating loop, the value of LCNTR is the last CURLCNTR value for the
loop.
3-30
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
LCNTR
AAAA AAAA
CURLCNTR
0XFFFF FFFF
1
LCNTR
AAAA AAAA
3
2
CURLCNTR
AAAA AAAA
LCNTR
BBBB BBBB
AAAA AAAA
CURLCNTR
BBBB BBBB
LCNTR
CCCC CCCC
4
5
6
AAAA AAAA
AAAA AAAA
AAAA AAAA
BBBB BBBB
BBBB BBBB
BBBB BBBB
CURLCNTR
CCCC CCCC
LCNTR
DDDD DDDD
CCCC CCCC
CCCC CCCC
CURLCNTR
DDDD DDDD
DDDD DDDD
LCNTR
EEEE EEEE
CURLCNTR
EEEE EEEE
LCNTR
FFFF FFFF
7
AAAA AAAA
BBBB BBBB
CCCC CCCC
DDDD DDDD
EEEE EEEE
CURLCNTR
FFFF FFFF
Figure 3-4. Pushing the Loop Counter Stack for Nested Loops
Interrupts and Sequencing
Another type of non-sequential program flow that the sequencer supports
is interrupt processing. Interrupts may stem from a variety of conditions,
ADSP-21160 SHARC DSP Hardware Reference
3-31
Interrupts and Sequencing
both internal and external to the processor. In response to an interrupt,
the sequencer processes a subroutine call to a predefined address, the
interrupt vector. The DSP assigns a unique vector to each type of
interrupt.
The DSP supports three prioritized, individually-maskable external interrupts, each of which can be either level- or edge-sensitive. External
interrupts occur when another device asserts one of the DSP’s interrupt
inputs (IRQ2-0). The DSP also supports internal interrupts. An internal
interrupt can stem from arithmetic exceptions, stack overflows, or circular
data buffer overflows. Several factors control the DSP’s response to an
interrupt. The DSP responds to an interrupt request if:
• The DSP is executing instructions or is in an Idle state
• The interrupt is not masked
• Interrupts are globally enabled
• A higher priority request is not pending
When the DSP responds to an interrupt, the sequencer branches program
execution with a Call to the corresponding interrupt vector address.
Within the DSP’s program memory, the interrupt vectors are grouped in
an area called the interrupt vector table. The interrupt vectors in this table
are spaced at 4-instruction intervals. For a list of interrupt vector addresses
and their associated latch and mask bits, see Table B-1 on page B-2. Each
interrupt vector has associated latch and mask bits. Table A-9 on
page A-18 lists the latch and mask bits.
To process an interrupt, the DSP’s program sequencer does the following.
1. Outputs the appropriate interrupt vector address
2. Pushes the current PC value (the return address) on to the PC stack
3. Pushes the current value of the ASTATx,y and MODE1 registers onto
the status stack (if the interrupt is IRQ2-0, timer, or VIRPT)
3-32
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
4. Sets the appropriate bit in the interrupt latch register (IRPTL)
5. Alters the interrupt mask pointer (IMASKP) to reflect the current
interrupt nesting state, depending on the nesting mode
At the end of the interrupt service routine, the sequencer processes the
return from interrupt (RTI) instruction and does following.
1. Returns to the address stored at the top of the PC stack
2. Pops this value off of the PC stack
3. Pops the status stack (if the ASTATx,y and MODE1 status registers
were pushed for the IRQ2-0, timer, or VIRPT interrupt)
4. Clears the appropriate bit in the interrupt latch register (IRPTL)
and interrupt mask pointer (IMASKP)
Except for reset, all interrupt service routines should end with a
return-from-interrupt (RTI) instruction. After reset, the PC stack is
empty, so there is no return address. The last instruction of the reset service routine should be a jump to the start of your program.
If software writes to a bit in IRPTL forcing an interrupt, the processor recognizes the interrupt in the following cycle, and two cycles of branching
to the interrupt vector follow the recognition cycle.
The DSP responds to interrupts in three stages: synchronization and
latching (1 cycle), recognition (1 cycle), and branching to the interrupt
ADSP-21160 SHARC DSP Hardware Reference
3-33
Interrupts and Sequencing
vector (2 cycles). Table 3-15, Table 3-16 on page 3-35, and Table 3-17
on page 3-36 show the pipelined execution cycles for interrupt processing.
Table 3-15. Pipelined Execution cycles For Interrupt During Single-Cycle
Instruction
Cycles
Fetch
Decode
Execute
1
n+1
n
n – 11
2
n+ 22
n+1nop3
n
3
v4
n+2NOP5
NOP
4
v+1
v
NOP
5
v+2
v+1
v
Note that n is the single-cycle instruction, and v is the interrupt vector instruction
1. Interrupt occurs
2. Interrupt recognized
3. n+1 pushed on PC stack; n+1 suppressed
4. Interrupt vector output
5. n+2 suppressed
3-34
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Table 3-16. Pipelined Execution Cycles For Interrupt During Instruction
With Conflicting PM Data Access (Instruction Not Cached)
Cycles
Fetch
Decode
Execute
1
n+1
n
n – 11
2
—2
n+1nop3
n
3
n+24
n+1nop5
NOP
4
v6
n+2NOP7
NOP
5
v+1
v
NOP
6
v+2
v+1
v
Note that n is the conflicting instruction, and v is the interrupt vector instruction
1. Interrupt occurs
2. Interrupt recognized, but not processed; PM data access
3. n+1 suppressed
4. Interrupt processed
5. n+1 suppressed
6. Interrupt vector output
7. n+1 pushed on PC stack; n+2 suppressed
ADSP-21160 SHARC DSP Hardware Reference
3-35
Interrupts and Sequencing
Table 3-17. Pipelined Execution Cycles for Interrupt During Delayed
Branch Instruction
Cycles
Fetch
Decode
Execute
1
n+1
n
n – 11
2
n+22
n+1
n
3
j
n+2
n+1
4
j+13
j nop4
n+2
5
v5
j+1NOP6
NOP
6
v+1
v
NOP
7
v+2
v+1
v
Note that n is the delayed branch instruction, j is the instruction at the branch address, and v is
the interrupt vector instruction
1. Interrupt occurs
2. Interrupt recognized, but not processed
3. Interrupt processed
4. For a Call, n+3 (return address) is pushed onto the PC stack; j suppressed
5. Interrupt vector output
6. j pushed on PC stack; j+1 suppressed
For most interrupts, internal and external, only one instruction is executed after the interrupt occurs (and before the two instructions aborted)
while the processor fetches and decodes the first instruction of the service
routine. Because of the one-cycle delay between an arithmetic exception
and the STKYx,y register update, there are two cycles after an arithmetic
exception occurs before interrupt processing starts. Table 3-18 on
page 3-37 lists the latency associated with the IRQ2-0 interrupts and the
multiprocessor vector interrupt.
3-36
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Table 3-18. Minimum Latency of the IRQ2-0 and VIRPT Interrupts
Interrupt
Minimum Latency
IRQ2-0
3 cycles
VIRPT
6 cycles
If nesting is enabled and a higher priority interrupt occurs immediately
after a lower priority interrupt, the service routine of the higher priority
interrupt is delayed by one additional cycle. This delay allows the first
instruction of the lower priority interrupt routine to be executed before it
is interrupted. For more information, “Nesting Interrupts” on page 3-43.
Certain DSP operations that span more than one cycle hold off interrupt
processing. If an interrupt occurs during one of these operations, the DSP
latches the interrupt, but delays its processing. The operations that have
delayed interrupt processing are as follows.
• A branch (Jump or Call/return) instruction and the following
cycle, whether it is an instruction (in a delayed branch) or a NOP
(in a non-delayed branch)
• The first of the two cycles used to perform a program memory data
access and an instruction fetch when the instruction is not cached
• The third-to-last iteration of a one-instruction loop
• The last iteration of a one-instruction loop executed once or twice
or of a two-instruction loop executed once, and the following cycle
(which is a NOP)
• The first of the two cycles used to fetch and decode the first
instruction of an interrupt service routine
ADSP-21160 SHARC DSP Hardware Reference
3-37
Interrupts and Sequencing
• Any waitstates for external memory accesses
• Any external memory access that is required when the DSP does
not have control of the external bus, during a host bus grant or
when the DSP is a bus slave in a multiprocessing system
Sensing Interrupts
The DSP supports two types of interrupt sensitivity—the signal shape that
triggers the interrupt. On interrupt pins (IRQ2-0), either the input signal’s
edge or level can trigger an external interrupt.
The DSP detects a level-sensitive interrupt if the signal input is low
(active) when sampled on the rising edge of CLKIN. A level-sensitive interrupt must go high (inactive), before the processor returns from the
interrupt service routine. If a level-sensitive interrupt is still active when
the DSP samples it after returning from its service routine, the DSP treats
the signal as a new request, repeating the same interrupt routine without
returning to the main program, assuming no higher priority interrupts are
active.
The DSP detects an edge-sensitive interrupt if the input signal is high
(inactive) on one cycle and low (active) on the next cycle when sampled on
the rising edge of CLKIN. An edge-sensitive interrupt signal can stay active
indefinitely without triggering additional interrupts. To request another
interrupt, the signal must go high, then low again.
Edge-sensitive interrupts require less external hardware compared to
level-sensitive requests, because there is never a need to negate the request.
An advantage of level-sensitive interrupts is that multiple interrupting
devices may share a single level-sensitive request line on a wired-OR basis,
allowing for easy system expansion.
The MODE2 register controls external interrupt sensitivity. Table A-3 on
page A-7 lists all bits in the MODE2 register.
3-38
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
The following bits in MODE2 control interrupt sensitivity.
• Interrupt 0 Sensitivity. Bit 0, (IRQ0E), directs the DSP to detect
IRQ0 as edge-sensitive (if 1) or level-sensitive (if 0).
• Interrupt 1 Sensitivity. Bit 1, (IRQ1E), directs the DSP to detect
IRQ1 as edge-sensitive (if 1) or level-sensitive (if 0).
• Interrupt 2 Sensitivity. Bit 2, (IRQ2E), directs the DSP to detect
IRQ2 as edge-sensitive (if 1) or level-sensitive (if 0).
The DSP accepts external interrupts that are asynchronous to the DSP’s
clock (CLKIN), allowing external interrupt signals to change at any time.
An external interrupt must be held low at least one CLKIN cycle to guarantee that the DSP samples the signal.
interrupts must meet the setup and hold time require External
ments relative to the rising edge of
. For information on
CLKIN
interrupt signal timing requirements, see the DSP’s data sheet.
Masking Interrupts
The sequencer supports interrupt masking—latching an interrupt, but not
responding to it. Except for the RESET and EMU interrupts, all interrupts are
maskable. If a masked interrupt is latched, the DSP responds to the
latched interrupt if it is later unmasked.
Interrupts can be masked globally or selectively. Bits in the MODE1, IMASK,
and LIRPTL registers control interrupt masking. Table A-2 on page A-3
lists the bits in MODE1, Table A-9 on page A-18 lists the bits in IMASK, and
Table A-10 on page A-24 lists the bits in LIRPTL.
ADSP-21160 SHARC DSP Hardware Reference
3-39
Interrupts and Sequencing
These bits control interrupt masking as follows.
• Global interrupt enable. MODE1, Bit 12, (IRPTEN), directs the DSP
to enable (if 1) or disable (if 0) all interrupts.
• Selective interrupt enable. IMASK, Bits 32-0, direct the DSP to
enable (if 1) or disable/mask (if 0) the corresponding interrupt.
• Selective link port interrupt enable. LIRPTL, Bits 21-16, (LPxMSK)
direct the DSP to enable (if 1) or disable/mask (if 0) the corresponding link port interrupt.
Except for the non-maskable interrupts and boot interrupts, all interrupts
are masked at reset. For booting, the DSP automatically unmasks and uses
either the external port (EPOI) or link port (LP4I) interrupt after reset,
depending on whether the ADSP-21160 DSP is booting from EPROM,
host, or link ports.
Latching Interrupts
When the DSP recognizes an interrupt, the DSP’s interrupt latch (IRPTL
and LIRPTL) registers latch the interrupts—set a bit to record that the
interrupt occurred. The bits in these registers indicate all interrupts that
are currently being serviced or are pending. Because these registers are
readable and writable, any interrupt (except reset) can be set or cleared in
software. Note that writing to the reset bit (bit 1) in IRPTL puts the processor into an illegal state.
When an interrupt occurs, the sequencer sets the corresponding bit in
IRPTL or LIRPTL. During execution of the interrupt’s service routine, the
DSP keeps this bit cleared—clearing the bit during every cycle to prevent
the same interrupt from being latched while its service routine is executing. After the return from interrupt (RTI), the sequencer stops clearing
the latch bit.
3-40
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
If necessary, it is possible to re-use an interrupt while it is being serviced.
For more information, see “Reusing Interrupts” on page 3-45.
The interrupt latch bits in IRPTL correspond to interrupt mask bits in the
IMASK register. In both registers, the interrupt bits are arranged in order of
priority. The interrupt priority is from 0 (highest) to 31 (lowest). Interrupt priority determines which interrupt is serviced first when more than
one occurs in the same cycle. Priority also determines which interrupts are
nested when the DSP has interrupt nesting enabled. For more information, see “Nesting Interrupts” on page 3-43.
While IRPTL latches interrupts for a variety of events, the LIRPTL register
contains latch and mask bits only for Link Port DMA interrupts. A logical
Or’ing of link port interrupts (masked-latch state) appears in the LPSUM bit
in the IRPTL register. Because the LPSUM bit has a corresponding mask bit
in the IMASK register, programs can use LPSUM for a second level of link
port interrupt masking.
Multiple events can cause arithmetic interrupts—fixed-point overflow
(FIXI) and floating-point overflow (FLTOI), underflow (FLTUI), and
invalid operation (FLTII). To determine which event caused the interrupt,
a program can read the arithmetic status flags in the STYKx or STKYy status
registers. Table A-5 on page A-13 lists the bits in these registers. Service
routines for arithmetic interrupts must clear the appropriate STKYx or
STKYy bits to clear the interrupt. If the bits are not cleared, the interrupt is
still active after the return from interrupt (RTI).
bits in
only apply in SIMD mode. For more informa Status
tion, see “Secondary Processing Element (PEy)” on page 2-35.
STKYy
One event can cause multiple interrupts. The timer decrementing to zero
causes two timer expired interrupts, TMZHI (high priority) and TMZLI (low
priority). This feature allows selection of the priority for the timer interrupt. Programs should unmask the timer interrupt with the desired
priority and leave the other one masked. If both interrupts are unmasked,
IRPTL latches both interrupts when the timer reaches zero, and the DSP
ADSP-21160 SHARC DSP Hardware Reference
3-41
Interrupts and Sequencing
services the higher priority interrupt first, then the lower priority
interrupt.
The IRPTL also supports software interrupts. When a program sets the
latch bit for one of these interrupts (SFT0I, SFT1I, SFT2I, or SFT3I), the
sequencer services the interrupt, and the DSP branches to the corresponding interrupt routine. Software interrupts have the same behavior as all
other maskable interrupts.
Stacking Status During Interrupts
To run in an interrupt driven system, programs depend on the DSP being
restored to its pre-interrupt state after an interrupt is serviced. The
sequencer’s status stack eases the return from interrupt process by eliminating some interrupt service overhead—register saves and restores.
The status stack is fifteen locations deep. The stack is full when all entries
are occupied, is empty when no entries are occupied, and is overflowed if a
push occurs when the stack is already full. Bits in the STKYx register indicate the status stack full and empty states. Table A-5 on page A-13 lists
the bits in the STYKx register. The STKYx bits that indicate status stack status are:
• Status stack overflow. Bit 23, (SSOV), indicates that the status stack
is overflowed (if 1) or not overflowed (if 0)—a sticky bit.
• Status stack empty. Bit 24, (SSEM), indicates that the status stack is
empty (if 1) or not empty (if 0)—not sticky, cleared by a Push.
For some interrupts (IRQ2-0, timer expired, and VIRPT), the sequencer
automatically pushes the ASTATx, ASTATy, and MODE1 registers onto the status stack. When the sequencer pushes an entry onto the status stack, the
DSP uses the MMASK register to set up MODE1 register.
The sequencer automatically pops the ASTATx, ASTATY, and MODE1 registers
from the status stack during the return from interrupt instruction (RTI).
3-42
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
In one other case—Jump (CI), the sequencer pops the stack. For more
information, see “Reusing Interrupts” on page 3-45.
Only the IRQ2-0, timer expired, and VIRPT interrupts cause the sequencer
to push an entry onto the status stack. All other interrupts require explicit
saves and restores of effected registers or require an explicit push or pop of
the stack (Push/Pop Sts).
Pushing ASTATx, ASTATy, and MODE1 preserves the status and control bit
settings, allowing a service routine to alter these bits with the knowledge
that the original settings are automatically restored upon the return from
the interrupt.
The top of the status stack contains the current values of ASTATx, ASTATy,
and MODE1. Reading and writing these registers does not move the stack
pointer. Explicit Push or Pop instructions do move the status stack
pointer.
Nesting Interrupts
The sequencer supports interrupt nesting—responding to another interrupt while a previous interrupt is being serviced. Bits in the MODE1, IMASKP,
and LIRPTL registers control interrupt nesting. Table A-2 on page A-3 lists
the bits in MODE1, Table A-9 on page A-18 lists the bits in IMASKP, and
Table A-10 on page A-24 lists the bits in LIRPTL.
ADSP-21160 SHARC DSP Hardware Reference
3-43
Interrupts and Sequencing
These bits control interrupt nesting as follows.
• Interrupt nesting enable. MODE1, Bit 11 (NESTM), directs the DSP to
enable (if 1) or disable (if 0) interrupt nesting.
• Interrupt Mask Pointer. IMASKP, 32 Bits, lists the interrupts in priority order and provides a temporary interrupt mask for each
nesting level.
• Link Port DMA Interrupt Mask Pointer. LIRPTL, Bits 21-16,
(LPxMSK), lists link port DMA interrupts in priority order and provides a temporary interrupt mask for each nesting level.
When interrupt nesting is disabled, a higher priority interrupt can not
interrupt a lower priority interrupt’s service routine. Other interrupts are
latched as they occur, but the DSP processes them after the active routine
finishes.
When interrupt nesting is enabled, a higher priority interrupt can interrupt a lower priority interrupt’s service routine. Lower interrupts are
latched as they occur, but the DSP process them after the nested routines
finish.
Programs should only change the interrupt nesting enable (NESTM) bit
while outside of an interrupt service routine or during the reset service
routine.
If nesting is enabled and a higher priority interrupt occurs immediately
after a lower priority interrupt, the service routine of the higher priority
interrupt is delayed by one cycle. This delay allows the first instruction of
the lower priority interrupt routine to be executed, before it is
interrupted.
When servicing nested interrupts, the DSP uses the interrupt mask
pointer (IMASKP) to create a temporary interrupt mask for each level of
interrupt nesting; the IMASK value is not effected. The DSP changes
3-44
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
each time a higher priority interrupt interrupts a lower priority service routine.
IMASKP
The bits in IMASKP correspond to the interrupts in order of priority. When
an interrupt occurs, the DSP sets its bit in IMASKP. If nesting is enabled,
the DSP uses IMASKP to generate a new temporary interrupt mask, masking all interrupts of equal or lower priority to the highest priority bit set in
IMASKP and keeping higher priority interrupts the same as in IMASK. When
a return from an interrupt service routine (RTI) is executed, the DSP
clears the highest priority bit set in IMASKP and generates a new temporary
interrupt mask, masking all interrupts of equal or lower priority to the
highest priority bit set in IMASKP. The bit set in IMASKP that has the highest
priority always corresponds to the priority of the interrupt being serviced.
If an interrupt re-occurs while its service routine is running and nesting is
enabled, the DSP updates IRPTL, but does not service the interrupt. The
DSP waits until the return from interrupt (RTI) completes, before vectoring to the service routine again.
If nesting is not enabled, the DSP masks out all interrupts and IMASKP is
not used, but the DSP still updates IMASKP to create a temporary interrupt
mask.
The interrupt controller uses the
register and the
 bits
of the
register. These bits should not be modified to
IMASKP
LPxMSKP
LIRPTL
ensure proper functioning of the interrupt controller.
Reusing Interrupts
Unless interrupt nesting is enabled, the DSP ignores and does not latch an
interrupt that re-occurs while its service routine is executing. When the
interrupt initially occurs, the sequencer sets the corresponding bit in
IRPTL. During execution of the service routine, the sequencer keeps this
bit cleared—the DSP clears the bit during every cycle, preventing the
same interrupt from being latched while its service routine is already
executing.
ADSP-21160 SHARC DSP Hardware Reference
3-45
Interrupts and Sequencing
If necessary, it is possible to re-use an interrupt while it is being serviced.
Using a Jump clear interrupt—Jump (CI)—instruction in the interrupt
service routine clears the interrupt, allowing its re-usage while the service
routing is executing.
The Jump (CI) instruction reduces an interrupt service routine to a normal subroutine, clearing the appropriate bit in the interrupt latch and
interrupt mask pointer and popping the status stack. After the Jump (CI)
instruction, the DSP stops automatically clearing the interrupt’s latch bit,
allowing the interrupt to latch again.
When returning from a subroutine entered with a Jump (CI) instruction,
a program must use a return loop re-entry—RTS (LR)—instruction. For
more information, see “Restrictions On Ending Loops” on page 3-22.
The following example shows an interrupt service routine that is reduced
to a subroutine with the (CI) modifier:
instr1; {interrupt entry from main program}
JUMP(PC,3) (DB,CI); {clear interrupt status}
instr3;
instr4;
instr5;
RTS (LR); {use LR modifier with return from subroutine}
Jump(PC,3)(DB,CI) instruction actually only continues linear
 The
execution flow by jumping to the location PC + 3 (
), with
instr5
the two intervening instructions (instr3, instr4) being executed
because of the delayed branch (DB). This Jump instruction is only
an example—a Jump (CI) can be to any location.
Interrupting IDLE
The sequencer supports placing the DSP in Idle—a special instruction
that halts the processor core in a low-power state, until an external interrupt (IRQ2-0), timer interrupt, DMA interrupt, or VIRPT vector interrupt
3-46
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
occurs. When executing an Idle instruction, the sequencer fetches one
more instruction at the current fetch address and then suspends operation.
The DSP’s I/O processor is not effected by the Idle instruction—DMA
transfers to or from internal memory continues uninterrupted.
The processor’s internal clock and timer (if it is enabled) continue to run
during Idle. When an external interrupt (IRQ2-0), timer interrupt, DMA
interrupt, or VIRPT vector interrupt occurs, the processor responds normally. After two cycles used to fetch and decode the first instruction of the
interrupt service routine, the processor continues executing instructions
normally.
Multiprocessing Interrupts
The sequencer supports a multiprocessor vector interrupt. The vector
interrupt (VIRPT) permits passing interprocessor commands in multiple-processor systems. This interrupt occurs when an external processor (a
host or another DSP) writes an address to the VIRPT register, inserting a
new vector address for VIRPT.
There is room in the VIRPT register for the vector address and data for the
service routine. Table A-18 on page A-48 lists the bits in the VIRPT
registers.
When servicing a VIRPT interrupt, the DSP automatically pushes the status
stack and executes the service routine located at the address specified in
VIRPT. During the return from interrupt (RTI), the DSP automatically
pops the status stack.
To flag that a VIRPT interrupt is pending, the DSP sets the VIPD bit in the
SYSTAT register when the external processor writes to the VIRPT register.
ADSP-21160 SHARC DSP Hardware Reference
3-47
Timer and Sequencing
Programs passing interprocessor commands must monitor VIPD to check if
the DSP can receive a new VIRPT address because:
• If an external processor writes VIRPT while a previous vector is
pending, the new VIRPT address replaces the previous pending one.
• If an external processor writes VIRPT while a previous vector is executing, the new VIRPT address does not execute (no new interrupt is
triggered).
When returning from a VIRPT interrupt, the DSP clears the VIPD bit. Note
that if a DSP writes to its own VIRPT register, the write is ignored.
Timer and Sequencing
The sequencer includes a programmable interval timer, which appears in
Figure 3-2 on page 3-4. Bits in the MODE2, TCOUNT, and TPERIOD registers
control timer operations. Table A-3 on page A-7 lists the bits in MODE2.
The bits that control the timer are:
• Timer enable. MODE2, Bit 5 (TIMEN), directs the DSP to enable (if 1)
or disable (if 0) the timer.
• Timer count. (TCOUNT) This register contains the decrementing
timer count value, counting down the cycles between timer
interrupts.
• Timer period. (TPERIOD) This register contains the timer period,
indicating the number of cycles between timer interrupts.
The TCOUNT register contains the timer counter. The timer decrements the
register each clock cycle. When the TCOUNT value reaches zero, the
timer generates an interrupt and asserts the TIMEXP output high for four
core cycles (when the timer is enabled) as shown in Figure 3-4 on
TCOUNT
3-48
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
page 3-31. On the clock cycle after TCOUNT reaches zero, the timer automatically reloads TCOUNT from the TPERIOD register.
The TPERIOD value specifies the frequency of timer interrupts. The number of cycles between interrupts is TPERIOD + 1. The maximum value of
32
TPERIOD is 2 – 1.
To start and stop the timer, programs use the MODE2 register’s TIMEN bit.
With the timer disabled (TIMEN=0), the program loads TCOUNT with an initial count value and loads TPERIOD with the number of cycles for the
desired interval. Then, the program enables the timer (TIMEN=1) to begin
the count.
When a program enables the timer, the timer starts decrementing the
TCOUNT register at the end of the next clock cycle. If the timer is subsequently disabled, the timer stops decrementing TCOUNT after the next clock
cycle as shown in Figure 3-5.
The timer expired event (TCOUNT decrements to zero) generates two interrupts, TMZHI and TMZLI. For information on latching and masking these
interrupts to select timer expired priority, see “Latching Interrupts” on
page 3-40.
ADSP-21160 SHARC DSP Hardware Reference
3-49
Stacks and Sequencing
TIMER ENABLE
SET TIMEN
TIMER ACTIVE
IN MODE2
CLKIN
TCOUNT=N
TCOUNT=N
TCOUNT=N-1
TIMER DISABLE
CLEAR TIMEN
TIMER INACTIVE
IN MODE2
CLKIN
TCOUNT=M-1
TCOUNT=M-2
TCOUNT=M-2
Figure 3-5. Timer Enable and Disable
As with other interrupts, the sequencer needs two cycles to fetch and
decode the first instruction of the timer expired service routine, before
executing the routine. The pipeline execution for the timer interrupt
appears in Table 3-15 on page 3-34.
Programs can read and write the TPERIOD and TCOUNT registers, using universal register transfers. Reading the registers does not effect the timer.
Note that an explicit write to TCOUNT takes priority over the sequencer’s
loading TCOUNT from TPERIOD and the timer’s decrementing of TCOUNT.
Also note that TCOUNT and TPERIOD are not initialized at reset; programs
should initialize these registers before enabling the timer.
Stacks and Sequencing
The sequencer includes a Program Counter (PC) stack, which appears in
Table 3-2 on page 3-6. At the start of a subroutine or loop, the sequencer
pushes return addresses for subroutines (Call/return instructions) and
3-50
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
top-of-loop addresses for loops (Do/Until) instructions onto the PC stack.
The sequencer pops the PC stack during a return from interrupt (RTI),
returns from subroutine (RTS), and loop termination.
The PC stack is 30 locations deep. The stack is full when all entries are
occupied, is empty when no entries are occupied, and is overflowed if a
push occurs when the stack is already full. Bits in the STKYx register indicate the PC stack full and empty states. Table A-5 on page A-13 lists the
bits in the STYKx register. The STKYx bits that indicate PC stack status are:
• PC stack full. Bit 21, (PCFL), indicates that the PC stack is full (if 1)
or not full (if 0)—not a sticky bit, cleared by a Pop.
• PC stack empty. Bit 22, (PCEM), indicates that the PC stack is empty
(if 1) or not empty (if 0)—not sticky, cleared by a Push.
The PC stack full condition causes a maskable interrupt (SOVFI). This
interrupt occurs when the PC stack has 29 locations filled (the almost full
state). The PC stack full interrupt occurs when one location is left, because
the PC stack full service routine needs that last location for its return
address.
The address of the top of the PC stack is available in the PC stack pointer
(PCSTKP) register. The value of PCSTKP is zero when the PC stack is empty,
is 1...30 when the stack contains data, and is 31 when the stack overflows.
This register is a readable and writable register. A write to PCSTKP takes
effect after a one-cycle delay. If the PC stack is overflowed, a write to
PCSTKP has no effect.
The overflow and full flags provide diagnostic aid only. Programs should
not use these flags for runtime recovery from overflow. Note that the status stack, loop stack overflow, and PC stack full conditions trigger a
maskable interrupt.
The empty flags can ease stack saves to memory. Programs can monitor
the empty flag when saving a stack to memory to determine when the DSP
has transferred all values.
ADSP-21160 SHARC DSP Hardware Reference
3-51
Conditional Sequencing
Conditional Sequencing
The sequencer supports conditional execution with conditional logic that
appears in Figure 3-2 on page 3-4. This logic evaluates conditions for conditional (If) instructions and loop (Do/Until) terminations. The
conditions are based on information from the arithmetic status registers
(ASTATx and ASTATy), the mode control 1 register (MODE1), the flag inputs
and the loop counter. For more information on arithmetic status, see
“Using Computational Status” on page 2-7. When in SIMD mode, conditional execution is effected by the arithmetic status of both processing
elements. For information on conditional sequencing in SIMD mode, see
“SIMD Mode and Sequencing” on page 3-55.
Each condition that the DSP evaluates has an assembler mnemonic. The
condition mnemonics for conditional instructions appear in Table 3-19
on page 3-53. For most conditions, the sequencer can test both true and
false states. For example, the sequencer can evaluate ALU equal-to-zero
(EQ) and ALU not-equal-to-zero (NZ).
To test conditions that do not appear in Table 3-19, a program can use
the Test Flag (TF) condition that is generated from a Bit Test Flag (BTF)
instruction. The TF flag is set or cleared as a result of a Bit Test or Bit
3-52
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
XOR instruction, which can test the contents of any of the DSP’s system
registers, including STKYx and STKYy.
Table 3-19. IF Condition and Do/Until Termination Mnemonics
Condition From
Description
True if…
Mnemonic
ALU
ALU = 0
AZ = 1
EQ
ALU  0
AZ = 0
NE
ALU > 0
footnote1
GT
ALU < zero
footnote2
LT
ALU  0
footnote3
GE
ALU  0
footnote4
LE
ALU carry
AC = 1
AC
ALU not carry
AC = 0
NOT AC
ALU overflow
AV = 1
AV
ALU not overflow
AV = 0
NOT AV
Multiplier overflow
MV = 1
MV
Multiplier not overflow
MV= 0
NOT MV
Multiplier sign
MN = 1
MS
Multiplier not sign
MN = 0
NOT MS
Shifter overflow
SV = 1
SV
Shifter not overflow
SV = 0
NOT SV
Shifter zero
SZ = 1
SZ
Shifter not zero
SZ = 0
NOT SZ
Bit test flag true
BTF = 1
TF
Bit test flag false
BTF = 0
NOT TF
Multiplier
Shifter
Bit Test
ADSP-21160 SHARC DSP Hardware Reference
3-53
Conditional Sequencing
Table 3-19. IF Condition and Do/Until Termination Mnemonics (Cont’d)
Condition From
Description
True if…
Mnemonic
Flag Input
Flag0 asserted
FI0 = 1
FLAG0_IN
Flag0 not asserted
FI0 = 0
NOT FLAG0_IN
Flag1 asserted
FI1 = 1
FLAG1_IN
Flag1 not asserted
FI1 = 0
NOT FLAG1_IN
Flag2 asserted
FI2 = 1
FLAG2_IN
Flag2 not asserted
FI2 = 0
NOT FLAG2_IN
Flag3 asserted
FI3 = 1
FLAG3_IN
Flag3 not asserted
FI3 = 0
NOT FLAG3_IN
Mode
Sequencer
1
2
3
4
Bus master true
BM
Bus master false
NOT BM
Loop counter expired (Do)
CURLCNTR = 1
LCE
Loop counter not expired
(If )
CURLCNTR  1
NOT ICE
Always false (Do)
Always
FOREVER
Always true (If )
Always
TRUE
ALU greater than (GT) is true if: [AF and (AN xor (AV and ALUSAT)) or (AF and AN)] or AZ = 0
ALU less than (LT) is true if: [AF and (AN xor (AV and ALUSAT)) or (AF and AN and AZ)] = 1
ALU greater equal (GE) is true if: [AF and (AN xor (AV and ALUSAT)) or (AF and AN and AZ)] = 0
ALU lesser or equal (LE) is true if: [ AF and (AN xor (AV and ALUSAT)) or (AF and AN)] or AZ = 1
The two conditions that do not have complements are LCE/Not LCE
(loop counter expired/not expired) and True/Forever. The context of
these condition codes determines their interpretation. Programs should
use True and Not LCE in conditional (IF) instructions. Programs should
use Forever and LCE to specify loop (Do/Until) termination. A Do Forever instruction executes a loop indefinitely, until an interrupt or reset
intervenes.
3-54
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
There are some restrictions on how programs may use conditions in
Do/Until loops. For more information, see “Restrictions On Ending
Loops” on page 3-22 and “Restrictions On Short Loops” on page 3-23.
bus master condition (BM) indicates whether the DSP is the
 The
current bus master in a multiprocessor system. To enable testing
this condition, a program must clear the MODE1 register’s Condition
Code Select (CSEL) bits. Otherwise, the bus master condition is
always false.
SIMD Mode and Sequencing
The DSP supports a Single-Instruction, Multiple-Data (SIMD) mode. In
this mode, both of the DSP’s processing elements (PEx and PEy) execute
instructions and generate status conditions. For more information on
SIMD computations, see “Secondary Processing Element (PEy)” on
page 2-35.
Because the two processing elements can generate different outcomes, the
sequencers must evaluate conditions from both elements (in SIMD mode)
for conditional (If) instructions and loop (Do/Until) terminations. The
DSP records status for the PEx element in the ASTATx and STKYx registers.
The DSP records status for the PEy element in the ASTATy and STKYy registers. Table A-4 on page A-9 lists the bits in ASTATx and ASTATy, and
Table A-5 on page A-13 lists the bits in STKYx and STKYy.
Even though the DSP has dual processing elements, the sequencer does
not have dual sets of stacks. There is one PC stack, one loop address stack,
and one loop counter stack. The status bits for stacks are in STKYx and are
not duplicated in STKYy. In SIMD mode, the status stack stores both
ASTATx and ASTATy. A status stack Push or Pop instruction in SIMD mode
effects both registers in parallel.
While in SIMD mode, the sequencer evaluates conditions from both PE’s
for conditional (If) and loop (Do/Until) instructions. Table 3-20
ADSP-21160 SHARC DSP Hardware Reference
3-55
SIMD Mode and Sequencing
summarizes how the sequencer resolves each conditional test when SIMD
mode is enabled.
Table 3-20. Conditional Execution Summary
Conditional Operation
Conditional Outcome Depends On …
Compute Operations
Executes in each PE independently depending on condition test in each PE
Branches and Loops
Executes in sequencer depending on And’ing condition
test on both PE’s.
Data Moves (from complementary
pair1 to complementary pair, including X<->Y swap)
Executes move in each PE (and/or memory) independently depending on condition test in each PE
Data Moves (from uncomplemented
universal register2 to complementary
pair1)
Executes move in each PE (and/or memory) independently depending on condition test in each PE; the
same uncomplemented universal is source for each move
Data Moves (from complementary
pair1 to uncomplemented register2)
Executes explicit move to uncomplemented universal register depending on condition test in PEx only; no implicit
move occurs
DAG Operations
Executes modify3 in DAG depending on Or’ing condition test on both PE’s
1
Complementary pairs are registers with SIMD complements, include PEx/y data registers and
USTAT1/2, USTAT3/4, ASTATx/y, STKYx/y, and PX1/2 universal registers. This also includes
internal memory for conditional execution.
2 Uncomplemented registers are universal registers that do not have SIMD complements.
3 Post-modify operations follow this rule, but pre-modify operations always occur despite outcome
Conditional Compute Operations
While in SIMD mode, a conditional compute operation may execute on
both PE’s, either one PE, or neither PE dependent on the outcome of the
status flag test. Flag testing is independently performed on each PE.
3-56
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Conditional Branches and Loops
The DSP executes a conditional branch (Jump or Call/return) or loop
(Do/Until) based on the result of And’ing the condition tests on both
PE’s. A conditional branch or loop in SIMD mode occurs only when the
condition is true in PEx and PEy.
Using complementary conditions (for example EQ and NE), programs
can produce an OR’ing of the condition tests for branches and loops in
SIMD mode. A conditional branch or loop that uses this technique should
consist of a series of conditional compute operations. These conditional
computes generate NOPs on the PE where a branch or loop does not execute. For more information on programming in SIMD mode, see
ADSP-21160 SHARC DSP Instruction Set Reference.
Conditional Data Moves
The execution of a conditional (If) data move (register-to-register and register-to/from-memory) instruction depends on three factors:
• The explicit data move depends on the evaluation of the conditional test in the PEx processing element
• The implicit data move depends on the evaluation of the conditional test in the PEy processing element
• Both moves depend on the types of registers used in the move
There are four cases for SIMD conditional data moves.
Case 1:
Complementary Register Pair Data Move
In this case data moves from a complementary register pair to a complementary register pair. The DSP executes the explicit move depending on
the evaluation of the conditional test in the PEx processing element and
ADSP-21160 SHARC DSP Hardware Reference
3-57
SIMD Mode and Sequencing
the implicit move depending on the evaluation of the conditional test in
the PEy processing element.
Example: Register to Memory Move – PEx Explicit Register
IF EQ DM(I0,M0) = R2;
For this instruction the DSP is operating in SIMD mode, a register in the
PEx data register file is the explicit register and I0 is pointing to an even
address in internal memory. Indirect addressing is shown in the instructions shown in this example. However, the same results occur using direct
addressing. The data movement resulting from the evaluation of the conditional test in the PEx and PEy processing elements is shown in
Table 3-21.
Table 3-21. Register to Memory Moves – Complementary Pairs
Condition
in PEx
Condition
in PEy
Result
AZx
AZy
Explicit
Implicit
0
0
NO data move occurs
NO data move occurs
0
1
NO data move occurs from r2
to location I0
s2 transfers to location (I0+1)
1
0
r2 transfers to location I0
NO data move occurs from s2
to location (I0+1)
1
1
r2 transfers to location I0
s2 transfers to location (I0+1)
Example: Register to Memory Move – PEy Explicit Register
IF EQ DM(I0,M0) = S2;
For this instruction the DSP is operating in SIMD mode, a register in the
PEy data register file is the explicit register and I0 is pointing to an even
address in internal memory. The data movement resulting from the
3-58
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
evaluation of the conditional test in the PEx and PEy processing elements
is shown in Table 3-22.
Table 3-22. Register to Register Moves – Complementary Pairs
Condition
in PEx
Condition
in PEy
Result
AZx
AZy
Explicit
Implicit
0
0
NO data move occurs
NO data move occurs
0
1
NO data move occurs from
s2 to location I0
r2 transfers to location I0+1
1
0
s2 transfers to location I0
NO data move occurs from r2 to
location I0+1
1
1
s2 transfers to location I0
r2 transfers to location I0+1
Examples: Register to Register Move Instructions
IF EQ R8 = R2;
IF EQ PX1 = R2;
IF EQ USTAT1 = R2;
For these instruction the DSP is operating in SIMD mode and registers in
the PEx data register file are used as the explicit registers. The data movement resulting from the evaluation of the conditional test in the PEx and
PEy processing elements is shown in Table 3-23.
ADSP-21160 SHARC DSP Hardware Reference
3-59
SIMD Mode and Sequencing
Table 3-23. Register to Register Moves – Complementary Pairs
Condition
in PEx
Condition
in PEy
Result
AZx
AZy
Explicit
Implicit
0
0
NO data move occurs
NO data move occurs
0
1
NO data move to registers
r9,px1,ustat1 occurs
s2 transfers to registers s9,px2
and ustat2
1
0
r2 transfers to registers r9,px1
and ustat1
NO data move to s9, px2, or
ustat2 occurs
1
1
r2 transfers to registers r9,px1,
and ustat1
s2 transfers to registers
s9,px2,and ustat2
Examples: Register to Register Move Instructions
IF EQ R8 = S2;
IF EQ PX1 = S2;
IF EQ USTAT1 = S2;
For these instructions the DSP is operating in SIMD mode and registers
in the PEy data register file are used as explicit registers. The data movement resulting from the evaluation of the conditional test in the PEx and
PEy processing elements is shown in Table 3-24.
3-60
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Table 3-24. Register to Register Moves – Complementary Register Pairs
Condition
in PEx
Condition
in PEy
Result
AZx
AZy
Explicit
Implicit
0
0
NO data move occurs
NO data move occurs
0
1
NO data move to registers s9,px r2 transfers to registers s9,px2,
and ustat1 occurs
and ustat2
1
0
s2 transfers to registers r9,px1
and ustat1
NO data move to registers
s9,px2, and ustat2 occurs
1
1
s2 transfers to registers r9,px1,
and ustat1
r2 transfers to registers s9,px2,
and ustat2
Case 2:
Uncomplemented to Complementary Register Move
In this case data moves from an uncomplemented register (Ureg without a
SIMD complement) to a complementary register pair. The DSP executes
the explicit move depending on the evaluation of the conditional test in
the PEx processing element. The DSP executes the implicit move depending on the evaluation of the conditional test in the PEy processing
element. In each processing element where the move occurs, the content
of the source register is duplicated in destination.
Example: Register to Register Move
IF EQ R1 = PX;
and
are complementary registers, the combined
 While
register has no complementary register. For more information, see
PX1
PX2
PX
“Internal Data Bus Exchange” on page 5-7.
ADSP-21160 SHARC DSP Hardware Reference
3-61
SIMD Mode and Sequencing
For this instruction the DSP is operating in SIMD mode. The data movement resulting from the evaluation of the conditional test in the PEx and
PEy processing elements is shown in Table 3-25.
Table 3-25. Complementary to Uncomplemented Register Move
Condition
in PEx
Condition
in PEy
Result
AZx
AZy
Explicit
0
0
r1 remains unchanged
s1 remains unchanged
0
1
r1 remains unchanged
s1 gets px value
1
0
r1 gets px value
s1 remains unchanged
1
1
r1 gets px value
s1 gets px value
Implicit
Case 3:
Complementary Register => Uncomplimentary Register
In this case data moves from a complementary register pair to an uncomplemented register. The DSP executes the explicit move to the
uncomplemented universal register, depending on the condition test in
the PEx processing element only. The DSP does not perform an implicit
move.
Example: Register to Register Move
IF EQ PX = R1;
For this instruction the DSP is operating in SIMD mode. The data movement resulting from the evaluation of the conditional test in the PEx and
PEy processing elements is shown in Table 3-26.
3-62
ADSP-21160 SHARC DSP Hardware Reference
Program Sequencer
Table 3-26. Complementary to Uncomplemented Move
Condition
in PEx
Condition
in PEy
Result
AZx
AZy
Explicit
0
0
r1 remains unchanged
no implicit move
s1 remains unchanged
0
1
r1 remains unchanged
no implicit move
s1 remains unchanged
1
0
r1 40-bit explicit move to px
no implicit move
s1 remains unchanged
1
1
r1 40-bit explicit move to px
no implicit move
s1 remains unchanged
Implicit
For more details on PX register transfers, refer to “Internal Data Bus
Exchange” on page 5-7.
Case 4:
Data Move Involves External Memory or IOP Memory Space
Conditional data moves from a complementary register pair to an uncomplemented register with an access to external memory space or IOP
memory space. This results in unexpected behavior and should not be
used.
Example: Register to Memory Move
IF EQ DM(I0,M0) = R2;
IF EQ DM(I0,M0) = S2;
For these instruction the DSP is operating in SIMD mode and the explicit
register is either a PEx register or PEy register. I0 points to either external
memory space or IOP memory space.
ADSP-21160 SHARC DSP Hardware Reference
3-63
SIMD Mode and Sequencing
Indirect addressing is shown in the instructions shown in this example.
However, the same results occur using direct addressing.
Conditional DAG Operations
Conditional post-modify DAG operations update the DAG register based
on OR’ing of the condition tests on both processing elements. Actual data
movement involved in a conditional DAG operation is based on independent evaluation of condition tests in PEx and PEy. Only the post modify
update is based on the OR’ing of the these conditional tests.
Conditional pre-modify DAG operations behave differently. The DAGs
always pre-modify an index, independent of the outcome of the condition
tests on each processing element.
3-64
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
4 DATA ADDRESS
GENERATORS
This chapter describes Data Address Generators (DAGs).
Overview
The DSP’s Data Address Generators (DAGs) generate addresses for data
moves to and from Data Memory (DM) and Program Memory (PM). By
generating addresses, the DAGs let programs refer to addresses indirectly,
using a DAG register instead of an absolute address. The DAGs architecture, which appears in Figure 4-1, supports several functions that
minimize overhead in data access routines. These functions include:
• Supply address and post-modify—provides an address during a
data move and auto-increments the stored address for the next
move.
• Supply pre-modified address—provides a modified address during
a data move without incrementing the stored address.
• Modify address—increments the stored address without performing a data move.
ADSP-21160 SHARC DSP Hardware Reference
4-1
Overview
DM OR PM DAT A B US
IMME DIAT E
VAL UE F R OM
INS T R UCT ION
32
I
R E GIS T E R S
32
32
32
M
R E GIS T E R S
L
R E GIS T E R S
B
R E GIS T E R S
8 X 32
8 X 32
8 X 32
8 X 32
32
MUX
P OS T -MODIF Y
ADDR E S S ING
MODUL US
L OGIC
ADD
32
B IT -R E VE R S E
(OP T IONAL )
PR E -MODIF Y
ADDR E S S ING
MUX
32
UP DAT E
32
ADDR E S S ADJUS T ME NT P E R WOR D S IZ E (S HOR T , NOR MAL , OR L ONG)
OP T IONAL B IT -R E VE R S E F OR I0-DAG1 & I8-DAG2
OP T IONAL B R OADCAS T F OR I1-DAG1 & I9-DAG2
32
32
MODE 1
MODE 2
PM ADDR E S S B US (DAG2 - I,M,L ,B 8-15)
S T K YX
DM ADDR E S S B US (DAG1 - I,M,L ,B 0-7)
Figure 4-1. Data Address Generator (DAG) Block Diagram
• Bit-reverse address—provides a bit-reversed address during a data
move without reversing the stored address.
• Broadcast data moves—performs dual data moves to complementary registers in each processing element to support SIMD mode.
4-2
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
As shown in Figure 4-1, each DAG has four types of registers. These registers hold the values that the DAG uses for generating addresses. The four
types of registers are:
• Index registers (I0-I7 for DAG1 and I8-I15 for DAG2). An index
register holds an address and acts as a pointer to memory. For
example, the DAG interprets dm(I0,0) and pm(I8,0) syntax in an
instruction as addresses.
• Modify registers (M0-M7 for DAG1 and M8-M15 for DAG2). A
modify register provides the increment or step size by which an
index register is pre- or post-modified during a register move. For
example, the dm(I0, M1) instruction directs the DAG to output the
address in register I0 then modify the contents of I0 using the M1
register.
• Length and Base registers (L0-L7 and B0-B7 for DAG1 and
L8-L15 and B8-B15 for DAG2). Length and base registers set up
the range of addresses and the starting address for a circular buffer.
For more information on circular buffers, see “Addressing Circular
Buffers” on page 4-12.
Setting DAG Modes
The MODE1 register controls the operating mode of the DAGs. Table A-2
on page A-3 lists all the bits in MODE1. The following bits in MODE1 control
Data Address Generator modes:
• Circular buffering enable. Bit 24 (CBUFEN) enables circular buffering (if 1) or disables circular buffering (if 0).
• Broadcast register loading enable, DAG1-I1. Bit 23 (BDCST1)
enables register broadcast loads to complementary registers from I1
indexed moves (if 1) or disables broadcast loads (if 0).
ADSP-21160 SHARC DSP Hardware Reference
4-3
Setting DAG Modes
• Broadcast register loading enable, DAG2-I9. Bit 22 (BDCST9)
enables register broadcast loads to complementary registers from I9
indexed moves (if 1) or disables broadcast loads (if 0)
• SIMD mode enable. Bit 21 (PEYEN) enables computations in
PEy—SIMD mode—(if 1) or disables PEy—SISD mode—(if 0).
For more information on SIMD mode, see “Secondary Processing
Element (PEy)” on page 2-35.
• Secondary registers for DAG2 lo, I,M,L,B8-11. Bit 6 (SRD2L)
Secondary registers for DAG2 hi, I,M,L,B12-15. Bit 5 (SRD2H)
Secondary registers for DAG1 lo, I,M,L,B0-3. Bit 4 (SRD1L)
Secondary registers for DAG1 hi, I,M,L,B4-7. Bit 3 (SRD1H)
These bits select the corresponding secondary register set (if 1) or
select the corresponding primary register set—the set that is available at reset—(if 0).
• Bit-reverse addressing enable, DAG1-I0. Bit 1 (BR0) enables
bit-reversed addressing on I0 indexed moves (if 1) or disables
bit-reversed addressing (if 0).
• Bit-reverse addressing enable, DAG2-I8. Bit 0 (BR8) enables
bit-reversed addressing on I8 indexed moves (if 1) or disables
bit-reversed addressing (if 0).
Circular Buffering Mode
The CBUFEN bit in the MODE1 register enables circular buffering—a mode in
which the DAG supplies addresses ranging within a constrained buffer
length (set with an L register), starting at a base address (set with a B register), and incrementing the addresses on each access by a modify value (set
with an M register).
4-4
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
previous SHARC DSP’s (ADSP-2106x DSPs), circular buffer On
ing is always enabled. For code compatibility, programs ported to
the ADSP-21160 DSP should include the instruction:
Bit Set Mode1 CBUFEN;
This instruction enables circular buffering. For more information on setting up and using circular buffers, see “Addressing Circular Buffers” on
page 4-12. When using circular buffers, the DAGs can generate an interrupt on buffer overflow (wrap around). For more information, see “Using
DAG Status” on page 4-9.
Broadcast Loading Mode
The BDCST1 and BDCST9 bits in the MODE1 register enable broadcast loading
mode—multiple register loads from a single load command. When the
BDCST1 bit is set (1), the DAG performs a dual data register load on
instructions that use the I1 register for the address. The DAG loads both
the named register (explicit register) in one processing element and loads
that register’s complementary register (implicit register) in the other processing element. The BDCST9 bit in the MODE1 register enables this feature
for the I9 register.
Enabling either DAG1 or DAG2 register load broadcasting has no effect
on register stores or loads to universal registers other than the register file
data registers. “Dual Processing Element Register Load Broadcasts” on
page 4-6 demonstrates the effects of a register load operation on both
ADSP-21160 SHARC DSP Hardware Reference
4-5
Setting DAG Modes
processing elements with register load broadcasting enabled. In Table 4-1
on page 4-6, note that Rx and Sx are complementary data registers.
Table 4-1. Dual Processing Element Register Load Broadcasts
1
Example
Instruction
syntax
Rx = DM(I1,Ma); {Syntax #1}
Rx = PM(I9,Mb); {Syntax #2}
Rx = DM(I1,Ma), Rx = PM(I9,Mb);
PEx explicit
operations
PEy implicit
operations
1
{Syntax #3}
Rx = DM(I1,Ma); {Explicit #1}
Rx = PM(I9,Mb); {Explicit #2}
Rx = DM(I1,Ma), Rx = PM(I9,Mb); {Explicit
Sx = DM(I1,Ma); {Implicit #1}
Sx = PM(I9,Mb); {Implicit #2}
Sx = DM(I1,Ma), Sx = PM(I9,Mb);
#3}
{Implicit #3}
The letters “a” and “b” (as in Ma or Mb) indicate numbers for modify registers in DAG1 and
DAG2. The letter “a”, which indicates a DAG1 register, can be replaced with 0 through 7. The
letter “b” indicates a DAG2 register and can be replaced with 8 through 15.
The
bit (SISD/SIMD mode select) does not influence broad cast
operations. Broadcast loading is particularly useful in SIMD
PEYEN
applications where the algorithm needs identical data loaded into
each processing element. For more information on SIMD mode (in
particular, a list of complementary data registers), see “Secondary
Processing Element (PEy)” on page 2-35.
Alternate (Secondary) DAG Registers
Each DAG has an alternate register set. To facilitate fast context switching, the DSP includes alternate register sets for data, results, and data
address generator registers. Bits in the MODE1 register control when alternate registers become accessible. While inaccessible, the contents of
alternate registers are not effected by DSP operations. Note that there is a
one cycle latency between writing to MODE1 and being able to access an
alternate register set. The alternate register sets for the DAGs are described
4-6
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
in this section. For more information on alternate data and results registers, see “Alternate (Secondary) Data Registers” on page 2-30.
Bits in the MODE1 register can activate alternate register sets within the
DAGs: the lower half of DAG1 (I,M,L,B0-3), the upper half of DAG1
(I,M,L,B4-7), the lower half of DAG2 (I,M,L,B8-11), and the upper half
of DAG2 (I,M,L,B12-15). Figure 4-2 shows the DAG’s primary and alternate register sets.
MODE 1 S E L E CT
B IT
SRD1L
SRD1H
DAG1 R E GIS T E R S (DAT A ME MOR Y)
I0
M0
L0
B0
I1
M1
L1
B1
I2
M2
L2
B2
I3
M3
L3
B3
I4
M4
L4
B4
I5
M5
L5
B5
I6
M6
L6
B6
M7
L7
B7
I7
DAG2 R E GIS T E R S (P R OGR AM ME MOR Y)
I8
M8
L8
I9
M9
L9
B9
I10
M10
L10
B10
I11
M11
L11
B11
I12
M12
L12
B12
I13
M13
L13
B 13
I14
M14
L14
B14
I15
M15
L15
B15
SRD2L
SRD2H
B8
Figure 4-2. DAG Primary and Alternate Registers
To share data between contexts, a program places the data to be shared in
one half of either the current DAG’s registers or the other DAG’s registers
and activates the alternate register set of the other half. The following
ADSP-21160 SHARC DSP Hardware Reference
4-7
Setting DAG Modes
example demonstrates how code should handle the one cycle of latency
from the instruction setting the bit in MODE1 to when the alternate registers
may be accessed:
BIT SET MODE1 SRD1L;/* activate alternate dag1 lo regs */
NOP;/* wait for access to alternates */
R0=dm(i0,m1);
Bit-Reverse Addressing Mode
The BR0 and BR8 bits in the MODE1 register enable bit-reverse addressing
mode—outputting addresses in reverse bit order. When BR0 is set (1),
DAG1 bit-reverses 32-bit addresses output from I0. When BR8 is set (1),
DAG2 bit-reverses 32-bit addresses output from I8. The DAGs only
bit-reverse the address output from I0 or I8; the contents of these registers
are not reversed. Bit-reverse addressing mode effects both pre-modify and
post-modify operations. The following example demonstrates how
bit-reverse mode effects address output:
Bit Set Mode1 BR0; /* enables bit-rev. addressing for DAG1 */
I0=0x8a000; /* loads I0 with the bit reverse of the */
buffer’s base address, DM(0x51000) */
M0=0x4000000; /* loads M0 with value for post-modify */
R1=DM(I0,M0); /* loads r1 with contents of DM address
*/
/* DM(0x51000), which is the bit-reverse of */
/* 0x8a000, then post modifies I0 for the next */
/* access with (0x8a000 + 0x4000000)=0x408a000, */
/* which is the bit-reverse of DM(0x51020) */
In addition to bit-reverse addressing mode, the DSP supports a bit-reverse
instruction (Bitrev). This instruction bit-reverses the contents of the
selected register. For more information on the Bitrev instruction, see
“Modifying DAG Registers” on page 4-17 or ADSP-21160 SHARC DSP
Instruction Set Reference.
4-8
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
Using DAG Status
As described in “Addressing Circular Buffers” on page 4-12, the DAGs
can provide addressing for a constrained range of addresses, repeatedly
cycling through this data (or buffer). A buffer overflow (or wrap around)
occurs each time the DAG circles past the buffer’s base address.
The DAGs can provide buffer overflow information when executing circular buffer addressing for I7 or I15. When a buffer overflow occurs (a
circular buffering operation increments the I register past the end of the
buffer), the appropriate DAG updates a buffer overflow flag in a sticky
status (STKYx) register. A buffer overflow can also generate a maskable
interrupt. Two ways to use buffer overflows from circular buffering are:
• Interrupts. Enable interrupts and use an interrupt service routine
to handle the overflow condition immediately. This method is
appropriate if it is important to handle all overflows as they occur;
for example in a “ping-pong” or swap I/O buffer pointers routine.
• STKYx registers. Use the Bit Tst instruction to examine overflow
flags in the STKY register after a series of operations. If an overflow
flag is set, the buffer has overflowed—wrapped around—at least
once. This method is useful when overflow handling is not critical.
DAG Operations
The DSP’s DAGs perform several types of operations to generate data
addresses. As shown in Figure 4-1 on page 4-2, the DAG registers and the
MODE1, MODE2, and STKYx registers all contribute to DAG operations. The
following sections provide details on DAG operations:
ADSP-21160 SHARC DSP Hardware Reference
4-9
DAG Operations
• “Addressing With DAGs” on page 4-10
• “Addressing Circular Buffers” on page 4-12
• “Modifying DAG Registers” on page 4-17
An important item to note from Figure 4-1 on page 4-2 is that the DAG
automatically adjusts the output address per the word size of the address
location (short word, normal word, or long word). This address adjustment lets internal memory use the address directly. For details on these
address adjustments, see “Access Word Size” on page 5-40.
mode, access word size, and data location (inter SISD/SIMD
nal/external) all influence data access operations. “Data Access
Options” on page 5-45
Addressing With DAGs
The DAGs support two types of modified addressing—generating an
address that is incremented by a value or a register. In pre-modify addressing, the DAG adds an offset (modifier), either an M register or an
immediate value, to an I register and outputs the resulting address.
Pre-modify addressing does not change (or update) the I register. The
other type of modified addressing is post-modify addressing. In post-modify addressing, the DAG outputs the I register value unchanged then adds
an M register or immediate value, updating the I register value. Figure 4-3
compares pre- and post-modify addressing.
4-10
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
POST-MODIFY
I REGISTER UPDATE
PRE-MODIFY
NO I REGISTER UPDATE
SYNTAX:
PM(MX, IX)
DM(MX, IX)
SYNTAX:
PM(IX, MX)
DM(IX, MX)
2. UPDATE
I
OUTPUT
1. OUTPUT
I
+
+
M
M
I+M
I+M
Figure 4-3. Data Address Generator (DAG) Block Diagram
The difference between pre-modify and post-modify instructions in the
DSP’s assembly syntax is the position of the index and modifier in the
instruction. If the I register comes before the modifier, the instruction is a
post-modify operation. If the modifier comes before the I register, the
instruction is a pre-modify without update operation. The following
instruction accesses the program memory location indicated by the value
in I15 and writes the value I15 + M12 to the I15 register:
R6 = PM(I15,M12); /* Post-modify addressing with update */
By comparison, the following instruction accesses the program memory
location indicated by the value I15 + M12 and does not change the value in
I15:
R6 = PM(M12,I15); /* Pre-modify addressing without update */
Modify (M) registers can work with any index (I) register in the same
DAG (DAG1 or DAG2). For a list of I and M registers and their DAGs, see
Figure 4-2 on page 4-7.
Instructions can use a number (immediate value), instead of an M register,
as the modifier. The size of an immediate value that can modify an I register depends on the instruction type. For all single data access operations,
modify immediate values can be up to 32 bits wide. Instructions that
ADSP-21160 SHARC DSP Hardware Reference
4-11
DAG Operations
combine DAG addressing with computations limit the size of the modify
immediate value. In these instructions (multifunction computations), the
modify immediate values can be up to 6 bits wide. The following example
instruction accepts up to 32-bit modifiers:
R1=DM(0x40000000,I1);
/* DM address = I1+0x4000 0000 */
The following example instruction accepts up to 6-bit modifiers:
F6=F1+F2,PM(I8,0x0B)=ASTAT; /* PM address = I8, I8=I8+0x0B */
Note that pre-modify addressing operations must not change the memory
space of the address. For example, pre-modifying an address in the DSP’s
internal memory space should not generate an address in external memory
space. For more information, see “Access Word Size” on page 5-40.
Addressing Circular Buffers
The DAGs support addressing circular buffers—a range of addresses containing data that the DAG steps through repeatedly, “wrapping around”
to repeat stepping through the range of addresses in a circular pattern. To
address a circular buffer, the DAG steps the index pointer (I register)
through the buffer, post-modifying and updating the index on each access
with a positive or negative modify value (M register or immediate value).
If the index pointer falls outside the buffer, the DAG subtracts or adds the
length of the buffer from or to the value, wrapping the index pointer back
to the start of the buffer. The DAG’s support for circular buffer addressing appears in Figure 4-1 on page 4-2, and an example of circular buffer
addressing appears in Figure 4-1.
The starting address that the DAG wraps around is called the buffer’s base
address (B register). There are no restrictions on the value of the base
address for a circular buffer.
buffering may only use post-modify addressing. The
 Circular
DAG’s architecture, as shown in Figure 4-1 on page 4-2, cannot
support pre-modify addressing for circular buffering, because circular buffering requires that the index be updated on each access.
4-12
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
It is important to note that the DAGs do not detect memory map overflow or underflow. If the address post-modify produces I+M > 0xffffffff or
I–M < 0, circular buffering may not function correctly. Also, the length of
a circular buffer should not let the buffer straddle the top of the memory
map. For more information on the DSP’s memory map, see
“ADSP-21160 DSP Memory Map” on page 5-12.
As shown in Figure 4-4 on page 4-13, programs use the following steps to
set up a circular buffer:
T HE F OL L OWING S YNT AX S E T S U P AND ACCE S S E S A CIR CU L AR B U F F E R WIT H:
L E NGT H = 11
B AS E ADDR E S S = 0X55000
MODIF IE R = 4
0
1
B IT S E T MODE 1 CB UF E N;
/* E NAB L E S CIR CU L AR B U F F E R ADDR E S S ING; J US T ONCE IN
P R OGR AM */
B 0 = 0X55000;
/* L OADS B 0 AND L 0 R E GIS T E R S WIT H B AS E ADDR E S S */
L 0 = 0XB ;
/* L OADS L 0 R E GIS T E R WIT H L E NGT H OF B U F F E R */
M1 = 0X4;
/* L OADS M1 WIT H MODIF IE R OR S T E P S IZ E */
L CNT R = 11, DO MY_CIR _B UF F E R UNT IL L CE ;
/* S E T S U P A L OOP CONT AINING B U F F E R
ACCE S S E S */
R 0 = DM(I0,M1);
/* AN ACCE S S W IT HIN T HE B U F F E R U S E S P OS T MODIF Y ADDR E S S ING
*/
... 1
/* OT HE R INS T R U CT IONS0 IN T HE MY_CIR _B U F F E R L OOP
0
0 */
MY_CIR _B U F F E R : NOP ;
/* E ND OF MY_CIR _B UF F E R L OOP */
4
1
1
1
7
2
2
2
3
3
3
3
4
4
4
2
4
5
5
6
6
6
7
7
7
8
8
8
9
9
7
8
3
9
9
10
10
5
6
5
2
10
10
5
8
9
6
11
10
THE COLUMNS ABOVE SHOW THE SEQUENCE IN ORDER OF LOCATIONS ACCESSED IN ONE PASS.
NOTE THAT "0" ABOVE IS ADDRESS DM(0X55000). THE SEQUENCE REPEATS ON SUBSEQUENT
PASSES.
Figure 4-4. Data Address Generator (DAG) Block Diagram
ADSP-21160 SHARC DSP Hardware Reference
4-13
DAG Operations
1. Enable circular buffering (Bit Set
is only needed once in a program.
Mode1 CBUFEN;).
This operation
2. Load the buffer’s base address into the B register. This operation
automatically loads the corresponding I register.
3. Load the buffer’s length into the corresponding L register. For
example, L0 corresponds to B0.
4. Load the modify value (step size) into an M register in the corresponding DAG. For example, M0 through M7 correspond to B0.
Alternatively, the program can use an immediate value for the
modifier.
After this set up, the DAGs use the modulus logic in Figure 4-1 on
page 4-2 to process circular buffer addressing.
On the ADSP-21160 DSP, programs enable circular buffering by setting
the CBUFEN bit in the MODE1 register. This bit has a corresponding mask bit
in the MMASK register. Setting the corresponding MMASK bit causes the
CBUFEN bit to be cleared following a push status instruction (Push Sts), the
execution of an external interrupt, timer interrupt, or vectored interrupt.
This feature lets programs disable circular buffering while in an interrupt
service routine that does not use circular buffering. By disabling circular
buffering, the routine does not need to save and restore the DAG’s B and
L registers.
Clearing the CBUFEN bit disables circular buffering for all data load and
store operations. The DAGs perform normal post-modify load and store
accesses instead, ignoring the B and L register values. Note that a write to
a B register modifies the corresponding I register, independent of the state
of the CBUFEN bit. The Modify instruction executes independent of the
state of the CBUFEN bit. The Modify instruction always performs circular
buffer modify of the index registers if the corresponding B and L registers
are set up, independent of the state of the CBUFEN bit.
4-14
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
previous SHARC DSP’s (ADSP-2106x family), circular buffer On
ing is always enabled. For code compatibility, programs ported to
the ADSP-21160 DSP should enable circular buffering
(CBUFEN=1).
On the first post-modify access to the buffer, the DAG outputs the I register value on the address bus then modifies the address by adding the
modify value. If the updated index value is within the buffer length, the
DAG writes the value to the I register. If the updated value is outside the
buffer length, the DAG subtracts (positive) or adds (negative) the L register value before writing the updated index value to the I register. In
equation form, these post-modify and wrap around operations work as
follows:
• If M is positive:
• Inew = Iold + M if Iold + M < Buffer base + length (end of
buffer)
• Inew = Iold + M – L if Iold + M  Buffer base + length (end of
buffer)
• If M is negative:
• Inew = Iold + M if Iold + M  Buffer base (start of buffer)
• Inew = Iold + M + L if Iold + M < Buffer base (start of buffer)
The DAGs use all four types of DAG registers for addressing circular buffers. These registers operate as follows for circular buffering:
The index (I) register contains the value that the DAG outputs on the
address bus.
• The modify (M) register contains the post-modify amount (positive or negative) that the DAG adds to the I register at the end of
each memory access. The M register can be any M register in the
same DAG as the I register and does not have to have the same
ADSP-21160 SHARC DSP Hardware Reference
4-15
DAG Operations
number. The modify value also can be an immediate value instead
of an M register. The size of the modify value, whether from an M
register or immediate, must be less than the length (L register) of
the circular buffer.
• The length (L) register sets the size of the circular buffer and the
address range that the DAG circulates the I register through. L
must be positive and cannot have a value greater than 231 – 1. If an
L register’s value is zero, its circular buffer operation is disabled.
• The base (B) register, or the B register plus the L register, is the
value that the DAG compares the modified I value with after each
access. When the B register is loaded, the corresponding I register
is simultaneously loaded with the same value. When I is loaded, B
is not changed. Programs can read the B and I registers
independently.
There is one set of registers (I7 and I15) in each DAG that can generate an
interrupt on circular buffer overflow (address wraparound). For more
information, see “Using DAG Status” on page 4-9.
When a program needs to use I7 or I15 without circular buffering and the
DSP has the circular buffer overflow interrupts unmasked, the program
should disable the generation of these interrupts by setting the B7/B15 and
L7/L15 registers to values that prevent the interrupts from occurring. If I7
were accessing the address range 0x1000–0x2000, the program could set
B7=0x0000 and L7=0xFFFF. Because the DSP generates the circular buffer
interrupt based on the wrap around equations on page 4-15, setting the L
register to zero does not necessarily achieve the desired results. If the program is using either of the circular buffer overflow interrupts, it should
avoid using the corresponding I register(s) (I7 or I15) where interrupt
branching is not needed.
In the case of circular buffer overflow interrupts, if CBUFEN = 1 and register
L7 = 0 (or L15 = 0), the CB7I (or CB15I) interrupt occurs at every change of
I7 (or I15) after the index register ( I7 or I15) crosses the base register (B7
4-16
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
or B15) value. This wrap-around behavior is independent of the context of
the DAG registers, both primary and alternate.
a Long word access, SIMD access, or Normal word access
 When
(with LW option) crosses the end of the circular buffer, the DSP
completes the access before responding to the end of buffer
condition.
Modifying DAG Registers
The DAGs support two operations that modify an address value in an
index register without outputting an address. These two operations,
address bit-reversal and address modify, are useful for bit-reverse addressing and maintaining pointers.
The Modify instruction modifies addresses in any DAG index register
(I0-I15) without accessing memory. If the I register’s corresponding B and
L registers are set up for circular buffering, a Modify instruction performs
the specified buffer wrap around (if needed). The syntax for Modify is
similar to post-modify addressing (index, then modifier). Modify accepts
either a 32-bit immediate values or an M register as the modifier. The following example adds 4 to I1 and updates I1 with the new value:
Modify(I1,4);
The Bitrev instruction modifies and bit-reverses addresses in any DAG
index register (I0-I15) without accessing memory. This instruction is
independent of the bit-reverse mode. The Bitrev instruction adds a 32-bit
immediate value to a DAG index register, bit-reverses the result, and
writes the result back to the same index register. The following example
adds 4 to I1, bit-reverses the result, and updates I1 with the new value:
Bitrev(I1,4);
ADSP-21160 SHARC DSP Hardware Reference
4-17
DAGs, Registers, and Memory
Addressing in SISD and SIMD Modes
Single-Instruction, Multiple-Data (SIMD) mode (PEYEN bit=1) does not
change the addressing operations in the DAGs, but it does change the
amount of data that moves during each access. The DAGs put the same
addresses on the address buses in SIMD and SISD modes. In SIMD
mode, the DSP’s memory and processing elements get data from the locations named (explicit) in the instruction syntax and complementary
(implicit) locations. For more information on data moves between registers, see “Secondary Processing Element (PEy)” on page 2-35. For more
information on data accesses and memory, see “Data Access Options” on
page 5-45.
DAGs, Registers, and Memory
DAG registers are part of the DSP’s universal register set. Programs may
load the DAG registers from memory, from another universal register, or
with an immediate value. Programs may store DAG registers’ contents to
memory or to another universal register.
The DAG’s registers support the bidirectional register-to-register transfers
that are described in “SIMD (Computational) Operations” on page 2-39.
When the DAG register is a source of the transfer, the destination can be a
register file data register. This transfer results in the contents of the single
source register being duplicated in complementary data registers in each
processing element.
Programs should use care in the case where the DAG register is a destination of a transfer from a register file data register source. Programs should
use a conditional operation to select either one processing element or neither as the source. Having both processing elements contribute a source
value results in the PEx element’s write having precedence over the PEy
element’s write.
4-18
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
In the case where a DAG register is both source and destination, the data
move operation executes the same as it would if SIMD mode were disabled (PEYEN cleared).
DAG Register-to-Bus Alignment
There are three word alignment cases for DAG registers and PM or DM
data buses: Normal word, Extended-precision Normal word, and Long
word.
The DAGs align normal word (32-bit) addressed transfers to the low order
bits of the buses. These transfers between memory and 32-bit DAG1or
DAG2 registers use the 64-bit DM and PM data buses. Figure 4-5 illustrates these transfers.
D M OR P M D AT A B U S
63
31
0
IMP L ICIT (NAME D + OR - 1)
D AG1 OR D AG2 R E GIS T E R S
31
0
31
0
E XP L ICIT (NAME D )
D AG1 OR D AG2 R E GIS T E R S
Figure 4-5. Normal Word (32-bit) DAG Register Memory Transfers
The DAGs align extended-precision normal word (40-bit) addressed
transfers or register-to-register transfers to bits 39-8 of the buses. These
transfers between a 40-bit data register and 32-bit DAG1 or DAG2 registers use the 64-bit DM and PM data buses. Figure 4-6 illustrates these
transfers.
ADSP-21160 SHARC DSP Hardware Reference
4-19
DAGs, Registers, and Memory
DM OR P M DAT A B U S
63
8
39
0X0000 00
0
0X00
31
0
DAG1 OR DAG2 R E GIS T E R S
Figure 4-6. DAG Register to Data Register Transfers
Long word (64-bit) addressed transfers between memory and 32-bit
DAG1 or DAG2 registers target double DAG registers and use the 64-bit
DM and PM data buses. Figure 4-7 illustrates how the bus works in these
transfers.
DM OR P M DAT A B U S
63
31
0
IMP L ICIT (NAME D + OR - 1)
DAG1 OR DAG2 R E GIS T E R S
31
0
31
0
E XP L ICIT (NAME D)
DAG1 OR DAG2 R E GIS T E R S
Figure 4-7. Long Word DAG Register to Data Register Transfers
If the Long word transfer specifies an even-numbered DAG register (e.g.,
I0 or I2), then the even numbered register value transfers on the lower
half of the 64-bit bus, and the even numbered register + 1 value transfers
on the upper half (bits 63-32) of the bus.
If the Long word transfer specifies an odd numbered DAG register (e.g.,
I1, or B3), the odd numbered register value transfers on the lower half of
the 64-bit bus, and the odd numbered register - 1 value (I0 or B2 in this
example) transfers on the upper half (bits 63-32) of the bus.
In both the even- and odd-numbered cases, the explicitly specified DAG
register sources or sinks bits 31-0 of the Long word addressed memory.
4-20
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
DAG Register Transfer Restrictions
The two types of transfer restrictions are hold-off conditions and illegal
conditions that the DSP does not detect.
For certain instruction sequences involving transfers to and from DAG
registers, an extra (Nop) cycle is automatically inserted by the processor.
When an instruction that loads a DAG register is followed by an instruction that uses the same DAG register for data addressing, modify
instructions, or indirect jumps, the DSP inserts an extra (Nop) cycle
between the two instructions. This hold-off happens because the same bus
is needed by both operations in the same cycle. So, the second operation
must be delayed. The following case causes a delay because it exhibits a
write/read dependency in which I0 is written in one cycle. The results of
that register write are not available to a register read for one cycle. Note
that if either instruction had specified I1, the stall would still occur,
because the DSP’s DAG register transfers can occur in pairs. The DAG
detects write/read dependencies with a register pair granularity:
I0=8;
DM(I0,M1)=R1;
Certain other sequences of instructions cause incorrect results on the DSP
and are flagged as errors by DSP assembler software. These types of
instructions can execute on the processor, but cause incorrect results:
• An instruction that stores a DAG register in memory using indirect
addressing from the same DAG, with or without update of the
index register. The instruction writes the wrong data to memory or
updates the wrong index register.
• Do not try these: DM(M2,I1)=I0; or DM(I1,M2)=I0;These
example instructions do not work because I0 and I1 are
both DAG1 registers.
ADSP-21160 SHARC DSP Hardware Reference
4-21
DAG Instruction Summary
• An instruction that loads a DAG register from memory using indirect addressing from the same DAG, with update of the index
register. The instruction will either load the DAG register or
update the index register, but not both.
• Do not try this: L2=DM(I1,M0);This example instruction
does not work because both L2 and I1 are DAG1 registers.
DAG Instruction Summary
Table 4-2, Table 4-3, Table 4-4, Table 4-5, Table 4-6, Table 4-7,
Table 4-8, and Table 4-9 list the DAG instructions. For more information
on assembly language syntax, see ADSP-21160 SHARC DSP Instruction
Set Reference. In these tables, note the meaning of the following symbols:
• I15-8 indicates a DAG2 index register: I15, I14, I13, I12, I11, I10,
I9, or I8, and I7-0 indicates a DAG1 index register I7, I6, I5, I4,
I3, I2, I1, or I0.
• M15-8 indicates a DAG2 modify register: M15, M14, M13, M12, M11,
M10, M9, or M8, and M7-0 indicates a DAG1 modify register M7, M6,
M5, M4, M3, M2, M1, or M0.
• Ureg indicates any universal register; For a list of the DSP’s universal registers, see Table A-1 on page A-2.
• Dreg indicates any data register; For a list of the DSP’s data registers, see the Data Register File registers that are listed in Table A-1
on page A-2.
• Data32 indicates any 32-bit value, and Data6 indicates any 6-bit
value
4-22
ADSP-21160 SHARC DSP Hardware Reference
Data Address Generators
Table 4-2. Post-Modify Addressing, Modified by M Register and
Updating I Register
DM(I7-0,M7-0)=Ureg (LW); {DAG1}
PM(I15-8,M15-8)=Ureg (LW); {DAG2}
Ureg=DM(I7-0,M7-0) (LW); {DAG1}
Ureg=PM(I15-8,M15-8) (LW); {DAG2}
DM(I7-0,M7-0)=Data32; {DAG1}
PM(I15-8,M15-8)=Data32; {DAG2}
Table 4-3. Post-Modify Addressing, Modified by 6-Bit Data and
Updating I Register
DM(I7-0,Data6)=Dreg; {DAG1}
PM(I15-8,Data6)=Dreg; {DAG2}
Dreg=DM(I7-0,Data6); {DAG1}
Dreg=PM(I15-8,Data6); {DAG2}
Table 4-4. Pre-Modify Addressing, Modified by M Register (No I Register
Update)
DM(M7-0,I7-0)=Ureg (LW); {DAG1}
PM(M15-8,I15-8)=Ureg (LW); {DAG2}
Ureg=DM(M7-0,I7-0) (LW); {DAG1}
Ureg=PM(M15-8,I15-8) (LW); {DAG2}
ADSP-21160 SHARC DSP Hardware Reference
4-23
DAG Instruction Summary
Table 4-5. Pre-Modify Addressing, Modified by 6-Bit Data (No I Register
Update)
DM(Data6,I7-0)=Dreg; {DAG1}
PM(Data6,I15-8)=Dreg; {DAG2}
Dreg=DM(Data6,I7-0); {DAG1}
Dreg=PM(Data6,I15-8); {DAG2}
Table 4-6. Pre-Modify Addressing, Modified by 32-Bit Data
(No I Register Update)
Ureg=DM(Data32,I7-0) (LW); {DAG1}
Ureg=PM(Data32,I15-8) (LW); {DAG2}
DM(Data32,I7-0)=Ureg (LW); {DAG1}
PM(Data32,I15-8)=Ureg (LW); {DAG2}
Table 4-7. Update (Modify) I Register, Modified by M Register
Modify(I7-0,M7-0); {DAG1}
Modify(I15-8,M15-8); {DAG2}
Table 4-8. Update (Modify) I Register, Modified by 32-Bit Data
Modify(I7-0,Data32); {DAG1}
Modify(I15-8,Data32); {DAG2}
Table 4-9. Bit-Reverse and Update I Register, Modified by 32-Bit Data
Bitrev(I7-0,Data32); {DAG1}
Bitrev(I15-8,Data32); {DAG2}
4-24
ADSP-21160 SHARC DSP Hardware Reference
Memory
5 MEMORY
The DSP contains a large, dual-ported internal memory and provides
access to external memory through the DSP’s external port. This chapter
describes the DSP’s memory and how to use it. For information on connecting and timing accesses to external memory, see “External Memory
Interface” on page 7-3.
Overview
There are 8 Mbits of internal memory space on the DSP. Within this
space, the ADSP-21160 DSP has 4 Mbits of memory, which is divided
into two 2 Mbit blocks: Block 0 and Block 1. The remaining, unpopulated 4 Mbits of the memory space are reserved on the ADSP-21160 DSP.
Table 5-1 shows the maximum number of data or instruction words that
can fit in a 2 Mbit internal memory block.
Table 5-1. Words Per 2 MBit Internal Memory Block
Word Type
Bits Per Word
Maximum Number of Words
Per 2 MBit block
Instruction
48-bits
42.67K words
Long Word Data
64-bits
32K words
Extended Precision Normal Word Data
40-bits
42.67K words
ADSP-21160 SHARC DSP Hardware Reference
5-1
Overview
Table 5-1. Words Per 2 MBit Internal Memory Block
Word Type
Bits Per Word
Maximum Number of Words
Per 2 MBit block
Normal Word Data
32-bits
64K words
Short Word Data
16-bits
128K words
There are 4 Gwords of external memory space that the DSP can address.
External memory connects to the DSP’s external port, which extends the
DSP’s 32-bit address and 64-bit data buses off the DSP. The DSP can
make 64-bit or 32-bit accesses to external memory for instructions or data.
Table 5-2 shows the access types and words for DSP external memory
accesses. The DSP’s DMA controller automatically packs external data
into the appropriate word width during data transfer.
Table 5-2. Internal-to-External Memory Word Transfers1
Word Type
Transfer Type
Instruction
48-bit word MSB justified within 64-bit transfer
Long Word Data
64-bit word in 64-bit transfer
Extended Precision Normal Word Data
40-bit word MSB justified within 64-bit transfer
Normal Word Data
32-bit word in 32-bit transfer
Short Word Data
Not supported
1
For external port word alignment, see Figure 7-1 on page 7-2
Most microprocessors use a single address and data bus for memory access.
This type of memory architecture is called Von Neumann architecture.
But, DSPs require greater data throughput than Von Neumann architecture provides, so many DSPs use memory architectures that have separate
buses for program and data storage. The two buses let the DSP get a data
word and an instruction simultaneously. This type of memory architecture
is called Harvard architecture.
5-2
ADSP-21160 SHARC DSP Hardware Reference
Memory
SHARC DSPs go a step farther by using a Super Harvard architecture.
This architecture has program and data buses, but provides a single, unified address space for program and data storage. While the Data Memory
(DM) bus only carries data, the Program Memory (PM) bus handles
instructions or data, allowing dual-data accesses.
DSP core and I/O processor accesses to internal memory are completely
independent and transparent to one another. Each block of memory can
be accessed by the DSP core and I/O processor in every cycle—no extra
cycles are incurred if the DSP core and the I/O processor access the same
block.
A memory access conflict can occur when the processor core attempts two
accesses to the same internal memory block in the same cycle. When this
conflict happens, an extra cycle is incurred. The DM bus access completes
first and the PM bus access completes in the following (extra) cycle.
During a single-cycle, dual-data access, the processor core uses the independent PM and DM buses to simultaneously access data from both
memory blocks. Though dual-data accesses provide greater data throughput, it is important to note some limitations on how programs may use
them. The limitations on single-cycle, dual-data accesses are:
• The two pieces of data must come from different memory blocks.
If the core tries to access two words from the same memory block
(over the same bus) for a single instruction, an extra cycle is
needed. For more information on how the buses access these
blocks, see “Internal Memory” on page 5-14
• The data access execution may not conflict with an instruction
fetch operation.
If the cache contains the conflicting instruction, the data access
completes in a single-cycle and the sequencer uses the cached
instruction. If the conflicting instruction is not in the cache, an
ADSP-21160 SHARC DSP Hardware Reference
5-3
Overview
extra cycle is needed to complete the data access and cache the
conflicting instruction. For more information, see “Instruction
Cache” on page 3-9.
Efficient memory usage relies on how the program and data are arranged
in memory and varies how the program accesses the data. For more information, see “Arranging Data in Memory” on page 5-84.
As shown in Table 5-1, the DSP has three internal buses connected to its
dual-ported memory, the Program Memory (PM) bus, Data Memory
(DM) bus, and I/O Processor (IO) bus. The PM bus and DM bus share
one memory port and the IO bus connects to the other port. Memory
accesses from the DSP’s core (computational units, data address generators, or program sequencer) use the PM or DM buses, while the I/O
processor uses the IO bus for memory accesses. Using the IO bus, the I/O
processor can provide data transfers between internal memory and the
DSP’s communication ports (link ports, serial ports, and external port)
without hindering the DSP core’s access to memory.
While the DSP’s internal memory is divided into blocks that can be
accessed by DAG1 and DAG2, the DSP’s external memory spaces is
divided into banks, which may be addressed by either data address generator. External memory banks also may be configured for size and access
waitstates. “External Memory” on page 5-20
The DSP core’s PM bus and DM bus and I/O processor’s External Port
(EP) bus can try to access multiprocessor memory space or external memory space in the same cycle. The DSP has a two level arbitration system to
handle this conflicting access. Arbitration stems from a priority convention and the state of the SYSCON register’s EBPRx bits. When arbitrating
between the processor core buses, the DM bus always has priority over the
PM bus. Arbitration between the winning core bus and I/O processor EP
bus depends on the priority set with the EBPRx bits. For more information
on setting this priority, see “External Bus Priority” on page 5-33.
5-4
ADSP-21160 SHARC DSP Hardware Reference
Memory
INTERNAL
(DSP) MEMORY
EXTERNAL
(SYSTEM) MEMORY
DATA
BLOCK 1
(NORMAL WORD 0X50000 - 0X5FFFF)
DATA
ADDRESS
ADDRESS
BLOCK 0
(NORMAL WORD 0X40000 - 0X4FFFF)
ADDRESS
DATA
ADDRESS
BANK 0
(STARTING AT NORMAL WORD
0X800000)
ADDRESS DATA
DATA
ANY TWO PATHS
SIMULTANEOUSLY
EXTERNAL PORT
ADDRESSES AND
DATA FOLLOW
PARALLEL PATHS
PM ADDRESS BUS
PM DATA BUS
18
64
32
64
64
32
64
32
64
PX BUS EXCHANGE REGISTER
64
DM ADDRESS BUS
DM DATA BUS
IO ADDRESS BUS
IO DATA BUS
IO ADDRESS
IO DATA
I/O PROCESSOR
EP
EP
ADDRESS DATA
Figure 5-1. ADSP-21160 Memory and Internal Buses Block Diagram
ADSP-21160 SHARC DSP Hardware Reference
5-5
Overview
Internal Address and Data Buses
Figure 5-1 on page 5-5 also shows that the PM buses, DM buses, and I/O
processor have access to the external bus (pins DATA63-0, ADDR31-0)
through the DSP’s external port. The external port provides access to system (off-DSP) memory and peripherals. This port also lets the DSP access
the internal memory of other DSPs if connected in a multiprocessing
system.
Almost without exception, the DSP’s three buses can access all memory
spaces, supporting all data sizes. There are three restrictions on the access
of buses to memory. The limitations on the PM, DM, and IO buses are as
follows:
• The PM, DM, and IO buses may only make Normal Word
addressing accesses to multiprocessor or external memory. For
more information, see “Multiprocessor Memory” on page 5-17.
• The IO bus may not access the I/O processor’s memory mapped
registers. For more information, see “I/O Processor”
• The IO bus may not use Short word addressing for DMA
operation.
Addresses for the PM and DM buses come from the DSP’s program
sequencer and Data Address Generators (DAGs). The program sequencer
and DAGs supply 32-bit addresses for locations throughout the DSP’s
memory spaces. The DAGs supply addresses for data reads and writes on
both the PM and DM address buses, while the program sequencer uses
only the PM address bus for sequencing execution.
Each DAG is associated with a particular data bus. DAG1 supplies
addresses over the DM bus and DAG2 supplies addresses over the PM
bus. For more information on address generation, see “Program
Sequencer” or “Data Address Generators”.
5-6
ADSP-21160 SHARC DSP Hardware Reference
Memory
Because the DSP’s internal memory is arranged in four 16-bit wide by
32K high columns, memory is addressable in widths that are multiples of
columns up to 64 bits: 1 column = 16-bit words, 2 columns = 32-bit
words, 3 columns = 48- or 40-bit words, and 4 columns = 64-bit words.
For more information on the how the DSP works with memory words, see
“Memory Organization and Word Size” on page 5-22.
The PM and DM data buses are 64 bits wide. Both data buses can handle
Long word (64-bit), Normal word (32-bit), Extended-precision Normal
word (40-bit), and Short word (16-bit) data, but only the PM data bus
carries Instruction words (48-bit).
At the processor’s external port, the DSP multiplexes the three memory
buses—PM, DM, and I/O—to create a single off-chip data bus (DATA
63-0) and address bus (ADDR 31-0).
Internal Data Bus Exchange
The data buses let programs transfer the contents of any register in the
DSP to any other register or to any internal memory location in a single
cycle. As shown in Figure 5-2, the PM Bus Exchange (PX) register permits
data to flow between the PM and DM data buses. The PX register can
work as one 64-bit register or as two 32-bit registers (PX1 and PX2). The
alignment of PX1 and PX2 within PX appears in Figure 5-3.
Combined PX Register
32 31
63
PX2
31
0
PX1
0
31
0
Figure 5-2. PM Bus Exchange (PX, PX1, and PX2) Registers
ADSP-21160 SHARC DSP Hardware Reference
5-7
Overview
PX1, PX2,
and the combined PX register are Universal registers (UREG) and
are accessible for register-to-register or memory-to-register transfers.
register-to-register transfers with data registers are either 40-bit transfers for the combined PX or 32-bit transfers for PX1 or PX2. Figure 5-3
shows the bit alignment for these types of transfers.
PX
Register File Transfer
Register File Transfer
40-bits
39
32-bits
0
40-bits
63
32-bits
0x0
Combined PX
8 7 0
39
24 23
0x0
0
31
0
PX1 or PX2
Figure 5-3. PX, PX1, and PX2 Register-to-Register Transfers
Figure 5-3 shows that:
• During a transfer between PX1 or PX2 and a data register file register
(DREG), the bus transfers the upper 32 bits of the register file and
zero fills the eight LSBs.
• During a transfer between the combined PX register and a register
file register, the bus transfers the upper 40 bits of PX and zero fills
the lower 24 bits.
register-to-memory transfers over the DM data bus are either 64-bit for
the combined PX or 32-bit transfers (on bits 31-0 of the bus) for PX1 or
PX2. Figure 5-4 shows these transfers.
PX
5-8
ADSP-21160 SHARC DSP Hardware Reference
Memory
DM (LW) or PM (LW)
Data Bus Transfer
DM or PM Data Bus Transfer
64-bits
63
31
0
63
31
Combined PX
31
0
32-bits
64-bits
63
32-bits
0x0
0
31
0
PX1 or PX2
Figure 5-4. PX, PX1, PX2 Register-to-Memory Transfers on DM (LW) or
PM (LW) Data Bus
The LW notation in Figure 5-4 draws attention to an important feature of
PX register-to-memory transfers over the PM or DM data bus for the combined PX register. PX transfers to memory are 48-bit (3-column) transfers
on bits 63-16 of the PM or DM data bus, unless forced to be 64-bit
(4-column) transfers with the LW (Long Word) mnemonic.
status of the memory block’s Internal Memory Data Width
 The
(
) setting does not effect this default transfer size for to
IMDWx
PX
internal memory.
Table 5-5 shows the default transfer size between PX and internal memory
over the PM or DM data bus.
ADSP-21160 SHARC DSP Hardware Reference
5-9
Overview
DM and PM Data Bus Transfer (not LW)
48-bits
0x0
31
8 7 0
48-bits
0x0
63
63
31
8 7 0
Combined PX
Figure 5-5. PX Register-to-Memory Transfers on PM Data Bus
(Without LW)
This default 3-column memory access for the PX register over the PM or
DM data bus has a particularly useful application in boot loading. If a
program was loading a series of instructions from external memory locations into internal memory, the program could use the following code:
i8 = start; {sets up dag2 to auto-increment on PM access}
m8 = 1;
i0 = source; {sets up dag1 to auto-increment on DM access}
m0 = 1;
lcntr = 1000, do external_load until lce; {sets up load loop}
px = dm(i0,m0);
external_load: pm(i8,m8) = px;
/* the loop moves from a range of external addresses */
/* (starting at source) to a range of internal addresses */
/* (starting at start) using the PX register. The external */
/* move is a 64-bit (long word, 4-column) access, and the */
/* internal move is a 48-bit (instruction word, 3-column) */
/* access. */
5-10
ADSP-21160 SHARC DSP Hardware Reference
Memory
jump start; {after load is complete, starts program}
Using the PX register for 48-bit moves over the PM data bus can be useful
for programs that handle non-standard loading operations. For information on more typical booting techniques, see “I/O Processor”. For more
information on using DAGs, see “DAG Operations” on page 4-9. For
more information on sequencing (Jump and Do/Until), see “Program
Sequencer”. For more information on how the DSP works with memory
words, see “Memory Organization and Word Size” on page 5-22.
For the previous example, a 64-bit transfer to the Port1 memory location
can be accomplished by adding the LW bit extension to the data move
instruction:
PM(Port1)=PX (LW); {move all 64 bits of PX to Port1}
bit extension does not alter the data alignment for the PX
 The
transfer. Specifying the bit extension on a write to internal
LW
LW
PX
memory generates a write to all four 16-bit columns of the memory
destination, rather than three 16-bit columns.
Some transfers over the DM and PM buses have fixed sizes (64- or 48-bit)
regardless whether the instruction includes the LW (Long Word) mnemonic. PM and DM data bus transfers that have fixed sizes include:
• All transfers between the PX and external memory are 64-bits
transfers.
• All transfers between the PX register and the I/O processor EPBx
registers are 64-bit transfers.
• All transfers between the PX register and the I/O processor LBUFx
registers are 48-bit transfers (most significant 48-bits of PX).
ADSP-21160 SHARC DSP Hardware Reference
5-11
ADSP-21160 DSP Memory Map
• All transfers between the PX register (or any other internal register/memory) and any I/O processor register (other than the EPBx or
LBUFx) are 32-bit transfers (least significant 32-bits of PX).
• All transfers between the PX register and data registers (R0-R15 or
S0-S15) are 40-bit transfers.
There is no implicit move when the combined PX register is used in SIMD
mode. For example in SIMD mode, the following moves could occur:
PX1 = R0; {R0 32-bit explicit move to PX1,
and R1 32-bit implicit move to PX2}
PX = R0; {R0 40-bit explicit move to PX,
but no implicit move for R1}
ADSP-21160 DSP Memory Map
The ADSP-21160 DSP’s memory map appears in Figure 5-6 and has
three memory spaces: internal memory space, multiprocessor memory
space, and external memory space.
5-12
ADSP-21160 SHARC DSP Hardware Reference
Memory
0XFFFF FFFF
UNBANKED
MS3
BANK 3
EXTERNAL
MEMORY
MS2
BANK 2
MS1
BANK 1
MS0
BANK 0
0X0080 0000
0X007F FFFF
0X0007 FFFF
RESERVED
(ALIASED)
INTERNAL
MEMORY
SPACE
BLOCK 1
MULTIPROCESSOR
MEMORY
SPACE
BLOCK 0
RESERVED
(I/O)
0X0010 0000
0X0000 0000
Figure 5-6. ADSP-21160 Memory Map
ADSP-21160 SHARC DSP Hardware Reference
5-13
ADSP-21160 DSP Memory Map
These spaces have the following definitions:
• Internal memory space. This space ranges from address
0x0000 0000 through 0x0007 FFFF (Normal word). Internal memory space refers to the DSP’s on-chip SRAM and memory mapped
registers.
• Multiprocessor memory space. This space ranges from address
0x0010 0000 through 0x007F FFFF (Normal word). Multiprocessor
memory space refers to the internal memory space of a group of
DSPs that are connected in a multiprocessor system.
• External memory space. This space ranges from address
0x0080 0000 through 0xFFFF FFFF (Normal word). External memory space refers to the off-chip memory or memory mapped
peripherals that are attached to the DSP’s external address
(ADDR31-0) and data (DATA63-0) buses.
The address ranges of the three memory spaces correspond to fields within
the 32-bit address on the PM, DM, or external address buses. For definitions of these bit fields, see “Program Counter Register (PC)” on
page A-28.
Internal Memory
The ADSP-21160 DSP’s internal memory space appears in Figure 5-7.
This memory space has four address regions.
• I/O processor memory mapped registers. This region ranges from
address 0x0000 0000 through 0x0000 00FF (Long word)
• Reserved (I/O) memory. This region ranges from address
0x0000 0100 through 0x0001 FFFF (Long word). These addresses
are not accessible.
5-14
ADSP-21160 SHARC DSP Hardware Reference
Memory
INTERNAL
MEMORY
LONG
WORD
(64-BIT)
0X003 FFFF
NORMAL
WORD
(32-BIT)
EXT. PREC. NORMAL
WORD (40-BIT)
OR
INSTRUCTION
WORD (48-BIT)
0X007 FFFF
0X007 FFFF
SHORT
WORD
(16-BIT)
0X00F FFFF
MISSING
0X007 AAAA
RESERVED
(ALIASED)
0X007 0000
MISSING
0X006 AAAA
0X003 0000
0X002 FFFF
0X006 0000
0X005 FFFF
0X006 0000
MISSING
0X00C 0000
0X00B FFFF
0X005 AAAA
BLOCK 1
0X002 8000
0X005 0000
0X002 7FFF
0X004 FFFF
0X005 0000
MISSING
0X00A 0000
0X009 FFFF
0X004 AAAA
BLOCK 0
0X002 0000
0X001 FFFF
0X004 0000
0X004 0000
0X008 0000
THE DIFFERENT TYPES OF ADDRESSING
ADDRESS THE SAME PHYSICAL MEMORY, BUT
USE DIFFERENT LEVELS OF GRANULARITY
(WORD SIZE).
RESERVED
(I/O)
0X000 0100
0X000 0000
I/O PROCESSOR REGISTERS
Figure 5-7. ADSP-21160 Internal Memory Space
ADSP-21160 SHARC DSP Hardware Reference
5-15
ADSP-21160 DSP Memory Map
• Block 0 memory. This region ranges from address
through 0x0002 7FFF (Long word).
0x0002 0000
• Block 1 memory. This region ranges from address 0x0002
through 0x0002 FFFF (Long word).
8000
• Reserved (aliased) memory. This region consists of a Block 0
aliased region from 0x0003 0000 through 0x0003 7FFF and a Block
1 aliased region from 0x0003 08000 through 0x0003 FFF (Long
word). Accesses to these two regions result in accesses to the corresponding addresses in Block 0 and Block 1.
The I/O processor’s memory-mapped registers control the system configuration of the DSP and I/O operations. For more information, see “I/O
Processor” These registers occupy consecutive 32-bit locations in this
region.
If a program uses Long word addressing (forced with the LW mnemonic) to
accesses this region, the access is only to the addressed 32-bit register,
rather than accessing two adjacent I/O processor registers. The register
contents are transferred on bits 31-0 of the data bus. There are a couple of
exceptions to this one-at-a-time I/O processor register access rule:
• Long word accesses to the external port data buffer locations (EPBx)
in SIMD mode access two adjacent 32-bit I/O registers.
• Long word accesses to external port buffer (EPBx) or link port buffer (LBUFx) locations using the PX register access two adjacent
32-bit I/O registers.
