Xilinx for DSP System Generator User Guide

Automatic Code Generation

Compiling MATLAB into an

FPGA

Importing a System Generator

Design into a Bigger System

Configurable Subsystems and

Notes for Higher Performance

Provides background on FPGAs, and discusses compilation, programming, and architectural considerations in the context of System Generator.

Describes several settings in which constructing designs in System Generator is useful.

Discusses System Generator's ability to implement device-specific hardware designs directly from a flexible, high-level, system modeling environment.

Discusses automatic code generation for System

Generator designs.

Describes how to use a subset of the MATLAB programming language to write functions that describe state machines and arithmetic operators.

Functions written in this way can be attached to blocks in System Generator and can be automatically compiled into equivalent HDL.

Discusses how to take the VHDL netlist from a System

Generator design and synthesize it in order to embed it into a larger design. Also shows how VHDL created by System Generator can be incorporated into a simulation model of the overall system.

Explains how to use configurable subsystems in

System Generator. Describes common tasks such as defining configurable subsystems, deleting and adding blocks, and using configurable subsystems to import compilation results into System Generator designs.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

FPGA Design

Processing a System Generator

Design with FPGA Physical

Design Tools

Resetting Auto-Generated Clock

Enable Logic

Design Styles for the DSP48

Using FDATool in Digital Filter

Applications

Generating Multiple Cycle-True

Islands for Distinct Clocks

Using ChipScope Pro Analyzer for Real-Time Hardware

Debugging

Suggests design practices in System Generator that lead to an efficient and high-performance implementation in an FPGA.

Describes how to take the low-level HDL produced by

System Generator and use it in tools like Xilinx's

Project Navigator, ModelSim, and Synplicity's

Synplify.

Describes the behavior of rate changing blocks from the System Generator library when the ce_clr signal is used for re-synchronization.

Describes three ways to implement and configure a

DSP48 (Xtreme DSP Slice) in System Generator

Demonstrates one way to specify, implement and simulate a FIR filter using the FDATool block.

Describes how to implement multi-clock designs in

System Generator

Demonstrated how to connect and use the Xilinx

Debug Tool called ChipScope

™

Pro within System

Generator

A Brief Introduction to FPGAs

A field programmable gate array (FPGA) is a general-purpose integrated circuit that is

“programmed” by the designer rather than the device manufacturer. Unlike an application-specific integrated circuit (ASIC), which can perform a similar function in an electronic system, an FPGA can be reprogrammed, even after it has been deployed into a system.

An FPGA is programmed by downloading a configuration program called a bitstream into static on-chip random-access memory. Much like the object code for a microprocessor, this bitstream is the product of compilation tools that translate the high-level abstractions produced by a designer into something equivalent but low-level and executable. Xilinx

System Generator pioneered the idea of compiling an FPGA program from a high-level

Simulink model.

An FPGA provides you with a two-dimensional array of configurable resources that can implement a wide range of arithmetic and logic functions. These resources include dedicated DSP blocks, multipliers, dual port memories, lookup tables (LUTs), registers, tristate buffers, multiplexers, and digital clock managers. In addition, Xilinx FPGAs contain sophisticated I/O mechanisms that can handle a wide range of bandwidth and voltage requirements. The Virtex™-4 and Virtex-II Pro family FPGAs include embedded microcontrollers (IBM PowerPC™ 405), and multi-gigabit serial transceivers. The compute and I/O resources are linked under the control of the bitstream by a programmable interconnect architecture that allows them to be wired together into systems.

FPGAs are high performance data processing devices. DSP performance is derived from the FPGA’s ability to construct highly parallel architectures for processing data. In contrast with a microprocessor or DSP processor, where performance is tied to the clock rate at which the processor can run, FPGA performance is tied to the amount of parallelism that can be brought to bear in the algorithms that make up a signal processing system. A combination of increasingly high system clock rates (current system frequencies of 100-200

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

A Brief Introduction to FPGAs

MHz are common today) and a highly-distributed memory architecture gives the system designer an ability to exploit parallelism in DSP (and other) applications that operate on data streams. For example, the raw memory bandwidth of a large FPGA running at a clock rate of 150 MHz can be hundreds of terabytes per second.

There are many DSP applications (e.g., digital up/down converters) that can be implemented only in custom integrated circuits (ICs) or in an FPGA; a von Neumann processor lacks both the compute capability and the memory bandwidth required.

Advantages of using an FPGA include significantly lower non-recurring engineering costs than those associated with a custom IC (FPGAs are commercial off-the-shelf devices), shorter time to market, and the configurability of an FPGA, which allows a design to be modified, even after deployment in an end application.

When working in System Generator, it is important to keep in mind that an FPGA has many degrees of freedom in implementing signal processing functions. You have, for example, the freedom to define data path widths throughout your system and to employ many individual data processors (e.g., multiply-accumulate engines), depending on system requirements. System Generator provides abstractions that allow you to design for an FPGA largely by thinking about the algorithm you want to implement. However, the more you know about the underlying FPGA, the more likely you are to exploit the unique capabilities an FPGA provides in achieving high performance.

The remainder of this topic is a brief introduction to some of the logic resources available in the FPGA, so that you gain some appreciation for the abstractions provided in System

Generator.

The figure above shows a physical view of a Virtex™-4 FPGA. To a signal DSP engineer, an

FPGA can be thought of as a 2-D array of logic slices striped with columns of hard macro blocks (block memory and arithmetic blocks) suitable for implementing DSP functions, embedded within a configurable interconnect mesh. In a Virtex™-4 FPGA, the DSP blocks

(shown in the next figure) can run in excess of 450 MHz, and are pitch-matched to dual port memory blocks (BRAMs) whose ports can be configured to a wide range of word sizes

(18 Kb total per BRAM). The Virtex™-4 SX55 device contains 512 such DSP blocks and

BRAMs. In System Generator, you can access all of these resources through arithmetic and

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator logic abstractions to build very high performance digital filters, FFTs, and other arithmetic and signal processing functions.

While the multiply-accumulate function supported by a Virtex™-4 DSP block is familiar to a DSP engineer, it is instructive to take a closer look at the Virtex™ FPGA family logic slice

(shown below), which is the fundamental unit of the logic fabric array.

Each logic slice contains two 4-input lookup tables (LUTs), two configurable D-flip flops, multiplexers, dedicated carry logic, and gates used for creating slice-based multipliers.

Each LUT can implement an arbitrary 4-input Boolean function. Coupled with dedicated logic for implementing fast carry circuits, the LUTs can also be used to build fast adder/subtractors and multipliers of essentially any word size. In addition to implementing Boolean functions, each LUT can also be configured as a 16x1 bit RAM or as a shift register (SRL16). An SRL16 shift register is a synchronously clocked 16x1 bit delay line with a dynamically addressable tap point.

In System Generator, these different memory options are represented with higher-level abstractions. Instead of providing a D-flip flop primitive, System Generator provides a register of arbitrary size. There are two blocks that provide abstractions of arbitrary www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

A Brief Introduction to FPGAs

width, arbitrary depth delay lines that map directly onto the SRL16 configuration. The delay block can be used for pipeline balancing, and can also be used as storage for time- division multiplexed (TDM) data streams. The addressable shift register (ASR) block, with a function depicted in the figure below, provides an arbitrary width, arbitrary depth tapped delay line. This block is of particular interest to the DSP engineer, since it can be used to implement tapped delay lines as well as sweeping through TDM data streams.

Although random access memories can be constructed either out of the BRAM or LUT

(RAM16x1) primitives, doing so can require considerable care to ensure most efficient mappings, and considerable clerical attention to detail to correctly assemble the primitives into larger structures. System Generator removes the need for such tasks.

For example, the dual port RAM (DPRAM) block shown in the figure below maps efficiently onto as many BRAM or RAM16x1 components on the device as are necessary to implement the desired memory. As can be seen from the mask dialog box for the DPRAM, the interface allows you to specify a type of memory (BRAM or RAM16x1), depth (data width is inferred from the Simulink signal driving a particular input port), initial memory contents, and other characteristics.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

In general, System Generator maps abstractions onto device primitives efficiently, freeing you from worrying about interconnections between the primitives. System Generator employs libraries of intellectual property (IP) when appropriate to provide efficient implementations of functions in the block libraries. In this way, you don’t always have to have detailed knowledge of the underlying FPGA details. However, when it makes sense to implement an algorithm using basic functions (e.g., adder, register, memory), System

Generator allows you to exploit your FPGA knowledge while reducing the clerical tasks of managing all signals explicitly.

System Generator library blocks and the mapping from Simulink to hardware are described in detail in subsequent topics of this documentation. There is a wealth of detailed information about FPGAs that can be found online at http://support.xilinx.com

, including data books, application notes, white papers, and technical articles.

Note to the DSP Engineer

System Generator extends Simulink to enable hardware design, providing high-level abstractions that can be automatically compiled into an FPGA. Although the arithmetic abstractions are suitable to Simulink (discrete time and space dynamical system simulation), System Generator also provides access to features in the underlying FPGA.

The more you know about a hardware realization (e.g., how to exploit parallelism and pipelining), the better the implementation you’ll obtain. Using IP cores makes it possible to have efficient FPGA designs that including complex functions like FFTs. System Generator also makes it possible to refine a model to more accurately fit the application.

Scattered throughout the System Generator documentation are notes that explain ways in which system parameters can be used to exploit hardware capabilities.

Note to the Hardware Engineer

System Generator does not replace hardware description language (HDL)-based design, but does makes it possible to focus your attention only on the critical parts. By analogy, most DSP programmers do not program exclusively in assembler; they start in a higherlevel language like C, and write assembly code only where it is required to meet performance requirements.

A good rule of thumb is this: in the parts of the design where you must manage internal hardware clocks (e.g., using the DDR or phased clocking), you should implement using

HDL. The less critical portions of the design can be implemented in System Generator, and then the HDL and System Generator portions can be connected. Usually, most portions of a signal processing system do not need this level of control, except at external interfaces.