As shown in Figure 5-7, the DSP can address memory in the Block 0,
Block 1, or the Block 0 and 1 aliased regions using Long word, Normal
word, or Short word addressing. The DSP interprets the addressing mode
from the address range for the access. Though there are multiple addressing modes for each memory region, these different modes are addressing
the same physical memory. For example, the Long word address
0x0002 0000 corresponds to the same locations as Normal word addresses
5-16
ADSP-21160 SHARC DSP Hardware Reference
Memory
and 0x0004 0001 and corresponds to the same locations as
Short word addresses 0x0008 0000, 0x0008 0001, 0x0008 0002, and
0x0008 0003.
0x0004 0000
Figure 5-7 also shows that there are gaps in the DSP’s memory map when
using Normal word addressing for 48-bit (instruction word) or 40-bit
(extended precision Normal word) accesses. These gaps of missing
addresses stem from the arrangement of this 3-column data in memory.
For more information, see “Memory Organization and Word Size” on
page 5-22.
Multiprocessor Memory
The ADSP-21160’s multiprocessor memory space appears in Figure 5-8.
This memory space has seven address regions that correspond to the internal memory of the DSPs in a multiprocessing system. Each of the
processors in such a system has a processor ID, which is set with the DSP’s
ID2-0 pins. The address regions by processor ID are:
• Internal memory of DSP with ID=001. This region ranges from
address 0x0010 0000 through 0x0017 FFFF.
• Internal memory of DSP with ID=010. This region ranges from
address 0x0020 0000 through 0x0027 FFFF.
• Internal memory of DSP with ID=011. This region ranges from
address 0x0030 0000 through 0x0037 FFFF.
• Internal memory of DSP with ID=100. This region ranges from
address 0x0040 0000 through 0x0047 FFFF.
• Internal memory of DSP with ID=101. This region ranges from
address 0x0050 0000 through 0x0057 FFFF.
ADSP-21160 SHARC DSP Hardware Reference
5-17
ADSP-21160 DSP Memory Map
0X07F FFFF
0X0067 FFFF
MULTIPROCESSOR
MEMORY
SPACE
0X0057 FFFF
0X0047 FFFF
0X0037 FFFF
0X0027 FFFF
0X0017 FFFF
BROADCAST
WRITE TO
ALL DSPS
(ID = 111)
RESERVED
(ALIASED)
BLOCK 1
BLOCK 1
BLOCK 1
BLOCK 1
BLOCK 1
BLOCK 1
BLOCK 1
BLOCK 0
BLOCK 0
BLOCK 0
BLOCK 0
BLOCK 0
BLOCK 0
BLOCK 0
INTERNAL
MEMORY
(ID = 110)
INTERNAL
MEMORY
(ID = 101)
INTERNAL
MEMORY
(ID = 100)
INTERNAL
MEMORY
(ID = 011)
INTERNAL
MEMORY
(ID = 010)
INTERNAL
MEMORY
(ID = 001)
RESERVED
(I/O)
0X0000 0000
0X0060 0000
0X0050 0000
0X0040 0000
0X030 0000
0X0020 0000
0X0010 0000
ALWAYS ADDRESSED
AS NORMAL WORD
Figure 5-8. ADSP-21160 Multiprocessor Memory Space
5-18
ADSP-21160 SHARC DSP Hardware Reference
Memory
• Internal memory of DSP with ID=110. This region ranges from
address 0x0060 0000 through 0x0067 FFFF.
• Broadcast write to internal memory of all DSPs (ID=111). This
region ranges from address 0x0070 0000 through 0x0077 FFFF.
It is important to note that programs may only use Normal word addressing in multiprocessor memory space. Long or Short word writes may
corrupt valid data, and Long or Short word reads return invalid data.
The address range of the access determines which DSP’s internal memory
is the multiprocessor memory access source or destination. Broadcast
writes (writes in the range 0x0070 0000 through 0x007F FFFF) access the
memory of all DSPs in the multiprocessing system.
Instead of using its own internal memory address range, a DSP can access
its memory through the DSP’s corresponding address range in multiprocessor memory space. In this case, the DSP reads or writes to its own
internal memory and does not make an access on the external system bus.
Note that such self-accesses through multiprocessor memory space may
only be accomplished with processor-core-generated addresses, not I/O
processor-generated addresses.
For more information on memory accesses in multiprocessor systems, see
“External Port”.
ADSP-21160 SHARC DSP Hardware Reference
5-19
ADSP-21160 DSP Memory Map
External Memory
The ADSP-21160’s external memory space appears in Figure 5-9.
0XFFFF FFFF
ALWAYS ADDRESSED
AS NORMAL WORD
UNBANKED
BANK 3
MS3
EPROM
(BOOT)
MEMORY
EXTERNAL
MEMORY
BANK 2
MS2
BANK 1
MS1
BANK 0
MS0
OPTIONAL
DRAM PAGES
MAP TO
BANK 0, AND
REQUIRE AN
EXTERNAL DRAM
CONTROLLER
BMS
DRAM
PAGES
0X0080 0000
Figure 5-9. ADSP-21160 External Memory Space
5-20
ADSP-21160 SHARC DSP Hardware Reference
Memory
The DSP accesses external memory space through the external port, which
multiplexes the processor core’s PM and DM buses and the I/O processor’s EP bus. To address this space, the DSP’s DAG1, DAG2, and I/O
processor generate 32-bit addresses over the DM, PM, and EP address
buses, allowing the DSP access to the complete 4 Gword memory map.
But, the program sequencer only generates 24-bit addresses over the PM
bus, limiting sequencing to the low 12 Mwords of the memory map.
As shown in Figure 5-9, the external memory space has five regions: 4
banks (Bank 0-3) and an unbanked region. The DSP controls access to the
banked regions with memory select lines (MS3-0) in addition to the memory address. Access to the unbanked region is controlled only by the
memory address. Each region of external memory may be configured for
address range and waitstates. For more information on configuring external memory banks, see “Setting Data Access Modes” on page 5-28.
For more information on accessing external memory, see “External Port”.
The external memory space can also accommodate an optional boot memory EPROM and optional paged DRAM. For more information, see
“Using Boot Memory” on page 5-29 and “External (Bank 0) DRAM Page
Size” on page 5-38.
Shadow Write FIFO
Because the DSP’s internal memory operates at high speeds, writes to the
memory do not go directly into the memory array, but rather to a
two-deep FIFO called the shadow write FIFO. This FIFO uses a non-read
cycle (either a write cycle, or a cycle in which there is no access of internal
memory) to load data from the FIFO into internal memory. When an
internal memory write cycle occurs, the FIFO loads any data from a previous write into memory and accepts new data. FIFO operation is normally
transparent, but there is one case in which programs need to intervene in
the operation of the shadow write FIFO: mixing 48-bit and 32-bit word
accesses to the same locations in memory.
ADSP-21160 SHARC DSP Hardware Reference
5-21
ADSP-21160 DSP Memory Map
The shadow FIFO cannot differentiate between the mapping of 48-bit
words and the mapping of 32-bit words. Examples of these mappings
appear in Figure 5-10 on page 5-23, Figure 5-11 on page 5-24,
Figure 5-12 on page 5-25, and Figure 5-13 on page 5-26. If a program
writes a 48-bit word to memory and then tries to read the data with a 16-,
32-, or 64-bit word access or writes a 16-, 32-, 64-bit word to memory
and tries to read the data with a 48-bit access, the shadow FIFO does not
intercept the read and returns incorrect data.
If a program must mix 48-bit or 40-bit accesses and 16-, 32-, or 64-bit
accesses to the same locations, the program must ensure that the FIFO is
flushed before attempting to read the data. The program flushes the FIFO
by performing two dummy writes or executing two instructions that do
not access the internal memory. These operations force the FIFO to automatically use the non-access cycles to push the write data.
Memory Organization and Word Size
The DSP’s internal memory is organized as four 16-bit wide by 32K high
columns. These columns of memory are addressable as a variety of word
sizes:
• 64-bit Long word data (4-columns)
• 48-bit instruction words or 40-bit extended precision Normal
word data (3-columns)
• 32-bit Normal word data (2-columns)
• 16-bit Short word data (1-column)
precision Normal word data is left-justified within a
 Extended
3-column location, using bits 47-8 of the location.
5-22
ADSP-21160 SHARC DSP Hardware Reference
Memory
Placing 32-Bit Words and 48-Bit Words
When the processor core or I/O processor addresses memory, the word
width of the access determines which columns within the memory are
accessed. For instruction words (48-bit) or extended precision Normal
word data (40-bit), the word width is 48 bits, and the access selects from
the memory’s 16-bit columns in groups of three. Because these sets of 3
column accesses are packed in a 4 column matrix, there are four rotations
of the columns for storing 48-bit data. The 3-column word rotations
within the 4-column matrix appear in Table 5-1 and Figure 5-10.
Rotation 3
Rotation 2
Addresses
Rotation 2
Rotation 1
0
Rotation 1
Rotation 0
15 0
15 0
15 0
15
Column 0
Column 3
Column 2
Column 1
Figure 5-10. 48-bit Word Rotations
For Long word (64-bit), Normal word (32-bit), and Short word (16-bit)
memory accesses, The DSP selects from fixed columns in memory. No
rotations of words within columns occur for these data types.
Figure 5-7 on page 5-15 shows the memory ranges for each data size in the
DSP’s internal memory. Figure 5-10 describes 48-bit word rotations.
Mixing 32-Bit and 48-Bit Words
The DSP’s memory organization lets programs freely place memory words
of all sizes (see “Memory Organization and Word Size” on page 5-22)
ADSP-21160 SHARC DSP Hardware Reference
5-23
ADSP-21160 DSP Memory Map
with few restrictions (see “Restrictions on Mixing 32-Bit and 48-Bit
Words” on page 5-26). This memory organization also lets programs mix
(place in adjacent addresses) words of all sizes. This section discusses how
to mix odd (3-column) and even (4-column) data words in the DSP’s
memory.
Transition boundaries between 48-bit (3-column) data and any other data
size, can occur at any 64-bit address boundary within either internal memory block. Depending on the ending address of the 48-bit words, there are
zero, one, or two empty locations at the transition between the 48-bit
(3-column) words and the 64-bit (4-column) words. These empty locations result from the column rotation for storing 48-bit words. The three
possible transition arrangements appear in Figure 5-11, Figure 5-12 on
page 5-25, and Figure 5-13 on page 5-26.
Transitioning from 48-bit to 32-bit
data with zero empty locations:
(48-bit word top address) MOD 4 = 3
32-bit word 3
32-bit word 2
32-bit word 1
32-bit word 0
48-bit word top
Addresses
48-bit word top-1
48-bit word top-2
0
48-bit word top-1
48-bit word top-2
48-bit word top-3
15 0
15 0
15 0
15
Column 0
Column 3
Column 2
Column 1
Figure 5-11. Mixed Instructions and Data With No Unused Locations
5-24
ADSP-21160 SHARC DSP Hardware Reference
Memory
Transitioning from 48-bit to 32-bit
data with one empty locations:
(48-bit word top address) MOD 4 = 0
32-bit word 3
32-bit word 2
32-bit word 1
32-bit word 0
Empty
48-bit word top
Addresses
48-bit word top-1
48-bit word top-2
0
48-bit word top-2
48-bit word top-3
15 0
15 0
15 0
15
Column 0
Column 3
Column 2
Column 1
Figure 5-12. Mixed Instructions and Data With One Unused Location
ADSP-21160 SHARC DSP Hardware Reference
5-25
ADSP-21160 DSP Memory Map
Transitioning from 48-bit to 32-bit
data with two empty locations:
(48-bit word top address) MOD 4 = 1
32-bit word 3
32-bit word 2
32-bit word 1
32-bit word 0
Empty
48-bit word top
Addresses
48-bit word top
Empty
48-bit word top-1
48-bit word top-2
0
48-bit word top-3
15 0
15 0
15 0
15
Column 0
Column 3
Column 2
Column 1
Figure 5-13. Mixed Instructions and Data With Two Unused Locations
Restrictions on Mixing 32-Bit and 48-Bit Words
There are some restrictions that stem from the memory column rotations
for 3-column data (48- or 40-bit words) and relate to the way that 3-column data can mix with 4-column data (32-bit words) in memory. These
restrictions apply to mixing 48- and 32-bit words, because the DSP uses a
Normal word address to access both of these types of data even though
48-bit data maps onto 3-columns of memory and 32-bit data maps onto
2-columns of memory.
When a system has a range of 3-column (48-bit) words followed by a
range of 2-column (32-bit) words, there is often a gap of empty 16-bit
locations between the two address ranges. The size of the address gap varies with the ending address of the range of 48-bit words. Because the
addresses within the gap alias to both 48- and 32-bit words, a 48-bit write
5-26
ADSP-21160 SHARC DSP Hardware Reference
Memory
into the gap corrupts 32-bit locations, and a 32-bit write into the gap corrupts 48-bit locations. The locations within the gap are only accessible
with Short word (16-bit) accesses.
Calculating the starting address for 4-column data that minimizes the gap
after 3-column data is a useful calculation for programs that are mixing 3and 4-column data.
Given the last address of the 3-column (48-bit) data, the starting address
of the 32-bit range that most efficiently uses memory can be determined
as follows:
• n is the number of contiguous 48-bit words the system has allocated in the internal memory block (n < 87,381)
• B is the base Normal word address of the internal memory block; if
{0 < n < 43,691} then B = 0x40000 else B = 0x50000
• m is the first 32-bit Normal word address to use after the end of
48-bit words
• m = B + 2 [(n MOD 43,690) – TRUNC((n MOD 43,690)
/ 4)]
Another useful calculation for programs that are mixing 3- and 4-column
data is to calculate the amount of 3-column data that minimizes the gap
before starting 4-column data. Given the starting address of the 4-column
(32-bit) data, the number of 48-bit words to allocate that most efficiently
uses memory can be determined as follows:
• m is the first 32-bit Normal word address after the end of 48-bit
words (0x3FFFF < m < 0x60000)
• B is the base Normal word address of the internal memory block; if
{0x3FFFF < m < 0x50000} then B = 0x40000 else B = 0x50000
ADSP-21160 SHARC DSP Hardware Reference
5-27
Setting Data Access Modes
• W is the number of offset words; if {B = 0x50000} then
W = 43,690 else W = 0
• n is the number of contiguous 48-bit words the system should allocate in the internal memory block
• n = TRUNC{4[(m - B) / 2] / 3]} + W
Setting Data Access Modes
The SYSCON, MODE1, MODE2, and WAIT registers control the operating mode
of the DSP’s memory. Table A-17 on page A-45 lists all the bits in
SYSCON, Table A-2 on page A-3 lists all the bits in MODE1, Table A-3 on
page A-7 lists all the bits in MODE2, and Table A-18 on page A-48 lists all
the bits in WAIT. The following bits in SYSCON, MODE1, MODE2, and WAIT registers control memory access modes:
• Boot Select Override. SYSCON Bit 1 (BSO) overrides normal usage of
MSx chip select lines in favor of the BMS select line for access to boot
memory instead of external memory (if 1) or allows normal access
to external memory with the MSx chip select lines (if 0).
• Internal Interrupt Vector Table. SYSCON Bit 2 (IIVT) forces placement of the interrupt vector table at address 0x0004 0000
regardless of booting mode (if 1) or allows placement of the interrupt vector table as selected by the booting mode (if 0).
• Internal Memory Block Data Width. SYSCON Bit 9 (IMDW0) and
Bit 10 (IMDW1) select the normal word data access size for internal
memory Block 0 and Block1. A block’s normal word access size is
fixed as 2-column (if IMDWx=0) or 3-column (if IMDWx=1).
• Memory Bank Size. SYSCON Bits 15-12 (MSIZE). This bit field
selects the size of the four external memory banks (Bank 3-0). The
external memory that is not allotted to a bank is part of the
Unbanked external memory region.
5-28
ADSP-21160 SHARC DSP Hardware Reference
Memory
• External Bus Priority. SYSCON Bits 18-17 (EBPRx). This bit field
selects the priority for the I/O processor’s EP bus when arbitrating
access to the DSP’s external port.
• Secondary processor element (PEy). MODE1 Bit 21 (PEYEN) enables
computations in PEy—SIMD mode—(if 1) or disables PEy—
SISD mode—(if 0).
• Broadcast register loads. MODE1 Bit 22 (BDCST9) and Bit 23
(BDCST1) enable broadcast register loads for memory transfers
indexed with I1 (if BDCST1=1) or indexed with I9 (if BDCST9=1).
• Illegal IOP register access enable. MODE2 Bit 20 (IIRAE) enables
detection of I/O processor register access (if 1) or disables detection
(if 0).
• Unaligned 64-bit memory access enable. MODE2 Bit 21 (U64MAE)
enables detection of uneven address memory access (if 1) or disables detection (if 0).
• External bank X access mode. WAIT Bits 1-0 (EB0AM), Bits 6-5
(EB1AM), Bits 11-10 (EB2AM), Bits 16-15 (EB3AM), and Bits 21-20
(UBAM). These bit fields select the access modes for the external
memory banks.
• External bank X waitstates. WAIT Bits 4-2 (EB0WS), Bits 9-7
(EB1WS), Bits 14-12 (EB2WS), Bits 19-17 (EB3WS), and Bits 24-22
(UBWS). These bit fields select the waitstates for the external memory banks.
• External bank 0 DRAM page size. WAIT Bits 27-25 (PAGSZ). This
bit field selects the page size for DRAM (allowed in Bank 0 only).
Using Boot Memory
As shown in Figure 5-9 on page 5-20, the DSP supports an external boot
EPROM mapped to external memory and selected with the BMS pin. The
ADSP-21160 SHARC DSP Hardware Reference
5-29
Setting Data Access Modes
boot EPROM provides one of the methods for automatically loading a
program in to the internal memory of the DSP after power-up or after a
software reset. This process is called booting. For information on boot
options and the booting process, see “Bootloading Through The External
Port” on page 6-76 or “Bootloading Through The Link Port” on
page 6-87. For information on systems with a boot EPROM, see “Booting
Single and Multiple Processors” on page 11-48.
Reading from Boot Memory
When the DSP boots from an EPROM, the DSP’s I/O processor only
loads 256 instructions automatically from EPROM. If the whole program
must be loaded into internal memory from the EPROM, the DSP must
gain access to the boot EPROM after the I/O processor completes the
automatic boot process. To manage continuing access to boot memory,
the DSP uses the Boot Select Override (BSO) bit in the SYSCON register.
Setting (=1) the BSO bit overrides the external memory selects and asserts
the DSP’s BMS pin for an external memory DMA transfer. For accessing
boot memory, the program first sets the BSO bit in SYSCON then sets up an
external port DMA channel to read the EPROM’s contents. The program
must unmask the DMA channel’s interrupt in the IMASK register; if using
external port DMA buffer zero (EP0I), the program could enable this
interrupt by initializing IMASK to 0x00008003. For more information on
external port DMA, see “Setting I/O Processor—EPort Modes” on
page 6-14.
While a program may use any external port DMA channel for accessing
boot memory, it is important to note that only DMA channel 10 has a
special packing mode for boot memory reads. By using DMA channel 10
to complete initial program loading, a program can take advantage of this
special packing mode.
When a program sets BSO, the DSP ignores the DMA channel’s packing
mode (PMODE) bits and forces 8-to-48 bit packing for reads. This special
8-bit packing mode is only available on DMA channel 10 during EPROM
5-30
ADSP-21160 SHARC DSP Hardware Reference
Memory
booting or on DMA reads when BSO is set. While one of the external port
DMA channels is making a DMA access to boot memory with the BSO bit
set, none of the other three channels may make a DMA access to external
(not boot) memory.
Only external port DMA transfers assert BMS when BSO is set; processor
core accesses to external memory always use the MSx pins. Because the processor core only accesses external (not boot) memory, programs can access
external memory in between DMA accesses to boot memory.
Writing to Boot Memory
In systems using write-able EEPROM or FLASH memory for boot memory, programs can write new data to the DSP’s boot memory using the
boot select override (BSO) pin. As described in “Reading from Boot Memory” on page 5-30, setting (=1) the BSO bit overrides the external memory
selects and asserts the DSP’s BMS pin for an external memory DMA
transfer.
To write to boot memory with the BMS asserted, programs must use DMA
channels 11, 12 or 13, but not DMA channel 10. With the BSO bit set,
programs should only use DMA channel 10 for reads.
When BSO is set, programs can use DMA channels 11-13 with any settings
in channel’s the DMACx register, any packing mode, and any data or
instruction.
boot memory is 8-bits wide and no 8-bit packing mode is
 Because
available for these writes, programs must use the shifter to place
data in the correct location for each write.
Internal Interrupt Vector Table
The default location of the ADSP-21160’s interrupt vector table depends
on the DSP’s booting mode. When the processor boots from an external
source (EPROM, host port, or link port booting), the vector table starts at
ADSP-21160 SHARC DSP Hardware Reference
5-31
Setting Data Access Modes
address 0x0004 0000 (Normal word). When the processor is in “no boot”
mode (runs from external memory location 0x0080 0000 without loading), the interrupt vector table starts at address 0x0080 0000.
The Internal Interrupt Vector Table (IIVT) bit in the SYSCON register overrides the default placement of the vector table. If IIVT is set (=1), the
interrupt table starts at address 0x0004 0000 (internal memory) regardless
of the booting mode.
Internal Memory Block Data Width
The DSP’s internal memory blocks use Normal word addressing to access
either single-precision 32-bit data or extended-precision 40-bit data. Programs select the data width independently for each internal memory block
using the Internal Memory Data Width (IMDW0 and IMDW1) bits in the
SYSCON register. If a block’s IMDWx bit is cleared (=0), Normal word
addressed accesses to the block access Normal word (32-bit) data. If a
block’s IMDWx bit is set (=1), Normal word addressed accesses to the block
access extended precision Normal word (40-bit) data. Reading or writing
40-bit data using a Normal word access to a memory block whose IMDWx
bit is cleared (=0) has the following results.
• If a program tries to write 40-bit data (for example, a data register-to-memory transfer), the transfer truncates the lower 8-bits
from the register; only writing 32 bits.
• If a program tries to read 40-bit data (for example, a memory-to-data register transfer), the transfer zero-fills the lower 8 bits
of the register; only reading 32 bits.
The Program Memory Bus Exchange (PX) register is the only exception to
these transfer rules—all loads/stores of the PX register are performed as
48-bit accesses unless forced to 64-bit access with the LW mnemonic. If
any 40-bit data must be stored in a memory block configured for 32-bit
words, the program should use the PX register to access the 40-bit data in
48-bit words. Programs should take care not to corrupt any 32-bit data
5-32
ADSP-21160 SHARC DSP Hardware Reference
Memory
with this type of access. “Restrictions on Mixing 32-Bit and 48-Bit
Words” on page 5-26
Long word ( ) mnemonic only effects Normal word address
 The
accesses and overrides all other factors (SIMD, IMDWx).
LW
Memory Bank Size
The DSP’s external memory space has four banks of equal, programmable
size. The remaining area of external memory that is not assigned to a bank
is called unbanked. Mapping peripherals into different banks lets systems
accommodate I/O devices with different timing requirements, because the
banked and unbanked regions have associated waitstate and access mode
settings. For more information, see “External Bank X Access Mode” on
page 5-36 and “External Bank X Waitstates” on page 5-37.
As shown in Figure 5-9 on page 5-20, Bank 0 starts at address
0x0080 0000 in external memory, and the Banks 1, 2, 3, and unbanked
regions follow. Whenever the DSP generates an address that is located
within one of the four banks, the DSP asserts the corresponding memory
select line (MS3-0).
The size of the memory banks ranges from 8 Kwords to 256 Mwords and
is always a power of two. The Memory Size (MSIZE) field of the SYSCON
register selects the memory banks size. The value in MSIZE is:
• MSIZE = log2 (desired bank size in words) - 13
External Bus Priority
The DSP’s internal bus architecture lets the PM bus, DM bus, and I/O
processor’s EP bus try to access multiprocessor memory space or external
memory space in the same cycle. This contending access produces a conflict that the DSP resolves with a two level arbitration policy. The
processor core’s DM bus always has priority over the PM bus. External
Bus Priority (EBPRx) bits in the SYSCON register control the further
ADSP-21160 SHARC DSP Hardware Reference
5-33
Setting Data Access Modes
arbitration between the winning core bus and the I/O processor’s EP bus.
The EBPRx field assigns priority as follows:
• If EBPR is 00, priority rotates between core and I/O processor buses.
• If EBPR is 01, the winning core bus has priority over the I/O processor bus.
• If EBPR is 10, the I/O processor bus has priority over the winning
core bus.
Secondary Processor Element (PEy)
When the PEYEN bit in the MODE1 register is set (=1), the DSP is in Single-Instruction, Multiple-Data (SIMD) mode. In SIMD mode, many data
access operations differ from the DSP’s default Single-Instruction, Single-Data (SISD) mode. These differences relate to doubling the amount of
data transferred for each data access.
Accesses in SIMD mode transfer both an explicit (named) location and an
implicit (un-named, complementary) location. The explicit transfers is a
data transfers between the explicit register and the explicit address, and
the implicit transfer is between the implicit register and the implicit
address.
For information on complementary (implicit) registers in SIMD mode
accesses, see “Secondary Processing Element (PEy)” on page 2-35. For
more information on complementary (implicit) memory locations in
SIMD mode accesses, see “Accessing Memory” on page 5-39.
Broadcast Register Loads
The DSP’s BDCST1 and BDCST9 bits in the MODE1 register control broadcast
register loading. When broadcast loading is enabled, the DSP writes to
complementary registers or complementary register pairs in each processing element on writes that are indexed with DAG1 register I1 (if
5-34
ADSP-21160 SHARC DSP Hardware Reference
Memory
=1) or DAG2 register I9 (if BDCST9 =1). Broadcast load accesses are
similar to SIMD mode accesses in that the DSP transfers both an explicit
(named) location and an implicit (un-named, complementary) location,
but broadcast loading only influences writes to registers and write identical data to these registers. Broadcast mode is independent of SIMD mode.
BDCST1
Table 5-3 on page 5-35 shows examples of explicit and implicit effects of
broadcast register loads to both processing elements. Note that broadcast
loading only effects loads of data registers (register file); broadcast loading
does not effect register stores or loads to other system registers. And,
broadcast loads only work on register loads; broadcast loading cannot be
used for memory writes. For more information on broadcast loading, see
“Accessing Memory” on page 5-39.
Table 5-3. Register Load Dual PE Broadcast Operation
Instruction
(Explicit, PEx Operation)
(Implicit, PEy operation)
Rx = dm(i1,ma);
Rx = pm(i9,mb);
Rx = dm(i1,ma), Ry = pm(i9,mb);
Sx = dm(i1,ma);
Sx = pm(i9,mb);
Sx = dm(i1,ma), Sy = pm(i9,mb);
Illegal I/O Processor Register Access
The DSP monitors for I/O processor register access if the Illegal I/O processor Register Access (IIRAE) bit in the MODE2 register is set (=1). When
detected, this condition is an input that can cause an Illegal Input Condition Detected (IICDI) interrupt if the interrupt is enabled in the IMASK
register.
The I/O processor’s DMA controller cannot generate the
 interrupt.
Only master (not slave) I/O register accesses are detectIICDI
able. For more information, see “Mode Control 2 Register
(MODE2)” on page A-6.
ADSP-21160 SHARC DSP Hardware Reference
5-35
Setting Data Access Modes
Unaligned 64-bit Memory Access
The DSP monitors for unaligned 64-bit memory accesses if the Unaligned
64-bit Memory Accesses (U64MAE) bit in the MODE2 register is set (=1). An
unaligned access is an odd numbered address Normal word access that is
forced to 64-bit with the LW mnemonic. When detected, this condition is
an input that can cause an Illegal Input Condition Detected (IICDI) interrupt if the interrupt is enabled in the IMASK register. For more
information, see “Mode Control 2 Register (MODE2)” on page A-6.
External Bank X Access Mode
The DSP has four modes for accessing external memory space. The External Bank Access Mode (EBxAM) fields in the WAIT register select how the
DSP uses waitstates and the acknowledge (ACK) pin to access each external
memory bank and unbanked region. The external bank access modes
appear in Table 5-4.
Table 5-4. External Bank Access Mode
EBxAM Field
External Bank Access Mode
00
Asynchronous—DSP RDH/L and WRH/L strobes change before CLKOUT’s
edge—accesses use the waitstate count setting from EBxWS and require external acknowledge (ACK), allowing a deasserted ACK to extend the access time.
01
Synchronous—DSP RDH/L and WRH/L strobes change on CLKOUT’s edge—
reads use the waitstate count setting from EBxWS (minimum EBxWS=001)
and require external acknowledge (ACK), allowing a deasserted ACK to extend
the read access time; writes are 0-wait state.
10
Synchronous—DSP RDH/L and WRH/L strobes change on CLKOUT’s edge—
reads use the waitstate count setting from EBxWS (minimum EBxWS=001)
and require external acknowledge (ACK), allowing a deasserted ACK to extend
the read access time; writes are 1-wait state.
11
Reserved
5-36
ADSP-21160 SHARC DSP Hardware Reference
Memory
External Bank X Waitstates
The DSP applies waitstates to each external memory access depending on
the bank’s external memory access mode (EBxAM). The External Bank
Waitstate (EBxWS) field in the WAIT register sets the number of waitstates
for each bank as shown in Table 5-5.
Table 5-5. External Bank Waitstates
EBxWS
# of Waitstates
Hold Time Cycle?1
000
0
no
001
1
no
010
2
yes
011
3
yes
100
4
yes
101
5
yes
110
6
yes
111
7
yes
1
Hold Cycle applies to asynchronous mode only.
Table 5-5 lists the hold time settings that EBxWS associates with external
memory accesses. A hold time cycle is an inactive bus cycle that the DSP
inserts automatically at the end of a read or write, allowing a longer hold
time for address and data. The address and data remain unchanged and are
driven for one cycle after the DSP deasserts the read or write strobes.
DSP applies hold time cycles regardless of the external bank
 The
access mode (
). For example, the asynchronous (ACK plus
EBxAM
waitstate mode) could also have an associated hold time cycle.
ADSP-21160 SHARC DSP Hardware Reference
5-37
Using Memory Access Status
External (Bank 0) DRAM Page Size
As shown in Figure 5-8 on page 5-18, the DSP supports a region of paged
DRAM mapped to the Bank 0 region of external memory. Systems placing
DRAM in this region require an external DRAM controller to manage
page access to the DRAM. For more information, see “DRAM Page
Boundary Detection” on page 7-15.
To support DRAM accesses in this region, the DSP detects page boundary
crossings and outputs the PAGE signal to the system’s DRAM controller.
The page boundaries depend on the type of DRAM in the system. For
correct operation, programs must configure the page size in the Page Size
(PAGESZ) field of the WAIT register. Table 5-6 shows the available PAGESZ
settings.
Table 5-6. External DRAM (Bank 0) Page Size
PAGSZ Field
DRAM Page Size
000
256 words
001
512 words
010
1024 words (1K)
011
2048 words (2K)
100
4096 words (4K)
101
8192 words (8K)
110
16384 words (16K)
111
32768 words (32K)
Using Memory Access Status
As described in “Illegal I/O Processor Register Access” on page 5-35 and
“Unaligned 64-bit Memory Access” on page 5-36, the DSP can provide
illegal access information for Long word or I/O register accesses. When
5-38
ADSP-21160 SHARC DSP Hardware Reference
Memory
these conditions occur, the DSP updates an illegal condition flag in a
sticky status (STKYx) register. Either of these two conditions can also generate a maskable interrupt. Two ways to use illegal access information are:
• Interrupts. Enable interrupts and use an interrupt service routine
to handle the illegal access condition immediately. This method is
appropriate if it is important to handle all illegal accesses as they
occur.
• STKYx registers. Use the Bit Tst instruction to examine illegal
condition flags in the STKY register after an interrupt to determine
which illegal access condition occurred.
Accessing Memory
The word width of DSP processor core accesses to internal memory vary
according to the following rules:
• 48-bit access for instruction words, extended precision Normal
word (40-bit) data, and PX register
• 64-bit access for Long word data, and Normal word (32-bit) or PX
register data with the LW mnemonic
• 32-bit access for Normal word (32-bit) data
• 16-bit access for Short word data
The DSP determines whether a Normal word access is 32- or 40-bit from
the internal memory block’s IMDWx setting. For more information, see
“Internal Memory Block Data Width” on page 5-32. While mixed
accesses of 48-bit words and 16-, 32-, or 64-bit words at the same address
are not allowed, mixed read/writes of 16-, 32-, and 64-bit words to the
same address are allowed. For more information, see “Restrictions on Mixing 32-Bit and 48-Bit Words” on page 5-26.
ADSP-21160 SHARC DSP Hardware Reference
5-39
Accessing Memory
The DSP’s DM and PM buses support 24 combinations of register-to-memory data access options. The following factors influence the
data access type:
• Size of words: Short word, Normal word, extended precision Normal word, or Long word
• Number of words: single- or dual-data move
• Mode of DSP: SISD, SIMD, or broadcast load
Access Word Size
The DSP’s internal memory accommodates the following word sizes:
• 48-bit instruction words
• 40-bit extended precision Normal word data
• 32-bit Normal word data
• 16-bit Short word data
The DSP’s external memory accommodates the following word sizes:
• 48-bit instruction words
• 32-bit Normal word data
To access words of memory, the DSP supports the following memory
access word sizes:
• 64-bit accesses, comprised of two consecutive 32-bit data words
• 48-bit accesses, for instruction fetches only
• 40-bit data word accesses
5-40
ADSP-21160 SHARC DSP Hardware Reference
Memory
• 32-bit data word accesses
• 16-bit data word accesses
Long Word (64-Bit) Accesses
A program makes a Long word (64-bit) access to internal memory, using
an access to a Long word address. Programs can also make a 64-bit access
through Normal word addressing with the LW mnemonic or through a PX
register move with the LW mnemonic. Programs may not use Long word
addressing to access multiprocessor memory space or external memory.
The address ranges for internal memory accesses appear in Figure 5-7 on
page 5-15.
When data is accessed using Long word addressing, the data is always
Long word aligned on 64-bit boundaries in internal memory space. When
data is accessed using Normal word addressing and the LW mnemonic, the
program should maintain this alignment by using an even Normal word
address (least significant bit of address =0). This register selection aligns
the Normal word address with a 64-bit boundary (Long word address).
All Long word accesses load or store two consecutive 32-bit data values.
The register file source or destination of a Long word access is a set of two
neighboring data registers in a processing element. In a forced Long word
access (uses the LW mnemonic), the even (Normal word address) location
moves to or from the explicit register in the neighbor-pair, and the odd
(Normal word address) location moves to or from the implicit register in
the neighbor-pair. For example, the following Long word moves could
occur:
DM(0x40000) = R0 (LW);
{The data in R0 moves to location DM(0x40000),
and the data in R1 moves to location DM(0x40001).}
R0 = DM(0x40003) (LW);
ADSP-21160 SHARC DSP Hardware Reference
5-41
Accessing Memory
{The data at location DM(0x40002) moves to R0,
and the data at location DM(0x40003) moves to R1.}
The example shows that R0 and R1 are a neighbor registers in the same
processing element. Table 5-7 lists the other neighbor register assignments
that apply to Long word accesses.
In un-forced Long word accesses, the DSP places the lower 32-bits of the
Long word in the named (explicit) register and places the upper 32-bits of
the Long word in the neighbor (implicit) register.
Table 5-7. Neighbor Registers for Long Word Accesses
PEx neighbor registers
PEy neighbor registers
r0
neighbors r1
s0
neighbors s1
r2
neighbors r3
s2
neighbors s3
r4
neighbors r5
s4
neighbors s5
r6
neighbors r7
s6
neighbors s7
r8
neighbors r9
s8
neighbors s9
r10
neighbors r11
s10
neighbors s11
r12
neighbors r13
s12
neighbors s13
r14
neighbors r15
s14
neighbors s15
Programs can monitor for unaligned 64-bit accesses by enabling the U64MAE bit. For more information, see “Unaligned 64-bit Memory Access” on
page 5-36.
Long word ( ) mnemonic only effects Normal word address
 The
accesses and overrides all other factors (
,
).
LW
PEYEN IMDWx
5-42
ADSP-21160 SHARC DSP Hardware Reference
Memory
Instruction Word (48-Bit) and Extended Precision Normal
Word (40-Bit) Accesses
The sequencer uses a 48-bit memory access for instruction fetches. Program can make 48-bit accesses with PX register moves, which default to
48-bit unless the LW mnemonic is part of the instruction.
A program makes an extended precision Normal word (40-bit) access to
internal memory using an access to a Normal word address when that
internal memory block’s IMDWx bit is set (=1) for 40-bit words. Programs
may not use extended precision Normal word addressing to access multiprocessor memory space or external memory. The address ranges for
internal memory accesses appear in Figure 5-7 on page 5-15. For more
information on configuring memory for extended precision Normal word
accesses, see “Internal Memory Block Data Width” on page 5-32.
The DSP transfers the 40-bit data to internal memory as a 48-bit value,
zero-filling the least significant 8 bits on stores and truncating these 8 bits
on loads. The register file source or destination of such an access is a single
40-bit data register.
Normal Word (32-Bit) Accesses
A program makes a Normal word (32-bit) access to internal memory using
an access to a Normal word address when that internal memory block’s
IMDWx bit is cleared (=0) for 32-bit words. Programs use Normal word
addressing to access all DSP memory spaces: internal, multiprocessor, and
external memory space. The address ranges for memory accesses appear in
Figure 5-7 on page 5-15, Figure 5-8 on page 5-18, and Figure 5-9 on
page 5-20.
ADSP-21160 SHARC DSP Hardware Reference
5-43
Accessing Memory
The register file source or destination of a Normal word access is a single
40-bit data register. The DSP zero-fills the least significant 8 bits on loads
and truncates these bits on stores.
memory space accesses using Normal word addressing and
 External
the LW mnemonic performs a forced 64-bit access.
Short Word (16-Bit) Accesses
A program makes a Short word (16-bit) access to internal memory, using
an access to a Short word address. Programs may not use Short word
addressing to access multiprocessor memory space or external memory.
The address ranges for internal memory accesses appear in Figure 5-7 on
page 5-15.
The register file source or destination of such an access is a single 40-bit
data register. The DSP zero-fills the least significant 8 bits on loads and
truncates these bits on stores. Depending on the value of the SSE bit in the
MODE1 system register, the DSP loads the register’s upper 16 bits by either:
• Zero-filling these bits if SSE=0
• Sign-extending these bits if SSE=1
SISD, SIMD, and Broadcast Load Modes
These three processing element modes influence memory accesses. For a
comparison of their effects, see the examples in “Data Access Options” on
page 5-45. For more information on SISD and SIMD modes, see “Secondary Processing Element (PEy)” on page 2-35.
Broadcast load mode is a hybrid between SISD and SIMD modes, transferring dual-data under special conditions. For examples of broadcast
transfers, see “Data Access Options” on page 5-45. For more information
on broadcast load mode, see “Broadcast Register Loads” on page 5-34.
5-44
ADSP-21160 SHARC DSP Hardware Reference
Memory
Single-and Dual-Data Accesses
The number of transfers that occur in a cycle influences the data access
operation. As described on “Overview” on page 5-1, the DSP supports
single-cycle, dual-data accesses to internal memory for register-to-memory
transfers. Though only available for transfers to data registers, dual-data
transfers are extremely useful, because they double the data throughput
over single-data transfers. For examples of data flow paths for single- and
dual-data transfers, see “Data Access Options” on page 5-45.
Data Access Options
Table 5-8 lists the DSP’s 24 possible memory transfer modes that stem
from the DSP’s data access options. When looking at Table 5-8, it is
important to note that Long and Short word addressing may not target
multiprocessor memory space or external memory space.
ADSP-21160 SHARC DSP Hardware Reference
5-45
Accessing Memory
Table 5-8. The 24 Possible Memory Transfer Modes
Access
Type
Single
Data
Access
Dual
Data
Access
DSP
Mode
Address Space
Long Word
Extended
Precision
Normal Word
Short Word
PM
DM
PM
DM
PM
DM
PM
DM
SISD
mode
LW
none
EW
none
NW
none
SW
none
none
LW
none
EW
none
NW
none
SW
SIMD
mode
LW
none
EW
none
LW
none
SWx2
none
none
LW
none
EW
none
LW
none
SWx2
B-cast
Load
LW
none
EW
none
NW
none
SW
none
none
LW
none
EW
none
NW
none
SW
SISD
mode
LW
LW
EW
EW
NW
NW
SW
SW
SIMD
mode
LW
LW
EW
EW
LW
LW
SWx2
SWx2
B-cast
Load
LW
LW
EW
EW
NW
NW
SW
SW
Symbols: LW = 64-bit data value (two 32-bit values), EW = 40-bit data value (48-bit value),
NW = 32-bit data value, SW = 16-bit data value, and SWx2 = two 16-bit data values.
Table 5-8 shows the transfer modes that stem from the following data
access options:
• The mode of the DSP: SISD, SIMD, or Broadcast Load
• The size of access words: Long, extended precision Normal word,
Normal word, or Short word
• The number of transferred words: single- or dual-data
5-46
ADSP-21160 SHARC DSP Hardware Reference
Memory
Table 5-8 provides a cross reference to examples of each memory access
option.
Table 5-9. Memory Transfer Modes Cross Reference
Access
Type
DSP
Mode
Address Space
Long Word
Extended
Precision
Normal Word
Short Word
Single
Data
Access
SISD
mode
on page 5-69
on page 5-62
on page 5-55
on page 5-47
SIMD
mode
on page 5-69
on page 5-64
on page 5-57
on page 5-48
B-cast
Load
Figure 5-23
Figure 5-23
Figure 5-21
Figure 5-16
SISD
mode
on page 5-69
on page 5-59
on page 5-59
on page 5-51
SIMD
mode
on page 5-69
on page 5-64
on page 5-59
on page 5-53
B-cast
Load
Figure 5-26
Figure 5-24
Figure 5-21
Figure 5-17
Dual
Data
Access
Short Word Addressing of Single Data in SISD Mode
Figure 5-14 on page 5-49 displays one possible SISD mode, single data,
Short word addressed access. For Short word addressing, the DSP treats
the data buses as four 16-bit Short word lanes. The 16-bit value for the
Short word access transfers using the least significant Short word lane of
the PM or DM data bus. The DSP drives the other Short word lanes of
the data buses with zeros.
In Figure 5-14 on page 5-49, the access targets PEx registers in a SISD
mode operation. This case accesses WORD X0 whose Short word address has
“00” for its least significant two bits of address. Other accesses within this
4-column location have addresses with least significant two bits of “01”,
“10”, or “11” and select WORD X1, WORD X2, or WORD X3 from memory
ADSP-21160 SHARC DSP Hardware Reference
5-47
Accessing Memory
respectively. The syntax targets register, RX, in PEx. The example would
target a PEy register if using the syntax SX.
The cross (†) in Figure 5-14 indicates that the DSP zero-fills or
sign-extends the most significant 16 bits of the data register while loading
the Short word value into a 40-bit data register. The selection depends on
the state of the SSE bit in the MODE1 system register. The least significant 8
bits of the data register are always zero.
Short Word Addressing of Single Data in SIMD Mode
Figure 5-15 displays one possible SIMD mode, single data, Short word
addressed access.
For Short word addressing, the DSP treats the data buses as four 16-bit
Short word lanes. The explicitly addressed (named in the instruction)
16-bit value transfers using the least significant Short word lane of the PM
or DM data bus. The implicitly addressed (not named in the instruction,
but inferred from the address in SIMD mode) Short word value transfers
using the 47-32 bit Short word lane of the PM or DM data bus. The DSP
drives the other Short word lanes of the PM or DM data buses with zeros.
In Figure 5-15, the explicit access targets the named register, RX, and the
implicit access targets that register’s complementary register, SX. This case
uses a PEx register with an RX mnemonic. If the syntax named a PEy register, SX, as the explicit target the DSP would use that register’s
complement, RX, as the implicit target. For more information on complementary registers, see “Secondary Processing Element (PEy)” on
page 2-35.
The cross (†) in Figure 5-15 indicates that the DSP zero-fills or
sign-extends the most significant 16 bits of the data register while loading
the Short word value into a 40-bit data register. The selection depends on
the state of the SSE bit in the MODE1 system register. The least significant 8
bits of the data register are always zero.
5-48
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y8
WORD Y7 WORD Y6
WORD Y5
WORD Y4
WORD Y3 WORD Y2
WORD Y1
WORD Y0
WORD X11 WORD X10 WORD X9 WORD X8
ADDRESS
WORD Y11 WORD Y10 WORD Y9
ADDRESS
BLOCK 1 (DM)
WORD X7
WORD X6
WORD X3
WORD X2 WORD X1
NO ACCESS
63-48
47-32
31-16
39-24
15-0
RB
23-8
7-0
63-48
47-32
31-16
15-0
DM DATA
BUS
0X0000
0X0000
0X0000
WORD X0
39-24
23-8
RA
39-24
WORD X0
SHORT WORD ACCESS
PM DATA
BUS
PEX REGISTERS
WORD X5 WORD X4
23-8
7-0
RY
7-0
RX
39-24
23-8
7-0
0X0000† WORD X0 0X00
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
7-0
SY
39-24
23-8
7-0
SX
39-24
23-8
7-0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(SHORT WORD X0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SISD, SHORT WORD, SINGLE-DATA TRANSFERS ARE:
UREG = PM(SHORT WORD ADDRESS);
UREG = DM(SHORT WORD ADDRESS);
PM(SHORT WORD ADDRESS) = UREG;
DM(SHORT WORD ADDRESS) = UREG;
Figure 5-14. Short Word Addressing of Single Data in SISD Mode
ADSP-21160 SHARC DSP Hardware Reference
5-49
Accessing Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y8
WORD Y7 WORD Y6
WORD Y5
WORD Y4
WORD Y3 WORD Y2
WORD Y1
WORD Y0
WORD X11 WORD X10 WORD X9 WORD X8
ADDRESS
WORD Y11 WORD Y10 WORD Y9
ADDRESS
BLOCK 1 (DM)
WORD X7
WORD X6
WORD X3
WORD X2 WORD X1
NO ACCESS
63-48
47-32
31-16
39-24
15-0
DM DATA
BUS
RB
23-8
7-0
63-48
47-32
31-16
15-0
0X0000
WORD X2
0X0000
WORD X0
RA
39-24
WORD X0
SHORT WORD ACCESS
PM DATA
BUS
PEX REGISTERS
WORD X5 WORD X4
23-8
7-0
RY
39-24
23-8
7-0
RX
39-24
23-8
7-0
0X0000† WORD X0 0X00
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
7-0
SY
39-24
23-8
7-0
SX
39-24
23-8
7-0
0X0000† WORD X2 0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(SHORT WORD X0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SIMD, SHORT WORD, SINGLE-DATA TRANSFERS ARE:
UREG = PM(SHORT WORD ADDRESS);
UREG = DM(SHORT WORD ADDRESS);
PM(SHORT WORD ADDRESS) = UREG;
DM(SHORT WORD ADDRESS) = UREG;
Figure 5-15. Short Word Addressing of Single Data in SIMD Mode
5-50
ADSP-21160 SHARC DSP Hardware Reference
Memory
Figure 5-15 shows the data path for one transfer, but in this mode it is
also important to note the pattern of iterative accesses. For Short word
accesses, the DSP accesses Short words sequentially in memory.
Table 5-10 shows the pattern of SIMD mode Short word accesses. For
more information on arranging data in memory to take advantage of this
access pattern, see “Arranging Data in Memory” on page 5-84.
Table 5-10. Short Word Addressing in SIMD Mode
Explicit Short Word Accessed
Implicit Short Word Accessed
Word X0 (Address two LSBs = 00)
Word X2 (Address two LSBs = 10)
Word X1 (Address two LSBs = 01)
Word X3 (Address two LSBs = 11)
Word X2 (Address two LSBs = 10)
Word X4 (Address two LSBs = 00)
Word X3 (Address two LSBs = 11)
Word X5 (Address two LSBs = 01)
Short Word Addressing of Dual-Data in SISD Mode
Figure 5-16 displays one possible SISD mode, dual-data, Short word
addressed access.
For Short word addressing, the DSP treats the data buses as four 16-bit
Short word lanes. The 16-bit values for Short word accesses transfer using
the least significant Short word lanes of the PM and DM data buses. The
DSP drives the other Short word lanes of the data buses with zeros. Note
that the accesses on both buses do not have to be the same word width.
SISD mode dual-data accesses can handle any combination of Short word,
Normal word, extended precision Normal word, or Long word accesses.
For more information, see “Mixed Word Width Addressing of Dual Data
in SISD Mode” on page 5-72.
In Figure 5-16, the access targets PEx registers in a SISD mode operation.
This case accesses WORD X0 in block 1 and WORD Y0 in block 0. Each of
these words has a Short word address with “00” for its least significant two
bits of address. Other accesses within these 4-column location have the
addresses with least significant two bits of “01”, “10”, or “11” and select
ADSP-21160 SHARC DSP Hardware Reference
5-51
Accessing Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y8
WORD Y7 WORD Y6
WORD Y5
WORD Y4
WORD Y3 WORD Y2
WORD Y1
WORD Y0
WORD X11 WORD X10 WORD X9 WORD X8
ADDRESS
WORD Y11 WORD Y10 WORD Y9
ADDRESS
BLOCK 1 (DM)
WORD X7
WORD X6
WORD X3
WORD X2 WORD X1
SHORT WORD ACCESS
PM DATA
BUS
39-24
63-48
47-32
31-16
15-0
0X0000
0X0000
WORD Y0
RB
23-8
7-0
63-48
47-32
31-16
DM DATA
BUS
0X0000
0X0000
0X0000
39-24
23-8
RA
39-24
23-8
39-24
SB
23-8
7-0
SA
39-24
23-8
15-0
WORD X0
RY
7-0
7-0
0X0000† WORD Y0 0X00
PEY REGISTERS
WORD X0
SHORT WORD ACCESS
0X0000
PEX REGISTERS
WORD X5 WORD X4
RX
39-24
23-8
0X0000† WORD X0 0X00
SY
7-0
39-24
23-8
7-0
7-0
SX
39-24
23-8
7-0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(SHORT WORD X0 ADDRESS), RY = PM(SHORT WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SISD, SHORT WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(SHORT WORD ADDRESS),
PM(SHORT WORD ADDRESS) = DREG,
DREG = DM(SHORT WORD ADDRESS);
DM(SHORT WORD ADDRESS) = DREG;
Figure 5-16. Short Word Addressing of Dual Data in SISD Mode
5-52
ADSP-21160 SHARC DSP Hardware Reference
Memory
WORD X/Y1, WORD X/Y2,
or WORD X/Y3 from memory respectively. The syntax targets registers, RX and RY, in PEx. The example would target PEy
registers if using the syntax SX or SY.
The cross (†) in Figure 5-16 indicates that the DSP zero-fills or
sign-extends the most significant 16 bits of the data register while loading
a Short word value into a 40-bit data register. The selection depends on
the state of the SSE bit in the MODE1 system register. The least significant 8
bits of the data register are always zero.
Short Word Addressing of Dual-Data in SIMD Mode
Figure 5-17 displays one possible SIMD mode, dual-data, Short word
addressed access.
For Short word addressing, the DSP treats the data buses as four 16-bit
Short word lanes. The explicitly addressed (named in the instruction)
16-bit values transfer using the least significant Short word lanes of the
PM and DM data bus. The implicitly addressed (not named in the
instruction, but inferred from the address in SIMD mode) Short word values transfer using the 47-32 bit Short word lanes of the PM and DM data
buses. The DSP drives the other Short word lanes of the PM and DM data
buses with zeros.
accesses on both buses do not have to be the same word width.
 The
SIMD mode dual-data accesses can handle combinations of Short
word and Normal word or extended precision Normal word and
Long word accesses. For more information, see “Mixed Word
Width Addressing of Dual Data in SIMD Mode” on page 5-74.
In Figure 5-17, the explicit accesses targets the named registers RX and RA,
and the implicit accesses target those register’s complementary registers, SX
and SA. This case uses a PEx registers with the RX and RA mnemonics. If
the syntax named PEy registers SX and SA as the explicit targets, the DSP
would use those registers’ complements, RX and RA, as the implicit targets.
ADSP-21160 SHARC DSP Hardware Reference
5-53
Accessing Memory
MEMORY
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y11 WORD Y10 WORD Y9
WORD Y8
WORD Y7 WORD Y6
WORD Y5
WORD Y4
WORD Y3 WORD Y2
WORD Y1
WORD Y0
ADDRESS
ADDRESS
BLOCK 0 (PM)
WORD X11 WORD X10 WORD X9 WORD X8
WORD X7
WORD X6
WORD X3
WORD X2 WORD X1
SHORT WORD ACCESS
PM DATA
BUS
39-24
63-48
47-32
31-16
15-0
WORD Y2
0X0000
WORD Y0
RB
23-8
7-0
0X0000†
PEY REGISTERS
39-24
23-8
7-0
63-48
47-32
31-16
15-0
0X0000
WORD X2
0X0000
WORD X0
RY
7-0
39-24
23-8
23-8
RX
39-24
23-8
7-0
0X0000† WORD X0 0X00
SA
39-24
7-0
WORD Y0 0X00
SB
23-8
DM DATA
BUS
RA
39-24
WORD X0
SHORT WORD ACCESS
0X0000
PEX REGISTERS
WORD X5 WORD X4
SY
7-0
39-24
0X0000† WORD Y2 0X00
23-8
7-0
SX
39-24
23-8
7-0
0X0000† WORD X2 0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(SHORT WORD X0 ADDRESS), RA = PM(SHORT WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SIMD, SHORT WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(SHORT WORD ADDRESS),
PM(SHORT WORD ADDRESS) = DREG,
DREG = DM(SHORT WORD ADDRESS);
DM(SHORT WORD ADDRESS) = DREG;
Figure 5-17. Short Word Addressing of Dual Data in SIMD Mode
5-54
ADSP-21160 SHARC DSP Hardware Reference
Memory
For more information on complementary registers, see “Secondary Processing Element (PEy)” on page 2-35.
The cross (†) in Figure 5-17 indicates that the DSP zero-fills or
sign-extends the most significant 16 bits of the data registers while loading
the Short word values into the 40-bit data registers. The selection depends
on the state of the SSE bit in the MODE1 system register. The least significant 8 bits of the data register are always zero.
Figure 5-17 shows the data path for one transfer, but in this mode it is
also important to note the pattern of iterative accesses. For Short word
accesses, the DSP accesses Short words sequentially in memory.
Table 5-10 on page 5-51 shows the pattern of SIMD mode Short word
accesses. For more information on arranging data in memory to take
advantage of this access pattern, see “Arranging Data in Memory” on
page 5-84.
32-Bit Normal Word Addressing of Single Data in SISD Mode
Figure 5-18 displays one possible SISD mode, single data, 32-bit normal
word addressed access.
For Normal word addressing, the DSP treats the data buses as two 32-bit
Normal word lanes. The 32-bit value for the Normal word access transfers
using the least significant Normal word lane of the PM or DM data bus.
The DSP drives the other Normal word lanes of the data buses with zeros.
In Figure 5-18, the access targets a PEx register in a SISD mode operation.
This case accesses WORD X0 whose Normal word address has “0” for its least
significant address bit. The other access within this 4-column location has
an addresses with a least significant bit of “1” and selects WORD X1 from
memory. The syntax targets register RX in PEx. The example would target
a PEy register if using the syntax SX.
For Normal word accesses, the DSP zero-fills least significant 8 bits of the
data register on loads and truncates these bits on stores to memory.
ADSP-21160 SHARC DSP Hardware Reference
5-55
Accessing Memory
MEMORY
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y11 WORD Y10 WORD Y9
WORD Y8
WORD Y7 WORD Y6
WORD Y5
WORD Y4
WORD Y3 WORD Y2
WORD Y1
WORD Y0
ADDRESS
ADDRESS
BLOCK 0 (PM)
WORD X11 WORD X10 WORD X9 WORD X8
WORD X7
WORD X6
WORD X3
WORD X2 WORD X1
SHORT WORD ACCESS
PM DATA
BUS
39-24
63-48
47-32
31-16
15-0
WORD Y2
0X0000
WORD Y0
RB
23-8
7-0
0X0000†
PEY REGISTERS
39-24
23-8
7-0
63-48
47-32
31-16
15-0
0X0000
WORD X2
0X0000
WORD X0
RY
7-0
39-24
23-8
23-8
RX
39-24
23-8
7-0
0X0000† WORD X0 0X00
SA
39-24
7-0
WORD Y0 0X00
SB
23-8
DM DATA
BUS
RA
39-24
WORD X0
SHORT WORD ACCESS
0X0000
PEX REGISTERS
WORD X5 WORD X4
SY
7-0
39-24
0X0000† WORD Y2 0X00
23-8
7-0
SX
39-24
23-8
7-0
0X0000† WORD X2 0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(SHORT WORD X0 ADDRESS), RA = PM(SHORT WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SIMD, SHORT WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(SHORT WORD ADDRESS),
PM(SHORT WORD ADDRESS) = DREG,
DREG = DM(SHORT WORD ADDRESS);
DM(SHORT WORD ADDRESS) = DREG;
Figure 5-18. Normal Word Addressing of Single Data in SISD Mode
5-56
ADSP-21160 SHARC DSP Hardware Reference
Memory
32-Bit Normal Word Addressing of Single Data in SIMD Mode
Figure 5-19 displays one possible SIMD mode, single data, Normal word
addressed access. Figure 5-19 shows the data path for one transfer, but in
this mode it is also important to note the pattern of iterative accesses. For
Normal word accesses, the DSP accesses Normal words sequentially in
memory. Table 5-11 on page 5-57 shows the pattern of SIMD mode Normal word accesses.
Table 5-11. Normal Word Addressing in SIMD Mode
Explicit Normal Word Accessed
Implicit Normal Word Accessed
Word X0 (Address LSB = 0)
Word X1 (Address LSB = 1)
Word X1 (Address LSB = 1)
Word X2 (Address LSB = 0)
For Normal word addressing, the DSP treats the data buses as two 32-bit
Normal word lanes. The explicitly addressed (named in the instruction)
32-bit value transfers using the least significant Normal word lane of the
PM or DM data bus. The implicitly addressed (not named in the instruction, but inferred from the address in SIMD mode) Normal word value
transfers using the most significant Normal word lane of the PM or DM
data bus. The DSP drives the other Normal word lanes of the data buses
with zeros.
In Figure 5-19, the explicit access targets the named register RX, and the
implicit access targets that register’s complementary register SX. This case
uses a PEx register with an RX mnemonic. If the syntax named a PEy register SX as the explicit target, the DSP would use that register’s complement,
RX, as the implicit target. For more information on complementary registers, see “Secondary Processing Element (PEy)” on page 2-35.
For Normal word accesses, the DSP zero-fills least significant 8 bits of the
data register on loads and truncates these bits on stores to memory. For
more information on arranging data in memory to take advantage of this
access pattern, see “Arranging Data in Memory” on page 5-84.
ADSP-21160 SHARC DSP Hardware Reference
5-57
Accessing Memory
MEMORY
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y5
WORD Y4
WORD Y3
WORD Y2
WORD Y1
WORD Y0
ADDRESS
ADDRESS
BLOCK 0 (PM)
WORD X5
WORD X4
WORD X3
WORD X2
WORD X1
WORD X0
NO ACCESS
63-48
47-32
NORMAL WORD ACCESS
31-16
15-0
63-48
DM DATA
BUS
PM DATA
BUS
RB
PEX REGISTERS
39-24
23-8
7-0
RA
39-24
23-8
7-0
47-32
31-16
WORD X1
15-0
WORD X0
RY
39-24
23-8
7-0
RX
39-24
23-8
WORD X0
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
7-0
SY
39-24
23-8
7-0
7-0
0X00
SX
39-24
23-8
WORD X1
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
7-0
0X00
PEY
REGISTER S
RX = DM(NORMAL WORD X0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SIMD, NORMAL WORD, SINGLE-DATA TRANSFERS ARE:
UREG = PM(NORMAL WORD ADDRESS);
UREG = DM(NORMAL WORD ADDRESS);
PM(NORMAL WORD ADDRESS) = UREG;
DM(NORMAL WORD ADDRESS) = UREG;
Figure 5-19. Normal Word Addressing of Single Data in SIMD Mode
5-58
ADSP-21160 SHARC DSP Hardware Reference
Memory
32-Bit Normal Word Addressing of Dual Data in SISD Mode
Figure 5-20 displays one possible SISD mode, dual data, 32-bit Normal
word addressed access.
For Normal word addressing, the DSP treats the data buses as two 32-bit
Normal word lanes. The 32-bit values for Normal word accesses transfer
using the least significant Normal word lanes of the PM and DM data
buses. The DSP drives the other Normal word lanes of the data buses with
zeros. Note that the accesses on both buses do not have to be the same
word width. SISD mode dual-data accesses can handle any combination of
Short word, Normal word, extended precision Normal word, or Long
word accesses. For more information, see “Mixed Word Width Addressing
of Dual Data in SISD Mode” on page 5-72.
In Figure 5-20, the access targets PEx registers in a SISD mode operation.
This case accesses WORD X0 in block 1 and WORD Y0 in block 0. Each of
these words has a Normal word address with “0” for its least significant
address bit. Other accesses within these 4-column locations have the
addresses with the least significant bit of “1” and select WORD X/Y1 from
memory. The syntax targets registers RX and RY in PEx. The example
would target PEy registers if using the syntax SX or SY.
For Normal word accesses, the DSP zero-fills least significant 8 bits of the
data register on loads and truncates these bits on stores to memory.
32-Bit Normal Word Addressing of Dual Data in SIMD Mode
Figure 5-21 displays one possible SIMD mode, dual data, 32-bit Normal
word addressed access.
For Normal word addressing, the DSP treats the data buses as two 32-bit
Normal word lanes. The explicitly addressed (named in the instruction)
32-bit values transfer using the least significant Normal word lane of the
PM or DM data bus. The implicitly addressed (not named in the instruction, but inferred from the address in SIMD mode) Normal word values
ADSP-21160 SHARC DSP Hardware Reference
5-59
Accessing Memory
MEMORY
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y5
WORD Y4
WORD Y3
WORD Y2
WORD Y1
WORD Y0
ADDRESS
ADDRESS
BLOCK 0 (PM)
WORD X5
WORD X4
WORD X3
WORD X2
WORD X1
WORD X0
NORMAL WORD ACCESS
PM DATA
BUS
63-48
47-32
0X0000
0X0000
15-0
WORD Y0
RB
PEX REGISTERS
39-24
31-16
NORMAL WORD ACCESS
23-8
7-0
63-48
47-32
DM DATA
BUS
0X0000
0X0000
39-24
23-8
RA
39-24
23-8
39-24
SB
23-8
7-0
SA
39-24
23-8
15-0
WORD X0
RY
7-0
WORD Y0
PEY REGISTERS
31-16
7-0
0X00
RX
39-24
23-8
WORD X0
7-0
0X00
SY
7-0
39-24
23-8
7-0
SX
39-24
23-8
7-0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(NORMAL WORD X0 ADDRESS), RY = PM(NORMAL WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SISD, NORMAL WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(NORMAL WORD ADDRESS),
PM(NORMAL WORD ADDRESS) = DREG,
DREG = DM(NORMAL WORD ADDRESS);
DM(NORMAL WORD ADDRESS) = DREG;
Figure 5-20. Normal Word Addressing of Dual Data in SISD Mode
5-60
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y5
WORD Y4
WORD Y3
WORD Y2
WORD Y1
WORD Y0
ADDRESS
ADDRESS
BLOCK 0 (PM)
WORD X5
WORD X4
WORD X3
WORD X2
WORD X1
WORD X0
NORMAL WORD ACCESS
63-48
PM DATA
BUS
PEX REGISTERS
39-24
47-32
31-16
WORD Y1
NORMAL WORD ACCESS
15-0
RB
23-8
7-0
23-8
39-24
SB
23-8
7-0
23-8
WORD Y1
31-16
15-0
WORD X0
RY
7-0
39-24
23-8
7-0
RX
39-24
23-8
WORD X0
0X00
SA
39-24
47-32
WORD X1
RA
39-24
WORD Y0
PEY REGISTERS
63-48
DM DATA
BUS
WORD Y0
7-0
0X00
SY
7-0
39-24
23-8
7-0
SX
39-24
23-8
WORD X1
0X00
7-0
0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(NORMAL WORD X0 ADDRESS), RY = PM(NORMAL WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SIMD, NORMAL WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(NORMAL WORD ADDRESS),
PM(NORMAL WORD ADDRESS) = DREG,
DREG = DM(NORMAL WORD ADDRESS);
DM(NORMAL WORD ADDRESS) = DREG;
Figure 5-21. Normal Word Addressing of Dual Data in SIMD Mode
ADSP-21160 SHARC DSP Hardware Reference
5-61
Accessing Memory
transfer using the most significant Normal word lanes of the PM and DM
data bus. The DSP drives the other Normal word lanes of the PM and
DM data buses with zeros. Note that the accesses on both buses do not
have to be the same word width. SIMD mode dual-data accesses can handle combinations of Short word and Normal word or extended precision
Normal word and Long word accesses. For more information, see “Mixed
Word Width Addressing of Dual Data in SIMD Mode” on page 5-74.
In Figure 5-21, the explicit access targets the named registers RX and RA,
and the implicit access targets those register’s complementary registers SX
and SA. This case uses a PEx registers with the RX and RA mnemonics. If
the syntax named PEy registers SX and SA as the explicit targets, the DSP
would use those registers’ complements RX and RA as the implicit targets.
For more information on complementary registers, see “Secondary Processing Element (PEy)” on page 2-35.
For Normal word accesses, the DSP zero-fills least significant 8 bits of the
data register on loads and truncates these bits on stores to memory.
Figure 5-21 shows the data path for one transfer, but in this mode it is
also important to note the pattern of iterative accesses. For Normal word
accesses, the DSP accesses Normal words sequentially in memory.
Table 5-11 on page 5-24 shows the pattern of SIMD mode Normal word
accesses. For more information on arranging data in memory to take
advantage of this access pattern, see “Arranging Data in Memory” on
page 5-84.
Extended Precision Normal Word Addressing of Single Data
Figure 5-22 displays one possible single data, 40-bit extended precision
Normal word addressed access. For extended precision Normal word
addressing, the DSP treats each data bus as a 40-bit extended precision
Normal word lane. The 40-bit value for the extended precision Normal
word access transfers using the most significant 40 bits of the PM or DM
data bus. The DSP drives the lower 24 bits of the data buses with zeros.
5-62
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y3
WORD Y2
WORD Y1
WORD Y1
WORD X3
WORD Y2
ADDRESS
ADDRESS
BLOCK 1 (DM)
WORD Y0
WORD X2
47-32
31-16
39-24
15-0
63-48
DM DATA
BUS
RB
23-8
7-0
23-8
7-0
47-32
31-16
WORD X0
RA
39-24
WORD X0
EXTENDED PRECISION NORMAL
WORD ACCESS
PM DATA
BUS
PEX REGISTERS
WORD X1
WORD X1
NO ACCESS
63-48
WORD X2
0X00
15-0
0X0000
RY
39-24
23-8
7-0
RX
39-24
23-8
7-0
WORD X0
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
7-0
SY
39-24
23-8
7-0
SX
39-24
23-8
7-0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(EXTENDED PRECISION NORMAL WORD X0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SISD OR SIMD, EXT. PREC. NORMAL WORD, SINGLE-DATA
TRANSFERS ARE:
UREG = PM(EXTENDED PRECISION NORMAL WORD ADDRESS);
UREG = DM(EXTENDED PRECISION NORMAL WORD ADDRESS);
PM(EXTENDED PRECISION NORMAL WORD ADDRESS) = UREG;
DM(EXTENDED PRECISION NORMAL WORD ADDRESS) = UREG;
Figure 5-22. Extended Precision Normal Word Addressing of Single Data
ADSP-21160 SHARC DSP Hardware Reference
5-63
Accessing Memory
In Figure 5-22, the access targets a PEx register in a SISD or SIMD mode
operation; extended precision Normal word single-data access operate the
same in SISD or SIMD mode. This case accesses WORD X0 with syntax that
targets register RX in PEx. The example would target a PEy register if using
the syntax SX.
Extended Precision Normal Word Addressing of Dual Data in
SISD Mode
Figure 5-23 displays one possible SISD mode, dual data, 40-bit extended
precision normal word addressed access.
For extended precision Normal word addressing, the DSP treats each data
bus as a 40-bit extended precision Normal word lane. The 40-bit values
for the extended precision Normal word accesses transfer using the most
significant 40 bits of the PM and DM data bus. The DSP drives the lower
24 bits of the data buses with zeros. Note that the accesses on both buses
do not have to be the same word width. SISD mode dual-data accesses can
handle any combination of Short word, Normal word, extended precision
Normal word, or Long word accesses. For more information, see “Mixed
Word Width Addressing of Dual Data in SISD Mode” on page 5-72.
In Figure 5-23, the access targets PEx registers in a SISD mode operation.
This case accesses WORD X0 in block 1 and WORD Y0 in block 0 with syntax
that targets registers RX and RY in PEx. The example would target a PEy
registers if using the syntax SX or SY.
Extended Precision Normal Word Addressing of Dual Data in
SIMD Mode
Figure 5-24 displays one possible SIMD mode, dual data, 40-bit extended
precision normal word addressed access.
For extended precision Normal word addressing, the DSP treats each data
bus as a 40-bit extended precision Normal word lane.
5-64
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y3
WORD X3
WORD Y2
WORD Y2
ADDRESS
ADDRESS
BLOCK 1 (DM)
WORD Y1
WORD Y1
WORD Y0
WORD X2
63-48
47-32
31-16
WORD Y0
PEX REGISTERS
39-24
0X00
7-0
63-48
DM DATA
BUS
0X0000
23-8
47-32
31-16
WORD X0
RA
39-24
WORD X0
EXTENDED PRECISION NORMAL
WORD ACCESS
15-0
RB
23-8
WORD X1
WORD X1
EXTENDED PRECISION NORMAL
WORD ACCESS
PM DATA
BUS
WORD X2
RY
7-0
39-24
23-8
7-0
RX
39-24
WORD Y0
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
15-0
0X00 0X0000
23-8
WORD X0
SY
7-0
39-24
23-8
7-0
7-0
SX
39-24
23-8
7-0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(EP NORMAL WORD X0 ADDR.), RY = PM(EP NORMAL WORD Y0 ADDR.);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SISD, EXTENDED PRECISION NORMAL WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(EXT. PREC. NORMAL WORD ADDRESS),
PM(EXT. PREC. NORMAL WORD ADDRESS) = DREG,
DREG = DM(EXT. PREC. NORMAL WORD ADDRESS);
DM(EXT. PREC. NORMAL WORD ADDRESS) = DREG;
Figure 5-23. Extended Precision Normal Word Addressing of Dual Data
in SISD Mode
ADSP-21160 SHARC DSP Hardware Reference
5-65
Accessing Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y3
WORD X3
WORD Y2
WORD Y2
ADDRESS
ADDRESS
BLOCK 1 (DM)
WORD Y1
WORD Y1
WORD Y0
WORD X2
63-48
PEX REGISTERS
39-24
47-32
31-16
WORD Y0
7-0
63-48
15-0
DM DATA
BUS
23-8
47-32
31-16
WORD X0
RA
39-24
WORD X0
EXTENDED PRECISION NORMAL
WORD ACCESS
0X00 0X0000
RB
23-8
WORD X1
WORD X1
EXTENDED PRECISION NORMAL
WORD ACCESS
PM DATA
BUS
WORD X2
15-0
0X00 0X0000
RY
7-0
39-24
23-8
7-0
RX
39-24
23-8
7-0
WORD X0
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
SY
7-0
39-24
23-8
7-0
SX
39-24
23-8
7-0
WORD Y0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(EP NORMAL WORD X0 ADDR.), SX = PM(EP NORMAL WORD Y0 ADDR.);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SIMD, EXTENDED PRECISION NORMAL WORD, DUAL-DATA TRANSFERS
ARE:
PEY DREG = PM(EP NORMAL WORD ADDRESS),
PM(EP NORMAL WORD ADDRESS) = PEY DREG,
PEX DREG = DM(EP NORMAL WORD ADDRESS);
DM(EP NORMAL WORD ADDRESS) = PEX DREG;
Figure 5-24. Extended Precision Normal Word Addressing of Dual Data
in SIMD Mode
5-66
ADSP-21160 SHARC DSP Hardware Reference
Memory
Because this word size approaches the limit of the data buses capacity, this
SIMD mode transfer only moves the explicitly addressed locations and
restricts data bus usage. The explicitly addressed (named in the instruction) 40-bit values transferred over the DM bus must source or sink a PEx
data register, and the explicitly addressed (named in the instruction)
40-bit values transferred over the PM bus must source or sink a PEy data
register; there are no implicit transfers in this mode. The 40-bit values for
the extended precision Normal word accesses transfer using the most significant 40 bits of the PM and DM data bus. The DSP drives the lower 24
bits of the data buses with zeros.
accesses on both buses do not have to be the same word width.
 The
This special case of SIMD mode dual-data accesses can handle any
combination of extended precision Normal word or Long word
accesses. For more information, see “Mixed Word Width Addressing of Dual Data in SIMD Mode” on page 5-74.
In Figure 5-24, the access targets PEx and PEy registers in a SIMD mode
operation. This case accesses WORD X0 in block 1 with syntax that targets
register RX in PEx and accesses WORD Y0 in block 0 with syntax that targets
register SX in PEy.
Long Word Addressing of Single Data
Figure 5-25 displays one possible single data, Long word addressed access.
For Long word addressing, the DSP treats each data bus as a 64-bit Long
word lane. The 64-bit value for the Long word access transfers using the
full width of the PM or DM data bus.
In Figure 5-25, the access targets a PEx register in a SISD or SIMD mode
operation; Long word single-data access operate the same in SISD or
SIMD mode. This case accesses WORD X0 with syntax that explicitly targets
register RX and implicitly targets its neighbor register RY in PEx. The
example would target PEy registers if using the syntax SX. For more
ADSP-21160 SHARC DSP Hardware Reference
5-67
Accessing Memory
MEMORY
BLOCK 0 (PM)
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD X2
ADDRESS
ADDRESS
WORD Y2
WORD Y1
WORD Y0
WORD X1
WORD X0
NO ACCESS
63-48
47-32
31-16
LONG WORD ACCESS
15-0
63-48
PM DATA
BUS
PEX REGISTERS
39-24
RB
23-8
PEY REGISTERS
39-24
47-32
DM DATA
BUS
7-0
RA
39-24
23-8
SB
23-8
7-0
7-0
23-8
7-0
15-0
RY
39-24
23-8
7-0
WORD X0, 63-32
0X00
39-24
7-0
SA
39-24
31-16
WORD X0
RX
39-24
23-8
WORD X0, 31-0
SY
23-8
7-0
0X00
SX
39-24
23-8
7-0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(LONG WORD X0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SISD OR SIMD, LONG WORD, SINGLE-DATA TRANSFERS ARE:
UREG = PM(LONG WORD ADDRESS);
UREG = DM(LONG WORD ADDRESS);
PM(LONG WORD ADDRESS) = UREG;
DM(LONG WORD ADDRESS) = UREG;
Figure 5-25. Long Word Addressing of Single Data
5-68
ADSP-21160 SHARC DSP Hardware Reference
Memory
information on how neighbor registers (listed in Table 5-7 on page 5-42)
work, see “Long Word (64-Bit) Accesses” on page 5-41.
Long Word Addressing of Dual Data in SISD Mode
Figure 5-26 displays one possible SISD mode, dual data, long word
addressed access.
For Long word addressing, the DSP treats each data bus as a 64-bit Long
word lane. The 64-bit values for the Long word accesses transfer using the
full width of the PM or DM data bus.
In Figure 5-26, the access targets PEx registers in a SISD mode operation.
This case accesses WORD X0 and WORD Y0 with syntax that explicitly targets
registers RX registers RA and implicitly targets their neighbor registers RY
and RB in PEx. The example would target PEy registers if using the syntax
SX and SA. For more information on how neighbor registers (listed in
Table 5-7 on page 5-42) work, see “Long Word (64-Bit) Accesses” on
page 5-41.
Programs must be careful not to explicitly target neighbor registers in this
case. While the syntax lets programs target these registers, one of the
explicit accesses targets the other access’s implicit target. The DSP resolves
this conflict by performing only the access with higher priority. For more
information on the priority order of data register file accesses, see “Data
Register File” on page 2-28.
Long Word Addressing of Dual Data in SIMD Mode
Figure 5-27 displays one possible SIMD mode, dual data, Long word
addressed access targeting internal memory space.
For Long word addressing, the DSP treats each data bus as a 64-bit Long
word lane. The 64-bit values for the Long word accesses transfer using the
full width of the PM or DM data bus.
ADSP-21160 SHARC DSP Hardware Reference
5-69
Accessing Memory
MEMORY
BLOCK 0 (PM)
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD X2
ADDRESS
ADDRESS
WORD Y2
WORD Y1
WORD X1
WORD X0
WORD Y0
LONG WORD ACCESS
LONG WORD ACCESS
63-48
47-32
PM DATA
BUS
PEX REGISTERS
39-24
15-0
63-48
PEY REGISTERS
7-0
0X00
23-8
WORD Y0, 31-0
SB
23-8
7-0
23-8
15-0
RY
7-0
39-24
23-8
0X00
WORD X0, 63-32
SA
39-24
31-16
WORD X0
RA
39-24
47-32
DM DATA
BUS
RB
23-8
WORD Y0, 63-32
39-24
31-16
WORD Y0
7-0
0X00
RX
39-24
23-8
WORD X0, 31-0
0X00
39-24
7-0
SY
7-0
39-24
23-8
7-0
7-0
SX
23-8
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(LONG WORD X0 ADDRESS), RA = PM(LONG WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SISD, LONG WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(LONG WORD ADDRESS),
PM(LONG WORD ADDRESS) = DREG,
DREG = DM(LONG WORD ADDRESS);
DM(LONG WORD ADDRESS) = DREG;
Figure 5-26. Long Word Addressing of Dual Data in SISD Mode
5-70
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 0 (PM)
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD X2
ADDRESS
ADDRESS
WORD Y2
WORD Y1
WORD Y0
WORD X1
WORD X0
LONG WORD ACCESS
63-48
47-32
PM DATA
BUS
63-48
7-0
23-8
SB
23-8
7-0
23-8
15-0
RY
7-0
39-24
23-8
RX
7-0
39-24
WORD X0, 63-32
0X00
WORD X0, 31-0
0X00
39-24
7-0
39-24
7-0
SA
39-24
31-16
WORD X0
RA
39-24
47-32
DM DATA
BUS
RB
23-8
PEY REGISTERS
39-24
15-0
WORD Y0
PEX REGISTERS
39-24
31-16
LONG WORD ACCESS
23-8
SY
7-0
23-8
WORD Y0, 63-32
0X00
7-0
SX
23-8
WORD Y0, 31-0
0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(LONG WORD X0 ADDRESS), SX = PM(LONG WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SIMD, LONG WORD, DUAL-DATA TRANSFERS ARE:
PEY DREG = PM(LONG WORD ADDRESS),
PM(LONG WORD ADDRESS) = PEY DREG,
PEX DREG = DM(LONG WORD ADDRESS);
DM(LONG WORD ADDRESS) = PEX DREG;
Figure 5-27. Long Word Addressing of Dual Data in SIMD Mode
ADSP-21160 SHARC DSP Hardware Reference
5-71
Accessing Memory
Because this word size approaches the limit of the data buses capacity, this
SIMD mode transfer only moves the explicitly addressed locations and
restricts data bus usage. The explicitly addressed (named in the instruction) 64-bit values transferred over the DM bus must source or sink a PEx
data register, and the explicitly addressed (named in the instruction)
64-bit values transferred over the PM bus must source or sink a PEy data
register; there are no implicit transfers in this mode.
In Figure 5-27, the access targets PEx and PEy registers in a SIMD mode
operation. This case accesses WORD X0 in block 1 with syntax that targets
register RX and its neighbor register RY in PEx and accesses WORD Y0 in
block 0 with syntax that targets register SX and its neighbor register SY in
PEy. For more information on how neighbor registers (listed in “Neighbor
Registers for Long Word Accesses” on page 5-42) work, see “Long Word
(64-Bit) Accesses” on page 5-41.
accesses on both buses do not have to be the same word width.
 The
This special case of SIMD mode dual-data accesses can handle any
combination of extended precision Normal word or Long word
accesses. “Mixed Word Width Addressing of Dual Data in SIMD
Mode” on page 5-74
Mixed Word Width Addressing of Dual Data in SISD Mode
Figure 5-28 displays an example of a mixed word width, dual data, SISD
mode access.
This example shows how the DSP transfers a Long word access on the DM
bus and transfers a Short word access on the PM bus. The memory architecture permits mixing all other combinations of dual-data SISD mode
5-72
ADSP-21160 SHARC DSP Hardware Reference
Memory
Short word, Normal word, extended precision Normal word, and Long
word accesses.
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y8
WORD Y7 WORD Y6
WORD Y5
WORD Y4
WORD Y3 WORD Y2
WORD Y1
WORD Y0
WORD X2
ADDRESS
WORD Y11 WORD Y10 WORD Y9
ADDRESS
BLOCK 1 (DM)
WORD X1
WORD X0
LONG WORD ACCESS
LONG WORD ACCESS
63-48
PM DATA
BUS
0X0000
PEX REGISTERS
39-24
47-32
0X0000
31-16
0X0000
15-0
WORD Y0
RB
23-8
7-0
63-48
23-8
39-24
SB
23-8
7-0
39-24
23-8
WORD X0, 63-32
SA
39-24
23-8
15-0
RY
7-0
0X0000† WORD Y0 0X00
PEY REGISTERS
31-16
WORD X0
RA
39-24
47-32
DM DATA
BUS
7-0
0X00
RX
39-24
23-8
WORD X0, 31-0
SY
7-0
39-24
23-8
7-0
7-0
0X00
SX
39-24
23-8
7-0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(LONG WORD X0 ADDRESS), RA = PM(SHORT WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SISD, MIXED WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(SHORT, NORMAL, EP NORMAL, LONG ADD),
PM(SHORT, NORMAL, EP NORMAL, LONG ADD) = DREG,
DREG = DM(SHORT, NORMAL, EP NORMAL, LONG ADD);
DM(SHORT, NORMAL, EP NORMAL, LONG ADD) = DREG;
Figure 5-28. Mixed Word Width Addressing of Dual Data in SISD Mode
ADSP-21160 SHARC DSP Hardware Reference
5-73
Accessing Memory
of conflicting dual access to the data register file, the DSP
 Inonlycaseperforms
the access with higher priority. For more information on how the DSP prioritizes accesses, see “Data Register File”
on page 2-28.
Mixed Word Width Addressing of Dual Data in SIMD Mode
Figure 5-29 displays an example of a mixed word width, dual data, SIMD
mode access.
This example shows how the DSP transfers a Long word access on the DM
bus and transfers an extended precision Normal word access on the PM
bus.
memory architecture permits mixing SIMD mode dual data
 The
Short word and Normal word accesses or extended precision Normal word and Long word accesses. No other combinations of
mixed word dual-data SIMD mode accesses are permissible.
Broadcast Load Access
Figure 5-30, Figure 5-31 on page 5-77, Figure 5-32 on page 5-78,
Figure 5-33 on page 5-79, Figure 5-34 on page 5-80, Figure 5-35 on
page 5-81, Figure 5-36 on page 5-82, and Figure 5-37 on page 5-83 provide examples of broadcast load accesses for single- and dual-data
transfers.
These examples show that the broadcast load’s memory and register access
is a hybrid of the corresponding non-broadcast SISD and SIMD mode
accesses. The exceptions to this relation are broadcast load dual-data,
extended precision Normal word and Long word accesses. These broadcast
accesses differ from their corresponding non-broadcast mode accesses.
5-74
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y3
WORD Y2
WORD Y2
WORD Y1
WORD Y1
WORD X2
ADDRESS
ADDRESS
BLOCK 1 (DM)
WORD X1
WORD Y0
WORD X0
EXTENDED PRECISION NORMAL
WORD ACCESS
63-48
PM DATA
BUS
PEX REGISTERS
39-24
31-16
0X00
7-0
23-8
7-0
23-8
15-0
RY
7-0
39-24
23-8
RX
7-0
39-24
WORD X0, 63-32
0X00
WORD X0, 31-0
0X00
39-24
7-0
39-24
7-0
SA
39-24
31-16
WORD X0
RA
39-24
47-32
DM DATA
BUS
0X0000
SB
23-8
63-48
15-0
RB
23-8
PEY REGISTERS
39-24
47-32
WORD Y0
LONG WORD ACCESS
23-8
SY
7-0
23-8
7-0
SX
23-8
WORD Y0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(LONG WORD X0 ADDRESS), SX = PM(EP NORMAL WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR SIMD, MIXED WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(ADDRESS),
PM(ADDRESS) = DREG,
DREG = DM(ADDRESS);
DM(ADDRESS) = DREG;
FOR A LIST OF
PERMISSIBLE
MIXED DUAL
ACCESS
COMBINATIONS,
SEE DISCUSSION
IN TEXT.
Figure 5-29. Mixed Word Width Addressing of Dual Data in SIMD Mode
ADSP-21160 SHARC DSP Hardware Reference
5-75
Accessing Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y8
WORD Y7 WORD Y6
WORD Y5
WORD Y4
WORD Y3 WORD Y2
WORD Y1
WORD Y0
WORD X11 WORD X10 WORD X9 WORD X8
ADDRESS
WORD Y11 WORD Y10 WORD Y9
ADDRESS
BLOCK 1 (DM)
WORD X7 WORD X6
WORD X3 WORD X2 WORD X1
NO ACCESS
63-48
47-32
31-16
39-24
15-0
RB
23-8
7-0
63-48
47-32
31-16
15-0
DM DATA
BUS
0X0000
0X0000
0X0000
WORD X0
39-24
23-8
RY
RA
39-24
WORD X0
SHORT WORD ACCESS
PM DATA
BUS
PEX REGISTERS
WORD X5 WORD X4
23-8
7-0
7-0
RX
39-24
23-8
7-0
0X0000† WORD X0 0X00
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
7-0
SY
39-24
23-8
7-0
SX
39-24
23-8
7-0
0X0000† WORD X0 0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(SHORT WORD X0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR BROADCAST, SHORT WORD, SINGLE-DATA TRANSFERS ARE:
UREG = PM(SHORT WORD ADDRESS);
UREG = DM(SHORT WORD ADDRESS);
PM(SHORT WORD ADDRESS) = UREG;
DM(SHORT WORD ADDRESS) = UREG;
Figure 5-30. Short Word Addressing of Single Data in Broadcast Load
5-76
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y11 WORD Y10 WORD Y9
WORD Y8
WORD Y7 WORD Y6
WORD Y5
WORD Y4
WORD Y3 WORD Y2
WORD Y1
WORD Y0
ADDRESS
ADDRESS
BLOCK 0 (PM)
WORD X11 WORD X10 WORD X9 WORD X8
WORD X7
WORD X6
WORD X3
WORD X2 WORD X1
SHORT WORD ACCESS
63-48
PM DATA
BUS
0X0000
0X0000
31-16
23-8
7-0
15-0
0X0000 WORD Y0
63-48
47-32
31-16
15-0
DM DATA
BUS
0X0000
0X0000
0X0000
WORD X0
39-24
23-8
RA
39-24
23-8
RY
7-0
7-0
0X0000† WORD Y0 0X00
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
WORD X0
SHORT WORD ACCESS
RB
PEX REGISTERS
39-24
47-32
WORD X5 WORD X4
23-8
RX
39-24
23-8
0X0000† WORD X0 0X00
SY
7-0
39-24
23-8
7-0
0X0000† WORD Y0 0X00
7-0
SX
39-24
23-8
7-0
0X0000† WORD X0 0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(SHORT WORD X0 ADDRESS), RY = PM(SHORT WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR BROADCAST, SHORT WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(SHORT WORD ADDRESS),
PM(SHORT WORD ADDRESS) = DREG,
DREG = DM(SHORT WORD ADDRESS);
DM(SHORT WORD ADDRESS) = DREG;
Figure 5-31. Short Word Addressing of Dual Data in Broadcast Load
ADSP-21160 SHARC DSP Hardware Reference
5-77
Accessing Memory
MEMORY
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y5
WORD Y4
WORD Y3
WORD Y2
WORD Y1
WORD Y0
ADDRESS
ADDRESS
BLOCK 0 (PM)
WORD X5
WORD X4
WORD X3
WORD X2
WORD X1
WORD X0
NO ACCESS
63-48
47-32
NORMAL WORD ACCESS
31-16
15-0
DM DATA
BUS
PM DATA
BUS
PEX REGISTERS
39-24
RB
23-8
7-0
63-48
47-32
0X0000
0X0000
23-8
7-0
15-0
WORD X0
RY
RA
39-24
31-16
39-24
23-8
7-0
RX
39-24
23-8
WORD X0
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
7-0
SY
39-24
23-8
7-0
7-0
0X00
SX
39-24
23-8
WORD X0
7-0
0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(NORMAL WORD X0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR BROADCAST, NORMAL WORD, SINGLE-DATA TRANSFERS ARE:
UREG = PM(NORMAL WORD ADDRESS);
UREG = DM(NORMAL WORD ADDRESS);
PM(NORMAL WORD ADDRESS) = UREG;
DM(NORMAL WORD ADDRESS) = UREG;
Figure 5-32. Normal Word Addressing of Single Data in Broadcast Load
5-78
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y5
WORD Y4
WORD Y3
WORD Y2
WORD Y1
WORD Y0
ADDRESS
ADDRESS
BLOCK 0 (PM)
WORD X5
WORD X4
WORD X3
WORD X2
WORD X1
WORD X0
NORMAL WORD ACCESS
PM DATA
BUS
63-48
47-32
0X0000
0X0000
PEX REGISTERS
39-24
31-16
NORMAL WORD ACCESS
15-0
RB
23-8
7-0
DM DATA
BUS
WORD Y0
63-48
47-32
0X0000
0X0000
23-8
7-0
39-24
23-8
WORD Y0
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
15-0
WORD X0
RY
RA
39-24
31-16
7-0
0X00
RX
39-24
23-8
WORD X0
7-0
0X00
SY
7-0
39-24
23-8
WORD Y0
7-0
0X00
SX
39-24
23-8
WORD X0
7-0
0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(NORMAL WORD X0 ADDRESS), RY = PM(NORMAL WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR BROADCAST, NORMAL WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(NORMAL WORD ADDRESS),
PM(NORMAL WORD ADDRESS) = DREG,
DREG = DM(NORMAL WORD ADDRESS);
DM(NORMAL WORD ADDRESS) = DREG;
Figure 5-33. Normal Word Addressing of Dual Data in Broadcast Load
ADSP-21160 SHARC DSP Hardware Reference
5-79
Accessing Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y3
WORD Y2
WORD Y1
WORD Y1
WORD X3
WORD Y2
ADDRESS
ADDRESS
BLOCK 1 (DM)
WORD Y0
WORD X2
47-32
31-16
39-24
15-0
63-48
DM DATA
BUS
RB
23-8
7-0
23-8
7-0
47-32
31-16
WORD X0
RA
39-24
WORD X0
EXTENDED PRECISION NORMAL
WORD ACCESS
PM DATA
BUS
PEX REGISTERS
WORD X1
WORD X1
NO ACCESS
63-48
WORD X2
0X00
15-0
0X0000
RY
39-24
23-8
7-0
RX
39-24
23-8
7-0
WORD X0
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
7-0
SY
39-24
23-8
7-0
SX
39-24
23-8
7-0
WORD X0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(EXTENDED PRECISION NORMAL WORD X0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR BROADCAST, EXTENDED NORMAL WORD, SINGLE-DATA TRANSFERS ARE:
UREG = PM(EP NORMAL WORD ADDRESS);
UREG = DM(EP NORMAL WORD ADDRESS);
PM(EP NORMAL WORD ADDRESS) = UREG;
DM(EP NORMAL WORD ADDRESS) = UREG;
Figure 5-34. Extended Precision Normal Word Addressing of Single Data
in Broadcast Load
5-80
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 0 (PM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD Y3
WORD Y2
WORD Y1
WORD Y1
WORD X3
WORD Y2
ADDRESS
ADDRESS
BLOCK 1 (DM)
WORD Y0
WORD X2
63-48
31-16
WORD Y0
PEX REGISTERS
39-24
47-32
0X00
7-0
63-48
DM DATA
BUS
0X0000
23-8
7-0
47-32
31-16
WORD X0
RA
39-24
WORD X0
EXTENDED PRECISION NORMAL
WORD ACCESS
15-0
RB
23-8
WORD X1
WORD X1
EXTENDED PRECISION NORMAL
WORD ACCESS
PM DATA
BUS
WORD X2
0X00 0X0000
RY
39-24
23-8
7-0
RX
39-24
WORD Y0
PEY REGISTERS
39-24
SB
23-8
7-0
SA
39-24
23-8
7-0
15-0
23-8
WORD X0
SY
39-24
23-8
WORD Y0
7-0
7-0
SX
39-24
23-8
7-0
WORD X0
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(EP NORMAL WORD X0 ADDR.), RY = PM(EP NORMAL WORD Y0 ADDR.);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR BROADCAST, EXTENDED NORMAL WORD, DUAL-DATA
TRANSFERS ARE:
DREG = PM(EP NORMAL WORD ADDRESS), DREG = DM(EPNORMAL WORD ADDRESS);
PM(EP NORMAL WORD ADDRESS) = DREG, DM(EP NORMAL WORD ADDRESS) = DREG;
Figure 5-35. Extended Precision Normal Word Addressing of Dual Data
in Broadcast Load
ADSP-21160 SHARC DSP Hardware Reference
5-81
Accessing Memory
MEMORY
BLOCK 0 (PM)
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD X2
ADDRESS
ADDRESS
WORD Y2
WORD Y1
WORD Y0
WORD X1
WORD X0
NO ACCESS
63-48
47-32
LONG WORD ACCESS
31-16
15-0
63-48
PM DATA
BUS
PEX REGISTERS
39-24
RB
23-8
7-0
RA
39-24
23-8
7-0
PEY REGISTERS
SB
23-8
7-0
23-8
7-0
15-0
RY
39-24
23-8
SA
39-24
31-16
WORD X0
WORD X0, 63-32
39-24
47-32
DM DATA
BUS
7-0
0X00
RX
39-24
23-8
WORD X0, 31-0
SY
39-24
23-8
WORD X0, 63-32
7-0
0X00
7-0
0X00
SX
39-24
23-8
WORD X0, 31-0
7-0
0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(LONG WORD X0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR BROADCAST, LONG WORD, SINGLE-DATA TRANSFERS ARE:
UREG = PM(LONG WORD ADDRESS);
UREG = DM(LONG WORD ADDRESS);
PM(LONG WORD ADDRESS) = UREG;
DM(LONG WORD ADDRESS) = UREG;
Figure 5-36. Long Word Addressing of Single Data in Broadcast Load
5-82
ADSP-21160 SHARC DSP Hardware Reference
Memory
MEMORY
BLOCK 0 (PM)
BLOCK 1 (DM)
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
WORD X2
ADDRESS
ADDRESS
WORD Y2
WORD Y1
WORD X1
WORD X0
WORD Y0
LONG WORD ACCESS
LONG WORD ACCESS
63-48
47-32
PM DATA
BUS
PEX REGISTERS
39-24
15-0
63-48
PEY REGISTERS
7-0
0X00
23-8
WORD Y0, 31-0
SB
23-8
WORD Y0, 63-32
7-0
0X00
23-8
WORD Y0, 31-0
15-0
RY
7-0
39-24
0X00
23-8
RX
7-0
39-24
WORD X0, 63-32
0X00
00
WORD X0, 31-0
0X00
7-0
39-24
7-0
39-24
7-0
0X00
WORD X0, 63-32
SA
39-24
31-16
WORD X0
RA
39-24
47-32
DM DATA
BUS
RB
23-8
WORD Y0, 63-32
39-24
31-16
WORD Y0
23-8
SY
23-8
0X00
7-0
SX
23-8
WORD X0, 31-0
0X00
THE ABOVE EXAMPLE SHOWS THE DATA FLOW FOR INSTRUCTION:
RX = DM(LONG WORD X0 ADDRESS), RA = PM(LONG WORD Y0 ADDRESS);
OTHER INSTRUCTIONS WITH SIMILAR DATA FLOWS FOR BROADCAST, LONG WORD, DUAL-DATA TRANSFERS ARE:
DREG = PM(LONG WORD ADDRESS),
PM(LONG WORD ADDRESS) = DREG,
DREG = DM(LONG WORD ADDRESS);
DM(LONG WORD ADDRESS) = DREG;
Figure 5-37. Long Word Addressing of Dual Data in Broadcast Load
ADSP-21160 SHARC DSP Hardware Reference
5-83
Arranging Data in Memory
Arranging Data in Memory
Each DSP’s access to internal memory gets data from either a 4-columns
(Long, Normal, or Short word) or 3-columns (instruction or extended
precision Normal word) memory location. For more information on how
the DSP accesses 4- or 3-column data, see “Memory Organization and
Word Size” on page 5-22.
To take advantage of the DSP’s data accesses to 4- and 3-column locations, programs must adjust the interleaving of data into memory
locations to accommodate the memory access mode. The following guidelines provide an overview of how programs should interleave data in
memory locations. For more information and examples, see ADSP-21160
SHARC DSP Instruction Set Reference:
• Programs can use odd or even modify values (1, 2, 3, …) to step
through a buffer in single-or dual-data, SISD or Broadcast load
mode regardless of the data word size (Long word, extended precision Normal word, Normal word, or Short word).
• Programs should use multiple of 4 modify values (4, 8, 12, …) to
step through a buffer of Short word data in single-or dual-data,
SIMD mode.
• Programs should use multiple of 2 modify values (2, 4, 6, …) to
step through a buffer of Normal word data in single- or dual-data
SIMD mode.
• Programs can use odd or even modify values (1, 2, 3, …) to step
through a buffer of Long word or extended precision Normal
word data in single- or dual-data, SIMD mode.
5-84
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
6 I/O PROCESSOR
The DSP’s I/O processor manages Direct Memory Accessing (DMA) of
DSP memory through the external, link, and serial ports. Each DMA
operation transfers an entire block of data. By managing DMA, the I/O
processor lets programs move data as a background task while using the
processor core for other DSP operations.
Overview
The I/O processor’s architecture, which appears in Figure 6-1 on
page 6-4, supports a number of DMA operations. These operations
include the following transfer types:
• Internal memory  external memory or external peripherals
• Internal memory  internal memory of other DSPs
• Internal memory  host processor
• Internal memory  serial port I/O
ADSP-21160 SHARC DSP Hardware Reference
6-1
Overview
• Internal memory  link port I/O
• External memory  external peripherals
chapter describes the I/O processor and how the I/O proces This
sor controls external port, link port, and serial port operations. For
information on connecting external devices to the external port,
link ports, or serial ports, see “External Port”, “Link Ports”, or
“Serial Ports”.
DMA transfers between internal memory and external memory, multiprocessor memory, or a host use the DSP’s external port. For these types of
transfers, a program sets up the DMA controller with the internal memory
buffer size and address, the address modifier, and the direction of transfer.
These DMA set up parameters are the Transfer Control Block (TCB) for
the DMA transfer. After setup, the DMA transfers begins when the program enables the channel and continues until the I/O processor transfers
the entire buffer to or from DSP memory.
Similarly, DMA transfers between internal memory and link or serial
ports have DMA parameters (a TCB). When the I/O processor performs
DMA between internal memory and one of these ports, the program sets
up the parameters and the I/O goes through the port instead of the external bus.
The direction (receive or transmit) of the I/O port determines the direction of data transfer. When the port receives data, the I/O processor
automatically transfers the data to internal memory. When the port needs
to transmit a word, the I/O processor automatically fetches the data from
internal memory.
The I/O processor also lets the DSP system perform DMA transfers
between an external device and external memory. This external to external
transfer only uses the external port and I/O processor.
6-2
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
External devices can control external port DMA transfers in two ways. If
the external device can handle bus mastership, the external device can
master reads or writes to DMA buffers on the DSP. External devices also
can assert a DMA Request input (DMARx) to request service.
To further minimize loading on the processor core, the I/O processor supports chained DMA operations. When using chained DMA, a program
can set up a DMA transfer to automatically set up and start the next DMA
transfer after the current one completes.
Figure 6-1 shows the DSP’s I/O processor, related ports, and buses.
Figure 6-8 on page 6-68 shows more detail on DMA channel data paths.
The Data Buffer Registers column in Figure 6-1 shows the data buffer registers for each port. These registers include:
• External Port Buffer registers (EBPx). These 64-bit buffers for the
external port have eight-position FIFOs for transmitting or receiving data when interfacing with a host or external devices such as
memory and memory mapped devices.
• Link Port Buffer registers (LBUFx). These buffers for the link ports
have two-position FIFOs for transmitting or receiving DMA data
when connected to another link port.
• Serial Port Receive Buffer registers (RXx). These receive buffers for
the serial ports have two-position FIFOs for receiving data when
connected to another serial device.
• Serial Port Transmit Buffer registers (TXx). These transmit buffers
for the serial ports have two position FIFOs for transmitting data
when connected to another serial device.
ADSP-21160 SHARC DSP Hardware Reference
6-3
Overview
INTERNAL MEMORY ADDRESS
INTERNAL MEMORY DATA
IOA BUS
IOD BUS
IRPTL, LIRPTL
DMASTAT
SERIAL
PORTS 1-0
II3-0, IM3-0,
C3-0, CP3-0,
GP3-0, DA3-0,
DB3-0
SRCTL1-0,
STCTL1-0
TX1-0, RX1-0
LINK
PORTS 5-0
II9-4, IM9-4,
C9-4, CP9-4,
GP9-4, DA9-4,
DB9-4
LCTL1-0, LAR,
LCOM
LBUF5-0
EXTERNAL
PORT DATA
II13-10, IM13-10,
C13-10, CP13-10,
GP13-10, EI13-10,
EM13-10,
EC13-10
SYSCON,
WAIT,
DMAC13-10,
EP3-0
DMARx
DMAGx
DMA
PARAMETER
REGISTERS
PORT, BUFFER, &
DMA CONTROL
REGISTERS
BUFFER
DATA
REGISTERS
EXTERNAL
PORT
ADDRESS
Figure 6-1. I/O Processor Block Diagram
6-4
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
The Port, Buffer, and DMA Control Registers column in Figure 6-1 on
page 6-4 shows the control registers for the ports and DMA channels.
These registers include:
• System Configuration register (SYSCON). This register configures
packing, priority, and word order for the external port.
• Waitstate and Access Mode register (WAIT). This register configures handshake, idle cycle insertion, and waitstate insertion for
external memory DMA accesses.
• External Port DMA Control registers (DMACx). These control registers for each external port DMA channel select the direction,
format, and handshake and enable chaining, transfer mode, and
DMA start.
• Link Port Common Controls register (LCOM). This register indicates link buffer packing and error status for link port operations.
• Link Port Assignment register (LAR). This register assigns link buffers to link ports for link port operations.
• Link Port Control registers (LCTLx). These control registers (each
register controls three link buffers) select the direction, word
width, and transfer rate and enable chaining, 2-D DMA mode, and
DMA start.
• Serial Port Receive Control registers (SRCTLx). These control registers for each port select the receive format; monitor FIFO status;
and enable chaining, 2-D DMA mode, and DMA start.
• Serial Port Transmit Control registers (STCTLx). These control registers for each port select the transmit format; monitor FIFO
status; and enable chaining, 2-D DMA mode, and DMA start.
ADSP-21160 SHARC DSP Hardware Reference
6-5
Overview
The DMA Parameter Registers column in Figure 6-2 on page 6-7 shows
the parameter registers for each DMA channel. These registers function
similarly to data address generator registers and include:
• Internal Index registers (IIx). An index register provides an internal memory address, acting as a pointer to the next internal
memory DMA read or write location.
• Internal Modify registers (IMx). A modify register provides the
signed increment by which the DMA controller post-modifies the
corresponding internal memory index register after the DMA read
or write.
• Count registers (Cx). A count register indicates the number of
words remaining to be transferred to or from internal memory on
the corresponding DMA channel.
• Chain Pointer registers (CPx). A chain pointer register holds the
starting address of the Transfer Control Block (parameter register
values) for the next DMA operation on the corresponding channel.
These registers also control whether the I/O processor generates an
interrupt when the current DMA process ends.
• General Purpose registers (GPx). A general purpose DMA register
holds an address or other value.
• Dimension A and B registers (DAx and DBx). Dimension registers
hold the counts for the A and B dimensions of a 2-dimensional
DMA. For more information on two-dimensional DMA, see
“Using Two-Dimensional Link Port DMA” on page 6-83 or
“Using Two-Dimensional Serial Port DMA” on page 6-91.
• External Index registers (EIx). An index register provides an external memory address, acting as a pointer to the next external
memory DMA read or write location.
6-6
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
• External Modify registers (EMx). A modify register provides the
increment by which the DMA controller post-modifies the corresponding external memory index register after the DMA read or
write.
• External Count registers (ECx). An external count register indicates
the number of words remaining to be transferred to or from external memory on the corresponding DMA channel.
R egister
Function
W idth
IIx
Internal Index R egister
18-bits*
D escription
A ddress of buffer in internal m em ory
IM x
Internal M odify R egister
16-bits
S tride for internal buffer
Cx
Internal C ount R egister
16-bits
Length of internal buffer
CPx
C hain Pointer R egister
19-bits*
C hain pointer for D M A chaining
GPx
G eneral P urpose R egister
18-bits
U ser definable
DBx
D im ension B R egister
16-bits
C ount of dim ension B buffer
DAx
D im ension A R egister
16-bits
C ount of dim ension A buffer
E Ix
E xternal Index R egister
32-bits
A ddress of buffer in external m em ory
EMx
E xternal M odify R egister
32-bits
S tride for external buffer
ECx
E xternal C ount R egister
32-bits
Length of external buffer
E xtern a l P o rt D M A c ha n n e ls o n ly
* O ffset by 0x40000 for internal addressing in norm al w ord space
Figure 6-2. ADSP-21160 DSP’s DMA Parameter Register
Figure 6-3 on page 6-8 shows a block diagram of the I/O processor’s
address generator (DMA controller). Table 6-1 on page 6-12 lists the
parameter registers for each DMA channel. The parameter registers are
uninitialized following a processor reset.
ADSP-21160 SHARC DSP Hardware Reference
6-7
Overview
DMA ADDRESS GENERATOR (INTERNAL ADDRESSES)
LOCAL BUS
INTERNAL
MEMORY
ADDRESS
II X
INDEX (ADDRESS)
IMX
MODIFIER
DBX
ONLY FOR 2-D DMA
+
MUX
POST-MODIFY
DMA WORD COUNTER
CX
COUNT
–1
LOCAL BUS
CPX
CHAIN POI NT ER
GPX
GENERAL PURPOSE
WORKING REG ISTER
FO R 2-D DMA
DAX
+
ONLY FOR 2-D DMA
MUX
DMA ADDRESS GENERATOR (EXTERNAL ADDRESSES)
LOCAL BUS
EXTERNAL
MEMORY
ADDRESS
EIX
EXT. INDEX (ADDRESS)
EMX
ECX
EXT. MODI FIER
EXT. COUNT
–1
+
+
POST-MODI FY
Figure 6-3. DMA Address Generator
The I/O processor generates addresses for DMA channels much the same
way that the Data Address Generators (DAGs) generate addresses for data
memory accesses. Each channel has a set of parameter registers including
an index register (IIx) and modify register (IMx) that the I/O processor
uses to address a data buffer in internal memory. The index register must
6-8
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
be initialized with a starting address for the data buffer. As part of the
DMA operation, the I/O processor outputs the address in the index register onto the DSP’s IO (I/O Address) bus and applies the address to
internal memory during each DMA cycle—a clock cycle in which a DMA
transfer is taking place.)
All addresses in the index (IIx) registers are offset by a value matching the
DSP’s first internal Normal word addressed RAM location, before the I/O
processor uses the addresses. For the ADSP-21160, this offset value is
0x0004 0000.
While DMA addresses must always be Normal word (32-bit) memory, the
internal memory data transfer sizes may be 64-, 48-, or 32-bits. External
memory data transfer sizes may be 64-, 32-, or 16-bits. The I/O processor
can transfer Short word data (16-bit) using the packing capability of the
external port and serial port DMA channels.
After transferring each data word to or from internal memory, the I/O
processor adds the modify value to the index register to generate the
address for the next DMA transfer and writes the modified index value to
the index register. The modify value in the IMx register is a signed integer,
which allows both increment and decrement modifies.
the I/O processor modifies the index register past the maximum
 If18-bit
value to indicate an address out of internal memory, the
index wraps around to zero. With the offset for the ADSP-21160
DSP, the wraparound address is 0x0004 0000.
Each DMA channel has a count register (Cx) that programs load with a
word count to be transferred. The I/O processor decrements the count
register after each DMA transfer on that channel. When the count reaches
zero, the I/O processor generates the interrupt for that DMA channel. For
more information on DMA interrupts, see “Using I/O Processor Status”
on page 6-53.
ADSP-21160 SHARC DSP Hardware Reference
6-9
Overview
program loads the count ( ) register with zero, the I/O proces sorIf a does
not disable DMA transfers on that channel. The I/O
Cx
processor interprets the zero as a request for 216 transfers. This
count occurs because the I/O processor starts the first transfer
before the testing the count value. The only way to disable a DMA
channel is to clear its DMA enable bit. For more information, see
“External Port Channel Transfer Modes” on page 6-21, “Link Port
Channel Transfer Modes” on page 6-48, or “Serial Port Channel
Transfer Modes” on page 6-52.
Each DMA channel also has a chain pointer register (CPx) and a general-purpose register (GPx). Chained DMA sequences are a set of multiple
DMA sequences, each autoinitializing the next in line. The location of the
parameters for the next sequence comes from the CPx register. These
parameters are called a Transfer Control Block, and they set up DMA
parameter values for autoinitializing the next DMA sequence in the chain.
Programs can use the GP register for any purpose, but usually programs
store the address of the previous TCB in this register during chained
DMA. For more information, see “Chaining DMA Processes” on
page 6-69.
The external port DMA channels each contain three additional parameter
registers, the external index register (EIx), external modify register (EMx),
and external count register (ECx). These three registers are not available for
the serial port and link port DMA channels. The I/O processor generates
32-bit external memory addresses using the EI, EM, and EC registers, during
DMA transfers between internal memory and external memory or devices.
Programs must load the register with the count of external bus
 transfers
in the DMA. If the external port is using word packing,
EC
the EC count differs from the number of words transferred in the
DMA.
Instead of the EI, EM, and EC register, the serial port and link port DMA
channels have the Dimension-A (DA) and Dimension-B (DB) registers. The
I/O processor uses these registers for dimension indices during
6-10
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
two-dimensional DMA operations. In one-dimensional DMA operations,
programs may also use DA and DB as general-purpose registers. For more
information, see “Using Two-Dimensional Link Port DMA” on page 6-83
or “Using Two-Dimensional Serial Port DMA” on page 6-91.
Memory mapped devices can communicate with the I/O processor using
an internal DMA request/grant handshake on an external port DMA
channel. Each channel has a single request and a single grant.
When a particular I/O port needs to perform transfers to or from internal
memory, the channel asserts a request. The I/O processor prioritizes this
request with all other valid DMA requests. The default channel priority is
DMA channel 0 as highest and DMA channel 13 as lowest. Table 6-1 on
page 6-12 lists the DMA channels in priority order. For more information, see “Managing DMA Channel Priority” on page 6-67.
When a channel becomes the highest priority requester, the I/O processor
services the channel’s request. In the next clock cycle, the I/O processor
starts the DMA transfer.
a DMA channel is disabled, the I/O processor does not service
 Ifrequests
for that channel, whether or not the channel has data to
transfer.
The DSP’s 14 DMA channels are numbered as shown in Table 6-1. This
table also shows the control, parameter, priority (DMA channel zero is
highest and channel 13 lowest) and data buffer registers that correspond
to each channel.
ADSP-21160 SHARC DSP Hardware Reference
6-11
Overview
Table 6-1. DMA Channel Registers: Controls, Parameters and
Buffers
DMA
Chan#
Control
Registers
Parameter Registers
Buffer
Register
Description
Channel
Priority
0
SRCTL0
II0, IM0, C0, CP0, GP0,
DB0, DA0
RX0
Serial Port 0
Receive
Highest
Priority
1
SRCTL1
II1, IM1, C1, CP1, GP1,
DB1, DA1
RX1
Serial Port 1
Receive
2
STCTL0
II2, IM2, C2, CP2, GP2,
DB2, DA2
TX0
Serial Port 0
Transmit
3
STCTL1
II3, IM3, C3, CP3, GP3,
DB3, DA3
TX1
Serial Port 1
Transmit
II4, IM4, C4, CP4, GP4,
DB4, DA4
LBUF0
Link Buffer 0
II5, IM5, C5, CP5, GP5,
DB5, DA5
LBUF1
Link Buffer 1
II6, IM6, C6, CP6, GP6,
DB6, DA6
LBUF2
Link Buffer 2
II7, IM7, C7, CP7, GP7,
DB7, DA7
LBUF3
Link Buffer 3
II8, IM8, C8, CP8, GP8,
DB8, DA8
LBUF4
Link Buffer 4
II9, IM9, C9, CP9, GP9,
DB9, DA9
LBUF5
Link Buffer 5
-
TCB Chain Loading
Requests1
-
External Accesses of Internal Memory
(Direct Reads, Direct
Writes) and IOP Registers2
4
5
LCTL0,
LAR,
LCOM
6
7
8
LCTL1,
LAR,
LCOM
9
10
DMAC10
II10, IM10, C10, CP10,
GP10, EI10, EM10, EC10
EPB0
Ext. Port FIFO
Buffer 0
113
DMAC11
II11, IM11, C11, CP11,
GP11, EI11, EM11, EC11
EPB1
Ext. Port FIFO
Buffer 1
6-12
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Table 6-1. DMA Channel Registers: Controls, Parameters and
Buffers (Cont’d)
DMA
Chan#
Control
Registers
Parameter Registers
Buffer
Register
Description
Channel
Priority
124
DMAC12
II12, IM12, C12, CP12,
GP12, EI12, EM12, EC12
EPB2
Ext. Port FIFO
Buffer 2
13
DMAC13
II13, IM13, C13, CP13,
GP13, EI13, EM13, EC13
EPB3
Ext. Port FIFO Lowest
Buffer 3
Priority
1
TCB chain loading is not associated with a specific DMA channel. TCB chain loading uses the
I/O bus and requires prioritization.
2 Direct reads and writes are not associated with a specific DMA channel. Direct reads and writes
use the I/O bus and require prioritization.
3 The DMAR1 and DMAG1 pins are handshake controls for DMA channel 11.
4 The DMAR2 and DMAG2 pins are handshake controls for DMA channel 12.
All of the I/O processor’s registers are memory-mapped in the DSP’s
internal memory, ranging from address 0x0000 0000 to 0x0000 00FF.
For more information on these registers, see “I/O Processor Registers” on
page A-33.
Because the I/O processor registers are memory-mapped, the DSP and
external processors (host or multiprocessor DSPs) have access to program
DMA operations. A processor sets up a DMA channel by writing the
transfer’s parameters to the DMA parameter registers. After the IIx, IMx,
and Cx registers (among others) are loaded with a starting source or destination address, an address modifier, and a word count, the processor is
ready to start the DMA.
The external ports, link ports, and serial ports each have a DMA enable bit
(DEN, LxDEN, or SDEN) in their channel control register. Setting this bit for a
DMA channel with configured DMA parameters starts the DMA on that
channel. If the parameters configure the channel to receive, the I/O processor transfers data words received at the buffer to the destination in
internal memory. If the parameters configure the channel to transmit, the
I/O processor transfers a word automatically from the source memory to
ADSP-21160 SHARC DSP Hardware Reference
6-13
Setting I/O Processor—EPort Modes
the channel’s buffer register. These transfers continue until the I/O processor transfers the selected number of words (count parameter).
start a new (non-chained) DMA sequence after the current one
 Tois finished,
programs must disable the channel (clear its
bit);
DEN
write new parameters to the II, IM, and C registers; then enable the
channel (set its DEN bit). For chained DMA operations, this disable-enable process is not necessary. For more information, see
“Chaining DMA Processes” on page 6-69.
Setting I/O Processor—EPort Modes
The SYSCON, WAIT, and DMACx registers control the external port operating
mode for the I/O processor. Table A-17 on page A-45 lists all the bits in
SYSCON, Table A-19 on page A-49 lists all the bits in WAIT, and Table A-21
on page A-55 lists all the bits in DMACx.
The following bits control external port I/O processor modes. Except for
the FLSH bit, the control bits in the DMACx registers have a one cycle effect
latency (take effect on the second cycle after change). The FLSH bit has a
two cycle effect latency. Programs should not modify an active DMA
channel’s DMACx register; other than to disable the channel by clearing the
DEN bit. For information on verifying a channel’s status with the DMASTAT
register, see “Using I/O Processor Status” on page 6-53.
Some other bits in SYSCON, WAIT, and DMACx setup non-DMA external port
features. For information on these features, see “Setting External Port
Modes” on page 7-2.
• Boot Select Override. SYSCON Bit 1 (BSO) This bit enables (if set,
=1) or disables (if cleared, =0) access to Boot Memory Space. When
BSO is set, the DSP uses the BMS select line (instead of MS3-0) to perform DMA channel 10 accesses of external memory. When BSO is
set, BMS will be asserted low for an external port DMA transfer and
6-14
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
all memory selects (MSx) will be disabled for DMA transfers. However, the memory selects are available (not disabled) to the DSP
core for external memory accesses.
• Host Packing Mode. SYSCON Bits 6-5 (HPM) These bits select the
external bus packing mode for host accesses:
000=no packing, 001=16-to-32/64, 010=16-to-48 (reset value),
011=32-to-48, 100=32-to-32/64.
• Host Most Significant Word First Packing Select. SYSCON Bit 7
(HMSWF) This bit selects the word packing order for host accesses as
most-significant-word first (if set, =1) or least-significant-word first
(if cleared, =0).
• Buffer Hang Disable. SYSCON Bit 16 (BHD) This bit controls
whether the processor core proceeds (hang disabled if set, =1) or is
held-off (hang enabled if cleared, =0) when the core tries to read
from an empty EPBx, TXx, or LBUFx buffer or tries to write to a full
EPBx, RXx, or LBUFx buffer.
• External Port DMA Channel Priority Rotation Enable. SYSCON Bit
19 (DCPR) This bit enables (rotates if set, =1) or disables (fixed if
cleared, =0) priority rotation among external port DMA channels
(channel 10-13).
• Handshake and Idle for DMA Enable. WAIT Bit 30 (HIDMA) This
bit enables (if set, =1) or disables (if cleared, =0) adding an idle
cycle after every memory access for DMAs with handshaking
(DMARx-DMAGx).
• External Port DMA Enable. DMACx Bit 0 (DEN) This bit enables (if
set, =1) or disables (if cleared, =0) DMA for the corresponding
external port FIFO buffer (EPBx).
• External Port DMA Chaining Enable. DMACx Bit 1 (CHEN) This bit
enables (if set, =1) or disables (if cleared, =0) DMA chaining for
the corresponding external port FIFO buffer (EPBx).
ADSP-21160 SHARC DSP Hardware Reference
6-15
Setting I/O Processor—EPort Modes
• External Port Transmit/Receive Select. DMACx Bit 2 (TRAN) This bit
selects the transfer direction (transmit if set, =1) (receive if cleared,
=0) for the corresponding external port FIFO buffer (EPBx).
• External Port Data Type Select. DMACx Bit 5 (DTYPE) This bit
selects the transfer data type (40/48=bit, 3-column if set, =1)
(32/64-bit, 4-column if cleared, =0) for the corresponding external
port FIFO buffer (EPBx).
• External Port Packing Mode. DMACx Bits 8-6 (PMODE) These bits
select the packing mode for the corresponding external port FIFO
buffer (EPBx) as follows: 000=No pack, 001=16 external to 32/64
internal packing, 010=16 external to 48 internal packing, 011=32
external to 48 internal packing, 100=32 external to 32/64 internal
packing, 101=110=111=reserved.
• Most Significant 16-bit Word First during packing. DMACx Bit 9
(MSWF) When the buffer’s PMODE is 001 or 010, this bit selects the
packing order of 16-bit words (most significant first set, =1) (least
significant first cleared, =0) for the corresponding external port
FIFO buffer (EPBx).
• Master Mode Enable. DMACx Bit 10 (MASTER) This bit enables (if
set, =1) or disables (if cleared, =0) master mode for the corresponding external port FIFO buffer (EPBx).
• Handshake Mode Enable. DMACx Bit 11 (HSHAKE) This bit enables
(if set, =1) or disables (if cleared, =0) handshake mode for the corresponding external port FIFO buffer (EPBx).
• External Handshake Mode Enable. DMACx Bit 13 (EXTERN) This bit
enables (if set, =1) or disables (if cleared, =0) external handshake
mode for the corresponding external port FIFO buffer (EPBx).
• External Port Bus Priority. DMACx Bit 15 (PRIO) This bit selects the
external bus access priority level (high if set, =1) (low if cleared, =0)
for the corresponding external port FIFO buffer (EPBx).
6-16
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Boot Memory DMA Mode
The BSO bit in the SYSCON register enables Boot Memory Select Override—
a mode in which the I/O processor supports DMA access to boot memory
space. When BSO is set, the DSP uses the BMS select line (instead of MS3-0)
to perform DMA channel 10 accesses of external memory.
When reading from 8-bit boot memory space, the DSP uses 8-to-48-bit
packing. Programs most often use this feature to finish loading programs
and data after the DSP completes its automatic 256-instruction boot-load.
When writing to 8-bit boot memory space, programs must use the shifter
to place ordered bytes for the transfer in bits 39-32 of each long word to
be written. Programs must use the shifter because there is no 8-to-48-bit
packing mode for external port writes. Programs most often use this feature to update writable boot memory data (flash memory or EEPROM).
External Port Buffer Modes
The HPM, HMSWF, PMODE, MSWF, and BHD bits in the SYSCON and DMACx registers select a buffer’s packing mode and disable buffer not-ready processor
core stalls. The packing mode bits (PMODE for DSP and HPM for host) select
the external bus width and word size for transfers. Table 6-2 shows the
available settings. Packed data or instructions are arranged in external
memory according to the memory address that stems from their word size.
For more information, see “Memory Organization and Word Size” on
page 5-22. When data or instructions in external memory are not packed,
ADSP-21160 SHARC DSP Hardware Reference
6-17
Setting I/O Processor—EPort Modes
the words are arranged in memory according to the external bus’ data
alignment. This data alignment appears in Figure 7-1 on page 7-2.
Table 6-2. DSP (PMODE) and Host (HPM) Packing Modes
PMODE or HPM
Packing Mode
000
No word packing (64-bit bus, 64-bit words)
001
16-to32/64-bit packing (16-bit bus, 32- or 64-bit words)
010
16-to-40/48-bit packing (16-bit bus, 40- or 48-bit words)
011
32-to-40/48-bit packing (32-bit bus, 40- or 48-bit words)
100
32-to-32/64-bit packing (32-bit bus, 32- or 64-bit words)
When packing is enabled, the DSP only uses the
and
 strobes
for accessing external memory, regardless of the least-signifRDH
WRH
icant-bit of the address.
The DSP (
) and host ( ) packing modes must match for
 correct
word-packing operations in host systems.
PMODE
HPM
When the packing mode (PMODE or HPM) is set for a 16-bit bus, programs
should set up the 16-bit word order. The 16-bit word order bits ( MSWF for
DSP and HMSWF for host) control the order of 16-bit words being packed
or unpacked in the 32-, 48-, or 64-bit word being transferred. If the MSWF
or HMSWF bit is set, the packing and unpacking is Most significant 16-bit
word first.
In addition to selecting the packing mode for external port DSP transfers,
programs must indicate the type of data in the transfer, using the Data
Type (DTYPE) bit. For more information, see “External Port Channel
Transfer Modes” on page 6-21.
The Buffer Hang Disable (BHD) bit lets the processor core proceed if the
core tries to read from an empty EPBx, TXx, or LBUFx buffer or tries to write
to a full EPBx, RXx, or LBUFx buffer. The processor core still performs buffer accesses when buffer hang is disabled (BHD=1). If the processor core
6-18
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
attempts to read from an empty receive buffer, the core gets a repeat of the
last value that was in the buffer. If the processor core attempts to write to
a full buffer, the core overwrites the last value that was written to the
buffer. Because these buffers are not initialized at reset, a read from a buffer that hasn't been filled since the reset returns an undefined value.
If an external port buffer’s
bit is set and DMA for that chan nel
is not enabled, the external port channel is in single-word,
INTIO
interrupt-driven transfer mode. For more information, see “Using
I/O Processor Status” on page 6-53.
External Port Channel Priority Modes
The DCPR and PRIO bits in the SYSCON and DMACx registers influence priority levels for an external port buffer and the external port in relation to
external port DMA channels and external bus arbitration. For more information on prioritization operations, see “Managing DMA Channel
Priority” on page 6-67.
Priority for DMA requests from external port channels can be fixed or
rotated. When the DMA Channel Priority Rotate (DCPR) bit is cleared, the
lowest number external port channel has the highest priority, ranging
from highest-priority channel 10 to lowest-priority channel 13.
When the DCPR bit is set, the priority levels rotate. High priority shifts to a
new channel after each single-word transfer. The I/O processor services a
single-word transfer then rotates priority to the next higher numbered
channel. Rotation continues until the I/O processor services all four
external port channels. Figure 6-4 illustrates this process as described in
the following steps:
1. At reset, external port channels have priority order—from high to
low—10, 11, 12, and 13.
2. The external port performs a single transfer on channel 11.
ADSP-21160 SHARC DSP Hardware Reference
6-19
Setting I/O Processor—EPort Modes
3. The I/O processor rotates channel priority, changing it to 12, 13,
10, and 11 (because rotating priority is enabled for this example,
DCPR=1).
LOWEST
PRIORITY
13
HIGHEST
PRIORITY
HIGHEST
PRIORITY
10
12
STEP 2
12
11
LOWEST
PRIORITY
11
STEP 3
13
10
ONE TRANSFER OCCURS ON CHANNEL 11 (STEP 2),
ROTATING CHANNEL 11'S PRIORITY TO THE LOWEST PRIORITY SLOT (STEP 3).
Figure 6-4. Rotating External Port DMA Channel Priority
though the external port channel DMA priority can rotate,
 Even
the interrupt priorities of all DMA channels are fixed.
When external port DMA channel priority is fixed (DCPR=0), channel 10
has the highest priority, and channel 13 has the lowest priority. But, programs can redefine this priority order by assigning one of the other
channels the highest priority. To change the fixed priority sequence of the
external port DMA channels, a program could use the following
procedure:
6-20
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
1. Disable all external port DMA channels except the one which is to
have lowest priority.
2. Select rotating priority.
3. Cause at least one transfer to occur on the enabled channel.
4. Disable rotating priority and re-enable all of the external port
DMA channels
After completing this procedure, the channel immediately after the
selected channel has the highest fixed priority.
In systems where multiple processors are using the external bus, the PRIO
bit raises the priority level for external port DMA transfers. When a channel’s PRIO bit is set, the I/O processor asserts the Priority Access (PA) pin
when that channel uses the external bus. The channel gets higher priority
in bus arbitration, allowing the DMA to complete more quickly.
Programs can also rotate priority between external port and link port
DMA channels. For more information, see “Link Port Channel Priority
Modes” on page 6-45.
External Port Channel Transfer Modes
The DEN, CHEN, TRAN, and DTYPE bits in the DMACx register enable DMA and
chained DMA and select the transfer direction and data type. The DMA
enable (DEN) and Chained DMA enable (CHEN) bits work together to select
an external port DMA channel’s transfer mode. Table 6-3 lists the modes.
ADSP-21160 SHARC DSP Hardware Reference
6-21
Setting I/O Processor—EPort Modes
Table 6-3. External Port DMA Enable Modes
CHEN
DEN
DMA Enable Mode Description
0
0
Channel disabled (chaining disabled, DMA disabled)
0
1
Single DMA mode (chaining disabled, DMA enabled)
1
0
Chain insertion mode (chaining enabled, DMA enabled, auto-chaining disabled); For more information, see “Chaining DMA Processes”
on page 6-69.
1
1
Chained DMA mode (chaining enabled, DMA enabled, auto-chaining enabled)
Because the external port is bidirectional, the I/O processor uses the
Transmit select (TRAN) bit to determine the transfer direction (transmit or
receive). Data flows from internal to external memory when in transmit
mode. In transmit mode, the I/O processor fills the channel’s EPBx buffer
when the channel’s DEN bit is set.
The Data Type (DTYPE) bit determines how the DMA channel accesses
columns of internal memory. If DTYPE is set, the data is 40- or 48-bit
words, and the I/O processor makes 3-column internal memory accesses.
If DTYPE is cleared, the data is 32- or 64-bit words, and the I/O processor
makes 4-column internal memory accesses. For more information, see
“Memory Organization and Word Size” on page 5-22.
 The
Width (
for the transfer overrides the Internal Memory Data
IMDWx) setting for the internal memory block.
DTYPE
External Port Channel Handshake Modes
The MASTER, HSHAKE, EXTERN, and HIDMA bits in the DMACx and WAIT registers select the channel’s DMA handshake and enable the hold cycles for
host DMA. Table 6-4 shows how the MASTER, HSHAKE, and EXTERN bits
work to select the channel’s DMA handshake mode.
6-22
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Table 6-4. External Port DMA Handshake Modes—DMACx MASTER
(M), HSHAKE (H), and EXTERN (E) Bits
EHM
DMA Mode of Operation
000
Slave Mode. The DSP responds to the buffer’s internal memory transfer activity
based on the buffer status in the FS field, generating a DMA request whenever the
buffer is not empty (on receive) or is not full (on transmit). During transmit
(TRAN=1), the DSP fills the EPBx buffer when the program enables the buffer
(DEN=1).
For more information, see “Slave Mode” on page 6-31.
001
Master Mode. The DSP attempts the internal memory DMA transfers indicated by
the DMA counter (Cx) based on the buffer status in the FS field, making transfers
whenever the buffer is not empty (on receive) or is not full (on transmit).
Systems using Master Mode should deassert corresponding DMA request inputs,
deasserting DMAR1 if channel 11 is in master mode and deasserting DMAR2 if channel
12 is in master mode.
For more information, see “Master Mode” on page 6-25.
010
Handshake Mode. When in this mode, the DSP generates a DMA request whenever
the external device asserts the DMARx pin, then the DSP asserts the DMAGx pin, transferring the data (and deasserting DMAGx) when the external devices deasserts the
DMARx pin.
Note that this mode only applies to external port buffers EPB1 and EPB2 and only
applies to DMA channels 11 and 12.
For more information, see “Handshake Mode” on page 6-34.
011
Paced Master Mode. The DSP attempts the internal memory DMA transfers indicated by the DMA counter (Cx), making transfers based on external DMA request
inputs. The DSP generates a DMA request whenever the external device asserts the
DMARx pin and controls the data transfer using the RDH/L or WRH/L and ACK pins
and applying the selected number of waitstates.
Note that this mode only applies to DMA channels 11 and 12.
For more information, see “Paced Master Mode” on page 6-30.
100
Reserved
101
Reserved
ADSP-21160 SHARC DSP Hardware Reference
6-23
Setting I/O Processor—EPort Modes
Table 6-4. External Port DMA Handshake Modes—DMACx MASTER
(M), HSHAKE (H), and EXTERN (E) Bits (Cont’d)
EHM
DMA Mode of Operation
110
External-Handshake Mode. The DSP responds to external memory DMA requests
based on external DMA request inputs. This mode is identical to Handshake Mode,
but applies to transfers between external memory and external devices.
When in this mode, the DSP generates a DMA request whenever the external device
asserts the DMARx pin, then the DSP asserts the DMAGx pin, transferring the data
(and deasserting DMAGx) when the external devices deasserts the DMARx pin.
Note that this mode only applies to external port buffers EPB1 and EPB2 and only
applies to DMA channels 11 and 12.
For more information, see “External-Handshake Mode” on page 6-40.
111
Reserved
For the Handshake and External-handshake modes shown in Table 6-4,
programs can insert an added idle cycle after every memory access. The
Handshake and Idle for DMA (HIDMA) bit in the WAIT register enables this
added cycle, which reduces bus contention from devices with slow
three-state timing or long recovery times.
Because external port DMA transfers can go between DSP internal memory and external memory, the I/O processor must generate addresses for
both memory spaces. The external port DMA channels have additional
parameter registers (EIx, EMx, ECx) for external memory access.
To support data packing options for external memory DMA transfers, the
EI and EM registers can generate addresses at a different rate than the internal address registers (II and IM). This separation is shown in Figure 6-8 on
page 6-68, which shows that the I/O processor has separate address generators for internal and external addresses. For this reason when packing is
used for external memory DMA, the external count (EC) register indicates
the number of external port transfers, not necessarily the number of internal memory words being transferred.
The DMA mode and other factors determine the size of the DMA data
transfer on the external port. These other factors include the EI, EM, and
6-24
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
parameters, the PMODE, DTYPE, and MAXBL values in DMACx, and the transfer capacity available in the EPBx data buffer employed in the transfer. The
internal I/O processor bus transfer size varies with the II, IM, and C
parameters, and the PMODE, DMA mode, DTYPE, and INT32 values in DMACx.
EC
The following sections describe these DMA modes and transfer sizes in
more detail:
• “Master Mode” on page 6-25
• “Paced Master Mode” on page 6-30
• “Slave Mode” on page 6-31
• “Handshake Mode” on page 6-34
• “External-Handshake Mode” on page 6-40
Master Mode
When the MASTER bit is set (=1) and the EXTERN and HSHAKE bits are cleared
(=0) in the channel’s DMACx register, the DMA channel is in master mode.
A channel in this mode can independently initiate internal or external
memory transfers.
mode applies to all external port DMA channels: 10, 11,
 Master
12, and 13.
To initiate a master mode DMA transfer, the DSP sets up the channel’s
parameter registers and sets the channel’s DMA enable (DEN) bit. A master
mode DMA channel performing internal memory to external memory
data transfer automatically performs enough transfers from internal memory to keep the EPBx buffer full. When the data transfer direction is
external to internal, a master mode DMA channel also performs enough
transfers from external memory to keep the EPBx buffer full.
I/O processor uses the , , and registers to access exter The
nal DSP memory in master mode DMA.
EI EM
ADSP-21160 SHARC DSP Hardware Reference
EC
6-25
Setting I/O Processor—EPort Modes
External Transfer Controls In Master Mode. In master mode, the DSP
determines the size of the external transfer from the channel’s PMODE bits
and EIx, EMx, and ECx registers. Table 6-1 on page 6-12 shows the packing
mode selected by the PMODE bits, and Table 6-5 on page 6-26 shows the
external transfer size in master mode that results from the combination of
the PMODE bits.
Table 6-5. Master Mode External Transfer Size
Transfer Size
64-bit1
64-bit2
32-bit
16-bit
PMODE
000
000
0003, 011, 100
001, 010
EI
64-bit aligned4
64-bit aligned
X5
X
EM
0 or 1
2
X
X
EC
even # of 32-bit
words, >= 2
# of 48-bit words X
# of 16-bit xfers
DTYPE
0
1
X
X
EPBx Depth
>1
>1
>=1
>=1
1
2
3
4
5
Including packed instructions
Including unpacked instructions or 40-bit data
For PMODE=000, even 32-bit addresses (EI[0]=0) access the lower 32-bits of the data bus.
For a 64-bit aligned address, EI[0]=0.
An X indicates any supported value.
64-bit External Transfers. To enable 64-bit transfers, PMODE must be set to
000. EI must be a 64-bit aligned Normal word address, because unaligned
64-bit external transfers are not supported. EM is restricted to values of 0
(to address a memory-mapped data FIFO such as the EPBx data buffer of
another DSP), or 1 (to increment through contiguous memory). EC contains the number of 32-bit words to transfer. For 64-bit transfers (only),
EC should be programmed to an even number. If EC must be set to an odd
value, the last transfer will be a 32-bit only transfer. There must be at least
two 32-bit EPBx FIFO entries available to support the 64-bit external
transfer.
6-26
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
64-bit External Burst Transfers. Burst transfers are a subset of 64-bit
transfers. In addition to the 64-bit transfer requirements described above,
bursting must be enabled by setting the MAXBL field in the DMAC. Also, the
burst truncates (or does not start), if the least significant bits of the 64-bit
address (ADDR bits 2-1) are both set (EI bits 2-1=11) (see the SBSRAM discussion in the External Memory chapter for more discussion on burst
address boundaries.)
Note that the external memory addressed by the burst transfer must map
to a memory bank configured for synchronous access mode (with the WAIT
register). The DMA programmer must ensure that the burst transfer does
not straddle, or cross, the external memory bank boundaries.
The following information applies to 64-bit burst transfers from
Table 6-5.
MAXBL=01
where:
•
must address a memory bank configured for synchronous access
modes (with WAIT register), and EIx must be 64-bit aligned (EIx
bits 2-1 may not be = 11).
EI
• Burst writes only supported in 1-wait write access mode. Bursts are
truncated at modulo4 boundaries of the 64-bit address EPBx Depth
>3 to support burst transfers
64-bit External Transfers of 48-bit Data/Instructions. Because the DSP’s
external bus does not support a 48-bit transfer size, programs must use
64-bit transfers sizes to move instructions. The two 64-bit transfer columns in Table 6-5 describe how to transfer packed instructions or
unpacked instructions.
Instructions are “packed” in external memory when the 3-column instructions are stored using all four 16-bit memory columns (same arrangement
as in internal memory). When instructions are packed in external
ADSP-21160 SHARC DSP Hardware Reference
6-27
Setting I/O Processor—EPort Modes
memory, the DSP cannot fetch and execute these instructions. The advantages of using packed instructions are that they take up 1/3 less memory
space than unpacked and the DSP can performs 1/3 fewer bus transfers to
DMA a block of these instructions than unpacked.
Instructions are “unpacked” in external memory when the 3-column
instructions are stored left-aligned, using three of the 16-bit memory columns. This arrangement matches the external port bus alignment shown
in Figure 7-1 on page 7-2. When instructions are unpacked in external
memory, the DSP can fetch and execute these instructions. For more
information, see “Executing Instructions From External Memory” on
page 7-48.
32-bit External Transfers. The DSP performs 32-bit transfers when
(No hardware packing mode), 011 (32-to-48-bit internal), or
100 (32-bit external-to-32-bit/64-bit internal). In PMODE=000 mode, the
external bus operation transfers 32-bits, instead of 64-bits if EIx, ECx, or
EMx do not match the 64-bit transfer conditions in Table 6-5.
PMODE=000
For 32-bit transfers in the PMODE=000 case, consecutive 32-bit transfers
access alternating high and low halves of the 64-bit data bus.
In PMODE=011or 100, all data transfers across the upper word of the data
bus (DATA63-32) as indicated in Figure 7-1 on page 7-2. This mode supports all values of EI, EM, and EC. EC contains the number of 32-bit words
to transfer. There must be at least one 32-bit EPBx FIFO entry available to
support the 32-bit external transfer.
16-bit External Transfers. The DSP performs 16-bit transfers when
PMODE=001 (16-bit external-to-32/64-bit internal) or 010 (16-bit external-to-48-bit internal). This mode supports all values of EI, EM, and EC. EC
is programmed to the number of 16-bit words to transfer. There must be
at least one 32-bit EPBx FIFO entry available to support the 16-bit external transfer. In PMODE=001, or 010, all data transfers across DATA47-32 as
indicated in Figure 7-1 on page 7-2.
6-28
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Internal Address/Transfer Size Generation. In master mode, the DSP
determines the size of the internal transfer from the channel’s PMODE bits
and IIx, IMx, and Cx registers. Table 6-2 on page 6-18 shows the packing
mode selected by the PMODE bits, and Table 6-6 shows the internal transfer
size in master mode that results from the combination of the PMODE bits.
Table 6-6. Master Mode Internal Transfer Size Determination
Transfer Size
64-bit1
48-bit
32-bit
PMODE
000, 001, 100
000, 010, 011
000, 001, 100
II
depends on IM2
X3
X
IM
-1 or 1
X
X
C
even # of 32-bit words # of 48-bit words
X
DTYPE
0
0 or 14
0
EPBx Depth
>1
>1
>=1
INT32
0
0
0 or 1
1
2
Including packed instructions.
If IMx is 1 for increment, IIx must be an even, 64-bit aligned Normal word address.
If IMx is -1 for decrement, IIx must be an odd, Normal word address.
3 X indicates any supported value.
4 DTYPE=1 for PMODE=000, 48-bit instruction transfers (unpacked).
DTYPE=0 for 48-bit packing modes.
64-bit Internal Transfers. To enable internal 64-bit transfers and increment the internal IIx pointer, programs must set IIx to match the IMx
selection as shown in Table 6-6. Cx contains the number of 32-bit words
to transfer, and should be set to an even # of 32-bit words. The DSP decrements Cx by 2 for each 64-bit transfer. For 64-bit transfers, PMODE must
be set to 000, 001 (16-bit-to-32/64-bit internal), or 100 (32-bit external-to-32/64-bit internal). DTYPE and INT32 must be cleared. There must
be at least two 32-bit EPBx FIFO entries available to support the 64-bit
external transfer.
ADSP-21160 SHARC DSP Hardware Reference
6-29
Setting I/O Processor—EPort Modes
48-bit Internal Transfers. The DSP can perform 48-bit internal transfers
for DMA of packed or unpacked 48-bit instructions. For more information on packed and unpacked instructions, see the discussion
on page 6-27.
Many applications can use internal 64-bit transfer for 48-bit instructions.
This technique can provide greater throughput than 48-bit internal transfers, but there are some restrictions. For more information on internal
64-bit transfers, see Table 6-6 and the discussion on page 6-29.
In either of the 48-bit internal transfer modes in Table 6-6 (PMODE=000
and DTYPE=1 or PMODE=010 or 011 and DTYPE=0), the DSP accesses the
memory using instruction alignment (3-column read or write) for the EPBx
buffer. In this case, IIx points to 48-bit words, and Cx counts the number
of 48-bit internal transfers.
32-bit Internal Transfers. The DSP performs according to the conditions
in Table 6-6. Under these additional conditions, the DSP performs 32-bit
transfers instead of 64- or 48-bit transfers: PMODE=000 (no hardware packing), 001 (16-bit external-to-32-bit internal), or 100 (32-bit
external-to-32-bit internal), and II is not aligned to a 64-bit boundary, or
IM is < -1, or > 1, or C is < 2, or EPBx depth < 2, or INT32 = 1, and DTYPE=0.
Paced Master Mode
When the MASTER and HSHAKE bits are set (=1) and the EXTERN bit is cleared
(=0) in the channel’s DMACx register, the DMA channel is in Paced Master
mode. A channel in this mode can independently initiate internal or external memory transfers.
Master mode applies only to external port DMA channels 11
 Paced
and 12.
In Paced Master mode, the DSP has the same control for address generation and transfer size as in master mode. For more information, see
“Master Mode” on page 6-25. The difference between these modes is that
6-30
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
in Paced Master mode external transfers are controlled and initiated
(paced) by the DMARx signal as in Handshake mode. For more information,
see “Handshake Mode” on page 6-34.
The DSP responds to the DMARx request only with the RDH/L, or WRH/L
strobes, depending on direction and data alignment. DMAGx is not asserted
in Paced Master mode. This method lets the DSP share the same buffer
between the I/O processor and processor core without external gating.
Paced Master mode accesses can be extended by the ACK input, by waitstates programmed in the WAIT register, and by holding the DMARx input
low.
Slave Mode
When the MASTER, HSHAKE, and EXTERN bits in the channel’s DMACx register
are cleared (=0), the DMA channel is in slave mode. A channel in this
mode cannot independently initiate external memory transfers.
To initiate a slave mode DMA transfer, an external device must read or
write the channel’s EPBx buffer. A slave mode DMA channel performing
internal to external data transfer automatically performs enough transfers
from internal memory to keep the EPBx buffer full. When the data transfer
direction is external to internal, a slave mode DMA channel does not initiate any internal DMA transfers until the external device writes data to the
channel’s EPBx buffer.
The I/O processor does not use the
 mode
DMA.
EI, EM,
and EC registers in slave
The following sequence describes a typical external to internal slave mode
DMA operation where an external device transfers a block of data into the
DSP’s internal memory:
1. The external device writes the DMA channel’s parameter registers
(IIx, IMx, and Cx) and DMACx control register, initializing the
channel.
ADSP-21160 SHARC DSP Hardware Reference
6-31
Setting I/O Processor—EPort Modes
2. The external device begins writing data to the EPBx buffer.
3. The EPBx buffer detects data is present and asserts an internal
DMA request to the I/O processor.
4. The I/O processor grants the request and performs the internal
DMA transfer, emptying the EPBx buffer FIFO.
If the internal DMA transfer is held off, the external device can continue
writing to the EPBx buffer because of its eight-deep FIFO. When the EPBx
FIFO becomes full, the DSP holds off the external device with the ACK signal (for synchronous accesses) or with the REDY signal (for asynchronous,
host-driven accesses). This hold-off state continues until the I/O processor
finishes the internal DMA transfer, freeing space in the EPBx buffer.
The following sequence describes a typical internal to external slave mode
DMA operation where an external device transfers a block of data from
the DSP’s internal memory:
1. The external device writes the DMA channel’s parameter registers
(IIx, IMx, and Cx) and DMACx control register, initializing the channel and automatically asserting an internal DMA request to the I/O
processor.
2. The I/O processor grants the request and performs the internal
DMA transfer, filling the EPBx buffers FIFO.
3. The external device begins reading data from the EPBx buffer.
4. The EPBx buffer detects that there is room in the buffer (it is now
“partially empty) and asserts another internal DMA request to the
I/O processor, continuing the process.
If the internal DMA transfers cannot fill the EPBx FIFO buffer at the same
rate as the external device empties it, the DSP holds off the external device
with the ACK signal (for synchronous accesses) or with the REDY signal (for
asynchronous, host-driven accesses) until valid data can be transferred to
the EPBx buffer.
6-32
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
The DSP only deasserts the
(or
) signal when the
 FIFO
buffer (or pad data buffer) is full during a write. The
ACK
REDY
EPBx
ACK (or
signal remains asserted at the end of a completed block transfer if the EPBx buffer is not full. For reads, the DSP deasserts the
ACK (or REDY) signal for each read to handle the latency of the read
versus posting the write to a buffer.
REDY)
In slave mode, the DSP determines the size of the transfer by decoding the
read and write (RDH/L and WRH/L) signals in addition to the channel’s
PMODE bits. Table 6-2 on page 6-18 shows the packing mode selected by
the PMODE bits, and Table 6-7 shows the transfer size in slave mode that
results from the combination of the read and write signals and PMODE bits.
Table 6-7. Slave Mode Transfer Size Determination
Transfer Size
(external internal)
64-bit
64-bit
32-bit
32-bit
32/64-bit
32-bit
48-bit
16-bit
32-bit1
32/64-bit2
16-bit
48-bit
PMODE
000
000
100
011
010
001
RDH




RDL



Not
supported
WRH




WRL

¥


DTYPE
0
0
0
1
2
1

0
Not
supported
1
External device must be connected to the upper half of the data bus (Data[63:32])
External device must be connected to Data[47:32])
ADSP-21160 SHARC DSP Hardware Reference
6-33
Setting I/O Processor—EPort Modes
DSP does not support 48-bit accesses to internal memory in
 The
slave mode.
Handshake Mode
When the MASTER and EXTERN bits are cleared (=0) and the HSHAKE bit is set
(=1) in the channel’s DMACx register, the DMA channel is in Handshake
mode. A channel in this mode cannot independently initiate external
memory transfers.
Handshake mode only applies to DMA channels 11 and 12.

To initiate a Handshake mode DMA transfer, an external device must
assert an external DMA request, asserting DMAR1 for access to EPB1 or
DMAR2 for access to EPB2. The buffers pass these request to the I/O processor, which prioritizes these requests with other internal DMA requests.
When the external DMA request has the highest priority, the I/O processor asserts an external DMA grant, asserting DMAG1 for EPB1 or DMAG2 for
EPB2. The grant signals the external device to read or write the EPBx buffer.
A Handshake mode DMA channel performing internal to external data
transfer automatically performs enough transfers from internal memory to
keep the EPBx buffer full. When the data transfer direction is external to
internal, a Handshake mode DMA channel does not initiate any internal
DMA transfers until the external devices writes data to the channel’s EPBx
buffer.
The I/O processor does not use the
 mode
DMA.
EI
or EM registers in Handshake
Other than the DMARx/DMAGx handshake, Handshake mode DMA operations follow almost the same process as slave mode DMA operations. The
exception is that in Handshake mode DMAs from internal to external
memory the external device must load the channel’s ECx register with the
number of external bus transfers.
6-34
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
In Handshake mode, the DSP determines the size of the transfer from the
channel’s parameter registers and PMODE bits. Table 6-2 on page 6-18
shows the packing mode selected by the PMODE bits, and Table 6-8 shows
the transfer size in slave mode that results from the combination of the
read and write signals and PMODE bits.
Table 6-8. Handshake mode Transfer Size Determination
Transfer Size
(external internal)
64-bit
64-bit
64-bit
48-bit
32-bit
32-bit
16-bit
16-bit
32/64-bit1
48-bit2
32/64-bit2
48-bit2
PMODE
000
000
100
011
001
001
IIx
64-bit
aligned
X
X
X
X
X
IMx
1
X
X
X
X
X
Cx
# of 32-bit
words3
# of 48-bit
words4
# of 32-bit
words
# of 48-bit
words
# of 32-bit
words
# of 48-bit
words
ECx
# of 32-bit
words1
3/4 * Cx
# of 32-bit
words
# of 32-bit
words
# of 16-bit
words
# of 16-bit
words
DTYPE
0
1
0
0
0
0
1
2
3
4
External device must be connected to the upper half of the data bus (Data[63:32])
External device must be connected to Data[47:32])
Must be an even number of 32-bit words
Should be a multiple of 4
Signal timing for Handshake mode does not delay the DMA operation.
The DMARx/DMAGx handshake operates asynchronously at up to the full
CLKIN speed of the DSP. For Handshake mode DMA, the DSP does not
assert the MS3-0 memory select lines (address strobes). For information on
DMARx/ DMAGx handshake timing, see Figure 6-5 on page 6-36.
The I/O processor uses the rising and falling edges of DMARx in the
handshake as prompts for DMA operations. On the falling
edge of DMARx, the edge signals the I/O processor to begin a DMA access.
DMARx/ DMAGx
ADSP-21160 SHARC DSP Hardware Reference
6-35
Setting I/O Processor—EPort Modes
DMAR rising edge allows
first DMAG to complete
DMAG has a waitstate because DMAR
remained asserted in cycle prior to the
DMAG assertion
CLKIN
First DMA
request
Second DMA
request
DMARx
DMAGx
DATA63-0
Bus
Transition
Cycle
(if not bus
master)
Data
valid
Data
valid
Data
valid
Data
valid
Figure 6-5. Handshake DMA Timing (Asynchronous Requests)
On the rising edge of DMARx, the edge signals the I/O processor to complete the DMA access.
The following sequence describes the process for requesting access to an
EPBx buffer in Handshake mode:
1. The external device asserts the buffer’s DMARx signal, placing an
external DMA request for access to the EPBx buffer.
2. The EPBx buffer detects the falling edge of the DMARx signal and
passes the external DMA request to the I/O processor, synchronizing the DMA operation with the processor’s system clock.
6-36
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
3. To be recognized in a particular cycle, the DMARx low transition
must meet the signal setup time from the DSP’s data sheet. If the
transition is slower than the setup time, the signal may not take
effect until the following cycle.
4. The I/O processor prioritizes the external DMA request with other
internal DMA requests. If the DSP is not already bus master, the
DSP arbitrates for the external bus when the external DMA request
has the highest priority, unless the EPBx buffer is blocked.
5. If the EPBx buffer is full during a write or empty during a read, the
buffer is blocked. The DSP does not begin external bus arbitration
until the I/O processor services the EPBx buffer, returning it to the
unblocked state empty for writing or full for reading.
6. The DSP becomes bus master and asserts DMAGx.
7. The DSP keeps DMAGx asserted until the cycle after the external
device deasserts DMARx. By holding DMARx asserted, the external
device holds the DSP until the external device is ready to proceed.
If the external device does not need to extend the DMA grant
cycle, the external device can deassert DMARx immediately (not waiting for DMAGx), providing the DMARx assertion time meets the timing
requirements from the DSP data sheet. The responding DMAGx in
this case is a short pulse, and the DSP only uses the external bus for
one cycle.
Because the I/O processor has a three-cycle DMA pipeline and a
seven-deep external request counter, the DSP can execute DMARx/DMAGx
handshake operations at up to the full CLKIN rate of the DSP. The I/O
processor has a three-cycle DMA pipeline, similar to the program
sequencer’s fetch–decode–execute instruction pipeline.
ADSP-21160 SHARC DSP Hardware Reference
6-37
Setting I/O Processor—EPort Modes
The I/O processor processes the DMA pipeline in the following stages:
• Recognizes the DMA request and arbitrates internal DMA priority
during the DMA fetch cycle.
• Generates the DMA address and arbitrates external bus access
during the DMA decode cycle.
• Transfers DMA data during the DMA execute cycle.
the I/O processor has a three-cycle DMA pipeline, there is
 aBecause
minimum delay of three cycles before the DSP asserts
. This
DMAGx
delay is in addition to any delay from internal DMA arbitration, so
the external device must not assume that the DMA grant can arrive
within two cycles even if higher priority DMA operations are disabled and the external bus is available for the transfer.
The I/O processor’s external request counter increments each time the
external device asserts DMARx and decrements each time the DSP replies by
asserting DMAGx. The external request counter records up to seven requests,
so the external device can make up to seven requests before the first one
has been serviced.
If the DSP cannot immediately service the DMA requests in the external
request counter, the DSP services the requests on a prioritized basis. The
external DMA device is responsible for keeping track of requests, monitoring grants, and pipelining the data when operating at full speed.
external device makes more than seven
without receiv Ifingthea grant,
the delayed grant causes unpredictable results.
DMARx
The DSP only asserts DMAGx for the number of DMARx requests indicated by
the external request counter. If the external devices makes more requests
than the count indicates, the DSP DMAGx assertions cannot match the
number of external device requests. To clear this mismatch, programs can
clear the buffer and the external request counter using the flush bit (FLSH)
in the channel’s DMACx control register.
6-38
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
To prevent holding off the DSP, the external device must service the
DSP’s data requirements when the DSP’s asserts the DMAGx grant signal.
The external device should immediately supply data for writes to the DSP
or immediately accept data on reads from the DSP. External interfaces can
handle this I/O by placing the data in an external FIFO. When performing DMA operations at the full CLKIN speed of the DSP, the system may
need a three-deep external FIFO to handle the latency between request
and grant. Programs on the external device can optimize operation of this
FIFO by issuing three requests rapidly and making the next requests conditional on when the DSP issues a grant.
The external devices must follow the conditions in Figure 6-6 on
page 6-40 when enabling or disabling Handshake mode for an external
port DMA channel.
The DSP ignores a disabled (transitioning from disabled to enabled)
DMA channel’s DMARx and DMAGx pins and ignores internal DMARx assertions for up to two processor core clock cycles after the instruction that
enables the channel in Handshake mode.
• The external devices must maintain DMARx deasserted (kept high,
not low or changing) during the instruction that enables DMA in
handshake mode. Before using the channel for the first time, programs flush the DMA channel, asserting the FLSH bit in the DMACx
control register. This action is not required during chain insertion.
• The DSP deasserts DMAGx if a program disables the channel while
DMARx and DMAGx are asserted (=0). This action clears the channel’s
active status bit, avoiding a potential deadlock condition.
ADSP-21160 SHARC DSP Hardware Reference
6-39
Setting I/O Processor—EPort Modes
CCLK
(core clock)
DMARx
Executing
Instruction
DMAR must be kept HI
during this instruction
Instruction enabling DMA by
setting DEN=1 and HSHAKE=1
in DMAC11 or DMAC12
DMARx ignored
Instruction
Instruction
Figure 6-6. DMARx Delay After Enabling Handshake DMA
DSPs in a multiprocessing cluster may share a DMAGx signal, because only
the bus master drives DMAGx. On the bus slaves, DMAGx is three-stated. This
state eliminates the need for external gating if more than one DSP or the
host needs to drive the DMA buffer. Systems may need a pull-up resistor
on this line if the host is not connected to the pin or does not drive it
when it acquires the bus. DMAGx has the same timing and transitions as the
RDH/L and WRH/L strobes in asynchronous access mode. DMAGx responds to
the SBTS and HBR signals in the same way as the read and write strobes. For
more information, see “External Port”.
External-Handshake Mode
When the MASTER bit is cleared (=0) and the HSHAKE and EXTERN bits are set
(=1) in the channel’s DMACx register, the DMA channel is in the
External-Handshake mode. A channel in this mode cannot independently
initiate external memory transfers.
External-handshake mode is identical to Handshake mode, except that
External-Handshake mode transfers data between external memory and an
external device. This section covers the differences between Handshake
mode and External-Handshake mode. For more information, see “Handshake Mode” on page 6-34.
6-40
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Handshake mode, External-Handshake mode only applies to
 Like
DMA channels 11 and 12.
To initiate an External-Handshake mode DMA transfer, an external
device must assert an external DMA request, asserting DMAR1 for access to
DMA channel 11or DMAR2 for access to DMA channel 12. The channels
pass these request to the I/O processor, which prioritizes these requests
with other internal DMA requests. When the external DMA request has
the highest priority, the I/O processor asserts an external DMA grant,
asserting DMAG1 for channel 11 or DMAG2 for channel 12. The grant signals
the external device to read or write the external bus. An External-Handshake mode DMA channel performing external to external data transfer
automatically generates external memory addresses and strobes for transfers between external memory and the external device.
Handshake mode, the I/O processor must use the
,
,
 Unlike
registers in External-Handshake mode DMA. Also unlike
and
EIx EMx
ECx
Handshake mode, the data for DMA channels 11 and 12 does not
pass through the EPB1 or EPB2 buffers.
During External-Handshake mode transfers, the I/O processor generates
external memory access cycles. DMARx and DMAGx operate the same as in
Handshake mode, but the DSP also outputs addresses, MS3-0 memory
selects, and RDH/L and WRH/L strobes, and responds to ACK. On external
memory writes, the DSP asserts DMAGx until the external device releases the
ACK line or any DSP waitstates expire. The external memory access by the
external devices responds as if the DSP processor core were making the
access. See “External Port” for information on connecting external devices
to the external port.
Because the I/O processor accesses external memory in External-Handshake mode, programs must load the DMA channel’s EIx, EMx, and ECx
parameter registers, and the DMACx PMODE bits. These settings let the I/O
processor generate the external memory addresses and word count.
ADSP-21160 SHARC DSP Hardware Reference
6-41
Setting I/O Processor—EPort Modes
mode does not support chained DMA inter External-handshake
rupts. Because no internal DMA transfers occur in
External-Handshake mode, the PCI bit in the channel’s CPx register
cannot disable the DMA interrupt. Programs must use the IMASK
register to mask this interrupt.
In External-Handshake mode, the DSP does not perform packing. The
DSP does determine the size of the transfer from the channel’s parameter
registers, PMODE bits. Table 6-9 shows the transfer size in slave mode that
results from the combination of the read and write signals and PMODE bits.
Table 6-9. External-Handshake Mode Transfer Size Determination
Transfer Size
(DSP device)
64-bit Memory
64-bit Memory
32-bit Memory


64-bit Device
64-bit Device
32-bit Device
32-bit Device
PMODE
000
000
000
100
EIx
X
constant
X
X1
EMx
12
03
Not 0 or 1
X
ECx
Even4
Even4
X5
X5
DTYPE
0
0
0
0


32-bit Memory
• For 64-bit transfers, the device must be connected to DATA63-0, and both RDH/L and WRH/L
are used.
• For 32-bit transfers, the device may be connected to either DATA63-32 or DATA31-0 and the
corresponding RDH/L or WRH/L is used depending on the address.
• X indicates any legal value.
1
2
3
4
5
6-42
For packed 32-bit transfers (PMODE=100), all ADDR bits are decoded.
For EMx=1, EIx is incremented by 2.
For EMx=0, EIx remains constant.
For 64-bit transfers, ECx is decremented by 2.
For 32-bit transfers, ECx is decremented by 1.
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Setting I/O Processor—LPort Modes
The SYSCON, LCOM, LAR, and LCTLx registers control the link ports operating
modes for the I/O processor. Table A-17 on page A-45 lists all the bits in
SYSCON, Table A-24 on page A-65 lists all the bits in LCOM, Table A-25 on
page A-67 lists all the bits in LAR, and Table A-23 on page A-63 lists all
the bits in LCTLx.
This section contains:
• “Link Port Buffer Modes” on page 6-45
• “Link Port Channel Priority Modes” on page 6-45
• “Link Port Channel Priority Modes” on page 6-45
The following bits control link port I/O processor modes. The control bits
in the LCTLx registers have a one cycle effect latency (take effect on the second cycle after change). Programs should not modify an active DMA
channel’s bits in the LCTLx register; other than to disable the channel by
clearing the LxDEN bit. For information on verifying a channel’s status
with the DMASTAT register, see “Using I/O Processor Status” on page 6-53.
• Link Port DMA Channel Priority Rotation Enable. SYSCON Bit 10
(LDCPR) This bit enables (rotates if set, =1) or disables (fixed if
cleared, =0) priority rotation among link port DMA channels
(channel 4-9).
• Link–External Port DMA Channel Priority Rotation Enable.
SYSCON Bit 12 (PRROT) This bit enables (rotates if set, =1) or disables (fixed if cleared, =0) priority rotation between link port DMA
channels (channel 4-9) and external port DMA channels (channel
10-13).
ADSP-21160 SHARC DSP Hardware Reference
6-43
Setting I/O Processor—LPort Modes
• Link port assignment for LBUFx. LAR Bits 2-0, 5-3, 8-6, 11-9,
14-12, and 17-15 (AxLB) Each 3-bit set assigns a link port to the
corresponding link buffer (LBUFx) as shown in Table 6-10 on
page 6-45.
• Link Buffer Enable. LCTL0 Bits 0, 10, and 20 and LCTL1 Bits 0, 10,
and 20 (LxEN) This bit enables (if set, =1) or disables (if cleared,
=0) the corresponding link buffer (LBUFx).
• Link Buffer DMA Enable. LCTL0 Bits 1, 11, and 21 and LCTL1 Bits
1, 11, and 21 (LxDEN) This bit enables (if set, =1) or disables (if
cleared, =0) DMA transfers for the corresponding link buffer
(LBUFx).
• Link Buffer DMA Chaining Enable. LCTL0 Bits 2, 12, and 22 and
LCTL1 Bits 2, 12, and 22 (LxCHEN) This bit enables (if set, =1) or
disables (if cleared, =0) DMA chaining for the corresponding link
buffer (LBUFx).
• Link Buffer Transfer Direction. LCTL0 Bits 3, 13, and 23 and
LCTL1 Bits 3, 13, and 23 (LxTRAN) This bit selects the transfer direction (transmit if set, =1) (receive if cleared, =0) for the
corresponding link buffer (LBUFx).
• Link Buffer Extended Word Size. LCTL0 Bits 4, 14, and 24 and
LCTL1 Bits 4, 14, and 24 (LxEXT) This bit selects the transfer
extended word size (48-bit if set, =1) (32-bit if cleared, =0) for the
corresponding link buffer (LBUFx). Programs must not change a
buffer’s LxEXT setting while the buffer is enabled.
• Link Buffer DMA 2-Dimensional. LCTL0 Bits 7, 17, and 27 and
LCTL1 Bits 7, 17, and 27 (LxDMA2D) This bit enables (if set, =1) or
disables (if cleared, =0) two-dimensional DMA transfers for the
corresponding link buffer (LBUFx).
Some other bits in LCTLx setup non-DMA link port features. For information on these features, see “Setting Link Port Modes” on page 8-5.
6-44
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Link Port Buffer Modes
The AxLB and LxEN bits in the LAR and LCTLx registers assign link ports to
link buffers and enable link buffers. Table 6-10 shows how the AxLB bits
in the LAR register assign link buffers to link ports.
To enable a link buffer, a program sets the buffer’s LxEN bit in LCTL0 or
LCTL1. To disable a link buffer, a program clears the buffer’s LxEN bit in
LCTL0 or LCTL1. The bit locations appear on page 6-43.
the DSP disables the buffer (
 When
low), the DSP clears the corresponding
transitions from high to
LxSTAT and LxRERR bits.
LxEN
Table 6-10. DSP Link Buffer-to-Link Port Assignments (AxLB)
Link Buffers
Link Ports
Buffer#
LAR Bits
0
1
2
3
4
5
NA1
0
2-0
000
001
010
011
100
101
111
1
5-3
000
001
010
011
100
101
111
2
8-6
000
001
010
011
100
101
111
3
11-9
000
001
010
011
100
101
111
4
14-12
000
001
010
011
100
101
111
5
17-15
000
001
010
011
100
101
111
1
NA indicates Not Assigned
Link Port Channel Priority Modes
The LDCPR and PRROT bits in the SYSCON register select priority levels for
the link port buffers in relation to the priority of other link port buffers
and the other I/O ports. The Link Port DMA Channel Priority Rotation
Enable (LDCPR) bit enables (rotates if set, =1) or disables (fixed if cleared,
=0) priority rotation among link port DMA channels (channel 4-9).
Rotating priority distributes link port DMA channel access to the IO bus.
ADSP-21160 SHARC DSP Hardware Reference
6-45
Setting I/O Processor—LPort Modes
When channel priority is rotating, the DSP arbitrates IO bus access
between contending link port DMA channels, forcing the channels to take
turns.
When channel priority is fixed, the lower numbered link port DMA channel (4-9) always has priority over the higher numbered channel when
contending for IO bus access. When LDCPR is set (rotating priority), high
priority shifts to a new channel after each single-word transfer. The order
of channel priority then rotates. After a single-word transfer, the I/O processor rotates priority to the next higher-numbered channel, and so on
until all six are serviced.
Figure 6-7 illustrates this process, according to the following example:
HIGHEST
PRIORITY
HIGHEST
PRIORITY
8
4
5
LOWEST
PRIORITY
9
STEP 2
6
9
LOWEST
PRIORITY
7
STEP 3
7
8
4
5
6
ONE TRANSFER OCCURS ON CHANNEL 7 (STEP 2),
ROTATING CHANNEL 7'S PRIORITY TO THE LOWEST PRIORITY SLOT (STEP 3).
Figure 6-7. Link Port Channel Rotating Priority Example
6-46
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
• At reset, link port channels have priority order—from high to
low—4, 5, 6, 7, 8, and 9.
• The link port performs a single transfer on channel 7.
• The I/O processor rotates channel priority, changing it to 8, 9, 4,
5, 6, and 7 (because rotating priority is enabled for this example,
LDCPR=1).
though the link port channel DMA priority can rotate, the
 Even
interrupt priorities of all DMA channels are fixed.
When a program uses fixed priority for the link port DMA channels, the
I/O processor assigns the highest priority to Channel 4 and the lowest priority to Channel 9. For a list of all fixed priority assignments, see the list
of DMA channels in Table 6-1 on page 6-12.
Programs can change the fixed priority order, assigning a different channel
to the highest priority. The following example shows how to change the
fixed priority sequence of the link port DMA channels:
• Disable all link port DMA channels except the one immediately
above the channel that is to have highest priority.
• Select rotating priority (by setting the LDCPR bit).
• Cause at least one transfer to occur on the enabled channel.
• Disable rotating priority and re-enable all of the link port DMA
channels.
• The channel immediately after the selected channel now has
the highest fixed priority.
Programs can also rotate priority between the link port and external port
DMA channels. The Link–External Port DMA Channel Priority Rotation
Enable (PRROT) bit enables (rotates if set, =1) or disables (fixed if cleared,
ADSP-21160 SHARC DSP Hardware Reference
6-47
Setting I/O Processor—LPort Modes
=0) priority rotation between link port DMA channels (channel 4-9) and
external port DMA channels (channel 10-13).
Rotating priority distributes link port and external port DMA channels’
access to the I/O bus. When channel priority is rotating, the DSP arbitrates IO bus access between contending link port and external port DMA
channels, forcing the channel types to take turns. When channel priority is
fixed, any link port DMA channel (4-9) always has priority over any external port DMA channel (10-13) when contending for IO bus access.
Link Port Channel Transfer Modes
The LxDEN, LxCHEN, LxTRAN, LxEXT, and LxDMA2D bits in the LCTLx register
enable link port DMA, chained DMA, and two-dimensional DMA and
select the transfer direction and format. The link DMA enable (LxDEN)
and link Chained DMA enable (LxCHEN) bits work together to select a link
port DMA channel’s transfer mode.
Table 6-11 lists the modes.
Table 6-11. Link Port DMA Enable Modes
LxCHEN
LxDEN
DMA Enable Mode Description
0
0
Channel disabled (chaining disabled, DMA disabled)
0
1
Single DMA mode (chaining disabled, DMA enabled)
1
0
Chain insertion mode (chaining enabled, DMA enabled,
auto-chaining disabled); For more information, see “Chaining
DMA Processes” on page 6-69.
1
1
Chained DMA mode (chaining enabled, DMA enabled,
auto-chaining enabled)
Because link ports are bidirectional, the I/O processor uses the link Transmit select (LxTRAN) bit to determine the transfer direction (transmit or
receive). Data flows from internal to external memory when in transmit
6-48
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
mode. In transmit mode, the I/O processor fills the channel’s LBUFx buffer
when the channel’s LxDEN bit is set.
The link Extend Word Size (LxEXT) bit determines how the DMA channel
accesses columns of internal memory. If LxEXT is set, the data is 40- or
48-bit words, and the I/O processor makes 3-column internal memory
accesses. If LxEXT is cleared, the data is 32-bit words, and the I/O processor makes 2-column internal memory accesses. For more information, see
“Memory Organization and Word Size” on page 5-22.
 The
Width (
LxEXT
for the transfer overrides the Internal Memory Data
setting for the internal memory block.
IMDWx)
The link buffer’s DMA 2-Dimensional (LxDMA2D) bit enables (if set, =1) or
disables (if cleared, =0) two-dimensional DMA transfers for the corresponding link buffer (LBUFx). When LxDMA2D is set, the channel’s DAx and
DBx registers control the two-dimensional DMA process. For more information, see “Using Two-Dimensional Link Port DMA” on page 6-83.
A link buffer’s
bit must be cleared (=0) for standard
 (one-dimension)
DMA operation.
LxDMA2D
Setting I/O Processor—SPort Modes
The SRCTLx and STCTLx registers control the serial port operating mode for
the I/O processor. Table A-28 on page A-74 lists all the bits in SRCTLx and
Table A-27 on page A-71 lists all the bits in STCTLx.
This section contains:
• “Serial Port Buffer Modes” on page 6-51
• “Serial Port Channel Priority Modes” on page 6-52
• “Serial Port Channel Transfer Modes” on page 6-52
ADSP-21160 SHARC DSP Hardware Reference
6-49
Setting I/O Processor—SPort Modes
The following bits control serial port I/O processor modes. The control
bits in the SRCTLx and STCTLx registers have a one cycle effect latency (take
effect on the second cycle after change). Programs should not modify an
active DMA channel’s bits in the SRCTLx or STCTLx registers; other than to
disable the channel by clearing the SDEN bit. To change an inactive serial
port’s operating mode, programs should clear a serial port’s control register before writing new settings to the control register. For information on
verifying a channel’s status with the DMASTAT register, see “Using I/O Processor Status” on page 6-53.
Some other bits in SRCTLx and STCTLx setup non-DMA serial port features. For information on these features, see “Setting Serial Port Modes”
on page 9-6.
• Serial Port Enable. SRCTLx and STCTLx Bit 0 (SPEN) This bit enables
(if set, =1) or disables (if cleared, =0) the corresponding serial port.
• Data Type Select. SRCTLx and STCTLx Bits 2-1 (DTYPE) These bits
select the data type formatting for normal and multi-channel
reception as follows: (normal/multichannel= format)
00/x0=Right-justify and zero-fill unused MSBs, 01/x1=Right-justify and sign-extend unused MSBs, 10/0x=Compand using -law,
11/1x=Compand using A-law.
• Serial Word Endian Select. SRCTLx and STCTLx Bit 3 (SENDN) This
bit selects little endian words (LSB first, if set, =1) or big endian
words (MSB first, if cleared, =0).
• Serial Word Length Select. SRCTLx and STCTLx Bits 8-4 (SLEN)
These bits select the word length (–1) in bits. Word sizes can be
from 3-bit (SLEN=2) to 32-bit (SLEN=31).
• 16-bit to 32-bit Word Packing Enable. SRCTLx and STCTLx Bit 9
(PACK) This bit enables (if set, =1) or disables (if cleared, =0) 16- to
32-bit word packing.
6-50
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
• Serial Port Receive DMA Enable. SRCTLx and STCTLx Bit 18 (SDEN)
This bit enables (if set, =1) or disables (if cleared, =0) the serial
port’s receive DMA.
• Serial Port Receive DMA Chaining Enable. SRCTLx and STCTLx Bit
19 (SCHEN) This bit enables (if set, =1) or disables (if cleared, =0)
serial port DMA chaining.
• Two Dimensional DMA Array Enable. SRCTLx and STCTLx Bit 21
(D2DMA) This bit enables (if set, =1) or disables (if cleared, =0)
two-dimensional serial DMA. This bit must be cleared for multichannel operation.
Serial Port Buffer Modes
The SPEN, SENDN, SLEN, and PACK bits in the SRCTLx and STCTLx registers
enable the serial port and select the transfer format. To enable a serial port
transmit or receive buffer, a program sets the buffer’s SPEN bit in the
SRCTLx or STCTLx register.
To disable a serial port transmit or receive buffer, a program clears the
buffer’s SPEN bit in the SRCTLx or STCTLx register.
a serial port buffer is enabled and DMA for that channel is not
 Ifenabled,
the serial port is in single-word, interrupt-driven transfer
mode. For more information, see “Using I/O Processor Status” on
page 6-53.
Each serial port buffer allows independent settings for the three transfer
format features: bit order, word length, and word packing.
For transferring little endian words (LSB first, if set, =1) to or from little
endian devices, the serial port buffers have a Serial Word Endian Select
(SENDN) bit. This bit selects little endian words (LSB first, if set, =1) or big
endian words (MSB first, if cleared, =0).
ADSP-21160 SHARC DSP Hardware Reference
6-51
Setting I/O Processor—SPort Modes
The Serial Word Length Select (SLEN) bit field selects the transfer word
length (-1) in bits. Word sizes can be from 3-bit (SLEN=2) to 32-bit
(SLEN=31).
If the serial word length is 16-bits or smaller, the serial port can pack two
of these words into the serial port buffer. The 16-bit to 32-bit Word Packing Enable (PACK) bit can enable this packing because the I/O processor
performs 32-bit transfers between the serial port buffers and DSP
memory.
In addition to selecting the endian, length, and packing modes for serial
port DSP transfers, programs must indicate the type of data in the transfer, using the Data Type (DTYPE) bit. For more information, see “Serial
Port Channel Transfer Modes” on page 6-52.
Serial Port Channel Priority Modes
Serial port DMA transfers always take priority over external port or link
port DMA transfers. For more information on prioritization operations,
see “Managing DMA Channel Priority” on page 6-67.
Serial Port Channel Transfer Modes
The SDEN, SCHEN, DTYPE, and D2DMA bits in the SRCTLx and STCTLx register
enable serial port DMA, chained DMA, and two-dimensional DMA and
select the format. The DMA enable (SDEN) and Chained DMA enable
(SCHEN) bits work together to select a serial port DMA channel’s transfer
mode. Table 6-12 lists the modes.
6-52
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Table 6-12. Serial Port DMA Enable Modes
SCHEN
SDEN
DMA Enable Mode Description
0
0
Channel disabled (chaining disabled, DMA disabled)
0
1
Single DMA mode (chaining disabled, DMA enabled)
1
0
Chain insertion mode (chaining enabled, DMA enabled,
auto-chaining disabled); For more information, see “Chaining
DMA Processes” on page 6-69.
1
1
Chained DMA mode (chaining enabled, DMA enabled,
auto-chaining enabled)
Because serial port buffers are bidirectional, the I/O processor does not
need an indicator to determine the transfer direction (transmit or receive).
Data flows from internal to external devices using a transmit (TXx) buffer.
When transmitting serial data as DMA, the I/O processor fills the channel’s TXx buffer when the channel’s SDEN bit is set.
The serial port buffer’s DMA 2-Dimensional (D2DMA) bit enables (if set,
=1) or disables (if cleared, =0) two-dimensional DMA transfers for the
corresponding serial port buffer (TXx or RXx). When 2DDMA is set, the channel’s DAx and DBx registers control the two-dimensional DMA process. For
more information, see “Using Two-Dimensional Serial Port DMA” on
page 6-91.
A serial port buffer’s
bit must be cleared (=0) for standard
 (one-dimension)
DMA operation.
D2DMA
Using I/O Processor Status
The I/O processor monitors the status of data transfers on DMA channels
and indicates status in the DMASTAT, IRPTL, and LIRPTL registers. Table A-9
on page A-18 lists all the bits in IRPTL, Table A-10 on page A-24 lists all
the bits in LIRPTL, and a discussion of DMASTAT appears in “DMA Channel
Status Register (DMASTAT)” on page A-60.
ADSP-21160 SHARC DSP Hardware Reference
6-53
Using I/O Processor Status
The I/O processor reports on DMA in progress, DMA complete, or DMA
channel not ready status:
• All DMA channels can be active or inactive. If a channel is active—
a DMA is in progress on that channel—the I/O processor indicates
the active status by setting the channel’s bit in the DMASTAT register.
• When an unchained (single-block) DMA process reaches completion on any DMA channel, the I/O processor generates that DMA
channel's interrupt—sets the DMA channel's interrupt latch bit in
the IRPTL or LIRPTL register. DMA process is complete when
• The count in Cx=0 (for slave mode and handshake modes).
• The count in ECx=0 (for external handshake mode).
• The count in Cx=0 and ECx=0 (for master mode and paced
master mode).
• When a DMA process in a chained DMA sequence reaches completion (the count in Cx=0) on any DMA channel, the I/O
processor generates an interrupt if the PCI bit in the channels CPx
register is set, except in external-handshake mode. The I/O processor also generates that DMA channel’s interrupt when the last
block in a chained DMA reaches completion whether or not PCI is
set.
• When a DMA channel’s buffer not being used for a DMA process,
the I/O processor can generate an interrupt on single word writes
or reads of the buffer. This interrupt service differs slightly for each
port. For more information on single-word interrupt-driven transfers, see “External Port Status” on page 6-57, “Link Port Status” on
page 6-60, and “Serial Port Status” on page 6-63.
Using the DMA Channel Status Register (DMASTAT), programs can check
which DMA channels are performing a DMA or chained DMA. For each
channel, the I/O processor sets the channel’s active status bit if DMA for
6-54
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
that channel is enabled and a DMA sequence is in progress on that channel. The I/O processor sets the channel’s chaining status bit if a chained
DMA sequence is in progress or pending on that channel.
is a one cycle latency between a change in DMA channel sta There
tus and the status update in the
register.
DMASTAT
As an alternative to interrupt-driven DMA, programs can poll the DMASTAT
register to determine when a single DMA sequence is done. To poll channel status, programs read DMASTAT. If both status bits for the channel are
cleared, the DMA sequence has completed.
chaining is enabled on a DMA channel programs should not use
 Ifpolling
to determine channel status. Polling could provide inaccurate information in this case because the next DMA sequence
might be under way by the time the polled status is returned.
During interrupt-driven DMA, programs use the interrupt mask bits in
the IMASK and LIRPTL registers to selectively mask DMA channel interrupts that the I/O processor latches into the IRPTL and LIRPTL registers.
Descriptions of these conditions appear on on page 6-53.
I/O processor only generates a DMA complete interrupt when
 The
the channel’s count register decrements to zero as a result of actual
DMA transfers. Writing zero to a count register does not generate
the interrupt.
A channel interrupt mask in IMASK and IRPTL masks out DMA complete
interrupts for a channel, but other types of interrupt masking are also
available. These other types of interrupt masking include:
ADSP-21160 SHARC DSP Hardware Reference
6-55
Using I/O Processor Status
• By clearing a channels PCI bit during chained DMA, programs
mask the DMA complete interrupt for a DMA processes within a
chained DMA sequence.
• By masking the LPISUM interrupt, programs mask out the logical
Or’ing of link port interrupt status.
• By masking the LSRQ interrupt, programs mask out link port service
requests to link ports that do not have an assigned link buffer.
These lower levels of interrupt masking let programs limit some of the
conditions that can cause DMA channel interrupts.
DMA channel has its own interrupt. Although the external
 Each
port and link port channel access priority can rotate, the interrupt
priorities of all DMA channels are fixed.
In DSP systems using I/O processor interrupts, an external device may
need to change the DSP’s interrupt mask. This task presents a challenge
because the IMASK register is not memory-mapped and is not directly
accessible to external devices through the external port. To read or write
IMASK through the external port, programs can set up an interrupt vector
routine to handle this task. The VIRPT vector interrupt register may be
used for this task.
The I/O processor can also generate DMA channel interrupts for I/O port
operations that do not use DMA. In this case, the I/O processor generates
a DMA interrupt when data becomes available at the receive buffer or
when the transmit buffer does not have new data to transmit. Generating
DMA interrupts in this fashion lets programs implement interrupt-driven
I/O under control of the processor core. Care is needed because multiple
interrupts can occur if several I/O ports transmit or receive data in the
same cycle.
For more information on these types of single-word interrupt-driven
transfers, see the external port discussion on page 6-59, link port discussion on page 6-63, or serial port discussion on page 6-65.
6-56
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
External Port Status
The I/O processor monitors the status of data transfers on the external
port. DMA channel status for the external port is described in “Using I/O
Processor Status” on page 6-53. This section describes external port specific status features, such as buffer status, buffer control, and single-word
interrupt-driven transfers.
Bits in the SYSTAT, SYSCON and DMACx registers indicate and control status
of external port buffers. Table A-20 on page A-51 lists all the bits in SYSTAT, Table A-17 on page A-45 lists all the bits in SYSCON, and Table A-21
on page A-55 lists all the bits in DMACx. The following bits influence external port buffer status:
• Host Packing Status. SYSTAT bits 15-14 (HPS). These bits indicate
the host’s packing status.
• Host Packing Status Flush. SYSCON Bit 8 (HPFLSH) This bit flushes
(when set, =1) settings for the direct write FIFO.
• External Port Packing Status. DMACx Bits 4-3 (PS). These bits indicate the corresponding FIFO buffer’s packing status.
• Single-Word Interrupt Enable. DMACx Bit 12 (INTIO). This bit
enables (if set, =1) or disables (if cleared, =0) single-word,
non-DMA, interrupt-driven transfers for the corresponding external port FIFO buffer (EPBx). To avoid spurious interrupts,
programs must not change a buffer’s INTIO setting while the buffer
is enabled.
• Flush DMA Buffers and Status. DMACx Bit 14 (FLSH). This bit
flushes (when set, =1) settings for the corresponding external port
FIFO buffer (EPBx).
• External Port FIFO Buffer Status. DMACx bit 17-16 (FS). These bits
indicate the corresponding FIFO buffer’s status.
ADSP-21160 SHARC DSP Hardware Reference
6-57
Using I/O Processor Status
The HPS and PS bits in the SYSTAT and DMACx registers indicate an external
buffer’s packing status. These bits are read-only, and the DSP clears these
bits when DEN is cleared (changes from 1 to 0). Table 6-13 shows the available settings.
Table 6-13. DSP (PS) and Host (HPS) Packing Status
PS or HPS
Packing Status
00
pack complete (reset value)
01
1st stage pack/unpack
10
2nd stage multi-stage pack/unpack
11
reserved
The FS bits in the DMACx registers indicate an external buffer’s FIFO status.
These bits are read-only. The DSP clears these bits when DEN is cleared
(changes from 1 to 0). Table 6-14 shows the available settings.
Table 6-14. External Port Buffer FIFO Status
FS
FIFO Buffer Status
00
buffer empty
01
buffer-not-full
10
buffer-not-empty
11
buffer full
For transmit (TRAN=1), buffer-not-full means that the buffer has space for
one Normal word, and buffer-not-empty means that the buffer has space
for two-or-more Normal words. For receive (TRAN=0), buffer-not-full
means that the buffer contains one Normal word, and buffer-not-empty
means that the buffer contains two-or-more Normal words. Any type of
full status (01, 10, or 11) in receive mode indicates that new (unread) data
is in the buffer.
6-58
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
The HPFLSH and FLSH bits in the SYSCON and DMACx registers flush an external buffer’s packing status, but these bits work differently. The HPFLSH bit
flushes (when set, =1) settings for the direct write FIFO. Flushing these
settings clears the HPS status bits in the SYSTAT register, clears (=0) the
channel’s DMA request counter, and clears (-0) any partially packed
words.
By comparison, the FLSH bit flushes (when set, =1) settings for the corresponding external port FIFO buffer (EPBx). Flushing these settings does
the following: Clears (=0) the FS and PS status bits, Clears (=0) the FIFO
buffer and DMA request counter, Clears (-0) any partially packed words.
When a program sets (=1) either HPFLSH or FLSH, the DSP flushes the settings and clears (=0) the flush bit. There is a two-cycle effect latency in
completing the flush operation. DSP programs must not set a buffer’s
FLSH during the same write that enables the buffer. Also, programs must
not set a buffer’s FLSH bit while the DMA channel is active. Programs
should determine the channel’s active status by reading the corresponding
bit in the DMASTAT register.
does not change on the master DSP during external port
 Status
DMA until the external portion is completed (i.e., the
buffers
EPBx
are emptied).
If in chain insertion mode ( =0,
=1), then channel chaining
 status
will never go to 1. Programs should test channel status to see
DEN
CHEN
if it is ready before re-writing the chain pointer (CPx).
The INTIO bit in the DMACx registers support single-word interrupt-driven
transfers for each corresponding external port buffer. These non-DMA
transfers are available under the following conditions:
ADSP-21160 SHARC DSP Hardware Reference
6-59
Using I/O Processor Status
• The external port DMA channel’s DEN bit is cleared (DMA
disabled).
• The external port DMA channel’s INTIO bit is set (enabling interrupt-driven I/O).
• The external port DMA channel’s buffer is “not empty” on an
external read or “not full” on an external write.
Under these conditions, the I/O processor generates that DMA channel’s
interrupt on the single word transfer to that channel’s external port buffer.
Link Port Status
The I/O processor monitors the status of data transfers on the link ports.
DMA channel status for the link ports is described in “Using I/O Processor Status” on page 6-53. This section describes link ports specific status
features, such as buffer status, buffer control, and single-word interrupt-driven transfers.
Bits in the LCOM and LSRQ registers indicate and control status of link port
buffers. Table A-24 on page A-65 lists all the bits in LCOM and Table A-26
on page A-68 lists all the bits in LSRQ. The following bits influence link
port buffer status.
• Link Buffer x Status. LCOM Bits 1-0, 3-2, 5-4, 7-6, 9-8, 11-10
(LxSTAT). These bits indicate the corresponding buffer’s status.
• Link Buffer x Receive Pack Error Status. LCOM Bits 26, 27, 28, 29,
30, 31 (LRERRx). These bits indicate the buffer’s packing status.
• Link Port x transmit mask. LSRQ Bit 4, 6, 8, 10, 12, 14 (LxTM).
These bits mask (if set, =1) or unmask (if cleared, =0) the L0TRQ
through L5TRQ status bits.
6-60
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
• Link Port x receive mask. LSRQ Bit 5, 7, 9, 11, 13, 15 (LxRM). These
bits mask (if set, =1) or unmask (if cleared, =0) the L0RRQ through
L5RRQ status bits.
• Link Port x transmit request status (read-only). LSRQ Bit 20, 22,
24, 26, 28, 30 (LxTRQ). If set (=1), these bits indicate that a link
port (0 through 5) is disabled, but has a request to transmit data.
• Link Port x receive request status (read-only). LSRQ Bit 21, 23, 25,
27, 29, 31 (LxRRQ). If set (=1), these bits indicate that a link port (0
through 5) is disabled, but has a request to receive data.
The LRERRx bits in the LCOM register indicate a link port buffer’s receive
packing status. When the buffer is ready to receive and pack a a new word,
the DSP clears (=0) LRERRx. If this bit remains set (=1) after the buffer
receives a word, a link transfer error (for example, a clock glitch) has
occurred. These bits are read-only, and the DSP clears these bits when
LxEN is cleared (changes from 1 to 0). Table 6-15 shows the available
settings.
Table 6-15. Link Port Buffer Receive Packing Status
LRERRx
Receive Packing Status
0
pack complete (reset value)
1
pack not complete
The LxSTAT bits in the LCOM register indicate a link buffer’s FIFO status.
When transmitting, these bits indicate when the buffer has space for more
data. When receiving, these status bits indicate when the buffer contains
new (unread) data. These bits are read-only. The DSP clears these bits
when LxEN is cleared (changes from 1 to 0) and empties the buffer.
Table 6-16 shows the available settings.
ADSP-21160 SHARC DSP Hardware Reference
6-61
Using I/O Processor Status
Table 6-16. Link Port Buffer FIFO Status
LxSTAT
FIFO Buffer Status
00
buffer empty
01
reserved
10
one word
11
buffer full
The LAR register lets programs assign link buffers to link ports. Table 6-10
on page 6-45 shows how the AxLB bits in the LAR register assign link buffers to link ports. Because this mapping allows link ports to be unassigned
(no buffer), the I/O processor has an interrupt (LSRQI) to notify programs
that an external device has made a read or write request on a disabled link
port.
When the an LSRQI interrupt is latched into the IRPTL register, programs
use the transmit (LxTRQ) and receive (LxRRQ) request bits in LSRQ register to
determine which port has a request. The LSRQ register’s bits indicate that:
• For a transmit request (LxTRQ=1), the LSRQI interrupt indicates that
the link port (0 through 5) is disabled, but another DSP has
requested more data by setting the link port’s acknowledge
(LxACK=1).
• For a receive request (LxRRQ=1), the LSRQI interrupt indicates that
the link port (0 through 5) is disabled, but another DSP has
requested to send data by setting the link port’s clock (LxCLK=1).
To control sources of link port service requests, the I/O processor lets programs mask these service requests. The LSRQ register provides mask bits for
transmit (LxTM) and receive (LxRM) link service requests.
6-62
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
The LxEN bits in the LCTLx register support single-word interrupt-driven
transfers for each corresponding link port buffer. These non-DMA transfers are available under these conditions:
• The link port DMA channel’s LxDEN bit is cleared (DMA disabled).
• The link port DMA channel’s LxEN bit is set (enabling the link
buffer).
• The link port DMA channel’s buffer is “not empty” on receive or
“not full” on transmit.
Under these conditions, the I/O processor generates that DMA channel’s
interrupt on the single word transfer to that channel’s link port buffer.
Serial Port Status
The I/O processor monitors the status of data transfers on the serial ports.
DMA channel status for the serial ports is described in “Using I/O Processor Status” on page 6-53. This section describes serial ports specific status
features, such as buffer status, transmit buffer underflow, receive buffer
overflow, and single-word interrupt-driven transfers.
Bits in the STCTLx and SRCTLx registers indicate and control status of serial
port buffers. Table A-27 on page A-71 lists all the bits in STCTLx and
Table A-28 on page A-74 lists all the bits in SRCTLx. The following bits
influence serial port buffer status:
• Transmit Underflow Status (sticky, read-only). STCTLx Bit 29
(TUVF). This bit indicates whether the serial transmit operation has
underflowed (if set, =1).
• Transmit Data Buffer Status (read-only). STCTLx Bits 31-30 (TXS).
These bits indicate the status of the serial port’s transmit buffer
(TXx).
ADSP-21160 SHARC DSP Hardware Reference
6-63
Using I/O Processor Status
• Receive Overflow Status (sticky, read-only). SRCTLx Bit 29 (ROVF).
This bit indicates whether the receive operation has overflowed (if
set, =1).
• Receive Data Buffer Status (read-only). SRCTLx Bits 31-30 (RXS).
These bits indicate the status of the serial port’s receive buffer
(RXx).
The TXS and RXS bits in the STCTLx and SRCTLx registers indicate a serial
port transmit (TXx) or receive (RXx) buffer’s FIFO status. Status bits are
read-only. Disabling the serial port (setting SPEN=0), clears the status bits
and empties the buffer. TXS and RXS may change state if the data is read or
written by the processor core while the serial port is disabled.
Table 6-17 shows the available settings.
Table 6-17. Serial Port Transmit and Receive Buffer FIFO Status
TXS or RXS
FIFO Buffer Status
00
buffer empty
01
reserved
10
partially full
11
buffer full
The TUVF and ROVF bits in the STCTLx and SRCTLx registers indicate a serial
port transmit underflow or receive overflow to the buffer’s FIFO. Status
bits are read-only. Disabling the serial port (setting SPEN=0), clears the status bits and empties the buffer. These overflow and underflow bits are
sticky; once set, they remain set regardless of buffer status until the serial
port is disabled.
6-64
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
The SPEN bit in the STCTLx or SRCTLx register support single-word interrupt-driven transfers for each corresponding serial port transmit or receive
buffer. These non-DMA transfers are available under these conditions:
• The serial port DMA channel’s SDEN bit is cleared (DMA disabled).
• The serial port DMA channel’s SPEN bit is set (enabling the serial
port transmit or receive buffer).
• The serial port DMA channel’s buffer is “not empty” on receive or
“not full” on transmit.
Under these conditions, the I/O processor generates that DMA channel’s
interrupt on the single word transfer to that channel’s link port buffer.
DMA Controller Operation
DMA sequences start in different ways depending on whether DMA
chaining is enabled. When chaining is not enabled, only the DMA enable
bit (DEN) allows DMA transfers to occur. A DMA sequence starts when
one of the following occurs:
• Chaining is disabled and the DMA enable bit (DEN) transitions
from low to high.
• Chaining is enabled, DMA is enabled (DEN=1), and the CPx register
address field is written with a non-zero value. In this case, TCB
chain loading of the channel parameter registers occurs first.
• Chaining is enabled, the CPx register address field is non-zero, and
the current DMA sequence finishes. Again, TCB chain loading
occurs.
ADSP-21160 SHARC DSP Hardware Reference
6-65
DMA Controller Operation
A DMA sequence ends when one of the following occurs:
• The count register decrements to zero (both C and EC for external
port channels).
• Chaining is disabled and the channel’s DEN bit transitions from
high to low. If the DEN bit goes low (=0) and chaining is enabled,
the channel enters chain insertion mode and the DMA sequence
continues. For more information, see “Inserting a TCB in an
Active Chain” on page 6-73.
a program sets the
bit (=1) after a single DMA finishes,
 When
the DMA sequence continues from where it left off (for
DEN
non-chained operations only). To start a new DMA sequence after
the current one is finished, a program must first clear the DEN
enable bit, write new parameters to the II, IM, and C registers, then
set the DEN bit to re-enable DMA. For chained DMA operations,
these steps are not necessary. For more information, see “Chaining
DMA Processes” on page 6-69.
a DMA operation completes and the count register is rewritten
 Ifbefore
the DMA enable bit is cleared, the DMA transfer will restart
at the new count.
Once a program starts a DMA process, the process is influenced by two
external controls: DMA channel priority and DMA chaining. For more
information, see For more information, see “Managing DMA Channel
Priority” on page 6-67. or “Chaining DMA Processes” on page 6-69.
6-66
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Managing DMA Channel Priority
The DMA channels for each of the DSP’s I/O ports negotiate channel priority with the I/O processor using an internal DMA request/grant
handshake. Each I/O port (link ports, serial ports, and external ports) has
one or more DMA channels, with each channel having a single request
and a single grant. When a particular channel needs to read or write data
to internal memory, the channel asserts an internal DMA request. The
I/O processor prioritizes the request with all other valid DMA requests.
When a channel becomes the highest priority requester, the I/O processor
asserts the channel’s internal DMA grant. In the next clock cycle, the
DMA transfer starts. Figure 6-8 shows the paths for internal DMA
requests within the I/O processor.
If a DMA channel is disabled ( ,
, or
bit =0), the I/O
 processor
does not issue internal DMA grants to that channel,
DEN LxDEN
SDEN
whether or not the channel has data to transfer.
Because more than one DMA channel can make a DMA request in a particular cycle, the I/O processor prioritizes DMA channel service. DMA
channel prioritization determines which channel can use the IOD (I/O
Data) bus to access memory. Default DMA channel priority is fixed prioritization by DMA channel type (serial ports, link ports, or external port).
Within the DMA channel types, the serial port DMA channels are always
fixed priority, the external port DMA channels may be either fixed or
rotated priority, and the link port DMA channels may be either fixed or
rotated priority.
Table 6-1 on page 6-12 lists the DMA channels in descending order of
priority. For information on programming link port or external port priority modes, see “External Port Channel Priority Modes” on page 6-19 or
“Link Port Channel Priority Modes” on page 6-45.