System Generator provides mechanisms to import HDL code into a design (see Importing

HDL Modules ) that are of particular interest to the HDL designer.

Another aspect of System Generator that is of interest to the engineer who designs using

HDL is its ability automatically generate an HDL testbench, including test vectors. This aspect is described in the topic

HDL Testbench .

Finally, the hardware co-simulation interfaces described in the topic

Using Hardware Co-

Simulation allow you to run a design in hardware under the control of Simulink, bringing

the full power of MATLAB and Simulink to bear for data analysis and visualization.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Design Flows using System Generator

System Generator can be useful in many settings. Sometimes you may want to explore an algorithm without translating the design into hardware. Other times you might plan to use a System Generator design as part of something bigger. A third possibility is that a System

Generator design is complete in its own right, and is to be used in FPGA hardware. This topic describes all three possibilities.

Algorithm Exploration

System Generator is particularly useful for algorithm exploration, design prototyping, and model analysis. When these are the goals, you can use the tool to flesh out an algorithm in order to get a feel for the design problems that are likely to be faced, and perhaps to estimate the cost and performance of an implementation in hardware. The work is preparatory, and there is little need to translate the design into hardware.

In this setting, you assemble key portions of the design without worrying about fine points or detailed implementation. Simulink blocks and MATLAB M-code provide stimuli for simulations, and for analyzing results. Resource estimation gives a rough idea of the cost of the design in hardware. Experiments using hardware generation can suggest the hardware speeds that are possible.

Once a promising approach has been identified, the design can be fleshed out. System

Generator allows refinements to be done in steps, so some portions of the design can be made ready for implementation in hardware, while others remain high-level and abstract.

System Generator's facilities for hardware co-simulation are particularly useful when portions of a design are being refined.

Implementing Part of a Larger Design

Often System Generator is used to implement a portion of a larger design. For example,

System Generator is a good setting in which to implement data paths and control, but is less well suited for sophisticated external interfaces that have strict timing requirements. In this case, it may be useful to implement parts of the design using System Generator, implement other parts outside, and then combine the parts into a working whole.

A typical approach to this flow is to create an HDL wrapper that represents the entire design, and to use the System Generator portion as a component. The non-System

Generator portions of the design can also be components in the wrapper, or can be instantiated directly in the wrapper.

Implementing a Complete Design

Many times, everything needed for a design is available inside System Generator. For such a design, pressing the Generate button instructs System Generator to translate the design into HDL, and to write the files needed to process the HDL using downstream tools. The files written include the following:

•

HDL that implements the design itself;

•

A clock wrapper that encloses the design. This clock wrapper produces the clock and clock enable signals that the design needs.

•

A HDL testbench that encloses the clock wrapper. The testbench allows results from

Simulink simulations to be compared against ones produced by a logic simulator.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

•

Project files and scripts that allow various synthesis tools, such as XST and Synplify

Pro to operate on System Generator HDL

•

Files that allow the System Generator HDL to be used as a project in Project

Navigator.

For details concerning the files that System Generator writes, see the topic Compilation

Results

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

System Generator allows device-specific hardware designs to be constructed directly in a flexible high-level system modeling environment. In a System Generator design, signals are not just bits. They can be signed and unsigned fixed-point numbers, and changes to the design automatically translate into appropriate changes in signal types. Blocks are not just stand-ins for hardware. They respond to their surroundings, automatically adjusting the results they produce and the hardware they become.

System Generator allows designs to be composed from a variety of ingredients. Data flow models, traditional hardware design languages (VHDL, Verilog, and EDIF), and functions derived from the MATLAB programming language, can be used side-by-side, simulated together, and synthesized into working hardware. System Generator simulation results are bit and cycle-accurate. This means results seen in simulation exactly match the results that are seen in hardware. System Generator simulations are considerably faster than those from traditional HDL simulators, and results are easier to analyze.

System Generator Blocksets

Signal Types

Bit-True and Cycle-True

Modeling

Timing and Clocking

Synchronization Mechanisms

Block Masks and Parameter

Passing

Resource Estimation

Describes how System Generator's blocks are organized in libraries, and how the blocks can be parameterized and used.

Describes the data types used by System Generator and ways in which data types can be automatically assigned by the tool.

Specifies the relationship between the Simulink-based simulation of a System Generator model and the behavior of the hardware that can be generated from it.

Describes how clocks are implemented in hardware, and how their implementation is controlled inside

System Generator. Explains how System Generator translates a multirate Simulink model into working clock-synchronous hardware.

Describes mechanisms that can be used to synchronize data flow across the data path elements in a high-level System Generator design, and describes how control path functions can be implemented.

Explains how parameterized systems and subsystems are created in Simulink.

Describes how to generate estimates of the hardware needed to implement a System Generator design.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

System Generator Blocksets

A Simulink blockset is a library of blocks that can be connected in the Simulink block editor to create functional models of a dynamical system. For system modeling, System

Generator blocksets are used like other Simulink blocksets. The blocks provide abstractions of mathematical, logic, memory, and DSP functions that can be used to build sophisticated signal processing (and other) systems. There are also blocks that provide interfaces to other software tools (e.g., FDATool, ModelSim) as well as the System

Generator code generation software.

System Generator blocks are bit-accurate and cycle-accurate. Bit-accurate blocks produce values in Simulink that match corresponding values produced in hardware; cycle-accurate blocks produce corresponding values at corresponding times.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

Xilinx Blockset

The Xilinx Blockset is a family of libraries that contain basic System Generator blocks.

Some blocks are low-level, providing access to device-specific hardware. Others are high- level, implementing (for example) signal processing and advanced communications algorithms. For convenience, blocks with broad applicability (e.g., the Gateway I/O blocks) are members of several libraries. Every block is contained in the Index library. The libraries are described below.

Library

Index

Basic Elements

Communication

Control Logic

Data Types

DSP

Math

Memory

Shared Memory

Tools

Description

Every block in the Xilinx Blockset.

ElementsStandard building blocks for digital logic

Forward error correction and modulator blocks, commonly used in digital communications systems

Blocks for control circuitry and state machines

Blocks that convert data types (includes gateways)

Digital signal processing (DSP) blocks

Blocks that implement mathematical functions

Blocks that implement and access memories

Blocks that implement and access Xilinx shared memories

“Utility” blocks, e.g., code generation (System Generator block), resource estimation, HDL co-simulation, etc

Note:

More information concerning blocks can be found in the topic Xilinx Blockset .

Xilinx Reference Blockset

The Xilinx Reference Blockset contains composite System Generator blocks that implement a wide range of functions. Blocks in this blockset are organized by function into different libraries. The libraries are described below.

Library

Communication

Control Logic

DSP

Imaging

Math

Description

Blocks commonly used in digital communications systems

LogicBlocks used for control circuitry and state machines

Digital signal processing (DSP) blocks

Image processing blocks

Blocks that implement mathematical functions

Each block in this blockset is a composite, i.e., is implemented as a masked subsystem, with parameters that configure the block.

You can use blocks from the Reference Blockset libraries as is, or as starting points when constructing designs that have similar characteristics. Each reference block has a

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator description of its implementation and hardware resource requirements. Individual documentation for each block is also provided in the topic Xilinx Reference Blockset .

Signal Types

In order to provide bit-accurate simulation of hardware, System Generator blocks operate on Boolean and arbitrary precision fixed-point values. By contrast, the fundamental scalar signal type in Simulink is double precision floating point. The connection between Xilinx blocks and non-Xilinx blocks is provided by gateway blocks. The gateway in converts a double precision signal into a Xilinx signal, and the gateway out converts a Xilinx signal into double precision. Simulink continuous time signals must be sampled by the Gateway In block.

Most Xilinx blocks are polymorphic, i.e., they are able to deduce appropriate output types based on their input types. When full precision is specified for a block in its parameters dialog box, System Generator chooses the output type to ensure no precision is lost. Sign extension and zero padding occur automatically as necessary. User-specified precision is usually also available. This allows you to set the output type for a block and to specify how quantization and overflow should be handled. Quantization possibilities include unbiased rounding towards plus or minus infinity, depending on sign, or truncation. Overflow options include saturation, truncation, and reporting overflow as an error.

Note:

System Generator data types can be displayed by selecting Format > Port Data Types in

Simulink. Displaying data types makes it easy to determine precision throughout a model. If, for example, the type for a port is Fix_11_9, then the signal is a two's complement signed 11-bit number www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

having nine fractional bits. Similarly, if the type is Ufix_5_3, then the signal is an unsigned 5-bit number having three fractional bits.

In the System Generator portion of a Simulink model, every signal must be sampled.

Sample times may be inherited using Simulink's propagation rules, or set explicitly in a block customization dialog box. When there are feedback loops, System Generator is sometimes unable to deduce sample periods and/or signal types, in which case the tool issues an error message. Assert blocks must be inserted into loops to address this problem.

It is not necessary to add assert blocks at every point in a loop; usually it suffices to add an assert block at one point to “break” the loop.

Note:

Simulink can display a model by shading blocks and signals that run at different rates with different colors (Format > Sample Time Colors in the Simulink pulldown menus). This is often useful in understanding multirate designs

Bit-True and Cycle-True Modeling

Simulations in System Generator are bit-true and cycle-true. To say a simulation is bit-true means that at the boundaries (i.e., interfaces between System Generator blocks and non-

System Generator blocks), a value produced in simulation is bit-for-bit identical to the corresponding value produced in hardware. To say a simulation is cycle-true means that at the boundaries, corresponding values are produced at corresponding times. The boundaries of the design are the points at which System Generator gateway blocks exist.

When a design is translated into hardware, Gateway In (respectively, Gateway Out) blocks become top-level input (resp., output) ports.

Timing and Clocking

Discrete Time Systems

Designs in System Generator are discrete time systems. In other words, the signals and the blocks that produce them have associated sample rates. A block’s sample rate determines how often the block is awoken (allowing its state to be updated). System Generator sets most sample rates automatically. A few blocks, however, set sample rates explicitly or implicitly.