The I/O processor determines which DMA channel has the highest priority internal DMA request during every cycle between each data transfer.
ADSP-21160 SHARC DSP Hardware Reference
6-67
DMA Controller Operation
DSP CORE
INTERNAL MEMORY
PM ADDRESS
ADDR
DM ADDRESS
ADDR
ON ASYNCHRONOUS WRITES, THE
SLAVE WRITE FIFO IS 4 DEEP.
ON SYNCHRONOUS WRITES, THE
SLAVE WRITE FIFO IS 2 DEEP.
ADDR DATA DATA DATA
IOA
PMD
DMD
EXTERNAL PORT
IOD
PMA
PM DATA
DMA
PMD
DM DATA
DMD
ADDR
32
EPD
EPA
I/O DATA BUS (IOD)
I/O ADDRESS BUS (IOA)
SLAVE WRITE
FIFO
(2/4 DEEP)
64
DATA
BUFFER
18
64
64
EXT. PORT DATA BUS (EPD)
32
EXT. PORT ADDR BUS (EPA)
I/O PROCESSOR
EXTERNAL PORT
FIFOS
EPB (8 DEEP)
INTERNAL DMA
ADDRESS
GENERATORS
GRNTS REQ.S
LINK PORT FIFOS
LBUF (2 DEEP)
EXTERNAL DMA
ADDRESS
GENERATORS
GRNTS REQ.S
DMAR
14
REQUESTS
GRANTS
INTERNAL
DMA
PRIORITZER
DMA CONTROLLER
14
SERIAL PORT
FIFOS
RX,TX (2 DEEP)
OTHER IOP
REGISTERS
REQUESTS
GRANTS
DMAG
EXTERNAL
DMA
PRIORITIZER
DMA CONTROLLER
LINK PORTS
DIRECT WRITE
FIFO (8 DEEP)
SERIAL PORTS
Figure 6-8. I/O Processor Internal Request and Grant Paths
Internal DMA channel arbitration differs from external bus arbitration.
For more information on external bus arbitration, see “Multiprocessor Bus
Arbitration” on page 7-98.
Processor core accesses of I/O processor registers, external direct accesses
of internal memory, and TCB chain loading are subject to the same prioritization scheme as the DMA channels. Applying this scheme uniformly
prevents I/O bus contention, because these accesses are also performed
6-68
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
over the internal I/O bus. TCB chain loading has a higher priority than
external port accesses. This TCB priority permits chained serial port
DMA, even when the external port is attempting an access in every cycle.
For more information, see “Chaining DMA Processes” on page 6-69.
If a DSP has all six link ports enabled and active at the same time, the
default priority scheme could hold off external port DMA channels for
extended periods of time. Because this hold off could have a significant
negative impact on external bus performance, the I/O processor permits
rotating DMA channel priority between the link port channel group and
external port channel group. For more information on using the PRROT bit
to rotate priority between link ports and the external port, see “Link Port
Channel Priority Modes” on page 6-45.
Chaining DMA Processes
DMA chaining lets the I/O processor automatically load DMA parameters
and start the next DMA when the current DMA finishes. This feature permits unlimited multiple DMA transfers without processor core
intervention. Using chaining, programs can set up multiple DMA operations, and each operation can have different attributes.
I/O processor responds by auto-initializing the channel’s
 The
parameter registers with the first TCB and starting the first
transfer.
To chain together multiple DMA operations, the I/O processor must load
the next Transfer Control Block (DMA parameters) into the DMA
parameter registers when the current DMA finishes (DMA count =0). The
chain pointer register (CPx) points to the next set of DMA parameters,
which are stored in internal memory. This process of loading the TCB
into the parameter registers is called TCB chain loading.
The chain pointer register should be cleared first before enabling chaining.
Two controls enable chained DMA. Each DMA channel has a chaining
enable bit (CHEN) in the channel’s control register. When set, the CHEN bit
ADSP-21160 SHARC DSP Hardware Reference
6-69
DMA Controller Operation
directs the I/O processor to use the CPx register for chained DMA. Programs start the chained DMA by writing a non-zero address to the CPx
register, directing the I/O processor to start the DMA with TCB chain
loading. Programs can disable chained DMA by writing all zeros to the
address field of the CPx register.
DMA operations may only occur within the same chan Chained
nel. The DSP does not support cross-channel chaining.
The CPx register is 19 bits wide, of which the lower 18 bits are the memory
address field. Like other I/O processor address registers, the CPx registers
value is offset to match the starting address of internal memory before
being used by the I/O processor. On the ADSP-21160 DSP, this offset
value is 0x0004 0000.
Bit 18 (the 19th bit) of the CPx register is Program Controlled Interrupts
(PCI) bit. If set, the PCI bit enables a DMA channel interrupt to occurs at
the completion of the current DMA sequence.
 (The
bit only effects DMA channels that have chaining enabled
CHEN =1). Also, interrupt requests enabled by the PCI bit are maskable with the IMASK register.
PCI
the
bit is not part of the memory address in the
 Because
register, programs must be careful when writing and reading
PCI
CPx
addresses to and from the register. To prevent errors, programs
should mask out the PCI bit (bit 18, the 19th bit) when copying
the address in CPx to another address register.
During chained DMA, the channel’s General Purpose (GP) register is a
useful place to point to the last completed DMA sequence. This practice
lets programs determine where the last full (or empty) data buffer is
located.
6-70
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Transfer Control Block (TCB) Chain Loading
During TCB chain loading, the I/O processor loads the DMA channel
parameter registers with values retrieved from internal memory. The
address in the CPx register points to the highest address of the TCB (containing the IIx parameter). The TCB values reside in consecutive memory
locations.
Table 6-18 shows the TCB-to-register loading sequence for the external
port, link port, and serial port DMA channels. The parameter order in the
table is the order that the I/O processor reads each word of the TCB and
loads it into the corresponding register. Programs must set up the TCB in
memory in the order shown in Table 6-18, placing the IIx parameter at
the address pointed to by the CPx register of the previous DMA operation
of the chain.
Table 6-18. TCB Chain Loading Sequence
Address1
External Port
Link Ports and Serial Ports
CPx + 0x0004 0000
IIx
IIx
CPx – 1 + 0x0004 0000
IMx
IMx
CPx – 2 + 0x0004 0000
Cx
Cx (and DAx for 2D DMA)2
CPx – 3 + 0x0004 0000
CPx
CPx
CPx – 4 + 0x0004 0000
GPx
GPx
CPx – 5 + 0x0004 0000
EIx
DBx (loaded during 2D DMA only)
CPx – 6 + 0x0004 0000
EMx
LPATH1 (mesh multiproc. links only)3
CPx – 7 + 0x0004 0000
ECx
LPATH2 (mesh multiproc. links only)3
CPx – 8 + 0x0004 0000
–
LPATH3 (mesh multiproc. links only)3
1 An “x” denotes the DMA channel number.
2 The DAx and DBx registers are not loaded during chaining in normal, one-dimensional DMA.
In 2D DMA operations, only DBx is loaded. The DAx register is automatically loaded with the
same value as the Cx register.
ADSP-21160 SHARC DSP Hardware Reference
6-71
DMA Controller Operation
3
The link transmit chain also downloads the LPATH1, LPATH2, and LPATH3 registers when
the LMSP bit in the LCOM control register is set, enabling mesh multiprocessing.
A TCB chain load request is prioritized like all other DMA operations.
The I/O processor latches a TCB loading request and holds it until the
load request has the highest priority. If multiple chaining requests are
present, the I/O processor services the TCB registers for the highest priority DMA channel first. A channel which is in the process of chain loading
cannot be interrupted by a higher priority channel.
For a list of DMA channels in priority order, see Table 6-1 on page 6-12.
For more information on DMA priority, see “Managing DMA Channel
Priority” on page 6-67.
Setting Up and Starting The Chain
To setup and initiate a chain of DMA operations, program use the following steps:
1. Set up all TCBs in internal memory.
2. Write to the appropriate DMA control register, setting the DEN
DMA enable bit to 1 and the CHEN chaining enable bit to 1.
3. Write the address containing the IIx register value of the first TCB
to the CPx register, starting the chain (see Figure 6-9).
The I/O processor responds by autoinitializing the channel’s parameter
registers with the first TCB and starting the first transfer. When the transfer finishes, the I/O processor begins the next TCB chain load if the
current chain pointer address is non-zero. The CPx address points to the
next TCB.
address field of the
registers is only 18 bits wide. If a pro The
gram writes a symbolic address to bit 18 of
, there may be a
CPx
CPx
conflict with the PCI bit. Programs should clear the upper bits of
the address, then AND in the PCI bit separately if needed.
6-72
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
18 17 16 15 14 13 12 11 10
9
8
7
6
5
4
3
2
1
0
CPx
PCI B IT
PR OGR AM CONT R OL L E D INT E R R UPT B IT
IF T HIS B IT IS S E T , T HE I/O P R OCE S S OR GE ME R AT E S A
DMA INT E R R UPT ON COMP L E T ION OF A CHAINE D DMA.
Figure 6-9. CPx Register
Inserting a TCB in an Active Chain
It is possible to insert a single DMA operation or another DMA chain
within an active DMA chain. Programs may need to perform insertion
when a high priority DMA requires service and cannot wait for the current chain to finish.
When DMA on a channel is disabled (DEN=0) and chaining on the channel
is enabled (CHEN=1), the DMA channel is in chain insertion mode. This
mode lets a program insert a new DMA or DMA chain within the current
chain without effecting the current DMA transfer. Programs should use
the following sequence to insert a DMA subchain while another chain is
active:
1. Enter chain insertion mode by setting CHEN=1 and DEN=0 in the
channel’s DMA control register.
2. The DMA interrupt indicates when the current DMA sequence has
completed.
3. Write the CPx register value into the CP position of the last TCB in
the new chain.
4. Enter chained DMA mode by setting DEN=1 and CHEN=1.
5. Write the start address of the first TCB of the new chain into the
CPx register.
ADSP-21160 SHARC DSP Hardware Reference
6-73
External Port DMA
Chain insertion mode operates the same as chained DMA mode (DEN=1,
CHEN=1), except that when the current DMA transfer ends, automatic
chaining is disabled and an interrupt request occurs. This interrupt
request is independent of the PCI bit state.
insertion should not be set up as an initial mode of opera Chain
tion. This mode should only be used to insert a DMA within an
active DMA chaining operation.
External Port DMA
The DSP support a number of DMA modes for external port DMA. The
following sections provide overviews of typical external port DMA
processes:
• “Setting up External Port DMA” on page 6-74
• “Bootloading Through The External Port” on page 6-76
Setting up External Port DMA
The method for setting up and starting an external port DMA sequence
varies slightly with the selection of transfer and DMA handshake for the
channel. For more information on transfer and DMA handshake modes,
see “External Port Channel Transfer Modes” on page 6-21 and “External
Port Channel Handshake Modes” on page 6-22. For more detailed information on external port DMA features, see “Setting I/O Processor—EPort
Modes” on page 6-14.
6-74
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
In general, the following sequence describes a typical external to internal
DMA operation where an external device transfers a block of data into the
DSP’s internal memory:
1. The DSP or host (depends on mode) writes the DMA channel’s
parameter registers (IIx, IMx, and Cx) and DMACx control register,
initializing the channel for receive (TRAN=0).
2. The DSP or host (depends on mode) sets (=1) the channel’s DEN bit
enabling the DMA process.
3. The external device begins writing data to the EPBx buffer (through
the external port).
4. Whether or not the DSP signals for this transfer to begin depends
on the mode.
5. The EPBx buffer detects data is present and asserts an internal
DMA request to the I/O processor.
6. The I/O processor grants the request and performs the internal
DMA transfer, emptying the EPBx buffer FIFO.
In general, the following sequence describes a typical internal to external
DMA operation where an external device transfers a block of data from
the DSP’s internal memory:
1. The DSP or host (depends on mode) writes the DMA channel’s
parameter registers (IIx, IMx, and Cx) and DMACx control register,
initializing the channel for transmit (TRAN=1).
2. The DSP or host (depends on mode) sets (=1) the channel’s DEN bit
enabling the DMA process. Because this is a transmit, setting DEN
automatically asserts an internal DMA request to the I/O
processor.
3. The I/O processor grants the request and performs the internal
DMA transfer, filling the EPBx buffer’s FIFO.
ADSP-21160 SHARC DSP Hardware Reference
6-75
External Port DMA
4. The external device begins reading data from the EPBx buffer
(through the external port).
5. Whether or not the DSP signals for this transfer to begin depends
on the mode.
6. The EPBx buffer detects that there is room in the buffer (it is now
“partially empty) and asserts another internal DMA request to the
I/O processor, continuing the process.
Bootloading Through The External Port
The DSP can boot from an EPROM or host processor through the external port. The DMAC10 control register is specially initialized for booting in
each case. Each booting mode packs boot data into 48-bit instructions.
EPROM and host boot use channel 10 of the I/O processor’s DMA controller to transfer the instructions to internal memory. For EPROM
booting, the DSP reads data from an 8-bit external EPROM. For host
booting, the DSP accepts data from a 16- or 32-bit host microprocessor
(or other external device).
After the boot process loads 256 words into memory locations 0x40000
through 0x400FF, the DSP begins executing instructions. Because most
DSP programs require more than 256 words of instructions and initialization data, the 256 words typically serve as a loading routine for the
application. Analog Devices supplies loading routines (Loader Kernels)
that can load entire programs. These routines come with the development
tools. For more information on Loader Kernels, see the development tools
documentation.
is important to note that DMA channel differences between the
 ItADSP-21160
DSP and previous SHARC DSPs (ADSP-2106x
DSPs) introduce some booting differences. Even with these differences, the ADSP-21160 DSP supports the same boot capability
and configuration as the ADSP-2106x DSPs.
6-76
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
DMAC default values differ because the ADSP-21160 DSP has
additional parameters and different DMA channel assignments.
The EPROM and Host boot modes use EPB0, DMA channel 10.
In EPROM booting, the alignment of the 8-bit port differs due to
the new 64-bit data path. The ADSP-21160 DSP boots from
DATA39-32 instead of DATA23-16 as on the ADSP-2106x DSPs.
For EPROM or host booting the ADSP-21160 DSP, the Program
sequencer automatically unmasks the DMA channel 10 channel
interrupt, initializing the IMASK register to 0x00008003.
The DSP determines the booting mode at reset from the EBOOT, LBOOT,
and BMS pin inputs. When EBOOT=1 and LBOOT=0, the DSP boots from an
EPROM through the External Port and uses BMS as the memory select output. When EBOOT=0, LBOOT=0, and BMS =1, the DSP boots from a host
through the External Port.
For a list showing how to select different boot modes, see the Boot Memory Select pin description on “Pin Descriptions” on page 11-3.
of the power-up booting modes, address
 When using any
should not contain a valid instruction since it is not
0x0004 0004
executed during the booting sequence. A NOP or IDLE instruction should be placed at this location.
In EPROM booting through the external port, an 8-bit wide boot
EPROM must be connected to data bus pins 39-32 (DATA39-32). The lowest address pins of the DSP should be connected to the EPROM’s address
lines. The EPROM’s chip select should be connected to BMS and its output
enable should be connected to RDH.
ADSP-21160 SHARC DSP Hardware Reference
6-77
External Port DMA
In a multiprocessor system, the BMS output is only driven by the
ADSP-21160 DSP bus master. This allows wire-OR’ing of multiple BMS
signals for a single common boot EPROM.
can boot any number of ADSP-21160 DSPs from a single
 Systems
EPROM, using the same code for each processor or differing code
for each.
During reset, the DSP’s ACK line is internally pulled high with a 2 k
equivalent resistor and is held high with an internal keeper latch. It is not
necessary to use an external pull-up resistor on the ACK line during booting
or at any other time.
When EPROM boot mode is configured, the External Port DMA Channel 10 (DMAC10) becomes active following reset; it is initialized to 0x04A1,
which allows external port DMA enable and selects DTYPE for instruction
words. The packing mode bits (PMODE) are ignored, BSO is set in SYSCON,
and 8-to-48 bit packing is forced with least-significant-word first.
The UBWS and UBAM fields of the WAIT register are initialized to perform
asynchronous access and generate seven wait states (eight cycles total) for
the EPROM access in unbanked external memory space. (Note that wait
states defined for unbanked memory are applied to BMS-asserted accesses.)
Table 6-19 shows how the DMA Channel 10 parameter registers are initialized at reset for EPROM booting. The count register (C10) is initialized
to 0x0100 for transferring 256 words to internal memory. The external
count register (EC10), which is used when external addresses are generated
by the DMA controller, is initialized to 0x0600 (that is, 0x0100 words with
six bytes per word).
6-78
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Table 6-19. DMA Channel 10 Parameter Register Initialization For
EPROM Booting
Parameter
Register
Initialization Value
II10
0x0004 0000
IM10
uninitialized (increment by 1 is automatic)
C10
0x0100 (256 instruction words)
CP10
uninitialized
GP10
uninitialized
EI10
0x0080 0000
EM10
uninitialized (increment by 1 is automatic)
EC10
0x0600 (256 words x 6 bytes/word)
At system start-up, when the DSP’s RESET input goes inactive, the following sequence occurs:
1. The DSP goes into an idle state, identical to that caused by the
IDLE instruction. The program counter (PC) is set to address
0x0004 0004.
2. The DMA parameter registers for channel 10 are initialized (as
shown in Table 6-19).
3.
BMS
becomes the boot EPROM chip select.
4. 8-bit Master Mode DMA transfers from EPROM to internal memory begin, on the external port data bus lines 39-32.
5. The external address lines (ADDR31-0) start at 0x0080
increment after each access.
0000
and
6. The RDH strobe asserts as in a normal memory access, with seven
wait states (eight cycles).
ADSP-21160 SHARC DSP Hardware Reference
6-79
Link Port DMA
The DSP’s DMA controller reads the 8-bit EPROM words, packs them
into 48-bit instruction words, and transfers them to internal memory until
256 words have been loaded. The EPROM is automatically selected by the
BMS pin; other memory select pins are disabled. The DMA external count
register (EC10) decrements after each EPROM transfer. When EC10
reaches zero, the following wake-up sequence occurs:
1. The DMA transfers stop.
2. The External Port DMA Channel 10 interrupt (EP0I) is activated.
3.
is deactivated and normal external memory selects are
activated.
BMS
4. The DSP vectors to the EP0I interrupt vector at 0x0004 0050.
At this point the DSP has completed its booting mode and is executing
instructions normally. The first instruction at the EP0I interrupt vector
location, address 0x0004 0050, should be an RTI (Return From Interrupt). This process returns execution to the reset routine at location
0x0004 0005 where normal program execution can resume. After reaching
this point, a program can write a different service routine at the EP0I vector location 0x0004 0050.
Link Port DMA
The DSP support a number of DMA modes for link port DMA. The following sections provide overviews of typical link port DMA processes:
• “Setting up Link Port DMA” on page 6-81
• “Using Two-Dimensional Link Port DMA” on page 6-83
• “Bootloading Through The Link Port” on page 6-87
6-80
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Setting up Link Port DMA
The method for setting up and starting an link port DMA sequence varies
slightly with the transfer mode for the channel. For more information on
DMA transfer modes, see “Link Port Channel Transfer Modes” on
page 6-48. For more detailed information on link port DMA features, see
“Setting I/O Processor—LPort Modes” on page 6-43.
In general, the following sequence describes a typical external to internal
DMA operation where an external device transfers a block of data into the
DSP’s internal memory using a link port:
1. The DSP or host (depends on mode) assigns the DMA channel’s
link buffer to a link port using the channel’s AxLB bits in the LAR
register.
2. The DSP or host (depends on mode) enables the DMA channel’s
link buffer, setting the buffer’s LxEN bit in the channel’s LCTLx register. The DSP or host selects a words size (32- or 40/48-bits) using
the LxEXT in the channel’s LCTLx register.
3. The DSP or host (depends on mode) writes the DMA channel’s
parameter registers (IIx, IMx, and Cx) and LCTLx control register,
initializing the channel for receive (LxTRAN=0).
4. The DSP or host (depends on mode) sets (=1) the channel’s LxDEN
bit enabling the DMA process.
5. The external device begins writing data to the LBUFx buffer
(through the link port).
6. The LBUFx buffer detects data is present and asserts an internal
DMA request to the I/O processor.
7. The I/O processor grants the request and performs the internal
DMA transfer, emptying the LBUFx buffer FIFO.
ADSP-21160 SHARC DSP Hardware Reference
6-81
Link Port DMA
In general, the following sequence describes a typical internal to external
DMA operation where an external device transfers a block of data from
the DSP’s internal memory using a link port:
1. The DSP or host (depends on mode) assigns the DMA channel’s
link buffer to a link port using the channel’s AxLB bits in the LAR
register.
2. The DSP or host (depends on mode) enables the DMA channel’s
link buffer, setting the buffer’s LxEN bit in the channel’s LCTLx register. The DSP or host selects a words size (32- or 40/48-bits) using
the LxEXT in the channel’s LCTLx register.
3. The DSP or host (depends on mode) writes the DMA channel’s
parameter registers (IIx, IMx, and Cx) and DMACx control register,
initializing the channel for transmit (LxTRAN=1).
4. The DSP or host (depends on mode) sets (=1) the channel’s LxDEN
bit enabling the DMA process. Because this is a transmit, setting
LxDEN automatically asserts an internal DMA request to the I/O
processor.
5. The I/O processor grants the request and performs the internal
DMA transfer, filling the LBUFx buffer’s FIFO.
6. The external device begins reading data from the LBUFx buffer
(through the link port).
7. The LBUFx buffer detects that there is room in the buffer (it is now
“partially empty) and asserts another internal DMA request to the
I/O processor, continuing the process.
6-82
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Using Two-Dimensional Link Port DMA
Two-dimensional DMA is available on all link port and serial port DMA
channels. This DMA mode lets programs DMA data in memory treating
the data as an array. Programs can use this row and column data arrangement in DSP algorithms that perform array operations. In this mode, the
DMA channel’s DAx, DBx, and GPx registers direct data placement operations and the operation of the Cx register changes slightly from
single-dimension DMA transfers.
To place a link port or serial port in two-dimensional DMA mode, a program must set the link port channel’s LxDMA2D bit in the LCTLx register or
the serial port channel’s D2DMA bit in the SRCTLx or SRCTLx register.
Figure 6-10 and Figure 6-11 show how registers operate for two-dimensional DMA. These figures also show how the I/O processor places 16- or
32-bit data from a two-dimensional DMA in memory.
In two-dimensional DMA, the I/O processor’s parameter registers operate
as follows:
• The Index register (IIx) initially holds the first address in the data
array. As the DMA progresses, the I/O processor updates IIx to
hold the current address by adding the data array X increment after
each transfer.
• The Modify register (IMx) holds the data array X (column) increment. The I/O processor uses IMx to modify the current address
(IIx) to point to the next element in the data array X dimension
(next column of array, not necessarily the next memory column).
• The Dimension-A (DAx) register holds the data array X initial count
(column count). At the beginning of each new row, the I/O processor uses DAx to load the Cx register with the number of columns in
the data array. For programming convenience, the I/O processor
writes the DAx register automatically whenever the processor core
writes a count to the Cx register.
ADSP-21160 SHARC DSP Hardware Reference
6-83
Link Port DMA
Viewing DSP memory as four 16-bit columns,
the 4x4 data array on the right would 2-D
DMA into the memory space shown below.
The callouts on this diagram indicate parameter
register values for the 2-D DMA operation,
transferring 16-bit words.
E
D
B
A
9
8
7
6
5
4
3
2
1
1
5
9
D
2
6
A
E
3
7
B
F
C
Y-DIRECTION (COLUMN)
F
0
4
8
C
GPx
THE GENERAL PURPOSE (GPX)
REGISTER CONTA INS THE
COUNT OF ARRA Y ELEM ENTS
REMAINING IN THE
Y-DIRECTI ON. HE RE, GPX=4
TO S TART AND DEC RE MENTS
AS E AC H ROW IS PROC ESSE D.
DBx
THE DIMENSION-B (DBX)
REGISTER CONTA INS THE
OFFSE T FROM THE ADDRESS
OF THE LAST ROW ELEMENT
TO THE E LEM ENT AT THE START
OF THE NEXT ROW . HERE,
DBX=1.
0
X-DI RE CTION (ROW )
Cx
DAx
IMx
IIx
THE C OUNT (CX) REGI STER
CONTAINS THE NUMBER OF
ARRAY ELEME NTS LEFT IN THE
X-DI REC TION (ROW). CX IS
RELOADE D FROM DAX W ITH
EACH NEW ROW. HERE, C X
STA RTS AS 4.
THE DIM ENSION-A
(DAX) RE GIS TER
C ONT AI NS THE
NUM BER OF ARRAY
ELEME NTS IN THE
X-DIRECTION. HERE,
DAX=4.
THE MODIFY (IMX)
REGISTER CONTAINS
THE X-DIRE CTION
INCREMEN T. HERE,
IMX=1.
THE INDE X (II X) REGISTER
I NITIALLY C ONTAINS THE
FIRST ADDRESS IN THE ARRAY
AND IS UPDATED TO I NDIC ATE
THE CURRENT ADDRESS BY
ADDING THE X INCRE MENT
AFTE R EACH TRANS FE R
Figure 6-10. Two-Dimensional DMA of a 4x4 Array of 16-bit Words
• The Count (Cx) register contains the number of data elements left
in the current row. At the beginning of each new row, the I/O processor loads Cx from DAx. When Cx decrements to zero, the I/O
processor goes to the next row.
6-84
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Viewing DSP memory as two 32-bit columns,
the 4x4 data array on the right would 2-D
DMA into the memory space shown below.
The callouts on this diagram indicate parameter
register values for the 2-D DMA operation,
transferring 32-bit words.
E
D
C
B
A
9
8
7
6
5
4
3
2
2
6
A
E
3
7
B
F
THE GENERAL PURPOSE (GPX)
REGISTER CONTA INS THE
COUNT OF ARRA Y ELEM ENTS
REMAINING IN THE
Y-DIRECTI ON. HE RE, GPX=4
TO S TART AND DEC RE MENTS
AS E AC H ROW IS PROC ESSE D.
DBx
THE DIMENSI ON-B (DBX)
REGISTER CONTA INS THE
OFFSE T FROM THE ADDRESS
OF THE LAST ROW ELEMENT
TO THE E LEM ENT AT THE START
OF THE NEXT ROW . HERE,
DBX=1.
X-DI RE CTION (ROW )
1
1
5
9
D
GPx
Y-DIRECTION (COLU MN)
F
0
4
8
C
0
Cx
DAx
IMx
IIx
THE C OUNT (CX) REGISTER
CONTAINS THE NUMBER OF
ARRAY ELEME NTS LEFT IN THE
X-DI REC TION (ROW). CX IS
RELOADE D FROM DAX W ITH
EACH NEW ROW. HERE, C X
STA RT S AS 4.
THE DIM ENSION-A
(DAX) RE GIS TER
C ONTAI NS THE
NUM BER OF ARRAY
ELEME NTS I N THE
X-DIRECTI ON. HERE,
DAX=4.
THE MODIFY (IMX)
REGISTER CONTAINS
THE X-DIRE CTION
INCREMEN T. HERE,
IMX=1.
THE INDE X (IIX) REGISTER
I NITIALLY C ONTAINS THE
FIRST ADDRESS IN THE ARRAY
AND IS UPDATED TO I NDIC ATE
THE CURRENT ADDRESS BY
ADDING THE X INCRE MENT
AFTE R EACH TRANS FE R
Figure 6-11. Two-Dimensional DMA of a 4x4 Array of 32-bit Words
ADSP-21160 SHARC DSP Hardware Reference
6-85
Link Port DMA
• The Dimension-B (DBx) register contains the data array Y increment (row increment). To start a new row, the I/O processor adds
the offset in DBx to the current address to point to the next element
in the Y dimension (first location in next row).
• The General Purpose (GPx) register contains the data array Y
Count. Initially, the GPx contains the number of data elements in
the Y dimension (number of rows). The I/O processor decrements
this value each time the X count register reaches zero. When Y
Count reaches zero, the two-dimensional DMA is finished. To do
one-dimensional DMA transfers in two-dimensional DMA mode,
programs must set the Y Count GPx to one.
a number of parameter registers must update, two DMA
 Because
cycles are required for a row change.
For two-dimensional DMA transfers, these register operations occur in the
following processor order:
1. During the first DMA cycle, the I/O processor performs the following steps:
a. Outputs the current address from the IIx register and starts
a DMA memory cycle
b. Adds the X Increment value from the IMx register to the current address in the IIx register
c. Decrements the X Count in the Cx register
d. Checks whether the X Count has decremented to zero; if so,
performs the second DMA cycle
6-86
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
2. During the second DMA cycle, the I/O processor performs the following steps:
a. Restores the X Count in the Cx register from the DAx register
b. Adds the Y Increment value in the DBx register to the current address in the IIx register
c. Decrements the Y Count in the GPx register
d. Checks whether the Y Count has decremented to zero; if so,
the DMA sequence is finished (the channel becomes
inactive)
loads the X Count ( ) register or Y Count ( ) with
 Ifzero,a program
the I/O processor does not disable DMA transfers on that
Cx
GPx
channel. The I/O processor interprets the zero as a request for 216
transfers. This count occurs because the I/O processor starts the
first transfer before the testing the count value. The only way to
disable a DMA channel is to clear its DMA enable bit.
For more information, see “External Port Channel Transfer
Modes” on page 6-21, “Link Port Channel Transfer Modes” on
page 6-48, or “Serial Port Channel Transfer Modes” on page 6-52.
Bootloading Through The Link Port
One of the DSP’s booting mode is booting the DSP through the link port.
Link port booting uses DMA channel 8 of the I/O processor to transfer
the instructions to internal memory. In this boot mode, the DSP receives
4-bit wide data in link buffer 4.
After the boot process loads 256 words into memory locations 0x40000
through 0x400FF, the DSP begins executing instructions. Because most
DSP programs require more than 256 words of instructions and initialization data, the 256 words typically serve as a loading routine for the
ADSP-21160 SHARC DSP Hardware Reference
6-87
Link Port DMA
application. Analog Devices supplies loading routines (Loader Kernels)
that load an entire program through the selected port. These routines
come with the development tools. For more information on Loader Kernels, see the development tools documentation.
channel differences between the ADSP-21160 DSP and pre DMA
vious SHARC DSPs (ADSP-2106x DSPs) introduce some booting
differences. Even with these differences, the ADSP-21160 DSPs
supports the same boot capability and configuration as the
ADSP-2106x DSPs. For link booting the ADSP-21160 DSP, the
Program sequencer automatically unmasks the DMA channel 8
interrupt, initializing the LIRPTL register to 0x00100000 and IMASK
register to 0x00004003.
The DSP determines the booting mode at reset from the EBOOT, LBOOT,
and BMS pin inputs. When EBOOT=0, LBOOT=1, and BMS=1, the DSP boots
through the Link Port. For a list showing how to select different boot
modes, see the Boot Memory Select pin description on Table 11-1 on
page 11-3.
of the power-up booting modes, address
 When using any
should not contain a valid instruction since it is not
0x0004 0004
executed during the booting sequence. A NOP or IDLE instruction should be placed at this location.
In Link Port Booting, the DSP gets boot data from another DSP’s link
port or four bit wide external device after system powerup.
The external device must provide a clock signal to the link port assigned
to link buffer 4. The clock can be any frequency, up to a maximum of the
DSP clock frequency. The clock’s falling edges strobe the data into the
link port. The most significant 4-bit nibble of the 48-bit instruction must
be downloaded first.
Table 6-20 shows how the DMA Channel 8 parameter registers are initialized at reset for Link Port booting. The count register (C8) is initialized to
0x0100 for transferring 256 words to internal memory. The LCTL and
6-88
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
link port control registers are overridden during link port booting to
allow link buffer 4 to receive 48-bit data.
LCOM
Table 6-20. DMA Channel 8 Parameter Register Initialization For Link
Port Booting
Parameter
Register
Initialization Value
II8
0x0004 0000
IM8
uninitialized (increment by 1 is automatic)
C8
0x0100 (256 instruction words)
CP8
uninitialized
GP8
uninitialized
DA
uninitialized
DB
uninitialized
In systems where multiple DSPs are not connected by the parallel external
bus, booting can be accomplished from a single source through the link
ports. To simultaneously boot all of the DSPs, a parallel common connection should be made to Link Buffer 4 on each of the processors. If only a
daisy chain connection exists between the processors’ link ports, then each
DSP can boot the next one in turn. Link Buffer 4 must always be used for
booting.
Serial Port DMA
The DSP support a number of DMA modes for link port DMA. The following sections provide overviews of typical serial port DMA processes:
• “Setting up Serial Port DMA” on page 6-90
• “Using Two-Dimensional Serial Port DMA” on page 6-91
ADSP-21160 SHARC DSP Hardware Reference
6-89
Serial Port DMA
Setting up Serial Port DMA
The method for setting up and starting an serial port DMA sequence varies slightly with the transfer mode for the channel. For more information
on DMA transfer modes, see “Serial Port Channel Transfer Modes” on
page 6-52. For more detailed information on serial port DMA features,
see “Setting I/O Processor—SPort Modes” on page 6-49.
In general, the following sequence describes a typical external to internal
DMA operation where an external device transfers a block of data into the
DSP’s internal memory using a serial port:
1. The DSP or host (depends on mode) enables the DMA channel’s
serial port, setting the port’s SPEN bit in the port’s SRCTLx register.
The DSP or host selects a words size using the DTYPE in the ports’s
SRCTLx register.
2. The DSP or host (depends on mode) writes the DMA channel’s
parameter registers (IIx, IMx, and Cx) and SRCTLx control register,
initializing the channel for receive.
3. The DSP or host (depends on mode) sets (=1) the channel’s SDEN
bit enabling the DMA process.
4. The external device begins writing data to the RXx buffer (through
the serial port).
5. The RXx buffer detects data is present and asserts an internal DMA
request to the I/O processor.
6. The I/O processor grants the request and performs the internal
DMA transfer, emptying the RXx buffer.
6-90
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
In general, the following sequence describes a typical internal to external
DMA operation where an external device transfers a block of data from
the DSP’s internal memory using a serial port:
1. The DSP or host (depends on mode) enables the DMA channel’s
serial port, setting the port’s SPEN bit in the port’s STCTLx register.
The DSP or host selects a words size using the DTYPE in the port’s
STCTLx register.
2. The DSP or host (depends on mode) writes the DMA channel’s
parameter registers (IIx, IMx, and Cx) and STCTLx control register,
initializing the channel for transmit.
3. The DSP or host (depends on mode) sets (=1) the channel’s SDEN
bit enabling the DMA process. Because this is a transmit, setting
SDEN automatically asserts an internal DMA request to the I/O
processor.
4. The I/O processor grants the request and performs the internal
DMA transfer, filling the TXx buffer.
5. The external device begins reading data from the TXx buffer
(through the serial port).
6. The TXx buffer detects that there is room in the buffer (it is now
“partially empty) and asserts another internal DMA request to the
I/O processor, continuing the process.
Using Two-Dimensional Serial Port DMA
Two-dimensional DMA is available on all link port and serial port DMA
channels. This DMA mode lets programs DMA data in memory treating
the data as an array. For more information, see “Using Two-Dimensional
Link Port DMA” on page 6-83.
ADSP-21160 SHARC DSP Hardware Reference
6-91
Optimizing DMA Throughput
Optimizing DMA Throughput
This section discusses overall DMA throughput when several DMA channels are trying to access internal or external memory at the same time.
Table 6-21 on page 6-93 summarizes the advantages of different system
configurations.
Internal Memory DMA
The DMA channels arbitrate for access to the DSP’s internal memory.
The DMA controller determines, on a cycle-by-cycle basis, which channel
is allowed access to the internal I/O bus and consequently which channel
will read or write to internal memory. The priority order of the DMA
channels appears in Table 6-1 on page 6-12.
Each DMA transfer takes one clock cycle even when different DMA channels are being allowed access on sequential cycles; i.e. there is no overall
throughput loss in switching between channels. Thus, four link port
DMA channels, each transferring one byte per cycle, would have the same
I/O transfer rate as one external port DMA channel transferring data to
internal memory on every cycle. Any combination of link ports, serial
ports, and external port transfers has the same maximum transfer rate.
6-92
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
External Memory DMA
When the DMA transfer is between DSP internal memory and external
memory, the external memory may have one or more wait states.
Figure 6-12 shows an example DMA hardware interface.
Table 6-21. Configurations for DSP—DSP (ADSP-2116x) DMA
DSP Configuration
(Data Source)
DSP Configuration
(Data Destination)
C/T1
Advantages, Disadvantages
Bus Master
DMA Master Mode
(MASTER= 1) TRAN=1,
EIx=address of EPBx buffer in destination, EMx= 0
Bus Slave
DMA Slave Mode
(MASTER= 0),
TRAN= 0
1
Advantage: Destination automatically generates interrupt upon
completion.
Disadvantage: DMA must be programmed on both source and destination.
Bus Master
DMA Master Mode
(MASTER= 1) TRAN=1,
EIx=MMS address in destination2, EMx=1
Bus Slave
Direct Write
1
Advantage: No programming
required for destination.
Disadvantage: No interrupt generated upon completion—source
must issue vector interrupt to destination.
Bus Slave
DMA Slave Mode
(MASTER= 0),
TRAN= 1
Bus Master
DMA Master Mode
(MASTER= 1),
TRAN=0, EIx=address
of EPBx buffer in
source, EMx=0
33
Advantage: Source automatically
generates interrupt upon completion.
Disadvantages: Slower throughput. DMA must be programmed
on both source and destination.
Bus Slave
Direct Read
Bus Master
DMA Master Mode
(MASTER= 1),
TRAN=0, EIx=MMS
address in source2,
EMx=1
43
Advantage: No programming
required for source.
Disadvantages: Slowest throughput. No interrupt generated upon
completion— destination must
issue vector interrupt to source.
1
2
3
C/T is throughput in cycles/transfer.
MMS= Multiprocessor Memory Space
Maximum burst throughput: 3-2-2-2, 4-2-2-2
ADSP-21160 SHARC DSP Hardware Reference
6-93
Optimizing DMA Throughput
External memory wait states, however, do not reduce the overall internal
DMA transfer rate if other channels have data available to transfer. In
other words, the DSP’s internal I/O data bus will not be held up by an
incomplete external transfer.
ADSP-2116X
5
BR2
ADDR23-0
BR1
DATA31-0
DMA
DATA
BUS
LATCH
16, 32, 48, OR 64
D
Q
DMA
READ
REQ.
OE
3
DMAR1
ID2-0
DMAG1
010
DMA
READ
GRANT
DMAR2
RDH/L
WRH/L
ACK
MS3-0
ADSP-2116X
5
3
COMMON REQ. LINE
HBR
COMMON GRANT LINE
DMAG2
HBG
16, 32,
48, OR
64
DMA
DATA
BUS
LATCH
BR1
ADDR23-0
Q
BR2
DATA31-0
OE
D
DMAR1
ID2-0
DMAG1
001
DMA
WRITE
REQ.
DMAR2
DMAG2
HBG
RDH/L
OE
HBR
WE
ACK
WRH/L
ACK
DMA
WRITE
GRANT
CS
MS3-0
ADDR
DATA
EXTERNAL MEMORY
Figure 6-12. Example DMA Hardware Interface
6-94
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
Notes on Figure 6-12:
Because DMARx and DMAGx are tied together, only one of the DSPs may have
DMA enabled at a time.
•
DMAGx
is only driven by the DSP bus master.
• The DMA Write Grant signal can be the combination of RDH/L and
MS3-0 instead of DMAG2 if paced master mode is used.
• The DMA Read Grant signal can be the combination of WRH/L and
MS3-0 instead of DMAG1 if paced master mode is used.
• DMA transfers may be to either DSP or to external memory (in
external handshake mode).
CLKIN
tHDGC
FETCH/DECODE CYCLES—2 CYCLES MIN.
DMARX
tSDRLC
DMAGX
tSDRHC
tWDGL
tVDATDGH
DATA31-0
VALID
Figure 6-13. DMAR and DMAG Timing
ADSP-21160 SHARC DSP Hardware Reference
6-95
Optimizing DMA Throughput
Notes on Figure 6-13:
•
DMARx setup times relate to the use of the signal in that cycle by the
DSP. DMA requests may be asserted asynchronously to CLKIN.
•
drives DATA63-0 if DSP is receiving. DMAGx latches
DATA63-0 if DSP is transmitting.
DMAGx
When data is to be transferred from internal to external memory, the
internal memory data is first placed in the external port’s EPBx buffer by
the DMA controller; the external memory access is then begun independently. (Likewise for external-to-internal DMA, the internal DMA
request will not be made until the external memory data is in the EPBx
buffer.) In both cases, the external DMA address generator—the EI and EM
parameter registers—maintains the external address until the data transfer
is completed. The internal and external address generators of a DMA
channel are decoupled and operate independently.
When EXTERN mode DMA transfers occur between an external device and
external memory, no internal resources of the DSP are utilized and internal DMA throughput is not affected.
System-Level Considerations
Slave mode DMA is useful in systems with a host processor because it
allows the host to access any DSP internal memory location while limiting
the address space the host must recognize—only the address space of the
DSP’s I/O processor registers. Slave mode DMA is also useful for
DSP-to-DSP DMA transfers.
Slave mode DMA has one drawback when interfacing to a slow host—the
fact that the external bus is held up during the transfer (whether initiated
by the DSP or the host) and no other transactions can proceed. To overcome this, the handshake DMA mode may be used.
6-96
ADSP-21160 SHARC DSP Hardware Reference
I/O Processor
In Handshake mode, the host does not have to master the bus in order to
make a DMA request, nor does the DSP (in master mode) have to wait on
the bus for the transfer to complete. Instead, the host asserts the DMARx
pin. When the DSP is ready to make the transfer, it can complete it in one
bus cycle. For more information, see “Handshake Mode” on page 6-34.
ADSP-21160 SHARC DSP Hardware Reference
6-97
Optimizing DMA Throughput
6-98
ADSP-21160 SHARC DSP Hardware Reference
External Port
7 EXTERNAL PORT
The DSP’s external port extends the DSP’s address and data buses
off-chip. Using these buses and external control lines, systems can interface the DSP with external memory, 16- or 32-bit host processors, and
other DSPs. Because many of the external port operations relate to external memory accessing or I/O processing, this chapter refers to the memory
and I/O processor chapters (“Memory” and “I/O Processor”) frequently.
Overview
This chapter describes connection and timing issues for the external port.
The main sections of this chapter describe the interfaces that are available
through the external port. These interfaces include:
• “External Memory Interface” on page 7-3
• “Host Processor Interface” on page 7-49
• “Multiprocessor (DSPs) Interface” on page 7-91
Data alignment through the external port is identical for these interfaces.
Figure 7-1 on page 7-2 shows the external port’s data alignment.
ADSP-21160 SHARC DSP Hardware Reference
7-1
Setting External Port Modes
DATA63-0
63
55
47
39
31
23
15
BYTE 7
7
0
BYTE 0
RDL/WRL
RDH/WRH
64-BIT LONG WORD, SIMD, OR DMA TRANSFERS
64-BIT TRANSFER FOR 48-BIT INSTRUCTION FETCH
64-BIT TRANSFER FOR 40-BIT EXT. PREC.
32-BIT NORMAL WORD(EVEN ADDR)
32-BIT NORMAL WORD (ODD ADDR)
RESTRICTED DMA, HOST, EPROM DATA ALIGNMENTS:
32-BIT PACKED
16-BIT PACKED
EPROM
Figure 7-1. External Port Word Alignment
Setting External Port Modes
The SYSCON, WAIT, and DMACx registers control the external port operating
mode. Table A-17 on page A-45 lists all the bits in SYSCON, Table A-19 on
page A-49 lists all the bits in WAIT, and Table A-21 on page A-55 lists all
the bits in DMACx. For information about setting up memory access modes
(synchronous versus asynchronous interface), see “Setting Data Access
Modes” on page 5-28.
7-2
ADSP-21160 SHARC DSP Hardware Reference
External Port
For information on setting DMA through the external port, see “Setting
I/O Processor—EPort Modes” on page 6-14. For information on using
external port interrupts, see “Using I/O Processor Status” on page 6-53.
is a 3:1 conflict resolution ratio at the external port interface
 There
(three internal buses to one external bus) in addition to the 2:1 or
greater clock ratio between the DSP’s internal clock and the external system clock. Systems that fetch instructions or data through
the external port must tolerate at least one cycle—and possibly
many additional cycles—of latency.
External Memory Interface
In addition to its on-chip SRAM, the DSP provides addressing of up to 4
gigawords of off-chip memory through its external port. This external
address space includes multiprocessor memory space—the on-chip memory of all other DSPs connected in a multiprocessor system—as well as
external memory space—the region for standard addressing of off-chip
memory.
Figure 7-2 shows how the buses and control signals extend off-chip, connecting to external memory.
The DSP’s memory control signals permit direct connection to fast static
RAM devices. Memory mapped peripherals and slower memories can also
connect to the DSP using a user-defined combination of programmable
waitstates and hardware acknowledge signals.
External memory can hold instructions and data. The external data bus
(DATA63-0) must be 64 bits wide to transfer 48-bit instructions and 40-bit
extended-precision floating-point data without data packing. If external
memory contains only data or packed instructions for transfer by DMA,
the external data bus width can be either 16 or 32 bits wide. In a 16- or
32-bit bus system, the DSP’s on-chip I/O processor unpacks incoming
ADSP-21160 SHARC DSP Hardware Reference
7-3
External Memory Interface
data and packs outgoing data. Figure 7-1 shows how the DSP transfers
different data word sizes over the external port.
ADSP-2116X
CLKIN
4
FLAG3-0
TIMEXP
LINK
DEVICES
(6 MAX)
(OPTIONAL)
LXACK
SERIAL
SERIAL
DEVICE
DEVICE
(OPTIONAL)
(OPTIONAL)
TCLK0
RCLK0
TFS0
RSF0
DT0
DR0
SERIAL
SERIAL
DEVICE
DEVICE
(OPTIONAL)
(OPTIONAL)
TCLK1
RCLK1
TFS1
RSF1
DT1
DR1
LXCLK
LXDAT7-0
RPBA
ID2-0
RESET
ADDR
CIF
DATA
BRST
ADDR31-0
ADDR
DATA63-0
DATA
OE
WE
RDX
WRX
ACK
ACK
CS
MS3-0
PAGE
SBTS
CLKOUT
DMAR1-2
BOOT
EPROM
(OPTIONAL)
MEMORY
AND
PERIPHERALS
(OPTIONAL)
DATA
3
EBOOT
LBOOT
IRQ2-0
CS
BMS
CLK_CFG3-0
ADDRESS
4
CONTROL
CLOCK
DMA DEVICE
(OPTIONAL)
DATA
DMAG1-2
CS
HBR
HBG
REDY
HOST
PROCESSOR
INTERFACE
(OPTIONAL)
BR1-6
ADDR
PA
DATA
JTAG
6
Figure 7-2. ADSP-21160 Processor System
7-4
ADSP-21160 SHARC DSP Hardware Reference
External Port
Table 7-1 defines the DSP pins used for interfacing to external memory.
Table 7-1. External Memory Interface Signals
Pin
Type
Function
ADDR 31-0
I/O/T
External Bus Address. The DSP outputs addresses for external memory and peripherals on these pins. In a multiprocessor system, the bus
master outputs addresses for read/writes of the internal memory or
IOP registers of other DSPs. The DSP inputs addresses when a host
processor or multiprocessing bus master is reading or writing its
internal memory or I/O processor registers.
DATA 63-0
I/O/T
External Bus Data. The DSP inputs and outputs data and instructions on these pins. 32-bit single-precision floating-point data and
32-bit fixed-point data is transferred over bits 63-32 or 31-0 of the
bus. 40-bit extended-precision floating-point data is transferred over
bits 63-24 of the bus. 16-bit short word data is transferred over bits
47-32 of the bus. Pull-up resistors on unused DATA pins are not necessary. In asynchronous access mode, read data is sampled by the rising edge of the read strobe (DATA 63-32 sampled with RDH,
DATA31-0 sampled with RDL). On write operations, the data is
driven from rising edge of CLKIN, before the write strobes are
asserted.
MS3-0
O/T
Memory Select Lines. These lines are asserted (low) as chip selects for
the corresponding banks of external memory. Memory bank size must
be defined in the DSP’s system control register (SYSCON). The
MS3-0 lines are decoded memory address lines that change at the
same time as the other address lines. When no external memory
access is occurring the MS3-0 lines are inactive. In asynchronous
access mode, the MSx signal is asserted for the whole access. In synchronous access mode, the MSx signal is only asserted until ACK is
sampled asserted. MS0 can be used with the PAGE signal to implement a bank of DRAM memory (Bank 0). In a multiprocessing system the MS3-0 lines are output by the bus master.
Unlike previous SHARC DSPs, strobe assertion for conditional
instructions occurs only when the instruction condition code evaluates as true.
CLKOUT
O/T
Synchronous output clock. Output clock signal at same rate as
CLKIN. Output by current bus master.
I (Input), S (Synchronous), o/d (Open Drain), O (Output), A (Asynchronous), a/d (Active
Drive), T (Three-state, when SBTS or HBR is asserted, or when the DSP is a bus slave)
ADSP-21160 SHARC DSP Hardware Reference
7-5
External Memory Interface
Table 7-1. External Memory Interface Signals (Cont’d)
Pin
Type
Function
PAGE
O/T
DRAM Page Boundary. The DSP asserts this pin to signal that an
external DRAM page boundary has been crossed. DRAM page size
must be defined in the DSP’s memory control register (WAIT).
DRAM can only be implemented in external memory Bank 0; the
PAGE signal can only be activated for Bank 0 accesses. In a multiprocessing system, PAGE is output by the bus master.
RDH/L
I/O/T
Read High, and Read Low Strobes. RDH indicates that a read of the
high word of the data bus (DATA63-32) is in progress. RDL indicates
that a read of the low word of the data bus (DATA31-0) is in progress.
As a master, the DSP asserts the strobe after the ADDR31-0 and
MS3-0 assert, unless the following bus operation is to the same bank
or multiprocessor memory and asserts the same strobe. Timing of the
deassertion of the strobe depends upon the access mode. In asynchronous access mode, the strobe is deasserted before the rising edge of
CLKIN. For an access to a bank in synchronous access mode, the
strobe is deasserted on the rising edge of CLKIN.
As a slave, the DSP samples this input to determine the type of bus
operation, as well as the size and data alignment for the transfer.
WRH/L
I/O/T
Write High, and Write Low Strobes. WRH indicates that a write on the
high word of the data bus (DATA63-32) is in progress. WRL indicates
that a write on the low word of the data bus (DATA31-0) is in progress.
As a master, the DSP asserts the strobe after the ADDR31-0 and
MS3-0 assert, unless the following bus operation is to the same bank
or multiprocessor memory and asserts the same strobe. Timing of the
deassertion of the strobe depends upon the access mode. In asynchronous access mode, the strobe is deasserted before the rising edge of
CLKIN. For an access to a bank in synchronous access mode, the
strobe is deasserted on the rising edge of CLKIN.
As a slave, the DSP samples this input to determine the type of bus
operation, as well as the size and data alignment for the transfer.
I (Input), S (Synchronous), o/d (Open Drain), O (Output), A (Asynchronous), a/d (Active
Drive), T (Three-state, when SBTS or HBR is asserted, or when the DSP is a bus slave)
7-6
ADSP-21160 SHARC DSP Hardware Reference
External Port
Table 7-1. External Memory Interface Signals (Cont’d)
Pin
Type
Function
CIF
O/T
Core Instruction Fetch. As a master, the DSP asserts (low) this output
when the program sequencer of the DSP is making an off-chip
instruction fetch (read) only. The address generated for this request is
a 48-bit instruction pointer. If the instruction fetch is to an address
in one of the external memory banks, the MSx output for that bank is
also asserted. This output has timing similar to the MS3-0signals.
BRST
I/O/T
Burst Transfer. This signal is asserted (high) by a bus master, to indicate that the current bus read or write transfers a block of data to
contiguous, incrementing, 64-bit aligned addresses, over multiple
cycles. Each individual data transfer requires an acknowledgment
(ACK assertion) from the slave addressed by the transfer. BRST is
asserted as an output by the DSP bus master in the cycle after the first
cycle in which ACK is sampled asserted.
As a synchronous slave, the DSP samples the BRST input to determine if a burst read transfer is in progress. The DSP slave does not
support burst write transfers.
When interfacing to SBSRAM gluelessly, this output should be connected to the ADSC input of the SBSRAMs (not ADV).
ACK
I/O/S
Memory Acknowledge. External devices can deassert ACK (low) to
add waitstates to an external memory access (including individual
transfers within a burst access). ACK is used by I/O devices, memory
controllers, or other peripherals to hold off completion of an external
memory access.
As a bus master, the DSP samples this input. In asynchronous access
mode, ACK is not sampled until the programmed number of waitstates for the access have been counted. For an access to a bank in
synchronous access mode, ACK is sampled each CLKIN cycle even
during programmed waitstate count. The DSP has a keeper latch on
its ACK pin that maintains the input at the level it was last driven.
ACK must be sampled high by the DSP before it asserts the strobe(s)
for a bus operation. Slaves must assert ACK before three-stating this
signal.
As a slave, the DSP deasserts ACK as an output, to add waitstates to a
synchronous access of its internal memory or IOP register space.
I (Input), S (Synchronous), o/d (Open Drain), O (Output), A (Asynchronous), a/d (Active
Drive), T (Three-state, when SBTS or HBR is asserted, or when the DSP is a bus slave)
ADSP-21160 SHARC DSP Hardware Reference
7-7
External Memory Interface
maximum flexibility when interfacing the DSP to 32-bit wide
 For
memory, connect the memory’s data lines to the DSP’s
DATA63-32
pins; do not connect the A0 pin. This alignment permits more
packing options and lets supports easier DMA to the external
memory. In DMA accesses to such memory, the DMA uses a stride
of two.
Figure 7-1 also shows how the DSP stores unpacked data and instructions
in the 64-bit wide external memory. The external memory map is organized such that consecutive addresses access adjacent 32-bit memory
locations. For off-chip instruction fetches, the program sequencer accesses
adjacent 48-bit wide memory locations.
ADSP-21160 external memory interface differs from previous
 The
SHARC DSPs. Compared to previous SHARC DSPs, the interface
has added signals that support burst transfers and the 64-bit data
bus. The synchronous interface delivers greater performance, while
the asynchronous interface remains similar to previous SHARC
DSPs. The external interface provides glueless support for many
asynchronous and/or synchronous devices, including other DSPs.
The DSP’s burst transfer protocol supports Synchronous Burst
SRAMs (SBSRAMs).
Because the memory sub-system uses a 64-bit wide data bus, the DSP has
high and low read and write strobes (RDH, RDL, WRH, WRL) to mask and
enable 32-bit normal word lanes on the DATA63-0 bus. Note that the least
significant bit, ADDR0, of the ADDR31-0 bus may be disregarded during DSP
external memory space accesses of 32-bit locations (CIF deasserted), as this
information is redundant with the strobes. For more information on packing modes in which the DSP only uses the RDH and WRH pins for accesses,
see Table 6-2 on page 6-7.
require the least significant address bit to support off-chip
 Systems
instruction execution by the core (Core Instruction Fetch,
,
CIF
asserted), DMA packing modes (including EPROM booting), and
host-DSP accesses.
7-8
ADSP-21160 SHARC DSP Hardware Reference
External Port
External memory can hold both instructions and data. The external memory must support the full width of the data bus (DATA63-0) to achieve
maximum performance. If the DSP DAGs generate external accesses to
Long word data (including 48-bit instructions or 40-bit Extended Precision Normal word data) or if the DSP accesses external memory while in
SIMD mode, the system must implement the full 64-bit external data bus.
Also, the system must support the full 64-bit external data bus if the DSP
makes burst DMA transfers.
ADSP-21160 DSP does not support direct data transfers of
 The
48-bit instructions or 40-bit extended precision data to or from
external memory.
For example:
dm(0x800100) = r0;
// moves 32 MSBs of r0
dm(0x800100) = r0 (LW); // moves 32 bits from r0
// and 32-bits from r1
To move instructions or 40-bit extended precision data to or from external memory, programs should use the PX register as an intermediate 64-bit
holding register. Also, programs can use the I/O processor to transfer this
data through an EPBx FIFO.
For example:
dm(0x800100) = px;
//moves 48 MSBs of px
ADSP-21160’s external PM address bus is 32 bits wide. The
 The
DSP’s DM address, PM address, and I/O processor can address the
entire 4-gigaword external memory space. The ADSP-21160’s program sequencer, like previous SHARC DSPs, only can address the
low 24-bits of address space.
ADSP-21160 SHARC DSP Hardware Reference
7-9
External Memory Interface
Banked External Memory
The DSP divides external memory into four equal-size, programmable
banks. By mapping peripherals into different banks, systems can accommodate I/O devices with different timing requirements. For information
on configuring these memory banks for waitstates and synchronous or
asynchronous access modes, see “Setting Data Access Modes” on
page 5-28.
the ADSP-21160 DSP, Bank 0 starts at address 0x0080 0000
 On
in external memory and is followed in order by Banks 1, 2, and 3.
When the DSP generates an address located within one of the four
banks, the DSP asserts the corresponding memory select line,
MS3-0.
The MS3-0 outputs serve as chip selects for memories or other external
devices, eliminating the need for external decoding logic. MS0 provides a
select line for an optional bank of DRAM memory, when used in combination with the PAGE signal. For more information, see “DRAM Page
Boundary Detection” on page 7-15.
The MS3-0 lines are decoded memory address lines that change at the same
time as the other address lines. When no external memory access is occurring, the MS3-0 lines are inactive.
previous SHARC DSPs, strobe assertion for conditional
 Unlike
instructions occurs only when the instruction condition code evaluates as true.
Unbanked External Memory
The region of external memory above Banks 0-3 is called unbanked external memory space. No MSx memory select line is asserted for accesses in
this address space. For information on configuring this unbanked memory
for waitstates and synchronous or asynchronous access modes, see “Setting
Data Access Modes” on page 5-28.
7-10
ADSP-21160 SHARC DSP Hardware Reference
External Port
Boot Memory
Most often, the DSP only asserts the BMS memory select line when the
DSP is reading from a boot EPROM. This line allows access to a separate
external memory space for booting. Unbanked memory waitstates and
mode are applied to BMS-selected accesses.
The BMS output is only driven by the DSP bus master. For more information on booting, see “Bootloading Through The External Port” on
page 6-76 or “Bootloading Through The Link Port” on page 6-87.
It is also possible to write to boot memory using BMS. For more information, see “Using Boot Memory” on page 5-29.
Idle Cycle
A bus idle cycle is an inactive bus cycle that the DSP automatically generates to avoid data bus driver conflicts. Such a conflict can occur when a
device with a long output disable time continues to drive after RDH/L is
deasserted while another device begins driving on the following cycle. Idle
cycles are also required to provide time for a slave in one bank to
three-state its ACK driver, before the slave in the next bank enables its ACK
driver in the synchronous access modes.
ADSP-21160 SHARC DSP Hardware Reference
7-11
External Memory Interface
Figure 7-3 shows idle cycle insertion between a synchronous read and a
zero-wait, synchronous write in cycle 3.
READ OP
1
2
IDLE CYCLE
WRITE OP
3
4
5
CLKIN
ADDRESS[31:0]
MS[3:0]
RDH
RDL
WRH
WRL
BRST
DATA[63:0]
ACK
Figure 7-3. Idle Cycle Example
7-12
ADSP-21160 SHARC DSP Hardware Reference
External Port
To avoid this conflict, the DSP generates an idle cycle in the following
cases.
• On a transition from a read operation to a write operation in the
same bank.
• On a transition from one bank, or multiprocessor memory ID space
to any other bank or multiprocessor slave ID space, independent of
access mode.
previous SHARC DSPs, the ADSP-21160 DSP does not
 Unlike
support idle cycle insertion on a page boundary crossing.
Data Hold Cycle
The data hold cycle is another configurable memory access feature for
adding cycles much like waitstates, as discussed in “Setting Data Access
Modes” on page 5-28. A hold time cycle is an inactive bus cycle that the
DSP automatically generates at the end of a read or write to allow a longer
hold time for address and data. The address, data (if a write), and bank
select (if in banked external memory) remain unchanged and are driven
for one cycle after the read or write strobes are deasserted. The DSP inserts
the data hold cycle only in asynchronous mode and only if the number of
programmed waitstates code is 010–111.
Figure 7-4 demonstrates a hold time cycle appended to an asynchronous
write access (EBxWS=011).
ADSP-21160 SHARC DSP Hardware Reference
7-13
External Memory Interface
HOLD TIME
CYCLE
WRITE OPERATION
1
2
3
4
5
CLKIN
ADDRESS[31:0]
MS[3:0]
WRH
WRL
DATA[63:0]
Figure 7-4. Hold Time Cycle Example
ADSP-21160 DSP does not append an Idle cycle after a Hold
 The
cycle.
Multiprocessor Memory Space Waitstates and Acknowledge
Multiprocessor memory space uses only the synchronous transfer protocols, using the zero-waitstate access for writes and a minimum 1-waitstate
access for reads. Slave DSPs deassert ACK if more access time is required.
DMA burst transfers are only defined for direct read access of a DSP
slave’s internal memory and reads from the external port buffers (EPBx).
For more information, see “Multiprocessor (DSPs) Interface” on
page 7-91.
The ADSP-21160 DSP does not support the
 ous
SHARC DSPs.
7-14
MMSWS
bit from previ-
ADSP-21160 SHARC DSP Hardware Reference
External Port
DRAM Page Boundary Detection
Applications with large amounts of data may want to use DRAM memory
for bulk storage. To simplify interfacing to page-mode or static-column
DRAMs, the DSP detects page boundary crossings and outputs the PAGE
signal to an external DRAM controller. Figure 7-5 shows an example DSP
system with DRAM.
Host
ADSP-2116x
HBR
HBG
REDY
MS0
DATA63-0
ADDR31-0
RDH/L
5
BR1
BR2
3
010
WRH/L
BR3-6 ACK
PAGE
SBTS
ID2-0
CS
11
DATA31-0
HBR
MS0
HBG
DATA63-0
REDY
OE
ADDR31-0
CS
22
RDH/L
WRH/L
5
BR 2-6
BR1
3
ID2-0
4M
DRAM
RAS CAS WR
DATA
ADSP-2116x
ADDRESS
6
001
ADDR10-0
ACK
RAS CAS WR
ADDR21-0
RD
WR
PAGE
ACK
PAGE
SBTS
TS
DRAM
Controller
Figure 7-5. Example DRAM Interface
ADSP-21160 SHARC DSP Hardware Reference
7-15
External Memory Interface
Different interfacing methods may be needed in some applications, especially if buffers are needed for the DRAM.
Page boundaries are user-defined. Boundaries must be programmed in the
WAIT register. For more information, see “Setting Data Access Modes” on
page 5-28. Automatic page boundary detection is provided by the DSP’s
PAGE signal. Systems may only place DRAM memory in bank 0 of external
memory—the PAGE signal is only active within bank 0. Programs write the
page size for page boundary detection in the PAGSZ field of the WAIT
register.
The DSP asserts the PAGE pin when an external access crosses a page
boundary and the address is within bank 0. The processor detects a
boundary crossing by comparing each address output for bank 0 to the
address of the last successful external access, which is stored in the I/O
processor ELAST register. If a memory access is aborted—for example, due
to a conditional write, the DSP does not assert the PAGE pin and does not
update the current page in ELAST. Also, the DSP does not assert the PAGE
pin or update the current page if the access is to multiprocessor memory
space or to any memory space other than bank 0 of external memory
space.
The PAGE pin remains asserted as long as the access is active. PAGE is not
asserted if no access is performed. The current page is automatically invalidated and the PAGE pin asserted upon the next external access if: 1) the
DSP loses mastership of the external bus to another DSP or to a host processor, or 2) the processor is reset. Programs should not read ELAST in the
cycle immediately after it is written, because it may be in the process of
updating.
The host bus request pin (HBR) is disabled when the PAGE pin is asserted.
Disabling HBR prevents the possibility of the DSP becoming a bus slave
through deadlock resolution while the DRAM controller is servicing a
page change.
7-16
ADSP-21160 SHARC DSP Hardware Reference
External Port
In page DRAM systems, the DSP may need to recover from DRAM page
fault conditions, using the Suspend Bus Three-State (SBTS) pin. External
devices can assert the DSP’s SBTS input to place the external bus address,
data, selects, burst, and strobes in a high-impedance state. This input is
sampled by the DSP on the rising edge of CLKIN. The DSP external bus
outputs three-state later in the cycle in which SBTS is sampled asserted. If
the DSP core attempts to access external memory while SBTS is asserted,
the processor halts and the memory access does not complete until SBTS is
sampled deasserted.
should only be used to recover from DRAM page faults or
 host processor/DSP
deadlock condition. For more information, see
SBTS
“Deadlock Resolution” on page 7-84. In the case of DRAM page
faults, SBTS allows the external DRAM controller to take control of
the external bus. SBTS three-states the signals.
Table 7-2. Signals SBTS Three-States
ADDR31-0
RDH/L
BRST
DMAG1
MS3-0
CIF
PAGE
DATA63-0
WRH/L
DMAG2
BMS
CLKOUT
When the DSP uses SBTS for resolving bus deadlock, SBTS operates differently than when a host processor uses and SBTS and HBR. For more
information see how the host processor uses SBTS and HBR as discussed at
the end of the section “Synchronous Burst Read Transfers” on page 7-65.
When SBTS is asserted, the DSP places the external bus address, data,
selects, and strobes in a high-impedance state for the following cycle. If an
external access is underway when SBTS is asserted, the access is held off (as
if ACK were deasserted), the bus is three-stated, and the memory access
continues in the cycle after the deassertion of SBTS. If SBTS is asserted
while no external access is occurring, the external bus pins are three-stated
and the DSP continues running until it tries to perform an external access.
ADSP-21160 SHARC DSP Hardware Reference
7-17
External Memory Interface
The DSP then halts. In this case, the memory access begins in the cycle
after the deassertion of SBTS.
When SBTS is deasserted, the DSP reasserts the RDH/L, WRH/L, and DMAGx
strobes—if they had been asserted prior to SBTS—after the external
address has become valid, asserting them at their normal timing within the
cycle. The waitstate counter is reset.
differs from HBR in that SBTS takes effect in the next cycle, even if an
external access is occurring but not finished. Systems should only use SBTS
when the external access is to a device such as a DRAM or cache memory
where the access must be held off in order to prepare for the access. Using
SBTS at other times—such as during DSP-to-DSP accesses or when DMAGx
is asserted—results in incorrect operation.
SBTS
Timing External Memory Accesses
Memory access timing for external memory space and multiprocessor
space is the same. For exact timing specifications, refer to ADSP-21160
DSP Microcomputer Data Sheet.
The DSP can interface to external memories and memory-mapped peripherals that operate asynchronously with respect to CLKIN. The DSP also
supports synchronous external memories and memory-mapped peripherals. Synchronous devices derive all of their bus timing with respect to
CLKIN of the DSP.
The synchronous interface mode supports DMA burst transfers, which
can significantly improve bus throughput for large, contiguous block
transfers. The synchronous interface protocols are compatible with Synchronous Burst SRAMS (SBSRAMs) from a variety of vendors. In a
multiprocessing system, the DSP must be the bus master in order to access
external memory.
7-18
ADSP-21160 SHARC DSP Hardware Reference
External Port
Asynchronous Mode Interface Timing
Figure 7-6 shows typical timing for an asynchronous read or write of
external memory. Here, the CLKIN clock signal appears only to indicate
that the access occurs within a single CLKIN cycle. All timing for the master
DSP is derived synchronously from CLKIN. The asynchronous slave mode
modifies the basic synchronous access to better support slaves whose timing is not derived from CLKIN.
CLKIN
ADDRESS[31:0]
READ/WRITE ADDRESS
MS[3:0]
RDH/L OR WRH/L
(WRITE) DATA[63:0]
WRITE DATA
(READ) DATA[63:0]
ACK
Figure 7-6. External Memory Asynchronous Access Cycle
ADSP-21160 SHARC DSP Hardware Reference
7-19
External Memory Interface
Figure 7-7 shows timing relationships employed by the asynchronous
external access mode.
1
2
CLKIN
ADDRESS[31:0]
A0
MS[3:0]
BRST
RDH
RDL
WRH
WRL
ACK
DATA[63:0]
D0
Figure 7-7. Asynchronous Access Timing Derivation
7-20
ADSP-21160 SHARC DSP Hardware Reference
External Port
In this mode, the following occurs.
• The strobes assert and deassert based on timing derived from an
internal clock whose frequency is twice that of the core clock. (This
differs from synchronous mode where the strobes assert from the
same edge.) The trailing edge timing is derived from the rising edge
of the internal version of CLKIN.
• The MSx memory select lines are held stable for the entire access.
(This differs from synchronous read or synchronous write—minimum 2-cycle—modes where the memory select lines are deasserted
after the first ACK-ed cycle of the transfer.)
• For read operations, DATA63-32 are sampled by the DSP on the rising edge of the RDH. DATA31-0 are sampled by the rising edge of RDL.
(This differs from synchronous mode where DATA63-0 are sampled
by the internal version of CLKIN.)
Asynchronous Mode Read – Bus Master
DSP bus master reads of external memory, in asynchronous mode, occur
with the following sequence of events as shown in Figure 7-6 on
page 7-51.
1. The DSP samples ACK synchronously. If ACK is asserted, the DSP
drives the read address and asserts a memory select signal (MS3-0) to
indicate the selected bank. A memory select signal is not deasserted
between successive accesses of the same memory bank. If ACK is
sampled deasserted, the DSP waits one CLKIN cycle and samples ACK
again.
2. The DSP asserts the read strobes. Strobe assertion is determined by
the size and alignment of the data transfer. For more information
on data alignment, see Figure 7-1 on page 7-2.
ADSP-21160 SHARC DSP Hardware Reference
7-21
External Memory Interface
3. The DSP checks whether waitstates are needed. If so, the memory
select and read strobe remain active for additional cycles. Waitstates are determined by a combination of the state of the external
acknowledge signal (ACK) and the internally programmed waitstate
count.
4. The DSP deasserts the read strobe(s) in the cycle where no further
waitstates are indicated. The data bus (DATA63-0) is sampled on the
rising edge of the read strobe(s).
5. If a Hold cycle is programmed for the accessed bank (via the EBxWS
parameter of the WAIT register), the address bus and memory selects
are held stable for an additional cycle. If initiating another read
memory access to the same bank, the DSP drives the address and
memory select for that access in the next cycle.
Asynchronous Mode Write – Bus Master
DSP bus master writes to external memory, in asynchronous mode, occur
with the following sequence of events as shown in Figure 7-5 on
page 7-15.
1. The DSP samples ACK synchronously. If ACK is asserted, the DSP
drives the write address and asserts a memory select signal (MS3-0)
to indicate the selected bank. A memory select signal is not deasserted between successive accesses of the same memory bank. The
DSP also drives the write data (DATA63-0). If ACK is sampled deasserted, the DSP waits one CLKIN cycle and samples ACK again.
2. The DSP asserts the write strobes. Strobe assertion is determined
by the size and alignment of the data transfer. For more information, Figure 7-1 on page 7-2.
3. The DSP checks whether waitstates are needed. If so, the memory
select and write strobes remain active for additional cycles. Waitstates are determined by the state of the external acknowledge
signal (ACK) and the internally programmed waitstate count.
7-22
ADSP-21160 SHARC DSP Hardware Reference
External Port
4. The DSP deasserts the write strobes near the end of the cycle where
no further waitstates are indicated.
5. The DSP three-states its data outputs, unless the next access is also
a write to the same bank, or if a Hold cycle is programmed for the
accessed bank using the EBxWS parameter of the WAIT register. If a
Hold cycle is inserted, the address bus, data bus, and memory
selects are held stable for an additional cycle. If initiating another
memory access to the same bank, the DSP drives the address, memory select for the next access in the following cycle.
Synchronous Mode Interface Timing
Any slave addressed by a DSP in a bank configured for synchronous transfer mode must use a clock with the same frequency and phase
characteristics to the clock which drives CLKIN on the DSP. The slave samples all inputs, and drives all outputs on the rising edge of this clock.
Except for zero-waitstate writes, the slave must assert ACK at least twice for
each access; once to acknowledge the address/command (strobe assertion)
and once (if not a burst) or more to acknowledge the data transfer.
The following notes apply to all synchronous access modes:
• A slave recognizes the start of a valid bus operation by synchronously sampling one or more of the strobes asserted and ACK
asserted—but not by this slave, which would indicate the end of
the transfer.
• For each of the non-burst, synchronous read/write accesses (except
zero-waitstate writes), the master recognizes the end of the access as
the cycle in which 1) the slave samples or drives data in response to
a valid operation driven by the master (read or write), 2) the slave
asserted ACK to the master {except for zero-waitstate write
ADSP-21160 SHARC DSP Hardware Reference
7-23
External Memory Interface
operations}, and 3) the number of waitstates for read or write
access to that bank have occurred—asserting ACK does not terminate the wait count early.
• The program must select a number of waitstates that is consistent
with the access time for the slave addressed by that external memory bank.
• For the zero-waitstate writes, the access can only be extended
beyond one clock cycle by deasserting ACK in the cycle of the transfer. This extension can occur on back-to-back writes in which ACK
is deasserted due to full write buffer capacity from the previous
write, or slaves can asynchronously deassert ACK in the first cycle.
• Deasserting ACK during the initial command phase does inhibit
waitstate count and change of bus signals. After the first ACK assertion, deasserting ACK for the data phase does not inhibit waitstate
counting.
• Only one slave (or driver for ACK) should be allocated per external
memory bank. More than one slave may introduce ACK drive
contention.
• The read/write strobes for an access do not assert until ACK sampled
asserted. This conditional strobe assertion delays the start of an
access until ACK is asserted by the previous slave. This sampling is
because the slave target of a single-cycle write operation may have
to deassert ACK in the cycle after the bus cycle, to stall further writes
to that slave. To provide a cycle for the previous slave to three-state
its ACK driver before the next slave drives ACK, the next operation to
a new bank must not launch on the bus.
7-24
ADSP-21160 SHARC DSP Hardware Reference
External Port
• Write/read access stalls (no state change, other than internal waitstate counting) on the bus if ACK is deasserted in cycle(s) of data
transfer.
• The last read/write operation must be ACK-ed before a transition to
a new bus master (BTC), bank, or multiprocessor space slave
occurs. The master always inserts an Idle cycle on this transition.
No pipelining can occur across these boundaries.
Synchronous Mode Read – Bus Master
An example synchronous read cycle appears in Figure 7-8.
1
2
CLKIN
ADDRESS 31-0
MS3-0
RDH
RDL
WRH
WRL
BRST
DATA 63:0
ACK
Figure 7-8. Typical Synchronous Read Timing
ADSP-21160 SHARC DSP Hardware Reference
7-25
External Memory Interface
Propagation delays are not shown in this timing diagram. Because a synchronous access requires a rising clock edge for the slave to sample the
asserted signals of the master (and for the master to sample slave), the
minimum read access in the synchronous mode is two cycles.
synchronous access mode, the waitstate selection in the
 Inregister
) must be 001 or greater.
=000 is not sup(
WAIT
EBxWS
EBxWS
ported in synchronous access mode.
This example demonstrates a minimum latency, one-waitstate, 32-bit
(normal word) read, from an even address in external memory (had the
32-bit access been to an odd 32-bit address, RDH would have asserted
instead of RDL.)
that do not support the entire 64-bit data bus width do not
 Slaves
have to connect to both read strobes. Also, slaves that do not support bursting protocols do not need to connect to the BRST signal.
Bus master synchronous reads from external memory occur with the following sequence of events as shown in.
1. (cycle 1 in Figure 7-8 on page 7-25) If ACK is sampled as asserted at
the beginning of cycle 1, the DSP drives the read address and
asserts a memory select signal (MS3-0) to indicate the selected bank.
The DSP asserts the RDH/RDL strobes to indicate the size and alignment of the requested data. The read strobes are not deasserted
between successive read accesses of the same memory bank. If the
size or alignment changes, strobe assertion also changes. Strobe
assertion is determined by the size and alignment of the data transfer. For more information on data alignment, see Figure 7-1 on
page 7-2.
2. (cycle 2) If ACK was sampled as deasserted at the beginning of the
cycle, the MSx strobes would remain asserted. If ACK was sampled
asserted (as shown in Figure 7-8), the MSx strobes would deassert.
The slave must be capable of detecting that MSx was asserted in
7-26
ADSP-21160 SHARC DSP Hardware Reference
External Port
cycle 1 and retain this information internally. If ACK was deasserted
by the previous slave (for a single-cycle write), deassertion of the
MSx is delayed.
3. (cycle 2) The DSP checks whether more than one waitstates are
needed. If so, the read strobes remain active for additional cycle(s).
Waitstates are determined by a combination of the state of the
external acknowledge signal (ACK) and the programmed waitstate
count.
4. (end of cycle 2) The data bus (DATA63-0) is sampled on the rising
edge of CLKIN.
5. (cycle 3) If initiating another read memory access to the same bank,
the DSP drives the address, memory select, and strobes for the next
access.
Figure 7-1 on page 7-HIDDEN shows back-to-back reads to the same
bank with the second access stalled for one cycle by the slave deasserting
ACK. This example assumes that the EBxWS=001 for this bank, indicating
one internal waitstate.
Synchronous Write, Zero-Waitstate Mode
Figure 7-9 shows typical synchronous write cycle timing. Propagation
delays are not shown in this timing diagram. Synchronous access requires
a rising clock edge for the slave to sample the asserted signals of the master
(and for master to sample slave). In the case of writes, the latency can be
reduced to a single cycle if the slave always latches the bus signals on each
clock cycle (it does not sample ACK). For example, the slave cannot sample
the bus, decode that it is being addressed as a slave, and sample the write
data of the bus in the following cycle.
ADSP-21160 SHARC DSP Hardware Reference
7-27
External Memory Interface
1
2
3
CLKIN
ADDRESS[31:0]
WRITE #1
WRITE #2
WRITE #1
WRITE #2
MS[3:0]
RDH
RDL
WRH
WRL
BRST
DATA[63:0]
STALL 2ND WRITE
ACK
Figure 7-9. Typical Synchronous Write Example
The slave samples the bus each cycle and decodes the sampled value to
determine if that slave was addressed by the write operation. If the slave’s
write queue goes full with that write, the slave deasserts ACK in the cycle
after the write operation transferred on the bus. Any subsequent bus operation (read or write) stalls until ACK is sampled asserted, as shown in
Figure 7-9.
7-28
ADSP-21160 SHARC DSP Hardware Reference
External Port
The example demonstrates a minimum latency, zero-waitstate, 64-bit
(Long word) write in cycle 1 followed by a write to the same bank that
stalls because ACK is deasserted in cycle 2 in response to the write in
cycle 1. The second access is a 32-bit write to an odd address in external
memory. If the 32-bit access went to an even 32-bit address, WRL would
have asserted instead of WRH.
The zero-waitstate write mode provides the highest performance if the
slave has sufficient write buffer storage. Systems should use this mode
where the slave can always accept one write transfer (unless it has ACK deasserted) and can generally accept more than one write. If the slave has only
one store buffer, such that it always deasserts ACK after the first write, the
one-waitstate write mode may be the better choice. The zero-waitstate
write mode is targeted towards ASIC/FPGA designs, which can likely
implement multiple write buffers (including DSP as a slave), and fully
pipelined synchronous devices such as SBSRAMs.
that do not support the entire 64-bit data bus width do not
 Slaves
have to connect to both write strobes. Also, slaves that do not support bursting protocols do not need to connect to the BRST signal.
Bus master synchronous writes to external memory occur with the following sequence of events as shown in Figure 7-9 on page 7-28.
1. (cycle 1 in Figure 7-9) If ACK is sampled asserted at the start of cycle
1, the DSP bus master drives the write address and asserts a memory select signal (MS3-0) to indicate the selected bank. The DSP
asserts the WRH/WRL strobe(s) to indicate the size and alignment of
the requested data. The write strobes are not deasserted between
successive writes accesses of the same memory bank. If the size or
alignment changes, strobe assertion also changes. Strobe assertion
is determined by the size and alignment of the data transfer. For
more information on data alignment, see Figure 7-1 on page 7-2.
ADSP-21160 SHARC DSP Hardware Reference
7-29
External Memory Interface
2. (cycle 1) The previous slave three-states ACK. The keeper latch on
the DSP master keeps ACK at the asserted value until driven by the
next slave. Note that the slave could have driven ACK through
cycle 1. Only one slave is supported per bank, and any bank transition has an Idle cycle inserted to provide time for the slave to
three-state ACK.
3. (cycle 2) The DSP is initiating another write memory access to the
same bank. It drives the address, memory select, and strobes for the
next access.
4. (cycle 2) The slave, having decoded that it received a valid write
operation in the previous cycle, detects that it cannot accept further bus operations until the (or an element in the) write queue
becomes available, so it deasserts ACK.
5. (cycle 3) The DSP samples ACK deasserted by the slave. It inserts
waitstates until ACK is sampled asserted. The write ends in the cycle
in which ACK is sampled asserted by the slave (end of cycle 3).
Figure 7-10 shows a zero waitstate write, followed by a synchronous read
from the same bank. The slave addressed by both accesses determines in
cycle 2 that it has no more write capacity. It deasserts ACK in this cycle, in
response to the write in cycle 1. In cycle 3, the slave determines that it is
now addressed by the master to perform a read and asserts ACK to acknowledge the transfer. The slave asserts ACK in cycle 4 when read data is
available to complete the data transfer. The memory select for the read
access is held asserted by the master until cycle 4, because ACK was deasserted in cycle 2. In this example, both operations use the full data bus
width, as indicated by both WRH/L and RDH/L strobes asserted in for the
write and the read.
7-30
ADSP-21160 SHARC DSP Hardware Reference
External Port
1
2
3
4
5
CLKIN
write address
ADDRESS[31:0]
read address
MS[3:0]
RDH
RDL
WRH
WRL
BRST
write data
read data
DATA[63:0]
ACK
Figure 7-10. Synchronous Write Followed by Synchronous Read Example
Synchronous Write, One Waitstate Mode
Because some synchronous slaves cannot support a free-running latch
function to capture zero-wait bus writes, the DSP also supports a minimum two-cycle (minimum one-waitstate) write access. This mode is set
using the bank Access Mode bits (EBxAM). For more information on access
modes, see Table A-19 on page A-49.
The one-waitstate, synchronous write access is shown in the second write
of Figure 7-11.
ADSP-21160 SHARC DSP Hardware Reference
7-31
External Memory Interface
WRITE #1
1
IDLE
2
WRITE #2, DIFFERENT BANK
3
4
CLKIN
ADDRESS[31:0]
MS[3:0]
RDH
RDL
WRH
WRL
BRST
DATA[63:0]
ACK
Figure 7-11. Asynchronous Write Followed by Synchronous Write One-Waitstate Mode
In this example, the first access is to a bank configured for asynchronous
writes (cycle 1). In Figure 7-11, this condition is shown by the deassertion
of the write strobes before the rising edge of CLKIN for cycle 2. In cycle 2, a
bank transition occurs, and an idle cycle is inserted to allow the slaves to
transition ownership of ACK. In cycle 3, the second write begins, to a new
bank configured for one-waitstate write access. The address and data are
held for a minimum of two cycles.
7-32
ADSP-21160 SHARC DSP Hardware Reference
External Port
Similar to the synchronous read, MSx deassert in the next cycle (cycle 4),
and the waitstate counter decrements if ACK is sampled asserted. The access
can be held off the bus by deasserting ACK in cycle 2, or extended by deasserting ACK in cycle 3 (unlikely for a synchronous slave) or cycle 4.
Synchronous Burst Mode Interface Timing
Synchronous burst mode provides improved performance on synchronous
operations, read operations in particular. The DSP supports a DMA-mastered (only) burst mode. If the addressed slave supports this burst transfer,
after the one or more waitstates associated with access to the first 64-bit
read data transfer, contiguous data can transfer on each subsequent clock
cycle, up to a maximum of four 64-bit transfers. Burst accesses support
only 64-bit data transfers. Partial data bus width transfers are not
supported.
For burst transfers, the master drives the address of the first access on the
bus during the entire burst transfer. The master does not increment the
address for the slave. The maximum length of the burst transfer is four.
So, slaves only need a 2-bit address incrementer to generate the offset
address from the address driven by the master on the bus. Burst length
determination as a function of initial address is shown in Table 7-3.
Table 7-3. Linear Burst Address Order
First Address[2:1]
(external)
Second Address
(internal)
Third Address (internal) Fourth Address
(internal)
00
01
10
11
01
10
11
Burst Terminated1
10
11
Burst Terminated1
11
Burst Terminated2
1
2
Master always terminates burst when internal address[2:1] = 11
Master transfers this case as a single synchronous access
ADSP-21160 SHARC DSP Hardware Reference
7-33
External Memory Interface
If the DMA channel has sufficient data to transfer, it initiates a new burst
transfer starting at ADDR2-1=00, 01, or 10 when it wins bus arbitration.
Bursts always terminate when ADDR2-1=11.
An example of a synchronous burst read, of length three appears in
Figure 7-12. Here, the bank employed in the transfer has 2 waitstates.
1
2
3
4
5
CLKIN
ADDRESS[31:0]
ADDRESS[2:1] = 01
MS[3:0]
RDH
RDL
WRH
WRL
BRST
DATA[63:0]
ACK
Figure 7-12. External Memory Synchronous Burst Read Example
Burst Length Determination
The DMA arbitration logic amortizes the initial access latency by bursting
up to the maximum burst length of four when possible, assuming the
7-34
ADSP-21160 SHARC DSP Hardware Reference
External Port
channel is burst enabled. When a DMA channel wins internal I/O processor arbitration, the channel drives the internal buses as with a non-burst
transfer. At the same time, the I/O processor detects whether it can perform a burst transfer, according to the following criteria.
1. The DMAC burst enable (MAXBL1-0) control bit field is set for that
DMA channel. For more information on setting up a burst transfer, see the 64-bit External Burst Transfers discussion
on page 6-27.
2. The EI register points to a 64-bit aligned address,
3. The EM register is set to 0 or 1. A value of 0 does not increment EI.
This feature is useful when bursting to or from a registered data
port, buffer, or register, such as the EPBx FIFOs of another DSP.
4. The EC register is >= 4 (four 32-bit words equals two 64-bit
transfers).
5. The EPB FIFO for that channel has at least four 32-bit words to
transfer for an external burst write or has at least four empty 32-bit
elements to receive data for an external burst read.
6. The two least significant bits of the 64-bit DMA channel external
address are not set (ADDR2-1 does not equal 11).
Burst Stall Criteria
If I/O processor determines that it can perform a burst transfer (according
to the burst length criteria), the arbitration between the processor core
and I/O processor locks or parks the effective arbitration grant to that
DMA channel until:
1. The DMA channel external ADDR2-1 = 11. By disconnecting the
burst on this boundary, a modulo4 (ADDR31-1) is effectively implemented, which is required by SBSRAMs, and other slaves with
ADSP-21160 SHARC DSP Hardware Reference
7-35
External Memory Interface
limited address incrementing capability. For DSP-based systems,
slaves only need a 2-bit counter to support the address incrementing function of the burst.
2. Space in the EPB FIFO drops to less than four 32-bit elements (if a
external bus read), or less than four valid 32-bit elements for external bus writes. This almost full or empty detection is required by
the master logic to deassert BRST on the cycle before the end of the
burst.
3.
EC
goes to < 4; the burst pin must negate at EC=2.
4.
and SBTS are asserted on the external bus, indicating the deadlock resolution case in which the DSP must three-state its outputs
and switch into slave mode. For more information, see “Deadlock
Resolution” on page 7-84. Assertion of either signal alone does not
terminate the burst early. HBR assertion does not receive an HBG
until the burst finishes. SBTS assertion causes the master to
three-state outputs and insert waitstates.
HBR
If any of these conditions occur, normal arbitration between the processor
core and I/O processor for the external bus occurs. If the same bursting
channel wins arbitration again, a new burst is initiated, introducing at
least one lost or dead cycle in the burst throughput for reads.
7-36
ADSP-21160 SHARC DSP Hardware Reference
External Port
When arbitration occurs, the DMA channel loses arbitration if any of the
following conditions are detected:
1. Higher priority external request for the bus:
a.
HBR
asserted.
b.
BRx
asserted and BMAX time out has occurred.
c.
BRx
asserted and PA asserted, but not by this master.
2. Higher priority internal I/O processor requester:
a. Processor core request (DAGs or program sequencer)
b. A higher priority request from another DMA channel or
direct read/write access causes this channel to lose arbitration. For more information, see “I/O Processor”
Synchronous Burst Reads
External memory synchronous burst reads occur with the following
sequence of events as shown in Figure 7-12 on page 7-34:
1. (cycle 1 in) If ACK is sampled asserted at the beginning of cycle 1,
the DSP drives the read address and asserts a memory select signal
(MS3-0) to indicate the selected bank.
2. (cycle 1) The DSP asserts both RDH/RDL strobes to indicate a 64-bit
read request of the slave.
3. (cycle 2) As with the non-burst synchronous read, the DSP deasserts the MSx output signal, asserts the BRST output signal and
enables waitstate counting if ACK is sampled asserted at the end of
cycle 1.
4. (cycle 2) The DSP checks whether more than one waitstates (2
waitstates for this example) are needed. If so, BRST and the read
strobes remain active for additional cycle(s).
ADSP-21160 SHARC DSP Hardware Reference
7-37
External Memory Interface
5. (cycle 3) The slave samples BRST asserted, informing it that the
master requests at least one more 64-bit transfer after the current
transfer is ACK-ed by the slave.
6. (cycle 3) The programmed number of waitstates (for example, 2)
have been counted, and the slave is driving 64-bits of valid data
and asserting the ACK signal. This ends the first access.
7. (cycle 4) The slave drives the next 64-bits of contiguous data and
asserts ACK. If the slave needs more time to service any one transfer
within the burst, it can deassert ACK to stall the bus transfer.
8. (cycle 4) The slave samples BRST asserted, informing it that the
master requests at least one more 64-bit transfer.
9. (cycle 5) The master deasserts BRST to inform the slave that this is
the last transfer of the burst. In this example, the master deasserts
BRST due to the address modulo4 function. The two LSBs of the
initial 64-bit address = 01. The slave increments the address as
01->10->11, the maximum offset it needs to support from the initial address.
10.(cycle 5) The slave drives valid data for the last transfer, and asserts
ACK.
11.(cycle 6) If initiating another burst read memory access to the same
bank, the DSP asserts the address, memory select, and strobes for
the next access. This introduces at least two dead cycles in the
back-to-back burst throughput, because the initial waitstate count
applies to the first access of the second burst.
12.(cycle 6) With BRST sampled deasserted, the slave concludes its service of the burst request by three-stating the DATA63-0 and ACK
drivers.
As a master, the DSP supports burst reads on each of the four external
port DMA channels. Each channel has an independent burst enable
7-38
ADSP-21160 SHARC DSP Hardware Reference
External Port
control field (MAXBL1-0). For more information on setting up a burst
transfer, see the 64-bit External Burst Transfers discussion on page 6-27.
As a slave, the DSP supports read bursts from internal memory or the EPBx
buffers (with the EPBx read). For more information, see “Multiprocessor
(DSPs) Interface” on page 7-91 and “Host Processor Interface” on
page 7-49.
reads of the
FIFO are destructive, the DSP slave must
 Because
deassert
on each transfer of the burst to guarantee that it samEPBx
ACK
ples the deasserted BRST input before committing the EPBx FIFO
read. If the system design employs a similar destructive read data
buffer, similar precautions should be employed if burst reads of the
buffer are supported.
Synchronous Burst Writes
The DSP can master burst read and write operations in the one-waitstate
write access mode (EBxAM=10) if one or more DMA channels are configured appropriately. The DSP can master non-burst, zero-waitstate, writes
every cycle. Burst write transfers are not supported in this access mode.
Synchronous external devices which require at least one cycle of write
access latency (for example, bus bridges, SDRAM controllers, and others)
may be able to optimize throughput for burst write operations, based on
the contiguous, incrementing block transfer information conveyed by the
burst protocol. Burst accesses support only 64-bit data transfers. Partial
data bus width transfers are not supported.
An example of a synchronous burst write appears in Figure 7-13. Here,
the bank employed in the transfer has the 1 waitstate mode, for the first
write of the burst.
ADSP-21160 SHARC DSP Hardware Reference
7-39
External Memory Interface
1
2
3
4
5
CLKIN
ADDRESS[31:0]
ADDRESS[2:1]=00
MS[3:0]
RDH
RDL
WRH
WRL
BRST
DATA[63:0]
ACK
Figure 7-13. External Memory Synchronous Burst Write Example
7-40
ADSP-21160 SHARC DSP Hardware Reference
External Port
External memory synchronous burst writes occur with the following
sequence of events as shown in Figure 7-13 on page 7-40.
1. (cycle 1) If ACK is sampled asserted at the start of cycle 1, the DSP
drives the write address and asserts a memory select signal (MS3-0)
to indicate the selected bank. The DSP also drives valid data in this
cycle. The DSP asserts both WRH/WRL strobes to indicate a 64-bit
write command to the slave.
2. (cycle 2) The slave samples the write command and address. At this
point, the slave does not see that a burst write is in progress—the
access looks identical to a non-burst synchronous write. If the slave
cannot accept the write command, it deasserts ACK in this cycle to
stall the bus until it can. In this example, it has buffer capacity to
accept all of the data of the burst, so ACK stays asserted.
3. (cycle 2) If ACK was sampled asserted at the start of the cycle, the
DSP asserts the BRST output signal and deasserts the MSx output
signal.
4. (cycle 3) The DSP samples ACK asserted by the slave at the start of
the cycle, so it increments the data bus to the second of four data
transfers within the burst.
5. (cycle 3) The slave samples BRST asserted at the start of the cycle,
informing it that the master is writing at least one more 64-bit
transfer. The slave samples the second of four data transfers within
the burst and asserts ACK.
6. (cycle 4) The DSP samples ACK asserted by the slave at the start of
the cycle, so it increments the data bus to the third of four data
transfers within the burst.
ADSP-21160 SHARC DSP Hardware Reference
7-41
External Memory Interface
7. (cycle 4) The slave samples BRST asserted at the start of the cycle,
informing it that the master is writing at least one more 64-bit
transfer. The slave also samples the third of four data transfers
within the burst, and asserts ACK. If the slave needs more time to
service any one transfer within the burst, it can deassert ACK to stall
the bus transfer.
8. (cycle 5) The DSP samples ACK asserted by the slave at the start of
the cycle, so it increments the data bus to the last of four data
transfers within the burst. The master deasserts BRST to inform the
slave that this is the last transfer of the burst.
9. (cycle 5) The slave samples BRST asserted at the start of the cycle,
informing it that the master is writing at least one more 64-bit
transfer. The slave samples the fourth of four data transfers within
the burst and asserts ACK.
10.(cycle 6) If initiating another write burst memory access to the
same bank, the DSP asserts the address, memory select, and strobes
for the next access. This introduces at least one dead cycle in the
back-to-back burst throughput, because the initial waitstate count
applies to the first access of the second burst.
11.(cycle 6) With BRST sampled deasserted, the slave concludes its service of the burst request by three-stating the ACK driver.
As a master, the DSP supports burst writes on each of the four external
port DMA channels. Each channel has an independent burst enable control field (MAXBL1-0). For more information on setting up a burst transfer,
see the 64-bit External Burst Transfers discussion on page 6-27.
ADSP-21160 DSP does not support burst writes. The
 AsDSPa slave,
supports single cycle writes, so burst writes would provide no
added performance improvement.
7-42
ADSP-21160 SHARC DSP Hardware Reference
External Port
Using External SBSRAM
The DSP can connect to a variety of synchronous burst static RAMs
(SBSRAMs) with a glueless interface—no external logic required. These
synchronous memories can provide high throughput, especially when
employing the burst read transfer modes. The DSP has features to support
SBSRAMs from a number of memory vendors.
The DSP can support flow-through, pipelined and ZBT SBSRAMs.
Where bus frequency and system organization features like trace lengths,
capacitive loading, and termination characteristics allow, using
flow-through devices delivers lower latency and higher system
performance.
The DSP can support SBSRAMs on any of the four external memory
banks. The DSP supports SBSRAM single transfer reads and writes and
SBSRAM burst read transfer operations.
Single cycle burst write transfers are not supported.

SBSRAM support is enabled by configuring the bank access mode (
EBxAM)
bits for synchronous, 1-cycle writes and waitstate (EBxWS) bits for 1 waitstate (flow-through SBSRAMs) or 2 waitstates (fully pipelined SBRAMs).
For more information on programming access modes and waitstates, see
the WAIT register bits in Table A-19 on page A-49.
If burst read transfer capability is needed, one or more of the external port
DMA channels must be configured appropriately. For more information
on setting up a burst transfer, see the 64-bit External Burst Transfers discussion on on page 6-27. Because burst transfers are controlled at the
DMA channel, the DMA sequence must make sure that the DMA burst
transfer addresses a memory bank or slave that supports the read burst
transfer.
Figure 7-14 and Table 7-4 on page 7-45 show how the DSP I/O should
be connected to the SBSRAM I/O. Table 7-4 assumes a 512KByte
ADSP-21160 SHARC DSP Hardware Reference
7-43
External Memory Interface
SBSRAM array consisting of one bank of two 3.3V, 32K x 32 devices. The
names of the SBSRAM signals may vary from one vendor to another.
ADSP-21160
SBSRAM
32Kx32
ADDR[15:0]
ADDR[31:1]
MS0
CE1
BRST
RDH
ADSC
CE
BWE
BW[4:1]
OE
RDL
WRH
ADSP
GW
WRL
LBO
ADV
DATA[63:32]
DATA[63:0]
CE2
DATA[31:0]
CLKIN
CLKIN
ZZ
SBSRAM
32Kx32
ADDR[15:0]
CE1
ADSC
OE
GW
DATA[31:0]
DATA[31:0]
CLKIN
ADSP
CE
BWE
BW[4:1]
LBO
ADV
CE2
ZZ
Figure 7-14. SBSRAM System Interface Example
7-14 is for illustrative purposes—actual system designs may
 Figure
differ and must be carefully analyzed to determine the actual system topology.
7-44
ADSP-21160 SHARC DSP Hardware Reference
External Port
Table 7-4. ADSP-21160 to SBSRAM Signal Mapping
DSP
SBSRAM
Comment
CLKIN
CLK
Both devices driven by same input clock
ADDR16-1
ADDR15-0
Read/Write strobes decode bit 0 of address
MSx
CE
Chip Enable, active low
BRST
ADSC
Address Status Controller, active low
RDH
OE
Asynchronous Output Enable of SBSRAM #1, active low
RDL
OE
Asynchronous Output Enable of SBSRAM #2, active low
WRH
GW
Global Write Enable of SBSRAM #1, active low
WRL
GW
Global Write Enable of SBSRAM #2, active low
DATA63-32
DATA31-0
I/O of SBSRAM #1 (High word of bus, odd address)
DATA31-0
DATA31-0
I/O of SBSRAM #2 (Low word of bus, even address)
No connect
CE
Chip Enable, active high, Always Asserted (Vdd)
No connect
CE2
Second Chip Enable, Always Asserted (GND)
No connect
ADSP
Always Deasserted (Vdd)
No connect
ADV
Always Asserted (GND)
No connect
BWE
Byte Write Enable, Always Deasserted (Vdd)
No connect
BW4-1
Byte Write Selects, Always Deasserted (Vdd)
No connect
LBO
Linear Burst Order, active low, always asserted (GND)
No connect
ZZ
Sleep Mode Enable, active high, always deasserted (GND)
The SBSRAM devices are fully synchronous devices, except for the output
enable. The DSP issues commands and updates the SBSRAM address
latches, as a controller, using the ADSC input of the SBSRAMs, rather than
the ADSP processor input. Using the ADSC SBSRAM input enables single
cycle writes and simplifies SBSRAM deselect operations.
By always asserting the ADV (advance address) input to the SBSRAM, the
device is always attempting to burst. This input is a do not care when ADSC
is asserted. Because the BRST/ADSC signal is always low for a single access or
ADSP-21160 SHARC DSP Hardware Reference
7-45
External Memory Interface
the first access of a burst, the SBSRAM always updates its address latches
correctly. For the subsequent transfers (up to three, after the initial access)
of a read burst, the SBSRAM samples BRST/ADSC high. The asserted ADV
correctly advances the internal address count of the SBSRAM.
Figure 7-15 on page 7-46 demonstrates a burst read of the flow-through
SBSRAM.
1
2
3
4
5
6
7
8
9
10
CLKIN
ADDRESS[31:0]
A0
B1
MS0 (CE)
DESELECT
CYCLE
BRST (ADSC)
C1
DESELECT
CYCLE
IDLE
CYCLE
RDH/L (OE)
WRH/L (GW)
ACK
DATA[63:0]
A0
A1
A2
A3
B1
C1
Figure 7-15. SBSRAM – Burst Read, Single Write, Single Read
The DSP issues four types of bus operations to the SBSRAMs, as shown in
Table 7-5.
7-46
ADSP-21160 SHARC DSP Hardware Reference
External Port
Table 7-5. SBSRAM Partial Truth Table
SBSRAM Operation
ADV1
GW
WRx
OE
RDx
I/O
L
X
H
L
Data
L
L
X
L
H
Hi-Z
Read cycle, continue
burst
X
H
L
H
L
Data
Deselect Cycle
H
L
X
X
X
Hi-Z
CE1
MSx
ADSC
Read cycle, begin burst
L2
Write cycle, begin burst
BRST
All other signal inputs held static per Figure 7-7
1
2
ADV statically held asserted, low
L=low, H=High, X=don’t care, Hi-Z=three-stated, high impedance output
Single read or write transfers, and the first transfer of a burst read, employ
the read or write cycle, begin burst bus operation. Burst write transfers are
not supported. The subsequent transfers (up to three) of a read burst
employ the read cycle, continue burst bus operation. The last cycle of any
read access performs a deselect bus operation to make sure that the
SBSRAM data buffers remain three-stated for accesses to other banks.
The write operations are achieved by configuring the appropriate bank of
DSP to synchronous minimum one-cycle write mode. The synchronous
read waitstate count should be programmed to one for flow-through
SBSRAMs, or two for fully pipelined SBSRAMs.
DSP’s page detection function is not needed for SRAM mem The
ory systems.
SBSRAMs are not stalled, or suspended, by assertion of
in this
 configuration.
Systems should not deassert
during any
ACK
ACK
SBSRAM access. The DSP has a weak pull-up device on ACK; ACK
does not need to be driven during an access to a slave which does
not or cannot control ACK.
ADSP-21160 SHARC DSP Hardware Reference
7-47
External Memory Interface
The read is followed by a single write to the SBSRAM, which is followed
by a single read of the SBSRAM. For burst operations, the deasserting
BRST is not required in the last cycle of the burst transfer. The DSP’s burst
protocols also support ASIC/FPGA systems in which the pipelined
end-of-burst indicator may be of value.
It is possible to increase the SBSRAM array size from the example. This
increase can come from using higher density devices or implementing
multiple banks of SBSRAM. Multiple banks are possible using the depth
expansion capability of the SBSRAMs and the multiple memory select
outputs of the DSP.
Executing Instructions From External Memory
For systems that execute instructions from external memory, the system
must include a bank of 48-bit or 64-bit wide memory that is allocated specifically to program memory. This dedicated bank for instructions is
required because fetch addresses from the program sequencer are pointers
to 48-bit locations—the DSP does not translate the fetch address into a
32-bit address for external memory. The system can select the instruction
bank using one of the memory select (MS3-0) pins or the Core Instruction
Fetch (CIF) pin.
performance is reduced significantly when executing instruc DSP
tions directly from external memory.
For the instruction bank, the system can use CIF as a separate bank select.
CIF has the same timing as the MS3-0 outputs, but CIF asserts only for an
instruction fetch from external memory (depends on fetch from
sequencer, not address in memory). If the instruction fetch occurs to the
address range of one of the external banks, the DSP also asserts the memory select for that bank.
The ADSP-21160 DSP supports the Core Instruction Fetch ( )
 pin
for executing instructions from external memory. This pin is
CIF
not available on the previous SHARC DSPs.
7-48
ADSP-21160 SHARC DSP Hardware Reference
External Port
To fetch instructions from external memory, the system uses either of the
following methods:
Connect a dedicated bank of 48-bit wide memory to DATA63-16 pins and
use CIF or MS3-0 as the memory selects. The DSP uses the full address bus
(ADDR31-0 including LSB of the address) to address this bank. A bank of
64-bit wide memory also can be used this way.
• Connect a bank of 64-bit wide memory, store the 48-bit instructions MSB aligned in this bank, and use external address
translation to generate the appropriate address on an instruction
fetch. The system can use CIF as an indication of the fetch.
either of the above methods, the DSP asserts both
for
 Using
instruction fetch accesses.
address translation in the second method is required to accom The
modate the unpacked instruction locations. The Program
RDH/L
Sequencer issues sequential addresses for each fetch, but each
unpacked instruction word uses two (32-bit) memory locations.
The DSP only asserts CIF during instruction fetches from external memory. Other types of external memory accesses (such as 40-bit data accesses
or DMA transfers of packed instructions) do not use CIF. For more information on 40-bit data accesses in external memory, see “Internal Data Bus
Exchange” on page 5-7.
For more information on packed data transfers from external memory, see
the packed data discussion on page 6-27.
Host Processor Interface
The DSP’s host interface supports connecting the DSP to 16- or 32-bit
microprocessor buses. By providing an address, a data bus, and memory
control signals—such as read, write and chip select—a host may access any
ADSP-21160 SHARC DSP Hardware Reference
7-49
Host Processor Interface
device on the DSP bus as if it were a memory. Figure 7-16 shows an example of how to connect a host processor to the DSP.
HBG
REDY
HBR
SYSTEM BUS INTERFACE
ADSP-21160
BR1-BR6
SYSTEM
DATA BUS
ADDR31-0
DATA63-0
3
000
ID2-0
CS
HBR
WRH/L
HBG
RDH/L
OE
T/R
HBG
REDY
WRITE
ACK
MS3-0
READ
EXTERNAL
MEMORY
ADDR
SYSTEM
ADDRESS
BUS
DATA
WE
OE
HBR
CS
ACK
CS
ADDRESS
COMPARATOR
"ADDRESS
VALID"
REDY
DSP BUS
ACK
SYSTEM
BUS
Figure 7-16. Example DSP-to-Host System Interface
7-50
ADSP-21160 SHARC DSP Hardware Reference
External Port
The DSP accommodates either synchronous or asynchronous data transfers, allowing the host to use a different clock frequency. Transfers at
speeds up to the full CLKIN clock rate are supported.
7-16 shows all lines needed for an asynchronous host inter Figure
face. Systems using a synchronous interface do not need or
,
CS
REDY
and the address comparator might not be needed, depending on
the host processor’s requirements.
Table 7-6 defines the DSP pins used in host processor interfacing.
Table 7-6. Host Interface Signals
Signal
Type
Definition
HBR
I/A
Host Bus Request. The host processor must assert HBR to request control
of the DSP’s external bus. When HBR is asserted in a multiprocessing system, the DSP that is bus master relinquishes the bus and asserts HBG. To
relinquish the bus, the DSP places the address, data, select, and strobe
lines in a high-impedance state. HBR has priority over all DSP bus
requests (BR1-6) in a multiprocessing system.
HBG
I/O
Host Bus Grant. HBG acknowledges an HBR bus request, indicating that
the host processor may take control of the external bus. HBG is asserted
(held low) by the DSP until HBR is released. In a multiprocessing system,
HBG is output by the DSP bus master and is monitored by all others.
CS
I/A
Chip Select. The host processor asserts CS to select the DSP (for asynchronous transfer protocol).
O (o/d)
Host Bus Acknowledge. The DSP deasserts REDY (low) to add waitstates
to an asynchronous access of its internal memory or IOP registers by a
host. This pin is open-drain output (o/d) by default, but can be programmed with the ADREDY bit of SYSCON register to be active drive
(a/d). REDY is only output if the CS and HBR inputs are asserted.
I/S
Suspend Bus Three-state. External devices can assert SBTS (low) to place
the external bus address, data, selects, and strobes in a high-impedance
state for the following cycle. If the DSP attempts to access external memory while SBTS is asserted, the processor halts and the memory access
does not complete until SBTS is deasserted. SBTS should only be used to
recover from PAGE faults or host processor/DSP deadlock.
REDY
SBTS
I=Input, S=Synchronous, (o/d)=Open Drain, O=Output, A=Asynchronous, (a/d)=Active Drive
ADSP-21160 SHARC DSP Hardware Reference
7-51
Host Processor Interface
The host accesses the DSP through the DSP’s external port. Figure 6-8 on
page 6-68 shows a block diagram of the external port, I/O processor, and
FIFO data buffers, illustrating the on-chip data paths for host-driven
transfers. The four external port DMA channels are available for use by
the host—DMA transfers of code and data can be performed with low
software overhead.
ADSP-21160 DSP supports the host interface protocols of the
 The
previous SHARC DSPs. Also, the ADSP-21160 DSP provides new
synchronous interface protocols that support the 64-bit data bus
and burst transfers of sequential data.
The host processor requests and controls the DSP’s external bus with the
host bus request (HBR) and host bus grant (HBG) signals. Host logic does
not need to duplicate the distributed multiprocessor arbitration protocol
of the DSPs.
After the host gets control of the DSP bus, the host may transfer data
either synchronously or asynchronously. The host bus may be 16, 32, or
64 bits wide for synchronous transfers, but only 16 or 32 bits wide for
asynchronous transfers.
For asynchronous transfers, the host also uses the chip select ( CS) and
ready (REDY) signals. After getting control of the bus, the host can directly
read and write the internal memory of the DSP. The host can also read
and write to any of the DSP’s I/O processor registers, including the EPBx
FIFO buffers. The host uses certain I/O processor registers to control and
monitor the DSP (such as SYSCON and SYSTAT) and to set up DMA transfers. DMA transfers are controlled by the DSP’s I/O processor after they
are set up by the host. In a multiprocessor system, the host can access the
internal memory and I/O processor registers of every DSP.
Data written to and read from the DSP can be packed or unpacked into
different word widths. When the width of the host bus is 16 or 32 bits,
the DSP can pack data into 32 or 48-bit words. The DSP attempts to
gather two 32-bit words into one single 64-bit internal transfer where
7-52
ADSP-21160 SHARC DSP Hardware Reference
External Port
possible. When the width of the host bus is 64 bits (synchronous transfer
modes only), the DSP can pack 48-bit instructions so four instructions are
transferred in three 64-bit transfers (maximum throughput), or the DSP
handles unpacked data so only 48-bits of the 64-bit transfer are treated as
valid data. The host packing mode control bits (HPM) in the SYSCON register
configure data packing and unpacking.
Acquiring the Bus
For a host processor to gain access to the DSP, the host must first assert
HBR, the host bus request signal. HBR has priority over all BRx multiprocessor bus requests. When asserted, HBR causes the current DSP master to give
up the bus to the host after the DSP finishes the current bus operation. If
the current operation is a burst transfer, the change in bus mastership
interrupts the transfer on a modulo4 boundary.
The current DSP bus master signals that it is transferring ownership of the
bus by asserting HBG (low) when the current bus operation ends. The cycle
in which control of the bus is transferred to the host is called a Host Transition Cycle (HTC).
Figure 7-7 on page 7-20 shows the timing for the host acquiring the bus.
HBG is asserted while the bus master releases control of the bus and remains
asserted until HBR is sampled deasserted by the DSP. The cycles in which
control of the bus is released by the bus master is called the DSP’s Bus
Transition Cycle (BTC). HBG freezes DSP multiprocessor bus arbitration
during the time that the host owns the bus. HBG may be used to enable the
host’s signal buffers, as shown in Figure 7-16 on page 7-50, Figure 7-22
on page 7-81, and Figure 7-23 on page 7-83. While HBG is asserted in a
multiprocessor system, the DSPs continue to assert their BRx outputs, as in
normal operation, but no BTCs occur. The current DSP bus master keeps
its BRx output asserted throughout the entire time the host controls the
bus.
ADSP-21160 SHARC DSP Hardware Reference
7-53
Host Processor Interface
1
2
3
4
5
6
7
8
CLKIN
HBR
CS
REDY
HBG
BRx
ADDR[31:0]
A0
Host Address
D0
Host Data
A1
MSx
BRST
WRH
WRL
RDH, RDL
DATA[63:0]
D1
ACK
Figure 7-17. Example Timing for Host Acquisition of Bus
7-54
ADSP-21160 SHARC DSP Hardware Reference
External Port
After the host gets control of the bus, the host can choose to perform synchronous or asynchronous transfers with the DSP or other system
components. To initiate asynchronous transfers, the host asserts (low) the
CS and HBR inputs of the DSP that it intends to access and performs the
read or write. The DSP does not respond to CS until HBG is asserted. To
initiate synchronous transfers, the host keeps all DSP CS pins deasserted
(high) and reads or writes to the DSPs’ multiprocessor memory space.
The host may also communicate directly with system peripherals, such as
SBSRAMs. These transfers occur using the protocol of the peripheral or
using the external handshake mode of DMA channels 11 and 12 to control the memory or peripheral. With DMA handshaking, the host only
needs to source or sink the data with the correct timing. Either of these
solutions may require additional hardware support for the host.
The host is responsible for driving the following signals during the HTC
in which it gains control of the bus: ADDR31-0, RDH, RDL, WRH, WRL, DMAGx (if
employed in the system), and PAGE (if the PAGE function is employed in
the system application). These signals must be driven by the host while the
host is bus master. Also, the host must drive or weakly pull up or down the
MS3-0, BRST, CLKOUT, DMAG1, and DMAG2 signals as required. The DSP bus
master three-states these lines, letting the host use them.
The DSP with device ID=000 or 001 enables internal pull-up devices on
the MS3-0, CIF, RDL, RDH, WRL, WRH, DMAR1, DMAR2, DMAG1, and DMAG2 signals.
The pull-up provides a weak current source to hold these signals in the
deasserted state when driven to that state.
system noise can cause these weakly driven signals
 Excessive
(
,
,
,
,
,
,
,
,
, and
MS3-0 CIF RDL RDH WRL WRH DMAR1 DMAR2 DMAG1
DMAG2)
to
be sampled asserted.
The DSP with device ID=000 or 001 enables its keeper latches on
and DATA63-0, BRST, PAGE, and CLKOUT, so these signals are
weakly pulled to the last value driven on them if any of these signals
remain undriven for multiple cycles.
ADDR31-0
ADSP-21160 SHARC DSP Hardware Reference
7-55
Host Processor Interface
During read-modify-write operations, the host should keep HBR asserted to
avoid temporary loss of bus mastership. HBR must remain asserted until
after the host completes the last data transfer.
The following restrictions apply to bus acquisition by the host.
• If HBR is asserted while the DSP is in reset, the DSP does not
respond with HBG until after reset and multiprocessor synchronization is completed.
• The host should keep HBR asserted until after the host completes its
last data transfer and is ready to give up bus ownership.
• If SBTS is asserted after HBR, the DSP enters slave mode and suspends any unfinished access to the external bus.
• In uniprocessor systems (with ID2-0=000), the host must assert CS
in the same cycle as HBR to initiate an asynchronous access.
• If the host is to execute both synchronous and asynchronous
accesses during a single bus grant, it must allow at least one cycle to
pass after the last access before switching CS.
• Synchronous accesses may not be used in systems with only one
DSP (with ID2-0=000).
• Synchronous burst writes to DSPs are not supported. Synchronous
burst reads of DSP internal memory and the I/O processor’s EPBx
FIFOs are supported.
After the host finishes its task, it can relinquish control of the bus by deasserting HBR. The DSP bus master responds by deasserting HBG in the cycle
after sampling HBR deasserted. In the cycle following deassertion of HBG,
the DSP bus master assumes control of the bus and normal multiprocessor
arbitration resumes.
7-56
ADSP-21160 SHARC DSP Hardware Reference
External Port
Asynchronous Transfers
To initiate asynchronous transfers after acquiring control of the DSP’s
external bus, the host must assert the CS input of the DSP to be accessed.
The host then drives the address of the memory location or I/O processor
register to access. To simplify the hardware requirements for external
interface logic, only the address bits shown in Table 7-7 need to be driven.
Table 7-7. Address Fields For Asynchronous Host Accesses
Address Bits1
Comments
ADDR7-0
Must be driven in all cases
ADDR16-8
Must be driven only if the S field indicates an internal memory access
ADDR19-17
S field2 – Must be driven “000” for IOP register accesses, “01m” for internal
memory normal word accesses, or”1nn” for short word accesses
and either
ADDR22-20
M field2 – Must be driven “000” to deselect other DSPs, if present,
or
ADDR31-23
1
2
E field2 – One of the lines 31–23 driven as “1”
Setup and hold times for these address lines are specified in the DSP’s data sheet.
For a complete description of these address fields, see “ADSP-21160 DSP Memory Map” on
page 5-12.
Table 7-7 applies to all asynchronous host access cases, including multiprocessor systems. Fewer address bits may need to be driven depending on
the system. For example in a uniprocessor DSP system, the host need not
drive the ADDR22-20 address pins.
Host direct reads and writes are possible for Normal word or Short word
data. Host access to Long word data on DSP slaves is not supported.
Normal words are accessed if the ADDR19-17 address pins are 01m, where
m is the most significant bit of the Normal word address. Short words are
ADSP-21160 SHARC DSP Hardware Reference
7-57
Host Processor Interface
accessed if the ADDR19-17 address pins are 1nn, where nn are the two most
significant bits of the Short word address.
When using asynchronous transfers and direct access to internal memory
is not required, only the lower 8 bits, ADDR7-0, need be supplied by the
host. The upper address bits can be configured as explained in Table 7-7.
ADSP-21160 DSP does not support the Instruction Word
 The
Transfer ( ) function from previous SHARC DSPs. 48-bit
IWT
instructions can be transferred by configuring the host packing
mode to one of the 48-bit internal transfer modes. For direct write
operations, the correct size information is tracked through the
FIFO.
Asynchronous write operations are latched at the I/O pads in a four-deep
FIFO buffer; this buffer is called the slave write FIFO and appears in
Figure 6-8 on page 6-68. This buffering allows previously written words
to be re-synchronized while a new word is being written and allows asynchronous writes to occur at up to the full CLKIN clock rate of the DSP.
A host may write to several DSPs simultaneously (a broadcast write), by
asserting each of their CS pins. Each DSP accepts the write as if it were the
only device being addressed. Because the REDY output is wire-OR’ed (if
configured as an open-drain output), REDY only appears asserted when all
selected DSPs are ready, unless REDY is actively pulled up. ACK is not active
when CS is asserted.
To eliminate the need for a host to drive the multiprocessor address lines
(ADDR22-20) in systems with only one DSP (ID2-0=000), such DSP does
not recognize synchronous accesses to these addresses. The host must
drive these address lines if the DSP’s ID2-0 is anything other than 000. To
account for buffer delays when sampling the REDY signal, systems must
make sure that REDY is properly re-synchronized by the host.
7-58
ADSP-21160 SHARC DSP Hardware Reference
External Port
Asynchronous Transfer Timing
When a DSP’s CS chip select is asserted (low), the selected DSP deasserts
the REDY signal. Refer to ADSP-21160 DSP Microcomputer Data Sheet for
exact timing specifications.
As shown in Figure 7-18, the DSP deasserts REDY in response to CS. The
host can assert CS before or after HBR is asserted, but the DSP does not
reassert REDY until after HBG is asserted and a RDH/L or WRH/L strobe is
applied. This condition is true only if a RDH/L or WRH/L strobe is active
when HBG is asserted. Otherwise, this timing is determined by the
tTRDYHG switching characteristic specified in the “Multiprocessor Bus
Request and Host Bus Request” timing data in the ADSP-21160 DSP
Microcomputer Data Sheet.
is asserted prior to the RDH/L or WRH/L being asserted and becomes
deasserted only if the DSP is not ready for the read or write to complete—
the only exception is when CS is first asserted. The REDY pin is an
open-drain output to facilitate interfacing to common buses. It can be
changed to an active-drive output by setting the ADREDY bit in the SYSCON
register.
REDY
ADSP-21160 DSP’s asynchronous transfer timing is similar to
 The
that of previous SHARC DSPs with one important difference. The
DSP has two read strobes (RDH and RDL) and two write strobes (WRH
and WRL). Each of these strobes enables or masks 32-bits of the
64-bit data bus. Only RDH and WRH are employed on asynchronous
transfers using the host packing mode support for 16/32-bit transfers. See “External Port Buffer Modes” on page 6-17.
ADSP-21160 SHARC DSP Hardware Reference
7-59
Host Processor Interface
HBR
CS
DRIVEN
BY HOST
HO ST
ADDRESS
VALID
HOST
WRITE
ADDRESS VALID
ADDRESS SETUP
HOST
READ
HOST
BUFFERS
TURN ON
HBG
RDH/L
WRH/ L
MSX
DRI VEN BY
DSP BUS
MASTER
DRIVEN
INACTIVE
BEFORE
THREE-STATE
BRX
ACK
DRIVEN BY
EACH DSP
REDY
DATA SETUP
REDY DEASSERTED
FOR A MIN OF
1 CYCLE
VALI D
DATA
VA LID D A TA
FR OM DS P
HOST TRI STATES
BEFORE
ASSERTING RD
DATA IS LATCHED IN
HO ST
DATA FROM HOST I S LATCHED INTO DSP ON WR RISING EDG E
O N RD RISING EDG E
HOST
BUS
TRANSITIO N T RANSITI ON
CYCLE (BTC) CYCLE (HTC)
Figure 7-18. Example Timing For Host Read and Write Cycles
7-60
ADSP-21160 SHARC DSP Hardware Reference
External Port
Figure 7-18 shows the timing of a host write cycle, including details of
data packing and unpacking. This timing applies to the example host
interface hardware shown in Figure 7-23 on page 7-83 and has the following sequence.
1. The host asserts the address. HBR and CS are decoded from the host
bus interface address comparator and do not need to be supplied
directly by the host. The selected DSP deasserts REDY immediately.
2. The host asserts WRH/L and drives data according to the timing
requirements specified in the ADSP-21160 DSP Microcomputer
Data Sheet.
3. The selected DSP asserts REDY when it is ready to accept the data.
This transition occurs after the current bus master has completed
its current transfer and has asserted HBG. HBG enables the host interface buffers to drive onto the DSP bus.
4. The host deasserts WRH/L when REDY is high and stops driving data.
5. The selected DSP latches data on the rising edge of WRH/L.
After the first word, the write sequence is:
1. The host asserts WRH/L and drives data according to the timing
requirements specified in the ADSP-21160 DSP Microcomputer
Data Sheet.
2. The DSP deasserts REDY if it is not ready to accept data.
3. The host deasserts WRH/L when REDY is high and stops driving data.
4. The selected DSP latches data on the rising edge of WRH/L.
More than one DSP may have its CS pin asserted at any one time during a
write, but not during a read because of bus conflicts.
ADSP-21160 SHARC DSP Hardware Reference
7-61
Host Processor Interface
Figure 7-18 on page 7-60 also shows the timing of a host read cycle. This
timing applies to the bus interface hardware in Figure 7-23 on page 7-83
and has the following sequence:
1. The host asserts the address. HBR and the appropriate CS line are
decoded by the host bus interface address comparator. The selected
DSP deasserts REDY immediately and asserts HBG.
2. The host asserts RDH/L.
3. The selected DSP drives data onto the bus and asserts REDY when
the data is available.
4. The host latches the data and deasserts RDH/L.
After the first word, the read sequence is:
1. The host asserts RDH/L.
2. The selected DSP deasserts REDY then asserts REDY, driving data
when it becomes available.
3. The host deasserts RDH/L when REDY is high and latches the data.
Synchronous Transfers
Synchronous transfers are defined by both master and slave deriving bus
timing (sampling bus inputs and driving bus outputs) from the same clock
input (same clock frequency and phase). Synchronous transfers potentially
offer significantly higher throughput than asynchronous transfers, but
may require additional synchronization logic.
To perform synchronous transfers, the CS input is not asserted and the
host must act like another DSP in a multiprocessor system. The host must
generate an address in the multiprocessor memory space of a DSP, assert
the RDH/L or WRH/L strobes (and BRST if a burst read transfer) and sink or
source data.
7-62
ADSP-21160 SHARC DSP Hardware Reference
External Port
For examples of synchronous transfers, see “External Memory Interface”
on page 7-3 and “Multiprocessor (DSPs) Interface” on page 7-91.
Synchronous accesses are not supported in uniprocessor systems in which
the ID2-0=000. Synchronous accesses are possible in uniprocessor systems
in which the ID2-0=001 or in multiprocessor systems. To perform synchronous accesses in a multiprocessor system, the host must drive the
address pins ADDR22-20 with a value of 1-7 to select one of the DSPs (by
its ID2-0) or one of the ADDR31-23 address pins must be driven high to
select an address in external memory. For more information on using
these address pins, see Table 7-7 on page 7-57.
For synchronous host transfers, the DSP uses its ACK signal—instead of
REDY—to add waitstates to an access. The host must wait for the DSP to
assert ACK. Synchronous accesses are not recognized during the Host Transition Cycle. This prevents any spurious access from occurring while the
external host buffers are starting to drive the address, data, and strobes.
may glitch during the HTC and should not be relied on until
 the following
cycle.
ACK
When a DSP is responding to a synchronous read access, it drives valid
data only in the cycle in which it asserts ACK. The DSP three-states the
data bus in the cycles following assertion of ACK in response to a synchronous host read, even if the host continues to assert RDH/RDL.
Synchronous Broadcast Writes
The timing in Figure 7-19 on page 7-64 demonstrates two synchronous
broadcast writes from the host. Broadcast writes address multiple slaves.
Master and slaves employ a different protocol for ACK-acknowledgment of
the broadcast write, relative to other bus operations. For broadcast writes,
slaves employ a wired-OR protocol in which they drive ACK deasserted
(only if required) in alternate cycles, starting with the cycle after the write
is driven on the bus. The master pre-charges ACK high in alternate cycles,
starting with the second cycle after the broadcast write.
ADSP-21160 SHARC DSP Hardware Reference
7-63
Host Processor Interface
1
2
BTC
HTC
3
4
5
6
7
8
9
CLKIN
BTC
HBR
HBG
BRx
ADDR31:0
BCast Write 1
BCast Write 2
Write 1
BCast Write 2
MSx
BRST
WRH/L
RDH/L
ACK
ACK (master out)
ACK (slaves)
DATA63:0
Figure 7-19. Synchronous Broadcast Write Example
In this example, the first broadcast write is accepted by all slaves at the end
of cycle 3. One or more slaves must stall any further access until one has
capacity to accept the next write operation. This stall is accomplished by
the slaves deasserting ACK in cycle 4. The master must pre-charge ACK in
cycle 5, the second cycle after the first broadcast write. All slaves can
accept more operations by cycle 6, and none of the slaves drive ACK deasserted again.
7-64
ADSP-21160 SHARC DSP Hardware Reference
External Port
The host samples ACK asserted in cycle 6 and completes the second broadcast write as shown by deasserting the write strobes in cycle 7.
Even though the host deasserted
in cycle 7, the DSP does not
 respond
by deasserting
until cycle 9.
HBR
HBG
Synchronous Burst Read Transfers
As a slave, the DSP supports synchronous burst read transfers from the
EPBx FIFOs or direct reads from internal memory. Burst write transfers are
not supported, because single-cycle, non-burst writes can provide as much
write bandwidth as burst writes. Burst reads are supported as contiguous,
aligned, 64-bit data transfers up to a maximum length of four 64-bit
transfers. The DSP slave increments the address if the burst read access is
from internal memory. The slave address increment function is only supported for ADDR2-1. The host cannot burst across a modulo4 (ADDR2-0)
address boundary as shown in Table 7-3 on page 7-33.
To perform a burst read transfer from an EPBx buffer, the host issues a
starting burst address pointing to one of the EPBx buffers. The DSP slave
does not increment an EPBx burst read address. The modulo4 (ADDR2-0)
address boundary restriction does not apply in this case.
burst transfers tie up the external bus and may prevent or sig Long
nificantly degrade system response to potentially higher priority
events. The DSP has no way of truncating or disconnecting a host
access.
 Ifring,
and HBR are asserted while an external DMA access is occurHBG is not asserted until the access is completed.
SBTS
The DSP also supports burst transfers, which can be truncated by assertion of HBR and SBTS. If the DMA transaction was a burst transfer, when
the host relinquishes control of the local bus, the DSP resumes the burst
transfer, starting at the address of the last operation that did not complete.
ADSP-21160 SHARC DSP Hardware Reference
7-65
Host Processor Interface
Slave Direct Reads and Writes
The host can directly access the internal memory and I/O processor registers of a DSP by reading or writing the appropriate address in
multiprocessor memory —known as a direct read or direct write access.
Each DSP bus slave monitors addresses driven on the external bus and
responds to any that fall within its region of multiprocessor memory.
These accesses are invisible to the slave DSP’s processor core. They do not
degrade internal memory or internal bus performance. Direct access is
important, because it lets the processor core continue program execution
uninterrupted.
The host can directly read or write the I/O processor registers to control
and configure the DSP or to set up DMA transfers.
IOP Shadow Registers
To ease host and multiprocessor system operations, the DSP I/O processor registers include registers that shadow or mirror some processor core
system registers, including the program counter (PC), and MODE2 registers.
These registers facilitate system start up and debug, by letting the host (or
another DSP in an multiprocessor system) interrogate these processor core
registers. These shadow registers are read only and lag the value of the registers they shadow by one internal core clock. For more information, see
“PC Shadow Register (PC_SHDW)” on page A-53 and “MODE2
Shadow Register (MODE2_SHDW)” on page A-53.
silicon revision field of the
shadow register
 The
is now used for differentiating between silicon revisions. These corMODE2
MODE2_SHDW
responding bits in the MODE2 (foreground) register are now
reserved. The application program must read the MODE2_SHDW register bits [31:25] to identify the silicon revision. MODE2_SHDW is a
memory-mapped IOP register whose address is 0x11. See
Figure 7-20 on page 7-67.
7-66
ADSP-21160 SHARC DSP Hardware Reference
External Port
MODE2_SHDW
0x11
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
0
1
0
0
0
0
1
0
0
0
0
0
0
0
0
0
PID 2-0**
Processor Identification
(Read Only - 21160 M)
PID4-3
Processor Identification
(Read Only)
Silicon Revision*
(Read Only - 21160M)
Revision 0.0 = 00
Revision 0.1 = 01
Revision 1.0/1.1 = 10
Revision 1.2 = 11
15 14 13 12 11 10 9
8
7
6
0
0
0
0
0
0
0
0
0
0
5
0
4
3
2
0
0 0
1
0
0
0
*These revisions apply to 21160M silicon only. For 21160N silicon, revision bits 29:28 are 00.
**This processor ID applies to 21160M silicon only. For 21160N silicon, the processor ID bits 27:25 are 100.
Figure 7-20. Mode2 Shadow Register
Instruction Transfers
For 16- or 32-bit host interfaces, the DSP can pack and unpack 48-bit
instruction or 40-bit Extended Precision Normal word data based on the
host packing mode selected with the HPM bits in the SYSCON register. Host
processors can achieve maximum throughput by transferring packed
instructions to or from internal memory, using the full 64-bit data bus
width synchronously. For more information, see the Packed versus
Unpacked instruction discussion on on page 6-27.
In packed 64-bit direct reads or writes of 48-bit memory, the program
must translate the addresses from an instruction word (Iword) pointer to a
Normal word (Nword) pointer as follows.
1.
Nword address= [(Iword address - Block base) * 3/2] + Block
base
ADSP-21160 SHARC DSP Hardware Reference
7-67
Host Processor Interface
2. If the host transfers a group of instructions to the DSP, the instructions go into DSP memory starting at an internal memory address
(for example, 0x40100). In internal memory, the instructions are
packed instructions—48-bit instructions 48-bit aligned—not padded to align to 64-bit boundaries.
3. Programs can find the Normal word starting address of the host
transfer to the DSP by scaling the least significant 16-bits of the
internal 48-bit instruction address by 3/2. For example, the Normal word address that corresponds to instruction address 0x40100
is:
Nword address= [(0x40100 - 0x40000) * 3/2] + 0x40000
= 0x40180
The ADSP-21160 DSP does not support the
 SHARC
DSPs.
IWT
bit from previous
Host Direct Writes and Reads
The DSP supports synchronous and asynchronous direct read and writes
by the host.
 Synchronous burst writes to ADSP-21160 slaves are not supported.
Direct Writes
When a direct write to a slave DSP occurs, the address, data, and control
are latched into the direct write FIFO. The FIFO supports up to eight,
64-bit wide direct writes, which can be performed with no stalls. If additional direct writes are attempted when the FIFO buffer is full, the DSP
deasserts ACK (or REDY) until the buffer is no longer full. The direct write
FIFO may be held off for up to four processor core cycles if all of the serial
port DMA channels are active, or up to nine core cycles per chain if DMA
chaining is occurring.
7-68
ADSP-21160 SHARC DSP Hardware Reference
External Port
Direct Write Latency
The DSP handles asynchronous and synchronous direct writes differently.
This difference influences the latency for the direct writes.
When a DSP bus slave receives data from an asynchronous direct write,
the DSP latches the data and address in a four-level FIFO buffer. For synchronous direct writes, this buffer is two-levels deep. This buffer is called
the slave write FIFO and appears in Figure 6-8 on page 6-68. In the following cycle, the slave write FIFO attempts to complete the write
internally. This buffering lets the host (or DSP bus master) perform writes
at the full clock rate.
slave DSP’s core cannot explicitly read the slave write FIFO.
 The
Also, the DSP cannot determine the slave write FIFO’s status.
Writes to the I/O processor registers from the slave write FIFO usually
occur in the following one or two cycles or when any current DMA transfer is completed. The write takes more than two cycles only if a direct
write in the previous cycle was held off by a full buffer.
If the slave write FIFO is full when a write is attempted, the DSP deasserts
ACK (or REDY) until the FIFO is not full. Unless higher priority on-chip
DMA transfers are occurring, the slave write FIFO usually empties out
within one cycle, creating a one-cycle write latency.
Slave reads are held off when there is data in the slave write FIFO—this
prevents false data reads and out-of-sequence operations.
The Direct Write Pending (DWPD) bit of the SYSTAT register indicates when
a direct write to internal memory is pending in the I/O processor’s direct
write FIFO or data is pending in the slave write FIFO (at the external port
I/O pins). Direct writes and I/O processor register accesses may be completed in different sequences. If the host performs a direct memory write
then writes to an I/O processor register on the DSP, the I/O processor
register write may complete before the direct write.
ADSP-21160 SHARC DSP Hardware Reference
7-69
Host Processor Interface
Direct Reads
When a direct read of a DSP occurs, the address is latched on-chip by the
I/O processor. If the access is asynchronous (CS asserted), REDY is deasserted asynchronously. If the access is synchronous (CS deasserted), ACK is
deasserted in the following CLKIN cycle. When the data is available, the
I/O processor drives the data and asserts REDY (or ACK). Direct reads cannot be pipelined like direct writes—direct reads only occur one at a time.
Direct reads have a maximum throughput of one access per every three
CLKIN cycles for synchronous I/O processor register reads or have one
access every four CLKIN cycles for synchronous internal memory reads. As a
slave, the DSP supports synchronous burst direct read accesses, which
improve throughput for internal memory reads. Maximum throughput for
synchronous burst direct read accesses is summarized in Table 7-8.
Table 7-8. Direct Read Latencies—for a 1:2 Clock Ratio
Access Type
Latency (CLKIN cycles)
Single Direct Read of I/O processor register
3
Single Direct Read of internal memory
4
Burst Direct Read of I/O processor registers (EPBx only)
3-2-2-2
Burst Direct Read of Internal memory
4-2-2-2
with the throughput advantage of burst read transfers, direct
 Even
reads are not the most efficient method of transferring data out of a
slave DSP. The highest throughput is achieved by forcing the slave
to become the master and having it write (or push) the data out.
The advantage of direct reads is that no programming of the slave
I/O processor is required.
7-70
ADSP-21160 SHARC DSP Hardware Reference
External Port
Broadcast Writes
Broadcast writes allow simultaneous transmission of data to all of the
DSPs in a multiprocessing system. The host processor (or master DSP)
can perform broadcast writes to the same memory location or I/O processor register on all of the slaves. Broadcast writes can be used to implement
reflective semaphores in a multiprocessing system. For more information,
see “Broadcast Writes” on page 7-115.
Shadow Write FIFO
Because the DSP’s internal memory must operate at high speeds, writes to
the memory do not go directly into the internal memory. Instead, writes
go to a two-deep FIFO called the shadow write FIFO.
When an internal memory write cycle occurs, the DSP loads the data in
the FIFO from the previous write into memory, and the new data goes
into the FIFO. This operation is transparent, because any reads of the last
two locations written are intercepted and routed to the FIFO.
the ADSP-21160 DSP’s shadow write FIFO automatically
 Because
pushes the write to internal memory as soon as the write does not
compete with a read, this FIFO’s operation is completely transparent to programs. Unlike previous SHARC DSPs, there is no need
for dummy writes to clear the FIFO when mixing 32- and 48-bit
writes.
Data Transfers Through the EPBx Buffers
In addition to direct reads and writes, the host processor can transfer data
to and from the DSP through the external port FIFO buffers, EPB0, EPB1,
EPB2, and EPB3. Each of these buffers, which are part of the I/O processor
register set, is an eight-location FIFO. These buffers support single-word
transfers, DMA transfers, and sequential burst accesses.
ADSP-21160 SHARC DSP Hardware Reference
7-71
Host Processor Interface
DMA transfers are handled internally by the DSP’s I/O processor, but single-word transfers must be handled by the processor core.
For information on external port transfers, see “External Port Channel
Transfer Modes” on page 6-21. For information on external port handshaking, see “External Port Channel Handshake Modes” on page 6-22.
For information on single-word interrupt-driven transfers through the
external port, see “Link Port Status” on page 6-60.
support debugging buffer transfers, the DSP has a Buffer Hang
 ToDisable
( ) bit. When set (=1), this bit prevents the processor
BHD
core from detecting a buffer-related stall condition, permitting
debugging of this type of stall condition. For more information, see
the BHD discussion on page on page 6-18.
DMA Transfers
The host processor can also set up DMA transfers to and from the DSP.
After the host gets control of the DSP, the host can access the on-chip
DMA control and parameter registers to set up an external port DMA
operation. DMA is the most efficient way to transfer blocks of data.
• DMA Transfers to Internal Memory. The host can set up external
port DMA channels to transfer data to and from DSP internal
memory.
• DMA Transfers to External Memory. The host can set up an external port DMA channel to transfer data directly to external memory
using the DMA request and grant lines (DMARx, DMAGx).
For more information, see “Setting up External Port DMA” on page 6-74.
The host may also use the
handshake signals for a
 DMA
transfer as a bus slave, but may not use DMA as a bus master
DMARx/DMAGx
while HBR retains control of the bus.
7-72
ADSP-21160 SHARC DSP Hardware Reference
External Port
Host Data Packing
The host interface uses the same data packing features as the I/O processor
uses. “External Port Buffer Modes” on page 6-17. The “32-bit Data Packing” on page 7-73 and “48-Bit Instruction Packing” on page 7-76 sections
describe timing for these data packing operations.
For direct reads and writes of DSP internal memory, the
 determine
the packing mode. For transfers to or from the
HPM
bits
data
EPBx
buffers, the packing mode is determined by the setting of the PMODE
bits in the DMACx control register of each external port buffer.
32-bit Data Packing
For a 16-bit host bus, the DSP latches incoming data on pins DATA47-32.
Similarly, the DSP drives outgoing data on DATA47-32 with the other lines
equal to zeroes. The sequence of events for 32-bit packing and unpacking
is different for writes and reads.
a 16-bit host bus, the endian format of the transfers is con For
trolled by the
bit in the
register. If
=0, the least
HMSWF
SYSCON
HMSWF
significant 16-bit word is packed first. If HMSWF=1, the most significant 16-bit word is packed first.
Figure 7-21 on page 7-74 shows example timing for host interface data
packing.
ADSP-21160 SHARC DSP Hardware Reference
7-73
Host Processor Interface
16/32 BIT PACKING
HOST
ADDRESS(15:1)
(SAME) WRITE
ADDRESS
WRITE
ADDRESS
READ
ADDRESS
(SAME) READ
ADDRESS
WR
RD
REDY
WRITE 1ST
WORD
INTO DSP
DATA47-31
WRITE 2ND
WORD
INTO DSP
READ 1ST
WORD
FROM DSP
READ 2ND
WORD
FROM DSP
VALID
VALID
VALID
VALID
HOST READ WITH
16/32 BIT PACKING
HOST WRITE WITH
16/32 BIT PACKING
32/48 BIT PACKING
ADDR31-0
WORD1
WORD1
WRITE ADDR WRITE ADDR
WORD2
WRITE
ADDR
WORD1
READ ADDR
WORD2
READ
ADDR
WORD2
READ ADDR
WR
RD
REDY
WRITE 2ND
WRITE 1ST
LONG WORD LONG WORD
INTO DSP
INTO DSP
DATA63-32
1ST
WORD
2ND
WORD
HOST WRITE WITH
32/48 BIT PACKING
3RD
WORD
READ 1ST
LONG WORD
INTO DSP
WORD1
READ 2ND
LONG
WORD
INTO DSP
WORD2
WORD3
HOST READ WITH
32/48 BIT PACKING
Figure 7-21. Example Timing for Host Interface Data Packing
7-74
ADSP-21160 SHARC DSP Hardware Reference
External Port
When a host reads a 32-bit word with 16-bit unpacking using the typical
bus interface hardware shown in Figure 7-23 on page 7-83, the following
sequence of events occurs as illustrated in Figure 7-21 on page 7-74.
• The host initiates a read cycle by driving an address, asserting CS if
the access is asynchronous, and asserting RDH (low).
• The selected DSP deasserts REDY, latches the address, and performs
an internal direct read to get the data.
• When the DSP has the data, it asserts REDY and drives the first
16-bit word.
• The host latches the data and deasserts RDH (high).
• The host initiates another read access, driving the address of the
data to be accessed then asserting RDH.
• The DSP transmits the second 16-bit word.
When a host writes a 32-bit word with 16-bit packing using the typical
bus interface hardware shown in Figure 7-23 on page 7-83, the following
sequence of events occurs as illustrated in Figure 7-21 on page 7-74.
• The host initiates a write cycle by driving the write address, asserting CS if the access is asynchronous, and asserting WRH (low).
• The DSP asserts REDY when it is ready to accept data.
• The host drives the address and the first 16-bit word and deasserts
WRH (high).
• The DSP latches the first 16-bit word.
• The host drives the address and initiates another write cycle for the
second 16-bit word by asserting WRH.
ADSP-21160 SHARC DSP Hardware Reference
7-75
Host Processor Interface
• When the DSP has accepted the second word, it performs an internal direct write to its memory (or memory-mapped I/O processor
register). If the DSP’s internal write has not completed by the time
another host access occurs, the DSP holds off that access with REDY.
If the DSP is waiting for another 16-bit word from the host to complete
the packed word, the HPS field in the SYSTAT register is non-zero. For more
information, see “Host Interface Status” on page 7-77.
there is only one packing buffer for the host interface, the
 Because
host must complete each packed transfer (or end the transfer and
purge the FIFO with the HPFLSH bit) before another is begun. For
more information, see “External Port Status” on page 6-57.
48-Bit Instruction Packing
The host can also download and upload 48-bit instructions over its 16- or
32-bit bus. The packing sequence for downloading DSP instructions from
a 32-bit host bus (HPM=11) takes 3 cycles for every 2 words, as illustrated
in Table 7-9. The 32-bit data is transferred on data bus lines 63-32
(DATA63-32). If an odd number of instruction words are transferred, the
packing buffer must be flushed by a dummy access to remove the unused
word.
Table 7-9. 32-to-48-Bit Word Packing (Host Bus  DSP)
Data Bus Lines 63-48
Data Bus Lines 47-32
First transfer
Word1, bits 47-32
Word1, bits 31-16
Second transfer
Word2, bits 15-0
Word1, bits 15-0
Third transfer
Word2, bits 47-32
Word2, bits 31-16
 The
7-76
HMSWF
bit of SYSCON is ignored for 32-to-48-bit packing.
ADSP-21160 SHARC DSP Hardware Reference
External Port
The packing sequence for downloading or uploading DSP instructions
over a 16-bit host bus takes 3 cycles for every word, as shown in
Table 7-10.
Table 7-10. 16-to-48-Bit Word Packing, HMSWF=1 (Host Bus  DSP)
Data Bus Pins 47-32
First transfer
Word1, bits 47-32
Second transfer
Word1, bits 31-16
Third transfer
Word1, bits 15-0
The
bit in
determines whether the most significant
 16-bit
word or least significant 16-bit word is packed first.
HMSWF
SYSCON
40-bit extended precision data may be transferred using the 48-bit packing mode. For more information on memory allocation for different word
widths, see “Memory Organization and Word Size” on page 5-22.
Host Interface Status
The SYSTAT register provides status information for host and multiprocessor systems. Table 7-14 on page 7-122 lists the status bits in the SYSTAT
register. For more information on the SYSTAT register, see Table A-20 on
page A-51.
Interprocessor Messages and Vector Interrupts
After getting control of the DSP, the host processor communicates with
the DSP by writing messages to the memory-mapped I/O processor registers. In a multiprocessor system, the host can access the internal memory
and I/O processor registers of every DSP.
The MSGRx registers are general-purpose registers that can be used for message passing between the host and DSP. They are also useful for
ADSP-21160 SHARC DSP Hardware Reference
7-77
Host Processor Interface
semaphores and resource sharing between multiple DSPs. The MSGRx and
VIRPT registers can be used for message passing in the following ways.
• Message Passing. The host can use any of the 8 message registers,
MSGR0 through MSGR7, to communicate with the DSP.
• Vector Interrupts. The host can issue a vector interrupt to the
DSP by writing the address of an interrupt service routine to the
VIRPT register. When serviced, this high priority interrupt causes
the DSP to branch to the service routine at that address.
The MSGRx and VIRPT registers also support shared-bus multiprocessing
through the external port. Because these registers may be shared resources
within a single DSP, conflicts may occur—your system software must prevent this. For further discussion of I/O processor register access conflicts,
see “I/O Processor Registers” on page A-33.
Message Passing (MSGRx)
Three possible software protocols that the host can use for communicating
with the DSP through the MSGRx message registers are: (1) vector-interrupt-driven, (2) register handshake, and (3) register write-back.
For the vector-interrupt-driven method, the host fills predetermined
MSGRx registers with data and triggers a vector interrupt by writing the
address of the service routine to VIRPT. The service routine should read the
data from the MSGRx registers and then write “0” into VIRPT to tell the host
it is done. The service routine also could use one of the DSP’s FLAG3-0
pins to tell the host it has finished.
For the register handshake method, four of the MSGRx registers should be
designated as follows: a receive register (R), a receive handshake register
(RH), a transmit register (T), and a transmit handshake register (TH). To
pass data to the DSP, the host would write data into T and then write a
“1” into TH. When the DSP sees a “1” in TH, it reads the data from T and
then writes back a “0” into TH. When the host sees a “0” in TH, it knows
7-78
ADSP-21160 SHARC DSP Hardware Reference
External Port
that the transfer is complete. A similar sequence of events occurs when the
DSP passes data to the host through R and RH.
The register write-back method is similar to register handshaking, but uses
only the T and R data registers. The host writes data to T. When the DSP
sees a non-zero value in T, it retrieves it and writes back a “0” to T. A similar sequence occurs when the host is receiving data. This simpler method
works well when the data to be passed does not include “0.”
Host Vector Interrupts (VIRPT)
Vector interrupts are used for interprocessor commands between the host
and a DSP or between two DSPs. When the external processor writes an
address to the DSP’s VIRPT register, the write triggers a vector interrupt.
For more information, see “Multiprocessing Interrupts” on page 3-47
To use the DSP’s vector interrupt feature, the host could perform the following sequence of actions.
1. Poll the DSP’s VIRPT register until the host reads a certain token
value (for example, zero).
2. Write the vector interrupt service routine address to VIRPT.
3. When the service routine is finished, the DSP should write the
token back into VIRPT to indicate that it is finished and that
another vector interrupt can be initiated.
A host using direct writes and vector interrupts should use the Direct
Write Pending (DWPD) bit to determine when a direct write to internal
memory is pending, because direct writes and I/O processor register
accesses may be completed in different sequences. If the host performs a
direct memory write to a DSP then writes to an I/O processor register on
the DSP, the I/O processor register write may complete before the direct
write. Because of this FIFO latency, direct writes performed just before
vector interrupt writes (to VIRPT) may be delayed until after the branch to
the interrupt vector.
ADSP-21160 SHARC DSP Hardware Reference
7-79
Host Processor Interface
To prevent unacceptable, latency-related sequencing, the host should
check that all direct writes have completed before writing to the DSP’s
VIRPT register. By polling the DSP’s DWPD bit and waiting for direct writes
to clear, the host can write to VIRPT with correct sequencing.
System Bus Interfacing
A DSP subsystem, consisting of several DSPs with local memory may be
viewed as one of several processors connected together by a system bus.
Examples of such systems are the EISA bus, PCI bus, or several DSP subsystems. The processors in such a system arbitrate for the system bus using
an arbitration unit. Each device on the bus that needs to become a bus
master must be able to drive a bus request signal and respond to a bus
grant signal. The arbitration unit determines which request to grant in any
given cycle.
Access to the DSP Bus—Slave DSP
Figure 7-22 on page 7-81 shows an example of a interface to a system bus
that isolates the local DSP bus from the system bus. When the system is
not accessing the DSPs, the local bus supports transfers between other
local DSPs and local external memory or devices.
7-80
ADSP-21160 SHARC DSP Hardware Reference
External Port
HBG
REDY
HBR
SYSTEM BUS INTERFACE
ADSP-21160
#2
5
3
010
SYSTEM
DATA BUS
BR2
ADDR31-0
BR1,
BR3-BR6
DATA63-0
ID2-0
CS
HBR
WRH/L
HBG
RDH/L
OE
T/R
HBG
WRITE
REDY
ACK
MS3-0
READ
EXTERNAL
MEMORY
ADDR
SYSTEM
ADDRESS
BUS
DATA
WE
OE
HBR
CS2
CS1
ACK
CS
ADDRESS
COMPARATOR
"ADDRESS
VALID"
ADSP-21160
#1
REDY
ACK
BR1
ADDR31-0
BR2-BR6
DATA63-0
SYSTEM BUS
5
3
001
ID2-0
CS
HBR
WRH/L
HBG
RDH/L
REDY
ACK
MS3-0
CLUSTER
BUS
Figure 7-22. Slave DSP System Bus Interface
ADSP-21160 SHARC DSP Hardware Reference
7-81
Host Processor Interface
When the system needs to access a DSP, the system executes a read or
write to the address range of the DSP subsystem. The external address
comparator detects a local access and asserts HBR and one of the appropriate CS lines. The DSP holds off the system bus with REDY until the DSP is
ready to accept the data. The HBG signal enables the system bus buffers.
The buffers’ direction for data is controlled by the read or write signals.
To avoid glitching the HBR line when addresses are changing, the address
comparator may be qualified by an enable signal from the system or qualified by the system read or write signals. These methods cause HBR to be
deasserted each time system read or write is deasserted or the address is
changed. Because these techniques deassert HBR with each access, the
overhead of an HTC occurs as part of each access. To avoid this type of
overhead, systems can latch HBG during long sequences of bus accesses.
Access to the System Bus—Master DSP
Figure 7-23 on page 7-83 shows a bidirectional system interface in which
the DSP subsystem can access the system bus by becoming a bus master.
Before beginning the access, the DSP first requests permission to become
the bus master by generating the System Bus Request signal (SBR). A bus
arbitration unit determines when to respond with the System Bus Grant
signal (SBG). Here, each system bus master generates and responds to its
own unique pair of signals.
The method a DSP uses to arbitrate for the system bus depends on
whether the access is from the DSP processor core or I/O processor. For
more information, see “Processor Core Access To System Bus” on
page 7-84 and “DSP DMA Access To System Bus” on page 7-88.
7-82
ADSP-21160 SHARC DSP Hardware Reference
External Port
HBG
REDY
HBR
SYSTEM BUS INTERFACE
FLAG0
5
010
3
ADSP-21160
#2
BR2
ADDR31-0
BR1,
BR3-BR6
DATA63-0
ID2-0
CS
HBR
WRH/L
HBG
RDH/L
REDY
ACK
SBTS
MS3-0
SYSTEM
DATA BUS
SYSTEM
BUS
GRANT
HBG
WRITE
READ
EXTERNAL
MEMORY
ADDR
DATA
SYSTEM
ADDRESS
BUS
WE
OE
ACK
HBR
CS2
CS1
CS
FLAG0
5
001
3
ADDRESS
COMPARATOR
ADSP-21160
#1
BR1
ADDR31-0
BR2-BR6
DATA63-0
ID2-0
CS
HBR
WRH/L
HBG
RDH/L
"ADDRESS
VALID"
ACK
REDY
SYSTEM BUS
REDY
ACK
SBTS
MS3-0
CLUSTER
BUS
SYSTEM
BUS
REQUEST
ACK
MS3-0
FLAG0 (1,2)
FLAG0 (1,2)
SBTS (1,2)
SYSTEM
BUS
GRANT
SBTS (1,2)
Figure 7-23. Bidirectional System Bus Interface
ADSP-21160 SHARC DSP Hardware Reference
7-83
Host Processor Interface
Processor Core Access To System Bus
The DSP core may arbitrate for the system bus by setting a flag and waiting for SBG on another flag. This technique has the benefit of not stalling
the local bus while waiting. If SBG is tied to an interrupt pin, the DSP can
continue processing while waiting.
Another method is for the DSP to attempt the access assuming that the
system bus is available. The DSP then either waits or aborts the access if
the bus is not available. The DSP begins the access to the system bus by
asserting one of the memory select lines, MS3-0. This assertion also asserts
SBR. If the system bus is not available (for example, SBG is deasserted), the
DSP should be held off with ACK. This approach is simple, but stalls the
DSP and the local bus when the system bus is accessed while it is busy. To
overcome this stall, programs can use the Type 10 instruction:
IF condition JUMP(addr), ELSE compute, DM(addr)=dreg;
This instruction aborts the bus access if the condition (SBG) is not true and
causes a branch to a try-again-later routine. This method works well if SBG
is asserted most of the time. If the Type 10 instruction is not used, a deadlock condition can result if an access is attempted before the bus is
granted.
The DSP samples
inputs at the
frequency, and
 outputs
must be held stable for at least one full
cycle.
FLAG
CLKIN
FLAG
CLKIN
Deadlock Resolution
When both the DSP subsystem and a host processor try to access the system bus in the same cycle, a bus deadlock occurs in which neither access
can complete. Normally, the master DSP responds to an HBR request by
asserting HBG after the completion of the current access. If the DSP is
accessing the system bus at the same time, HBG is not asserted, because this
current access cannot complete. This condition results in a deadlock in
which neither access can complete. The deadlock may be broken by asserting the Suspend Bus Three-state (SBTS) input for one or more cycles after
7-84
ADSP-21160 SHARC DSP Hardware Reference
External Port
the deadlock is detected—when the system bus to local bus buffer is
requested from both sides.
The combination of SBTS and HBR forces the master DSP to suspend its
core access and assert HBG. This lets the system access to the local bus proceed. The combination of HBR and SBTS should be applied only when a
deadlock is caused by a DSP access to the system bus. The SBTS signal
should not be used when there is a local bus transfer, because the WRH/L
signal is asserted twice—once before the SBTS is asserted and once after the
access resumes. For DSP-to-DSP transfers on the local bus, this double
assertion violates the slave timing requirements.
The signals ACK, HBG, REDY, and the data bus are all active in slave mode. If
the DSP was performing an external access (which did not complete) in
the same cycle that SBTS and HBR were asserted, the access is suspended
until SBTS and HBR are both deasserted again. The timing in Figure 7-24
on page 7-86 shows example timing for the deadlock resolution case on a
synchronous host interface.
previous SHARC DSPs,
and
can be used together
 Asfor with
and
are
host/DSP deadlock resolution. However, if
SBTS
HBR
SBTS
HBR
asserted while bus lock is set, the DSP three-states its bus signals,
but does not go into slave mode.
When a host processor uses SBTS and HBR for deadlock resolution, SBTS
operates differently than when the DSP uses only SBTS. For more information on how the DSP uses SBTS, see the discussion on page 7-17.
The signals
and
can be used together for host/DSP core
 deadlock
resolution but not for DMA transfers or bursts.
SBTS
HBR
ADSP-21160 SHARC DSP Hardware Reference
7-85
Host Processor Interface
DSP Read
BTC
Dead
HTC
Host Access
BTC
Read Resumed
CLKIN
HBR
SBTS
HBG
BRx
A0
ADDR[31:0]
Host Address
A0
Driven by Host
MSx
RDH/L
WRH/L
ACK for synchronous host
ACK
D0
DATA[63:0]
Host Data
D0
Figure 7-24. Deadlock Resolution Example Timing
7-86
ADSP-21160 SHARC DSP Hardware Reference
External Port
The following sequence of actions allows the host processor to suspend an
ongoing DSP access and gain access to its internal resources, provided
that: (1) the access originates from the DSP’s core, not the DMA controller, (2) a DRAM page miss is not detected for that memory access, and
(3) bus lock is not enabled.
1. The host interface asserts HBR.
2. The host interface determines whether the request has created a
deadlock. If a deadlock is detected, the host asserts SBTS for one or
more cycles.
HBR should be asserted first. SBTS should be asserted only after
detecting the deadlock. SBTS should remain asserted at least until
the DSP asserts HBG. SBTS should be deasserted before HBR deasserts.
SBTS should be a stable signal that follows a proper setup and hold
requirement as described in the ADSP-21160 DSP Microcomputer
Data Sheet.
3. The DSP sees asserted SBTS with HBR. If SBTS is asserted in cycle 0,
the DSP samples it in cycle 1 and aborts the ongoing access in the
same cycle.
4. The host three-states both RDH/L and WRH/L in cycle 1.
5. In cycle 3, HBG is asserted. The host acquires full control of the bus
and may access any of the DSPs or peripherals on the bus. Remove
SBTS before HBR is deasserted.
6. The host performs the accesses as bus master. After completing the
accesses, the host deasserts HBR.
7. The DSP waits for ACK to be asserted. If ACK is sampled asserted in
cycle 0, HBG is deasserted in cycle 1. In cycle 2, the DSP starts the
access (which it aborted to give mastership to host) as a fresh
access. All the strobes fire as if it is a new access. The wait counter
is also reset.
ADSP-21160 SHARC DSP Hardware Reference
7-87
Host Processor Interface
8. The DSP becomes bus master again.
When SBTS is used for deadlock resolution, an asserted ACK is not
needed to assert HBG. The host, however, may need a valid ACK signal for synchronous accesses. ACK must be asserted to deassert HBG
and return the bus mastership back to the DSP. ACK is then used for
normal DSP transfer requirements.
DSP DMA Access To System Bus
Using the SBTS and HBR inputs to resolve a system bus deadlock, as
described in “Deadlock Resolution” on page 7-84, cannot be done for
DMA transfers because after a DMA word transfer has begun in the DSP,
it must be completed (for example, it must receive the ACK signal). If SBTS
and HBR are asserted during a DMA access, the HBG pin is not asserted until
the access cycle has completed. If the single DMA access is not allowed to
complete, a deadlock condition may result.
To prevent system bus deadlock when using DMA, programs must make
sure that SBG has been asserted before the DMA sequence begins. If a
higher priority access is needed, the DMA sequence may be held off (by
asserting HBR) at any time after a word has been transferred. Systems must
make sure that SBG is asserted before HBR is deasserted to prevent the possibility of another deadlock occurring. When the DMA sequence is
complete, the DMA interrupt service routine should clear the external SBR
flag.
Because the system bus is likely to be considerably slower than the DSP
local bus, performance on the local bus may be considerably improved by
using handshake mode DMA. In this case, the SBG signal is tied to the
DMA request line, DMARx. The local and system bus accesses are only initiated when the system bus is available.
a FIFO in the system interface unit, to allow DMA data
 Using
from the local bus to be posted, may also increase performance on
the local bus when using a slow system bus.
7-88
ADSP-21160 SHARC DSP Hardware Reference
External Port
Multiprocessing with Local Memory
Figure 7-25 shows how several DSP subsystems may be connected
together on a system bus for high throughput. The gate array implements
bus arbitration when the system bus is accessed. The buffers isolate the
DSP local buses from the system bus.
The example system in Figure 7-25 works in the following way.
• A DSP requests the system bus with SBR when it asserts the MS2
line. The gate array arbitrates between the SBR lines and then
enables the highest priority group by asserting SBG, which is tied to
FLAG0.
• The master DSP may connect to system memory or to other DSP
groups. When the bus buffer is enabled, the read or write strobe
enables should be asserted with a delay to allow the address to
stabilize.
• To access a DSP slave in another group, the master DSP addresses
that group’s multiprocessor memory space. The gate array detects
group multiprocessor memory space from three high-order address
bits and asserts the HBR line for the selected group. When HBG is
asserted, the gate array enables the slave’s bus buffer. The
high-order group address bits are cleared by the buffer to allow the
group to decode the address as local multiprocessor memory space.
The access is synchronous because the CS line is not asserted. The
single waitstate option for the bus should be enabled.
• If two groups access each other in the same cycle, a deadlock may
occur. The SBTS pin may be used to clear the deadlock.
DSP To Microprocessor Interface
A DSP without external memory may connect to a host microprocessor’s
bus. Depending on the microprocessor’s I/O capabilities, the interface
may not require any buffers. This type of connection assumes that the
ADSP-21160 SHARC DSP Hardware Reference
7-89
Host Processor Interface
3
SYSTEM BUS ADDRESS
SYSTEM BUS
ADSP-21160
#2
SBR
MS2
SBG
FLAG0
ADDR
LOCAL BUS
BUFFER
ENABLE
ACK
DATA
HBR
HBG
SBTS
SYSTEM
MEMORY
LOCAL
MEMORY
SYSTEM
BUS
ARBITRATION
DATA
ADDR
(GATE ARRAY)
ADSP-21160
#1
SBR
MS2
SBG
FLAG0
DATA63-0
ADDR31-0
3 SYSTEM BUS ADDRESS
ADDR
LOCAL BUS
ENABLE
BUFFER
ACK
DATA
HBR
HBG
SBTS
LOCAL
MEMORY
Figure 7-25. DSP Subsystems On A System Bus
DSP can execute its application from internal memory most of the time
and only occasionally needs to request an external access.
The host microprocessor should always keep the HBR request asserted
unless it sees BR1 asserted (for the BRx line of the DSP with ID=001). The
host can then deassert HBR to allow the DSP to perform an external access
7-90
ADSP-21160 SHARC DSP Hardware Reference
External Port
when the host is ready to give up its bus. Usually, the host can read or
write to the DSP as needed. The host accesses the DSP by asserting the CS
signal and handshaking with REDY. Host Bus Grant (HBG) need not be used
in this system.
Multiprocessor (DSPs) Interface
The ADSP-21160 supports connecting to other ADSP-21160s to create
multiprocessing DSP systems. This support includes:
• Distributed, on-chip arbitration for the shared external bus
• A unified multiprocessor address space that makes the internal
memory and I/O processor registers of all DSPs directly accessible
to each DSP (and host interface)
• Dedicated hardware support for interprocessor communication
(for example, reflective semaphores)
• Dedicated, point-to-point communication channels between DSPs
using the link ports
Figure 7-26 on page 7-93 illustrates a basic multiprocessing system. In a
multiprocessor system with several DSPs sharing the external bus, any of
the processors can become the bus master. The bus master has control of
the bus, which consists of the DATA63-0, ADDR31-0, and associated control
lines.
ADSP-21160 SHARC DSP Hardware Reference
7-91
Multiprocessor (DSPs) Interface
Table 7-11 on page 7-92 shows the external port signals that are needed in
multiprocessor DSP arbitration and communication.
Table 7-11. Signal For Cluster Multiprocessor Systems
Signal Types
Signals
Synchronization:
CLKIN, RESET
Arbitration:
BR6-1, PA1
Bused Information:
ADDR31-0, DATA63-0
Master Controls:
RDH, RDL, WRH, WRL,
Slave Control:
ACK
Host Interface:2
HBR, HBG, CS3,
1
2
3
BRST
REDY3, SBTS
Optional, only needed if Priority Access function is used
Optional, only needed if Host Interface is used.
Optional, only needed if asynchronous Host Interface employed.
The internal memory and I/O processor registers of the system’s DSPs
comprise the multiprocessor memory space. Multiprocessor memory space
is mapped into the unified address space of each DSP. For more information, see the multiprocessor memory map in Figure 5-8 on page 5-18.
After a DSP becomes the bus master, it can directly read and write the
internal memory of any other slave DSP. The master can also read and
write to any of the slave’s I/O processor registers, including their external
port FIFO data buffers. For example, the master DSP may write to a
slave’s I/O processor registers to set up DMA transfers or to send a vector
interrupt.
Multiprocessing System Architectures
Multiprocessor systems typically use one of two schemes to communicate
between processor nodes. One scheme uses dedicated point-to-point
7-92
ADSP-21160 SHARC DSP Hardware Reference
CONTROL
ADDRESS
DATA
ADDRESS
DATA
ADSP-2116X #6
ADSP-2116X #5
ADSP-2116X #4
CONTROL
External Port
ADSP-2116X #3
CLKIN
RESET
RPBA
3
011
ID2-0
ADDR31-0
DATA63-0
CONTROL
PA
BR1, BR3-6
BR3
5
ADSP-2116X #2
CLKIN
RESET
RPBA
3
ID2-0
010
ADDR31-0
DATA63-0
CONTROL
PA
BR1, BR3-6
BR2
5
ADSP-2116X #1
CLOCK
CLKIN
RESET
RESET
RPBA
3
001
ID2-0
CONTROL
ADDR31-0
ADDR
DATA63-0
DATA
RDX
WRX
ACK
MS3-0
OE
WE
ACK
CS
BMS
PAGE
CS
ADDR
SBTS
BRST
CLKOUT
DATA
CS
HBR
HBG
REDY
PA
BR2-6
BR1
ADDR
5
GLOBAL
MEMORY
AND
PERIPHERALS
(OPTIONAL)
BOOT
EPROM
(OPTIONAL)
HOST
PROCESSOR
INTERFACE
(OPTIONAL)
DATA
Figure 7-26. ADSP-21160 Multiprocessor System
ADSP-21160 SHARC DSP Hardware Reference
7-93
Multiprocessor (DSPs) Interface
communication channels. In the other scheme, nodes communicate
through a single shared global memory over a parallel bus.
The DSP supports point-to-point communication—data flow multiprocessing—through its six link ports. Also, the DSP supports a shared
parallel bus communication—cluster multiprocessing—through its link
ports and external port. For more information on data flow multiprocessing, see “Data Flow Multiprocessing” below and “Data Flow
Multiprocessing With Link Ports” on page 8-22. For more information on
cluster multiprocessing, see “Cluster Multiprocessing” on page 7-95.
Data Flow Multiprocessing
Data flow multiprocessing works for applications requiring high computational bandwidth, but requiring only limited flexibility. The program
partitions its algorithm sequentially across multiple processors and passes
data through a line of processors, as shown in Figure 7-26.
ADSP-21160
LINK
PORT
LINK
PORT
ADSP-21160
LINK
PORT
LINK
PORT
ADSP-21160
LINK
PORT
LINK
PORT
Figure 7-27. Data Flow Multiprocessing
The DSP provides complete support for data flow multiprocessing applications, because the DSP eliminates the need for interprocessor data
FIFOs and external memory. The internal memory of the DSP is usually
large enough to contain both code and data for most applications using
data-flow system-topology. All that a data flow system requires are a
number of DSPs and point-to-point signals connecting them. This design
yields savings in complexity, board space, and system cost. For more information on connecting multiple DSPs using link ports, see “Host Processor
Access To Link Buffers” on page 8-11.
7-94
ADSP-21160 SHARC DSP Hardware Reference
External Port
Cluster Multiprocessing
Cluster multiprocessing works for applications where a fair amount of
flexibility is required. This flexibility is needed when a system must be
able to support a variety of different tasks, some of which may be running
concurrently. The cluster multiprocessing configuration is shown in
Figure 7-28. Also, the DSP has an on-chip host interface that lets a cluster
be interfaced to a host processor or another cluster.
ADSP-21160
LINK
PORT
LINK
PORT
EXTERNAL PORT
ADSP-21160
LINK
PORT
LINK
PORT
EXTERNAL PORT
ADSP-21160
LINK
PORT
LINK
PORT
EXTERNAL PORT
BULK
MEMORY
Figure 7-28. Cluster Multiprocessing
Cluster multiprocessing systems include multiple DSPs connected to a
parallel bus that supports interprocessor access of on-chip memory and
access to shared global memory. In a typical cluster of DSPs, up to six processors and a host can arbitrate for the bus. The on-chip bus arbitration
logic lets these processors share the common bus.
The DSP’s features (such as large internal memory, link ports, and external port FIFOs) help eliminate the need for any extra hardware in the
cluster multiprocessor configuration. External memory—local and
global—can frequently be eliminated in this type of system.
ADSP-21160 SHARC DSP Hardware Reference
7-95
Multiprocessor (DSPs) Interface
The DSP supports fixed and rotating priority schemes. Other supported
techniques include bus locking, timed release, DMA prioritization, and
core processor access preemption of background DMA transfers. The
on-chip arbitration logic lets transitions in bus mastership take up to only
one cycle of overhead. Bus requests are generated implicitly when a processor accesses an external address. Because each processor monitors all
bus requests and applies the same priority logic to the requests, each can
independently determine who is the next bus master.
After getting mastership of the bus, a DSP can access external memory and
the internal memory and I/O processor registers of all other DSPs (slaves)
in the system. A DSP can directly transfer data to another DSP or set up a
DMA channel to transfer the data. The DSPs are mapped into a common
memory map—to identify the address space of each DSP within the unified memory map of the system cluster. Also, each DSP has a unique ID.
The DSP’s I/O processor registers, internal memory, and external memory
are all part of the unified address space. This shared on-chip memory
eliminates the need to use external memory for message passing between
DSPs and simplifies software communications. DSPs can write directly
into each other’s memory, saving an extra transfer step.
The multiprocessor communication between DSPs is eased with the
broadcast write feature, which lets a DSP write to all processors simultaneously. This broadcast supports reflective semaphores, where a processor
polls its own internal copy of the semaphore and only uses the external bus
for a broadcast write to all other processors when it wants to change it.
This reduces communications traffic on the external bus.
The cluster configuration allows the DSPs to have a very fast
node-to-node data transfer rate. Clusters also allow a simple, efficient software-communication model.
7-96
ADSP-21160 SHARC DSP Hardware Reference
External Port
For example, all of the required setup operations for a DMA transfer can
be accomplished by a single DSP on one side of the transfer. The other
processor is not interrupted until the DMA transfer is complete.
DSP’s internal memory facilitates I/O in multiprocessor sys The
tems. The on-chip, dual-ported RAM supports full-speed
inter-DSP transfers concurrent with dual accesses by the DSP’s
processor core. No cycles are stolen from the processor core, and
the processor’s full performance is maintained during these
accesses.
Link Port Data Transfers In A Cluster. A bottleneck exists within the
cluster because only two DSPs can communicate over the shared bus
during each cycle—other DSPs are held off until the bus is released.
Because the DSP can also perform point-to-point link port transfers
within a cluster, systems can eliminate this bottleneck by setting up data
communication through the link ports. Data links between DSPs can be
dynamically set up and initiated over the common bus. All six link ports
can operate simultaneously on each DSP.
A disadvantage of the link ports is that individual transfers occur at a
much lower rate than that of the shared parallel bus. Because the link
ports’ 8-bit data path is smaller than the processor’s native word size, the
transfer of each word requires multiple clock cycles. Link ports may also
require more software overhead and complexity because they must be set
up on both sides of the transfers before they can occur.
SIMD Multiprocessing. For certain classes of applications such as radar
imaging, a Single-Instruction, Multiple-Data (SIMD) array of DSPs may
be the most efficient topology to coordinate a large number of DSPs in a
single system. The SIMD array of Figure 7-28 on page 7-95 consists of
multiple DSPs connected in a two- or three-dimensional mesh. The data
link ports provide nearest neighbor communications and through-routing
of data. A single master DSP provides the instruction stream that the array
executes. Data flow in and out the array can be managed through multiple
serial port streams.
ADSP-21160 SHARC DSP Hardware Reference
7-97
Multiprocessor (DSPs) Interface
Multiprocessor Bus Arbitration
Multiple DSPs can share the external bus with no additional arbitration
logic. Arbitration logic is included on-chip to allow the connection of up
to six DSPs and a host processor. The DSP accomplishes bus arbitration
through the BR1-6, HBR, and HBG signals. BR1-6 arbitrate between multiple
DSPs, and HBR/HBG pass control of the bus from the DSP bus master to the
host and back. The priority scheme for bus arbitration is determined by
the setting of the RPBA pin.
Table 7-12 defines the DSP pins used in multiprocessing systems.
Table 7-12. Multiprocessing DSP Pins
Signal
Type
Definition
BR6-1
I/O/S
Multiprocessing Bus Requests. Used by multiprocessing DSPs to
arbitrate for bus mastership. A DSP only drives its own BRx line (corresponding to the value of its ID2-0 inputs) and monitors all others.
In a multiprocessor system with less than six DSPs, the unused BRx
pins should be tied high; the processor’s own BRx line must not be
tied high or low because it is an output. Note that ID=00x device
enables keeper latch/pull-up devices on certain signals. For more
information, see Table 11-1 on page 11-3.
ID2-0
I
Multiprocessing ID. Determines which multiprocessing bus request
(BR1—BR6) is used by DSP. ID=001 corresponds to BR1, ID=010
corresponds to BR2, etc. (ID=000 is used in single-processor systems.) These lines are a system configuration selection which should
be hardwired or only changed at reset.
I = Input, S = Synchronous, (o/d) = Open Drain; O = Output, A = Asynchronous,
(a/d) = Active Space
7-98
ADSP-21160 SHARC DSP Hardware Reference
External Port
Table 7-12. Multiprocessing DSP Pins (Cont’d)
Signal
Type
Definition
RPBA
I
Rotating Priority Bus Arbitration Select. When RPBA is high, rotating priority for multiprocessor bus arbitration is selected. When
RPBA is low, fixed priority is selected. This signal is a system configuration selection which must be set to the same value on every DSP.
If the value of RPBA is changed during system operation, it must be
changed in the same CLKIN cycle on every DSP.
PA
(a/d)
I/O/S
Priority Access. The DSP slave may assert the PA signal to interrupt
background DMA transfers and gain access to the external bus. This
signal is asserted when a DSP slave’s processor core requests the bus
or if an external DMA channel requests the bus with the DMACx
PRIO control bit set. The PA signal is an active drive output, which
may be asserted (low) by one or more slaves. It is deasserted (high) by
the master. A protocol is employed to avoid driver contention.
I = Input, S = Synchronous, (o/d) = Open Drain; O = Output, A = Asynchronous,
(a/d) = Active Space
The ID2-0 pins provide a unique identity for each DSP in a multiprocessing system. The first DSP should be assigned ID=001, the second should
be assigned ID=010, and so on. One of the DSPs must be assigned ID=001
in order for the bus synchronization scheme to function properly.
DSP with
 The
during reset.
ID=001
holds the external bus control lines stable
When the ID2-0 inputs of a DSP are equal to 001, 010, 011, 100, 101, or
110, the DSP configures itself for a multiprocessor system and maps its
internal memory and I/O processor registers into the multiprocessor
memory space. ID=000 configures the DSP for a single-processor system.
ID=111 is reserved and should not be used.
A DSP in a multiprocessor system can determine which processor is the
current bus master, by reading the CRBM2-0 bits of the SYSTAT register.
These bits give the value of the ID2-0 inputs of the current bus master.
ADSP-21160 SHARC DSP Hardware Reference
7-99
Multiprocessor (DSPs) Interface
Conditional instructions can be written, depending on whether the DSP is
the current bus master in a multiprocessor system. The assembly language
mnemonic for this condition code is Bm, and its complement is Not Bm
(not bus master). The Bm condition indicates whether the DSP is the current bus master. For more information, see “Conditional Sequencing” on
page 3-52. To use the bus master condition, the condition code select
(CSEL) field in the MODE1 register must be zero or the condition is always
evaluated as false.
Bus Arbitration Protocol
The Bus Request (BR1-6) pins are connected between each DSP in a multiprocessing system, with the number of BRx lines used equal to the
number of DSPs in the system. Each processor drives the BRx pin that corresponds to its ID2-0 inputs and monitors all others. If less than six DSPs
are used in the system, the unused BRx pins should be tied high.
When one of the slave DSPs needs to become bus master, it automatically
initiates the bus arbitration process by asserting its BRx line at the beginning of the cycle. Later in the same cycle, the DSP samples the value of the
other BRx lines.
The cycle in which mastership of the bus is passed from one DSP to
another is called a Bus Transition Cycle (BTC). A bus transition cycle
occurs when the current bus master’s BRx pin is deasserted and one or
more of the slave’s BRx pins is asserted. The bus master can retain bus mastership by keeping its BRx pin asserted. Also, the bus master does not
always lose bus mastership when it deasserts its BRx line—another BRx line
must be asserted by one or more of the slaves at the same time. In this
case, when no other BRx is asserted, the master does not lose any bus
cycles.
By observing all of the BRx lines, each DSP can detect when a bus transition cycle occurs and which processor has become the new bus master. A
bus transition cycle is the only time that bus mastership is transferred.
7-100
ADSP-21160 SHARC DSP Hardware Reference
External Port
After conditions determine that a bus transition cycle is going to occur,
every DSP in the system evaluates the priority of the BRx lines asserted
within that cycle. For a description of bus arbitration priority, see “Bus
Arbitration Priority (RPBA)” on page 7-104. The DSP with the highest
priority request becomes the bus master on the following cycle, and all of
the DSPs update their internal record to indicate which DSP is the current bus master. This information can be read from the current bus master
field, CRBM, of the SYSTAT register. Figure 7-29 shows typical timing for
bus arbitration.
ADSP-21160
#1 IS BUS
BTC
MASTER
ADSP-21160 #2 IS BUS MASTER
BTC
CLKIN
BUS REQUESTS:
BR1
BRX SAMPLED
AT THIS POINT
BR2
EXECUTE
FLOW
BUS
ACTIVE
ADSP-21160 WITH ID=1:
INTERNAL
OPERATION
INTERNAL
OPERATION
INTERNAL
OPERATION
INTERNAL
OPERATION
HOLD SIGNAL
STABLE
EXTERNAL
ACCESS
INTERNAL
OPERATION
PERFORM
ACCESS
UNDRIVEN
ADSP-21160 WITH ID=2:
EXECUTE
FLOW
BUS
ACTIVE
INTERNAL
OPERATION
UNDRIVEN
EXTERNAL
ACCESS
PERFORM
ACCESS
EXTERNAL
ACCESS
INTERNAL
OPERATION
EXTERNAL
ACCESS
INTERNAL
OPERATION
INTERNAL
OPERATION
PERFORM
ACCESS
HOLD SIGNAL
STABLE
PERFORM
ACCESS
HOLD SIGNAL
STABLE
UNDRIVEN
Figure 7-29. Bus Arbitration Timing
ADSP-21160 SHARC DSP Hardware Reference
7-101
Multiprocessor (DSPs) Interface
The actual transfer of bus mastership is accomplished by the current bus
master three-stating the external bus—DATA63-0, ADDR31-0, RDH, RDL, WRH,
WRL, BRST, MS3-0, CIF, PAGE, HBG, DMAG2-1—at the end of the bus transition cycle and the new bus master driving these signals at the beginning of
the next cycle. The bus strobes (RDH, RDL, WRH, and WRL) and MS3-0 are
driven high (inactive) before three-stating occurs. ACK must be sampled
high by the new master before it starts a new bus operation. For more
information, see Figure 7-30 on page 7-103.
During bus transition cycle delays, execution of external accesses are
delayed. When one of the slave DSPs needs to perform an external read or
write, it automatically initiates the bus arbitration process by asserting its
BRx line. This read or write is delayed until the processor receives bus mastership. If the read or write was generated by the DSP’s processor core
(not the I/O processor), program execution stops on that DSP until the
instruction is completed.
The following steps occur as a slave acquires bus mastership and performs
an external read or write over the bus as shown in Figure 7-30.
1. The slave determines that it is executing an instruction which
requires an off-chip access. It asserts its BRx line at the beginning of
the cycle. Extra cycles are generated by the core processor (or I/O
processor) until the slave acquires bus mastership.
2. To acquire bus mastership, the slave waits for a bus transition cycle
in which the current bus master deasserts its BRx line. If the slave
has the highest priority request in the bus transition cycle, it
becomes the bus master in the next cycle. If not, it continues
waiting.
3. At the end of the bus transition cycle the current bus master
releases the bus, and the new bus master starts driving.
7-102
ADSP-21160 SHARC DSP Hardware Reference
External Port
1
2
3
4
5
6
CLKIN
BRX
OPTIONAL
HIGHEST PRIORITY REQUESTER BECOMES BUS MASTER
ADDR
VALID
VALID
MSX
VALID
VALID
RDH
MS, STROBES DRIVEN INACTIVE BEFORE TRISTATE
RDL
WRH
WRL
BRST
pull-up HOLDS ACK ASSERTED
ACK
MINIMUM 2-CYCLE SYNC READ
- SLAVE DEASSERTS ACK IN 2ND
CYCLE IF NEEDED
DATA
BTC
VALID
VALID
SYNC WRITE
SYNC READ
BTC DOES NOT
ACCESS
ACCESS
OCCUR IF NO OTHER
BRS ASSERTED
Figure 7-30. Bus Request and Read/Write Timing
ADSP-21160 SHARC DSP Hardware Reference
7-103
Multiprocessor (DSPs) Interface
During the CLKIN cycle in which the bus master deasserts its BRx output, it
three-states its outputs in case another bus master wins arbitration and
enables its drivers in the next CLKIN cycle. If the current bus master retains
control of the bus in the next cycle, it enables its bus drivers, even if it has
no bus operation to run.
The DSP with ID=00x enables internal keeper latches, or pull-up devices,
on key signals, including the address and data buses, strobes, and ACK.
These devices provide a weak current source or sink—approximate 20K
impedance—to keep these signals from drifting near input receiver thresholds when all drivers are three-stated. For more information, see
Table 11-1 on page 11-3.
When the bus master stops using the bus, its BRx line is deasserted, allowing other DSPs to arbitrate for mastership if they need it. If no other DSPs
are asserting their BRx line when the master deasserts its BRx, the master
retains control of the bus and continues to drive the memory control signals until: 1) it needs to use the bus again, or 2) another DSP asserts its
BRx line.
waits to be a master for a DMA transfer, it asserts
 While. If athatslaveslave’s
core accesses the DA group registers, the
is
BRx
BRx
deasserted during that access.
Bus Arbitration Priority (RPBA)
To resolve competing bus requests, there are two available priority
schemes: fixed and rotating. The RPBA pin selects the scheme. When RPBA
is high, rotating priority bus arbitration is selected, and when RPBA is low,
fixed priority is selected.
The RPBA pin must be set to the same value on each DSP in a multiprocessing system. If the value of RPBA is changed during system operation, it
must be changed synchronously to CLKIN and must meet a setup time that
lets all DSPs recognize the change in the same cycle. The priority scheme
changes in that (same) cycle.
7-104
ADSP-21160 SHARC DSP Hardware Reference
External Port
In the fixed priority scheme, the DSP with the lowest ID number among
the competing bus requests becomes the bus master. If, for example, the
processor with ID=010 and the processor with ID=100 request the bus
simultaneously, the processor with ID=010 becomes bus master in the following cycle.
DSP knows the of the other processors requesting the bus,
 Each
because their corresponds to the
line being used.
ID
ID
BRx
The rotating priority scheme gives roughly equal priority to each DSP.
When rotating priority is selected, the priority of each processor is reassigned after every transfer of bus mastership. Highest priority is rotated
from processor to processor as if they were arranged in a circle—the DSP
located next to (one place down from) the current bus master is the one
that receives highest priority. Table 7-13 shows an example of how rotating priority changes on a cycle-by-cycle basis.
Table 7-13. Rotating Priority Arbitration Example
Cycle Number
Hardwired Processor IDs and Priority1
ID1
ID2
ID3
ID4
ID5
ID6
12
M
1
2-BR
3
4
5
2
4
5-BR
M-BR
1
2
3
3
4
5-BR
M
1
2
3
4
5-BR
M
1
2
3
4-BR
53
1-BR
2
3
4
5
M
1 The following symbols appear in these cells: 1-5 = assigned priority, M = bus mastership (in that
cycle), BR = requesting bus mastership with BRx
2 Initial priority assignments
3 Final priority assignments
ADSP-21160 SHARC DSP Hardware Reference
7-105
Multiprocessor (DSPs) Interface
Mastership Timeout Bus
In either the fixed or rotating priority scheme, systems may need to limit
how long a bus master can retain the bus. Systems can limit bus mastership by forcing the bus master to deassert its BRx line after a specified
number of CLKIN cycles and giving the other processors a chance to acquire
bus mastership.
To setup a bus master timeout, a program must load the BMAX register with
the maximum number of CLKIN cycles (minus 2) that the DSP can retain
bus mastership:
BMAX
= (maximum # of bus mastership CLKIN cycles) – 2
 Internal processor clock cycles are a multiple of
CLKIN
cycles.
The minimum value for BMAX is 2, which lets the processor retain bus mastership for 4 CLKIN cycles. Setting BMAX=1 is not allowed. To disable the
bus master timeout function, set BMAX=0.
Each time a DSP acquires bus mastership, its BCNT register is loaded with
the value in BMAX. BCNT is then decremented in every CLKIN cycle that the
master performs a read or write over the bus and any other (slave) DSPs
are requesting the bus. Any time the bus master deasserts its BRx line, BCNT
is reloaded from BMAX.
When BCNT decrements to zero, the bus master first completes its off-chip
read/write and then deasserts its own BRx (any new off-chip accesses are
delayed)—this allows transfer of bus mastership. If the ACK signal is holding off an access when BCNT reaches zero, bus mastership is not
relinquished until the access can complete.
If BCNT reaches zero while a burst transfer is in progress, the bus master
completes the burst transfer before deasserting its BRx output. If BCNT
reaches zero while bus lock is active, the bus master does not deassert its
7-106
ADSP-21160 SHARC DSP Hardware Reference
External Port
line until bus lock is removed. If HBR is being serviced, BCNT stops decrementing and continues only after HBR is deasserted.
BRx
lock is enabled by the
bit in the
register. For more
 Bus
information, see “Bus Lock and Semaphores” on page 7-118.
BUSLK
MODE2
Priority Access
The Priority Access signal (PA) lets external bus accesses by a slave DSP
take priority over ongoing DMA transfers. Normally when external port
DMA transfers are in progress, the slave DSPs cannot use the external bus
until the DMA transfer is finished. By asserting its PA pin, the slave DSP
can acquire the bus without waiting for the DMA operation to complete.
The PA signal can also be asserted by a slave with a high-priority DMA
access pending on the external bus.
If the PA signal is not used in a multiprocessor system, the DSP bus master
does not give up the bus to another DSP until: (1) a cycle in which it does
not perform an external bus access or (2) a bus timeout. If a slave DSP
needs to send a high priority message or perform an important data transfer, it normally must wait until any DMA operation completes. Using the
PA signal lets the slave perform its higher priority bus access with less
delay.
Each of the DMACx registers has a PRIO bit that raises that DMA channel to
a higher priority than all other internal DMA channels that do not have
the PRIO bit set. Unless configured differently with the EBPR bit in the
SYSCON register, this channel still has lower priority (internally) than the
core. Programs should be careful to minimize the number of DMA channels enabled to high priority status in the multiprocessor system, because
both core and (external) high priority DMA requests from slaves are arbitrated at the same priority level.
For example, a slave core cannot arbitrate bus ownership away from a high
priority DMA transfer unless the bus timeout (BMAX function) occurs.
ADSP-21160 SHARC DSP Hardware Reference
7-107
Multiprocessor (DSPs) Interface
When PA is asserted, the current DSP bus master deasserts its BRx output,
and gives up the bus, provided:
• Its core does not have an external access pending, and
• None of its external bus DMA channels have pending high-priority
bus requests.
All DSP slaves also deassert their BRx outputs, if each slave meets the same
provisions. The current bus master never asserts PA, because it already has
control of the bus. If the current master detects a condition that would
assert PA while it is bus master, it performs that high priority operation
before giving up bus ownership.
In the CLKIN cycle after PA has been asserted, only the DSP slaves with a
pending high priority access have their bus requests asserted. Bus arbitration proceeds as usual with the highest priority device becoming the
master when the previous bus master releases its BRx output.
The new master samples all BRx inputs after gaining bus mastership—
during the cycle that follows the BTC. If no other bus requests are
asserted, the master is the only device driving PA, and the master deasserts
and three-state PA in this cycle as shown in Figure 7-31.
If the master samples other BRx inputs as asserted, multiple devices are
driving PA, and the new bus master cannot deassert PA. The new bus master three-states its PA driver in this case. All DSP slaves recognize the cycle
following the BTC. They do not assert PA during this cycle, unless they
were already driving their BR and PA outputs in the BTC.
7-108
ADSP-21160 SHARC DSP Hardware Reference
External Port
This behavior is demonstrated in Figure 7-32.
1
2
BR1-5
3
4
All ADSP-21160s that do not
have core access pending
remove their BRx
BR6
{
{
PA
Slaves cannot assert
BTC
PA in this cycle
Bus Master samples
all other BRx negated
and negates PA
Figure 7-31. Example PA Deassertion
All ADSP-21160s that
do not have core access
pending remove their BRx
1
2
3
4
BR1-6
{
{
PA
BTC
Bus Master samples
other BRx asserted
and three-states (only) PA
Slaves continue
to assert PA in
this cycle
Figure 7-32. Example of PA Driven by Multiple Slaves
ADSP-21160 SHARC DSP Hardware Reference
7-109
Multiprocessor (DSPs) Interface
Bus Synchronization After Reset
When a multiprocessing system is reset (RESET asserted), the bus arbitration logic on each processor must synchronize, making sure that only one
DSP drives the external bus. One DSP must become the bus master, and
all other processors must recognize which one it is before actively arbitrating for the bus. The bus synchronization scheme also lets the system safely
bring individual DSPs into and out of reset.
only difference between the soft and hard reset is that the
 The
external bus arbitration does not get affected by a soft reset (no bus
synchronization at soft reset). The PLL also does not get reset at
soft reset.
One of the DSPs in the system must be assigned ID=001 in order for the
bus synchronization scheme to function properly. This processor also
holds the external bus control lines stable during reset. Bus arbitration
synchronization is disabled if the DSP is in a single-processor system
(ID=000).
To synchronize their bus arbitration logic and define the bus master after
a system reset, the multiple DSPs obey the following rules:
• All DSPs except the one with ID=001 deassert their BRx line during
reset. They keep their BRx deasserted for at least two cycles after
reset and until their bus arbitration logic is synchronized.
• After reset, a DSP considers itself synchronized when it detects a
cycle in which only one BRx line is asserted. The DSP identifies the
bus master by recognizing which BRx is asserted and updates its
internal record to indicate the current master.
• The DSP with ID=001 asserts its BRx (BR1) during reset and for at
least two cycles after reset. If no other BRx lines are asserted during
these cycles, the DSP with ID=001 drives the memory control
7-110
ADSP-21160 SHARC DSP Hardware Reference
External Port
signals to prevent them from glitching. Although it is asserting its
BRx and driving the memory control signals during these cycles,
this DSP does not perform reads or writes over the bus.
If the DSP with ID=001 is synchronized by the end of the two cycles following reset, it becomes the bus master. If it is not synchronized at this
time, it deasserts its BRx (BR1) and waits until it is synchronized.
a DSP has synchronized itself, it sets the
 Whenregister.
BSYN
bit in the SYS-
TAT
If one DSP comes out of reset after the others have synchronized and
started program execution, that DSP may not be able to synchronize
immediately (for example, if it detects more than one BRx line asserted). If
the un-synchronized processor tries to execute an instruction with an
off-chip read or write, it cannot assert its BRx line to request the bus and
execution is delayed until it can synchronize and correctly arbitrate for the
bus.
Synchronization cannot occur while HBG is asserted, because bus arbitration is suspended while the bus is controlled by a host. If HBR is asserted
immediately after reset and no bus arbitration has taken place, the DSP
with ID=001 is considered to be the last bus master.
The DSP with ID=001 maintains correct logic levels on the RDH/L, WRH/L,
MS3-0, CIF, PAGE, and HBG signals during reset. Because the “001” processor can be accidently reset by an erroneous write to the soft reset bit (SRST)
of the SYSCON register, it behaves in the following manner during reset.
• While it is in reset, the DSP with ID=001 attempts to gain control
of the bus by asserting BR1.
• While it is in reset, the DSP with ID=001 drives the RDH/L, WRH/L,
MS3-0, CIF, DMAG1, DMAG2, PAGE, and HBG signals only if it determines that it has control of the bus. For the DSP to decide it has
ADSP-21160 SHARC DSP Hardware Reference
7-111
Multiprocessor (DSPs) Interface
control of the bus, two conditions must be true: 1) BR1 was asserted
and no other BRx lines were asserted in the previous cycle, and
2) HBG was deasserted in the previous cycle.
Timing differences occur during processor reset (RESET) or software reset
(SRST bit in SYSCON register = 1) deassertion (MS3-0, HBG, DMAGx, RDx, WRx,
CIF, PAGE, BRST) and threestate (FLAG3-0, LxCLK, LxACK, LxDAT7-0, ACK,
REDY, PA, TFSx, RFSx, TCLKx, RCLKx, DTx, BMS, TDO, EMU, DATA). These timings occur asynchronously to CLKIN and may not meet the specifications
published in the ADSP-21160 DSP Microcomputer Data Sheet Timing
Requirements and Switching Characteristics tables. Refer to ADSP-21160
DSP Microcomputer Data Sheet for more information.
The DSP with ID=001 continues to drive the RDH/L, WRH/L, MS3-0, CIF,
DMAG1, DMAG2, PAGE, and HBG signals for two cycles after reset, as long as
neither HBG nor any other BRx lines are asserted. At the end of the second
cycle it assumes bus mastership (if it is synchronized), and normal bus
arbitration begins in the following cycle. If it is not synchronized, it deasserts BR1, stops driving the memory control signals and does not arbitrate
for the bus until it becomes synchronized.
Although the bus synchronization scheme allows individual processors to
be reset, the DSP with ID=001 may fail to drive the memory control signals if it is in reset while any other processors are asserting their BRx line.
If the DSP with ID=001 has asserted HBG while it is in reset, it is synchronized when RESET is deasserted. This lets the host start using the bus while
the DSPs are still in reset.
If a host processor attempts to reset the DSP bus master (which is driving
the HBG output), the host immediately loses control of the bus.
During reset, the ACK line is pulled high internally by the DSP bus master
with a 2 k equivalent resistor.
7-112
ADSP-21160 SHARC DSP Hardware Reference
External Port
Booting Another DSP
If the system uses one DSP to boot another DSP over the cluster bus, the
master DSP must do the following to communicate to the slave DSP
(DSP that boots) through the external port interface:
1. Program the PMODE field in DMAC10 of the booting DSP for no
packing.
2. Make 48-bit writes to EPB0 on the booting DSP.
Multiprocessor Direct Writes and Reads
A DSP bus master has the same type of access as a host processor to read
or write the internal memory and I/O processor registers of a slave DSP. A
DSP bus master or host processor can access the slave by reading and writing to the appropriate address in multiprocessor memory space—this is
called a direct read or direct write access. For more information, see “Slave
Direct Reads and Writes” on page 7-66.
Each DSP bus slave monitors addresses driven on the external bus and
responds to any that fall within its region of multiprocessor memory
space. These accesses are invisible to the slave DSP’s processor core. They
do not degrade internal memory or internal bus performance as seen by
the core. This feature lets the processor core continue program execution
uninterrupted.
The DSP bus master can directly read and write the slave’s I/O processor
registers (for example, SYSCON, SYSTAT) to send a vector interrupt or to set
up DMA transfers.
The DSP supports 64-bit direct writes through normal word (32-bit)
address space. The master can perform a 64-bit DMA to a slave by asserting Normal word addresses in multiprocessor slave space with the stride
set to 2. The master can also do 64-bit direct writes by asserting Normal
ADSP-21160 SHARC DSP Hardware Reference
7-113
Multiprocessor (DSPs) Interface
word addresses in multiprocessor slave space with the LW mnemonic set
or with SIMD enabled.
IOP Shadow Registers
In a multiprocessing system, read access to another DSP’s PC or MODE2 register provides useful information. The DSP’s I/O processor registers
include registers that shadow or mirror the program counter (PC), and
MODE2 registers. For more information, see “IOP Shadow Registers” on
page 7-66.