Note:

For an in-depth explanation of Simulink discrete time systems and sample times, consult the

Using Simulink reference manual from the MathWorks, Inc.

A simple System Generator model illustrates the behavior of discrete time systems.

Consider the model shown below. It contains a gateway that is driven by a Simulink source

(Sine Wave), and a second gateway that drives a Simulink sink (Scope).

The Gateway In block is configured with a sample period of one second. The Gateway Out block converts the Xilinx fixed-point signal back to a double (so it can analyzed in the

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Simulink scope), but does not alter sample rates. The scope output below shows the unaltered and sampled versions of the sine wave.

Multirate Models

System Generator supports multirate designs, i.e., designs having signals running at several sample rates. System Generator automatically compiles multirate models into hardware. This allows multirate designs to be implemented in a way that is both natural and straightforward in Simulink.

Rate-Changing Blocks

System Generator includes blocks that change sample rates. The most basic rate changers are the Up Sample and Down Sample blocks. As shown in the figure below, these blocks explicitly change the rate of a signal by a fixed multiple that is specified in the block’s dialog box.

Other blocks (e.g., the Parallel To Serial and Serial To Parallel converters) change rates implicitly in a way determined by block parameterization.

Consider the simple multirate example below. This model has two sample periods, SP1 and SP2. The Gateway In dialog box defines the sample period SP1. The Down Sample block causes a rate change in the model, creating a new rate SP2 which is half as fast as SP1. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

Hardware Oversampling

Some System Generator blocks are oversampled, i.e., their internal processing is done at a rate that is faster than their data rates. In hardware, this means that the block requires more than one clock cycle to process a data sample. In Simulink such blocks do not have an observable effect on sample rates.

One block that can be oversampled is the DAFIR FIR filter. An oversampled DAFIR processes samples serially, thus running at a higher rate, but using less hardware.

Although blocks that are oversampled do not cause an explicit sample rate change in

Simulink, System Generator considers the internal block rate along with all other sample rates when generating clocking logic for the hardware implementation. This means that you must consider the internal processing rates of oversampled blocks when you specify the Simulink system period value in the System Generator block dialog box.

Asynchronous Clocking

System Generator focuses on the design of hardware that is synchronous to a single clock.

It can, under some circumstances, be used to design systems that contain more than one clock. This is possible provided the design can be partitioned into individual clock domains with the exchange of information between domains being regulated by dual port memories and FIFOs. System Generator fully supports such multi-clock designs, including the ability to simulate them in Simulink and to generate complete hardware descriptions.

Details are discussed in the topic

Generating Multiple Cycle-True Islands for Distinct

Clocks . The remainder of this topic focuses exclusively on the clock-synchronous aspects

of System Generator. This discussion is relevant to both single-clock and multiple-clock designs.

Synchronous Clocking

As shown in the figure below, when you use the System Generator token to compile a design into hardware, there are three clocking options for Multirate implementation: (1)

Clock Enables (the default), (2) Hybrid DCM-CE, and (3) Expose Clock Ports.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

The Clock Enables Option

When System Generator compiles a model into hardware with the Clock Enable option selected, System Generator preserves the sample rate information of the design in such a way that corresponding portions in hardware run at appropriate rates. In hardware,

System Generator generates related rates by using a single clock in conjunction with clock enables, one enable per rate. The period of each clock enable is an integer multiple of the period of the system clock.

Inside Simulink, neither clocks nor clock enables are required as explicit signals in a

System Generator design. When System Generator compiles a design into hardware, it uses the sample rates in the design to deduce what clock enables are needed. To do this, it employs two user-specified values from the System Generator block: the Simulink system period and FPGA clock period. These numbers define the scaling factor between time in a

Simulink simulation, and time in the actual hardware implementation. The Simulink system period must be the greatest common divisor (gcd) of the sample periods that appear in the model, and the FPGA clock period is the period, in nanoseconds, of the system clock. If p represents the Simulink system period, and c represents the FPGA system clock period, then something that takes kp units of time in Simulink takes k ticks of the system clock (hence kc nanoseconds) in hardware.

To illustrate this point, consider a model that has three Simulink sample periods 2, 3, and

4. The gcd of these sample periods is 1, and should be specified as such in the Simulink

System Period field for the model. Assume the FPGA Clock Period is specified to be 10ns.

With this information, the corresponding clock enable periods can be determined in hardware.

In hardware, we refer to the clock enables corresponding to the Simulink sample periods 2,

3, and 4 as CE2, CE3, and CE4, respectively. The relationship of each clock enable period to the system clock period can be determined by dividing the corresponding Simulink sample period by the Simulink System Period value. Thus, the periods for CE2, CE3, and

CE4 equal 2, 3, and 4 system clock periods, respectively. A timing diagram for the example clock enable signals is shown below:

The Hybrid DCM-CE Option

If the implementation target is an FPGA with a Digital Clock Manager (DCM), you can choose to drive the clock tree with a DCM. The DCM option is desirable when high fanout on clock enable nets make it difficult to achieve timing closure.

System Generator instantiates the DCM in a top-level HDL clock wrapper and configures the DCM to provide up to three clock ports at different rates for Virtex ™ -4 and Virtex ™ -5 and up to two clock ports for Spartan-3A DSP . If the design has more clock ports than the DCM can support, the remaining clocks are supported with the CE (clock enable) configuration.

The mapping of rates to the DCM outputs is done according to the following priority scheme:

CLK0 > CLK2x > CLKdv > CLKfx. The DCM supports the higher clock rates first.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

A dcm_reset input port is exposed on the top-level wrapper to allow the external design to reset the DCM after bitstream configuration. A dcm_locked output port is also exposed to help the external design synchronize the input data with the single clk input port.

Known Limitations

: The following System Generator blocks are not supported by the

Hybrid DCM-CE Option:

•

Clock Enable Probe

•

Clock Probe

•

DAFIR

•

Downsample - when the Sample option First value of the frame is selected

•

FIR Compiler - when the core rate is not equal to the input sample rate

•

Parallel to Serial- when the Latency option is specified as 0 (zero)

•

Time Division De-Multiplexer

•

Time Division Multiplexer

•

Upsample - when the Copy samples (otherwise zeros are inserted) option is not selected.

Note:

For Release 10.1.2, the Hybrid DCM-CE option was tested in hardware using Virtex

™

-4 and

Virtex

™

-5 platforms at 400MHz and the Spartan-3A DSP platform at 190MHz.

The Expose Clock Ports Option

When you select this option, System Generator creates a top-level wrapper that exposes a clock port for each rate. You can then manually instantiate a clock generator outside the design to drive the clock ports.

Tutorial Example: Using the Hybrid DCM-CE Option

The following step-by-step example will show you how to select the Hybrid DCM-CE option, netlist the HDL design, implement the design in ISE™, simulate the design and examine the files and reports to verify that the DCM is properly instantiated and configured.

The hybrid_dcm_ce_case1.mdl design example is located at the following pathname

<sysgen_tree>/examples/clocking_options/hybrid_dcm_ce_case1/hybrid

_dcm_ce_case1.mdl

Open the model in MATLAB and observe the following blocks:

•

Addressable Shift Register (ASR)

: used to implement the input delay buffer. The address port runs n times faster than the data port, where n is the number of the filter taps (5 for this example)

•

Coefficient ROM

: used to store the filter coefficients

•

Counter

: used to generate addresses for the ROM and ASR

•

Comparator

: used to generate the reset and enable signals

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

•

MAC Engine

: used as a Multiply-Accumulator operator for the filter

Double-click on the System Generator token to bring up the following dialog box:

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

As shown, select Hybrid DCM-CE, then click Generate. After a few moments, a subdirectory named hdl_netlist is created in the current working directory containing the generated files.

In the MATLAB Current Directory window, double-click on the file hybrid_dcm_ce_case1_sysgen.log. As shown below, the DCM clocks are listed first (highest rates first), followed by the CE driven clocks.

Launch ISE, then load the ISE project at pathname

./hdl_netlist_dcm/hybrid_dcm_ce_case1_dcm_mcw.ise

Under the Project Navigator Processes tab, double-click on Implement Design.

From the Project Navigator Sources tab, do the following: a.

Double-click on the file hybrid_dcm_ce_case1_dcm_mcw.vhd, then scroll down to view the DCM component declaration as shown below by the VHDL code snippet: b. Observe that System Generator automatically infers and instantiates the DCM instance and its parameters according to the required clock outputs.

Close the VHDL file.

Next, you are going to examine the clock propagation by examining the ISE timing report.

First, you must generate the report.

Select the following: Processes tab > Implement Design > Place & Route > Generate

Post-Place & Route Static Timing

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Once finished, double -click on Analyze Post-Place & Route Static Timing and you should see the information in the figure below:

This design is comprised of six clock rates – 1, 2, 4, 8, 20, 40 with respect to the 10 ns global clock constraint. The timing report validates the correct clock generation and propagation by System Generator as follows:

♦

DCM-based clocks

: clk_1 (CLK0 ->10 ns), clk_2 (CLKFX ->20 ns) , clk_4 (CLKDIV

->40 ns) generated by the DCM based on the 10 ns global clock input

Clock Enable-based clocks

: ce_8 (80 ns), ce_20 (200 ns), ce_40 (400 ns) generated by clock enables based on the clk_4 clock input

Next you want to perform a behavior simulation using the ModelSim.

As shown in the following figure, move to the Sources for dialog box in the Sources window, then select Behavioral Simulation www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

Note:

System Generator automatically creates the top-wrapper VHDL testbench, script file and input/output stimulus data files. The Processes tab changes and displays according to the

Sources type being selected.