Instruction Transfers
Multiprocessor instruction transfers to or from internal memory of DSP
should use 64-bit transfers for maximum performance (described below).
If 48-bit internal transfers are required, one of the slave EPBx FIFOs must
be employed, using the packing mode function (PMODE) of the DMA
channel.
Maximum throughput is achieved by transferring packed instructions to
or from internal memory, using the full 64-bit data bus width synchronously. For more information, see the Packed versus Unpacked
instruction discussion on on page 6-27.
For packed 64-bit direct reads or writes of 48-bit memory the addresses
must be translated as during a host processor transfer. For more information, see “Instruction Transfers” on page 7-67.
Direct Writes
When a direct write to a slave DSP occurs, the address, data, and control
are latched into a dedicated direct write FIFO. For more information, see
“Direct Writes” on page 7-68.
7-114
ADSP-21160 SHARC DSP Hardware Reference
External Port
Direct Write Latency
The DSP handles asynchronous and synchronous direct writes differently.
This difference influences the latency for the direct writes. For more information, see “Direct Write Latency” on page 7-69.
Direct Reads
When a direct read of a DSP occurs, the address is latched on-chip by the
I/O processor at the end of the first CLKIN cycle. ACK is deasserted in the
following CLKIN cycle. When the data is available, the I/O processor drives
the data and asserts REDY (or ACK). Direct reads cannot be pipelined like
direct writes—they only occur one at a time. See “Direct Reads” on
page 7-70.
Broadcast Writes
Broadcast writes allow simultaneous transmission of data to all of the
DSPs in a multiprocessing system. The master DSP can perform broadcast
writes to the same memory location or I/O processor register on all of the
slaves. During broadcast writes, the master also writes to itself unless the
broadcast is a DMA write. Broadcast writes can be used to implement
reflective semaphores in a multiprocessing system. For more information,
see “Bus Lock and Semaphores” on page 7-118. Broadcast writes also can
simultaneously transfer code or data to multiple processors.
The highest region of multiprocessor memory space, addresses
0x0070 0000 to 0x007F FFFF, is used for broadcast writes. When a write
address falls within this region, each DSP slave responds by accepting the
access; the master DSP also accepts its own broadcast write. A read cycle
generated in the broadcast write region reads the corresponding location
in that processor’s internal memory and does not assert the processor’s
BRx.
Figure 7-33 shows the timing for a typical broadcast write. In this example, the first broadcast write is accepted by all slaves in cycle 1. This
ADSP-21160 SHARC DSP Hardware Reference
7-115
Multiprocessor (DSPs) Interface
broadcast write fills the write buffer capacity of one or more of the DSP
slaves for less than three CLKIN cycles. The second broadcast write stalls on
the bus until write capacity is available in all of the slaves—as indicated by
none of the slaves deasserting ACK in cycle 4. The master—having sampled
ACK deasserted at the end of cycle 2—pre-charges ACK in cycle 3. The master samples ACK asserted at the end of cycle 4, indicating that all slaves have
accepted the second broadcast write.
1
2
3
4
CLKIN
ADDR
DATA
BCast Write 1
Write 1
BCast Write 2
BCast Write 2
WRH/L
ACK
ACK (master)
ACK (slaves)
Figure 7-33. Broadcast Write Timing Example
Because the master DSP must wait for a broadcast write to complete on all
of the slaves, the acknowledge signal is handled differently to prevent
drive conflicts on the ACK line. A wired-OR acknowledge signal is implemented to respond to broadcast writes.
7-116
ADSP-21160 SHARC DSP Hardware Reference
External Port
This protocol operates as follows.
1. The DSP does not assert the strobes for broadcast write, unless it
samples ACK asserted at the end of the previous cycle.
2. The synchronous broadcast write completes on the bus in the first
cycle if ACK is sampled asserted at the end of that cycle.
3. In the first cycle of the broadcast write and in all succeeding odd
cycles, a slave DSP deasserts ACK if it is not ready to allow the
broadcast write to complete on the bus. If it is ready, it does not
drive the ACK line.
4. During all succeeding even cycles in which the broadcast write is
not finished, the slave DSPs do not drive ACK. Instead, the master
DSP drives (for example, pre-charges) ACK high and must continue
the write. (Iterate steps 3 and 4).
In most cases, the ACK signal is high, and the DSP slaves are ready to
accept data at the start of the broadcast write—the write completes in one
cycle. If the ACK signal is low or one of the slaves is not ready to accept the
data, the broadcast write takes a minimum of three cycles.
DSP with =00x enables a keeper latch on the
line to pre The
vent the signal from drifting. This eliminates any power
ID
ACK
consumption caused by the signal drifting to the switching point
and improves the robustness of broadcast writes. Multiprocessor
systems that use broadcast writes should keep the ACK signal line as
free of noise as possible.
Shadow Write FIFO
Because the DSP’s internal memory must operate at high speeds, writes to
the memory do not go directly into the memory array, but rather to a
two-deep FIFO called the shadow write FIFO. The operation of this
ADSP-21160 SHARC DSP Hardware Reference
7-117
Multiprocessor (DSPs) Interface
FIFO is transparent to program execution. For more information, see
“Shadow Write FIFO” on page 7-71.
Data Transfers Through the EPBx Buffers
The DSP bus master can transfer data to and from the slave DSPs through
the external port FIFO buffers, EPB0, EPB1, EPB2, and EPB3. Each of these
buffers, which are part of the I/O processor register set, is an eight-location FIFO. Both single-word transfers and DMA transfers can be
performed through the EPBx buffers. DMA transfers are handled internally
by the DSP’s I/O processor, but single-word transfers must be handled by
the DSP core. For more information, see “Data Transfers Through the
EPBx Buffers” on page 7-71.
The DSP supports synchronous burst read transfers from the EPBx FIFOs,
or direct read from internal memory, as a slave. Burst write transfers are
not supported. Burst reads are supported as contiguous, aligned, 64-bit
data transfers up to a maximum length of four 64-bit transfers. The DSP
slave increments the address if the burst read access is from internal memory space only. The slave address increment function is only supported for
ADDR2-1 (similar to SBSRAMs).
To perform a burst read transfer from an EPBx buffer, the DSP master
issues a starting burst address pointing to one of the EPBx buffer addresses
in I/O processor control register space. The DSP slave does not increment
an EPBx burst read address, and the master DSP limits the burst transfer
length to the modulo4 address boundary restriction.
Bus Lock and Semaphores
Semaphores can be used in multiprocessor systems to allow the processors
to share resources such as memory or I/O. A semaphore is a flag that can
be read and written by any of the processors sharing the resource. The
value of the semaphore tells the processor when it can access the resource.
7-118
ADSP-21160 SHARC DSP Hardware Reference
External Port
Semaphores are also useful for synchronizing the tasks being performed by
different processors in a multiprocessing system.
With the use of its bus lock feature, the DSP has the ability to read and
modify a semaphore in a single indivisible operation—a key requirement
of multiprocessing systems.
Because both external memory and each DSP’s internal memory (and I/O
processor registers) are accessible by every other DSP, semaphores can be
located almost anywhere. Read-modify-write operations on semaphores
can be performed if all of the DSPs obey two simple rules:
• A DSP must not write to a semaphore unless it is the bus master.
This is especially important if the semaphore is located in the
DSP’s own internal memory or I/O processor registers.
• When attempting a read-modify-write operation on a semaphore,
the DSP must have bus mastership for the duration of the
operation.
Both rules apply when a DSP uses its bus lock feature, which retains its
mastership of the bus and prevents the other processors from simultaneously accessing the semaphore.
Bus lock is requested by setting the BUSLK bit in the MODE2 register. When
this happens, the DSP initiates the bus arbitration process in the usual
fashion, by asserting its BRx line. When it becomes bus master, it locks the
bus (for example, retains bus mastership) by keeping its BRx line asserted
even when it is not performing an external read or write. Host Bus
Request (HBR) is also ignored during a bus lock. When the BUSLK bit is
cleared, the DSP gives up the bus by deasserting its BRx line.
While the BUSLK bit is set, the DSP can determine if it has acquired bus
mastership by executing a conditional instruction with the Bus Master
(Bm) or Not Bus Master (Not Bm) condition codes, for example:
IF NOT BM JUMP(PC,0); /* wait for bus mastership */
ADSP-21160 SHARC DSP Hardware Reference
7-119
Multiprocessor (DSPs) Interface
If it has become the bus master, the DSP can proceed with the external
read or write. If not, it can clear its BUSLK bit and try again later.
A read-modify-write operation is accomplished with the following steps.
1. Request bus lock by setting the BUSLK bit in MODE2.
2. Wait for bus mastership to be acquired.
3. Wait until Direct Write Pending (DWPD) is zero.
4. Read the semaphore, test it, then write to it.
Locking the bus prevents other processors from writing to the semaphore
while the read-modify-write is occurring. After bus mastership is acquired,
the Direct Write Pending (DWPD) bit’s status in SYSTAT must be checked to
ensure that a semaphore write by another processor is not pending.
is reflective, located in the DSP’s internal mem Iforytheor semaphore
an I/O processor register, the processor must write to it only
when it has bus lock.
Bus lock can be used in combination with broadcast writes to implement
reflective semaphores in a multiprocessing system. The reflective semaphore should be located at the same address in internal memory or I/O
processor register of each DSP. To check the semaphore, each DSP simply
reads from its own internal memory. To modify the semaphore, a DSP
requests bus lock and then performs a broadcast write to the semaphore
address on every DSP, including itself. Before modifying the semaphore,
the DSP must re-check it to verify that another processor has not changed
it. With reflective semaphores, the external bus is used only for updating
the semaphore, not for reading it. This technique reduces bus traffic.
Interprocessor Messages and Vector Interrupts
The DSP bus master can communicate with slave DSPs by writing messages to their I/O processor registers. The MSGR0-7 registers are
7-120
ADSP-21160 SHARC DSP Hardware Reference
External Port
general-purpose registers which can be used for convenient message passing between DSPs. They are also useful for semaphores and resource
sharing between multiple DSPs. The MSGRx and VIRPT registers can be
used for message passing in the following ways.
• Message Passing. The master DSP can communicate with a slave
DSP by writing and/or reading any of the 8 message registers,
MSGR0-MSGR7, on the slave.
• Vector Interrupts. The master DSP can issue a vector interrupt to a
slave by writing the address of an interrupt service routine to the
slave’s VIRPT register. This causes an immediate high-priority interrupt on the slave which, when serviced, causes it to branch to the
specified service routine.
The MSGRx and VIRPT registers also support the host processor interface.
Because these registers may be shared resources within a single DSP, conflicts may occur—system software must prevent this. For further
discussion of I/O processor register access conflicts, see “I/O Processor
Registers” on page A-33
Message Passing (MSGRx)
There are three methods by which the DSP bus master can communicate
with a slave through the MSGRx message registers: 1) vector-interrupt
driven, 2) register handshake, and 3) register write-back. These techniques
are the same as for a host processor. For more information, see “Message
Passing (MSGRx)” on page 7-78.
Vector Interrupts (VIRPT)
Vector interrupts are used for interprocessor commands between two
DSPs or between a host and the DSP. When the external processor writes
an address to the DSP’s VIRPT register, a vector interrupt is caused. Vector
interrupts operate the same for host and multiprocessor systems. For more
information, see “Host Vector Interrupts (VIRPT)” on page 7-79.
ADSP-21160 SHARC DSP Hardware Reference
7-121
Multiprocessor (DSPs) Interface
Multiprocessor Interface Status
The SYSTAT register provides status information for host and multiprocessor systems. Table 7-14 shows the status bits in this register. For more
information on the SYSTAT register, see Table A-20 on page A-51.
Table 7-14. SYSTAT System Status Register
Bit #
Name
Definition
0
HSTM
Host Mastership – Indicates whether the host processor has been
granted control of the bus.
1=Host is bus master
0=Host is not bus master
1
BSYN
Bus Synchronization – Indicates when the DSP’s bus arbitration
logic is synchronized after reset. For more information, see “Bus
Synchronization After Reset” on page 7-110.
1=Bus arbitration logic is synchronized
0=Bus arbitration logic is not synchronized
[3:2]
[6:4]
reserved
CRBM
7
[10:8]
reserved
IDC
11
12
7-122
Current Bus Master – Indicates the ID of the DSP that is the current
bus master. If CRBM is equal to the ID of this DSP then it is the
current bus master. CRBM is only valid for ID 2-0 > 0 (greater than
zero). When ID 2-0 =000, CRBM is always 1.
ID Code – Indicates the ID 2-0 inputs of this DSP.
reserved
DWPD
Direct Write Pending – Indicates when a direct write to the DSP’s
internal memory is pending. The DWPD bit is cleared when the
direct write has been completed. (Direct writes may be delayed for
several cycles if DMA chaining is underway or if higher priority
DMA requests occur. Maximum delay is 12 cycles.)
1=Direct write pending
0=No direct write pending
ADSP-21160 SHARC DSP Hardware Reference
External Port
Table 7-14. SYSTAT System Status Register (Cont’d)
Bit #
Name
Definition
13
VIPD
Vector Interrupt Pending – Indicates that a pending vector interrupt
has not yet been serviced. The VIPD bit is set when the VIRPT register is written to and is cleared upon return from the interrupt service routine. The host processor (or other DSP) that issued the
vector interrupt should monitor this bit to determine when the service routine has been completed (and when a new vector interrupt
may be issued).
1=Vector interrupt pending
0=No vector interrupt pending
[15:14]
HPS
Host Packing Status – Indicates when host word packing is completed or, if not, what stage of the packing/unpacking process is taking place.
00=process complete
01=First stage of process
10=Second stage of process
[19:16]
CRAT
Processor Core Clock (CCLK)-to-CLKIN clock ratio, as determined
by the CLK_CFG0-3 inputs
[31:20]
reserved
ADSP-21160 SHARC DSP Hardware Reference
7-123
Multiprocessor (DSPs) Interface
7-124
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
8 LINK PORTS
This chapter describes the ADSP-21160 DSP’s link ports. The DSP has
six 8-bit wide link ports, which can connect to other DSPs’ or peripherals’
link ports.
Overview
The six ADSP-21160 DSP’s bidirectional link ports have eight data lines,
an acknowledge line, and a clock line. Link ports can operate at frequencies up to the same speed as the DSP’s internal clock, letting each port
transfer up to 8 bits of data per internal clock cycle.
Link ports also:
• Operate independently and simultaneously.
• Pack data into 32- or 48-bit words; this data can be directly read by
the DSP or DMA-transferred to or from on-chip memory.
• Are accessible by the external host processor, using direct reads and
writes.
• Have double-buffered transmit and receive data registers.
ADSP-21160 SHARC DSP Hardware Reference
8-1
Overview
• Include programmable clock/acknowledge controls for link port
transfers. Each link port has its own dedicated DMA channel.
• Provide high-speed, point-to-point data transfers to other DSP
processors, allowing differing types of interconnections between
multiple DSPs, including 1-, 2- and 3-dimensional arrays.
DSP’s link ports are logically (but not electrically)
 ADSP-21160
compatible with previous SHARC DSP (ADSP-2106x DSPs) link
ports. For more information, see “Link Data Path (and Compatibility) Modes” on page 8-7.
Table 8-1 on page 8-2 lists the pins associated with each link port. Each
link port consists of eight data lines (LxDAT7-0), a link clock line (LxCLK),
and a link acknowledge line (LxACK). The LxCLK line allows asynchronous
data transfers and the LxACK line provides handshaking. When configured
as a transmitter, the port drives both the data and LxCLK lines. When configured as a receiver, the port drives the LxACK line.
Figure 8-1 shows link port connections.
Table 8-1. Link Port Pins
Link Port Pin(s)
Link Port Function
LxDAT7-0
Link Port x Data
LxCLK
Link Port x Clock
LxACK
Link Port x Acknowledge
“x” denotes the link port number, 0-5
8-2
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
TRANSMITTER
LXDAT7-0
EACH
LINK
PORT
RECEIVER
8
LXDAT7-0
LXCLK
LXCLK
LXACK
LXACK
EACH
LINK
PORT
“X” DENOTES THE LINK PORT NUMBER, 0-5.
Figure 8-1. Link Port Pin Connections
Link Port To Link Buffer Assignment
There are six buffers, LBUF0-5, that buffer the data flow through the link
ports. These buffers are independent of the link ports and may be connected to any of the six link ports. The link ports receive and transmit
data on their LxDAT7-0 data pins. Any of the six link buffers may be
assigned to handle data for a particular link port. The data in the link buffers can be accessed with DMA or processor core control.
 “Link Port x” does not necessarily connect to “Link Buffer x.”
ADSP-21160 SHARC DSP Hardware Reference
8-3
Overview
Figure 8-2 shows a block diagram of the link ports and link buffers.
LINK BUFFERS 0-5
LBUF0
LBUF1
LBUF2
LBUF3
LDAT7-0
4/8
EXTERNAL PACKING REGISTER
4/8
LINK
PORTS
0-5
LBUF4
LBUF5
6/10
LP0
LP1
CROSS-BAR
CONNECTION
MX
32/48
LP2
LP3
LP4
LINK CLOCK
(1X, .5X, .33X, OR .25X)
LP5
INTERNAL REGISTER
LAR
LINK
ASSIGNMENT
REGISTER
32/48
32/48/64
DM DATA BUS
PM DATA BUS
I/O DATA BUS
Figure 8-2. Link Ports and Buffers
The Link Assignment Register (LAR) assigns the link buffer-to-port connections. Memory-to-memory transfers may be accomplished by assigning
the same link port to two buffers, setting up a loopback mode.
For details on the LAR register, see “Link Port Assignment Register (LAR)”
on page A-67.
 Assigning more than two buffers to one port will disable the port.
8-4
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
Link Port DMA Channels
DMA channels 4-9 support buffers 0-5. The buffer channel pairings are
listed in Table 8-2.
Table 8-2. DMA Channel/Link Buffer Pairing
DMA Channel #
Link Buffer Supported
DMA Channel 4
Link Buffer 0
DMA Channel 5
Link Buffer 1
DMA Channel 6
Link Buffer 2
DMA Channel 7
Link Buffer 3
DMA Channel 8
Link Buffer 4
DMA Channel 9
Link Buffer 5
For more information, see “Setting I/O Processor—LPort Modes” on
page 6-43.
Link Port Booting
Systems may boot the DSP through a link port. For more information, see
“Bootloading Through The Link Port” on page 6-87.
Setting Link Port Modes
The SYSCON, LCOM, LAR, and LCTLx registers control the link ports operating
modes for the I/O processor. Table A-17 on page A-45 lists all the bits in
SYSCON, Table A-24 on page A-65 lists all the bits in LCOM, Table A-25 on
page A-67 lists all the bits in LAR, and Table A-22 on page A-63 lists all
the bits in LCTLx.
ADSP-21160 SHARC DSP Hardware Reference
8-5
Setting Link Port Modes
The following bits control link port modes. Some other bits in the SYSCON,
LCOM, LAR, and LCTLx registers setup DMA and I/O processor related link
port features. For information on these features, see For more information, see “Setting I/O Processor—LPort Modes” on page 6-43..
• Link Buffer Mesh Multiprocessing. LCOM Bit 20 (LMSP) This bit
enables (if set, =1) or disable (if cleared, =0) mesh multiprocessing.
• Link Path (Mesh Multiprocessing) Delay. LCOM Bit 22-21 (LPATHD)
These bits apply change over delays when changing to the next
LPATH register as follows: 00=no additional delay, 01=1 additional
delay, 10=2 additional delays, 11=3 additional delays.
Link Port Clock Divisor. LCTL0 Bits 6-5, 16-15, and 26-25 and
LCTL1 Bits 6-5, 16-15, and 26-25 (LxCLKD) These bits select the
transfer clock divisor for the corresponding link buffer (LBUFx).
The transfer clock equals the DSP’s internal clock (CCLK) divided
by LxCLKD, where LxCLKD is: 01=1, 10=2, 11=3, or 00=4.
• Link Port pull-down Resistor Disable/Enable. LCTL0 Bits 8, 18,
and 28 and LCTL1 Bits 8, 18, and 28 (LxPDRDE) This bit disables (if
set, =1) or enables (if cleared, =0) the internal pull-down resistors
on the LxCLK, LxACK, and LxDAT7-0 pins of the corresponding link
port; this bit applies to the port, which is not necessarily the port
assigned to the corresponding link buffer (LBUFx).
link ports are bussed together and you have the Link
 IfPortmultiple
pull-down Resistor enabled on all the processors, you will
heavily load the line. Ensure that you have only one DSP enabling
this functionality.
Link Port Data Path Width. LCTL0 Bits 9, 19, and 29 and LCTL1
Bits 9, 19, and 29 (LxDPWID) This bit selects the link port data path
width (8-bit if set, =1) (4-bit if cleared, =0) for the corresponding
link buffer (LBUFx).
8-6
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
internal clock (
) is the
frequency multiplied
 byThea DSP’s
clock ratio (
). For more information, see the clock
CCLK
CLKIN
CLK_CFG3-0
ratio discussion on page 11-8.
link buffers are enabled or disabled, the I/O processor may
 When
generate unwanted interrupt service requests if Link Service
Request (LSRQ) interrupts are unmasked. To avoid unwanted interrupts, programs should mask the LSRQ interrupts while enabling or
disabling link buffers. For more information, see “Using Link Port
Interrupts” on page 8-12.
Link Data Path (and Compatibility) Modes
The link ports can transmit and received data using all 8 of the link port’s
data pins (LxDAT7-0) or the 4 lower data pins (LxDAT3-0). The LxDPWID bit
in the LCTLx register selects the link port data path width (8-bit if set, =1)
(4-bit if cleared, =0).
When
is cleared (4-bit data path), the ADSP-21160 DSP
 can
be connected to link ports of previous SHARC DSPs
LxDPWID
(ADSP-2106x DSP family). The link port receiver must run at the
same speed or faster than the transmitter. Connecting to an
ADSP-2106x DSP may require that the ADSP-21160 DSP be configured for 1/2 core clock rate operation. For more information, see
“Using Link Port Handshake Signals” on page 8-7.
Using Link Port Handshake Signals
The LxCLK and LxACK pins of each link port allow handshaking for asynchronous data communication between DSPs. Other devices that follow
the same protocol may also communicate with these link ports. The DSP
link ports are backward compatible with the SHARC link ports for basic
transfers, including LSRQ functions.
ADSP-21160 SHARC DSP Hardware Reference
8-7
Using Link Port Handshake Signals
A link-port-transmitted word consists of 4 bytes (for a 32-bit word) or 6
bytes (for a 48-bit word). The transmitter asserts the clock (LxCLK) high
with each new byte of data. The falling edge of LxCLK is used by the
receiver to latch the byte. The receiver asserts LxACK when it is ready to
accept another word in the buffer. The transmitter samples LxACK at the
beginning of each word transmission (that is, after every 4 or 6 bytes). If
LxACK is deasserted at that time, the transmitter does not transmit the new
word. For more information, see Figure 8-3 on page 8-8.
The transmitter leaves LxCLK high and continues to drive the first byte if
LxACK is deasserted. When LxACK is eventually asserted again, LxCLK goes
low and begins transmission of the next word. If the transmit buffer is
empty, LxCLK remains low until the buffer is refilled, regardless of the state
of LxACK.
LC LK STAYS HI GH A T BYTE 0 IF LA C K IS SAM PLED LOW ON PREVIOUS
LC LK R I SI NG EDGE— LC LK HIGH I N DIC A TES A STALL
LXC LK
M INIM UM LA CK SET-UP TIM E
LXAC K
LAC K WILL R EASSERT
AS SOON AS THE LIN K
BUFF ER IS "N OT F ULL"
LXA CK M AY
DEASSER T A FTER
BYTE 0
LXDA T7-0
BYTE 1
BY TE 2
BYTE 3
REC EIVER WI LL AC C EPT REM A INI N G BYTES IN THE
C URR ENT WO RD EV EN IF LA C K IS DEA SSER TED.
THE TRA NSM ITTER WILL NOT SEN D THE FO LLOW ING WO RD.
TRA N SM ITTER SA M PLES LA CK
HER E TO DETERM IN E W HETHER
TO TRA NSM IT N EXT WO RD
BYTE 0 (M SBS)
TR AN SMIT D ATA FO R N EXT
W ORD IS H ELD UNTIL
LAC K IS A SSER TED
Figure 8-3. Link Port Handshake Timing (32-Bit Mode)
8-8
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
The receive buffer may fill if a higher priority DMA, core I/O processor
register access, direct read, direct write or chain loading operation is
occurring. LxACK may deassert when it anticipates the buffer may fill.
LxACK is reasserted by the receiver as soon as the internal DMA grant signal has occurred, freeing a buffer location.
Data is latched in the receive buffer on the falling edge of LxCLK. The
receive operation is purely asynchronous and can occur at any frequency
up to the processor clock frequency. When a link port is not enabled,
LxDAT7-0, LxCLK and LxACK are three-stated. When a link port is enabled
to transmit, the data pins are driven with whatever data is in the output
buffer, LxCLK is driven high and LxACK is three-stated. When a port is
enabled to receive, the data pins and LxCLK are three-stated and LxACK is
driven high.
To allow a transmitter and a receiver to be enabled (assigned and link buffer enabled) at different times, LxACK, LxCLK, and LxDAT7-0 may be held
low with the 50K internal pull-down resistors if LxPDRDE is cleared when
the Link Port is disabled. If the transmitter is enabled before the receiver,
LxACK is low and the transmission is held off. If the receiver is enabled
before the transmitter, LxCLK is held low by the pull-down and the receiver
is held off. If many link ports are bused together, the systems may need to
enable only one of the internal resistors to pull down each bused pin, so
the bused lines are not pulled down too strongly or too heavily loaded.
, and
- should not be left unconnected unless
 external, pull-down
resistors are used.
LxACK LxCLK
LxDAT7 0
ADSP-21160 SHARC DSP Hardware Reference
8-9
Using Link Buffers
Using Link Buffers
Each link buffer consists of an external and an internal 48-bit register. For
more information, see Figure 8-2 on page 8-4. When transmitting, the
internal register is used to accept core data or DMA data from internal
memory. When receiving, the external register performs the packing and
unpacking for the link port, most significant nibble or byte first. These
two registers form a two-stage FIFO for the LBUFx buffer. Two writes (32or 48-bit) can occur to the register by the DMA or the core, before it signals a full condition. As each word is unpacked and transmitted, the next
location in the FIFO becomes available and a new DMA request is made.
If the register becomes empty, the LxCLK signal is deasserted. When transmitting, only the number of words written are transmitted.
Full/empty status for the link buffer FIFOs is given by the LxSTAT bits of
the LCOM register. This status is cleared for a link buffer when its LxEN
enable bit is cleared in the LCTLx register.
During receiving, the external buffer is used to pack the receive link port
data (most significant nibble or byte first) and pass it to the internal register before DMA-transferring it to internal memory. This buffer is a
two-deep FIFO. If the DSP’s DMA controller does not service it before
both locations are filled, the LxACK signal is deasserted.
The link buffer width may be selected to be either 32 or 48 bits. This
selection is made individually for each buffer with the LxEXT bits in the
LCTLx register. For 40-bit extended precision data or 48-bit instruction
transfers, the width must be set to 48 bits.
8-10
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
Core Processor Access To Link Buffers
In applications where the latency of link port DMA transfers to and from
internal memory is too long, or where a process is continuous and has no
block boundaries, the DSP processor core may read or write link buffers
directly using the full or empty status bit of the link buffer to automatically pace the operation. The full or empty status of a particular LBUFx
buffer can be determined by reading the LCOM control/status register.
DMA should be disabled when using this capability (LxDEN=0).
If a read is attempted from an empty receive buffer, the core stalls (hangs)
until the link port completes reception of a word. If a write is attempted
to a full transmit buffer, the core stalls until the external device accepts the
complete word. Up to four words (2 in the receiver and 2 in the transmitter) may be sent without a stall before the receiver core or DMA must read
a link buffer register.
support debugging buffer transfers, the DSP has a Buffer Hang
 ToDisable
( ) bit. When set (=1), this bit prevents the processor
BHD
core from detecting a buffer-related stall condition, permitting
debugging of this type of stall condition.
For more information, see the BHD discussion on page -18.
Host Processor Access To Link Buffers
The link buffers can also be accessed by the external host processor, using
direct reads and writes. When the host reads or writes to these buffers, the
word width is determined only by the host packing mode, as selected by
the HPM bits in the SYSCON register.
ADSP-21160 SHARC DSP Hardware Reference
8-11
Using Link Port DMA
Using Link Port DMA
DMA channels 4-9 support link buffers 0-5. A maskable interrupt is generated when the DMA block transfer has completed. For more
information on link port interrupts, see “Using Link Port Interrupts” on
page 8-12. For more information on link port DMA, see “Link Port
DMA” on page 6-80.
previous SHARC DSPs, there are no shared DMA channels
 Unlike
on the ADSP-2153x DSP. Each link port buffer has its own dedicated DMA channel.
In chained DMA operations, the DSP automatically sets up another DMA
transfer when the current DMA operation completes. The chain pointer
register (CPx) is used to point to the next set of buffer parameters stored in
memory. The DSP’s DMA controller automatically downloads these buffer parameters to set up the next DMA sequence. For information on
setting up DMA chaining, see “Chaining DMA Processes” on page 6-69.
Using Link Port Interrupts
Three types of interrupts are dedicated to the link ports:
The I/O processor generates a DMA channel interrupt when a DMA
block transfer through the link port with DMA enabled (LxDEN=1)
finishes.
• The I/O processor generates a DMA channel interrupt when DMA
for the link buffer channel is disabled (LxDEN=0) and the buffer is
not full (for transmit) or the buffer is not empty (for receive).
• The I/O processor generates a Link Services Request (LSRQ) interrupt when an external source accesses a disabled link port—
unassigned link port or assigned port with buffer disabled.
8-12
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
Registers control link port interrupt latching and masking. The LIRPTL
register is the individual link port interrupt latch/mask register, and the
IRPTL and IMASK registers control global link port DMA interrupt latching
and masking. For more information, see “Link Port Interrupt Register
(LIRPTL)” on page A-24, “Interrupt Latch Register (IRPTL)” on
page A-18, and “Interrupt Mask Register (IMASK)” on page A-23.
Link Port Interrupts With DMA Enabled
A link port interrupt is generated when the DMA operation is done—
when the block transfer has completed and the DMA count register is
zero.
One way programs can use this interrupt is to send additional control
information at the end of a block transfer. Because the receive DMA buffer is empty when the DMA block has completed, the external bus master
can send up to two additional words to the slave DSP’s buffer, which has
space for the two words. The slave’s same interrupt vector associated with
the completion of the Link Port DMA could then read the buffer and use
these control words to determine the next course of action.
Link Port Interrupts With DMA Disabled
If DMA is disabled for a link port buffer, then the buffer may be written
or read by the DSP core as a memory-mapped I/O processor register.
If the DMA is disabled but the associated link buffer is enabled, then a
maskable interrupt is generated whenever a receive buffer is not empty or
when a transmit buffer is not full. This interrupt is the same interrupt vector associated with the completion of the DMA block transfers.
The interrupt latch bit in LIRPTL may be unmasked by the corresponding
mask bit in the same register. When initially enabling the mask bit, the
corresponding latch bit in LIRPTL should be cleared first to clear out any
request that may have been inadvertently latched.
ADSP-21160 SHARC DSP Hardware Reference
8-13
Using Link Port Interrupts
The interrupt service routine should test the buffer status after each read
or write to check when the buffer is empty or full, in order to determine
when it should return from interrupt. This will reduce the number of
interrupts it must service.
Link Port Service Request Interrupts (LSRQ)
Link port service requests let a disabled (unassigned or assigned with buffer disabled) link port cause an interrupt when an external access is
attempted. The transmit and receive request status bits of the LSRQ register
let a DSP determine if another DSP is attempting to send or receive data
through a particular link port. This lets two processors communicate
without prior knowledge of the transfer direction, link port number, or
exactly when the transfer is to occur.
When LxACK or LxCLK is asserted externally, a link service request (LSR) is
generated in a disabled (unassigned or assigned with buffer disabled) link
port. LSRs are not generated for a link port that is disabled by loopback
mode. Each LSR is gated by mask bits before being latched in the LSRQ
register. The six possible receive LSRs and the six possible transmit LSRs
are gated by mask bits and then OR’ed together to generate the link service request interrupt. The LSRQ interrupt request may be masked by the
LSRQI mask bit of the IMASK register. When the mask bit is set, the interrupt is allowed to pass into the interrupt priority encoder. A diagram of
this logic appears in Figure 8-4 on page 8-14.
LSRQ
LXRRQ
LXTRQ
IMASK, LSRQI
LINK
SERVICE
REQUEST
INTERRUPT
LSR STATUS
LSR MASK
IRPT L, LSRQI
MODE1, I RPTEN
Figure 8-4. Logic For Link Port Interrupts
8-14
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
In Figure 8-4, the
and
inputs stand for status bits in
 the
register. For transmit requests,
=1 indicates the folLxTRQ
LxRRQ
LSRQ
LxTRQ
lowing status: LxACK=1, LxTM=1, and LxEN=0. For receive requests,
LxRRQ=1 indicates the following status: LxCLK=1, LxRM=1, and
LxEN=0.
The interrupt routine must read the LSRQ register to determine which link
port to service and whether it is a transmit or receive request. LSR interrupts have a latency of two cycles. Note that the link service request
interrupt is different from the link receive and transmit interrupt—this is
also true in IMASK.
The 32-bit LSRQ register holds the masked link status of each link port and
the corresponding interrupt mask bits. The link service request status of
the port is set whenever the port is not enabled and one of LxACK or LxCLK
is asserted high. The LSRQ status bits are read-only. Table A-26 on
page A-68 shows the individual bits of the LSRQ register.
which link port to service, programs can transfer
 To determine
to a register (in the register file) then use the leading 0s
LSRQ
Rx
detect instruction:
Rn=LEFTZ Rx
Here, Rn indicates which link port is active in order of priority.
If link service requests are in use, they should be masked out when the
assigned link buffers are being enabled, disabled, or when the link port is
being unassigned in LAR, otherwise spurious service requests may be
generated.
This need for masking is due to a delay before LxCLK or LxACK (if already
asserted) signals are pulled (if pull-downs enabled) or driven externally (if
pull-downs disabled) below logic threshold. During this delay, these signals are sampled asserted and generate an LSRQ.
ADSP-21160 SHARC DSP Hardware Reference
8-15
Detecting Errors On Link Transmissions
the possibility of spurious interrupts, programs should
 Tomaskavoid
the
interrupt or the appropriate request bit in the
LSRQ
LSRQ
register and allow an appropriate delay before unmasking. Alternatively, programs can mask the LSRQ interrupt and poll the
appropriate request status bit until it is cleared and then unmask
the interrupt.
Detecting Errors On Link Transmissions
Transmission errors on the link ports may be detected by reading the
LRERRx bits in the LCOM register. These bits reflects the status of each nibble or byte counter. The LRERRx bit is zero if the pack counter of the
corresponding link buffer is zero—a multiple of 8 or 12 nibbles or bytes
have been received. If LRERR is high when a transmission has completed,
then an error occurred during transmission.
DMA word count provides an exact count of the number of
 The
words to be transferred.
To allow checking of this status, the transmitter and receiver should follow a protocol such as the following:
• Transmitter Protocol—To make use of the LRERRx status, one additional dummy word should always be transmitted at the end of a
block transmission. The transmitter must then deselect the link
port and re-enable as a receiver to allow the receiver to send an
appropriate message back to the transmitter.
• Receiver Protocol—When the receiver has received the data block,
indicated by a the same interrupt vector associated with the completion of the Link Port DMA, it checks that it has received an
additional word in the link buffer and then reads the LRERR bit.
The receiver may then clear the link buffer (LxEN=0) and transmit
the appropriate message back to the transmitter on the same, or a
different, link port.
8-16
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
Using Token Passing With Link Ports
Two DSPs that communicate using a link port need to know which of
them is currently the transmitter and which is the receiver, otherwise they
might both try to transmit at the same time. Token passing is a protocol
that can help the DSPs alternate control. Figure 8-5 on page 8-18 shows a
Token Passing Flow Chart describing a software protocol for token passing through the link ports. A good example of this protocol is available in
the Engineer-to-Engineer Note EE-16: Using Token Passing to Control SHARC Link Port Bi-directional Communication on the Analog
Devices website.
In token passing, the token is a software flag that passes between the processors. At reset, the token (flag) is set to reside in the link port of one
device, making it the master and the transmitter. When a receiver link
port (slave) wants to become the master, it may assert its LxACK line
(request data) to get the master’s attention. The master knows, through
software protocol, whether it is supposed to respond with actual data or
whether it is being asked for the token.
ADSP-21160 SHARC DSP Hardware Reference
8-17
Using Token Passing With Link Ports
ORIGINAL SLAVE
ORIGINAL MASTER
• DMA TRANSFER COMPLETE
• LBUF DISABLED
• LSRQ INTERRUPT ENABLED
• DMA TRANSFER COMPLETE
• LBUF DISABLED
• LBUF RX NON-DMA ENABLED
• LACK ASSERTION CAUSES LSRQ INTERRUPT
• LBUF TX NON-DMA ENABLED
• SEND TRW 4 TIMES TO FILL LBUF FIFOS ON BOTH SIDES
• CHECK LCOM FOR SLAVE READ OF TRW
BEFORE ACCEPTANCE TEST
• READ LBUF
• TEST FOR TRW
• CHECK LCOM TO SEE IF SLAVE ACCEPTED
TOKEN BY EMPTYING FIFOS IN AN ALLOTTED
TIME PERIOD
• ACCEPT TOKEN BY EMPTYING LBUF FIFOS
THROUGH 3 MORE READS WITHIN THE
ALLOTTED TIME PERIOD
• SETUP LBUF FOR RX NON-DMA TO ACCEPT
DMA SIZE
• SETUP LBUF FOR RX DMA AND DMA COMPLETE IRQ
• DISABLE LBUF AND LSRQ INTERRUPT
• POLL LSRQ STATUS FOR LINK PORT TRANSMIT
REQUEST TO BE SURE THAT THE ORIGINAL
MASTER IS NOW A SLAVE
• LACK ASSERTION ASSURES THAT IT IS SAFE
TO BEGIN TRANSMITTING
• SETUP LBUF FOR TX NON-DMA TO SEND
DMA SIZE
• SETUP LBUF FOR TX DMA AND DMA
COMPLETE INTERRUPT
• DMA TRANSFER COMPLETE
• SETUP LBUF FOR TX NON-DMA
• DMA TRANSFER COMPLETE
• SETUP LBUF FOR RX NON-DMA
Figure 8-5. Token Passing Flow Chart
8-18
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
The token release word can be any user-defined value. Since both the
transmitter and receiver are expecting a code word, this need not be exclusive of normal data transmission.
If the master wishes to give up the token, it may send back a user-defined
token release word and thereafter clear its token flag. Simultaneously, the
slave examines the data sent back and if it is the token release word, the
slave will set its token, and can thereafter transmit. If the received data is
not the token release word, then the slave must assume the master was
beginning a new transmission.
Through software protocol, the master can also ask to receive data by
sending the token release word without the LxACK (data request) going low
first. Figure 8-5 shows a flow chart of the example code’s protocol.
To use the example, the example code is to be loaded on both the original
master and the original slave. The code is ID intelligent for multiprocessor
systems: ID1 is the original master (transmitter) and ID2 is original slave
(receiver). The master transmits a buffer via DMA through LPORT0 using
LBUF3 and the slave receives through LPORT0 using LBUF2. The slave then
requests the token by generating an LSRQ interrupt in the disabled link
port of the master (LPORT0). The master responds by sending the token
release word and waiting to see if it is accepted. The slave checks to see
that it is the token release word and accepts the token by emptying the
master’s link buffer FIFO within a predetermined amount of time.
If the token is accepted the slave becomes the master and transmits a buffer of data to the new slave. If the token is rejected, the master transmits a
second buffer. When complete, the original master will finish by setting
up LBUF2 to receive without DMA, and the original slave sets up LBUF3 to
transmit without DMA.
ADSP-21160 SHARC DSP Hardware Reference
8-19
Using Token Passing With Link Ports
The following is a list of the major areas of concern when a program
implements a software protocol scheme for token passing:
• The program must make sure that both link buffers are not enabled
to transmit at the same time. In the event that this is allowed, data
may be transmitted and lost due to the fact that neither link port is
driving LxACK.
• In the example, the LSRQ register status bits are polled to ensure
that the master becomes the slave before the slave becomes the master, avoiding the two transmitter conflict.
• The program must make sure that the link interrupt selection
matches the application. If a status detection scheme using the status bits of the LSRQ register is to be used, it is important to note the
following:
If a link port that is configured to receive is disabled while LxACK is
asserted, there is an RC delay before the 50K pull-down resistor
on LxACK (if enabled) can pull the value below logic threshold.
If a link port that is configured to transmit is disabled while LxCLK
is asserted, there is an RC delay before the 50K pull-down resistor on LxCLK (if enabled) can pull the value below logic threshold.
If the appropriate request status bit is unmasked in the LSRQ register (in this instance), then an LSR is latched and the LSRQ interrupt
may be serviced, even though unintended, if enabled.
• The program must make sure that synchronization is not disrupted
by unrelated influences at critical sections where timing control
loops are used to synchronize parallel code execution. Disabling of
nested interrupts is one techniques to control this.
8-20
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
Designing Link Port Systems
The DSPs link ports support I/O with peripherals and other DSP link
ports. While link ports require few connections, there are a number of
design issues that systems using these ports must accommodate.
Terminations For Link Transmission Lines
The link ports are designed to allow long distance connections to be made
between the driver and the receiver. This is possible because the links are
self-synchronizing—the clock and data are transmitted together. Only relative delay, not absolute delay between clock and data is of importance.
In addition, the LACK signal inhibits transmission of the next word, not of
the current nibble or byte. For example, the current word is always
allowed to complete transmission. This allows delays of 3 to 5 cycles for
the LxACK signal to reach the transmitter.
The links are designed to drive transmission lines with characteristic
impedances of 50 or greater. A higher transmission line impedance
reduces the on-chip effect of driver impedance variations, for distances
longer than about six inches. It is recommended that an external series termination resistor be used at each link port pin to absorb reflections from
the open circuit at the destination. The external resistor should be selected
such that its value (plus the internal resistance of the driver) be equal to
the characteristic impedance of the transmission line.
example, a system with a typical internal drive resistance of
 For
10 and a characteristic impedance of 50 should use a link port
pin resistor of 40.
ADSP-21160 SHARC DSP Hardware Reference
8-21
Designing Link Port Systems
Peripheral I/O Using Link Ports
The example shown in Figure 8-6 on page 8-23 shows how a multiprocessing system can use link ports to connect to local memories and I/O
devices. An ASIC implements the interface between the link port and
DRAM or an I/O device. This minimal hardware solution frees the DSP’s
external bus for other shared-bus communication. The DRAM and ASIC
may be implemented on a single 10-pin SIMM module.
Accesses to the DRAM over a link is most efficient under DMA control.
The ASIC receives DMA control information from the link port and sets
up the access to the DRAM. It unpacks 16-bit data words from the
DRAM or packs 8-bit bytes from the link. At the end of the DMA transfer, an interrupt lets the DSP send new control information to the ASIC.
The ASIC always reverts to receive mode at the end of a transfer. The
LxACK signal is deasserted by the ASIC whenever a page change, memory
refresh cycle, or any other access to the DRAM occurs.
Memory modules may be shared by multiple DSPs when the link port is
bused. Each link port supports 100 Mbyte per second access throughput
for either instructions or data. The ASIC is responsible for generating the
clock when transmitting to the DSP. The ASIC is also responsible for generating sequential DMA addresses based on a start address and word
count.
Data Flow Multiprocessing With Link Ports
Figure 8-7 on page 8-24 shows examples of different link port communications schemes.
8-22
ADSP-21160 SHARC DSP Hardware Reference
Link Ports
LINK
INTERFACE
ASIC
DRAM 0
20 MHZ CYCLE
16
ADSP-21160
L0DAT7-0
L0CLK
L0ACK
10
I/O DEVICE
EXTERNAL
PORT
DATA
LINK BUS 0
LINK PORT 0
ADDRESS &
CONTROL
LINK PORT 1
EXTERNAL
MEMORY
DRAM 1
LINK
INTERFACE
ASIC
20 MHZ CYCLE
16
L1DAT7-0
L1CLK
L1ACK
10
LINK BUS 1
CLK
ADSP-21160
EXTERNAL
PORT
DATA
LINK PORT 0
ADDRESS &
CONTROL
DMA DEVICE
HOST
LINK PORT 1
CLK
Figure 8-6. Local DRAM With Link Ports
ADSP-21160 SHARC DSP Hardware Reference
8-23
Designing Link Port Systems
For more information on the multiprocessor interface, see “Multiprocessing System Architectures” on page 7-92.
LINK
PORTS
LINK
PORTS
LINK
PORTS
LINK
PORTS
LINK
PORTS
LINK
PORTS
ADSP-21160
ADSP-21160
ADSP-21160
ADSP-21160
ADSP-21160
ADSP-21160
EXTERNAL
PORT
EXTERNAL
PORT
EXTERNAL
PORT
DATAFLOW
SHARC CLUSTER
LINK
PORTS
LINK
PORTS
LINK
PORTS
ADSP-21160
ADSP-21160
ADSP-21160
EXTERNAL
PORT
EXTERNAL
PORT
EXTERNAL
PORT
ADSP-21160
ADSP-21160
ADSP-21160
LINK
PORTS
LINK
PORTS
LINK
PORTS
LINK
PORTS
LINK
PORTS
LINK
PORTS
ADSP-21160
ADSP-21160
ADSP-21160
EXPANDING CLUSTERS
2D MESH
Figure 8-7. Link Port Communication Examples
8-24
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
9 SERIAL PORTS
This chapter describes ADSP-21160 DSP’s serial ports.
Overview
The DSP has two independent, synchronous serial ports, SPORT0 and
SPORT1, that provide an I/O interface to a wide variety of peripheral
devices. Each serial port has its own set of control registers and data buffers. With a range of clock and frame synchronization options, the
SPORTs allow a variety of serial communication protocols and provide a
glueless hardware interface to many industry-standard data converters and
CODECs.because channels 0 through
The serial ports can operate at 1/2 the full clock rate of the processor, providing each with a maximum data rate of n/2 Mbit/s, where n equals the
processor clock frequency. Independent transmit and receive functions
provide greater flexibility for serial communications. Serial port data can
be automatically transferred to and from on-chip memory using DMA
block transfers. Each of the serial ports offers a TDM (time division multiplexed) multichannel mode.
Serial port clocks and frame syncs can be internally generated by the DSP
or received from an external source. The serial ports can operate with little-endian or big-endian transmission formats, with word lengths
ADSP-21160 SHARC DSP Hardware Reference
9-1
Overview
selectable from 3 to 32 bits. They offer selectable synchronization and
transmit modes and optional -law or A-law companding in hardware.
The serial ports offer the following features and capabilities:
• Provides independent transmit and receive functions
• Transfers data words up to 32 bits in length, either MSB-first or
LSB-first
• Double-buffers data—both receive and transmit functions have a
data buffer register and a shift register—the double-buffering provides additional time to service the SPORT
• Compands (compression/decompression) A-law and -law hardware companding on transmitted and received words.
• Internally generates serial clock and frame sync signals—in a wide
range of frequencies—or accepts clock and frame synch input from
an external source
• Performs interrupt-driven, single-word transfers to and from
on-chip memory controlled by the DSP core
• Executes DMA transfers to and from on-chip memory—each
SPORT can automatically receive and transmit an entire block of
data
• Permits chaining of DMA operations for multiple data blocks
• Has a multichannel mode for TDM interfaces—each SPORT can
receive and transmit data selectively from channels of a time-division-multiplexed serial bitstream—this mode can be useful for T1
interfaces
9-2
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
Table 9-1 shows the pins of each serial port:
Table 9-1. Serial Port Pins
SPORT0 Pins
SPORT1 Pins
Description
DT0
DT1
Transmit Data
TCLK0
TCLK1
Transmit Clock
TFS0
TFS1
Transmit Frame Sync
DR0
DR1
Receive Data
RCLK0
RCLK1
Receive Clock
RFS0
RFS1
Receive Frame Sync
A serial port receives serial data on its DR input and transmits serial data on
its DT output. It can receive and transmit simultaneously for full duplex
operation.
Serial communications are synchronized to a clock signal—every data bit
must be accompanied by a clock pulse. Each serial port can generate or
receive its own transmit clock signal (TCLK) and receive clock signal (RCLK).
Internally-generated serial clock frequencies are configured in the TDIVx
and RDIVx registers.
In addition to the serial clock signal, data may be signalled by a frame synchronization signal. The framing signal can occur either at the beginning
of an individual word or at the beginning of a block of words. The configuration of frame synch signals depends upon the type of serial device
connected to the DSP. Each serial port can generate or receive its own
transmit frame sync signal (TFS) and receive frame sync signal (RFS). Internally-generated frame sync frequencies are configured in the TDIVx and
RDIVx registers.
Figure 9-1 shows a block diagram of a serial port. Data to be transmitted
is written to the TX buffer. The data is (optionally) compressed in hardware, then automatically transferred to the transmit shift register. The
ADSP-21160 SHARC DSP Hardware Reference
9-3
Overview
data in the shift register is then shifted out on the SPORT’s DT pin, synchronous to the TCLK transmit clock. If framing signals are used, the TFS
signal indicates the start of the serial word transmission. The DT pin is
always driven (for example, three-stated) if the serial port is enabled
(SPEN=1 in the STCTLx control register), unless it is in multichannel mode
and an inactive time slot occurs. For more information, see “Multichannel
Operation” on page 9-28.
DM DAT A B US
P M DAT A B US
I/O DAT A B US
32
32
32
T Xx
T R ANS MIT B UF F E R
R Xx
R E CE IVE DAT A B U F F E R
HAR DWAR E COMP ANDING
(COMP R E S S ION)
HAR DWAR E COMP ANDING
(COMPR E S S ION)
32
32
S E R IAL P OR T
CONT R OL
T R ANS MIT S HIF T R E GIS T E R
DT x
T CL K x
T FS x
RF Sx
R E CE IVE S HIF T R E GIS T E R
R CL K x
DR x
Figure 9-1. Serial Port Block Diagram
The receive portion of the SPORT shifts in data from the DR pin, synchronous to the RCLK receive clock. If framing signals are used, the RFS signal
indicates the beginning of the serial word being received. When an entire
word is shifted in, the data is (optionally) expanded, then automatically
transferred to the RX buffer.
9-4
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
DSP SPORTs are not UARTs and cannot be used to commu The
nicate with an RS-232 device or any other asynchronous
communications protocol. One way to implement RS-232-compatible communications with the DSP is to use two of the FLAG
pins as asynchronous data receive and transmit signals. For an
example of how to do this, see Chapter 11 “Software UART” in the
Digital Signal Processing Applications Using The ADSP-2100 Family, Volume 2.
SPORT Interrupts
Each serial port has a transmit DMA interrupt and a receive DMA interrupt. When serial port DMA is not enabled, interrupts occur based on the
SPORT transmit or receive FIFO status. If on the transmit side the FIFO
is empty or on the receive side the FIFO is full, interrupts are generated.
The priority of the serial port interrupts is shown in Table 9-2.
Table 9-2. SPORT Interrupts
Interrupt Name
Interrupt
SPR0I
SPORT0 Receive DMA Channel Highest Priority
SPR1I
SPORT1 Receive DMA Channel
SPT0I
SPORT0 Transmit DMA Channel
SPT1I
SPORT1 Transmit DMA Channel Lowest Priority
The interrupt names are defined in the def21160.h include file supplied with the ADSP-21xxx
DSP Development Software.
SPORT Interrupts occur on the second system clock (
 the
last bit of the serial word is latched in or driven out.
CLKIN)
ADSP-21160 SHARC DSP Hardware Reference
after
9-5
Overview
SPORT Reset
There are two ways to reset the serial ports: a hardware reset using the
RESET pin of the processor, and a software reset accomplished by clearing
the serial port’s enable bit (SPEN) in the STCTLx and SRCTLx control registers. Each method has a different effect on the serial port.
A hardware reset disables the serial ports by clearing the STCTLx and
control registers (including the SPEN enable bits) and the TDIVx and
RDIVx frame sync divisor registers. Any ongoing operations are aborted.
SRCTLx
A software reset of the SPEN enable bit(s) disables the serial port(s) and
aborts any ongoing operations. Status bits are also cleared. The serial ports
are ready to start transmitting or receiving data two CLKIN cycles after they
are enabled (in the STCTLx or SRCTLx control register). No serial clocks are
be lost from this point on.
only difference between the soft (setting the
bit in
 The
register and hard reset (
pin) is that the external bus arbitraSRST
SYSCON
RESET
tion does not get affected by a soft reset. That is, there is no bus
synchronization at soft reset. The PLL also does not get reset at soft
reset.
9-6
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
Setting Serial Port Modes
The registers used to control and configure the serial ports are part of the
IOP register set. Each SPORT has its own set of the control registers and
data buffers, as shown in Table 9-3.
Table 9-3. SPORT Registers
Register Name*
Function
STCTLx
SPORT Transmit Control Register
TXx
Transmit Data Buffer
TDIVx
Transmit Clock and Frame Sync Divisors
MTCSx
Multichannel Transmit Select
MTCCSx
Multichannel Transmit Compand Select
SRCTLx
SPORT Receive Control Register
RXx
Receive Data Buffer
RDIVx
Receive Clock and Frame Sync Divisors
MRCSx
Multichannel Receive Select
MRCCSx
Multichannel Receive Compand Select
SPATHx
SPORT Path Length (for mesh multiprocessing)
KEYWDx
SPORT Receive Comparison
KEYMASKx
SPORT Receive Comparison Mask
An asterisk (*) indicates x = 0, 1.
These control registers are describe in detail in the following sections:
• “SPORT Serial Transmit Control Registers (STCTLx)” on
page A-71
• “SPORT Serial Receive Control Registers (SRCTLx)” on
page A-73
ADSP-21160 SHARC DSP Hardware Reference
9-7
Setting Serial Port Modes
• “SPORT Transmit Buffer Registers (TXx)” on page A-73
• “SPORT Receive Buffer Registers (RXx)” on page A-76
• “SPORT Transmit Divisor Registers (TDIVx)” on page A-76
• “SPORT Transmit Count Registers (TCNTx)” on page A-77
• “SPORT Receive Divisor Registers (RDIVx)” on page A-77
• “SPORT Receive Count Registers (RCNTx)” on page A-78
• “SPORT Transmit Select Registers (MTCSx)” on page A-78
• “SPORT Receive Select Registers (MRCSx)” on page A-78
• “SPORT Transmit Compand Registers (MTCCSx)” on page A-79
• “SPORT Receive Compand Register (MRCCSx)” on page A-79
• “SPORT Receive Comparison and Mask Registers (KEYWDx and
KEYMASKx)” on page A-80
• “SPORT Serial Path Length Registers (SPATHx)” on page A-80
These sections show the memory-mapped address and reset initialization
value of each SPORT control register. All of the registers are 32 bits wide.
The SPORT control registers are programmed by writing to the appropriate address in memory. The symbolic names of the registers and individual
control bits can be used in DSP programs—the #define definitions for
these symbols are contained in the file def21160.h which is provided in
the INCLUDE directory of the ADSP-21xxx DSP Development Software.
The def21160.h file is shown in the Control/Status Registers appendix of
this manual. All control and status bits in the SPORT registers are active
high unless otherwise noted.
9-8
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
Because the SPORT registers are memory-mapped they cannot be written
with data coming directly from memory. They must instead be written
from (or read into) DSP core registers, usually one of the general-purpose
universal registers of the register file (R15–R0). For instance, the SPORT
control registers can also be written or read by external devices (for example, another DSP or a host processor) to set up a serial port DMA
operation.
Transmit and Receive Control Registers
(STCTL, SRCTL)
The main control registers for each serial port are the transmit control register, STCTLx, and the receive control register, SRCTLx. These registers are
defined in Table A-27 on page A-71 and Table A-28 on page A-74. When
changing operating modes, a serial port control register should be cleared
(for example, written with all zeros) before the new mode is written to the
register.
The Transmit Underflow Status bit (TUVF) is set whenever the TFS signal
occurs (from either external or internal source) while the TX buffer is
empty. The internally generated TFS may be suppressed whenever TX is
empty by clearing the DITFS control bit (DITFS=0).
When DITFS=0, the default, the transmit frame sync signal (TFS) is dependent upon new data being present in the TX buffer—the TFS signal is only
generated for new data. Setting DITFS to 1 selects data-independent frame
syncs. This causes the TFS signal to be generated whether or not new data
is present, transmitting the contents of the TX buffer regardless. Serial port
DMA typically keeps the TX buffer full, and when the DMA operation is
complete the last word in TX is continuously transmitted.
The TXS status bits indicate whether the TX buffer is full (11), empty (00),
or partially full (10). To test for space in TX, test for TXS0 (bit 30) equal to
zero. To test for the presence of any data in TX, test for TXS1 (bit 31) equal
to one.
ADSP-21160 SHARC DSP Hardware Reference
9-9
Setting Serial Port Modes
The SRCTLx and STCTLx registers control the serial ports operating modes
for the I/O processor. Table A-28 on page A-74 lists all the bits in SRCTLx
and Table A-27 on page A-71 lists all the bits in STCTLx.
The following bits control serial port modes. Some other bits in the
SRCTLx and STCTLx registers are used to setup DMA and I/O processor
related serial port features. For information on these features, see “Setting
I/O Processor—SPort Modes” on page 6-49.
• Internal Transmit Clock Select. SRCTLx Bit 10 (ICLK) This bit
selects the internal receive clock (if set, =1) or external receive clock
(if cleared, =0).
• Clock Rising Edge Select. SRCTLx Bit 12 (CKRE) This bit select
whether the serial port uses the rising edge (if set, =1) or falling
edge (if cleared, =0) of the clock signal for sampling data and the
frame sync.
• Receive Frame Sync Required Select. SRCTLx Bit 13 (RFSR) This bit
selects whether the serial port requires (if set, =1) or does not
require (if cleared, =0) a receive frame synch.
• Internal Receive Frame Sync Select. SRCTLx Bit 14 (IRFS) This bit
selects whether the serial port uses an internal RFS (if set, =1) or
uses an external RFS (if cleared, =0).
• Data Independent Receive Frame Sync Select. SRCTLx Bit 15
(DIRFS) This bit selects whether the serial port uses a data-independent RFS (synch at selected interval, if set, =1) or uses a
data-dependent RFS (synch when data in RX, if cleared, =0).
• Active Low Receive Frame Synch Select. SRCTLx Bit 16 (LRFS) This
bit selects an active low RFS (if set, =1) or active high RFS (if
cleared, =0).
9-10
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
• Late Receive Frame Sync Select. SRCTLx Bit 17 (LAFS) This bit
selects a late RFS (RFS during first bit, if set, =1) or an early RFS (RFS
before first bit, if cleared, =0). This bit must be cleared for multichannel operation.
• Serial Port Loopback Enable. SRCTLx Bit 22 (SPL) This bit enables
(if set, =1) or disables (if cleared, =0) serial port loopback mode.
This bit must be cleared for multichannel operation.
• Multichannel Enable. SRCTLx Bit 23 (MCE) This bit enables
(if set, =1) or disables (if cleared, =0) multichannel serial port
mode.
• Number of Multi Channels (–1) Select. SRCTLx Bits 28-24 (NCHN)
These bits select the number of channels (–1) for a multichannel
serial port. The number of channels can be from 1 (NCHN=0) to 32
(NCHN=31).
• Transmit Frame Sync Required Select. STCTLx Bit 13 (TFSR) This
bit selects whether the serial port requires (if set, =1) or does not
require (if cleared, =0) a transfer frame synch.
• Internal Transmit Frame Sync Select. STCTLx Bit 14 (ITFS) This
bit selects whether the serial port uses an internal TFS (if set, =1) or
uses an external TFS (if cleared, =0).
• Data Independent Transmit Frame Sync Select. STCTLx 15 (DITFS)
This bit selects whether the serial port uses a data-independent TFS
(synch at selected interval, if set, =1) or uses a data-dependent TFS
(synch when data in TX, if cleared, =0).
• Active Low Transmit Frame Synch Select. STCTLx Bit 16 (LTFS)
This bit selects an active low TFS (if set, =1) or active high TFS (if
cleared, =0).
ADSP-21160 SHARC DSP Hardware Reference
9-11
Setting Serial Port Modes
• Late Transmit Frame Sync Select. STCTLx Bit 17 (LAFS) This bit
selects a late TFS (TFS during first bit, if set, =1) or an early TFS (TFS
before first bit, if cleared, =0).
• Multichannel Transmit Frame Sync Delay Select. STCTLx Bits
23-20 (MFD) These bits select the delay in serial clock cycles
between the TFS and the first data bit. When MFD=0, the TFS and
first data bit are concurrent. The maximum value is MFD=16.
• Current Channel Selected (read-only). STCTLx Bits 28-24 (CHNL)
These bits indicate which channel the DSP has selected for the
serial port’s transmission in multichannel mode.
Register Writes and Effect Latency
SPORT register writes are internally completed at the end of the same
CLKIN cycle in which they occur. The register is read back the newly written value on the very next cycle. When a read of one of the STCTLx or
SRCTLx control registers is immediately followed by a write to that register,
the write may take two cycles to complete.
After a write to a SPORT register, control and mode bit changes generally
take effect in the second CLKIN cycle after the write is completed. The
serial ports are ready to start transmitting or receiving two CLKIN cycles
after they are enabled (in the STCTLx or SRCTLx control register). No serial
clocks are lost from this point on.
Transmit and Receive Data Buffers (TX, RX)
and TX1 are the transmit data buffers for SPORT0 and SPORT1.
They are 32-bit buffers which must be loaded with the data to be transmitted; the data is loaded either by the DMA controller or by the program
running on the DSP core. RX0 and RX1 are the receive data buffers for
SPORT0 and SPORT1. They are 32-bit buffers which are automatically
loaded from the receive shifter when a complete word has been received.
TX0
9-12
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
Word lengths of less than 32 bits are right-justified in the receive and
transmit buffers.
The TX buffers act like a two-location FIFO because they have a data register plus an output shift register as shown in Figure 9-1 on page 9-4. Two
32-bit words may be stored in TX at any one time. When the TX buffer is
loaded and any previous word has been transmitted, the buffer contents
are automatically loaded into the output shifter. An interrupt is generated
when the output shifter has been loaded, signifying that the TX buffer is
ready to accept the next word (for example, the TX buffer is “not full”).
This interrupt does not occur if serial port DMA is enabled or if the corresponding mask bit in the IMASK register is set.
The transmit underflow status bit (TUVF) is set in the transmit control register when a transmit frame synch occurs and no new data has been loaded
into TX. The TUVF status bit is “sticky” and is only cleared by disabling the
serial port.
The RX buffers act like a three-location FIFO because they have two data
registers plus an input shift register. Two complete 32-bit words can be
stored in RX while a third word is being shifted in. The third word overwrites the second if the first word has not been read out (by the DSP core
or the DMA controller). When this happens, the receive overflow status
bit (ROVF) is set in the receive control register. Almost three complete
words can be received without the RX buffer being read before overflow
occurs. The overflow status is generated on the last bit of third word. The
ROVF status bit is “sticky” and is only cleared by disabling the serial port.
An interrupt is generated when the RX buffer has been loaded with a
received word (for example, the RX buffer is “not empty”). This interrupt
is masked out if serial port DMA is enabled or if the corresponding bit in
the IMASK register is set.
If your DSP program causes the core processor to attempt a read from an
empty RX buffer or a write to a full TX buffer, the access is delayed until the
buffer is accessed by the external I/O device. (This delay is called a core
ADSP-21160 SHARC DSP Hardware Reference
9-13
Setting Serial Port Modes
processor hang.) If it is not known whether the core processor can access
the RX or TX buffer without a hang, the buffer’s full or empty status should
be read first (in STCTLx or SRCTLx) to determine if the access can be made.
support debugging buffer transfers, the DSP has a Buffer Hang
 ToDisable
( ) bit. When set (=1), this bit prevents the processor
BHD
core from detecting a buffer-related stall condition, permitting
debugging of this type of stall condition. For more information, see
the BHD discussion on page -18.
The status bits in STCTLx and SRCTLx are updated during reads and writes
from the core processor even when the serial port is disabled. The serial
port should be disabled when writing to the RX buffer or reading from the
TX buffer.
Clock and Frame Sync Frequencies (TDIV, RDIV)
The TDIVx and RDIVx registers contain divisor values which determine the
frequencies for internally generated clocks and frame syncs. These registers
are defined in “SPORT Transmit Divisor Registers (TDIVx)” on
page A-76 and “SPORT Receive Divisor Registers (RDIVx)” on
page A-77.
f CCLK
f RCLK = -------------------------------------------2  RCLKDIV + 1 
f CCLK
f TCLK = -------------------------------------------2  TCLKDIV + 1 
The maximum serial clock frequency is equal to 1/2 the DSP’s internal
clock (CCLK) frequency, which occurs when xCLKDIV is set to zero. Use the
9-14
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
following equation to determine the value of xCLKDIV to use, given the
CCLK frequency and desired serial clock frequency:
f CCLK
RCLKDIV = --------------------- – 1
2  f RCLK 
f CCLK
TCLKDIV = --------------------- – 1
2  f TCLK 
internal clock (
) is the
frequency multiplied
 byThea DSP’s
clock ratio (
). For more information, see the clock
CCLK
CLKIN
CLK_CFG3-0
ratio discussion on page 11-8.
and RFSDIV specify how many transmit or receive clock cycles are
counted before generating a TFS or RFS pulse (when the frame synch is
internally generated). In this way a frame sync can be used to initiate periodic transfers. The counting of serial clock cycles applies to either
internally or externally generated serial clocks.
TFSDIV
The formula for the number of cycles between frame synch pulses is:
# of serial clocks between frame syncs = xFSDIV + 1
ADSP-21160 SHARC DSP Hardware Reference
9-15
Setting Serial Port Modes
Use the following equations to determine the value of xFSDIV to use, given
the serial clock frequency and desired frame sync frequency:
f TCLK
TFSDIV = ------------- – 1
f TFS
f RCLK
RFSDIV = ------------- – 1
f RFS
The frame sync is continuously active if xFSDIV=0. The value of xFSDIV
should not be less than the serial word length minus one (the value of the
SLEN field in the transmit or receive control register), as this may cause an
external device to abort the current operation or cause other unpredictable
results. If the serial port is not being used, the xFSDIV divisor can be used
as a counter for dividing an external clock or for generating a periodic
pulse or periodic interrupt. The serial port must be enabled for this mode
of operation to work.
should be exercised when operating with externally gener Caution
ated transmit clocks near the frequency of 1/2 the DSP’s internal
clock. There is a delay between when the clock arrives at the TCLKx
pin and when data is output—this delay may limit the receiver’s
speed of operation. Refer to the data sheet for exact timing specifications. For reliable operation, it is recommended that full-speed
serial clocks only be used when receiving with an externally generated clock and externally generated frame sync (ICLK=0, IRFS=0).
Externally-generated late transmit frame syncs also experience a delay
from when they arrive to when data is output—this can also limit the
maximum serial clock speed. Refer to the data sheet for exact timing
specifications.
9-16
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
The serial ports handle word lengths of 3 to 32 bits, but transmitting or
receiving words smaller than 7 bits at 1/2 the full clock rate of the DSP
may cause incorrect operation when DMA chaining is enabled. Chaining
disables the DSP’s internal I/O bus for several cycles while the new TCB
parameters are being loaded. Receive data may be lost (for example, overwritten) during this period.
Data Word Formats
The format of the data words transmitted over the serial ports is configured by the DTYPE, SENDN, SLEN, and PACK bits of the STCTLx and SRCTLx
control registers.
Word Length
The serial ports handle word lengths of 3 to 32 bits. The word length is
configured in the 5-bit SLEN field in the STCTLx and SRCTLx control registers. The value of SLEN is equal to the word length minus one:
SLEN = Serial Word Length – 1
The SLEN value should not be set to zero or one. Words smaller than 32
bits are right-justified in the RX and TX buffers, residing in the least significant bit positions.
Transmitting or receiving words smaller than 7 bits at 1/2 the full clock
rate of the DSP may cause incorrect operation when DMA chaining is
enabled. Chaining disables the DSP’s internal I/O bus for several cycles
while the new TCB parameters are being loaded. Receive data may be lost
(for example, overwritten) during this period.
Endian Format
Endian format determines whether the serial word is transmitted
MSB-first or LSB-first. Endian format is selected by the SENDN bit in the
STCTLx and SRCTLx control registers. When SENDN=0, serial words are
ADSP-21160 SHARC DSP Hardware Reference
9-17
Setting Serial Port Modes
transmitted (or received) MSB-first. When SENDN=1, serial words are
transmitted (or received) LSB-first.
Data Packing and Unpacking
Received data words of 16 bits or less may be packed into 32-bit words,
and 32-bit words being transmitted may be unpacked into 16-bit words.
Word packing and unpacking is selected by the PACK bit in the SRCTLx and
STCTLx control registers.
When PACK=1 in the receive control register (SRCTLx), two successive
words received are packed into a single 32-bit word.
When PACK=1 in the transmit control register (STCTLx), each 32-bit word
is unpacked and transmitted as two 16-bit words.
The first 16-bit (or smaller) word is right-justified in bits 15-0 of the
packed word, and the second 16-bit (or smaller) word is right-justified in
bits 31-16. This applies for both receive (packing) and transmit (unpacking) operations. Companding may be used when word packing or
unpacking is being used.
When serial port data packing is enabled, the transmit and receive interrupts are generated for the 32-bit packed words, not for each 16-bit word.
16-bit received data is packed into 32-bit words and stored
 When
in normal word space in DSP internal memory, the 16-bit words
can be read or written with short word space addresses.
9-18
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
Data Type
The DTYPE field of the STCTLx and SRCTLx control registers specifies one of
four data formats (for non-multichannel operation):
Table 9-4. DTYPE and Data Formatting (non-multichannel)
DTYPE
Data Formatting
00
Right-justify, zero-fill unused MSBs
01
Right-justify, sign-extend into unused MSBs
10
Compand using -law
11
Compand using A-law
These formats are applied to serial data words loaded into the RX and TX
buffers. TX data words are not actually zero-filled or sign-extended,
because only the significant bits are transmitted.
For multichannel operation, the companding selection and MSB-fill selection is independent:
Table 9-5. DTYPE and Data Formatting (multichannel)
DTYPE
Data Formatting
x0
Right-justify, zero-fill unused MSBs
x1
Right-justify, sign-extend into unused MSBs
0x
Compand using -law
1x
Compand using A-law
Linear transfers occur if the channel is active but companding is not
selected for that channel. Companded transfers occur if the channel is
active and companding is selected for that channel. The multichannel
compand select registers, MTCCSx and MRCCSx, are used to specify which
ADSP-21160 SHARC DSP Hardware Reference
9-19
Setting Serial Port Modes
transmit and receive channels are companded. For more information, see
“Channel Selection Registers” on page 9-32.
Transmit sign extension is selected by bit 0 of DTYPE in the STCTLx register
and is common to all transmit channels. Receive sign extension is selected
by bit 0 of DTYPE in the SRCTLx register and is common to all receive channels. If bit 0 of DTYPE is set, sign extension occurs on selected channels that
do not have companding selected. If this bit is not set, the word
contains 0s in the MSBs.
Companding
Companding (compressing/expanding) is the process of logarithmically
encoding and decoding data to minimize the number of bits that must be
sent. The DSP serial ports support the two most widely used companding
algorithms, A-law and -law, performed according to the CCITT G.711
specification. The type of companding can be selected independently for
each SPORT. Companding is selected by the DTYPE field of the STCTLx
and SRCTLx control registers.
When companding is enabled, the data in the RX0 or RX1 buffer is the
right-justified, sign-extended expanded value of the eight LSBs received. A
write to TX0 or TX1 causes the 32-bit value to be compressed to eight LSBs
(sign-extended to the width of the transmit word) before it is transmitted.
If the 32-bit value is greater than the 13-bit A-law or 14-bit -law maximum, it is automatically compressed to the maximum value.
Because the values in the TX and RX buffers are actually companded
in-place, the companding hardware can be used without transmitting (or
receiving) any data, for example during testing or debugging. This operation requires a single cycle of overhead, as described below. For
companding to execute properly, program the SPORT registers prior to
loading data values into the SPORT buffers.
9-20
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
To compand data in-place, without transmitting, use the following
sequence of operations:
1. Enable companding in the DTYPE field of the STCTLx transmit control register.
2. Write a 32-bit data word to TX. (The companding is calculated in
this cycle.)
3. Wait one cycle. A NOP instruction can be used to do this; if a NOP is
not inserted, the DSP core is held off for one cycle anyway. This
allows the serial port companding hardware to reload TX with the
companded value.
4. Read the 8-bit companded value from TX.
To expand data in-place, the same sequence of operations is used but with
RX rather than TX. When expanding data in this way, be sure that the serial
word length (SLEN) is set appropriately in the SRCTLx control register.
With companding enabled, interfacing the DSP serial port to a codec
requires little additional programming effort. If companding is not
selected, there are two formats available for received data words of fewer
than 32 bits: one that fills unused MSBs with zeros, and another that
sign-extends the MSB into the unused bits. For more information, see
“Data Type” on page 9-19.
Clock Signal Options
Each serial port has a transmit clock signal (TCLKx) and a receive clock signal (RCLKx). The clock signals are configured by the ICLK and CKRE bits of
the STCTLx and SRCTLx control registers. Serial clock frequency is configured in the TDIVx and RDIVx registers.
receive clock pin may be tied to the transmit clock if a single
 The
clock is desired for both input and output.
ADSP-21160 SHARC DSP Hardware Reference
9-21
Setting Serial Port Modes
Both transmit and receive clocks can be independently generated internally or input from an external source. The ICLK bit of the STCTLx and
SRCTLx control registers determines the clock source.
When ICLK=1, the clock signal is generated internally by the DSP and the
TCLKx or RCLKx pins are outputs. The clock frequency is determined by the
value of the serial clock divisor (TCLKDIV or RCLKDIV) in the TDIVx or RDIVx
registers.
When ICLK=0, the clock signal is accepted as an input on the TCLKx or
RCLKx pins, and the serial clock divisors in the TDIVx/ RDIVx registers are
ignored. The externally generated serial clock need not be synchronous
with the DSP system clock.
Frame Sync Options
Framing signals indicate the beginning of each serial word transfer. The
framing signals for each serial port are TFS (transmit frame synchronization) and RFS (receive frame synchronization). A variety of framing
options are available; these options are configured in the serial port control registers. The TFS and RFS signals of a serial port are independent and
are separately configured in the control registers.
Framed Versus Unframed
The use of frame sync signals is optional in serial port communications.
The TFSR (transmit frame sync required) and RFSR (receive frame sync
required) control bits determine whether frame sync signals are required.
These bits are located in the STCTLx and SRCTLx control registers.
When TFSR=1 or RFSR=1, a frame sync signal is required for every data
word. To allow continuous transmitting from the DSP, each new data
word must be loaded into the TX buffer before the previous word is shifted
out and transmitted. For more information, see “Data-Independent
Transmit Frame Sync” on page 9-27.
9-22
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
When TFSR=0 or RFSR=0, the corresponding frame sync signal is not
required. A single frame sync is needed to initiate communications but is
ignored after the first bit is transferred. Data words are then transferred
continuously, unframed.
DMA is enabled in this mode, with frame syncs not
 When
required, DMA requests may be held off by chaining or may not be
serviced frequently enough to guarantee continuous unframed data
flow.
Figure 9-2 illustrates framed serial transfers, which have the following
characteristics:
xCLK
FRAMED
DATA
UNFRAMED
DATA
B
3
B
2
B
1
B
0
B
3
B
2
B
1
B
0
B
3
B
2
B
3
B
2
B
1
B
0
B
1
B
0
B
3
B
2
B
1
Figure 9-2. Framed Versus Unframed Data
•
TFSR and RFSR bits in STCTLx, SRCTLx control registers determine
framed or unframed mode.
• Framed mode requires a framing signal for every word. Unframed
mode ignores framing signal after first word.
• Unframed mode is appropriate for continuous reception.
• Active-low or active-high frame syncs selected with LTFS and LRFS
bits of STCTLx, SRCTLx control registers.
ADSP-21160 SHARC DSP Hardware Reference
9-23
Setting Serial Port Modes
Internal Versus External Frame Syncs
Both transmit and receive frame syncs can be independently generated
internally or input from an external source. The ITFS and IRFS bits of the
STCTLx and SRCTLx control registers determine the frame sync source.
When ITFS=1 or IRFS=1, the corresponding frame sync signal is generated
internally by the DSP and the TFSx pin or RFSx pin is an output. The frequency of the frame sync signal is determined by the value of the frame
sync divisor (TFSDIV or RFSDIV) in the TDIVx or RDIVx registers.
When ITFS=0 or IRFS=0, the corresponding frame sync signal is accepted
as an input on the TFSx pin or RFSx pins, and the frame sync divisors in
the TDIVx/RDIVx registers are ignored.
All of the various frame sync options are available whether the signal is
generated internally or externally.
Active Low Versus Active High Frame Syncs
Frame sync signals may be either active high or active low (for example,
inverted). The LTFS and LRFS bits of the STCTLx and SRCTLx control registers determine the frame syncs’ logic level:
• When LTFS=0 or LRFS=0, the corresponding frame sync signal is
active high.
• When LTFS=1 or LRFS=1, the corresponding frame sync signal is
active low.
Active high frame syncs are the default. The LTFS and LRFS bits are initialized to 0 after a processor reset.
9-24
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
Sampling Edge For Data and Frame Syncs
Data and frame syncs can be sampled on either the rising or falling edges
of the serial port clock signals. The CKRE bit of the STCTLx and SRCTLx control registers selects the sampling edge.
For transmit data and frame syncs, setting CKRE=1 in STCTLx selects the rising edge of TCLKx. CKRE=0 selects the falling edge. Note that data and
frame sync signals change state on the clock edge that is not selected.
For receive data and frame syncs, setting CKRE=1 in SRCTLx selects the rising edge of RCLKx. CKRE=0 selects the falling edge.
The transmit and receive functions of two serial ports connected together,
for example, should always select the same value for CKRE so that any internally generated signals are driven on one edge and any received signals are
sampled on the opposite edge.
Early Versus Late Frame Syncs
Frame sync signals can occur during the first bit of each data word (“late”)
or during the serial clock cycle immediately preceding the first bit
(“early”). The LAFS bit of the STCTLx and SRCTLx control registers configures this option.
When LAFS=0, early frame syncs are configured; this is the normal mode
of operation. In this mode, the first bit of the transmit data word is available (and the first bit of the receive data word is latched) in the serial clock
cycle after the frame sync is asserted, and the frame sync is not checked
again until the entire word has been transmitted (or received). (In multichannel operation, this is the case when frame delay is 1.)
If data transmission is continuous in early framing mode (for example, the
last bit of each word is immediately followed by the first bit of the next
word), then the frame sync signal occurs during the last bit of each word.
Internally generated frame syncs are asserted for one clock cycle in early
framing mode.
ADSP-21160 SHARC DSP Hardware Reference
9-25
Setting Serial Port Modes
When LAFS=1, late frame syncs are configured; this is the alternate mode
of operation. In this mode, the first bit of the transmit data word is available (and the first bit of the receive data word is latched) in the same serial
clock cycle that the frame sync is asserted. (In multichannel operation, this
is the case when frame delay is zero.) Receive data bits are latched by serial
clock edges, but the frame sync signal is only checked during the first bit
of each word. Internally generated frame syncs remain asserted for the
entire length of the data word in late framing mode. Externally generated
frame syncs are only checked during the first bit.
Figure 9-3 illustrates the two modes of frame signal timing:
•
LAFS bits of STCTLx, SRCTLx control registers. LAFS=0 for early
frame syncs, LAFS=1 for late frame syncs.
• Early framing: frame sync precedes data by one cycle. Late framing:
frame sync checked on first bit only.
• Data transmitted MSB-first (SENDN=0) or LSB-first (SENDN=1).
• Frame sync and clock generated internally or externally.
xCLK
LATE
FRAME
SYNC
EARLY
FRAME
SYNC
DATA
B3
B2
B1
B0
Figure 9-3. Normal Versus Alternate Framing
9-26
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
Data-Independent Transmit Frame Sync
Normally the internally generated transmit frame sync signal (TFS) is output only when the TX buffer has data ready to transmit. The DITFS mode
(data-independent transmit frame sync) allows the continuous generation
of the TFS signal, with or without new data. The DITFS bit of the STCTLx
control register configures this option.
When DITFS=0, the internally generated TFS is only output when a new
data word has been loaded into the TX buffer. Once data is loaded into TX,
it is not transmitted until the next TFS is generated. This mode of operation allows data to be transmitted only at specific times.
When DITFS=1, the internally generated TFS is output at its programmed
interval regardless of whether new data is available in the TX buffer. Whatever data is present in TX is retransmitted with each assertion of TFS. The
TUVF transmit underflow status bit (in the STCTLx control register) is set
when this occurs (for example, when old data is retransmitted). The TUVF
status bit is also set if the TX buffer does not have new data when an externally generated TFS occurs. Note that in this mode of operation, the first
internally generated TFS is delayed until data has been loaded into the TX
buffer.
If the internally generated TFS is used, a single write to the TX data register
is required to start the transfer.
SPORT Loopback
When the SPL bit (SPORT loopback) is set in the SRCTLx receive control
register, the serial port is configured in an internal loopback connection.
The loopback configuration allows the serial ports to be tested internally.
When loopback is configured, the DRx, RCLKx, and RFSx signals of the
receive section of the SPORT are internally connected to the DTx, TCLKx,
and TFSx signals of the transmit section.
ADSP-21160 SHARC DSP Hardware Reference
9-27
Setting Serial Port Modes
The DTx, TCLKx, and TFSx signals are active and are available at their
respective pins, while the DRx, RCLKx, and RFSx pins are ignored by the
DSP.
transmit clock and transmit frame sync options may be used
 Only
in loopback mode—programs must ensure that the serial port is set
up correctly in the STCTLx and SRCTLx control registers. Multichannel mode is not allowed.
Multichannel Operation
The DSP serial ports offer a multichannel mode of operation which allows
the SPORT to communicate in a time-division-multiplexed (TDM) serial
system. In multichannel communications, each data word of the serial bit
stream occupies a separate channel— each word belongs to the next consecutive channel so that, for example, a 24-word block of data contains
one word for each of 24 channels.
The serial port can automatically select words for particular channels
while ignoring the others. Up to 32 channels are available for transmitting
or receiving—each SPORT can receive and transmit data selectively from
any of the 32 channels. In other words, the SPORT can do any of the following on each channel:
• transmit data
• receive data
• transmit and receive data, or
• do nothing
Data companding and DMA transfers can also be used in multichannel
mode.
9-28
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
The DT pin is always driven (for example, not three-stated) if the serial
port is enabled (SPEN=1 in the STCTLx control register), unless it is in multichannel mode and an inactive time slot occurs.
Note that (in multichannel mode) the TCLKx pin is always an input and
must be connected to its corresponding RCLKx pin.
Figure 9-4 shows example timing for a multichannel transfer, which have
the following characteristics:
WORD 0
WORD 1
WORD 2
SCLK
DR
B3
B2
B1
B0
B3
IGNORED
B2
RFS
DT
B3
B2
B1
B0
B3
B2
TFS
Figure 9-4. Multichannel Operation
• Uses TDM method where serial data is sent or received on different channels sharing the same serial bus.
• The number of channels is selected with the NCH bits of SRCTLx:
NCH=(# of channels) – 1
• Can independently select transmit and receive channels.
•
RFS
•
TFS
signal start of frame.
is used as “Transmit Data Valid” for external logic; active only
during transmit channels.
• Example: Receive on channels 0 and 2. Transmit on channels 1
and 2.
ADSP-21160 SHARC DSP Hardware Reference
9-29
Setting Serial Port Modes
Frame Syncs in Multichannel Mode
All receiving and transmitting devices in a multichannel system must have
the same timing reference. The RFS signal is used for this reference, indicating the start of a block (or frame) of multichannel data words.
When multichannel mode is enabled on a SPORT, both the transmitter
and receiver use RFS as a frame sync. This is true whether RFS is generated
internally or externally. The RFS signal is used to synchronize the channels
and restart each multichannel sequence. RFS assertion occurs the beginning of the channel 0 data word.
is used as a transmit data valid signal which is active during transmission of an enabled word. Because the serial port’s DTx pin is three-stated
when the time slot is not active, the TFS signal specifies whether or not DTx
is being driven by the DSP. The DSP drives TFS in multichannel mode
whether or not ITFS is cleared.
TFS
After the TX transmit buffer is loaded, transmission begins and the TFS signal is generated. When serial port DMA is being used, this may happen
several cycles after the multichannel transmission is enabled. If a deterministic start time is required, the TX buffer should be preloaded.

is normally left unconnected in multichannel mode, and the
RFS pins of the serial port(s) are usually connected together.
TFS
Multichannel Control Bits in STCTL, SRCTL
The STCTLx and SRCTLx control registers contain several bits used to enable
and configure multichannel operations. Multichannel mode is enabled by
setting the MCE bit in the SRCTLx control register:
• When MCE=1, multichannel operation is enabled.
• When MCE=0, all multichannel operations are disabled.
9-30
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
Multichannel operation is activated three cycles after MCE is set. Internally
generated frame sync signals activate four cycles after MCE is set.
Setting the MCE bit enables multichannel operation for both receive and
transmit sides of the SPORT. A transmitting SPORT must be in multichannel mode if the receiving SPORT is in multichannel mode.
The SLEN bits determine the serial bit length of the transmit and receive
data words. The SLEN bit settings in the STCTLx register (bits 4-8) should
match the SLEN bit settings in the SRCTLx register.
The number of channels used in multichannel operation is selected by the
5-bit NCH field in the SRCTLx control register. NCH should be set to the
actual number of channels minus one:
NCH = Number of Channels – 1
The 5-bit CHNL field in the STCTLx control register indicates which channel
is currently selected during multichannel operation. This field is a
read-only status indicator. CHNL(4:0) increments modulo NCH(4:0) as
each channel is serviced.
The 4-bit MFD field in the STCTLx control register specifies a delay between
the frame sync pulse and the first data bit in multichannel mode. The
value of MFD is the number of serial clock cycles of the delay. Multichannel
frame delay allows the processor to work with different types of T1 interface devices.
A value of zero for MFD causes the frame sync to be concurrent with the
first data bit. The maximum value allowed for MFD is 15. A new frame sync
may occur before data from the last frame has been received, because
blocks of data occur back to back.
A multichannel frame delay of at least one should be used when the DSP
is generating frame syncs for the multichannel system and the serial clock
of the system is equal to CLKIN (the processor clock). If MFD is not set to at
least one, the master DSP in a multiprocessing system does not recognize
ADSP-21160 SHARC DSP Hardware Reference
9-31
Setting Serial Port Modes
the first frame sync after multichannel operation is enabled. All succeeding
frame syncs are recognized normally.
Channel Selection Registers
Specific channels can be individually enabled or disabled to select which
words are received and transmitted during multichannel communications.
Data words from the enabled channels are received or transmitted, while
disabled channel words are ignored. Up to 32 channels are available for
transmitting and up to 32 channels for receiving.
The multichannel selection registers are used to enable and disable individual channels. The registers for each serial port are as shown in
Table 9-6.
Table 9-6. Multichannel Selection Registers
Register Name
Function
MTCSx
Multichannel Transmit Select—specifies the active transmit channels
MRCSx
Multichannel Receive Select—specifies the active receive channels
MTCCSx
Multichannel Transmit Compand Select—specifies which active transmit
channels are companded
MRCCSx
Multichannel Receive Compand Select—specifies which active receive
channels are companded
Each register has 32 bits, corresponding the 32 channels. Setting a bit
enables that channel so that the serial port selects its word from the multiple-word block of data (for either receive or transmit). For example,
setting bit 0 selects word 0, setting bit 12 selects word 12, and so on.
Setting a particular bit to 1 in the MTCSx register causes the serial port to
transmit the word in that channel’s position of the data stream. Clearing
the bit to 0 in the MTCSx register causes the serial port’s DT (data transmit)
pin to three-state during the time slot of that channel.
9-32
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
Setting a particular bit to 1 in the MRCSx register causes the serial port to
receive the word in that channel’s position of the data stream; the received
word is loaded into the RX buffer. Clearing the bit to 0 in the MRCSx register causes the serial port to ignore the data.
Companding may be selected on a per-channel basis. The MTCCSx and
MRCCSx registers are used to specify companding for any active channels.
Setting a bit to 1 in these registers causes the data to be companded. A-law
or -law companding is selected with the DTYPE bit 1 in the STCTLx and
SRCTLx control registers.
SPORT Receive Comparison Registers
On the DSP, two sets of registers aid multiprocessor communications
when using multichannel mode (MCE=1) through the serial ports. These
32-bit registers are the Receive Comparison (KEYWDx) registers and the
Receive Comparison Mask (KEYMASKx) registers.
Table 9-7 shows the MCE setting as well as the bits in the SRCTL register
that control the operation of Receive Comparison.
Table 9-7. Receive Comparison Selection
IMODE
(Bit 15)
IMAT
(Bit 20)
Operation
0
x
Receive comparison disabled
1
0
Accept receive data if the KEYWD comparison is false
1
1
Accept receive data if the KEYWD comparison is true
The KEYWD0 or KEYWD1 register stores the pattern to be matched with the
incoming data. The corresponding KEYMASK0 or KEYMASK1 register specifies
which of the bits in the received data should be compared. Setting KEYMASKx bit (=1) masks the corresponding bit in KEYWDx register, disabling its
comparison.
ADSP-21160 SHARC DSP Hardware Reference
9-33
Setting Serial Port Modes
The processor receiving the data compares it with the data in the KEYWDx
register. Depending on the comparison results, the received data is
accepted or ignored. If accepted, the receiver requests—based on the setting to the SRCTL register—a DMA transfer to internal memory or
generates an interrupt.
When receive comparison is enabled, companding is disabled on the transmitter and receiver. The MTCCSx register, which selects multichannel
companding when receive comparison is disabled, determines whether the
DSP performs a KEYWD comparison for the enabled received channels. If
the MTCCSx bit for a particular channel is '0,' the processor does not perform a comparison and always accepts the receive data on that channel. If
the MTCCSx bit for a particular channel is '1,' the processor performs the
comparison and accepts (or rejects) the receive data, depending on the
result of the comparison and IMAT setting in the SRCTLx register.
The receive comparison feature lets the DSP's SPORTs generate a DMA
request or an interrupt when the received data matches a specified condition on a specified channel in multichannel mode. Without this feature,
the SPORT would interrupt the processor every time data was received
and the processor would be required to check if the data was meant for it
or not. It is possible that most of the time the data being sent is not meant
for the processor. With the receive comparison feature, the SPORT on a
particular processor can be programmed to interrupt only on messages
meant for that processor.
As a receive comparison example, consider four DSPs (A, B, C, and D)
which use SPORT0 (in multichannel mode) for interprocessor communication. Channels 0, 1, 2, and 3 are used respectively by A, B, C, and D to
transmit control information between the processors. Channels 4 through
10, 11 through 17, 18 through 24, and 25 through 31 are used respectively by A, B, C, and D to transmit data.
Because channels 0 through 3 are used to send control information
between the processors, the comparisons for incoming data is enabled only
for these channels. Initially, channels 4 through 31 may have receive
9-34
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
disabled. For this example, consider communication between processors A
and B only. The keyword for comparison is programmable; in this example, processor B can check for the keyword START TRANSMIT TO B.
Processor B can check for this keyword as follows.
1. Set the KEYWD register to START
TRANSMIT TO B.
2. Clear bits 31:16 of the KEYMASK register to 0 and set the other bits
to 1.
This step enables comparison only for bits 31:16. So, assume that
the code for START TRANSMIT TO B only uses bits 31:16 and bits
15:0 indicate the source of the transmission and the data channels.
3. Set bits 15 and 20 of the SRCTL register to 1.
This step enables the SPORT to generate an interrupt or DMA
request only if the incoming data matches the KEYWD.
4. Set bits 0 through 3 of the Transmit Compand Channel Selector
register to 1 and clear the remaining bits to 0.
This step enables comparison only on channels 0 through 3.
Until it receives the START TRANSMIT TO B keyword, processor B ignores all
transmissions that it receives. When processor A wants to send data to B,
it sends this keyword on channel 0. When receive comparison on processor B recognizes the START TRANSMIT TO B keyword, the SPORT
interrupts processor B. Then, processor B analyzes the remaining 16-bits,
determining that the source is processor A and the data is on channels 4
through 10.
Because processor A is using channels 4 through 10 to transmit data, processor B enables receive channels 4 through 10 and sends a “READY TO
RECEIVE DATA” message to processor A, using channel 1. After processor A receives this message, it sends the data on channels 4 through 10.
ADSP-21160 SHARC DSP Hardware Reference
9-35
Moving Data Between SPORTS and Memory
If the transfer protocol uses a fixed number of bytes in each message, processor B could send back a checksum message to processor A after
receiving A's message, confirming that the data transferred accurately.
Moving Data Between SPORTS and
Memory
Transmit and receive data can be transferred between the DSP serial ports
and on-chip memory in one of two ways, with single-word transfers or
with DMA block transfers. Both methods are interrupt-driven, using the
same internally generated interrupts.
When serial port DMA is not enabled in the STCTLx or SRCTLx control registers, the SPORT generates an interrupt every time it has received a data
word or has started to transmit a data word. SPORT DMA provides a
mechanism for receiving or transmitting an entire block of serial data
before the interrupt is generated. The DSP’s on-chip DMA controller
handles the DMA transfer, allowing the processor core to continue running until the entire block of data is transmitted or received. Service
routines can then operate on the block of data rather than on single words,
significantly reducing overhead.
DMA Block Transfers
The DSP’s on-chip DMA controller allows automatic DMA transfers
between internal memory and the two serial ports. There are four DMA
channels for serial port operations—each SPORT has one channel for
receiving data and one for transmitting data. The serial port DMA channels are numbered as follows:
• DMA Channel 0 – SPORT0 Receive
• DMA Channel 1 – SPORT1 Receive
9-36
ADSP-21160 SHARC DSP Hardware Reference
Serial Ports
• DMA Channel 2 – SPORT0 Transmit
• DMA Channel 3 – SPORT1 Transmit
The SPORT DMA channels are assigned higher priority than all other
DMA channels (for example, link ports and the external port) because of
their relatively low service rate and their inability to hold off incoming
data. Having higher priority causes the SPORT DMA transfers to be performed first when multiple DMA requests occur in the same cycle.
Although the DMA transfers are always performed with 32-bit words, the
serial ports can handle word sizes from 3 to 32 bits. If the serial words are
16 bits or smaller, they can be packed into 32-bit words for each DMA
transfer; this is configured by the PACK bit of the STCTLx and SRCTLx control registers. When serial port data packing is enabled (PACK=1), the
transmit and receive interrupts are generated for the 32-bit packed words,
not for each 16-bit word.
The following sections present an overview of serial port DMA operations;
some additional details are covered in the DMA chapter of this manual.
• For information on SPORT DMA Channel Setup, see “Setting up
Serial Port DMA” on page 6-90.
• For information on SPORT DMA Parameter Registers, see “Setting I/O Processor—SPort Modes” on page 6-49.
• For information on SPORT DMA Chaining, see “Chaining DMA
Processes” on page 6-69.
Single-Word Transfers
Individual data words may also be transmitted and received by the serial
ports, with interrupts occurring as each 32-bit word is transmitted or
received. When a serial port is enabled and DMA is disabled (in the
STCTLx or SRCTLx control registers), the SPORT DMA interrupts are generated in this way—whenever a complete 32-bit word has been received in
ADSP-21160 SHARC DSP Hardware Reference
9-37
SPORT Pin/Line Terminations
the RX buffer, or whenever the TX buffer is not full. Single-word interrupts
can be used to implement interrupt-driven I/O on the serial ports.
Whenever the DSP core’s program reads a word from a serial port’s RX
buffer or writes a word to its TX buffer, the buffer’s full/empty status
should first be checked in order to avoid hanging the DSP core. (This can
also happen to an external device, for example a host processor, when it is
reading or writing a serial port buffer.) The full/empty status can be read
in the RXS bits of the SRCTLx register or the TXS bits of the STCTLx register.
Reading from an empty RX buffer or writing to a full TX buffer causes the
DSP (or external device) to hang, waiting for the status to change.
support debugging buffer transfers, the DSP has a Buffer Hang
 ToDisable
( ) bit. When set (=1), this bit prevents the processor
BHD
core from detecting a buffer-related stall condition, permitting
debugging of this type of stall condition. For more information, see
the BHD discussion on page -18.
Multiple interrupts can occur if both SPORTs transmit or receive data in
the same cycle. Any interrupt can be masked out in the IMASK register; if
the interrupt is later enabled in IMASK, the corresponding interrupt latch
bit in IRPTL must be cleared in case the interrupt has occurred in the
meantime.
When serial port data packing is enabled (PACK=1 in the STCTLx or SRCTLx
control registers), the transmit and receive interrupts are generated for the
32-bit packed words, not for each 16-bit word.
SPORT Pin/Line Terminations
The DSP has very fast drivers on all output pins including the serial ports.
If connections on the data, clock, or frame sync lines are longer than six
inches, you should consider using a series termination for strip lines on
point-to-point connections. This may be necessary even when using
low-speed serial clocks, because of the edge rates.
9-38
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
10 JTAG TEST EMULATION PORT
A boundary scan allows a system designer to test interconnections on a
printed circuit board with minimal test-specific hardware. The scan is
made possible by the ability to control and monitor each input and output
pin on each chip through a set of serially scannable latches. Each input
and output is connected to a latch, and the latches are connected as a long
shift register so that data can be read from or written to them through a
serial test access port (TAP).
The ADSP-21160 DSP contains a test access port compatible with the
industry-standard IEEE 1149.1 (JTAG) specification. Only the IEEE
1149.1 features specific to the ADSP-21160 DSPs are described here. For
more information, see the IEEE 1149.1 specification and other the documents listed in “References” on page 10-55.
The boundary scan allows a variety of functions to be performed on each
input and output signal of the ADSP-21160 DSP. Each input has a latch
that monitors the value of the incoming signal and can also drive data into
the chip in place of the incoming value. Similarly, each output has a latch
that monitors the outgoing signal and can also drive the output in place of
the outgoing value. For bidirectional pins, the combination of input and
output functions is available.
Every latch associated with a pin is part of a single serial shift register path.
Each latch is a master/slave type latch with the controlling clock provided
externally. This clock (TCK) is asynchronous to the ADSP-21160 system
clock (CLKIN).
ADSP-21160 SHARC DSP Hardware Reference
10-1
The ADSP-21160 emulation features let you halt the processor at a
pre-defined point. You then examine the state of the processor, execute
arbitrary code, restore the original state, and continue execution.
ADSP-21160 emulation features are a superset of the
 The
ADSP-21060 emulation features. All emulation features supported
by previous SHARC DSPs are supported on the ADSP-21160
DSP, except the ICSA output signal and function. The set of features on which EZ-ICE® designs rely are supported in an identical
fashion on ADSP-21160 DSP. The ADSP-21160 DSP can be used
with the ADSP-2106x SHARC EZ-ICE hardware.
There are several changes/extensions to the base functionality of the
ADSP-21060 emulation capability, which require changes in the EZ-ICE
software for ADSP-21160 DSP support. These extensions include:
• The emulation breakpoint address start/end registers have moved
from UREG space to IOP register space. This change did not effect
the TSTEMU block directly, only the address decodes to gain access to
it.
•
has been added to the IR decode space. This shift register
provides access to the full 64-bit wide PX register of ADSP-21160
DSP.
EMU64PX
• A memory test shift register has been added to the IR decode space.
This feature is for Analog Devices internal use ONLY.
• Addition of the MTST (Memory TeST) bit in the EMUCTL register.
This feature is for Analog Devices internal use ONLY. EMUCTL is 40
bits wide on the ADSP-21160 DSP.
Several on chip facilities are directly accessed through the JTAG interface.
These facilities are listed in Table 10-2 on page 10-4. Other emulation
facilities are only indirectly accessible. To indirectly access the facilities
that do not appear in Table 10-2, scan the instruction which moves data
10-2
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
of interest to/from the PX register, scan the PX data (if the instruction is a
PX read), let the core execute the instruction, then scan the PX register out
(if the instruction was a PX write).
The breakpoint start/end registers are mapped into the IOP register space
of the ADSP-21160 DSP. For specific addresses, see “Register and Bit
#Defines File (def21160.h)” on page A-81. The EMUN, EMUCLK, and EMUCLK2 registers occupy the same UREG address space as on the ADSP-2106
DSPs.
These facilities are read only by the ADSP-21160 processor core in normal
operation.
JTAG Test Access Port
The emulator uses JTAG boundary scan logic for ADSP-21160 communications and control. This JTAG logic consists of a state machine, a five
pin Test Access Port (TAP), and shift registers. The state machine and
pins conform to the IEEE 1149.1 specification. The TAP pins appear in
Table 10-1.
Table 10-1. JTAG Test Access Port (TAP) Pins
Pin
Function
TCK
(input) Test Clock: pin used to clock the TAP state machine.1
TMS
(input) Test Mode Select: pin used to control the TAP state machine sequence.2
TDI
(input) Test Data In: serial shift data input pin.
TDO
(output) Test Data Out: serial shift data output pin.
TRST
(input) Test Logic Reset: resets the TAP state machine
1
2
Asynchronous with CLKIN
Synchronous to CLKIN
ADSP-21160 SHARC DSP Hardware Reference
10-3
Instruction Register
A BSDL file for the ADSP-21160 DSP is available on Analog Devices’
website. Set your browser to:
http://www.analog.com/dsp
Refer to the IEEE 1149.1 JTAG specification for detailed information on
the JTAG interface. The many sections of this appendix assume a working
knowledge of the JTAG specification.
Instruction Register
The instruction register allows an instruction to be shifted into the processor. This instruction selects the test to be performed and/or the test data
register to be accessed. The instruction register is 5 bits long with no parity bit. A value of 10000 binary is loaded (LSB nearest TDO) into the
instruction register whenever the TAP reset state is entered.
Table 10-2 lists the binary code for each instruction. Bit 0 is nearest TDO
and bit 4 is nearest TDI. No data registers are placed into test modes by
any of the public instructions. The instructions affect the ADSP-21160
DSP as defined in the 1149.1 specification. The optional instructions
RUNBIST, IDCODE and USERCODE are not supported by the ADSP-21160
DSP.
Table 10-2. JTAG Instruction Register Codes
43210
Register
Instruction
11111
Bypass
BYPASS
Public
00000
Boundary
EXTEST
Public
10000
Boundary
SAMPLE
Public
01000
EMUPMD
EMULATION
11000
Boundary
INTEST
Public
00100
EMUCTL
EMULATION
Private
10100
EMUPX
EMULATION
10-4
Comment
48-bit scan length
48-bit shift register
Type
Private
Private
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-2. JTAG Instruction Register Codes (Cont’d)
43210
Register
Instruction
Comment
Type
10110
EMU64PX
EMULATION
64-bit shift register
Private
01100
EMUSTAT
EMULATION
Private
11100
BRKSTAT
EMULATION
Private
00010
EMUPC
EMULATION
Private
10101
MEMTST
TEST
All
others
Reserved
Reserved
Memory test
Private
Private
The entry under “Register” is the serial scan path, either Boundary or
Bypass in this case, enabled by the instruction. Figure 10-1 on page 10-6
shows these register paths. The 1-bit Bypass register is fully defined in the
1149.1 specification. For more information on the Boundary register, see
“Boundary Register” on page 10-17.
No special values need be written into any register prior to selection of any
instruction. As Table 10-2 on page 10-4 shows, certain instructions are
reserved for emulator use. For more information, see Table 10-7 on
page 10-18.
EMUPMD Shift Register
The EMUPMD serial shift register is located in the system unit. EMUPMD is 48
bits wide and is accessed by the emulator through TAP. When the TAP
enters the UPDATE state and EMUPMD is selected, a 48-bit slave register is
updated from EMUPMD. EMUPMD’s purpose is to force the ADSP-21160 DSP
to execute emulator supplied instructions. The register accomplishes this
by driving the instruction bus while in emulation space.
ADSP-21160 SHARC DSP Hardware Reference
10-5
Instruction Register
B OU NDAR Y R E GIS T E R
643
2
644
1
645
0
B YP AS S R E GIS T E R
T DI
1
T DO
4
0
3
1
2
INS T R U CT ION R E GIS T E R
Figure 10-1. Serial Scan Paths
10-6
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
EMUPX Shift Register
The EMUPX serial shift register is located in the system unit. EMUPX is a
48-bits wide and is accessed by the emulator through the TAP. When the
TAP goes into the UPDATE state and EMUPX is selected, the most significant 48-bits of PX is updated from EMUPX. When the TAP goes into the
CAPTURE state and EMUPX is selected, EMUPX is updated with the most significant 48-bits of PX. The EMUPX register is used to transfer data between
the emulator and the target system.
is provided for backwards compatibility with the SHARC ICE
hardware. PX is a 64-bit wide register. To provide compatibility, only the
most significant 48 bits of PX are mapped to EMUPX. 48-bit instructions,
and 40-bit extended precision data, are always aligned to the most significant bit. When transferring 32-bit data to/from PX register, PX2 must be
specified as the source/destination to ensure that the 32 bits is aligned to
the most significant bit.
EMUPX
EMU64PX Shift Register
The EMU64PX serial shift register is located in the system unit. EMU64PX is
64-bits wide and is accessed by the emulator through the TAP. When the
TAP goes into the UPDATE state and EMU64PX is selected, PX is updated
from EMU64PX. When the TAP goes into the CAPTURE state and EMU64PX
is selected, EMU64PX is updated from PX. The EMU64PX register transfers data
between the emulator and the target system. The most significant 48 bits
of EMU64PX are redundantly available in EMUPX.
ADSP-21160 SHARC DSP Hardware Reference
10-7
Instruction Register
EMUPC Shift Register
The EMUPC serial shift register is located in the system unit. EMUPC is 24-bits
wide and is accessed by the emulator through the TAP. The EMUPC register
captures addresses from the PC register. This data can be used to statistically profile the user’s code. Addresses cannot be forced into the PC
register from the EMUPC register.
EMUCTL Shift Register
The EMUCTL serial shift register is located in the system unit. EMUCTL is
40-bits wide and is accessed by the emulator through the TAP. EMUCTL
controls all of the ADSP-21160 emulation functionality. Table 10-3 lists
EMUCTL’s bits and describes their functionality.
Table 10-3. EMUCTL (Emulation Control) Register Definition
Bit #
Name
Function
0
EMUENA
Emulator Function Enable.
The EMUENA bit enables ADSP-21160 emulation functions.
(0=ignore breakpoints and emulator interrupts, 1=respond to breakpoints and emulator interrupts)
1
EIRQENA
Emulator Interrupt Enable.
The EIRQENA bit enables the emulation logic to recognize external
emulator interrupts. (0=disable, 1=enable)
2
BKSTOP
Enable Autostop on Breakpoint.
The BKSTOP bit enables the ADPS-21160 DSP to generate an external emulator interrupt when any breakpoint event occurs.
(0=disable, 1=enable)
3
SS
Enable Single Step Mode. The SS bit enables single-step operation.
(0=disable, 1=enable)
4
SYSRST
Software Reset of the ADSP-21160 DSP.
The SYSRST bit resets the ADSP-21160 DSP in the same manner as
the external RESET pin. The SYSRST bit must be cleared by the emulator.(0=normal operation, 1=reset)
10-8
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-3. EMUCTL (Emulation Control) Register Definition (Cont’d)
Bit #
Name
Function
5
ENBRKOUT
Enable the BRKOUT pin.
The ENBRKOUT bit enables the BRKOUT pin operation.
(0=BRKOUT pin at high-impedance state, 1=BRKOUT pin enabled)
6
IOSTOP
Stop IOP DMAs in EMU space.
The IOSTOP bit disables all DMA requests when the DSP is in emulation space. Data that is currently in the EP, LINK, or SPORT DMA
buffers is held there unless the internal DMA request was already
granted. IOSTOP causes incoming data to be held off and outgoing
data to cease. Because SPORT receive data cannot be held off, it is
lost and the overrun bit is set. The direct write buffer (internal memory write) and the EP pad buffer are allowed to flush any remaining
data to internal memory. (0=IO continues, 1=IO Stops)
7
EPSTOP
Stop I/O Processor EP operation in emulation space.
The EPSTOP bit disables all EP requests when the DSP is in emulation space. After an emulation interrupt is acknowledged, EPSTOP
deasserts ACK (deasserts REDY if host access) to prevent further data
from being accepted if the EP is accessed. The emulator may clear
this bit—allowing I/O to continue and the bus to clear—so that the
emulator may use the EP (through BR and bus lock). Note that the
EP bus clears only if accesses are direct writes or IOP register writes,
because all other IOP functions are halted. The EP bus does not clear
if accesses to any of the DMA buffers are extended due to a buffer full
or empty condition.
(0=EP IO continues, 1=EP IO Stops)
8
NEGPA1
Negate program memory data address breakpoint.
The NEG bits enable breakpoint events if the address is greater than
the end register value OR less than the start register value. This function is useful to detect index range violations in user code. (0=disable
breakpoint, 1=enable breakpoint)
9
NEGDA1
Negate data memory address breakpoint #1. For more information,
see NEGPA1 bit description on page 10-9.
10
NEGDA2
Negate data memory address breakpoint #2. For more information,
see NEGPA1 bit description on page 10-9.
11
NEGIA1
Negate instruction address breakpoint #1. For more information, see
NEGPA1 bit description on page 10-9.
ADSP-21160 SHARC DSP Hardware Reference
10-9
Instruction Register
Table 10-3. EMUCTL (Emulation Control) Register Definition (Cont’d)
Bit #
Name
Function
12
NEGIA2
Negate instruction address breakpoint #2. For more information, see
NEGPA1 bit description on page 10-9.
13
NEGIA3
Negate instruction address breakpoint #3. For more information, see
NEGPA1 bit description on page 10-9.
14
NEGIA4
Negate instruction address breakpoint #4. For more information, see
NEGPA1 bit description on page 10-9.
15
NEGIO1
Negate I/O address breakpoint. For more information, see NEGPA1
bit description on page 10-9.
16
NEGEP1
Negate EP address breakpoint. For more information, see NEGPA1
bit description on page 10-9.
17
ENBPA
Enable program memory data address breakpoints.
The ENB bits enable each breakpoint group. Note that when the
ANDBKP bit is set, breakpoint types not involved in the generation
of the effective breakpoint must be disabled. (0=disable breakpoints,
1=enable breakpoints)
18
ENBDA
Enable data memory address breakpoints. For more information, see
ENBPA bit description on page 10-10.
19
ENBIA
Enable instruction address breakpoints. For more information, see
ENBPA bit description on page 10-10.
20
ENBIO
Enable I/O address breakpoint. For more information, see ENBPA
bit description on page 10-10.
21
ENBEP
Enable external port address breakpoint. For more information, see
ENBPA bit description on page 10-10.
22-23
PA1MODE
PA1 breakpoint triggering mode. The breakpoint triggering mode
bits trigger on the following conditions:
ModeTriggering condition
00Breakpoint is disabled
01WRITE accesses only
10READ accesses only
11any access
24-25
DA1MODE
DA1 breakpoint triggering mode. For more information, see PA1MODES bit description on page 10-10.
10-10
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-3. EMUCTL (Emulation Control) Register Definition (Cont’d)
Bit #
Name
Function
26-27
DA2MODE
DA2 breakpoint triggering mode. For more information, see PA1MODES bit description on page 10-10.
28-29
IO1MODE
IO1 breakpoint triggering mode. For more information, see PA1MODES bit description on page 10-10.
30-31
EP1MODE
EP1 breakpoint triggering mode. For more information, see PA1MODES bit description on page 10-10.
32
ANDBKP
AND composite breakpoints.
The ANDBKP bit enables AND’ing of each breakpoint type to generate an effective breakpoint from the composite breakpoint signals.
(0=OR breakpoint types, 1=AND breakpoint types)
33
RESERVED
The ICSA function and DMDSEL bit used by that function not supported on ADSP-21160 DSP.
34
NOBOOT
No power-up boot on reset.
The NOBOOT bit forces the ADSP-21160 DSP into the No boot
mode. In this mode, the processor does not boot load, but begins
fetching instructions from 0x0080 0004 in external memory.
(0=disable, 1=force No boot mode)
35
TMODE
Test mode enable. The TMODE bit is for Analog Devices’ usage
only. Do NOT set this bit. (0=normal operation)
36
BHO
Buffer Hang Override bit.
The BHO control bit overrides the BHD bit in SYSCON, disabling
BHD’s control over core access of data buffer behavior. Note that the
default (reset) state of BHD is now set for ADSP-21160 DSP, a
change from ADSP-2106x DSPs.
(0=normal BHD operation, 1=override BHD operation)
37
MTST
Memory Test Enable Bit.
The MTST bit enables scanning of data for to the latches used for
memory test. (0=normal operation, 1=enable memory test)
38, 39
Reserved
Reserved
ADSP-21160 SHARC DSP Hardware Reference
10-11
Instruction Register
EMUSTAT Shift Register
The EMUSTAT serial shift register is located in the system unit. EMUSTAT is
8-bits wide and is accessed by the emulator through the TAP. This register
is updated by the ADSP-21160 DSP when the TAP is in the CAPTURE
state.
The emulator reads EMUSTAT to determine the state of the ADSP-21160
DSP. None of the bits in this register can be written by the emulator. All
bits are active high. Table 10-4 lists the EMUSTAT register’s bits.
Table 10-4. EMUSTAT (Emulation Status) Register Definition
Bit
Name
Function (If bit=1...)
0
EMUSPACE
Indicates that the next instruction is to be fetched from the emulator.
1
EMUREADY
Indicates that the ADSP-21160 has finished executing the previous
emulator instruction.
2
INIDLE
Indicates that the ADSP-21160 was in IDLE prior to the latest emulator interrupt.
3
COMHALT
Indicates a core access to a SPORT or a LINK is hung because of an
external device.
4
EPHALT
Indicates a core access to a DMA buffer is hung because of the external port.
5-7
Reserved
BRKSTAT Shift Register
The BRKSTAT serial shift register is located in the system unit. BRKSTAT is a
16 bits wide and is accessed by the emulator through the TAP. This register monitors the status of the emulation breakpoints and is updated on
every clock cycle. None of the bits of this register can be written by the
emulator. Table 10-5 lists the BRKSTAT register’s bits.
10-12
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
A high bit indicates a breakpoint hit. When a breakpoint hit occurs, the
register ceases updating. Stopping allows the emulator to see which breakpoint was triggered. When the ADSP-21160 DSP leaves emulation space
the BRKSTAT register is cleared and resumes updating. All status bits are
synchronized to TCLK before being scanned out.
Table 10-5. BRKSTAT (Breakpoint Status) Register Definition
Bit #
Name
Function (If bit=1...)
0
STATPA
Program Memory Data breakpoint hit
1
STATDA0
Data Memory breakpoint hit
2
STATDA1
Data Memory breakpoint hit
3
STATIA0
Instruction Address breakpoint hit
4
STATIA1
Instruction Address breakpoint hit
5
STATIA2
Instruction Address breakpoint hit
6
STATIA2
Instruction Address breakpoint hit
7
STATIO
I/O Address breakpoint hit
8
STATEP
EP Address breakpoint hit
9-15
Reserved
MEMTST Shift Register
The MEMTST serial shift register is for Analog Devices’ usage only.
not attempt to use this register—incorrect usage of this feature
 Do
can result in permanent damage to the ADSP-21160 DSP being
tested.
ADSP-21160 SHARC DSP Hardware Reference
10-13
Instruction Register
PSx, DMx, IOx, and EPx (Breakpoint) Registers
The PSx, DMx, IOx, and EPx (Breakpoint) registers are located in the I/O
Processor register set. The emulation breakpoint registers are not user
accessible and can be written only when the ADSP-21160 DSP is in emulation space or test mode. The breakpoint registers vary in size according
to the address type: instruction (24-bit address), data (32-bit address), or
I/O data (19-bit address). Table 10-6 shows the sizes.
The ADSP-21160 DSP contains nine sets of emulation breakpoint registers. Each set consists of a start and end register which describe an address
range, with the start register setting the lower end of the address range.
Each breakpoint set monitors a particular address bus. When a valid
address is in the address range, than a breakpoint signal is generated. The
address range includes the start and end addresses.
The nine breakpoint sets are grouped into five types: instruction (IA),
DM data (DA), PM data (PA), IO data (IO), and EP data (EP). The individual breakpoint signals in each type are OR’ed together to create five
composite breakpoint signals.
These composite signals can be optionally AND’ed or OR’ed together to
create the effective breakpoint event signal used to generate an emulator
interrupt. The ANDBKP bit in the EMUCTL register selects the function used.
Each breakpoint type has an enable bit in the EMUCTL register. When set,
these bits add the specified breakpoint type into the generation of the
effective breakpoint signal. If cleared, the specified breakpoint type is not
used in the generation of the effective breakpoint signal. This allows the
user to trigger the effective breakpoint from a subset of the breakpoint
types.
10-14
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
To provide further flexibility, each individual breakpoint can be programmed to trigger if the address is in range AND one of these conditions
is met: READ access, WRITE access, ANY access, or NO access. The control bits for this feature are also located in EMUCTL. For more information,
see PA1MODES bit description on page 10-10.
The address ranges of the emulation breakpoint registers are negated by
setting the appropriate renege negation bits in the EMUCTL register. For
more information, see NEGPA1 bit description on page 10-9. Each breakpoint can be disabled by setting the start address larger than the end
address.
Four of the breakpoints monitor the instruction address. Two monitor the
data memory address. One monitors the program memory data address,
one monitors the I/O address bus and one monitors the EP address bus.
The instruction address breakpoints monitor the address of the instruction being executed, not the address of the instruction being fetched. If
the current execution is aborted, the breakpoint signal does not occur even
if the address is in range. Data address breakpoints (DA and PA only) are
also ignored during aborted instructions.
The nine breakpoint sets appear in Table 10-6.
Table 10-6. PSx, DMx, IOx, and EPx (Breakpoint) Registers
Register
Function
Group1
PSA1S
Instruction Address Start #1
IA
PSA1E
Instruction Address End #1
IA
PSA2S
Instruction Address Start #2
IA
PSA2E
Instruction Address End #2
IA
PSA3S
Instruction Address Start #3
IA
PSA3E
Instruction Address End #3
IA
PSA4S
Instruction Address Start #4
IA
ADSP-21160 SHARC DSP Hardware Reference
10-15
Instruction Register
Table 10-6. PSx, DMx, IOx, and EPx (Breakpoint) Registers (Cont’d)
Register
Function
Group1
PSA4E
Instruction Address End #4
IA
DMA1S
Data Address Start #1
DA
DMA1E
Data Address End #1
DA
DMA2S
Data Address Start #2
DA
DMA2E
Data Address End #2
DA
PMDAS
Program Data Address Start
PA
PMDAE
Program Data Address End
PA
IOAS
I/O Address Start
IO
IOAE
I/O Address End
IO
EPAS
External Port Address Start
EP
EPAE
External Port Address End
EP
1
Group IA=24-bit addresses, Groups DA, PA, and EP=32-bit addresses, Group IO=19-bit addresses.
EMUN Register
The EMUN (Nth event counter) register is located in the I/O Processor register set. The EMUN register is not user accessible and can be written only
when the ADSP-21160 DSP is in emulation space. EMUN is read-only from
normal-space and can be written only when the ADSP-21160 DSP is in
emulation space. The Nth event counter allows an emulation breakpoint
to occur on the Nth occurrence of the breakpoint event. This is accomplished by writing the desired Nth value to the EMUN register in UREG
space. This register can be read from normal space, but it can be written
only in emulation space. The counter decrements on each occurrence of
the breakpoint event, asserting the interrupt when the counter is equal to
zero and the hardware breakpoint event occurs.
10-16
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
EMUCLK and EMUCLK2 Registers
The EMUCLK (clock counter) and EMUCLK2 (clock counter scaling) registers
are located in the universal (UREG) register set. EMUCLK and EMUCLK2 are not
user accessible and can be written only when the ADSP-21160 DSP is in
emulation space. These registers are read-only from normal-space and can
be written only when the ADSP-21160 DSP is in emulation space.
The Emulation Clock Counter consists of a 32-bit count register (EMUCLK)
and a 32-bit scaling register (EMUCLK2). The EMUCLK counts clock cycles
while the user has control of the ADSP-21160 DSP and stops counting
when the emulator gains control. These registers let you gauge the amount
of time spent executing a particular section of code. The EMUCLK2 register
extends the time EMUCLK can count by incrementing each time the EMUCLK
value rolls over to zero. The combined emulation clock counter can count
accurately for thousands of hours.
EMUIDLE Instruction
The EMUIDLE instruction places the ADSP-21160 DSP in the IDLE state
and triggers an emulator interrupt. This operation lets you use the EMUIDLE instruction to be used as a software breakpoint. When EMUIDLE is
executed, the emulation clock counter immediately halts.
In-Circuit Signal Analyzer (ICSA) Function
This function is NOT supported in ADSP-21160 DSP.
Boundary Register
The Boundary register is 655 bits long. Table 10-7 defines the latch type
and function of each position in the scan path. The positions are numbered with 0 being the first bit output (closest to TDO) and 654 being the
last (closest to TDI).
ADSP-21160 SHARC DSP Hardware Reference
10-17
Boundary Register
Notes on Boundary registers:
• Scan position 0 (L0DAT0) is the end is closest to TDO (scan in first)
• Scan position 654 (SPARE); this end is closest to TDI (scan in last)
• Output Enables:
• 1 = Drive the associated signals during the EXTEST and
INTEST instructions
• 0 = Three-state the associated signals during the EXTEST and
INTEST instructions
can be sampled but not controlled (read-only). CLKIN continues to clock the ADSP-21160DSP no matter which instruction is
enabled.
CLKIN
Table 10-7. JTAG Boundary
Register
Scan
#
Signal Name
Latch Type
0
L0DAT(0)
Output
1
L0DAT(0)
Output enable
2
L0DAT(0)
Input
3
L0DAT(1)
Output
4
L0DAT(1)
Output enable
5
L0DAT(1)
Input
6
L0DAT(2)
Output
7
L0DAT(2)
Output enable
10-18
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
8
L0DAT(2)
Input
9
L0DAT(3)
Output
10
L0DAT(3)
Output enable
11
L0DAT(3)
Input
12
L0ACK
Output
13
L0ACK
Output enable
14
L0ACK
Input
15
L0CLK
Output
16
L0CLK
Output enable
17
L0CLK
Input
18
L0DAT(4)
Output
19
L0DAT(4)
Output enable
20
L0DAT(4)
Input
21
L0DAT(5)
Output
22
L0DAT(5)
Output enable
23
L0DAT(5)
Input
24
L0DAT(6)
Output
25
L0DAT(6)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-19
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
26
L0DAT(6)
Input
27
L0DAT(7)
Output
28
L0DAT(7)
Output enable
29
L0DAT(7)
Input
30
DT0
Output
31
DT0
Output enable
32
DT0
No function
33
TCLK0
Output
34
TCLK0
Output enable
35
TCLK0
Input
36
TFS0
Output
37
TFS0
Output enable
38
TFS0
Input
39
RFS0
Output
40
RFS0
Output enable
41
RFS0
Input
42
RCLK0
Output
43
RCLK0
Output enable
10-20
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
44
RCLK0
Input
45
DR0
Output
46
DR0
Output enable
47
DR0
Input
48
DR1
Output
49
DR1
Output enable
50
DR1
Input
51
RCLK1
Output
52
RCLK1
Output enable
53
RCLK1
Input
54
RFS1
Output
55
RFS1
Output enable
56
RFS1
Input
57
TFS1
Output
58
TFS1
Output enable
59
TFS1
Input
60
TCLK1
Output
61
TCLK1
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-21
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
62
TCLK1
Input
63
DT1
Output
64
DT1
Output enable
65
DT1
No function
66
TIMEXP
Output
67
TIMEXP
No function
68
TIMEXP
No function
69
FLAG0
Output
70
FLAG0
Output enable
71
FLAG0
Input
72
FLAG1
Output
73
FLAG1
Output enable
74
FLAG1
Input
75
FLAG2
Output
76
FLAG2
Output enable
77
FLAG2
Input
78
FLAG3
Output
79
FLAG3
Output enable
10-22
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
80
FLAG3
Input
81
IRQ0_B
No function
82
IRQ0_B
No function
83
IRQ0_B
Input
84
IRQ1_B
No function
85
IRQ1_B
No function
86
IRQ1_B
Input
87
IRQ2_B
No function
88
IRQ2_B
No function
89
IRQ2_B
Input
90
RPBA
No function
91
RPBA
No function
92
RPBA
Input
93
RESET_B
No function
94
RESET_B
No function
95
RESET_B
Input
96
EMU_B
Output
97
EMU_B
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-23
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
98
EMU_B
No function
99
DATA(0)
Output
100
DATA(0)
Output enable
101
DATA(0)
Input
102
DATA(1)
Output
103
DATA(1)
Output enable
104
DATA(1)
Input
105
DATA(2)
Output
106
DATA(2)
Output enable
107
DATA(2)
Input
108
DATA(3)
Output
109
DATA(3)
Output enable
110
DATA(3)
Input
111
DATA(4)
Output
112
DATA(4)
Output enable
113
DATA(4)
Input
114
DATA(5)
Output
115
DATA(5)
Output enable
10-24
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
116
DATA(5)
Input
117
DATA(6)
Output
118
DATA(6)
Output enable
119
DATA(6)
Input
120
DATA(7)
Output
121
DATA(7)
Output enable
122
DATA(7)
Input
123
DATA(8)
Output
124
DATA(8)
Output enable
125
DATA(8)
Input
126
DATA(9)
Output
127
DATA(9)
Output enable
128
DATA(9)
Input
129
DATA(10)
Output
130
DATA(10)
Output enable
131
DATA(10)
Input
132
DATA(11)
Output
133
DATA(11)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-25
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
134
DATA(11)
Input
135
DATA(12)
Output
136
DATA(12)
Output enable
137
DATA(12)
Input
138
DATA(13)
Output
139
DATA(13)
Output enable
140
DATA(13)
Input
141
DATA(14)
Output
142
DATA(14)
Output enable
143
DATA(14)
Input
144
DATA(15)
Output
145
DATA(15)
Output enable
146
DATA(15)
Input
147
DATA(16)
Output
148
DATA(16)
Output enable
149
DATA(16)
Input
150
DATA(17)
Output
151
DATA(17)
Output enable
10-26
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
152
DATA(17)
Input
153
DATA(18)
Output
154
DATA(18)
Output enable
155
DATA(18)
Input
156
DATA(19)
Output
157
DATA(19)
Output enable
158
DATA(19)
Input
159
DATA(20)
Output
160
DATA(20)
Output enable
161
DATA(20)
Input
162
DATA(21)
Output
163
DATA(21)
Output enable
164
DATA(21)
Input
165
DATA(22)
Output
166
DATA(22)
Output enable
167
DATA(22)
Input
168
DATA(23)
Output
169
DATA(23)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-27
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
170
DATA(23)
Input
171
DATA(24)
Output
172
DATA(24)
Output enable
173
DATA(24)
Input
174
DATA(25)
Output
175
DATA(25)
Output enable
176
DATA(25)
Input
177
DATA(26)
Output
178
DATA(26)
Output enable
179
DATA(26)
Input
180
DATA(27)
Output
181
DATA(27)
Output enable
182
DATA(27)
Input
183
DATA(28)
Output
184
DATA(28)
Output enable
185
DATA(28)
Input
186
DATA(29)
Output
187
DATA(29)
Output enable
10-28
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
188
DATA(29)
Input
189
DATA(30)
Output
190
DATA(30)
Output enable
191
DATA(30)
Input
192
DATA(31)
Output
193
DATA(31)
Output enable
194
DATA(31)
Input
195
DATA(32)
Output
196
DATA(32)
Output enable
197
DATA(32)
Input
198
DATA(33)
Output
199
DATA(33)
Output enable
200
DATA(33)
Input
201
DATA(34)
Output
202
DATA(34)
Output enable
203
DATA(34)
Input
204
DATA(35)
Output
205
DATA(35)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-29
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
206
DATA(35)
Input
207
DATA(36)
Output
208
DATA(36)
Output enable
209
DATA(36)
Input
210
DATA(37)
Output
211
DATA(37)
Output enable
212
DATA(37)
Input
213
DATA(38)
Output
214
DATA(38)
Output enable
215
DATA(38)
Input
216
DATA(39)
Output
217
DATA(39)
Output enable
218
DATA(39)
Input
219
DATA(40)
Output
220
DATA(40)
Output enable
221
DATA(40)
Input
222
DATA(41)
Output
223
DATA(41)
Output enable
10-30
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
224
DATA(41)
Input
225
DATA(42)
Output
226
DATA(42)
Output enable
227
DATA(42)
Input
228
DATA(43)
Output
229
DATA(43)
Output enable
230
DATA(43)
Input
231
DATA(44)
Output
232
DATA(44)
Output enable
233
DATA(44)
Input
234
DATA(45)
Output
235
DATA(45)
Output enable
236
DATA(45)
Input
237
DATA(46)
Output
238
DATA(46)
Output enable
239
DATA(46)
Input
240
DATA(47)
Output
241
DATA(47)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-31
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
242
DATA(47)
Input
243
CLK_CFG0
No function
244
CLK_CFG0
No function
245
CLK_CFG0
Input
246
CLK_CFG1
No function
247
CLK_CFG1
No function
248
CLK_CFG1
Input
249
CLK_CFG2
No function
250
CLK_CFG2
No function
251
CLK_CFG2
Input
252
CLK_CFG3
No function
253
CLK_CFG3
No function
254
CLK_CFG3
Input
255
CLKOUT
Output
256
CLKOUT
Output enable
257
CLKOUT
No function
258
DATA(48)
Output
259
DATA(48)
Output enable
10-32
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
260
DATA(48)
Input
261
DATA(49)
Output
262
DATA(49)
Output enable
263
DATA(49)
Input
264
DATA(50)
Output
265
DATA(50)
Output enable
266
DATA(50)
Input
267
DATA(51)
Output
268
DATA(51)
Output enable
269
DATA(51)
Input
270
DATA(52)
Output
271
DATA(52)
Output enable
272
DATA(52)
Input
273
DATA(53)
Output
274
DATA(53)
Output enable
275
DATA(53)
Input
276
DATA(54)
Output
277
DATA(54)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-33
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
278
DATA(54)
Input
279
DATA(55)
Output
280
DATA(55)
Output enable
281
DATA(55)
Input
282
DATA(56)
Output
283
DATA(56)
Output enable
284
DATA(56)
Input
285
DATA(57)
Output
286
DATA(57)
Output enable
287
DATA(57)
Input
288
DATA(58)
Output
289
DATA(58)
Output enable
290
DATA(58)
Input
291
DATA(59)
Output
292
DATA(59)
Output enable
293
DATA(59)
Input
294
DATA(60)
Output
295
DATA(60)
Output enable
10-34
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
296
DATA(60)
Input
297
DATA(61)
Output
298
DATA(61)
Output enable
299
DATA(61)
Input
300
DATA(62)
Output
301
DATA(62)
Output enable
302
DATA(62)
Input
303
DATA(63)
Output
304
DATA(63)
Output enable
305
DATA(63)
Input
306
ADDR(2)
Output
307
ADDR(2)
Output enable
308
ADDR(2)
Input
309
ADDR(3)
Output
310
ADDR(3)
Output enable
311
ADDR(3)
Input
312
ADDR(4)
Output
313
ADDR(4)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-35
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
314
ADDR(4)
Input
315
ADDR(5)
Output
316
ADDR(5)
Output enable
317
ADDR(5)
Input
318
ADDR(6)
Output
319
ADDR(6)
Output enable
320
ADDR(6)
Input
321
ADDR(7)
Output
322
ADDR(7)
Output enable
323
ADDR(7)
Input
324
ADDR(8)
Output
325
ADDR(8)
Output enable
326
ADDR(8)
Input
327
ADDR(9)
Output
328
ADDR(9)
Output enable
329
ADDR(9)
Input
330
ADDR(10)
Output
331
ADDR(10)
Output enable
10-36
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
332
ADDR(10)
Input
333
ADDR(11)
Output
334
ADDR(11)
Output enable
335
ADDR(11)
Input
336
ADDR(12)
Output
337
ADDR(12)
Output enable
338
ADDR(12)
Input
339
ADDR(13)
Output
340
ADDR(13)
Output enable
341
ADDR(13)
Input
342
ADDR(14)
Output
343
ADDR(14)
Output enable
344
ADDR(14)
Input
345
ADDR(15)
Output
346
ADDR(15)
Output enable
347
ADDR(15)
Input
348
ADDR(16)
Output
349
ADDR(16)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-37
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
350
ADDR(16)
Input
351
ADDR(17)
Output
352
ADDR(17)
Output enable
353
ADDR(17)
Input
354
ADDR(18)
Output
355
ADDR(18)
Output enable
356
ADDR(18)
Input
357
ADDR(19)
Output
358
ADDR(19)
Output enable
359
ADDR(19)
Input
360
ADDR(20)
Output
361
ADDR(20)
Output enable
362
ADDR(20)
Input
363
ADDR(21)
Output
364
ADDR(21)
Output enable
365
ADDR(21)
Input
366
ADDR(22)
Output
367
ADDR(22)
Output enable
10-38
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
368
ADDR(22)
Input
369
ADDR(23)
Output
370
ADDR(23)
Output enable
371
ADDR(23)
Input
372
ADDR(24)
Output
373
ADDR(24)
Output enable
374
ADDR(24)
Input
375
ADDR(25)
Output
376
ADDR(25)
Output enable
377
ADDR(25)
Input
378
ADDR(26)
Output
379
ADDR(26)
Output enable
380
ADDR(26)
Input
381
ADDR(27)
Output
382
ADDR(27)
Output enable
383
ADDR(27)
Input
384
ADDR(28)
Output
385
ADDR(28)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-39
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
386
ADDR(28)
Input
387
ADDR(29)
Output
388
ADDR(29)
Output enable
389
ADDR(29)
Input
390
ADDR(30)
Output
391
ADDR(30)
Output enable
392
ADDR(30)
Input
393
ADDR(31)
Output
394
ADDR(31)
Output enable
395
ADDR(31)
Input
396
ID0
No function
397
ID0
No function
398
ID0
Input
399
ID1
No function
400
ID1
No function
401
ID1
Input
402
ID2
No function
403
ID2
No function
10-40
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
404
ID2
Input
405
ADDR(0)
Output
406
ADDR(0)
Output enable
407
ADDR(0)
Input
408
ADDR(1)
Output
409
ADDR(1)
Output enable
410
ADDR(1)
Input
411
BRST
Output
412
BRST
Output enable
413
BRST
Input
414
BMS_B
Output
415
BMS_B
Output enable
416
BMS_B
Input
417
MS_B(0)
Output
418
MS_B(0)
Output enable
419
MS_B(0)
No function
420
MS_B(1)
Output
421
MS_B(1)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-41
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
422
MS_B(1)
No function
423
MS_B(2)
Output
424
MS_B(2)
Output enable
425
MS_B(2)
No function
426
MS_B(3)
Output
427
MS_B(3)
Output enable
428
MS_B(3)
No function
429
CIF_B
Output
430
CIF_B
Output enable
431
CIF_B
No function
432
CS_B
No function
433
CS_B
No function
434
CS_B
Input
435
WRH_B
Output
436
WRH_B
Output enable
437
WRH_B
Input
438
WRL_B
Output
439
WRL_B
Output enable
10-42
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
440
WRL_B
Input
441
RDH_B
Output
442
RDH_B
Output enable
443
RDH_B
Input
444
RDL_B
Output
445
RDL_B
Output enable
446
RDL_B
Input
447
DMAG1_B
Output
448
DMAG1_B
Output enable
449
DMAG1_B
Input
450
DMAG2_B
Output
451
DMAG2_B
Output enable
452
DMAG2_B
Input
453
DMAR1_B
Output
454
DMAR1_B
Output enable
455
DMAR1_B
Input
456
DMAR2_B
Output
457
DMAR2_B
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-43
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
458
DMAR2_B
Input
459
LBOOT
No function
460
LBOOT
No function
461
LBOOT
Input
462
EBOOT
No function
463
EBOOT
No function
464
EBOOT
Input
465
L5DAT(0)
Output
466
L5DAT(0)
Output enable
467
L5DAT(0)
Input
468
L5DAT(1)
Output
469
L5DAT(1)
Output enable
470
L5DAT(1)
Input
471
L5DAT(2)
Output
472
L5DAT(2)
Output enable
473
L5DAT(2)
Input
474
L5DAT(3)
Output
475
L5DAT(3)
Output enable
10-44
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
476
L5DAT(3)
Input
477
L5ACK
Output
478
L5ACK
Output enable
479
L5ACK
Input
480
L5CLK
Output
481
L5CLK
Output enable
482
L5CLK
Input
483
L5DAT(4)
Output
484
L5DAT(4)
Output enable
485
L5DAT(4)
Input
486
L5DAT(5)
Output
487
L5DAT(5)
Output enable
488
L5DAT(5)
Input
489
L5DAT(6)
Output
490
L5DAT(6)
Output enable
491
L5DAT(6)
Input
492
L5DAT(7)
Output
493
L5DAT(7)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-45
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
494
L5DAT(7)
Input
495
L4DAT(0)
Output
496
L4DAT(0)
Output enable
497
L4DAT(0)
Input
498
L4DAT(1)
Output
499
L4DAT(1)
Output enable
500
L4DAT(1)
Input
501
L4DAT(2)
Output
502
L4DAT(2)
Output enable
503
L4DAT(2)
Input
504
L4DAT(3)
Output
505
L4DAT(3)
Output enable
506
L4DAT(3)
Input
507
L4ACK
Output
508
L4ACK
Output enable
509
L4ACK
Input
510
L4CLK
Output
511
L4CLK
Output enable
10-46
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
512
L4CLK
Input
513
L4DAT(4)
Output
514
L4DAT(4)
Output enable
515
L4DAT(4)
Input
516
L4DAT(5)
Output
517
L4DAT(5)
Output enable
518
L4DAT(5)
Input
519
L4DAT(6)
Output
520
L4DAT(6)
Output enable
521
L4DAT(6)
Input
522
L4DAT(7)
Output
523
L4DAT(7)
Output enable
524
L4DAT(7)
Input
525
L3DAT(0)
Output
526
L3DAT(0)
Output enable
527
L3DAT(0)
Input
528
L3DAT(1)
Output
529
L3DAT(1)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-47
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
530
L3DAT(1)
Input
531
L3DAT(2)
Output
532
L3DAT(2)
Output enable
533
L3DAT(2)
Input
534
L3DAT(3)
Output
535
L3DAT(3)
Output enable
536
L3DAT(3)
Input
537
L3ACK
Output
538
L3ACK
Output enable
539
L3ACK
Input
540
L3CLK
Output
541
L3CLK
Output enable
542
L3CLK
Input
543
L3DAT(4)
Output
544
L3DAT(4)
Output enable
545
L3DAT(4)
Input
546
L3DAT(5)
Output
547
L3DAT(5)
Output enable
10-48
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
548
L3DAT(5)
Input
549
L3DAT(6)
Output
550
L3DAT(6)
Output enable
551
L3DAT(6)
Input
552
L3DAT(7)
Output
553
L3DAT(7)
Output enable
554
L3DAT(7)
Input
555
SBTS_B
No function
556
SBTS_B
No function
557
SBTS_B
Input
558
PAGE
Output
559
PAGE
Output enable
560
PAGE
No function
561
PA_B
Output
562
PA_B
Output enable
563
PA_B
Input
564
REDY
Output
565
REDY
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-49
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
566
REDY
No function
567
ACK
Output
568
ACK
Output enable
569
ACK
Input
570
BR_B(1)
Output
571
BR_B(1)
Output enable
572
BR_B(1)
Input
573
BR_B(2)
Output
574
BR_B(2)
Output enable
575
BR_B(2)
Input
576
BR_B(3)
Output
577
BR_B(3)
Output enable
578
BR_B(3)
Input
579
BR_B(4)
Output
580
BR_B(4)
Output enable
581
BR_B(4)
Input
582
BR_B(5)
Output
583
BR_B(5)
Output enable
10-50
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
584
BR_B(5)
Input
585
BR_B(6)
Output
586
BR_B(6)
Output enable
587
BR_B(6)
Input
588
HBR_B
No function
589
HBR_B
No function
590
HBR_B
Input
591
HBG_B
Output
592
HBG_B
Output enable
593
HBG_B
Input
594
L2DAT(0)
Output
595
L2DAT(0)
Output enable
596
L2DAT(0)
Input
597
L2DAT(1)
Output
598
L2DAT(1)
Output enable
599
L2DAT(1)
Input
600
L2DAT(2)
Output
601
L2DAT(2)
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-51
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
602
L2DAT(2)
Input
603
L2DAT(3)
Output
604
L2DAT(3)
Output enable
605
L2DAT(3)
Input
606
L2ACK
Output
607
L2ACK
Output enable
608
L2ACK
Input
609
L2CLK
Output
610
L2CLK
Output enable
611
L2CLK
Input
612
L2DAT(4)
Output
613
L2DAT(4)
Output enable
614
L2DAT(4)
Input
615
L2DAT(5)
Output
616
L2DAT(5)
Output enable
617
L2DAT(5)
Input
618
L2DAT(6)
Output
619
L2DAT(6)
Output enable
10-52
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
620
L2DAT(6)
Input
621
L2DAT(7)
Output
622
L2DAT(7)
Output enable
623
L2DAT(7)
Input
624
L1DAT(0)
Output
625
L1DAT(0)
Output enable
626
L1DAT(0)
Input
627
L1DAT(1)
Output
628
L1DAT(1)
Output enable
629
L1DAT(1)
Input
630
L1DAT(2)
Output
631
L1DAT(2)
Output enable
632
L1DAT(2)
Input
633
L1DAT(3)
Output
634
L1DAT(3)
Output enable
635
L1DAT(3)
Input
636
L1ACK
Output
637
L1ACK
Output enable
ADSP-21160 SHARC DSP Hardware Reference
10-53
Boundary Register
Table 10-7. JTAG Boundary
Register (Cont’d)
Scan
#
Signal Name
Latch Type
638
L1ACK
Input
639
L1CLK
Output
640
L1CLK
Output enable
641
L1CLK
Input
642
L1DAT(4)
Output
643
L1DAT(4)
Output enable
644
L1DAT(4)
Input
645
L1DAT(5)
Output
646
L1DAT(5)
Output enable
647
L1DAT(5)
Input
648
L1DAT(6)
Output
649
L1DAT(6)
Output enable
650
L1DAT(6)
Input
651
L1DAT(7)
Output
652
L1DAT(7)
Output enable
653
L1DAT(7)
Input
654
SPARE
No function
10-54
ADSP-21160 SHARC DSP Hardware Reference
JTAG Test Emulation Port
Device Identification Register
No device identification register is included in the ADSP-21160 DSP.
Built-in Self-test Operation (BIST)
No self-test functions are supported by the ADSP-21160 DSP.
Private Instructions
Table 10-2 on page 10-4 lists the private instructions that are reserved for
emulation and memory test. The ADSP-21160 EZ-ICE emulator uses the
TAP and boundary scan as a way to access the processor in the target system. The EZ-ICE emulator requires a target board connector for access to
the TAP. For more information, see “Designing for JTAG Emulation” on
page 11-24.
References
• IEEE Standard 1149.1-1990. Standard Test Access Port and
Boundary-Scan Architecture.
To order a copy, contact IEEE at 1-800-678-IEEE.
• Maunder, C.M. and R. Tulloss. Test Access Ports and Boundary
Scan Architectures.
IEEE Computer Society Press, 1991.
• Parker, Kenneth. The Boundary Scan Handbook.
Kluwer Academic Press, 1992.
ADSP-21160 SHARC DSP Hardware Reference
10-55
References
• Bleeker, Harry, P. van den Eijnden, and F. de Jong. Boundary-Scan Test—A Practical Approach.
Kluwer Academic Press, 1993.
• Hewlett-Packard Co. HP Boundary-Scan Tutorial and BSDL Reference Guide.
(HP part# E1017-90001.) 1992.
10-56
ADSP-21160 SHARC DSP Hardware Reference
System Design
11 SYSTEM DESIGN
The DSP supports many system design options. The options implemented
in a system are influenced by cost, performance, and system requirements.
This chapter provides the following system design information:
• “DSP Pin Descriptions” on page 11-2
• “Designing for JTAG Emulation” on page 11-24
• “Conditioning Input Signals” on page 11-34
• “Designing For High Frequency Operation” on page 11-36
• “Booting Single and Multiple Processors” on page 11-48
Other chapters also discuss system design issues. Some other locations for
system design information include:
• “Setting External Port Modes” on page 7-2
• “Setting Link Port Modes” on page 8-5
• “Setting Serial Port Modes” on page 9-6
ADSP-21160 SHARC DSP Hardware Reference
11-1
DSP Pin Descriptions
DSP Pin Descriptions
This section describes the pins of the DSP and shows how these signals
can be used in a DSP system. Figure 11-1 illustrates how the pins are used
in a single-processor system. Figure 7-26 on page 7-93 shows a system diagram illustrating pin connections in an DSP multiprocessor cluster.
ADSP-2116X
CLKIN
4
FLAG3-0
TIMEXP
LINK
DEVICES
(6 MAX)
(OPTIONAL)
LXCLK
LXACK
LXDAT7-0
SERIAL
SERIAL
DEVICE
DEVICE
(OPTIONAL)
(OPTIONAL)
TCLK0
RCLK0
TFS0
RSF0
DT0
DR0
SERIAL
SERIAL
DEVICE
(OPTIONAL)
(OPTIONAL)
TCLK1
RCLK1
TFS1
RSF1
DT1
DR1
RPBA
ID2-0
RESET
ADDR
CIF
DATA
BRST
ADDR31-0
ADDR
DATA63-0
DATA
OE
OE
WE
WE
RDX
WRX
ACK
ACK
CS
MS3-0
PAGE
SBTS
CLKOUT
DMAR1-2
BOOT
EPROM
(OPTIONAL)
MEMORY
AND
PERIPHERALS
(OPTIONAL)
DATA
3
EBOOT
LBOOT
IRQ2-0
CS
CS
BMS
CLK_CFG3-0
ADDRESS
4
CONTROL
CLOCK
DMA DEVICE
(OPTIONAL)
DATA
DMAG1-2
CS
HBR
HBG
REDY
HOST
PROCESSOR
INTERFACE
(OPTIONAL)
BR1-6
ADDR
PA
DATA
JTAG
6
Figure 11-1. Single-Processor DSP System
11-2
ADSP-21160 SHARC DSP Hardware Reference
System Design
DSP pin definitions are listed in Table 11-1. The following symbols
appear in the Type column of Table 11-1:
Table 11-1. Pin Descriptions
Pin
Type
Function
ADDR31-0
I/O/T
External Bus Address. The DSP outputs addresses for external memory and peripherals on these pins. In a multiprocessor system, the bus
master outputs addresses for read/writes of the internal memory or I/O
processor registers of other DSPs. The DSP inputs addresses when a
host processor or multiprocessing bus master is reading or writing its
internal memory or I/O processor registers. A keeper latch on the
DSP’s ADDR31-0 pins maintains the input at the level it was last driven
(only enabled on the DSP with ID2-0=00x).
DATA63-0
I/O/T
External Bus Data. The DSP inputs and outputs data and instructions
on these pins. Pull-up resistors on unused DATA pins are not necessary.
A keeper latch on the DSP’s DATA63-0 pins maintains the input at the
level it was last driven (only enabled on the DSP with ID2-0=00x).
MS3-0
O/T
Memory Select Lines. These outputs are asserted (low) as chip selects
for the corresponding banks of external memory. Memory bank size
must be defined in the SYSCON control register. The MS3-0 outputs are
decoded memory address lines. In asynchronous access mode, the
MS3-0 outputs transition with the other address outputs. In synchronous access modes, the MS3-0 outputs assert with the other address
lines. They deassert after the first CLKIN cycle in which ACK is sampled
asserted. MS3-0 has a 20k internal pull up resistor that is enabled on
the DSP with ID2-0=00x.
CIF
O
Core Instruction Fetch. Signal is active low when an external instruction fetch is performed. Driven by bus master only. Three-state when
host is bus master. CIF has a 20k internal pull up resistor that is
enabled on the DSP with ID2-0=00x.
RDL
I/O/T
Memory Read Low Strobe. RDL is asserted whenever DSP reads from
the low word of external memory or from the internal memory of
other DSPs. External devices, including other DSPs, must assert RDL
for reading from low word of DSP internal memory. In a multiprocessing system, RDL is driven by the bus master RDL has a 20k internal
pull up resistor that is enabled on the DSP with ID2-0=00x.
ADSP-21160 SHARC DSP Hardware Reference
11-3
DSP Pin Descriptions
Table 11-1. Pin Descriptions (Cont’d)
Pin
Type
Function
RDH
I/O/T
Memory Read High Strobe. RDH is asserted whenever DSP reads from
the high word of external memory or from the internal memory of
other DSPs. External devices, including other DSPs, must assert RDH
for reading from the high word of DSP internal memory. In a multiprocessing system, RDH is driven by the bus master. RDH has a 20k
internal pull up resistor that is enabled on the DSP with ID2-0=00x.
WRL
I/O/T
Memory Write Low Strobe. WRL is asserted when DSP writes to the
low word of external memory or internal memory of other DSPs.
External devices must assert WRL for writing to DSP's low word of
internal memory. In a multiprocessing system, WRL is driven by the bus
master. WRL has a 20k internal pull up resistor that is enabled on the
DSP with ID2-0=00x.
WRH
I/O/T
Memory Write High Strobe. WRH is asserted when DSP writes to the
high word of external memory or internal memory of other DSPs.
External devices must assert WRH for writing to DSP's high word of
internal memory. In a multiprocessing system, WRH is driven by the bus
master. WRH has a 20k internal pull up resistor that is enabled on the
DSP with ID2-0=00x.
BRST
I/O/T
Sequential burst access. BRST is asserted by DSP or a host to indicate
that data associated with consecutive addresses is being read or written.
A slave device samples the initial address and increments an internal
address counter after each transfer. The incremented address is not
pipelined on the bus. If the burst access is a read from the DSP by a
host, DSP increments the address automatically as long as BRST is
asserted. BRST is asserted after the initial access of a burst transfer. It is
asserted for every cycle after that, except for the last data request cycle
(denoted by RDx or WRx asserted and BRST negated). A keeper latch on
the DSP’s BRST pin maintains the input at the level it was last driven
(only enabled on the DSP with ID2-0=00x).
PAGE
O/T
DRAM Page Boundary. The DSP asserts this pin to signal that an
external DRAM page boundary has been crossed. DRAM page size
must be defined in the DSP's memory control register (WAIT).
DRAM can only be implemented in external memory Bank 0. The
PAGE signal can only be activated for Bank 0 accesses. In a multiprocessing system PAGE is output by the bus master. A keeper latch on the
DSP’s PAGE pin maintains the output at the level it was last driven
(only enabled on the DSP with ID2-0=00x).
11-4
ADSP-21160 SHARC DSP Hardware Reference
System Design
Table 11-1. Pin Descriptions (Cont’d)
Pin
Type
Function
ACK
I/O/S
Memory Acknowledge. External devices can deassert ACK (low) to add
wait states to an external memory access. ACK is used by I/O devices,
memory controllers, or other peripherals to hold off completion of an
external memory access. The DSP deasserts ACK as an output to add
wait states to a synchronous access of its internal memory. ACK has a
2k internal pull up resistor that is enabled on the DSP with
ID2-0=00x
SBTS
I/S
Suspend Bus and Three-State. External devices can assert SBTS (low)
to place the external bus address, data, selects, and strobes in a high
impedance state for the following cycle. If the DSP attempts to access
external memory while SBTS is asserted, the processor halts and the
memory access does not complete until SBTS is deasserted. SBTS
should only be used to recover from host processor/DSP deadlock or
used with a DRAM controller.
IRQ2-0
I/A
Interrupt Request Lines. These are sampled on the rising edge of
CLKIN and may be either edge-triggered or level-sensitive.
FLAG3-0
I/O/A
Flag Pins. Each is configured via control bits as either an input or output. As an input, it can be tested as a condition. As an output, it can be
used to signal external peripherals.
TIMEXP
O
Timer Expired. Asserted for four core clock cycles when the timer is
enabled, and TCOUNT decrements to zero.
HBR
I/A
Host Bus Request. Must be asserted by a host processor to request
control of the DSP's external bus. When HBR is asserted in a multiprocessing system, the DSP that is bus master relinquishes the bus and
asserts HBG. To relinquish the bus, the DSP places the address, data,
select, and strobe lines in a high impedance state. HBR has priority over
all DSP bus requests (BR6-1) in a multiprocessing system.
HBG
I/O
Host Bus Grant. Acknowledges an HBR bus request, indicating that the
host processor may take control of the external bus. HBG is asserted
(held low) by the ADSP-21160 until HBR is released. In a multiprocessing system, HBG is output by the DSP bus master and is monitored by
all others.
After HBR is asserted and before HBG is deasserted, HBG will float for one
CLKIN cycle. To avoid erroneous grants, HBG should be pulled up with
a 20k to 50k external resistor.
ADSP-21160 SHARC DSP Hardware Reference
11-5
DSP Pin Descriptions
Table 11-1. Pin Descriptions (Cont’d)
Pin
Type
Function
CS
I/A
Chip Select. Asserted by host processor to select the DSP.
REDY
O (O/D) Host Bus Acknowledge. The DSP deasserts REDY (low) to add waitstates to a host access when CS and HBR inputs are asserted.
DMAR1
I/A
DMA Request 1 (DMA Channel 11). Asserted by external port devices
to request DMA services. DMAR1 has a 20k internal pull up resistor
that is enabled on the DSP with ID2-0=00x.
DMAR2
I/A
DMA Request 2 (DMA Channel 12). Asserted by external port devices
to request DMA services. DMAR2 has a 20k internal pull up resistor
that is enabled on the DSP with ID2-0=00x
DMAG1
O/T
DMA Grant 1 (DMA Channel 11). Asserted by DSP to indicate that
the requested DMA starts on the next cycle. Driven by bus master
only. DMAG1 has a 20k internal pull up resistor that is enabled on the
DSP with ID2-0=00x
DMAG2
O/T
DMA Grant 2 (DMA Channel 12). Asserted by DSP to indicate that
the requested DMA starts on the next cycle. Driven by bus master
only. DMAG2 has a 20k internal pull up resistor that is enabled on the
DSP with ID2-0=00x
BR6-1
I/O/S
Multiprocessing Bus Requests. Used by multiprocessing DSPs to arbitrate for bus mastership. Each DSP only drives its own BRx line (corresponding to the value of its ID2-0 inputs) and monitors all others. In a
multiprocessor system with less than six DSPs, the unused BRx pins
should be pulled high. The processor's own BRx line must not be
pulled high or low because it is an output.
ID2-0
I
Multiprocessing ID. Determines which multiprocessing bus request
(BR1-BR6) is used by each DSP. ID = 001 corresponds to BR1,
ID = 010 corresponds to BR2, and so on. Use ID = 000 in single-processor systems. These lines are a system configuration selection which
should be hardwired or only changed at reset.
RPBA
I/S
Rotating Priority Bus Arbitration Select. When RPBA is high, rotating
priority for multiprocessor bus arbitration is selected. When RPBA is
low, fixed priority is selected. This signal is a system configuration
selection that must be set to the same value on every DSP. If the value
of RPBA is changed during system operation, it must be changed in the
same CLKIN cycle on every DSP.
11-6
ADSP-21160 SHARC DSP Hardware Reference
System Design
Table 11-1. Pin Descriptions (Cont’d)
Pin
Type
Function
PA
I/O/T
Priority Access. Asserting its PA pin allows an DSP bus slave to interrupt background DMA transfers and gain access to the external bus. PA
is connected to all DSPs in the system. If access priority is not required
in a system, the PA pin should be left unconnected. DMAR1 has a 20k
internal pull up resistor that is enabled on the DSP with ID2-0=00x
DTx
O
Data Transmit (Serial Ports 0, 1). Each DT pin has a 50k internal
pull-up resistor.
DRx
I
Data Receive (Serial Ports 0, 1). Each DR pin has a 50k internal
pull-up resistor.
TCLKx
I/O
Transmit Clock (Serial Ports 0, 1). Each TCLK pin has a 50k internal
pull-up resistor.
RCLKx
I/O
Receive Clock (Serial Ports 0, 1). Each RCLK pin has a 50k internal
pull-up resistor.
TFSx
I/O
Transmit Frame Sync (Serial Ports 0, 1).
RFSx
I/O
Receive Frame Sync (Serial Ports 0, 1).
LxDAT7-0
I/O
Link Port Data (Link Ports 0-5). Each LxDAT pin has an internal
pull-down resistor that is enabled or disabled by the LPDRD bit of the
LCTL0-1 register.
LxCLK
I/O
Link Port Clock (Link Ports 0-5). Each LxCLK pin has an internal
pull-down resistor that is enabled or disabled by the LPDRD bit of the
LCTL0-1 register.
LxACK
I/O
Link Port Acknowledge (Link Ports 0-5). Each LxACK pin has an internal pull-down resistor that is enabled or disabled by the LPDRD bit of
the LCTL0-1 register.
EBOOT
I
EPROM Boot Select. For a description of how this pin operates, see
the BMS pin description. This signal is a system configuration selection
that should be hardwired.
LBOOT
I
Link Boot. For a description of how this pin operates, see the BMS pin
description. This signal is a system configuration selection that should
be hardwired.
ADSP-21160 SHARC DSP Hardware Reference
11-7
DSP Pin Descriptions
Table 11-1. Pin Descriptions (Cont’d)
Pin
Type
Function
BMS
I/O/T
Boot Memory Select. Serves as an output or input as selected with the
EBOOT and LBOOT pins as shown below. This input is a system configuration selection that should be hardwired.
EBOOTLBOOT BMS Booting Mode
10OutputEPROM (Connect BMS to
EPROM chip select.)
001 (Input)Host Processor
011 (Input)Link Port
000 (Input)No Booting. Processor executes
from external memory
010 (Input)Reserved
11x (Input)Reserved
CLKIN
I
Local Clock In. CLKIN is the DSP clock input. The DSP external port
cycles at the frequency of CLKIN. The CLKIN frequency is multiplied
by a ratio (CLK_CFG3-0) to select the instruction cycle rate, which is
programmable at powerup. CLKIN may not be halted, changed, or
operated below the specified frequency.
CLK_
CFG3-0
I
Core/CLKIN Ratio Control. DSP core clock (instruction cycle) rate is
equal to n x CLKIN where n is user selectable to 2, 3, or 4, using the
CLK_CFG3-0 inputs.
CLKOUT
O/T
Local Clock Out. CLKOUT is driven at the CLKIN frequency by the
DSP. This output is three-stated by setting the COD bit in the SYSCON
register. A keeper latch on the DSP’s CLKOUT pin maintains the output
at the level it was last driven (only enabled on the DSP with
ID2-0=00x).
RESET
I/A
Processor Reset. Resets the DSP to a known state and begins execution
at the program memory location specified by the hardware reset vector
address. The RESET input must be asserted (low) at power-up.
The only difference between the soft and hard reset is that the external
bus arbitration is not affected by a soft reset. The PLL does not get
reset by a soft reset.
TCK
I
Test Clock (JTAG). Provides a clock for JTAG boundary scan.
TMS
I/S
Test Mode Select (JTAG). Used to control the test state machine.
TMS has an internal pull-up resistor.
TDI
I/S
Test Data Input (JTAG). Provides serial data for the boundary scan
logic. TDI has an internal pull-up resistor.
11-8
ADSP-21160 SHARC DSP Hardware Reference
System Design
Table 11-1. Pin Descriptions (Cont’d)
Pin
Type
Function
TDO
O
Test Data Output (JTAG). Serial scan output of the boundary scan
path.
TRST
I/A
Test Reset (JTAG). Resets the test state machine. TRST must be
asserted (pulsed low) after power-up or held low for proper operation
of the DSP. TRST has an internal pull-up resistor.
EMU
O (o/d)
Emulation Status. Must be connected to the DSP target board connector only.
VDDINT
P
Processor Core Power Supply. Nominally +2.5V DC (ADSP-21160M
DSP) or +1.9V DC (ADSP-21160N DSP). (40 pins).
VDDEXT
P
I/O Power Supply; Nominally +3.3V DC. (43 pins).
AVDD
P
Analog Power Supply; Nominally +2.5V DC (ADSP-21160M DSP)
or +1.9V DC (ADSP-21160N DSP). It supplies the DSP’s internal
PLL (clock generator). This pin has the same specifications as VDDINT,
except that added filtering circuitry is required. For more information
on supply specifications, see ADSP-21160 DSP Microcomputer Data
Sheet.
AGND
G
Analog Power Supply Return.
GND
G
Power Supply Return. (82 pins).
NC
Do Not Connect. Reserved pins which must be left open and unconnected. (9 pins).
A
Asynchronous
G
Ground
I
Input
O
Output
P
Power Supply
S
Synchronous
(a/d)
Active Drive
ADSP-21160 SHARC DSP Hardware Reference
11-9
DSP Pin Descriptions
(o/d)
Open Drain
T
Three-State (when SBTS is asserted or DSP is bus slave)
Figure 11-2 shows how different data word sizes are transferred over the
external port.
Inputs identified as synchronous (S) must meet timing requirements with
respect to CLKIN (or with respect to TCK for TMS, TDI). Inputs identified as
asynchronous (A) can be asserted asynchronously to CLKIN (or to TCK for
TRST).
DATA63-0
63
55
47
39
31
23
15
BYTE 7
7
0
BYTE 0
RDL/WRL
RDH/WRH
64-BIT LONG WORD, SIMD, OR DMA TRANSFERS
64-BIT TRANSFER FOR 48-BIT INSTRUCTION FETCH
64-BIT TRANSFER FOR 40-BIT EXT. PREC.
32-BIT NORMAL WORD(EVEN ADDR)
32-BIT NORMAL WORD (ODD ADDR)
RESTRICTED DMA, HOST, EPROM DATA ALIGNMENTS:
32-BIT PACKED
16-BIT PACKED
EPROM
Figure 11-2. External Port Data Alignment
11-10
ADSP-21160 SHARC DSP Hardware Reference
System Design
Unused inputs should be tied or pulled to VDD or GND, except for the
following:
•
ADDR31-0, DATA63-0, BRST, PAGE, CLKOUT
—These pins have a
logic-level hold circuit enabled on the DSP with ID2-0=00x that
prevents input from floating internally.
•
PA,
•
LxDAT7-0 (LxPRDE=0), LxCLK, LxACK—These
pins have an internal
pull-down resistor that is controlled by bit settings in the LCTLx
register.
•
DTx, DRx, TCLKx, RCLKx,EMU—These
ACK, MS3-0, CIF, RDH/L, WRH/L, DMARx, DMAGx— These pins
have a pull-up resistor enabled on the DSP with ID2-0=00x.
pins have a 50k pull-up
resistor.
•
TMS, TRST, TDI—These
pins have a 20k pull-up resistor.
input of the JTAG interface must be asserted (pulsed
 The
low) or held low after power-up for proper operation of the DSP.
TRST
Do not leave this pin unconnected.
Additional Notes:
• In single-processor systems, the DSP owns the external bus during
reset and does not perform bus arbitration to gain control of the
bus.
• Operation of the RDH/L and WRH/L signals changes when CS is
asserted by a host processor. For more information, see “Asynchronous Transfers” on page 7-57 and “Synchronous Transfers” on
page 7-62.
• Except during a Host Transition Cycle (HTC), the RDH/L and
WRH/L strobes should not be deasserted (low-to-high transition)
while ACK or REDY are deasserted (low)—the DSP hangs if this
happens.
ADSP-21160 SHARC DSP Hardware Reference
11-11
DSP Pin Descriptions
• In multiprocessor systems, the ACK signal is an input to the DSP
bus master and does not float when it is not being driven, because
the bus master maintains a pull-up on the pin. During reset, the
ACK pin is pulled high internally with a 2 k equivalent resistor by
the DSP bus master and is held high with the internal pull-up resistor. It is not necessary to use an external pull-up resistor on the ACK
line during booting or at any other time.
• For multiprocessor systems, PAGE is guaranteed to be asserted for
the first true access after acquiring bus mastership. PAGE is not
updated or asserted for multiprocessor memory space accesses or
external memory space accesses to any bank other than Bank 0.
• The HBR input is disabled during any access in which the PAGE signal is asserted. This prevents the possibility of the DSP becoming a
bus slave while a DRAM controller is servicing a page change.
Pin States At Reset
Table 11-2 shows the DSP pin states during and after reset.
Table 11-2. Post RESET Pin States
Pin
Type
State During and After RESET
ADDR31-0
I/O/T
Driven1
MS3-0
O/T
Driven High1
CIF
O/T
Driven High1
RDH
I/O/T
Driven High1
RDL
I/O/T
Driven High1
WRH
I/O/T
Driven High1
WRL
I/O/T
Driven High1
BRST
I/O/T
Driven Low1
11-12
ADSP-21160 SHARC DSP Hardware Reference
System Design
Table 11-2. Post RESET Pin States (Cont’d)
Pin
Type
State During and After RESET
PAGE
O/T
Driven Low1
CLKOUT
O/T
Driven
ACK
I/O/S/T
Pulled High by Bus Master (w/ 2 k internal pull-up resistor)1
HBG
I/O/S/T
Driven High1
DMAG1
O/T
Driven High1
DMAG2
O/T
Driven High
BR6-1
I/O
BR1
DATA63-0
I/O/T
Three-state1
SBTS
I/S
Input; causes the master to three-state during reset2
IRQ2-0
I/A
Inputs2
FLAG3-0
I/O/A
Inputs2
TIMEXP
O
Driven Low2
HBR
I/A
Input2
CS
I
Input2
REDY (o/d)
O
Three-state2
DMAR1
I
Input2
DMAR2
I
Input2
ID2-0
I
Inputs2
RPBA
I/S
Input2
I/O
Three-state2
EBOOT
I
Input2
LBOOT
I
Input2
BMS
I/O/T
Input2
PA
(o/d)
Driven Low if Bus Master, Otherwise Driven High1
ADSP-21160 SHARC DSP Hardware Reference
11-13
DSP Pin Descriptions
Table 11-2. Post RESET Pin States (Cont’d)
Pin
Type
State During and After RESET
CLKIN
I
Input
RESET
I/A
Input2
DTx
O
Three-state (for multichannel)3
DRx
I
Input3
TCLKx
I/O
Three-state3
RCLKx
I/O
Three-state3
TFSx
I/O
Three-state3
RFSx
I/O
Three-state3
LxDAT7-0
I/O
Three-state3
LxCLK
I/O
Three-state3
LxACK
I/O
Three-state3
TCK
I
Input4
TMS
I/S
Input4
TDI
I/S
Input4
TDO
O
Three-state4
TRST
I/A
Input4
EMU
O (o/d)
Three-state4
1
2
3
4
11-14
Driven only by DSP bus master, otherwise three-stated
Bus master independent
Serial ports and link ports
JTAG interface
ADSP-21160 SHARC DSP Hardware Reference
System Design
Clock Derivation
The DSP employs a phase-locked loop on-chip, to provide clocks that
switch at higher frequencies than the system clock (CLKIN). The PLL-based
clocking methodology employed on the DSP influences the clock frequencies and behavior for the serial, link, and external ports; in addition to the
processor core and internal memory. In each case, the DSP PLL provides a
de-skewed clock to the port logic and I/O pins.
For the external port, this clock is fedback to the PLL, such that the external port clock always switches at the CLKIN frequency. The PLL provides
internal clocks that switch at multiples of the CLKIN frequency for the
internal memory, processor core, link and serial ports, and to modify the
external port timing as required (for example, read/write strobes in asynchronous modes). The ratio of processor core clock frequency and
CLKIN/external port clock frequency is determined by the CLK_CFG3-0 pins
(as shown in Table 11-3 on page 11-17), during reset.
core clock ratio cannot be altered dynamically. The DSP must
 The
be reset to alter the clock ratio.
The PLL provides a clock that switches at the processor core frequency to
the serial and link ports. Each of the serial and link ports can be programmed to operate at clock frequencies derived from this clock. The two
serial ports' transmit and receive clocks are divided down from the processor core clock frequency by setting the TDIVx and RDIVx registers
appropriately. For more information, see “SPORT Transmit Divisor Registers (TDIVx)” on page A-76 and “SPORT Receive Divisor Registers
(RDIVx)” on page A-77.
Each of the six link port clock frequencies are determined by programming the LxCLKDx parameters in the LCTL registers. For more information,
see “Link Port Buffer Control Registers (LCTLx)” on page A-62.
The following shows an example clock derivation:
ADSP-21160 SHARC DSP Hardware Reference
11-15
DSP Pin Descriptions
Definition of terms:
tCK = CLKIN clock period (and external port clock period)
tCCLK = (processor) core clock period
tLCLK = link port clock period
tSCLK = serial port clock period
Clock ratios:
cRTO = core/CLKIN ratio, (2, 3, or 4:1, determined by CLK_CFG)
lRTO = lport/core clock ratio (1:4, 1:2, 1:3, or 1:1, determined
by LxCLKD)
sRTO = sport/core clock ratio (wide range determined by xCLKDIV)
Determining clock period:
tCCLK = (tCK) / cRTO
tLCLK = (tCCLK) * lRTO
tSCLK = (tCCLK) * sRTO
RESET and CLKIN
The DSP receives its clock input on the CLKIN pin. The processor uses an
on-chip phase-locked loop to generate its internal clock, which is a multiple of the CLKIN frequency. Because the phase-locked loop requires some
time to achieve phase lock, CLKIN must be valid for a minimum time
period during reset before the RESET signal can be deasserted. For information on minimum clock setup, see the DSP’s data sheet.
11-16
ADSP-21160 SHARC DSP Hardware Reference
System Design
Table 11-3 describes the internal clock to CLKIN frequency ratios supported by DSP:
Table 11-3. Clock Configuration Definition
CLK_CFG3-0
Core/CLKIN ratio
0010
2:1
0011
3:1
0100
4:1
1111
Reserved
All others
Reserved
Table 11-4 demonstrates the internal core clock switching frequency,
across a range of CLKIN (external port bus) frequencies. The minimum
operational range for any given frequency is constrained by the operating
range of the phase lock loop. Note that the goal in selecting a particular
clock ratio for the DSP application is to provide the highest internal frequency, given a CLKIN frequency.
Table 11-4. Selecting Core to CLKIN Ratio
CLKIN Input (MHz)
25
33.3
40
50
Clock Ratios
Core CLK (MHz) 
2:1
50
66.6
80
80 (21160M)
100 (21160N)
3:1
75
100
N/A
N/A
4:1
100
N/A
N/A
N/A
Input Synchronization Delay
The DSP has several asynchronous inputs: RESET, TRST, HBR, CS, DMAR1,
DMAR2, IRQ2-0, and FLAG3-0 (when configured as inputs). These inputs can
ADSP-21160 SHARC DSP Hardware Reference
11-17
DSP Pin Descriptions
be asserted in arbitrary phase to the processor clock, CLKIN. The DSP synchronizes the inputs prior to recognizing them. The delay associated with
recognition is called the synchronization delay.
Any asynchronous input must be valid prior to the recognition point to be
recognized in a particular cycle. If an input does not meet the setup time
on a given cycle, it may be recognized in the current cycle or during the
next cycle.
To ensure recognition of an asynchronous input, it must be asserted for at
least one full processor cycle plus setup and hold time, except for RESET,
which must be asserted for at least four processor cycles. The minimum
time prior to recognition (the setup and hold time) is specified in the
DSP’s data sheet.
Interrupt and Timer Pins
The DSP’s external interrupt pins, flag pins, and timer pin can be used to
send and receive control signals to and from other devices in the system.
Hardware interrupt signals are received on the IRQ2-0 pins. Interrupts can
come from devices that require the DSP to perform some task on demand.
A memory-mapped peripheral, for example, can use an interrupt to alert
the processor that it has data available. For more information, see “Interrupts and Sequencing” on page 3-31.
The TIMEXP output is generated by the on-chip timer. It indicates to other
devices that the programmed time period has expired. For more information, see “Timer and Sequencing” on page 3-48.
Flag Pins
The FLAG3-0 pins allow single-bit signalling between the DSP and other
devices. For example, the DSP can raise an output flag to interrupt a host
processor. Each flag pin can be programmed to be either an input or output. In addition, many DSP instructions can be conditioned on a flag’s
11-18
ADSP-21160 SHARC DSP Hardware Reference
System Design
input value, enabling efficient communication and synchronization
between multiple processors or other interfaces.
The flags are bidirectional pins, each with the same functionality. The
FLGxO bits in the MODE2 register program the direction of each flag pin. For
more information, see “Mode Control 2 Register (MODE2)” on
page A-6.
Flag Inputs
When a flag pin is programmed as an input, its value is stored in a bit in
the FLAGS register. The bit is updated in each cycle with the input value
from the pin. Flag inputs can be asynchronous to the DSP clock, so there
is a one-cycle delay before a change on the pin appears in FLAGS (if the rising edge of the input misses the setup requirement for that cycle). For
more information, see “Flag Value Register (FLAGS)” on page A-27.
An flag bit is read-only if the flag is configured as an input. Otherwise, the
bit is readable and writable. The flag bit states are conditions that code
can specify in conditional instructions.
Flag Outputs
When a flag is configured as an output, the value on the pin follows the
value of the corresponding bit in the FLAGS register. A program can set or
clear the flag bit to provide a signal to another processor or peripheral.
The FLAG outputs transition on rising edge of CLKIN. Because the processor
core operates at least twice the frequency of CLKIN, the programmer must
hold the FLAG state stable for at least one full CLKIN period, to insure that
the output changes state. Figure 11-3 describes the relationship between
instruction execution and a Flag pin, when the processor core to bus clock
ratio is set to 2:1. Figure 11-3 also describes the flag in/out process. Note
that at least two instructions execute each CLKIN cycle.
ADSP-21160 SHARC DSP Hardware Reference
11-19
DSP Pin Descriptions
bit set MODE2 FLG0;
bit clr FLAGS FLG0;
bit set FLAGS FLG0;
nop;
/*
/*
/*
/*
/*
bit clr FLAGS FLG0;
/*
/*
/*
/*
/*
bit clr MODE2 FLG0;
Nop;
nop;
1S T CL K IN
CYCL E :
1st cycle: set FLAG0 to output in Mode2 */
clear FLAG0 */
1st cycle: set FLAG0 output high */
2nd cycle: FLAG register updated here */
A NOP indicates a NOP or another instruction
not related to FLAG. */
2nd cycle: clear FLAG0 output */
earliest assertion of FLAG0 output, depends on CLKIN phase*/
3rd cycle: set FLAG0 back to input */
3rd cycle: */
4th cycle: earliest deassertion of FLAG0 output */
2ND CL K IN
CY CL E :
3R D CL K IN
CY CL E :
OU T PU T E NAB LE D
FL AG HI GH
4T H CL KI N
CY CL E :
5T H CLK IN
CY CL E :
CL K IN
F L AG L OW
OU T PU T DIS AB L E D, I NP U T S AMP L E D
F L AGX
OU T PU T VAL ID
Figure 11-3. Flag Timing (At 2:1 Clock Ratio)
JTAG Interface Pins
The JTAG test access port consists of the TCK, TMS, TDI, TDO, and TRST
pins. The JTAG port can be connected to a controller that performs a
boundary scan for testing purposes. This port is also used by the Analog
Devices (ADI) family of emulators to access on-chip emulation features.
To allow the use of the emulator, a connector for its in-circuit probe must
be included in the target system. For more information, see “Designing
for JTAG Emulation” on page 11-24.
If TRST is not asserted (or held low) at power-up, the JTAG port is in an
undefined state that may cause the DSP to drive out on I/O pins that
11-20
ADSP-21160 SHARC DSP Hardware Reference
System Design
would normally be three-stated at reset. TRST can be held low with a
jumper to ground on the target board connector. For more information,
see Figure 11-9 on page 11-29.
Dual-Voltage Power-up Sequencing
The ADSP-21160 dual-voltage processor has special considerations
related to power-up. Note that these are general recommendations, and
specifics details on dual voltage power supply systems is beyond the scope
of this book. When the system power is activated through the DSP's dual
power supply system, both supplies should be brought up as quickly as
possible. Ideally, the two supplies, VDDEXT and VDDINT should be powered
up simultaneously. Many commercially available dual supply regulators
address simultaneous power-up requirements of the core and I/O. When
designing a dual supply system, the designer should consider the relative
voltage and ramp-up timing between the core and I/O voltages in order to
avoid potential issues with system and DSP long-term reliability.
The ADSP-21160 DSP’s I/O pads have a network of internal diodes to
protect the DSP from damage by electrostatic discharge. These protection
diodes connect the +2.5V DC (ADSP-21160M DSP) or +1.9V DC
(ADSP-21160N DSP) core and 3.3V I/O supplies internally. Figure 11-4
shows how a network of protection diodes isolates the internal supplies
and provides ESD protection for the I/O pins.
During the power-up sequence of the DSP, differences in the ramp up
rates and activation time between the two supplies can cause current to
flow in the I/O ESD protection circuitry. When applying power separately
to the VDDINT or VDDEXT pins, take precautions to prevent or limit the maximum current and duration conducted through the isolation diodes if the
unpowered pins are at ground potential.
Since the ESD protection diodes connect the +2.5V DC (ADSP-21160M
DSP) or +1.9V DC (ADSP-21160N DSP) core and 3.3V I/O supplies
internally, these diodes can be damaged at any time the +2.5V DC
ADSP-21160 SHARC DSP Hardware Reference
11-21
Dual-Voltage Power-up Sequencing
(ADSP-21160M DSP) or +1.9V DC (ADSP-21160N DSP) core supply
voltage is present without the presence of the 3.3V I/O supply.
V DDINT
V DDEXT
(3.3V)
(2.5V for 21 160M)
(1.9V for 21 160N)
ADSP-21160 DSP
INPUT
IO PIN
INTE RNAL
LOG IC
O UTPUT
Figure 11-4. Protection Diodes and IO Pin ESD Protection
ESD protection diodes connect the +2.5V DC
 The
(ADSP-21160M DSP) or +1.9V DC (ADSP-21160N DSP) core
and the 3.3V I/O supplies internally. Improper supply sequencing
can cause damage to the ESD protection circuitry. If the +2.5V DC
(ADSP-21160M DSP) or +1.9V DC (ADSP-21160N DSP) supply
is active for prolonged periods of time before the 3.3V I/O supply
is activated, there is a significant amount of loading on the I/O
pins. Damage occurs because the I/O will be powered from the
+2.5V DC (ADSP-21160M DSP) or +1.9V DC (ADSP-21160N
DSP) supply through the ESD diodes.
To prevent this damage to the ESD diode protection circuitry, Analog
Devices recommends including a bootstrap Schottky diode. The bootstrap
Schottky diode connected between the +2.5V DC (ADSP-21160M DSP)
11-22
ADSP-21160 SHARC DSP Hardware Reference
System Design
or +1.9V DC (ADSP-21160N DSP) and 3.3V power supplies protects the
ADSP-21160 DSPs from partially powering the 3.3V supply. Including a
Schottky diode will shorten the delay between the supply ramps and thus
prevent damage to the ESD diode protection circuitry. With this technique, of the +2.5V DC (ADSP-21160M DSP) or +1.9V DC
(ADSP-21160N DSP) rail rises ahead of the 3.3V rail, the Schottky diode
pulls the 3.3V rail along with the +2.5V DC (ADSP-21160M DSP) or
+1.9V DC (ADSP-21160N DSP) rail.
Figure 11-5 shows a basic block diagram of the Schottky diode connected
between the core and I/O voltage regulators and the DSP.
DC input
source
3.3V I/O
Voltage
Regulator
VDDEXT
ADSP-21160
2.5V (21160M)
1.9V (21160N)
Core Voltage
Regulator
VDDINT
Figure 11-5. Dual +2.5V DC (21160M) or +1.9V DC (21160N)/3/3V
Supplies with Schottky Diode
The anode of the diode must be connected to the +2.5V DC
(ADSP-21160M DSP) or +1.9V DC (ADSP-21160N DSP) supply. The
diode must have a forward biased voltage of 0.6V or less and must have a
current rating sufficient to supply the 3.3V rail of the system. The use of a
Schottky diode is the recommended method suggested by Analog Devices.
For recommendations on managing power-up sequencing for the core I/O
dual voltage supply, refer to the “Power-up Sequencing” specifications in
ADSP-21160 SHARC DSP Microcomputer Data Sheet.
ADSP-21160 SHARC DSP Hardware Reference
11-23
Designing for JTAG Emulation
Designing for JTAG Emulation
The DSP Analog Devices DSP Tools product line of JTAG emulator is a
development tool for debugging programs running in real time on DSP
target system hardware. The Analog Devices DSP Tools product line of
JTAG emulators provides a controlled environment for observing, debugging, and testing activities in a target system by connecting directly to the
target processor through its JTAG interface.
Because the Analog Devices DSP Tools product line of JTAG emulator
controls the target system’s DSP through the processor’s IEEE 1149.1
JTAG Test Access Port (TAP), non-intrusive in-circuit emulation is
assured. The emulator uses the TAP to access the internals of the DSP,
allowing the developer to load code, set breakpoints, observe variables,
observe memory, examine registers, etc. The DSP must be halted to send
data and commands, but once an operation is completed by the emulator,
the DSP system is set running at full speed with no impact on system timings emulator does not impact target loading or timing. The emulator’s
in-circuit probe connects to a variety of host computers (PCI bus, or USB)
with plug-in boards.
Target systems must have a 14-pin connector in order to accept the Analog Devices DSP Tools product line of JTAG emulator in-circuit probe, a
14-pin plug.
Designs must add this connector to the target board if the board is
intended for use with the ADSP-21160 JTAG Emulator. The total trace
length between the JTAG connector and the furthest device sharing the
Emulator’s JTAG pins should be limited to 15 inches maximum for guaranteed operation. This length restriction must include the emulator’s
JTAG signals, which are routed to one or more ADSP-21160 devices, or a
combination of ADSP-21160 devices and other JTAG devices on the
chain.
11-24
ADSP-21160 SHARC DSP Hardware Reference
System Design
Target Board Connector
The emulator interface to an Analog Devices’ JTAG DSP is a 14-pin
header, as shown in Figure 11-6. The customer must supply this header
on their target board in order to communicate with the emulator. The
interface consists of a standard dual row 0.025" square post header, set on
0.1" x 0.1" spacing, with a minimum post length of 0.235". Pin 3 is the
key position used to prevent the pod from being inserted backwards. This
pin must be clipped on the target board.
The clearance (length, width, and height) around the header must be as
shown in Figure 11-10. Maintain a minimum length of 0.15" and width
of 0.10" for the target board header. The pod connector attaches the target board header in this area. Therefore, there must be clearance to attach
and detach this connector. See the “DSP JTAG Pod Connector” on
page 11-31 for detailed drawings of the pod connector.
Figure 11-6. Emulator Interface for Analog Devices’ JTAG DSPs
As can be seen in Figure 11-6, there are two sets of signals on the header,
including the standard JTAG signals TMS, TCK, TDI, TDO, TRST, EMU used for
emulation purposes (via an emulator). Secondary JTAG signals BTMS,
ADSP-21160 SHARC DSP Hardware Reference
11-25
Designing for JTAG Emulation
BTCK, BTDI,
and BTRST are provided for optional use for board-level
(boundary scan) testing. While they are rarely used, the “B” signals should
be connected to a separate on-board JTAG boundary scan controller, if
they are used. If the “B” signals will not be used, tie all of them to ground
as shown in Figure 11-7.
alternately be activated (for some older silicon) to
 (+5V,can+3.3V,
or +2.5V) using a 4.7K resistor, as described in preBTCK
VCC
vious documents. Tying the signal to ground is universal and will
work for all silicon.
When the emulator is not connected to this header, jumpers should be
placed across BTMS, BTCK, BTRST, and BTDI as shown in Figure 11-7. This
holds the JTAG signals in the correct state to allow the DSP to run freely.
All the jumpers should be removed when connecting the emulator to the
JTAG header.
Figure 11-7. JTAG Target Board Connector With No Local Boundary
Scan
11-26
ADSP-21160 SHARC DSP Hardware Reference
System Design
For a list of the state of each standard JTAG signal refer to Table 11-5.
Use the following legend: O = Output, I = Input, and NU = Not Used
Table 11-5. State of Standard JTAG Signals
Signal
Description
Emulator
DSP
TMS
Test Mode Select
O
I
TCK
Test Clock (10 MHz)
O
I
TRST
Test Reset
O
I
TDI
Test Data In
O
I
TDO
Test Data Out
I
O
EMU
Emulation Pin
I
O (Open Drain)
CLKIN
DSP Clock Input
NU
I
The DSP CLKIN signal is the clock signal line (typically 30 MHz or
greater) that connects an oscillator to all DSPs in multiple DSP systems
requiring synchronization. In order for synchronous DSP operations to
work correctly the CLKIN signal on all the DSPs must be the same signal
and the skew between them must be minimal (use clock drivers, or other
means) - see the DSP user guide for more CLKIN details.
Note that the CLKIN signal is not used by the emulator and can cause noise
problems if connected to the JTAG header. Legacy documents show it
connected to pin 4 of the JTAG header. Pin 4 should be tied to ground on
the 14-pin JTAG header (do not connect the JTAG header pin to the DSP
CLKIN signal). If you have already connected it to the JTAG header pin,
and are experiencing noise from this signal, simply clip this pin on the
14-pin JTAG header.
The final connections between a single DSP target and the emulation
header (within 6 inches) are shown in Figure 11-8. A 4.7K pull-up resistor has been added on TCK, TDI and TMS for increased noise resistance.
ADSP-21160 SHARC DSP Hardware Reference
11-27
Designing for JTAG Emulation
Figure 11-8. Single DSP Connection to the JTAG Header
If a design uses more than one DSP (or other JTAG device in the scan
chain), or if the JTAG header is more than 6 inches from the DSP, use a
buffered connection scheme as shown in Figure 11-9 (no local boundary
scan mode shown). To keep signal skew to a minimum, be sure the buffers
are all in the same physical package (typical chips have 6, 8, or 16 drivers).
Using a buffer that includes a series of resistors such as the 74ABT2244
family can reduce ringing on the JTAG signal lines.
For low voltage applications (3.3V, 2.5V, and 1.9V I/O), the 74ALVT,
and 74AVC logic families is useful. Also, note the position of the pull-up
resistor on EMU. This is required since the EMU line is an open drain signal.
If more than one DSP (or JTAG device) is on the target (in the scan
chain), you must buffer the JTAG header. This will keep the signals clean
11-28
ADSP-21160 SHARC DSP Hardware Reference
System Design
and avoid noise problems that occur with longer signal traces (ultimately
resulting in reliable emulator operation).
Figure 11-9. Multiple DSP Connection to JTAG Header
Although the theoretical number of devices that can be supported (by the
software) in one JTAG scan chain is large (50 devices or more) it is not
recommended that you use more than eight physical devices in one scan
chain. A physical device could however contain many JTAG devices such
as inside a multi-chip module. The recommendation of not more than
eight physical devices is mostly due to the transmission line effects that
appear in long signal traces, and based on some field-collected empirical
data.
The best approach for large numbers of physical devices is to break the
chain into several smaller independent chains, each with their own JTAG
header and buffer. If this is not possible, at least add some jumpers that
can reduce the number of devices in one chain for debug purposes, and
pay special attention in the layout stage for transmission line effects.
ADSP-21160 SHARC DSP Hardware Reference
11-29
Layout Requirements
Layout Requirements
All JTAG signals (TCK, TMS, TDI, TDO, EMU, TRST) should be treated as critical route signals. Specify a controlled impedance requirement for each
route (value depends on your circuit board, typically 50-75). Keeping
crosstalk and inductance to a minimum on these lines by using a good
ground plane and by routing away from other high noise signals such as
clock lines is also important. Keep these routes as short and clean as possible, and keep the bused signals (TMS, TCK, TRST, EMU) as close to the same
length as possible.
JTAG TAP relies on the state of the TMS line and the TCK
 The
clock signal. If these signals have glitches (due to ground bounce,
crosstalk, etc.) unreliable emulator operation will result. When
experiencing emulator problems, look at these signals using a
high-speed digital oscilloscope. These lines must be clean, and may
require special termination schemes. If you are buffering the JTAG
header (most customers will) you must provide signal termination
appropriate for your target board (series, parallel, R/C, etc.).
Power Sequence for Emulation
The power-on sequence for your target and emulation system is as follows:
1. Apply power to the emulator first, then to the target board. This
ensures that the JTAG signals are in the correct state for the DSP
to run free.
2. Upon power-on, the emulator drives the TRST signal low, keeping
the DSP TAP in the test-logic-reset state, until the emulation software takes control.
Removal of power should be done in reverse: Turn off power to the target
board, then to the emulator.
11-30
ADSP-21160 SHARC DSP Hardware Reference
System Design
Additional JTAG Emulator References
The IEEE 1149.1 JTAG standard is sponsored by the Test Technology
Standards Committee of the IEEE Computer Society, and published by
the IEEE. The latest versions at the time of this publication are IEEE
Standard. 1149.1-1990 and IEEE Standard 1149.1a-1993. To order a
copy, call the IEEE at 1 800 678 4333 in the US and Canada, 1 908 981
1393 outside of the US and Canada. Visit the IEEE standards web site at
http://standards.ieee.org/.
Pod Specifications
This section contains design details on various emulator pod designs by
the Analog Devices DSP Tools product line. The emulator pod is the
device that connects directly to the DSP target board 14-pin JTAG
header. See also Engineer to Engineer Notes EE-68.
DSP JTAG Pod Connector
Figure 11-10 details the dimensions of the JTAG pod connector at the
14-pin target end. Figure 11-11 displays the keep-out area for a target
board header. The keep-out area allows the pod connector to properly seat
onto the target board header. This board area should contain no components (chips, resistors, capacitors, etc.). The dimensions are referenced to
the center of the 0.25" square post pin.
ADSP-21160 SHARC DSP Hardware Reference
11-31
Pod Specifications
Figure 11-10. JTAG Pod Connector (14-pin Target End)
Figure 11-11. Keep-out Area For a Target Board Header
DSP 3.3V Pod Logic
A portion of the Analog Devices DSP Tools product line 3.3V emulator
pod interface is shown in Figure 11-12. This figure describes the driver
circuitry of the emulator pod. As can be seen, TMS, TCK and TDI are driven
with a 33 series resistor. TRST is driven with a 100 series resistor. TDO
and CLKIN are terminated with an optional 91/120 parallel terminator.
EMU is pulled up with a 4.7K resistor. The 74LVT244 chip drives the signals at 3.3V, with a maximum current rating of ±32mA.
11-32
ADSP-21160 SHARC DSP Hardware Reference
System Design
Figure 11-12. 3.3V JTAG Pod Driver Logic
Parallel terminate the TMS, TCK, TRST, and TDI lines locally on your target
board, if needed, since they are driven by the pod with sufficient current
drive (±32mA). In order to use the terminators on the TDO line (CLKIN is
not used), you MUST have a buffer on your target board JTAG header.
The DSP is not capable of driving the parallel terminator load directly
with TDO. Assuming the proper buffers are included, use the optional parallel terminators by placing a jumper on J2.
DSP 2.5V Pod Logic
A portion of the Analog Devices DSP Tools product line 2.5V emulator
pod interface is shown in Figure 11-13. This figure describes the driver
circuitry of the emulator pod. As can be seen, TMS, TCK, and TDI are driven
with a 33 series resistor. TRST is driven with a 100 series resistor. TDO is
pulled up with a 4.7K resistor and terminated with an optional parallel
terminator that can be configured by the user. EMU is pulled up with a
4.7K resistor.
ADSP-21160 SHARC DSP Hardware Reference
11-33
Conditioning Input Signals
The CLKIN signal is not used and not connected inside the pod. The 74ALVT16244 chip drives the signals at 2.5V, with a maximum current rating
of ±8mA.
Figure 11-13. 2.5V JTAG Pod Driver Logic
You can terminate the TMS, TCK, TRST, and TDI lines locally on your target
board, if needed, as long as the terminator's current use does not exceed
the driver's maximum current supply (±8mA). In order to use the terminator on the TDO line, include a buffer on your target board JTAG header.
The DSP is not capable of driving a parallel terminator load (typically
50-75) directly with TDO. Assuming you have the proper buffers, you
may use the optional parallel terminator by adding the appropriate resistors and placing a jumper on J2.
Conditioning Input Signals
The DSP is a CMOS device. It has input conditioning circuits which simplify system design by filtering or latching input signals to reduce
susceptibility to glitches or reflections.
11-34
ADSP-21160 SHARC DSP Hardware Reference
System Design
The following sections describe why these circuits are needed and their
effect on input signals.
A typical CMOS input consists of an inverter with specific N and P device
sizes that cause a switching point of approximately 1.4V. This level is
selected to be the midpoint of the standard TTL interface specification of
V IL =0.8V and V IH =2.0V. This input inverter, unfortunately, has a fast
response to input signals and external glitches wider than about 1 ns. Filter circuits and hysteresis are added after the input inverter on some DSP
inputs, as described in the following sections.
Link Port Input Filter Circuits
The DSP’s link port input signals have on-chip filter circuits rather than
glitch rejection circuits. Filtering is not used on most signals because it
delays the incoming signal and the timing specifications.
Filtering is implemented only on the link port data and clock inputs. This
is possible because the link ports are self-synchronized. The clock and data
are sent together. It is not the absolute delay but rather the relative delay
between clock and data that determines performance margin.
By filtering both LxCLK and LxDAT3-0 with identical circuits, response to
glitches and reflections are reduced but relative delay is unaffected.
The filter has the effect of ignoring a full strength pulse (a glitch) narrower
than approximately 2 ns. Glitches that are not full strength can be somewhat wider. The link ports do not use glitch rejection circuits because they
can be used with longer, series-terminated transmission lines where the
reflections do not occur near the signal transitions.
LxCLK
ADSP-21160 SHARC DSP Hardware Reference
11-35
Designing For High Frequency Operation
RESET Input Hysteresis
Hysteresis is used only on the RESET input signal. Hysteresis causes the
switching point of the input inverter to be slightly above 1.4V for a rising
edge and slightly below 1.4V for a falling edge. The value of the hysteresis
is approximately ± 0.1V. The hysteresis is intended to prevent multiple
triggering of signals which are allowed to rise slowly, as might be expected
on a reset line with a delay implemented by an RC input circuit. Hysteresis is not used to reduce the effect of ringing on DSP input signals with
fast edges, because the amount of hysteresis that can be used on a CMOS
chip is too small to make much difference. The small amount of hysteresis
allowable is due to the restrictions on the tolerance of the V IL and V IH
TTL input levels under worst case conditions. Refer to the DSP’s data
sheet for exact specifications.
Designing For High Frequency Operation
Because the DSP processor can operate at very fast clock frequencies, signal integrity and noise problems must be considered for circuit board
design and layout. The following sections discuss these topics and suggest
various techniques to use when designing and debugging DSP systems.
Initial versions of the DSP are specified for operation at 50 MHz
and 40 MHz internal clocks; the following information is based on these
CLKIN frequencies. Refer to the DSP’s data sheet for current clock speed
specifications.
All DSP synchronous behavior is specified to CLKIN. System designers are
encouraged to clock synchronous peripherals/memory (which are attached
to the DSP external port) with this same clock source (or a different
low-skew output from the same clock driver). Alternatively, the clock output (CLKOUT) from the DSP may be employed to clock synchronous
peripherals/memory.
11-36
ADSP-21160 SHARC DSP Hardware Reference
System Design
Note the following behavior for CLKOUT:
• The DSP whose ID2-0=000 (uniprocessor), or 001 drives CLKOUT
during reset.
•
is specified relative to CLKIN in the DSP’s data sheet. When
using this output to clock system components, the phase and jitter
terms associated with this output must be treated as additional derating factors in determining specs.
CLKOUT
• The bus master drives CLKOUT. In an MP system, this clock has
multiple sources, including host logic, if present. The system component clocked by CLKOUT must be able to tolerate one or more
cycles in which CLKOUT is held high, as bus ownership changes.
Also, if the host logic operates asynchronously to the DSP CLKOUT
frequency, the system component clocked by CLKOUT must be able
to both operate at that host frequency, and handle the electrical
characteristics of the CLKOUT transition to that asynchronous frequency domain.
• For systems not needing CLKOUT as a clock source, CLKOUT may be
used to identify the current bus master. This requires that the outputs not be tied together. If and when this debug feature is not
needed, the CLKOUT output can be disabled by setting the COD bit in
the SYSCON register.
Clock Specifications and Jitter
The clock signal must be free of ringing and jitter. Clock jitter can easily
be introduced in a system where more than one clock frequency exists.