1. Select

2. Double Click

10. Simulate the design, as shown above, by double-click on Simulate Behavioral Model in the Processes window

11. After the simulation is finished, you should be able to observe the simulation waveforms as shown in the figure below:

All DCM clocks are included in the top-level wrapper testbench file

(hybrid_dcm_ce_case1_dcm_mcw_tb.vhd) – clk_1, clk_2 and clk_4.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Summary

When you select the Hybrid DCM-CE option, System Generator automatically infers and instantiates a DCM without further manual intervention. In addition, the tool intelligently generates different clock rates by using a combination of DCM and CE clock generation algorithms and by assigning appropriate clock rates to either the DCM or CE in order to obtain optimal Quality of Results and low power consumption. You do not have to set attributes or specify DCM clock outputs. You should expect minimal clock skew when selecting the Hybrid DCM-CE option compared to the Clock Enables option alone.

Tutorial Example: Using the Expose Clock Ports Option

The following step-by-step example will show you how to select the Expose Clock Ports option, netlist the HDL design, implement the design in ISE, simulate the design, then examine the files and reports to verify the design.

The expose_clock_ports_case1 design example is located at the following pathname

<sysgen_tree>/examples/clocking_options/expose_clock_ports_case1/e xpose_clock_ports_case1.mdl

Open the model in MATLAB and observe the following blocks:

•

Addressable Shift Register (ASR)

: used to implement the input delay buffer. The address port runs n times faster than the data port, where n is the number of the filter taps (5 for this example)

•

Coefficient ROM

: used to store the filter coefficients

•

Counter

: used to generate addresses for the ROM and ASR

•

Comparator

: used to generate the reset and enable signals

•

MAC Engine

: used as a Multiply-Accumulator operator for the filter

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

Double-click on the System Generator token to bring up the following dialog box:

As shown above, select Expose Clock Ports, then click Generate. After a few moments, a sub-directory named hdl_netlist is created in the current working directory containing the generated files.

Launch ISE, then load the ISE project at pathname

./hdl_netlist/expose_clock_ports_case1_mcw.ise

Under the Project Navigator Processes tab, double-click on Implement Design.

From the Project Navigator Sources tab, do the following: a.

double-click on the file expost_clock_ports_case1_mcw.vhd, then scroll down to view the entity named expose_clock_ports_mcw, as shown below: b. Observe that System Generator infers the clocks based on the different rates in the design and brings the clock ports to the top-level wrapper. Since this design contains two clock rates, clocks clk_1 and clk_5 are pulled to the top-level wrapper. This will allow you to directly drive the multiple synchronous clocks from outside the System Generator design.

Close the VHDL file.

Next you want to perform a behavior simulation using the ModelSim.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

As shown below, move to the Sources for dialog box in the Sources window, then select Behavioral Simulation

Note:

System Generator automatically creates the top-wrapper VHDL testbench, script file and input/output stimulus data files. The Processes tab changes and displays according to the

Sources type being selected.

1. Select

2. Double Click

Simulate the design, as shown above, by double-click on Simulate Behavioral Model in the Processes window

After the simulation is finished, you should be able to observe the simulation waveforms as shown in the figure below:

Summary

When you select the Expose Clock Ports option, System Generator automatically infers the correct clocks from the design rates and exposes the clock ports in the top-level wrapper.

The clock rates are determined by the same methodology when you use the Clock Enables option. You can now drive the exposed clock ports from an external synchronous clock source.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

Synchronization Mechanisms

System Generator does not make implicit synchronization mechanisms available. Instead, synchronization is the responsibility of the designer, and must be done explicitly.

Valid Ports

System Generator provides several blocks (in particular, a FIFO) that can be used for synchronization. Several blocks provide input (respectively, output) ports that specify when an input (resp., output) sample is valid. Such ports can be chained, affording a primitive form of flow control. Blocks with such ports include the FFT, FIR, and Viterbi.

Indeterminate Data

Indeterminate values are common in many hardware simulation environments. Often they are called “don’t cares” or “Xs”. In particular, values in System Generator simulations can be indeterminate. A dual port memory block, for example, can produce indeterminate results if both ports of the memory attempt to write the same address simultaneously.

What actually happens in hardware depends upon effectively random implementation details that determine which port sees the clock edge first. Allowing values to become indeterminate gives the system designer greater flexibility. Continuing the example, there is nothing wrong with writing to memory in an indeterminate fashion if subsequent processing does not rely on the indeterminate result.

HDL modules that are brought into the simulation through HDL co-simulation are a common source for indeterminate data samples. System Generator presents indeterminate values to the inputs of an HDL co-simulating module as the standard logic vector 'XXX . .

. XX'.

Indeterminate values that drive a Gateway Out become what are called NaNs. (NaN abbreviates “not a number”.) In a Simulink scope, NaN values are not plotted. Conversely,

NaNs that drive a Gateway In become indeterminate values. System Generator provides an Indeterminate Probe block that allows for the detection of indeterminate values. This probe cannot be translated into hardware.

In System Generator, any arithmetic signal can be indeterminate, but Boolean signals cannot be. If a simulation reaches a condition that would force a Boolean to become indeterminate, the simulation is halted and an error is reported. Many Xilinx blocks have control ports that only allow Boolean signals as inputs. The rule concerning indeterminate

Booleans means that such blocks never see an indeterminate on a control port

A UFix_1_0 is a type that is equivalent to Boolean except for the above restriction concerning indeterminate data.

Block Masks and Parameter Passing

The same scoping and parameter passing rules that apply to ordinary Simulink blocks apply to System Generator blocks. Consequently, blocks in the Xilinx Blockset can be parameterized using MATLAB variables and expressions. This capability makes possible highly parametric designs that take advantage of the expressive and computational power of the MATLAB language.

Block Masks

In Simulink, blocks are parameterized through a mechanism called masking. In essence, a block can be assigned mask variables whose values can be specified by a user through dialog

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator box prompts or can be calculated in mask initialization commands. Variables are stored in a mask workspace. A mask workspace is local to the blocks under the mask and cannot be accessed by external blocks.

Note:

It is possible for a mask to access global variables and variables in the base workspace. To access a base workspace variable, use the MATLAB evalin function. For more information on the

MATLAB and Simulink scoping rules, refer to the manuals titled Using MATLAB and Using Simulink from The Mathworks, Inc.

Parameter Passing

It is often desirable to pass variables to blocks inside a masked subsystem. Doing so allows the block’s configuration to be determined by parameters on the enclosing subsystem. This technique can be applied to parameters on blocks in the Xilinx blockset whose values are set using a listbox, radio button, or checkbox. For example, when building a subsystem that consists of a multiply and accumulate block, you can create a parameter on the subsystem that allows you to specify whether to truncate or round the result. This parameter will be called trunc_round as shown in the figure below.

As shown below, in the parameter editing dialog for the accumulator and multiplier blocks, there are radio buttons that allow either the truncate or round option to be selected.

In order to use a parameter rather than the radio button selection, right click on the radio button and select: “Define With Expression”. A MATLAB expression can then be used as the parameter setting. In the example below, the trunc_round parameter from the www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

System-Level Modeling in System Generator

subsystem mask can be used in both the accumulator and multiply blocks so that each block will use the same setting from the mask variable on the subsystem.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Resource Estimation

System Generator supplies tools that estimate the FPGA hardware resources needed to implement a design. Estimates include numbers of slices, lookup tables, flip-flops, block memories, embedded multipliers, I/O blocks and tristate buffers. These estimates make it easy to determine how design choices affect hardware requirements. To estimate the resources needed for a subsystem, drag a Resource Estimator block into the subsystem, double-click on the estimator, and press the Estimate button.

Automatic Code Generation

System Generator automatically compiles designs into low-level representations. The ways in which System Generator compiles a model can vary, and depend on settings in the

System Generator block. In addition to producing HDL descriptions of hardware, the tool generates auxiliary files. Some files (e.g., project files, constraints files) assist downstream tools, while others (e.g., VHDL testbench) are used for design verification.

Compiling and Simulating Using the System Generator Block

Compilation Results

HDL Testbench

Describes how to use the System Generator block to compile designs into equivalent low-level HDL.

Describes the low-level files System Generator produces when HDL Netlist is selected on the System

Generator block and Generate is pushed.

Describes the VHDL testbench that System Generator can produce.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Automatic Code Generation

Compiling and Simulating Using the System Generator Block

System Generator automatically compiles designs into low-level representations. Designs are compiled and simulated using the System Generator block. This topic describes how to use the block.

Before a System Generator design can be simulated or translated into hardware, the design must include a System Generator block. When creating a new design, it is a good idea to add a System Generator block immediately. The System Generator block is a member of the Xilinx Blockset’s Basic Elements and Tools libraries. As with all Xilinx blocks, the

System Generator block can also be found in the Index library.

A design must contain at least one System Generator block, but can contain several System

Generator blocks on different levels (one per level). A System Generator block that is underneath another in the hierarchy is a slave; one that is not a slave is a master. The scope of a System Generator block consists of the level of hierarchy into which it is embedded and all subsystems below that level. Certain parameters (e.g. Simulink System Period) can be specified only in a master.

Once a System Generator block is added, it is possible to specify how code generation and simulation should be handled. The block’s dialog box is shown below:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Compilation Type and the Generate Button

Pressing the Generate button instructs System Generator to compile a portion of the design into equivalent low-level results. The portion that is compiled is the sub-tree whose root is the subsystem containing the block. (To compile the entire design, use a System

Generator block placed at the top of the design.) The compilation type (under

Compilation

) specifies the type of result that should be produced. The possible types are

•

Two types of Netlists, HDL Netlist and NGC Netlist

•

Bitstream

- produces an FPGA configuration bitstream that is ready to run in a hardware FPGA platform

•

EDK Export Tool

- for exporting to the Xilinx Embedded Development Kit various varieties of hardware co-simulation, and

•

Timing Analysis

- a report on the timing of the design.

HDL Netlist is the type used most often. In this case, the result is a collection of HDL and

EDIF files, and a few auxiliary files that simplify downstream processing. The collection is ready to be processed by a synthesis tool (e.g., XST), and then fed to the Xilinx physical design tools (i.e., ngdbuild, map, par, and bitgen) to produce a configuration bitstream for

a Xilinx FPGA. The files that are produced are described in more detail in Compilation

Results

NGC Netlist is similar to HDL Netlist but the resulting files are NGC files instead of HDL files.

When the type is a variety of hardware co-simulation, then System Generator produces an

FPGA configuration bitstream that is ready to run in a hardware FPGA platform. The particular platform depends on the variety chosen. For example, when the variety is

Hardware Co-simulation > XtremeDSP Development Kit > PCI and USB

, then the bitstream is suitable for the XtremeDSP board (available for separate purchase from

Xilinx). System Generator also produces a hardware co-simulation block to which the bitstream is associated. This block is able to participate in Simulink simulations. It is functionally equivalent to the portion of the design from which it was derived, but is implemented by its bitstream. In a simulation, the block delivers the same results as those produced by the portion, but the results are calculated in working hardware.

Note:

It is possible to customize the list of compilation types. See the topic Hardware Co-Simulation

Installation for details.

The remaining compilation parameters are described in the table below. Some are available only when the compilation type is HDL Netlist. For example, the clock pin location cannot be chosen for a hardware co-simulation compilation because it is fixed in each hardware FPGA platform.

Control

Part

Target Directory

Description

Defines the FPGA part to be used.

Defines where System Generator should write compilation results.

Because System Generator and the FPGA physical design tools typically create many files, it is best to create a separate target directory, i.e., a directory other than the directory containing your

Simulink model files. The directory can be an absolute path (e.g. c:\netlist) or a path relative to the directory containing the model

(e.g. netlist).

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Automatic Code Generation

Control

Synthesis Tool

Hardware Description

Language

Create Testbench

Description

Specifies the tool to be used to synthesize the design. The possibilities are Synplify, Synplify Pro and Xilinx XST.

Specifies the language to be used for HDL netlist of the design. The possibilities are VHDL and Verilog.

This instructs System Generator to create an HDL testbench.

Simulating the testbench in an HDL simulator compares Simulink simulation results with ones obtained from the compiled version of the design. To construct test vectors, System Generator simulates the design in Simulink, and saves the values seen at gateways. The top

HDL file for the testbench is named <name>_testbench.vhd/.v, where <name> is a name derived from the portion of the design being tested and the extension is dependent on the hardware description language.

Import as

Configurable

FPGA Clock Period

Tells System Generator to do two things: 1) Construct a block to which the results of compilation are associated, and 2) Construct a configurable subsystem consisting of the block and the original subsystem from which the block was derived. See

Configurable

Subsystems and System Generator for details.

Defines the period in nanoseconds of the system clock.

The value need not be an integer. The period is passed to the Xilinx implementation tools through a constraints file, where it is used as the global PERIOD constraint. Multicycle paths are constrained to integer multiples of this value.

Clock Pin Location

Multirate implementation

Defines the pin location for the hardware clock. This information is passed to the Xilinx implementation tools through a constraints file.

Clock Enables (default)

: Creates a clock enable generator circuit to drive a multirate design.

Hybrid DCM-CE

: Creates a clock wrapper with a DCM that can drive up to three clock ports at different rates for Virtex

™

-4 and

Virtex ™ -5 and up to two clock ports for Spartan-3A DSP. The mapping of rates to the DCM output ports is done using the following priority scheme: CLK0 > CLK2x > CLKdv > CLKfx. The

DCM honors the higher clock rates first. If the design contains more clocks than the DCM can handle, the remaining clocks are implemented using the Clock Enable configuration.

A reset input port is exposed on the DCM clock wrapper to allow resetting the DCM and a locked output port is exposed to help the external design synchronize the input data with the single clk input pin.

Expose Clock Ports

: This option exposes multiple clock ports on the top-level of the System Generator design so you can apply multiple synchronous clock inputs from outside the design.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Control

DCM input clock period(ns)

Provide clock enable clear pin

Description

Specify if different than the FPGA clock period(ns) option

(system clock). The FPGA clock period (system clock) will then be derived from this hardware-defined input.

This instructs System Generator to provide a ce_clr port on the top-level clock wrapper. The ce_clr signal is used to reset the clock enable generation logic. Capability to reset clock enable generations logic allows designs to have dynamic control for specifying the beginning of data path sampling. See the topic for details.

Simulink System Period

You must specify a value for Simulink System Period in the System Generator block dialog box. This value tells the underlying rate, in seconds, at which simulations of the design should run. The period must evenly divide all sample periods in the design. For example, if the design consists of blocks whose sample periods are 2, 6, and 8, then the largest acceptable sample period is 2, though other values such as 1 and 0.5 are also acceptable. Sample periods arise in three ways: some are specified explicitly, some are calculated automatically, and some arise implicitly within blocks that involve internal rate changes. For more information on how the system period setting affects the hardware

clock, refer to Timing and Clocking

Before running a simulation or compiling the design, System Generator verifies that the period evenly divides every sample period in the design. If a problem is found, System

Generator opens a dialog box suggesting an appropriate value. Clicking the button labeled

Update

instructs System Generator to use the suggested value. To see a summary of period conflicts, click the button labeled View Conflict Summary. If you allow System Generator to update the period, you must restart the simulation or compilation.

It is possible to assemble a System Generator model that is inconsistent because its periods cannot be reconciled. (For example, certain blocks require that they run at the system rate.

Driving an up-sampler with such a block produces an inconsistent model.) If, even after updating the system period, System Generator reports there are conflicts, then the model is inconsistent and must be corrected.

The period control is hierarchical; see the discussion of hierarchical controls below for details.

Block Icon Display

The options on this control affect the display of the block icons on the model. After compilation (which occurs when Generating, Simulating, or by pressing Control-D) of the model various information about the block in your model can be displayed, depending on which option is chosen.

•

Default—basic information about port directions are shown

•

Sample rates—the sample rates of each port are shown

•

Pipeline stages—the number of pipeline stages are shown

•

HDL port names—the names of the ports are shown

•

Input data types—the input data types for each port are shown

•

Output data types—output data types for each port are shown

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Automatic Code Generation

Hierarchical Controls

The Simulink System Period control (see the topic

Simulink System Period

above) on the

System Generator block is hierarchical. A hierarchical control on a System Generator block applies to the portion of the design within the scope of the block, but can be overridden on other System Generator blocks deeper in the design. For example, suppose Simulink

System Period

is set in a System Generator block at the top of the design, but is changed in a System Generator block within a subsystem S. Then that subsystem will have the second period, but the rest of the design will use the period set in the top level.

Compilation Results

In topic discusses the low-level files System Generator produces when HDL Netlist is selected on the System Generator block and Generate is clicked. The files consist of HDL,

NGC and EDIF that implement the design. In addition, System Generator produces auxiliary files that simplify downstream processing, e.g., bringing the design into Project

Navigator, simulating using an HDL simulator, and synthesizing using various synthesis tools. All files are written to the target directory specified on the System Generator block. If no testbench is requested, then the key files produced by System Generator are the following:

File Name or Type

<design>.vhd/.v

<design>_cw.vhd/.v

.edn

and .ngc files globals

<design>_cw.xcf (or .ncf)

<design>_cw.ise

hdlFiles synplify_<design>.prj, or xst_<design>.pr

vcom.do

Description

This contains most of the HDL for the design

This is a HDL wrapper for <design>_files.vhd/.v. It drives clocks and clock enables.

Besides writing HDL, System Generator runs CORE

Generator

™

(coregen) to implement portions of the design.

Coregen writes EDIF files whose names typically look something like multiplier_virtex2_6_0_83438798287b830b.edn

Other required files may be supplied as .ngc files.

This file consists of key/value pairs that describe the design.

The file is organized as a Perl hash table so that the keys and values can be made available to Pearl scripts using Perl evals.

This contains timing and port location constraints. These are used by the Xilinx synthesis tool XST and the Xilinx implementation tools. If the synthesis tool is set to something other than XST, then the suffix is changed to .ncf.

This allows the HDL and EDIF to be brought into the Xilinx project management tool Project Navigator.

This contains the full list of HDL files written by System

Generator. The files are listed in the usual HDL dependency order.

These files allow the design to be compiled by the synthesis tool you specified.

This script can be used in ModelSim to compile the HDL for a behavioral simulation of the design.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

If a testbench is requested, then, in addition to the above, System Generator produces files that allow simulation results to be compared. The comparisons are between Simulink simulation results and corresponding results from ModelSim. The additional files are the following:

File Name or Type

Various .dat files

<design>_tb.vhd/.v

vsim.do

pn_behavioral.do, pn_postmap.do, pn_postpar.do, pn_posttranslate.do

Description

These contain the simulation results from Simulink.

This is a testbench that wraps the design. When simulated in

ModelSim, this testbench compares simulation results from

Simulink against those produced by ModelSim.

This script can be used in ModelSim to run a testbench simulation.

These files allow various ModelSim simulations to be started inside Project Navigator.

Using the System Generator Constraints File

When a design is compiled, System Generator produces a constraints file that tells downstream tools how to process the design. This enables the tools to produce a higher quality implementation, and to do so using considerably less time. Constraints supply the following:

•

The period to be used for the system clock;

•

The speed, with respect to the system clock, at which various portions of the design must run;

•

The pin locations at which ports should be placed;

•

The speed at which ports must operate.

The file format depends on the synthesis tool that is specified in the System Generator block. When XST is selected, the file is written in the XCF format; for Synplify and Synplify

Pro, the NCF format is used. The file name ends with.xcf or .ncf, as appropriate.

System Clock Period

The system clock period (i.e., the period of the fastest hardware clock in the design) can be specified in the System Generator block. System Generator writes this period to the constraints file. Downstream tools use the period as a goal when implementing the design.