ADSP-21160 SHARC DSP Hardware Reference
11-37
Designing For High Frequency Operation
High frequency jitter on the clock to the DSP may result in abbreviated
internal cycles.
FREQUENCY 1
CLOCK
a
ADSP-21160
S
NO CONNECT
NO CONNECT
Figure 11-14. Reducing Clock Jitter and Ring
share a clock buffer IC with a signal of a different clock fre Never
quency. This introduces excessive jitter.
As shown in Figure 11-14, keep the portions of the system that operate at
different frequencies as physically separate as possible.
Clock Distribution
There must be low clock skew between DSPs in a multiprocessor cluster
when communicating synchronously on the external bus. The clock must
be routed in a controlled-impedance transmission line that can be properly terminated at either the end of the line or the source.
Figure 11-15 illustrates end-of-line termination for the clock. End-of-line
termination is not usually recommended unless the distance between the
processors is extremely small, because devices that are at a different wire
distance from each other receive a skewed clock. This is due to the propagation delay of a PCB transmission line, which is typically 5 to 6
inches/ns.
11-38
ADSP-21160 SHARC DSP Hardware Reference
System Design
CLOCK
+5 V
50⍀TRANSMISSION LINE
180⍀
1.4V
70⍀
a
a
a
ADSP-21160
ADSP-21160
ADSP-21160
S
S
S
Figure 11-15. Do Not Use End-Of-Line Termination for the Clock
Figure 11-16 illustrates source termination for the clock. Source termination allows delays in each path to be identical. Each device must be at the
end of the transmission line because only there does the signal have a single transition. The traces must be routed so that the delay through each is
matched to the others. Line impedance higher than 50 may be used, but
clock signal traces should be in the PCB layer closest to the ground plane
to keep delays stable and crosstalk low. More than one device may be at
the end of the line, but the wire length between them must be short and
the impedance (capacitance) of these must be kept high. The matched
inverters must be in the same IC and must be specified for a low skew (< 1
ns) with respect to each other. This skew should be as small as possible
because it subtracts from the margin on most specifications.
ADSP-21160 SHARC DSP Hardware Reference
11-39
Designing For High Frequency Operation
OCT AL INVE R T E R
ACT Q240
(NAT IONAL S E MICONDUCT OR )
OR
IDT 49F CT 805/A
OR
CY7C992
40⍀
50⍀ TRANSMISSION LINE
a
ADS P -21160
S
40⍀
CLOCK
50⍀ TRANSMISSION LINE
a
ADS P -21160
S
40⍀
a
50⍀ TRANSMISSION LINE
ADS P -21160
S
BUFFER DRIVE IMPEDANCE = 10⍀
A S E P AR AT E B UF F E R AND T R ANS MIS S ION
L INE IS NE E DE D F OR E ACH GR OUP OF
PR OCE S S OR S T HAT AR E F UR T HE R T HAN
4 INCHE S F R OM E ACH OT HE R .
Figure 11-16. Use Source Termination to Distribute the Clock
Point-To-Point Connections
A series termination resistor may be added near the pin for point-to-point
connections. This is typically used for link port applications when distances are greater than 6 inches as shown in Figure 11-17. For more
specific guidance on related issues, see the reference source in “Recommended Reading” on page 11-47 for suggestions on transmission line
termination and see ADSP-21160 SHARC DSP Microcomputer Data Sheet
for output drivers' rise and fall time data.
For link port operation at the full internal clock rate it is important to
maintain low skew between the data (LxDAT7-0) and clock (LxCLK).
11-40
ADSP-21160 SHARC DSP Hardware Reference
System Design
Although the DSP’s serial ports may be operated at a slow rate, the output
drivers still have fast edge rates and may require source termination for
longer distances.
50⍀ TRANSMISSION
LINE, LONGER THAN
6" (15.25 cm)
a
ADS P -21160
D RIVE R
IMPE DANCE
= 17 
33⍀
ON 
OP E N
CIRCUIT
 OF F
L INK PORT
T R ANS MIT TE R
L INK POR T
T RANS MIT T E R
S
ADS P -21160
S
DRIVER
IMPEDANCE
= 17⍀
OPEN
CIRCUIT
OFF
ON
LINK PORT
TRANSMITTER
a
33⍀
REFLECTED WAVE IS
ABSORBED AT THE SOURCE.
LINK PORT
RECEIVER
Figure 11-17. Source Termination For Point-To-Point Connections
Signal Integrity
The capacitive loading on high-speed signals should be reduced as much
as possible. Loading of buses can be reduced by using a buffer for devices
that operate with wait states, for example DRAMs. This reduces the
capacitance on signals tied to the zero-wait-state devices, allowing these
signals to switch faster and reducing noise-producing current spikes.
Signal run length (inductance) should also be minimized to reduce ringing. Extra care should be taken with certain signals such as the read and
write strobes (RDH/L, WRH/L) and acknowledge (ACK). In a multiprocessor
cluster, each DSP can drive the read or write strobes. In this case, some
damping resistance should be put in the signal path if the line length is
ADSP-21160 SHARC DSP Hardware Reference
11-41
Designing For High Frequency Operation
greater than 6 inches. This is at the expense of additional signal delay. The
time budget for these signals should be carefully analyzed.
Two possible damping arrangements between four DSPs are shown in
Figure 11-18 and Figure 11-19.
In Figure 11-18, a star connection of resistors is used. Each DSP can drive
the signal (for example, RDH/L or WRH/L strobe). Trace lengths should be
minimized. Experiment with the optimal resistance value and placement;
for example, near the processor or near the common connection. This
adds signal delay.
a
a
A D S P -2 1 1 6 0
A D S P -2 1 1 6 0
S
S
1 0⍀
1 0⍀
1 0⍀
1 0⍀
S T A R C O N N E C T IO N D A M P IN G R E S IS T O R S
a
a
A D S P -2 1 1 6 0
A D S P -2 1 1 6 0
S
S
Figure 11-18. Star Connection Damping Resistors
In the example of Figure 11-19, where processors 1 and 2, and 3 and 4 are
close to each other, a single damping resistor between the processor pairs
helps damp out reflections. Experiment with the resistor value. The two
processor groups have a skew with respect to each other.
11-42
ADSP-21160 SHARC DSP Hardware Reference
System Design
a
a
A DSP-21160
A DSP-21160
S
S
2 0⍀
SIN G L E D A M PIN G R E S IS T O R B E T W E EN P R O C E S S O R G R O U P S
a
a
A DSP-21160
A DSP-21160
S
S
Figure 11-19. Single Damping Resistor Between Processor Groups
Another solution to multiple drivers where longer distances are involved is
to have a single transmission line that is terminated at both ends. This
arrangement is shown in Figure 11-20. The stubs to the processors must
be kept as short as possible. Each device driver sees an impedance of 25,
but this resistor is biased at 1.4V, so the drive from the DSPs is sufficient
for TTL levels. To reduce power dissipation in the system and in each
DSP, this should only be used, if necessary, for signals such as the RDH/L or
WRH/L strobe. With this arrangement, the signals are skewed, but well
behaved.
ADSP-21160 SHARC DSP Hardware Reference
11-43
Designing For High Frequency Operation
+5 V
a
a
a
ADSP-21160
ADSP-21160
ADSP-21160
S
S
S
+5 V
180⍀
1.4V
180⍀
50⍀ TRANSMISSION LINE,
LONGER THAN 10" (25.4 cm)
1.4V
70⍀
70⍀
a
a
a
ADSP-21160
ADSP-21160
ADSP-21160
S
S
S
Figure 11-20. Single Transmission Line Terminated At Both Ends
Other Recommendations and Suggestions
These recommendations and suggestions are:
• Use more than one ground plane on the PCB to reduce crosstalk.
Be sure to use lots of vias between the ground planes. One VDD
plane for each supply is sufficient. These planes should be in the
center of the PCB.
• Keep critical signals such as clocks, strobes, and bus requests on a
signal layer next to a ground plane and away from or layout perpendicular to other non-critical signals to reduce crosstalk. For
example, data outputs switch at the same time that BRx inputs are
sampled; if the layout permits crosstalk between them, the system
could have problems with bus arbitration.
11-44
ADSP-21160 SHARC DSP Hardware Reference
System Design
• Position the processors on both sides of the board to reduce area
and distances if possible.
• Design for lower transmission line impedances to reduce crosstalk
and to allow better control of impedance and delay.
• Use of 3.3V peripheral components and power supplies to help
reduce transmission line problems, because the receiver switching
voltage of 1.4V is close to the middle of the voltage swing. In addition, ground bounce and noise coupling is less.
• Experiment with the board and isolate crosstalk and noise issues
from reflection issues. This can be done by driving a signal wire
from a pulse generator and studying the reflections while other
components and signals are passive.
Decoupling Capacitors and Ground Planes
Ground planes must be used for the ground and power supplies. Designs
should use a minimum of eight bypass capacitors (six 0.1 F and two
0.01 F ceramic). The capacitors should be placed very close to the
VDDEXT and VDDINT pins of the package as shown in Figure 11-21. Use
short and fat traces for this. The ground end of the capacitors should be
tied directly to the ground plane inside the package footprint of the DSP
(underneath it, on the bottom of the board), not outside the footprint. A
surface-mount capacitor is recommended because of its lower series inductance. Connect the power plane to the power supply pins directly with
minimum trace length. The ground planes must not be densely perforated
ADSP-21160 SHARC DSP Hardware Reference
11-45
Designing For High Frequency Operation
with vias or traces as their effectiveness is reduced. In addition, there
should be several large tantalum capacitors on the board.
a
ADSP-21160
S
CASE 1:
BYPASS CAPACITORS ON NON-COMPONENT
(BOTTOM) SIDE OF BOARD, BENEATH DSP PACKAGE
CASE 2:
BYPASS CAPACITORS ON COMPONENT (TOP) SIDE
OF BOARD, AROUND DSP PACKAGE
Figure 11-21. Bypass Capacitor Placement
can use either bypass placement case shown in
 Designs
Figure 11-21 or combinations of the two. Designs should try to
minimize signal feedthroughs that perforate the ground plane.
11-46
ADSP-21160 SHARC DSP Hardware Reference
System Design
Oscilloscope Probes
When making high-speed measurements, be sure to use a “bayonet” type
or similarly short (< 0.5 inch) ground clip, attached to the tip of the oscilloscope probe. The probe should be a low-capacitance active probe
with 3 pF or less of loading. The use of a standard ground clip with 4
inches of ground lead causes ringing to be seen on the displayed trace and
makes the signal appear to have excessive overshoot and undershoot.
A 1 GHz or better sampling oscilloscope is needed to see the signals
accurately.
Recommended Reading
The text High-Speed Digital Design: A Handbook of Black Magic is recommended for further reading. This book is a technical reference that covers
the problems encountered in state-of-the-art, high-frequency digital circuit design, and is an excellent source of information and practical ideas.
Topics covered in the book include:
• High-Speed Properties of Logic Gates
• Measurement Techniques
• Transmission Lines
• Ground Planes and Layer Stacking
• Terminations
• Vias
• Power Systems
• Connectors
• Ribbon Cables
ADSP-21160 SHARC DSP Hardware Reference
11-47
Booting Single and Multiple Processors
• Clock Distribution
• Clock Oscillators
High-Speed Digital Design: A Handbook of Black Magic, Johnson and Graham, Prentice Hall, Inc., ISBN 0-13-395724-1
Booting Single and Multiple Processors
Programs can be automatically downloaded to the internal memory of an
DSP after power-up or after a software reset. This process is called booting. The DSP supports three booting modes: EPROM, host, and link
port. For cases when the DSP must execute instructions from external
memory without booting, a “No boot” mode may also be configured. For
information on the setup and DMA processes for booting a single processor, see “Bootloading Through The External Port” on page 6-76 and
“Bootloading Through The Link Port” on page 6-87.
Multiprocessor systems can be booted from a host processor, from external
EPROM, through a link port, or from external memory.
Multiprocessor Host Booting
To boot multiple DSP processors from a host, each DSP must have its
EBOOT, LBOOT, and BMS pins configured for host booting: EBOOT=0,
LBOOT=0, and BMS=1. After system powerup, each DSP is in the idle state
and the BRx bus request lines are deasserted. The host must assert the HBR
input and boot each DSP by:
11-48
ADSP-21160 SHARC DSP Hardware Reference
System Design
• Asserting its CS pin (for asynchronous)
• Writing to the multiprocessor memory space location (for
synchronous)
• Downloading instructions as described in “Booting Another DSP”
on page 7-113
Multiprocessor EPROM Booting
There are two methods of booting a multiprocessor system from an
EPROM. Processors perform the following steps in these methods:
• Arbitrate for the bus
• DMA the 256 word boot stream, after becoming bus master
• Release the bus
• Execute the loaded instructions
All DSPs Boot in Turn from a Single EPROM
The BMS signals from each DSP may be wire-OR’ed together to drive the
chip select pin of the EPROM. Each DSP can boot in turn, according to
its priority. When the last one has finished booting, it must inform the
others (which may be in the idle state) that program execution can begin
(if all DSPs are to begin executing instructions simultaneously).
An example system that uses this processors-take-turns technique appears
in Figure 11-22. When multiple DSPs boot from one EPROM, the DSPs
can boot either identical code or different code from the EPROM. If the
processors load differing code, a jump table (based on processor ID) can
be used to select the code for each processor.
ADSP-21160 SHARC DSP Hardware Reference
11-49
Booting Single and Multiple Processors
EBOOT
LBOOT
ADDR
ADDR31-0
DATA39-32
DATA
DATA63-0
RDH
ADSP-21160
(S1)
RD
ACK
BMS
CS
EPROM
ADDR31-0
EBOOT
LBOOT
HERE, MULTIPLE SHARCS BOOT
FROM THE SAME EPROM.
DATA63-0
RDH
FOR THIS CONFIGURATION, THE
LOADER ROUTINE USES A JUMP
TABLE.
ACK
EBOOT
LBOOT
THIS TABLE INDICATES THE
ADDRESS OF THE IMAGE THAT
LOADS INTO EACH PROCESSOR.
DATA
ADDRESS
BMS
CONTROL
ADSP-21160
(S2)
THE PROCESSORS CAN LOAD THE
SAME IMAGE OR INDIVIDUAL
IMAGES.
ADDR31-0
DATA63-0
RDH
ACK
ADSP-21160
(S6)
BMS
Figure 11-22. DSPs-Take-Turns Booting from an EPROM
One DSP is Booted, which then Boots the Others
The EBOOT pin of the DSP with IDx=1 must be set high for EPROM booting. All other DSPs should be configured for host booting (EBOOT=0,
LBOOT=0, and BMS=1), which leaves them in the idle state at startup and
allows the DSP with IDx=1 to become bus master and boot itself. Only the
BMS pin of DSP #1 is connected to the chip select of the EPROM. When
DSP #1 has finished booting, it can boot the remaining DSPs by writing
to their external port DMA buffer 0 (EPB0) via multiprocessor memory
11-50
ADSP-21160 SHARC DSP Hardware Reference
System Design
space. An example system that uses this one-boots-others technique
appears in Figure 11-23.
EBOOT
LBOOT
ADSP-21160
(S1, MASTER)
ADDR
ADDR31-0
DATA39-32
DATA63-0
DATA
RDH
RD
ACK
BMS
CS
EPROM
BMS
ADDR31-0
EBOOT
LBOOT
DATA63-0
RDH
BMS
EBOOT
LBOOT
DATA
CONTROL
ADDRESS
ACK
ADSP-21160
(S2, SLAVE)
ADDR31-0
DATA647-16
RDH
ACK
ADSP-21160
(S6, SLAVE)
Figure 11-23. DSP-Boots-Others from an EPROM
Multiprocessor Link Port Booting
In systems where multiple DSPs are not connected by the parallel external
bus, booting can be accomplished from a single source through the link
ports. To simultaneously boot all of the DSPs, a parallel common connection should be made to link port buffer 4 (LBUF4) on each of the
processors. If only a daisy chain connection exists between the processors’
ADSP-21160 SHARC DSP Hardware Reference
11-51
Booting Single and Multiple Processors
link ports, then each DSP can boot the next one in turn. Link Buffer 4
must always be used for booting.
Multiprocessor Booting From External Memory
If external memory contains a program after reset, then the DSP with
IDx=1 should be set up for no boot mode. It begins executing from address
0x0080 0004 in external memory. When booting has completed, the
other DSPs may be booted by DSP #1 if they are set up for host booting,
or they can begin executing out of external memory if they are set up for
no boot mode. Multiprocessor bus arbitration allows this booting to occur
in an orderly manner. The bus arbitration sequence after reset is described
in “Multiprocessor Bus Arbitration” on page 7-98.
11-52
ADSP-21160 SHARC DSP Hardware Reference
Registers
A REGISTERS
The DSP has general purpose and dedicated registers in each of its functional blocks. The register reference information for each functional block
includes bit definitions, initialization values, and (for I/O processor registers) memory mapped address. Information on each type of register is
available at the following locations:
• “Control and Status System Registers” on page A-2
• “Processing Element Registers” on page A-15
• “Program Sequencer Registers” on page A-17
• “Data Address Generator Registers” on page A-32
• “I/O Processor Registers” on page A-33
When writing DSP programs, it is often necessary to set, clear, or test bits
in the DSP’s registers. While these bit operations can all be done by referring to the bit’s location within a register or (for some operations) the
register’s address with a hexadecimal number, it is much easier to use symbols that correspond to the bit’s or register’s name. For convenience and
consistency, Analog Devices provides a header file that provides these bit
and registers definitions. CrossCore Embedded Studio provides processor-specific header files in the SHARC/include directory. An #include file is
provided with the VisualDSP++ tools and can be found in the
ADSP-21160 SHARC DSP Hardware Reference
A-1
Control and Status System Registers
VisualDSP/processortype/include
directory.For more information, see
the “Register and Bit #Defines File (def21160.h)” on page A-81.
registers have reserved bits. When writing to a register, pro Many
grams may only clear (write zero to) the register’s reserved bits.
Control and Status System Registers
The DSP’s control and status system registers configure how the processor
core operates and indicate the status of many processor core operations. In
the ADSP-21160 SHARC DSP Instruction Set Reference, these registers are
referred to as System Registers (SREG), which are a subset of the DSP’s
Universal Registers (UREG). Not all registers are valid in all assembly language instructions. In the assembly syntax descriptions, the register group
name (UREG, SREG, and others) indicates which type of register is valid
within the instruction’s context. Table A-1 lists the processor core’s control and status registers with their initialization values. Descriptions of
each register follow. Other system registers (SREG) are in the I/O processor. For more information, see “I/O Processor Registers” on page A-33.
Table A-1. Control and Status System Registers (SREG and UREG)
Register Name and Page Reference
Initialization After Reset
“Mode Control 1 Register (MODE1)” on page A-3
0x0000 0000
“Mode Mask Register (MMASK)” on page A-5
0x0020 0000
“Mode Control 2 Register (MODE2)” on page A-6
0xnn00 00001
“Arithmetic Status Registers (ASTATx and ASTATy)” on page A-8 0x0000 0000
“Sticky Status Registers (STKYx and STKYy)” on page A-12
0x0540 0000
“User-Defined Status Registers (USTATx)” on page A-15
0x0000 0000
1
A-2
MODE2 bits 31-25 are the processor ID and silicon revision number, so the initialization value
varies with the DSP’s ID2-0 pins’ input and the silicon revision.
ADSP-21160 SHARC DSP Hardware Reference
Registers
Mode Control 1 Register (MODE1)
This is a non-memory mapped, universal, system register (UREG and SREG).
The reset value for this register is 0x0000 0000.
Table A-2. Mode Control 1 Register (MODE1) Bit Definitions
Bit(s)
Name
Definition
0
BR8
Bit Reverse Addressing For Index I8 Enable. This bit enables (bit
reversed if set, =1) or disables (normal if cleared, =0) bit reversed
addressing for accesses that are indexed with DAG2 register I8.
1
BR0
Bit Reverse Addressing For Index I0 Enable. This bit enables (bit
reversed if set, =1) or disables (normal if cleared, =0) bit reversed
addressing for accesses that are indexed with DAG1 register I0.
2
SRCU
Secondary Registers For Computational Units Enable. This bit
enables (use secondary if set, =1) or disables (use primary if cleared,
=0) secondary result (MR) registers in the computational units.
3
SRD1H
Secondary Registers For DAG1 High Enable. This bit enables (use
secondary if set, =1) or disables (use primary if cleared, =0) secondary
DAG1 registers for the upper half (I, M, L, B7-4) of the address generator.
4
SRD1L
Secondary Registers For DAG1 Low Enable. This bit enables (use secondary if set, =1) or disables (use primary if cleared, =0) secondary
DAG1 registers for the lower half (I, M, L, B3-0) of the address generator.
5
SRD2H
Secondary Registers For DAG2 High Enable. This bit enables (use
secondary if set, =1) or disables (use primary if cleared, =0) secondary
DAG2 registers for the upper half (I, M, L, B15-12) of the address
generator.
6
SRD2L
Secondary Registers For DAG2 Low Enable. This bit enables (use secondary if set, =1) or disables (use primary if cleared, =0) secondary
DAG2 registers for the lower half (I, M, L, B11-8) of the address generator.
7
SRRFH
Secondary Registers For Register File High Enable. This bit enables
(use secondary if set, =1) or disables (use primary if cleared, =0) secondary data registers for the upper half (R15-8) of the computational
units.
ADSP-21160 SHARC DSP Hardware Reference
A-3
Control and Status System Registers
Table A-2. Mode Control 1 Register (MODE1) Bit Definitions (Cont’d)
Bit(s)
Name
9-8
Reserved
10
SRRFL
Secondary Registers For Register File Low Enable. This bit enables
(use secondary if set, =1) or disables (use primary if cleared, =0) secondary data registers for the lower half (R7-0) of the computational
units.
11
NESTM
Nesting Multiple Interrupts Enable. This bit enables (nest if set, =1)
or disables (no nesting if cleared, =0) interrupt nesting in the interrupt
controller.
When interrupt nesting is disabled, a higher priority interrupt can not
interrupt a lower priority interrupt’s service routine. Other interrupts
are latched as they occur, but the DSP process them after the active
routine finishes.
When interrupt nesting is enabled, a higher priority interrupt can
interrupt a lower priority interrupt’s service routine. Lower interrupts
are latched as they occur, but the DSP process them after the nested
routines finish.
12
IRPTEN
Global Interrupt Enable. This bit enables (if set, =1) or disables (if
cleared, =0) all maskable interrupts.
13
ALUSAT
ALU Saturation Select. This bit selects whether the computational
units saturate results on positive or negative fixed-point overflows (if
1) or return unsaturated results (if 0).
14
SSE
Fixed-point Sign Extension Select. This bit selects whether the computational units sign extend short-word, 16-bit data (if 1) or zero-fill
the upper 32 bits (if 0).
15
TRUNC
Truncation Rounding Mode Select. This bit selects whether the computational units round results with round-to-zero (if 1) or
round-to-nearest (if 0).
16
RND32
Rounding For 32-bit Floating-point Data Select. This bit selects
whether the computational units round floating-point data to 32 bits
(if 1) or round to 40 bits (if 0).
18-17
CSEL
Bus Master Code Selection. These bits indicate whether the DSP processor has control of the external bus as follows: 00=DSP is bus master
or 01, 10, 11=DSP is not bus master.
20-19
A-4
Definition
Reserved
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-2. Mode Control 1 Register (MODE1) Bit Definitions (Cont’d)
Bit(s)
Name
Definition
21
PEYEN
Processor Element Y Enable. This bit enables computations in PEy—
SIMD mode—(if 1) or disables PEy—SISD mode—(if 0).
When set, Processing Element Y (computation units and register files)
accepts instruction dispatches. When cleared, Processing Element Y
goes into a low power mode.
22
BDCST9
Broadcast Register Loads Indexed With I9 Enable. This bit enables
(broadcast I9 if set, =1) or disables (no I9 broadcast if cleared, =0)
broadcast register loads for loads that use the data address generator I9
index.
When the BDCST9 bit is set, data register loads from the PM data bus
that use the I9 DAG2 index register are “broadcast” to a register or
register pair in each PE.
23
BDCST1
Broadcast Register Loads Indexed With I1 Enable. This bit enables
(broadcast I1 if set, =1) or disables (no I1 broadcast if cleared, =0)
broadcast register loads for loads that use the data address generator I1
index.
When the BDCST1 bit is set, data register loads from the DM data
bus that use the I1 DAG1 index register are “broadcast” to a register or
register pair in each PE.
24
CBUFEN
Circular Buffer Addressing Enable. This bit enables (circular if set,
=1) or disables (linear if cleared, =0) circular buffer addressing for buffers with loaded I, M, B, and L data address generator register.
31-25
Reserved
Mode Mask Register (MMASK)
This is a non-memory mapped, universal, system register (UREG and SREG).
The reset value for this register is 0x0020 0000.
Each bit in the MMASK register corresponds to a bit in the MODE1 register.
Bits that are set in MMASK are used to clear bits in MODE1 when the DSP's
status stack is pushed. This effectively disables different modes upon servicing an interrupt, or when executing a Push Sts instruction.
ADSP-21160 SHARC DSP Hardware Reference
A-5
Control and Status System Registers
The DSP’s status stack will be pushed in two cases:
• When you execute a Push
Sts
instruction explicitly in your code.
• When an IRQ2-0 timer expires or a VIRPT interrupt occurs.
Example:
Before the Push Sts instruction, MODE1 is set to 0x01202811. This MODE1
value corresponds to the following settings being enabled:
• Bit Reversing for I8
• Secondary Registers for DAG2 (high)
• Interrupt Nesting, ALU Saturation
• Processor Element Y (SIMD)
• Circular Buffering
is set to 0x0020 2001 indicating that you want to disable ALU Saturation, SIMD, and bit reversing for I8 after pushing the status stack. The
value in MODE1 after Push Sts is 0x0100 0810. The other settings that were
previously in MODE1 remain the same. The only bits that are affected are
those that are set both in MMASK and in MODE1. These bits are cleared after
the status stack is pushed.
MMASK
Note also that the reset value of MMASK is 0x0020 0000. If you do not make
any changes to the MMASK register, the default setting will automatically
disable SIMD when servicing any of the hardware interrupts mentioned
above, or during any push of the status stack.
Mode Control 2 Register (MODE2)
This is a non-memory mapped, universal, system register (UREG and SREG).
The reset value for this register is 0xmm00 0000. Because bits 31-25 in this
register are the DSP ID and silicon revision, the reset value varies with the
A-6
ADSP-21160 SHARC DSP Hardware Reference
Registers
system setting and silicon revision. Bits 31-25 of the MODE2 register are also
readable in the MODE2_SHDW register. For more information, see “MODE2
Shadow Register (MODE2_SHDW)” on page A-53.
Table A-3. Mode Control 2 Register (MODE2) Bit Definitions
Bit
Name
Definition
0
IRQ0E
IRQ0
1
IRQ1E
IRQ1
2
IRQ2E
IRQ2
3
Sensitivity Select. This bit selects sensitivity for IRQ0 as
edge-sensitive (if set, =1) or level-sensitive (if cleared, =0).
Sensitivity Select. This bit selects sensitivity for IRQ1 as
edge-sensitive (if set, =1) or level-sensitive (if cleared, =0).
Sensitivity Select. This bit selects sensitivity for IRQ2 as
edge-sensitive (if set, =1) or level-sensitive (if cleared, =0).
Reserved
4
CADIS
Cache Disable. This bit disables the instruction cache (if set, =1) or
enables the cache (if cleared, =0).
5
TIMEN
Timer Enable. This bit enables the timer (starts, if set, =1) or disables the timer (stops, if cleared, =0).
6
BUSLK
Bus Lock Request. This bit requests bus lock (DSP retains bus mastership, if set, =1) or does not request bus lock (normal bus mastering, if cleared, =0).
14-7
Reserved
15
FLG0O
FLAG0 Output Select. This bit selects the I/O direction for FLAG0
as an output (if set, =1) or an input (if cleared, =0).
16
FLG1O
FLAG1 Output Select. This bit selects the I/O direction for FLAG1
as an output (if set, =1) or an input (if cleared, =0).
17
FLG2O
FLAG2 Output Select. This bit selects the I/O direction for FLAG2
as an output (if set, =1) or an input (if cleared, =0).
18
FLG3O
FLAG3 Output Select. This bit selects the I/O direction for FLAG3
as an output (if set, =1) or an input (if cleared, =0).
19
CAFRZ
Cache Freeze. This bit freezes the instruction cache (retains contents, if set, =1) or thaws the cache (allows new input, if cleared,
=0).
ADSP-21160 SHARC DSP Hardware Reference
A-7
Control and Status System Registers
Table A-3. Mode Control 2 Register (MODE2) Bit Definitions (Cont’d)
Bit
Name
Definition
20
IRAE
I/O Processor Register Access Enable. This bit enables detection of
I/O processor register accesses (if set, =1) or disables detection of
I/O processor register accesses (if cleared, =0).
If IRAE is set, the DSP flags an access by setting the IRA bit in the
STKYx register. For more information, see “IRA” on page A-14.
21
U64MAE
Unaligned 64-bit Memory Access Enable. This bit enables detection of unaligned long word accesses (if set, =1) or disables detection
of unaligned long word accesses (if cleared, =0).
If U64MAE is set, the DSP flags an unaligned long word accesses by
setting the U64MA bit in the STKYx register. For more information, see “U64MA” on page A-14.
31-22
Reserved
Arithmetic Status Registers (ASTATx and ASTATy)
These are non-memory mapped, universal, system registers (UREG and
SREG). The reset value for these registers is 0x0000 0000. Each processing
element has its own ASTAT register. ASTATx indicates status for PEx operations, and ASTATy indicates status for PEy operations (see Table A-4).
If a program loads the
register manually, there is a one cycle
 effect
latency before the new value in
can be used in a conASTATx
ASTATx
ditional instruction
A-8
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-4. Arithmetic Status Registers (ASTATx/y)
Bit Definitions
Bit(s)
Name
Definition
0
AZ
ALU Zero/Floating-Point Underflow. This bit indicates whether the last
ALU operation’s result was zero (if set, =1) or non-zero (if cleared, =0).
The ALU updates AZ for all fixed-point and floating-point ALU operations. AZ can also indicate a floating-point underflow. During an ALU
underflow—indicated by a set (=1) AUS bit in the STKYx/y register, the
DSP sets AZ if the floating-point result is smaller than can be represented in the output format.
1
AV
ALU Overflow. This bit indicates whether the last ALU operation’s
result overflowed (if set, =1) or did not overflow (if cleared, =0).
The ALU updates AV for all fixed-point and floating-point ALU operations. For fixed-point results, the DSP sets AV and the AOS bit in the
STKYx/y register when the XOR of the two most significant bits is a 1.
For floating-point results, the DSP sets AV and the AVS bit is the
STKYx/y register when the rounded result overflows (unbiased exponent
> 127).
2
AN
ALU Negative. This bit indicates whether the last ALU operation’s result
was negative (if set, =1) or positive (if cleared, =0).
The ALU updates AN for all fixed-point and floating-point ALU operations.
3
AC
ALU fixed-point Carry. This bit indicates whether the last ALU operation had a carry out of most significant bit of the result (if set, =1) or had
no carry (if cleared, =0).
The ALU updates AC for all fixed-point operations. The DSP clears AC
during fixed-point logic operations: PASS, MIN, MAX, COMP, ABS,
and CLIP. The ALU reads the AC flag for fixed-point accumulate operations: addition with carry and fixed-point subtraction with carry.
4
AS
ALU X-Input Sign (for ABS and MANT). This bit indicates whether
the last ALU ABS or MANT operation’s input was negative (if set, =1) or
positive (if cleared, =0).
The ALU updates AS for only fixed-point and floating-point ABS and
the MANT operations. The ALU clears AS for all operations other than
ABS and MANT.
ADSP-21160 SHARC DSP Hardware Reference
A-9
Control and Status System Registers
Table A-4. Arithmetic Status Registers (ASTATx/y)
Bit Definitions (Cont’d)
Bit(s)
Name
Definition
5
AI
ALU Floating-Point Invalid Operation. This bit indicates whether the
last ALU operation’s input was invalid (if set, =1) or valid (if cleared, =0).
The ALU updates AI for all fixed-point and floating-point ALU operations. The DSP sets AI and the AIS bit in the STKYx/y register if the
ALU operation:
• Receives a NAN input operand
• Adds opposite-signed Infinities
• Subtracts like-signed Infinities
• Overflows during a floating-point to fixed-point conversion when saturation mode is not set
• Operates on an Infinity when saturation mode is not set
6
MN
Multiplier Negative. This bit indicates whether the last multiplier operation’s result was negative (if set, =1) or positive (if cleared, =0).
The multiplier updates MN for all fixed-point and floating-point multiplier operations.
7
MV
Multiplier Overflow. This bit indicates whether the last multiplier operation’s result overflowed (if set, =1) or did not overflow (if cleared, =0).
The multiplier updates MV for all fixed-point and floating-point multiplier operations. For floating-point results, the DSP sets MV and the
MVS bit in the STKYx/y register if the rounded result overflows (unbiased exponent > 127). For fixed-point results, the DSP sets MV and the
MOS bit in the STKYx/y register if the result of the multiplier operation
is:
• Twos-complement, fractional with the upper 17 bits of MR not all
zeros or all ones
• Twos-complement, integer with the upper 49 bits of MR not all zeros
or all ones
• Unsigned, fractional with the upper 16 bits of MR not all zeros
• Unsigned, integer with the upper 48 bits of MR not all zeros
If the multiplier operation directs a fixed-point result to an MR register,
the DSP places the overflowed portion of the result in MR1 and MR2
for an integer result or places it in MR2 only for a fractional result.
A-10
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-4. Arithmetic Status Registers (ASTATx/y)
Bit Definitions (Cont’d)
Bit(s)
Name
Definition
8
MU
Multiplier Floating-Point Underflow. This bit indicates whether the last
multiplier operation’s result underflowed (if set, =1) or did not underflow (if cleared, =0).
The multiplier updates MU for all fixed-point and floating-point multiplier operations. For floating-point results, the DSP sets MU and the
MUS bit in the STKYx/y register if the floating-point result underflows
(unbiased exponent < –126). Denormal operands are treated as Zeros,
therefore they never cause underflows. For fixed-point results, the DSP
sets MU and the MUS bit in the STKYx/y register if the result of the
multiplier operation is:
• Twos-complement, fractional: upper 48 bits all zeros or all ones, lower
32 bits not all zeros
• Unsigned, fractional: upper 48 bits all zeros, lower 32 bits not all zeros
If the multiplier operation directs a fixed-point, fractional result to an
MR register, the DSP places the underflowed portion of the result in
MR0.
9
MI
Multiplier Floating-Point Invalid Operation. This bit indicates whether
the last multiplier operation’s input was invalid (if set, =1) or valid (if
cleared, =0).
The multiplier updates MI for floating-point multiplier operations. The
DSP sets MI and the MIS bit in the STKYx/y register if the ALU operation:
• Receives a NAN input operand
• Receives an Infinity and Zero as input operands
10
AF
ALU Floating-Point Operation. This bit indicates whether the last ALU
operation was floating-point (if set, =1) or fixed-point (if cleared, =0).
The ALU updates AF for all fixed-point and floating-point ALU operations.
11
SV
Shifter Overflow. This bit indicates whether the last shifter operation’s
result overflowed (if set, =1) or did not overflow (if cleared, =0). The
shifter updates SV for all shifter operations. The DSP sets SV if the
shifter operation:
• Shifts the significant bits to the left of the 32-bit fixed-point field
• Tests, sets, or clears a bit outside of the 32-bit fixed-point field
• Extracts a field that is past or crosses the left edge of the 32-bit
fixed-point field
• Performs a LEFTZ or LEFTO operation that returns a result of 32
ADSP-21160 SHARC DSP Hardware Reference
A-11
Control and Status System Registers
Table A-4. Arithmetic Status Registers (ASTATx/y)
Bit Definitions (Cont’d)
Bit(s)
Name
Definition
12
SZ
Shifter Zero. This bit indicates whether the last shifter operation’s result
was zero (if set, =1) or non-zero (if cleared, =0).
The shifter updates SZ for all shifter operations. The DSP also sets SZ if
the shifter operation performs a bit test on a bit outside of the 32-bit
fixed-point field.
13
SS
Shifter Input Sign. This bit indicates whether the last shifter operation’s
input was negative (if set, =1) or positive (if cleared, =0).
The shifter updates SS for all shifter operations.
17-14
18
Reserved
BTF
23-19
31-24
Bit Test Flag for System Registers. This bit indicates whether the last
system register bit manipulation operation Bit Tst operation was true (if
set, =1) or false (if cleared, =0).
The DSP sets BTF when the bit(s) in a system register and value in the
Bit Tst instruction match. The DSP also sets BTF when the bit(s) in a
system register and value in the Bit Xor instruction match.
Reserved
CACC
Compare Accumulation Shift Register. Bit 31 of CACC indicates which
operand was greater during the last ALU compare operation: X input (if
set, =1) or Y input (if cleared, =0). The other seven bits in CACC form a
right-shift register, each storing a previous compare accumulation result.
With each new compare, the DSP right shifts the values of CACC, storing the newest value in bit 31 and the oldest value in bit 24.
Sticky Status Registers (STKYx and STKYy)
These are non-memory mapped, universal, system registers (UREG and
SREG). The reset value for these registers is 0x0000 0000. Each processing
element has its own STKY register. STKYx indicates status for PEx operations
and indicates status for some program sequencer stacks. The STKYy register
only indicates status for PEy operations.Table A-5 lists bits for both STKYx
and STKYy, noting with an  the bits that apply only to STKYx.
A-12
ADSP-21160 SHARC DSP Hardware Reference
Registers
not clear themselves after the condition they flag is no
 longerbitstrue.do They
remain “sticky” until cleared by the program.
STKY
The DSP sets a STKY bit in response to a condition. For example, the DSP
sets the AUS bit in the STKY register when an ALU underflow set AZ in the
ASTAT register. The DSP clears AZ if the next ALU operation does not
cause an underflow, but AUS remains set until a program clears the STKY
bit. Interrupt service routines should clear their interrupt’s corresponding
STKY bit so the DSP can detect a re-occurrence of the condition. For
example, an interrupt service routine for the floating-point underflow
exception interrupt (FLTUI) would clear the AUS bit in the STKY register
near the beginning of the routine.
Table A-5. Sticky Status Registers (STKYx/y) Bit Definitions
Bit(s)
Name
Definition
At right: + shows bits in both STKYx/y
x shows bits in STKYx only


0
AUS
ALU Floating-Point Underflow. This bit is a sticky indicator for
the ALU AZ bit. For more information, see “AZ” on page A-9.
1
AVS
ALU Floating-Point Overflow. This bit is a sticky indicator for the +
ALU AV bit. For more information, see “AV” on page A-9.
2
AOS
ALU Fixed-Point Overflow. This bit is a sticky indicator for the
ALU AV bit. For more information, see “AV” on page A-9.
4-3
+
+
Reserved
5
AIS
ALU Floating-Point Invalid Operation. This bit is a sticky indica- +
tor for the ALU AI bit. For more information, see “AI” on
page A-10.
6
MOS
Multiplier Fixed-Point Overflow. This bit is a sticky indicator for +
the multiplier MV bit. For more information, see “MV” on
page A-10.
7
MVS
Multiplier Floating-Point Overflow. This bit is a sticky indicator
for the multiplier MV bit. For more information, see “MV” on
page A-10.
ADSP-21160 SHARC DSP Hardware Reference
+
A-13
Control and Status System Registers
Table A-5. Sticky Status Registers (STKYx/y) Bit Definitions (Cont’d)
Bit(s)
Name
Definition
At right: + shows bits in both STKYx/y
x shows bits in STKYx only


8
MUS
Multiplier Floating-Point Underflow. This bit is a sticky indicator +
for the multiplier MU bit. For more information, see “MU” on
page A-11.
9
MIS
Multiplier Floating-Point Invalid Operation. This bit is a sticky
indicator for the multiplier MI bit. For more information, see
“MI” on page A-11.
16-10
+
Reserved
17
CB7S
DAG1 Circular Buffer 7 Overflow. This bit indicates whether a
x
circular buffer being addressed with DAG1 register I7 has overflowed (if set, =1) or has not overflowed (if cleared, =0). A circular
buffer overflow occurs when DAG circular buffering operation
increments the I register past the end of buffer.
18
CB15S
DAG2 Circular Buffer 15 Overflow. This bit indicates whether a x
circular buffer being addressed with DAG2 register I15 has overflowed (if set, =1) or has not overflowed (if cleared, =0). A circular
buffer overflow occurs when DAG circular buffering operation
increments the I register past the end of buffer.
19
IRA
IOP Register Access. This bit indicates whether a core, host, or
x
multiprocessor access to I/O processor registers has occurred (if 1)
or has not occurred (if 0).
20
U64MA
Unaligned 64-bit Memory Access. This bit indicates whether a
x
Normal word access with the LW mnemonic addressing an uneven
memory address has occurred (if 1) or has not occurred (if 0).
21
PCFL
PC Stack Full. This bit indicates whether the PC stack is full (if 1) x
or not full (if 0)—Not a sticky bit, cleared by a Pop.
22
PCEM
PC Stack Empty. This bit indicates whether the PC stack is empty x
(if 1) or not empty (if 0)—Not sticky, cleared by a Push.
23
SSOV
Status Stack Overflow. This bit indicates whether the status stack
is overflowed (if 1) or not overflowed (if 0)—A sticky bit.
x
24
SSEM
Status Stack Empty. This bit indicates whether the status stack is
empty (if 1) or not empty (if 0)—Not sticky, cleared by a Push.
x
A-14
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-5. Sticky Status Registers (STKYx/y) Bit Definitions (Cont’d)
Bit(s)
Name
Definition
At right: + shows bits in both STKYx/y
x shows bits in STKYx only


25
LSOV
Loop Stack Overflow. This bit indicates whether the loop counter x
stack and loop stack are overflowed (if 1) or not overflowed (if
0)—A sticky bit.
26
LSEM
Loop Stack Empty. This bit indicates whether the loop counter
stack and loop stack are empty (if 1) or not empty (if 0)—Not
sticky, cleared by a Push.
31-27
x
Reserved
User-Defined Status Registers (USTATx)
These are non-memory mapped, universal, system registers (UREG and
SREG). The reset value for these registers is 0x0000 0000. The USTATx registers are user-defined, general-purpose status registers. Programs can use
these 32-bit registers with bitwise instructions (set, clear, test, and others).
Often, programs use these registers for low-overhead, general-purpose
flags or for temporary 32-bit storage of data.
Processing Element Registers
Except for the PX register, the DSP’s processing element registers store
data for each element’s ALU, multiplier, and shifter. The inputs and outputs for processing element operations go through these registers. The PX
register lets programs transfer data between the data buses, but cannot be
an input or output in a calculation.
ADSP-21160 SHARC DSP Hardware Reference
A-15
Processing Element Registers
Table A-6. Processing Element Universal Registers (UREG)
Register Name and Page Reference
Initialization After Reset
“Data File Data Registers (Rx, Fx, Sx)” on page A-16
Undefined
“Multiplier Results Registers (MRxF, MRxB)” on page A-16
Undefined
“Program Memory Bus Exchange Register (PX)” on page A-17
Undefined
Data File Data Registers (Rx, Fx, Sx)
These are non-memory mapped, universal, data registers (UREG and DREG).
Each of the DSP’s processing elements has a data register file—a set of
40-bit data registers that transfer data between the data buses and the
computation units. These registers also provides local storage for operands
and results.
The R, F, and S prefixes on register names do not effect the 32-bit or
40-bit data transfer; the naming convention determines how the ALU,
multiplier, and shifter treat the data and determines which processing element’s data registers are being used.
For more information on how to use these registers, see “Data Register
File” on page 2-28.
Multiplier Results Registers (MRxF, MRxB)
These are non-memory mapped, universal, data registers (UREG and DREG).
Each of the DSP’s multipliers has a primary or foreground (MRF) register
and alternate or background (MRB) results register. Fixed-point operations
place 80-bit results in the multiplier’s foreground MRF register or background MRB register, depending on which is active.
For more information on selecting the result register, see “Alternate (Secondary) Data Registers” on page 2-30. For more information on result
register fields, see “Data Register File” on page 2-28.
A-16
ADSP-21160 SHARC DSP Hardware Reference
Registers
Program Memory Bus Exchange Register (PX)
These are non-memory mapped, universal registers (UREG only). The PM
Bus Exchange (PX) register permits data to flow between the PM and DM
data buses. The PX register can work as one 64-bit register or as two 32-bit
registers (PX1 and PX2).
For more information on data alignment of PX1 and PX2 within PX and PX
register usage, see “Internal Data Bus Exchange” on page 5-7.
Program Sequencer Registers
The DSP’s program sequencer registers direct the execution of instructions. These registers include support for:
• Instruction pipeline
• Program and loop stacks
• Timer
• Interrupt mask and latch
Table A-7. Program Sequencer System Registers (UREG and SREG)
Register
Initialization After Reset
“Interrupt Latch Register (IRPTL)” on page A-18
0x0000 0000 (cleared)
“Interrupt Mask Register (IMASK)” on page A-23
0x0000 0003
“Interrupt Mask Pointer Register (IMASKP)” on page A-23
0x0000 0000 (cleared)
“Link Port Interrupt Register (LIRPTL)” on page A-24
0x0000 0000 (cleared)
“Flag Value Register (FLAGS)” on page A-27
0x0000 000n1
1
FLAGS bits 0-3 are equal to the values of the FLAG0-3 input pins after reset; the flag pins are
configured as inputs after reset.
ADSP-21160 SHARC DSP Hardware Reference
A-17
Program Sequencer Registers
Table A-8. Program Sequencer Universal Registers (UREG only)
Register
Initialization After Reset
“Program Counter Register (PC)” on page A-28
Undefined
“Program Counter Stack Register (PCSTK)” on page A-29
Undefined
“Program Counter Stack Pointer Register (PCSTKP)” on
page A-29
Undefined
“Fetch Address Register (FADDR)” on page A-30
Undefined
“Decode Address Register (DADDR)” on page A-30
Undefined
“Loop Address Stack Register (LADDR)” on page A-30
Undefined
“Current Loop Counter Register (CURLCNTR)” on page A-31
Undefined
“Loop Counter Register (LCNTR)” on page A-31
Undefined
“Timer Period Register (TPERIOD)” on page A-31
Undefined
“Timer Count Register (TCOUNT)” on page A-31
Undefined
Interrupt Latch Register (IRPTL)
This is a non-memory mapped, universal, system register (UREG and SREG).
The reset value for this register is 0x0000 0000. The IRPTL register indicates latch status for interrupts.
Table A-9. Interrupt Latch Register (IRPTL) Bit Definitions
Bit(s)
Name
Definition
0
EMUI
Emulator Interrupt. This bit indicates whether an EMUI interrupt is
latched and is pending (if set, =1) or no EMUI interrupt is pending
(if cleared, =0). An EMUI interrupt occurs on reset and when an
external device asserts the EMU pin.
1
RSTI
Reset Interrupt. This bit indicates whether an RSTI interrupt is
latched and is pending (if set, =1) or no RSTI interrupt is pending (if
cleared, =0). An RSTI interrupt occurs on reset as an external device
asserts the RESET pin.
A-18
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-9. Interrupt Latch Register (IRPTL) Bit Definitions (Cont’d)
Bit(s)
Name
Definition
2
IICDI
Illegal Input Condition Detected. This bit indicates whether an
IICD interrupt is latched and is pending (if set, =1) or no IICD
interrupt is pending (if cleared, =0). An IICD interrupt occurs when
a TRUE results from the logical OR'ing of the Illegal I/O Processor
Register Access (IRA) and Unaligned 64-bit Memory Access bits in
the STKYx registers. For more information, see “IRA” on page A-14
and “U64MA” on page A-14.
3
SOVFI
Stack Overflow/Full. This bit indicates whether an SOVFI interrupt
is latched and is pending (if set, =1) or no SOVFI interrupt is pending (if cleared, =0). An SOVFI interrupt occurs when a stack in the
program sequencer overflows or is full. For more information see
“PCFL” on page A-14, “SSOV” on page A-14, and “LSOV” on
page A-15.
4
TMZHI
Timer Expired High Priority. This bit indicates whether a TMZHI
interrupt is latched and is pending (if set, =1) or no TMZHI interrupt is pending (if cleared, =0). A TMZHI interrupt occurs when the
timer decrements to zero. Note that this event also triggers a TMZLI
interrupt. The following control timer operations:
• The TCOUNT register contains the timer counter. The timer decrements the TCOUNT register each clock cycle.
• The TPERIOD value specifies the frequency of timer interrupts.
The number of cycles between interrupts is TPERIOD + 1. The
maximum value of TPERIOD is
232 – 1.
• The TIMEN bit in the MODE2 register starts and stops the timer.
Because the timer expired event (TCOUNT decrements to zero) generates two interrupts, TMZHI and TMZLI, programs should
unmask the timer interrupt with the desired priority and leave the
other one masked.
5
VIRPTI
Multiprocessor Vector Interrupt. This bit indicates whether a
VIRPTI interrupt is latched and is pending (if set, =1) or no VIRPTI
interrupt is pending (if cleared, =0). A VIRPTI interrupt occurs
when one of the DSPs in a multiprocessor system writes an address
(the vector) to the DSP’s VIRPT register.
6
IRQ2I
IRQ2 Hardware Interrupt. This bit indicates whether an IRQ2I interrupt is latched and is pending (if set, =1) or no IRQ2I interrupt is
pending (if cleared, =0). An IRQ2I interrupt occurs when an external
device asserts the IRQ2 pin.
ADSP-21160 SHARC DSP Hardware Reference
A-19
Program Sequencer Registers
Table A-9. Interrupt Latch Register (IRPTL) Bit Definitions (Cont’d)
Bit(s)
Name
Definition
7
IRQ1I
IRQ1 Hardware Interrupt. This bit indicates whether an IRQ1I interrupt is latched and is pending (if set, =1) or no IRQ1I interrupt is
pending (if cleared, =0). An IRQ1I interrupt occurs when an external
device asserts the IRQ1 pin.
9
IRQ0I
IRQ0 Hardware Interrupt. This bit indicates whether an IRQ0I interrupt is latched and is pending (if set, =1) or no IRQ0I interrupt is
pending (if cleared, =0). An IRQ0I interrupt occurs when an external
device asserts the IRQ0 pin.
9
Reserved
10
SPR0I
SPORT Receive 0. This bit indicates whether a SPR0I interrupt is
latched and is pending (if set, =1) or no SPR0I interrupt is pending
(if cleared, =0). A SPR0I interrupt occurs two cycles after the last bit
of an input the serial word is latched into RX0.
11
SPR1I
SPORT Receive 1. This bit indicates whether a SPR1I interrupt is
latched and is pending (if set, =1) or no SPR1I interrupt is pending
(if cleared, =0). A SPR1I interrupt occurs two cycles after the last bit
of an input the serial word is latched into RX1.
12
SPT0I
SPORT Transmit 0. This bit indicates whether a SPT0I interrupt is
latched and is pending (if set, =1) or no SPT0I interrupt is pending
(if cleared, =0). An SPT0I interrupt occurs two cycles after the last
bit of an output the serial word is latched from TX0.
13
SPT1I
SPORT Transmit 1. This bit indicates whether a SPT1I interrupt is
latched and is pending (if set, =1) or no SPT1I interrupt is pending
(if cleared, =0). An SPT1I interrupt occurs two cycles after the last
bit of an output the serial word is latched from TX1.
14
LPISUMI
Link Buffer DMA Summary. This bit indicates whether an LPISUMI interrupt is latched and is pending (if set, =1) or no LPISUMI
interrupt is pending (if cleared, =0). An LPISUMI interrupt occurs
when a TRUE results from the logical Or’ing of unmasked link port
interrupts, which are configured in the LIRPTL register.
A-20
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-9. Interrupt Latch Register (IRPTL) Bit Definitions (Cont’d)
Bit(s)
Name
Definition
15
EP0I
External Port Buffer 0 DMA. This bit indicates whether an EP0I
interrupt is latched and is pending (if set, =1) or no EP0I interrupt is
pending (if cleared, =0). An EP0I interrupt occurs when the external
port buffer’s DMA is disabled (DEN=0) and either:
• The buffer set to receive (TRAN=0), and the buffer is not empty
• The buffer set to transmit (TRAN=1), and the buffer is not full
16
EP1I
External Port Buffer 1 DMA. This bit indicates whether an EP1I
interrupt is latched and is pending (if set, =1) or no EP1I interrupt is
pending (if cleared, =0). For more information, see “EP0I” on
page A-21.
17
EP2I
External Port Buffer 2 DMA. This bit indicates whether an EP2I
interrupt is latched and is pending (if set, =1) or no EP2I interrupt is
pending (if cleared, =0). For more information, see “EP0I” on
page A-21.
18
EP3I
External Port Buffer 3 DMA. This bit indicates whether an EP3I
interrupt is latched and is pending (if set, =1) or no EP3I interrupt is
pending (if cleared, =0). For more information, see “EP0I” on
page A-21.
19
LSRQI
Link Port Service Request. This bit indicates whether an LSRQI
interrupt is latched and is pending (if set, =1) or no LSRQI interrupt
is pending (if cleared, =0). An LSRQI interrupt occurs when an external source accesses an unassigned link port or accesses an assigned
link port that has its link buffer disabled.
20
CB7I
DAG1 Circular Buffer 7 Overflow. This bit indicates whether a
CB7I interrupt is latched and is pending (if set, =1) or no CB7I
interrupt is pending (if cleared, =0). For more information, see
“CB7S” on page A-14.
21
CB15I
DAG2 Circular Buffer 15 Overflow. This bit indicates whether a
CB15I interrupt is latched and is pending (if set, =1) or no CB15I
interrupt is pending (if cleared, =0). For more information, see
“CB15S” on page A-14.
22
TMZLI
Timer Expired (Low Priority). This bit indicates whether a TMZLI
interrupt is latched and is pending (if set, =1) or no TMZLI interrupt
is pending (if cleared, =0). For more information, see “TMZHI” on
page A-19.
ADSP-21160 SHARC DSP Hardware Reference
A-21
Program Sequencer Registers
Table A-9. Interrupt Latch Register (IRPTL) Bit Definitions (Cont’d)
Bit(s)
Name
Definition
23
FIXI
Fixed-Point Overflow. This bit indicates whether a FIXI interrupt is
latched and is pending (if set, =1) or no FIXI interrupt is pending (if
cleared, =0). For more information, see “AOS” on page A-13.
24
FLTOI
Floating-Point Overflow. This bit indicates whether a FLTOI interrupt is latched and is pending (if set, =1) or no FLTOI interrupt is
pending (if cleared, =0). For more information, see “AVS” on
page A-13.
25
FLTUI
Floating-Point Underflow. This bit indicates whether a FLTUI interrupt is latched and is pending (if set, =1) or no FLTUI interrupt is
pending (if cleared, =0). For more information, see “AUS” on
page A-13.
26
FLTII
Floating-Point Invalid Operation. This bit indicates whether a
FLTII interrupt is latched and is pending (if set, =1) or no FLTII
interrupt is pending (if cleared, =0). For more information, see “AIS”
on page A-13.
27
SFT0I
User Software Interrupt 0. This bit indicates whether a SFT0I interrupt is latched and is pending (if set, =1) or no SFT0I interrupt is
pending (if cleared, =0). An SFT0I interrupt occurs when a program
sets (=1) this bit.
28
SFT1I
User Software Interrupt 1. This bit indicates whether a SFT1I interrupt is latched and is pending (if set, =1) or no SFT1I interrupt is
pending (if cleared, =0). For more information, see “SFT0I” on
page A-22.
29
SFT2I
User Software Interrupt 2. This bit indicates whether a SFT2I interrupt is latched and is pending (if set, =1) or no SFT2I interrupt is
pending (if cleared, =0). For more information, see “SFT0I” on
page A-22.
30
SFT3I
User Software Interrupt 3. This bit indicates whether a SFT3I interrupt is latched and is pending (if set, =1) or no SFT3I interrupt is
pending (if cleared, =0). For more information, see “SFT0I” on
page A-22.
31
A-22
Reserved
ADSP-21160 SHARC DSP Hardware Reference
Registers
Interrupt Mask Register (IMASK)
This is a non-memory mapped, universal, system register (UREG and SREG).
The reset value for this register is 0x0000 0003. Each bit in the IMASK register corresponds to a bit with the same name in the IRPTL registers. The
bits in IMASK unmask (enable if set, =1) or mask (disable if cleared, =0) the
interrupts that are latched in the IRPTL register. Except for RESET, all interrupts are maskable.
When IMASK masks an interrupt, the masking disables the DSP’s response
to the interrupt. The IRPTL register still latches an interrupt even when
masked, and the DSP responds to that latched interrupt if it is later
unmasked.
Interrupt Mask Pointer Register (IMASKP)
This is a non-memory mapped, universal, system register (UREG and SREG).
The reset value for this register is 0x0000 0000. Each bit in the IMASKP register corresponds to a bit with the same name in the IRPTL registers. This
register supports an interrupt nesting scheme that lets higher priority
events interrupt an interrupt service routine and keeps lower priority
events from interrupting.
When interrupt nesting is enabled (NESTM=1 in the MODE1 register), the bits
in IMASKP mask lower priority and unmask higher priority interrupts than
the interrupt that is currently being serviced. The IRPTL register still
latches a lower priority interrupt even when masked, and the DSP
responds to that latched interrupt if it is later unmasked.
When interrupt nesting is disabled (NESTM=0 in the MODE1 register), the
bits in IMASKP mask all interrupts while an interrupt is currently being serviced. The IRPTL register still latches these interrupts even when masked,
and the DSP responds to the highest priority latched interrupt after servicing the current interrupt.
For more information, see “NESTM” on page A-4.
ADSP-21160 SHARC DSP Hardware Reference
A-23
Program Sequencer Registers
Link Port Interrupt Register (LIRPTL)
This is a non-memory mapped, universal, system register (UREG and SREG).
The reset value for these registers is 0x0000 0000. The LIRPTL register
indicates latch status, select masking, and displays mask pointers for link
port interrupts.
Note that the
bit in the
 OR’ing
of the link port latch bits (
LPISUM
IRPTL
register contains a logical
For more informa-
LIRPTL5-0).
tion, see “LPISUMI” on page A-20.
Table A-10. Link Port Interrupt Latch, Mask, and Mask Pointer Register
(LIRPTL) Bit Definitions
Bit
Name
Definition
0
LP0
Link Port Buffer 0 Interrupt. This bit indicates whether an LP0
interrupt is latched and is pending (if set, =1) or no LP0 interrupt is
pending (if cleared, =0). An LP0 interrupt occurs when the link port
buffer’s DMA is disabled (DEN=0) and either:
• The buffer set to receive (TRAN=0), and the buffer is not empty
• The buffer set to transmit (TRAN=1), and the buffer is not full
Note that the LPx bit is set whether the link port is enabled in core
mode or DMA mode.
1
LP1
Link Port Buffer 1 Interrupt. This bit indicates whether an LP1
interrupt is latched and is pending (if set, =1) or no LP1 interrupt is
pending (if cleared, =0). For more information, see “LP0” on
page A-24.
2
LP2
Link Port Buffer 2 Interrupt. This bit indicates whether an LP2
interrupt is latched and is pending (if set, =1) or no LP2 interrupt is
pending (if cleared, =0). For more information, see “LP0” on
page A-24.
3
LP3
Link Port Buffer 3 Interrupt. This bit indicates whether an LP3
interrupt is latched and is pending (if set, =1) or no LP3 interrupt is
pending (if cleared, =0). For more information, see “LP0” on
page A-24.
A-24
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-10. Link Port Interrupt Latch, Mask, and Mask Pointer Register
(LIRPTL) Bit Definitions (Cont’d)
Bit
Name
Definition
4
LP4
Link Port Buffer 4 Interrupt. This bit indicates whether an LP4
interrupt is latched and is pending (if set, =1) or no LP4 interrupt is
pending (if cleared, =0). For more information, see “LP0” on
page A-24.
5
LP5
Link Port Buffer 5 Interrupt. This bit indicates whether an LP5
interrupt is latched and is pending (if set, =1) or no LP5 interrupt is
pending (if cleared, =0). For more information, see “LP0” on
page A-24.
15-6
Reserved
16
LP0MSK
Link Buffer 0 DMA Interrupt Mask. This bit unmasks the LP0
interrupt (if set, =1) or masks the LP0 interrupt (if cleared, =0). For
more information on how interrupt masking works, see “Interrupt
Mask Register (IMASK)” on page A-23.
17
LP1MSK
Link Buffer 1 DMA Interrupt Mask. This bit unmasks the LP1
interrupt (if set, =1) or masks the LP1 interrupt (if cleared, =0). For
more information on how interrupt masking works, see “Interrupt
Mask Register (IMASK)” on page A-23.
18
LP2MSK
Link Buffer 2 DMA Interrupt Mask. This bit unmasks the LP2
interrupt (if set, =1) or masks the LP2 interrupt (if cleared, =0). For
more information on how interrupt masking works, see “Interrupt
Mask Register (IMASK)” on page A-23.
19
LP3MSK
Link Buffer 3 DMA Interrupt Mask. This bit unmasks the LP3
interrupt (if set, =1) or masks the LP3 interrupt (if cleared, =0). For
more information on how interrupt masking works, see “Interrupt
Mask Register (IMASK)” on page A-23.
20
LP4MSK
Link Buffer 4 DMA Interrupt Mask. This bit unmasks the LP4
interrupt (if set, =1) or masks the LP4 interrupt (if cleared, =0). For
more information on how interrupt masking works, see “Interrupt
Mask Register (IMASK)” on page A-23.
21
LP5MSK
Link Buffer 5 DMA Interrupt Mask. This bit unmasks the LP5
interrupt (if set, =1) or masks the LP5 interrupt (if cleared, =0). For
more information on how interrupt masking works, see “Interrupt
Mask Register (IMASK)” on page A-23.
23-22
Reserved
ADSP-21160 SHARC DSP Hardware Reference
A-25
Program Sequencer Registers
Table A-10. Link Port Interrupt Latch, Mask, and Mask Pointer Register
(LIRPTL) Bit Definitions (Cont’d)
Bit
Name
Definition
24
LP0MSKP
Link Buffer 0 DMA Interrupt Mask Pointer. When the DSP is servicing another interrupt, this bit indicates whether the LP0 interrupt is masked (if set, =1) or the LP0 interrupt is unmasked (if
cleared, =0). For more information on how interrupt mask pointers
works, see “Interrupt Mask Pointer Register (IMASKP)” on
page A-23.
25
LP1MSKP
Link Buffer 1 DMA Interrupt Mask Pointer. When the DSP is servicing another interrupt, this bit indicates whether the LP1 interrupt is masked (if set, =1) or the LP1 interrupt is unmasked (if
cleared, =0). For more information on how interrupt mask pointers
works, see “Interrupt Mask Pointer Register (IMASKP)” on
page A-23.
26
LP2MSKP
Link Buffer 2 DMA Interrupt Mask Pointer. When the DSP is servicing another interrupt, this bit indicates whether the LP2 interrupt is masked (if set, =1) or the LP2 interrupt is unmasked (if
cleared, =0). For more information on how interrupt mask pointers
works, see “Interrupt Mask Pointer Register (IMASKP)” on
page A-23.
27
LP3MSKP
Link Buffer 3 DMA Interrupt Mask Pointer. When the DSP is servicing another interrupt, this bit indicates whether the LP3 interrupt is masked (if set, =1) or the LP3 interrupt is unmasked (if
cleared, =0). For more information on how interrupt mask pointers
works, see “Interrupt Mask Pointer Register (IMASKP)” on
page A-23.
28
LP4MSKP
Link Buffer 4 DMA Interrupt Mask Pointer. When the DSP is servicing another interrupt, this bit indicates whether the LP4 interrupt is masked (if set, =1) or the LP4 interrupt is unmasked (if
cleared, =0). For more information on how interrupt mask pointers
works, see “Interrupt Mask Pointer Register (IMASKP)” on
page A-23.
A-26
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-10. Link Port Interrupt Latch, Mask, and Mask Pointer Register
(LIRPTL) Bit Definitions (Cont’d)
Bit
Name
Definition
29
LP5MSKP
Link Buffer 5 DMA Interrupt Mask Pointer. When the DSP is servicing another interrupt, this bit indicates whether the LP5 interrupt is masked (if set, =1) or the LP5 interrupt is unmasked (if
cleared, =0). For more information on how interrupt mask pointers
works, see “Interrupt Mask Pointer Register (IMASKP)” on
page A-23.
31-30
Reserved
Flag Value Register (FLAGS)
This is a non-memory mapped, universal, system register (UREG and SREG).
The reset value for these registers is 0x0000 0000. The FLAGS register indicates the state of the FLGx pins. When a FLGx pin is an output, the DSP
outputs a high when a program sets the pin’s bit in FLAGS. The I/O direction (input or output) selection of each bit is controlled by its FLGxO bit in
the MODE2 register. For more information, see “FLG0O” on page A-7.
Table A-11. Link Port Interrupt Latch, Mask, and Mask Pointer Register
(LIRPTL) Bit Definitions
Bit
Name
Definition
0
FLG0
FLAG0 Value. This bit indicates the state of the FLAG0 pin,
whether the pin is high (if set, =1) or low (if cleared, =0).
1
FLG1
FLAG1 Value. This bit indicates the state of the FLAG1 pin,
whether the pin is high (if set, =1) or low (if cleared, =0).
2
FLG2
FLAG2 Value. This bit indicates the state of the FLAG2 pin,
whether the pin is high (if set, =1) or low (if cleared, =0).
3
FLG3
FLAG3 Value. This bit indicates the state of the FLAG3 pin,
whether the pin is high (if set, =1) or low (if cleared, =0).
31-4
Reserved
ADSP-21160 SHARC DSP Hardware Reference
A-27
Program Sequencer Registers
Program Counter Register (PC)
This is a non-memory mapped, universal register (UREG only). The Program Counter register is the last stage in the fetch-decode-execute
instruction pipeline and contains the 24-bit address of the instruction that
the DSP executes on the next cycle. The PC couples with the Program
Counter Stack, PCSTK, which stores return addresses and top-of-loop
addresses. All addresses generated by the sequencer are 24-bit program
memory instruction addresses.
As shown in Figure A-1, the address buses can handle 32-bit addresses,
but the program sequencer only generates 24-bit addresses over the PM
bus. Because the sequencer generates 24-bit addresses, sequencing is limited to the low 12 Mwords of the DSP’s 4 Gword memory map.
PM and DM Address Buses and DAGs Can Handle 32-Bit Addresses
Program Sequencer Handles 24-Bit Addresses
E Field
M Field S Field
Bits 19-17, System (Internal) Memory
Bits 22-20, Multiprocessor Memory
Bits 31-23, External Memory
Three fields in the address
identify the type of memory
being addressed.
Figure A-1. PM and DM Bus Addresses Versus Sequencing Addresses
A-28
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-12 describes the three fields that appear in Figure A-1. The content of the External (E), Multiprocessor (M), and System (S) fields in the
address route the data or instruction access to the memory space.
Table A-12. PM and DM Address Bus E, M, and S Fields
Bit Field
Description
E
External — Values in this field have the following meaning:
non-zero: the address is in external memory; with the E bits active remaining
bits [22-0] are a valid address.
all zeros: the address is in the DSP’s internal memory or in the internal memory
of another ADSP-21160 DSP (M and S activated).
M
Multiprocessor — Values in this field have the following meaning:
non-zero: ID of another ADSP-21160
111: broadcast write to internal memory of all ADSP-21160s
000: address in the DSP’s own internal memory
S
System — Values in this field have the following meaning:
000: address of an IOP register
001: address in Long Word Addressing space
01x: address in Normal Word Addressing space
1xx: address in Short Word Addressing space
Program Counter Stack Register (PCSTK)
This is a non-memory mapped, universal register (UREG only). The Program Counter Stack register contains the address of the top of the PC
stack. This register is a readable and writable register.
Program Counter Stack Pointer Register (PCSTKP)
This is a non-memory mapped, universal register (UREG only). The Program Counter Stack Pointer register contains the value of PCSTKP. This
value is zero when the PC stack is empty, is 1...30 when the stack contains
data, and is 31 when the stack overflows. This register is readable and
writable. A write to PCSTKP takes effect after a one-cycle delay. If the PC
stack is overflowed, a write to PCSTKP has no effect.
ADSP-21160 SHARC DSP Hardware Reference
A-29
Program Sequencer Registers
Fetch Address Register (FADDR)
This is a non-memory mapped, universal register (UREG only). The Fetch
Address register is the first stage in the fetch-decode-execute instruction
pipeline and contains the 24-bit address of the instruction that the DSP
fetches from memory on the next cycle.
Decode Address Register (DADDR)
This is a non-memory mapped, universal register (UREG only). The Decode
Address register is the second stage in the fetch-decode-execute instruction
pipeline and contains the 24-bit address of the instruction that the DSP
decodes on the next cycle.
Loop Address Stack Register (LADDR)
This is a non-memory mapped, universal register (UREG only). The Loop
Address Stack is six levels deep by 32 bits wide. The 32-bit word of each
level consists of a 24-bit loop termination address, a 5-bit termination
code, and a 2-bit loop type code:
Table A-13. Loop Address Stack Register (LADDR)
Bits
Value
0-23
loop termination address
24-28
termination code
29
reserved (always reads 0)
30-31
loop type code
00 = arithmetic condition-based (not LCE)
01 = counter-based, length 1
10 = counter-based, length 2
11 = counter-based, length > 2
A-30
ADSP-21160 SHARC DSP Hardware Reference
Registers
Current Loop Counter Register (CURLCNTR)
This is a non-memory mapped, universal register (UREG only). The Current Loop Counter register provides access to the loop counter stack and
tracks iterations for the Do/Until LCE loop being executed.
For more information on how to use CURLCNTR, see “Loop Counter Stack”
on page 3-28.
Loop Counter Register (LCNTR)
This is a non-memory mapped, universal register (UREG only). The Loop
Counter register provides access to the loop counter stack and holds the
count value before the Do/Until LCE loop is executed. For more information on how to use LCNTR, see “Loop Counter Stack” on page 3-28.
Timer Period Register (TPERIOD)
This is a non-memory mapped, universal register (UREG only). The Timer
Period register contains the decrementing timer count value, counting
down the cycles between timer interrupts. For more information on how
to use the timer, see “Timer and Sequencing” on page 3-48.
Timer Count Register (TCOUNT)
This is a non-memory mapped, universal register (UREG only). The Timer
Count register contains the timer period, indicating the number of cycles
between timer interrupts. For more information on how to use the timer,
see “Timer and Sequencing” on page 3-48.
ADSP-21160 SHARC DSP Hardware Reference
A-31
Data Address Generator Registers
Data Address Generator Registers
The DSP’s Data Address Generator (DAG) registers hold data addresses,
modify values, and circular buffer configurations. Using these registers,
the DAGs can automatically increment addressing for ranges of data locations (a buffer).
Table A-14. Data Address Generator Universal Registers (UREG only)
Register
Initialization After Reset
“Index Registers (Ix)” on page A-32
Undefined
“Modify Registers (Mx)” on page A-32
Undefined
“Length and Base Register (Lx, Bx)” on page A-33
Undefined
Index Registers (Ix)
These are non-memory mapped, universal registers (UREG only). The Data
Address Generators store addresses in Index registers (I0-I7 for DAG1
and I8-I15 for DAG2). An index register holds an address and acts as a
pointer to memory.
For more information, see “Overview” in Chapter 4, Data Address
Generators.
Modify Registers (Mx)
These are non-memory mapped, universal registers (UREG only). The Data
Address Generators update stored addresses using Modify registers (M0-M7
for DAG1 and M8-M15 for DAG2). A modify register provides the increment or step size by which an index register is pre- or post-modified
during a register move.
For more information, see “Overview” in Chapter 4, Data Address
Generators.
A-32
ADSP-21160 SHARC DSP Hardware Reference
Registers
Length and Base Register (Lx, Bx)
These are non-memory mapped, universal registers (UREG only). The Data
Address Generators control circular buffering operations with Length and
Base registers (L0-L7 and B0-B7 for DAG1 and L8-L15 and B8-B15 for
DAG2). Length and base registers setup the range of addresses and the
starting address for a circular buffer.
For more information, see “Overview” in Chapter 4, Data Address
Generators.
I/O Processor Registers
The I/O processor’s registers are accessible as part of the DSP’s memory
map. Table A-16 on page A-36 lists the I/O processor’s memory mapped
registers in address order and provides a cross reference to a description of
each register. These registers occupy addresses 0x00 through 0xFF of the
memory map and control I/O operations, including:
• External port DMA
• Link port DMA
• Serial port DMA
processor registers have a one cycle effect latency (changes take
 I/O
effect on the second cycle after the change).
Because the I/O processor’s registers are part of the DSP’s memory map,
buses access these registers as locations in memory. While these registers
act as memory mapped locations, they are separate from the DSP’s internal memory and have different bus access. One bus can access one I/O
processor register from one I/O processor register group at a time.
Table A-15 lists the I/O processor register groups.
ADSP-21160 SHARC DSP Hardware Reference
A-33
I/O Processor Registers
When there is contention among the buses for access to registers in the
same I/O processor register group, the DSP arbitrates register access as
follows:
• External Port (EP) bus accesses (highest priority)
• Data Memory (DM) bus
• Program Memory (PM) bus
• I/O processor (IO) bus (lowest priority)
The bus with highest priority gets access to the I/O processor register
group, and the other buses are held off from accessing that I/O processor
register group until that access been completed.
There is one exception to this access contention rule. The IO bus and EP
bus can simultaneously access the DB (DMA buffer) group of registers,
allowing DMA transfers to internal memory at full speed.
Table A-15. I/O Processor Register Groups
Register Group
I/O Processor Registers In This Group
System Control (SC)
Registers
SYSCON, VIRPT, WAIT, SYSTAT, MSGR0, MSGR1, MSGR2,
MSGR3, MSGR4, MSGR5, MSGR6, MSGR7, BMAX, BCNT,
ELAST, PC_SHDW, MODE2_SHDW
DMA Address (DA)
Registers
II4, IM4, C4, CP4, GP4, DB4, DA4, II5, IM5, C5, CP5, GP5, DB5,
DA5, II6, IM6, C6, CP6, GP6, EI6, EM6, EC6, II7, IM7, C7, CP7,
GP7, EI7, EM7, EC7, II8, IM8, C8, CP8, GP8, EI8, EM8, EC8, II9,
IM9, C9, CP9, GP9, EI9, EM9, EC9, II0, IM0, C0, CP0, GP0,
DB0, DA0, II1, IM1, C1, CP1, GP1, DB1, DA1, II2, IM2, C2, CP2,
GP2, DB2, DA2, II3, IM3, C3, CP3, GP3, DB3, DA3, DMASTAT
A-34
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-15. I/O Processor Register Groups
Register Group
I/O Processor Registers In This Group
DMA Buffer (DB)
Registers
EPB0, EPB1, EPB2, EPB3, DMAC6, DMAC7, DMAC8, DMAC9
Link and Serial Port (LSP) LBUF0, LBUF1, LBUF2, LBUF3, LBUF4, LBUF5, LCTL, LCOM,
Registers
LAR, LSRQ, LPATH1, LPATH2, LPATH3, LPCNT, CNST1,
CNST2, STCTL0, SRCTL0, TX0, RX0, TDIV0, RDIV0, MTCS0,
MRCS0, MTCCS0, MRCCS0, SPATH0, KEYWD0, KEYMASK0,
STCTL1, SRCTL1, TX1, RX1, TDIV1, RDIV1, MTCS1, MRCS1,
MTCCS1, MRCCS1, SPATH1, KEYWD1, KEYMASK1
Because the I/O processor registers are memory-mapped, the DSP’s architecture does not allow programs to directly transfer data between these
registers and other memory locations, except as part of a DMA operation.
To read or write I/O processor registers, programs must use the processor
core registers. The following example code shows a value being transferred
from memory to the USTAT1 register, then the value is transferred to the
I/O processor WAIT registers.
USTAT2= 0x108421; /* 1st instr. to be executed after reset */
DM(WAIT)=USTAT2; /* Set external memory waitstates to 0
*/
The register names for I/O processor registers are not part of the DSP’s
assembly syntax. To ease access to these registers, programs should use the
#include command to incorporate a file containing the registers’ symbolic
names and addresses. An example #include file appears in the “Register
and Bit #Defines File (def21160.h)” on page A-81.
ADSP-21160 SHARC DSP Hardware Reference
A-35
I/O Processor Registers
Table A-16. I/O Processor Registers Memory Map
Register Address
Register Name
Initialization
After Reset
Register Group
Page Cross
Reference
0x00
SYSCON
0x0001 0010
SC
on page A-45
0x01
VIRPT
0x0004 0014
SC
on page A-48
0x02
WAIT
0x01ce 739c
SC
on page A-48
0x03
SYSTAT
0x000 0nn0
SC
on page A-50
0x04
EPB0
ni
DB
on page A-52
0x06
EPB1
ni
DB
on page A-52
0x08
MSGR0
ni
SC
on page A-52
0x09
MSGR1
ni
SC
on page A-52
0x0a
MSGR2
ni
SC
on page A-52
0x0b
MSGR3
ni
SC
on page A-52
0x0c
MSGR4
ni
SC
on page A-52
0x0d
MSGR5
ni
SC
on page A-52
0x0e
MSGR6
ni
SC
on page A-52
0x0f
MSGR7
ni
SC
on page A-52
0x10
PC_SHDW
ni
SC
on page A-53
0x11
MODE2_
SHDW
ni
SC
on page A-53
0x12 – 0x13
Reserved
0x14
EPB2
ni
DB
on page A-52
0x16
EPB3
ni
DB
on page A-52
0x18
BMAX
0x0000 0000
SC
on page A-54
0x19
BCNT
0x0000 0000
SC
on page A-54
Reserved
Notes: An “ni” in the Initialization column indicates that the register is Not Initialized.
For information on Register Groups, see Table A-15 on page A-34.
An * denotes that initialization depends on the booting mode. For more information, see
“Bootloading Through The External Port” on page 6-76 or “Bootloading Through The Link
Port” on page 6-87.
A-36
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-16. I/O Processor Registers Memory Map (Cont’d)
Register Address
Register Name
Initialization
After Reset
Register Group
Page Cross
Reference
0x1a
Reserved
0x1b
ELAST
ni
SC
on page A-54
0x1c
DMAC10
ni*
DB
on page A-54
0x1d
DMAC11
0x0000 0000
DB
on page A-54
0x1e
DMAC12
0x0000 0000
DB
on page A-54
0x1f
DMAC13
0x0000 0000
DB
on page A-54
0x20 – 0x2f
Reserved
0x30
II4
ni
DA
on page A-58
0x31
IM4
ni
DA
on page A-59
0x32
C4
ni
DA
on page A-59
0x33
CP4
ni
DA
on page A-59
0x34
GP4
ni
DA
on page A-60
0x35
DB4
ni
DA
on page A-60
0x36
DA4
ni
DA
on page A-60
0x37
DMASTAT
ni
DA
on page A-60
0x38
II5
ni
DA
on page A-58
0x39
IM5
ni
DA
on page A-59
0x3a
C5
ni
DA
on page A-59
0x3b
CP5
ni
DA
on page A-59
0x3c
GP5
ni
DA
on page A-60
0x3d
DB5
ni
DA
on page A-60
0x3e
DA5
ni
DA
on page A-60
Reserved
Reserved
Notes: An “ni” in the Initialization column indicates that the register is Not Initialized.
For information on Register Groups, see Table A-15 on page A-34.
An * denotes that initialization depends on the booting mode. For more information, see
“Bootloading Through The External Port” on page 6-76 or “Bootloading Through The Link
Port” on page 6-87.
ADSP-21160 SHARC DSP Hardware Reference
A-37
I/O Processor Registers
Table A-16. I/O Processor Registers Memory Map (Cont’d)
Register Address
Register Name
Initialization
After Reset
Register Group
Page Cross
Reference
0x3f
Reserved
ni
DA
Reserved
0x40
II10
ni*
DA
on page A-58
0x41
IM10
ni*
DA
on page A-59
0x42
C10
ni*
DA
on page A-59
0x43
CP10
ni*
DA
on page A-59
0x44
GP10
ni*
DA
on page A-60
0x45
EI10
ni*
DA
on page A-61
0x46
EM10
ni*
DA
on page A-61
0x47
EC10
ni*
DA
on page A-62
0x48
II11
ni
DA
on page A-58
0x49
IM11
ni
DA
on page A-59
0x4a
C11
ni
DA
on page A-59
0x4b
CP11
ni
DA
on page A-59
0x4c
GP11
ni
DA
on page A-60
0x4d
EI11
ni
DA
on page A-61
0x4e
EM11
ni
DA
on page A-61
0x4f
EC11
ni
DA
on page A-62
0x50
II12
ni
DA
on page A-58
0x51
IM12
ni
DA
on page A-59
0x52
C12
ni
DA
on page A-59
0x53
CP12
ni
DA
on page A-59
0x54
GP12
ni
DA
on page A-60
Notes: An “ni” in the Initialization column indicates that the register is Not Initialized.
For information on Register Groups, see Table A-15 on page A-34.
An * denotes that initialization depends on the booting mode. For more information, see
“Bootloading Through The External Port” on page 6-76 or “Bootloading Through The Link
Port” on page 6-87.
A-38
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-16. I/O Processor Registers Memory Map (Cont’d)
Register Address
Register Name
Initialization
After Reset
Register Group
Page Cross
Reference
0x55
EI12
ni
DA
on page A-61
0x56
EM12
ni
DA
on page A-61
0x57
EC12
ni
DA
on page A-62
0x58
II13
ni
DA
on page A-58
0x59
IM13
ni
DA
on page A-59
0x5a
C13
ni
DA
on page A-59
0x5b
CP13
ni
DA
on page A-59
0x5c
GP13
ni
DA
on page A-60
0x5d
EI13
ni
DA
on page A-61
0x5e
EM13
ni
DA
on page A-61
0x5f
EC13
ni
DA
on page A-62
0x60
II0
ni
DA
on page A-58
0x61
IM0
ni
DA
on page A-59
0x62
C0
ni
DA
on page A-59
0x63
CP0
ni
DA
on page A-59
0x64
GP0
ni
DA
on page A-60
0x65
DB0
ni
DA
on page A-60
0x66
DA0
ni
DA
on page A-60
0x68
II1
ni
DA
on page A-58
0x69
IM1
ni
DA
on page A-59
0x6a
C1
ni
DA
on page A-59
0x6b
CP1
ni
DA
on page A-59
Notes: An “ni” in the Initialization column indicates that the register is Not Initialized.
For information on Register Groups, see Table A-15 on page A-34.
An * denotes that initialization depends on the booting mode. For more information, see
“Bootloading Through The External Port” on page 6-76 or “Bootloading Through The Link
Port” on page 6-87.
ADSP-21160 SHARC DSP Hardware Reference
A-39
I/O Processor Registers
Table A-16. I/O Processor Registers Memory Map (Cont’d)
Register Address
Register Name
Initialization
After Reset
Register Group
Page Cross
Reference
0x6c
GP1
ni
DA
on page A-60
0x6d
DB1
ni
DA
on page A-60
0x6e
DA1
ni
DA
on page A-60
0x6f
Reserved
0x70
II2
ni
DA
on page A-58
0x71
IM2
ni
DA
on page A-59
0x72
C2
ni
DA
on page A-59
0x73
CP2
ni
DA
on page A-59
0x74
GP2
ni
DA
on page A-60
0x75
DB2
ni
DA
on page A-60
0x76
DA2
ni
DA
on page A-60
0x78
II3
ni
DA
on page A-58
0x79
IM3
ni
DA
on page A-59
0x7a
C3
ni
DA
on page A-59
0x7b
CP3
ni
DA
on page A-59
0x7c
GP3
ni
DA
on page A-60
0x7d
DB3
ni
DA
on page A-60
0x7e
DA3
ni
DA
on page A-60
0x7f
Reserved
0x80
II6
ni
DA
on page A-58
0x81
IM6
ni
DA
on page A-59
0x82
C6
ni
DA
on page A-59
Reserved
Reserved
Notes: An “ni” in the Initialization column indicates that the register is Not Initialized.
For information on Register Groups, see Table A-15 on page A-34.
An * denotes that initialization depends on the booting mode. For more information, see
“Bootloading Through The External Port” on page 6-76 or “Bootloading Through The Link
Port” on page 6-87.
A-40
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-16. I/O Processor Registers Memory Map (Cont’d)
Register Address
Register Name
Initialization
After Reset
Register Group
Page Cross
Reference
0x83
CP6
ni
DA
on page A-59
0x84
GP6
ni
DA
on page A-60
0x85
DB6
ni
DA
on page A-60
0x86
DA6
ni
DA
on page A-60
0x88
II7
ni
DA
on page A-58
0x89
IM7
ni
DA
on page A-59
0x8a
C7
ni
DA
on page A-59
0x8b
CP7
ni
DA
on page A-59
0x8c
GP7
ni
DA
on page A-60
0x8d
DB7
ni
DA
on page A-60
0x8e
DA7
ni
DA
on page A-60
0x8f
Reserved
0x90
II8
ni
DA
on page A-58
0x91
IM8
ni
DA
on page A-59
0x92
C8
ni
DA
on page A-59
0x93
CP8
ni
DA
on page A-59
0x94
GP8
ni
DA
on page A-60
0x95
DB8
ni
DA
on page A-60
0x96
DA8
ni
DA
on page A-60
0x98
II9
ni
DA
on page A-58
0x99
IM9
ni
DA
on page A-59
0x9a
C9
ni
DA
on page A-59
Reserved
Notes: An “ni” in the Initialization column indicates that the register is Not Initialized.
For information on Register Groups, see Table A-15 on page A-34.
An * denotes that initialization depends on the booting mode. For more information, see
“Bootloading Through The External Port” on page 6-76 or “Bootloading Through The Link
Port” on page 6-87.
ADSP-21160 SHARC DSP Hardware Reference
A-41
I/O Processor Registers
Table A-16. I/O Processor Registers Memory Map (Cont’d)
Register Address
Register Name
Initialization
After Reset
Register Group
Page Cross
Reference
0x9b
CP9
ni
DA
on page A-59
0x9c
GP9
ni
DA
on page A-60
0x9d
DB9
ni
DA
on page A-60
0x9e
DA9
ni
DA
on page A-60
0x9f–0xbf
Reserved (emulation control registers)
0xc0
LBUF0
ni
LSP
on page A-62
0xc2
LBUF1
ni
LSP
on page A-62
0xc4
LBUF2
ni
LSP
on page A-62
0xc6
LBUF3
ni
LSP
on page A-62
0xc8
LBUF4
ni
LSP
on page A-62
0xca
LBUF5
ni
LSP
on page A-62
0xcc
LCTL0
0x00000000
LSP
on page A-62
0xcd
LCTL1
0x00000000
LSP
on page A-62
0xce
LCOM
0x00000000
LSP
on page A-65
0xcf
LAR
0x0002C688
LSP
on page A-67
0xd0
LSRQ
0x0000 0000
LSP
on page A-68
0xd1
LPATH1
ni
LSP
on page A-70
0xd2
LPATH2
ni
LSP
on page A-70
0xd3
LPATH3
ni
LSP
on page A-70
0xd4
LPCNT
ni
LSP
on page A-70
0xd5
CNST1
ni
LSP
on page A-71
0xd6
CNST2
ni
LSP
on page A-71
Notes: An “ni” in the Initialization column indicates that the register is Not Initialized.
For information on Register Groups, see Table A-15 on page A-34.
An * denotes that initialization depends on the booting mode. For more information, see
“Bootloading Through The External Port” on page 6-76 or “Bootloading Through The Link
Port” on page 6-87.
A-42
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-16. I/O Processor Registers Memory Map (Cont’d)
Register Address
Register Name
Initialization
After Reset
Register Group
Page Cross
Reference
0xd7 – 0xdf
Reserved
0xe0
STCTL0
0x0000 0000
LSP
on page A-71
0xe1
SRCTL0
0x0000 0000
LSP
on page A-73
0xe2
TX0
ni
LSP
on page A-73
0xe3
RX0
ni
LSP
on page A-76
0xe4
TDIV0
ni
LSP
on page A-76
0xe5
TCNT0
ni
LSP
on page A-77
0xe6
RDIV0
ni
LSP
on page A-77
0xe7
RCNT0
ni
LSP
on page A-78
0xe8
MTCS0
ni
LSP
on page A-78
0xe9
MRCS0
ni
LSP
on page A-78
0xea
MTCCS0
ni
LSP
on page A-79
0xeb
MRCCS0
ni
LSP
on page A-79
0xec
KEYWD0
ni
LSP
on page A-80
0xed
KEYMASK0
ni
LSP
on page A-80
0xee
SPATH0
0x0000 0001
LSP
on page A-80
0xef
SPCNT0
0x0000 0001
LSP
on page A-80
0xf0
STCTL1
0x0000 0000
LSP
on page A-71
0xf1
SRCTL1
0x0000 0000
LSP
on page A-73
0xf2
TX1
ni
LSP
on page A-73
0xf3
RX1
ni
LSP
on page A-76
0xf4
TDIV1
ni
LSP
on page A-76
Reserved
Notes: An “ni” in the Initialization column indicates that the register is Not Initialized.
For information on Register Groups, see Table A-15 on page A-34.
An * denotes that initialization depends on the booting mode. For more information, see
“Bootloading Through The External Port” on page 6-76 or “Bootloading Through The Link
Port” on page 6-87.
ADSP-21160 SHARC DSP Hardware Reference
A-43
I/O Processor Registers
Table A-16. I/O Processor Registers Memory Map (Cont’d)
Register Address
Register Name
Initialization
After Reset
Register Group
Page Cross
Reference
0xf5
TCNT1
ni
LSP
on page A-77
0xf6
RDIV1
ni
LSP
on page A-77
0xf7
RCNT1
ni
LSP
on page A-78
0xf8
MTCS1
ni
LSP
on page A-78
0xf9
MRCS1
ni
LSP
on page A-78
0xfa
MTCCS1
ni
LSP
on page A-79
0xfb
MRCCS1
ni
LSP
on page A-79
0xfc
KEYWD1
ni
LSP
on page A-80
0xfd
KEYMASK1
ni
LSP
on page A-80
0xfe
SPATH1
0x0000 0001
LSP
on page A-80
0xff
SPCNT1
0x0000 0001
LSP
on page A-80
Notes: An “ni” in the Initialization column indicates that the register is Not Initialized.
For information on Register Groups, see Table A-15 on page A-34.
An * denotes that initialization depends on the booting mode. For more information, see
“Bootloading Through The External Port” on page 6-76 or “Bootloading Through The Link
Port” on page 6-87.
A-44
ADSP-21160 SHARC DSP Hardware Reference
Registers
System Configuration Register (SYSCON)
This register’s address is 0x00. The reset value for this register is 0x10, configuring the HPM bits for 16-to-32/64 bit packing (see Table A-17).
Table A-17. System Configuration Register (SYSCON) Bit
Definitions
Bit(s)
Name
Definition
0
SRST
Software Reset. This bit resets (when set, =1) the DSP. When a program sets (=1) SRST, the DSP responds to the non-maskable RSTI
interrupt and clears (=0) SRST.
1
BSO
Boot Select Override. This bit enables (if set, =1) or disables (if
cleared, =0) access to Boot Memory Space. When BSO is set, the
DSP uses the BMS select line (instead of MS3-0) to perform DMA
channel 10 accesses of external memory. The DSP uses 8- to 48-bit
packing when reading from 8-bit boot memory space, but does no
packing on writes to this space.
For appropriate byte alignment on DMA writes to boot memory
space, programs must use the shifter to place the ordered bytes for the
transfer in bits 39-32 of each internal Long word address.
2
IIVT
Internal Interrupt Vector Table. This bit forces placement of the
interrupt vector table at address 0x0004 0000 regardless of booting
mode (if 1) or allows placement of the interrupt vector table as
selected by the booting mode (if 0).
3
Reserved
6-4
HPM
Host Packing Mode. These bits select the external bus packing mode
for host accesses as follows: 000=no packing, 001=16-to-32/64 (reset
value), 010=16-to-48, 011=32-to-48, 100=32-to-32/64
7
HMSWF
Host Most Significant Word First Packing Select. This bit selects the
word packing order for host accesses as most-significant-word first (if
set, =1) or least-significant-word first (if cleared, =0).
ADSP-21160 SHARC DSP Hardware Reference
A-45
I/O Processor Registers
Table A-17. System Configuration Register (SYSCON) Bit
Definitions (Cont’d)
Bit(s)
Name
Definition
8
HPFLSH
Host Packing Status Flush. This bit flushes (when set, =1) settings
for the direct write FIFO. Flushing these settings does the following:
• Clears (=0) the HPS status bits in the SYSTAT register
• Clears (=0) the channel’s DMA request counter
• Clears (-0) any partially packed words
When a program sets (=1) HPFLSH, the DSP flushes the settings and
clears (=0) HPFLSH. There is a two-cycle effect latency in completing the flush operation.
Programs must not set the buffer’s HPFLSH during the same write
that enables the buffer. Also, programs must not set the HPFLSH bit
while the DMA channel is active. Programs should determine the
channel’s active status by reading the corresponding bit in the
DMASTAT register.
9
IMDW0
Internal Memory Block 0 Data Width. This bit selects the Normal
word data access size for internal memory Block 0 as 40-bit data (if
set, =1) or 32-bit data (if cleared, =0).
10
IMDW1
Internal Memory Block 1 Data Width. This bit selects the Normal
word data access size for internal memory Block 1 as 40-bit data (if
set, =1) or 32-bit data (if cleared, =0).
11
ADREDY
Active Drive REDY. This bit selects line driver type for the DSP’s
REDY pin as active drive (a/d) (if set, =1) or open drain (o/d) (if
cleared, =0).
15-12
MSIZE
Memory Bank Size. These bits select the size of the four external
memory banks (Bank 3-0). The external memory that is not allotted
to a bank is part of the Unbanked external memory region. The formula for external memory bank size is:
MSIZE=log2(desired bank size in words) - 13
16
BHD
Buffer Hang Disable. This bit controls whether the processor core
proceeds (hang disabled if set, =1) or is held-off (hang enabled if
cleared, =0) when the core tries to read from an empty EPBx, Tx, or
LBUFx buffer or tries to write to a full EPBX, Rx, or LBUFx buffer.
A-46
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-17. System Configuration Register (SYSCON) Bit
Definitions (Cont’d)
Bit(s)
Name
Definition
18-17
EBPR
External Bus Priority. These bits select the priority for the I/O processor’s EP bus when arbitrating access to the DSP’s external port as
follows: 00—priority rotates between DM or PM and IO buses, 01—
the winning DM or PM bus has priority over the IO bus, 10—the IO
bus has priority over the winning DM or PM bus.
19
DCPR
External Port DMA Channel Priority Rotation Enable. This bit
enables (rotates if set, =1) or disables (fixed if cleared, =0) priority
rotation among external port DMA channels (channel 10-13).
20
LDCPR
Link Port DMA Channel Priority Rotation Enable. This bit enables
(rotates if set, =1) or disables (fixed if cleared, =0) priority rotation
among link port DMA channels (channel 4-9).
21
PRROT
Link–External Port DMA Channel Priority Rotation Enable. This
bit enables (rotates if set, =1) or disables (fixed if cleared, =0) priority
rotation between link port DMA channels (channel 4-9) and external
port DMA channels (channel 10-13).
22
COD
CLKOUT Disable. This bit enables (if set, =1) or disables (if cleared,
=0) DSP clock output on the CLKOUT pin. If enabled, the DSP
outputs the clock signal on CLKOUT. If disabled, the DSP
three-states the CLKOUT pin.
This bit is the only way to control the CLKOUT pin.
31-23
Reserved
ADSP-21160 SHARC DSP Hardware Reference
A-47
I/O Processor Registers
Vector Interrupt Address Register (VIRPT)
This register’s address is 0x01. The reset value for this register is
0x0004 0014 (see Table A-18). The sequencer uses the VIRPT register to
support multiprocessor vector interrupts. The vector interrupt (VIRPTI)
permits passing interprocessor commands in multiple-processor systems.
This interrupt occurs when an external processor (a host or another DSP)
writes an address to the VIRPT register, inserting a new vector address for
VIRPTI.
Table A-18. Vector Interrupt Address Register (VIRPT) Bit Definitions
Bit(s)
Name
Definition
23-0
VIRPTA
Vector Interrupt Address. These bits contain the multiprocessor
interrupt’s vector (address). When an external processor loads an
address into this register, the DSP pushes the status stack and starts
executing the routine at the vector address.
31-25
VIRPTD
Vector Interrupt (optional) Data. These bits contain optional data
that the external processor may pass to the interrupt service routine.
External Memory Waitstate and Access Mode
Register (WAIT)
This register’s address is 0x02. The reset value for this register is
0x01ce739c, which equates to the following DSP external memory settings: asynchronous access mode for all external memory banks, seven
waitstates with a hold cycle for all accesses to external memory banks,
external DRAM page size of 256 words (if installed), and disable idle
cycle for DMA handshake (see Table A-19).
A-48
ADSP-21160 SHARC DSP Hardware Reference
Registers
Table A-19. External Memory Setup Register (WAIT) Bit
Definitions
Bit(s)
Name
Definition
1-0
EB0AM
External Bank 0 Access Mode. These bits select the access mode
for external memory Bank 0 as follows:
EBxAM=External Bank Access Mode
00=Asynchronous—DSP RDH/L and WRH/L strobes change before
CLKOUT’s edge—accesses use the waitstate count setting from
EBxWS and require external acknowledge (ACK), allowing a deasserted ACK to extend the access time.
01=Synchronous—DSP RDH/L and WRH/L strobes change on CLKOUT’s edge—reads use the waitstate count setting from EBxWS
(minimum EBxWS=001); writes are 0-wait state.
10=Synchronous—DSP RDH/L and WRH/L strobes change on CLKOUT’s edge—reads use the waitstate count setting from EBxWS
(minimum EBxWS=001); writes are 1-wait state.
11=Reserved
4-2
EB0WS
External Bank 0 Waitstates. These bit fields select the waitstates
for external memory Bank 0 as follows:
EBxWS# of WaitstatesHold Time Cycle?
0000no
0011no
0102yes
0113yes
1004yes
1015yes
1106yes
1117yes
Note that Hold Cycles applies to asynchronous mode only.
6-5
EB1AM
External Bank 1 Access Mode. (see EB0AM definition)
9-7
EB1WS
External Bank 1 Waitstates. (see EB0WS definition)
11-10
EB2AM
External Bank 2 Access Mode. (see EB0AM definition)
14-12
EB2WS
External Bank 2 Waitstates. (see EB0WS definition)
16-15
EB3AM
External Bank 3