Multicycle Path Constraints

Many designs consist of parts that run at different clock rates. For the fastest part, the system clock period is used. For the remaining parts, the clock period is an integer multiple of the system clock period. It is important that downstream tools know what speed each part of the design must achieve. With this information, efficiency and effectiveness of the tools are greatly increased, resulting in reduced compilation times and improved hardware realizations. The division of the design into parts, and the speed at which each part must run, are specified in the constraints file using multicycle path constraints.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Automatic Code Generation

IOB Timing and Placement Constraints

When translated into hardware, System Generator's Gateway In and Gateway Out blocks become input and output ports. The locations of these ports and the speeds at which they must operate can be entered in the Gateway In and Out parameter dialog boxes.

See the descriptions of the Gateway In block and the Gateway Out block for more information. Port location and speed are specified in the constraints file by IOB timing.

Constraints Example

The figure below shows a small multirate design and the constraints System Generator produces for it.

The up sampler doubles the rate, and the down sampler divides the rate by three. Assume the system clock period is 10 ns. Then the clock periods are 10 ns for the FIR, 20 ns for the input register, and 30 ns for the output register. The following text describes the constraints that convey this information.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

The lines that indicate the system clock period is10 ns are the following:

# Global period constraint

NET "clk" TNM_NET = "clk_392b7670";

TIMESPEC "TS_clk_392b7670" = PERIOD "clk_392b7670" 10.0 ns HIGH 50 %;

To build timing constraints, the blocks in the design are partitioned into timing groups.

Two blocks are in the same timing group if and only if they run at the same sample rate. In this design there are three timing groups, corresponding to the three rates. The nature of constraints dictates that no name is needed for the fastest group. The remaining groups are named ce_2_392b7670_group and ce_3_392b7670_group; they correspond to periods 20 ns and 30 ns respectively.

The FIR runs at the system (i.e., fastest) rate and therefore is constrained using the global period constraint shown above. The logic used to generate clocks always runs at the system rate and is also constrained to the system rate.

The ce_2_392b7670_group consists of the blocks that operate at half the system rate, i.e., the input register and the up sampler. Every block in the group is driven by the clock enable net named ce2_sysgen. The constraints that define the group are the following:

# ce_2_392b7670_group and inner group constraint

Net "ce_2_sg_x0*" TNM_NET = "ce_2_392b7670_group";

TIMESPEC "TS_ce_2_392b7670_group_to_ce_2_392b7670_group" = FROM

"ce_2_392b7670_group" TO "ce_2_392b7670_group" 20.0 ns;

Note:

A wildcard character is added to the net name to constrain any additional copies of this net that may be generated when clock enable logic is replicated. The maximum fanout of a clock enable net can be controlled in the synthesis tool.

The ce_3_392b7670_group operates at one third the system rate. It contains the down sampler and the output register, and is defined in a similar manner to the ce2_group.

# ce_3_392b7670_group and inner group constraint

Net "ce_3_sg_x0*" TNM_NET = "ce_3_392b7670_group";

TIMESPEC "TS_ce_3_392b7670_group_to_ce_3_392b7670_group" = FROM

"ce_3_392b7670_group" TO "ce_3_392b7670_group" 30.0 ns;

Group to group constraints establish relative speeds. Here are the constraints that relate the speeds of ce_2_392b7670_group and ce_3_392b7670_group:

# Group-to-group constraints

TIMESPEC "TS_ce_2_392b7670_group_to_ce_3_392b7670_group" = FROM

"ce_2_392b7670_group" TO "ce_3_392b7670_group" 20.0 ns;

TIMESPEC "TS_ce_3_392b7670_group_to_ce_2_392b7670_group" = FROM

"ce_3_392b7670_group" TO "ce_2_392b7670_group" 20.0 ns;

Port timing requirements can be set in the parameter dialog boxes for gateways. These requirements are translated into port constraints such as those shown below. In this example, the 3-bit din input is constrained to operate at its gateway's sample rate

(corresponding to a period of 20 ns). The "FAST" attributes indicate the ports should be implemented using hardware that reduces delay. The reduction comes at a cost of increased noise and power consumption.

# Offset in constraints

NET "din(0)" OFFSET = IN : 20.0 : BEFORE "clk";

NET "din(0)" FAST;

NET "din(1)" OFFSET = IN : 20.0 : BEFORE "clk";

NET "din(1)" FAST;

NET "din(2)" OFFSET = IN : 20.0 : BEFORE "clk";

NET "din(2)" FAST;

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Automatic Code Generation

Selecting Specify IOB Location Constraints for a gateway allows port locations to be specified. The locations must be entered as a cell array of strings in the box labeled IOB

Pad Locations

. Locations are package-specific; in this example a Virtex™-E 2000 in a

FG680 package is used. The location constraints for the din bus are provided in the dialog box as "{'D35', 'B36', 'C35' }". This is translated into constraints in the .xcf (or .ncf) file in the following way:

# Loc constraints

NET "din(2)" LOC = "D35";

NET "din(1)" LOC = "B36";

NET "din(0)" LOC = "C35";

Clock Handling in HDL

This topic describes how System Generator handles hardware clocks in the HDL it generates. Assume the design is named <design>, and <design> is an acceptable HDL identifier. When System Generator compiles the design, it writes a collection of HDL entities or modules, the topmost of which is named <design>, and is stored in a file named <design>.vhd/.v.

Clock and clock enables appear in pairs throughout the HDL. Typical clock names are

clk_1, clk_2, and clk_3, and the names of the companion clock enables are ce_1, ce_2, and

ce_3 respectively. The name tells the rate for the clock/clock enable pair; logic driven by

clk_1 and ce_1 runs at the system (i.e., fastest) rate, while logic driven by (say) clk_2 and

ce_2 runs at half the system rate. Clocks and clock enables are not driven in the entity or module named <design> or any subsidiary entities; instead, they are exposed as top-level input ports

Of course, there must be a way to generate these clocks and clock enables. System

Generator produces a separate clock wrapper (written to a file named

<design>_cw.vhd/.v

) to do this. This wrapper is external to the files described above.

The idea is to make the HDL flexible. In some applications, the files described above are added to a larger design, but the clock wrapper is omitted. In this case, you are responsible for generating clocks and clock enables, but a finer degree of control is obtained. If, on the other hand, the clock wrapper is suitable for the application, then include it. As an additional convenience, System Generator generates a DCM wrapper (written to a file named <design>_dw_vhd/_v) that encloses the clock wrapper. The DCM wrapper deskews the hardware FPGA clock. When incorporating System Generator HDL into a larger design, each of the following is possible:

•

Use the HDL that System Generator writes, but exclude the clock and DCM wrappers;

•

Use the HDL that System Generator writes, and use the clock wrapper, but exclude the DCM wrapper.

•

Use the HDL that System Generator writes, and use both the clock and DCM wrappers.

If you want to use the DCM wrapper in your work, rename the file by replacing the final

_vhd or _v with .vhd or .v.

The names of the clocks and clock enables in System Generator HDL suggest that clocking is completely general, but this is not the case. To illustrate this, assume a design has clocks named clk_1 and clk_2, and companion clock enables named ce_1 and ce_2 respectively.

The reader might expect that working hardware could be produced if the ce_1 and ce_2 signals were tied high, and clk_2 were driven by a clock signal whose rate is half that of

clk_1. For most System Generator designs this does not work. Instead, clk_1 and clk_2 must

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator be driven by the same clock, ce_1 must be tied high, and ce_2 must vary at a rate half that of clk_1 and clk_2.

The clock wrapper consists of two components: one for the design itself, and one clock

driver component that generate clocks and clock enables. The clock driver is contained in a file named <design>_cw.vhd/.v. The logic within the <design>_cw generates the ce_x

signals. The optional ce_clr port would be generated if the design was generated by selecting Provide clock enable clear pin on the System Generator block. The ports that are not clocks or clock enables are passed through to the exterior of the clock wrapper.

Schematically, the clock wrapper looks like the diagram below.

Note:

The clock wrapper exposes a port named ce. The port does nothing except to serve as a companion to the clk port on the wrapper. The reason for having the port is to allow the clock wrapper to be used as a black box in System Generator designs.

Core Caching

System Generator uses cores produced by Xilinx CORE Generator™ (coregen) to implement parts of designs. Generating cores can be expensive, so System Generator caches previously generated ones. Before coregen is called, System Generator looks in the cache, and if the core has already been generated, System Generator reuses it.

By default, the cache is the directory $TEMP/sg_core_cache. And by default, System

Generator caches no more than 2,000 cores. When the limit is reached, System Generator deletes cached cores to make room for new ones.

Note:

Environment variables can be used to change the location of the cache and the cache size limit. The variables are described below.

Environment Variable

SGCORECACHE

SGCORECACHELIMIT

Description

Location to store cached files. Setting this variable to a string of blanks instructs System Generator not to cache cores.

Maximum number of cores to cache.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

HDL Testbench

Ordinarily, System Generator designs are bit and cycle-accurate, so Simulink simulation results exactly match those seen in hardware. There are, however, times when it is useful to compare Simulink simulation results against those obtained from an HDL simulator. In particular, this makes sense when the design contains black boxes. The Create Testbench checkbox in the System Generator block makes this possible.

Suppose the design is named <design>, and a System Generator block is placed at the top of the design. Suppose also that in the block the Compilation field is set to HDL Netlist, and the Create Testbench checkbox is selected. When the Generate button is clicked,

System Generator produces the usual files for the design, and in addition writes the following:

A file named <design>_tb.vhd/.v that contains a testbench HDL entity;

Various .dat files that contain test vectors for use in an HDL testbench simulation.

Scripts vcom.do and vsim.do that can be used in ModelSim to compile and simulate the testbench, comparing Simulink test vectors against those produced in HDL.

System Generator generates the .dat files by saving the values that pass through gateways. In the HDL simulation, input values from the .dat files are stimuli, and output values are expected results. The testbench is simply a wrapper that feeds the stimuli to the

HDL for the design, then compares HDL results against expected ones.

Compiling MATLAB into an FPGA

System Generator provides direct support for MATLAB through the MCode block. The

MCode block applies input values to an M-function for evaluation using Xilinx's fixedpoint data type. The evaluation is done once for each sample period. The block is capable of keeping internal states with the use of persistent state variables. The input ports of the block are determined by the input arguments of the specified M-function and the output ports of the block are determined by the output arguments of the M-function. The block provides a convenient way to build finite state machines, control logic, and computation heavy systems.

In order to construct an MCode block, an M-function must be written. The M-file must be in the directory of the model file that is to use the M-file or in a directory in the MATLAB path.

This tutorial provides ten examples that use the MCode block:

•

Example 1 Simple Selector

shows how to implement a function that returns the maximum value of its inputs;

•

Example 2 Simple Arithmetic Operations

shows how to implement simple arithmetic operations;

•

Example 3 Complex Multiplier with Latency

shows how to build a complex multiplier with latency;

•

Example 4 Shift Operations shows how to implement shift operations;

•

Example 5 Passing Parameters into the MCode Block

shows how to pass parameters into a MCode block;

•

Example 6 Optional Input Ports shows how to implement optional input ports on an

MCode block;

•

Example 7 Finite State Machines shows how to implement a finite state machine;

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

•

Example 8 Parameterizable Accumulator shows how to build a parameterizable

accumulator;

•

Example 9 FIR Example and System Verification

shows how to model FIR blocks and how to do system verification;

•

Example 10 RPN Calculator shows how to model a RPN calculator – a stack machine;

•

Example 11

Example of disp Function

shows how to use disp function to print variable values.

The first two examples are in the mcode_block_tutorial.mdl file of the examples/mcode_block directory in your installation of the System Generator software.

Examples 3 and 4 are in the mcode_block_tutorial2.mdl file. Examples 5 and 6 are in the mcode_block_tutorial3.mdl file. Examples 7 and 8 are in the mcode_block_tutorial4.mdl file. Example 9 is mcode_block_verify_fir.mdl. Example 10 is in mcode_block_rpn_calculator.mdl.

Simple Selector

This example is a simple controller for a data path, which assigns the maximum value of two inputs to the output. The M-function is specified as the following and is saved in an Mfile xlmax.m: function z = xlmax(x, y) if x > y z = x; else z = y; end

The xlmax.m file should be either saved in the same directory of the model file or should be in the MATLAB path. Once the xlmax.m has been saved to the appropriate place, you should drag a MCode block into your model, open the block parameter dialog box, and enter xlmax into the MATLAB Function field. After clicking the OK button, the block has two input ports x and y, and one output port z.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

The following figure shows what the block looks like after the model is compiled. You can see that the block calculates and sets the necessary fixed-point data type to the output port.

Simple Arithmetic Operations

This example shows some simple arithmetic operations and type conversions. The following shows the xlSimpleArith.m file, which specifies the xlSimpleArith Mfunction. function [z1, z2, z3, z4] = xlSimpleArith(a, b)

% xlSimpleArith demonstrates some of the arithmetic operations

% supported by the Xilinx MCode block. The function uses xfix()

% to create Xilinx fixed-point numbers with appropriate

% container types.%

% You must use a xfix() to specify type, number of bits, and

% binary point position to convert floating point values to

% Xilinx fixed-point constants or variables.

% By default, the xfix call uses xlTruncate

% and xlWrap for quantization and overflow modes.

% const1 is Ufix_8_3 const1 = xfix({xlUnsigned, 8, 3}, 1.53);

% const2 is Fix_10_4 const2 = xfix({xlSigned, 10, 4, xlRound, xlWrap}, 5.687); z1 = a + const1; z2 = -b - const2; z3 = z1 - z2;

% convert z3 to Fix_12_8 with saturation for overflow z3 = xfix({xlSigned, 12, 8, xlTruncate, xlSaturate}, z3);

% z4 is true if both inputs are positive z4 = a>const1 & b>-1;

This M-function uses addition and subtraction operators. The MCode block calculates these operations in full precision, which means the output precision is sufficient to carry out the operation without losing information.

One thing worth discussing is the xfix function call. The function requires two arguments: the first for fixed-point data type precision and the second indicating the value.

The precision is specified in a cell array. The first element of the precision cell array is the type value. It can be one of three different types: xlUnsigned, xlSigned, or xlBoolean.

The second element is the number of bits of the fixed-point number. The third is the binary point position. If the element is xlBoolean, there is no need to specify the number of bits and binary point position. The number of bits and binary point position must be specified in pair. The fourth element is the quantization mode and the fifth element is the overflow

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator mode. The quantization mode can be one of xlTruncate, xlRound, or xlRoundBanker.

The overflow mode can be one of xlWrap, xlSaturate, or xlThrowOverflow.

Quanitization mode and overflow mode must be specified as a pair. If the quantizationoverflow mode pair is not specified, the xfix function uses xlTruncate and xlWrap for signed and unsigned numbers. The second argument of the xfix function can be either a double or a Xilinx fixed-point number. If a constant is an integer number, there is no need to use the xfix function. The Mcode block converts it to the appropriate fixed-point number automatically.

After setting the dialog box parameter MATLAB Function to xlSimpleArith, the block shows two input ports a and b, and four output ports z1, z2, z3, and z4.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

M-functions using Xilinx data types and functions can be tested in the MATLAB command window. For example, if you type: [z1, z2, z3, z4] = xlSimpleArith(2, 3) in the MATLAB command window, you'll get the following lines:

UFix(9, 3): 3.500000

Fix(12, 4): -8.687500

Fix(12, 8): 7.996094

Bool: true

Notice that the two integer arguments (2 and 3) are converted to fixed-point numbers automatically. If you have a floating-point number as an argument, an xfix call is required.

Complex Multiplier with Latency

This example shows how to create a complex number multiplier. The following shows the xlcpxmult.m

file which specifies the xlcpxmult function.

function [xr, xi] = xlcpxmult(ar, ai, br, bi) xr = ar * br - ai * bi; xi = ar * bi + ai * br;

The following diagram shows the sub-system:

Two delay blocks are added after the MCode block. By selecting the option Implement

using behavioral HDL

on the Delay blocks, the downstream logic synthesis tool is able to perform the appropriate optimizations to achieve higher performance.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Shift Operations

This example shows how to implement bit-shift operations using the MCode block. Shift operations are accomplished with multiplication and division by powers of two. For example, multiplying by 4 is equivalent to a 2-bit left-shift, and dividing by 8 is equivalent to a 3-bit right-shift. Shift operations are implemented by moving the binary point position and if necessary, expanding the bit width. Consequently, multiplying a Fix_8_4 number by

4 results in a Fix_8_2 number, and multiplying a Fix_8_4 number by 64 results in a

Fix_10_0 number.

The following shows the xlsimpleshift.m file which specifies one left-shift and one right-shift: function [lsh3, rsh2] = xlsimpleshift(din)

% [lsh3, rsh2] = xlsimpleshift(din) does a left shift

% 3 bits and a right shift 2 bits.

% The shift operation is accomplished by

% multiplication and division of power

% of two constant.

lsh3 = din * 8; rsh2 = din / 4;

The following diagram shows the sub-system after compilation:

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

Passing Parameters into the MCode Block

This example shows how to pass parameters into the MCode block. An input argument to an M-function can be interpreted either as an input port on the MCode block, or as a parameter internal to the block.

The following M-code defines an M-function xl_sconvert is contained in file xl_sconvert.m

: function dout = xl_sconvert(din, nbits, binpt) proto = {xlSigned, nbits, binpt}; dout = xfix(proto, din);

The following diagram shows a subsystem containing two MCode blocks that use Mfunction xl_sconvert. The arguments nbits and binpt of the M-function are specified differently for each block by passing different parameters to the MCode blocks. The parameters passed to the MCode block labeled signed convert 1 cause it to convert the input data from type Fix_16_8 to Fix_10_5 at its output. The parameters passed to the MCode block labeled signed convert2 causes it to convert the input data from type

Fix_16_8

to Fix_8_4 at its output.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

To pass parameters to each MCode block in the diagram above, you can click the Edit

Interface

button on the block GUI then set the values for the M-function arguments. The mask for MCode block signed convert 1 is shown below:

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

The above interface window sets the M-function argument nbits to be 10 and binpt to be 5. The mask for the MCode block signed convert 2 is shown below:

The above interface window sets the M-function argument nbits to be 8 and binpt to be

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Optional Input Ports

This example shows how to use the parameter passing mechanism of MCode blocks to specify whether or not to use optional input ports on MCode blocks.

The following M-code, which defines M-function xl_m_addsub is contained in file xl_m_addsub.m

: function s = xl_m_addsub(a, b, sub) if sub s = a - b; else s = a + b; end

The following diagram shows a subsystem containing two MCode blocks that use Mfunction xl_m_addsub.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

The Block Interface Editor of the MCode block labeled add is shown in below.

As a result, the add block features two input ports a and b; it performs full precision addition. Input parameter sub of the MCode block labeled addsub is not bound with any value. Consequently, the addsub block features three input ports: a, b, and sub; it performs full precision addition or subtraction based on the value of input port sub.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Finite State Machines

This example shows how to create a finite state machine using the MCode block with internal state variables. The state machine illustrated below detects the pattern 1011 in an input stream of bits.

The M-function that is used by the MCode block contains a transition function, which computes the next state based on the current state and the current input. Unlike example 3 though, the M-function in this example defines persistent state variables to store the state of the finite state machine in the MCode block. The following M-code, which defines function detect1011_w_state is contained in file detect1011_w_state.m: function matched = detect1011_w_state(din)

% This is the detect1011 function with states for detecting a

% pattern of 1011.

seen_none = 0; % initial state, if input is 1, switch to seen_1 seen_1 = 1; % first 1 has been seen, if input is 0, switch

% seen_10 seen_10 = 2; % 10 has been detected, if input is 1, switch to

% seen_1011 seen_101 = 3; % now 101 is detected, is input is 1, 1011 is

% detected and the FSM switches to seen_1

% the state is a 2-bit register persistent state, state = xl_state(seen_none, {xlUnsigned, 2, 0});

% the default value of matched is false matched = false; switch state case seen_none if din==1 state = seen_1; else state = seen_none; end case seen_1 % seen first 1 if din==1 www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

state = seen_1; else state = seen_10; end case seen_10 % seen 10 if din==1 state = seen_101; else

% no part of sequence seen, go to seen_none

state = seen_none; end case seen_101 if din==1 state = seen_1; matched = true; else state = seen_10; matched = false; end end

The following diagram shows a state machine subsystem containing a MCode block after compilation; the MCode block uses M-function detect1101_w_state.

Parameterizable Accumulator

This example shows how to use the MCode block to build an accumulator using persistent state variables and parameters to provide implementation flexibility. The following Mcode, which defines function xl_accum is contained in file xl_accum.m: function q = xl_accum(b, rst, load, en, nbits, ov, op, feed_back_down_scale)

% q = xl_accum(b, rst, nbits, ov, op, feed_back_down_scale) is

% equivalent to our Accumulator block.

binpt = xl_binpt(b); init = 0; precision = {xlSigned, nbits, binpt, xlTruncate, ov}; persistent s, s = xl_state(init, precision); q = s; if rst if load

% reset from the input port s = b; else

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

% reset from zero s = init; end else if ~en else

% if enabled, update the state if op==0 s = s/feed_back_down_scale + b; else s = s/feed_back_down_scale - b; end end end

The following diagram shows a subsystem containing the accumulator MCode block using

M-function xl_accum. The MCode block is labeled MCode Accumulator. The subsystem also contains the Xilinx Accumulator block, labeled Accumulator, for comparison purposes. The MCode block provides the same functionality as the Xilinx

Accumulator block; however, its mask interface differs in that parameters of the MCode block are specified with a cell array in the Function Parameter Bindings parameter.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

Optional inputs rst and load of block Accum_MCode1 are disabled in the cell array of the

Function Parameter Bindings parameter. The block mask for block MCode Accumulator is shown below:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

The example contains two additional accumulator subsystems with MCode blocks using the same M-function, but different parameter settings to accomplish different accumulator implementations.

FIR Example and System Verification

This example shows how to use the MCode block to model FIRs. It also shows how to do system verification with the MCode block.

The model contains two FIR blocks. Both are modeled with the MCode block and both are synthesizable. The following are the two functions that model those two blocks.

function y = simple_fir(x, lat, coefs, len, c_nbits, c_binpt, o_nbits, o_binpt) coef_prec = {xlSigned, c_nbits, c_binpt, xlRound, xlWrap}; out_prec = {xlSigned, o_nbits, o_binpt}; coefs_xfix = xfix(coef_prec, coefs); persistent coef_vec, coef_vec = xl_state(coefs_xfix, coef_prec); persistent x_line, x_line = xl_state(zeros(1, len-1), x); persistent p, p = xl_state(zeros(1, lat), out_prec, lat); sum = x * coef_vec(0); for idx = 1:len-1 sum = sum + x_line(idx-1) * coef_vec(idx); sum = xfix(out_prec, sum); end y = p.back; p.push_front_pop_back(sum); x_line.push_front_pop_back(x); function y = fir_transpose(x, lat, coefs, len, c_nbits, c_binpt, o_nbits, o_binpt) coef_prec = {xlSigned, c_nbits, c_binpt, xlRound, xlWrap}; out_prec = {xlSigned, o_nbits, o_binpt}; coefs_xfix = xfix(coef_prec, coefs); persistent coef_vec, coef_vec = xl_state(coefs_xfix, coef_prec); www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

persistent reg_line, reg_line = xl_state(zeros(1, len), out_prec); if lat <= 0 error('latency must be at least 1'); end lat = lat - 1; persistent dly, if lat <= 0 y = reg_line.back; else dly = xl_state(zeros(1, lat), out_prec, lat); y = dly.back; dly.push_front_pop_back(reg_line.back); end for idx = len-1:-1:1 reg_line(idx) = reg_line(idx - 1) + coef_vec(len - idx - 1) * x; end reg_line(0) = coef_vec(len - 1) * x;

The parameters are configured as following:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

In order to verify that the functionality of two blocks are equal, we also use another MCode block to compare the outputs of two blocks. If the two outputs are not equal at any given time, the error checking block will report the error. The following function does the error checking: function eq = error_ne(a, b, report, mod) persistent cnt, cnt = xl_state(0, {xlUnsigned, 16, 0}); switch mod case 1 eq = a==b; case 2 eq = isnan(a) || isnan(b) || a == b; case 3 eq = ~isnan(a) && ~isnan(b) && a == b; otherwise eq = false; error(['wrong value of mode ', num2str(mod)]); end if report if ~eq error(['two inputs are not equal at time ', num2str(cnt)]); end end cnt = cnt + 1;

The block is configured as following:

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

RPN Calculator

This example shows how to use the MCode block to model a RPN calculator which is a stack machine. The block is synthesizable.

The following function models the RPN calculator.

function [q, active] = rpn_calc(d, rst, en) d_nbits = xl_nbits(d);

% the first bit indicates whether it's a data or operator is_oper = xl_slice(d, d_nbits-1, d_nbits-1)==1; din = xl_force(xl_slice(d, d_nbits-2, 0), xlSigned, 0);

% the lower 3 bits are operator op = xl_slice(d, 2, 0);

% acc the the A register persistent acc, acc = xl_state(0, din);

% the stack is implemented with a RAM and

% an up-down counter persistent mem, mem = xl_state(zeros(1, 64), din); persistent acc_active, acc_active = xl_state(false, {xlBoolean}); persistent stack_active, stack_active = xl_state(false, ...

{xlBoolean}); stack_pt_prec = {xlUnsigned, 5, 0}; persistent stack_pt, stack_pt = xl_state(0, {xlUnsigned, 5, 0});

% when en is true, it's action

OP_ADD = 2;

OP_SUB = 3;

OP_MULT = 4;

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

OP_NEG = 5;

OP_DROP = 6; q = acc; active = acc_active; if rst acc = 0; acc_active = false; stack_pt = 0; elseif en if ~is_oper

% enter data, push if acc_active stack_pt = xfix(stack_pt_prec, stack_pt + 1); mem(stack_pt) = acc; stack_active = true; else acc_active = true; end acc = din; else if op == OP_NEG

% unary op, no stack op acc = -acc; elseif stack_active b = mem(stack_pt); switch double(op) case OP_ADD acc = acc + b; case OP_SUB acc = b - acc ; case OP_MULT acc = acc * b; case OP_DROP acc = b; end stack_pt = stack_pt - 1; elseif acc_active acc_active = false; acc = 0; end end end stack_active = stack_pt ~= 0;

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling MATLAB into an FPGA

Example of disp Function

The following MCode function shows how to use the disp function to print variable values.

function x = testdisp(a, b) persistent dly, dly = xl_state(zeros(1, 8), a); persistent rom, rom = xl_state([3, 2, 1, 0], a); disp('Hello World!'); disp(['num2str(dly) is ', num2str(dly)]); disp('disp(dly) is '); disp(dly); disp('disp(rom) is '); disp(rom); a2 = dly.back; dly.push_front_pop_back(a); x = a + b; disp(['a = ', num2str(a), ', ', ...

'b = ', num2str(b), ', ', ...

'x = ', num2str(x)]); disp(num2str(true)); disp('disp(10) is'); disp(10); disp('disp(-10) is'); disp(-10); disp('disp(a) is '); disp(a); disp('disp(a == b)'); disp(a==b);

The Enable print with disp option must be checked.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Here are the lines that are displayed on the MATLAB console for the first simulation step.

mcode_block_disp/MCode (Simulink time: 0.000000, FPGA clock: 0)

Hello World!

num2str(dly) is [0.000000, 0.000000, 0.000000, 0.000000, 0.000000,

0.000000, 0.000000, 0.000000] disp(dly) is type: Fix_11_7, maxlen: 8, length: 8,

0: binary 0000.0000000, double 0.000000,

1: binary 0000.0000000, double 0.000000,

2: binary 0000.0000000, double 0.000000,

3: binary 0000.0000000, double 0.000000,

4: binary 0000.0000000, double 0.000000,

5: binary 0000.0000000, double 0.000000,

6: binary 0000.0000000, double 0.000000,

7: binary 0000.0000000, double 0.000000, disp(rom) is type: Fix_11_7, maxlen: 4, length: 4,

0: binary 0011.0000000, double 3.0,

1: binary 0010.0000000, double 2.0,

2: binary 0001.0000000, double 1.0,

3: binary 0000.0000000, double 0.0, a = 0.000000, b = 0.000000, x = 0.000000

1 disp(10) is type: UFix_4_0, binary: 1010, double: 10.0

disp(-10) is type: Fix_5_0, binary: 10110, double: -10.0

disp(a) is type: Fix_11_7, binary: 0000.0000000, double: 0.000000

disp(a == b) type: Bool, binary: 1, double: 1

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Importing a System Generator Design into a Bigger System

A System Generator design is often a sub-design that is incorporated into a larger HDL design. This topic shows how to embed two System Generator designs into a larger design and how VHDL created by System Generator can be incorporated into the simulation model of the overall system.

Starting with Release 10.1, System Generator introduces a new integration flow between

System Generator (Sysgen) and Project Navigator (ProjNav). This first phase of integration concentrates on the following areas:

•

Allows you to add a System Generator design as a sub-level to a larger design

•

Consolidates and associates System Generator constraints to the top-level design

•

Enables you to perform certain design iterations between Project Navigator and the

System Generator design

HDL Netlist Compilation

Selecting the HDL Netlist compilation target from the System Generator token instructs

System Generator to generate HDL along with other related files such as NGC files and

EDIF files that implement the design. In addition, System Generator produces auxiliary files that simplify downstream processing such as bringing the design into Project

Navigator, simulating the design using an HDL simulator, and performing logic synthesis using various logic synthesis tools. See the topic

System Generator Compilation Types for

more details.

Starting with Release 10.1, the System Generator project information is encapsulated in the file <design_name>_cw.sgp or <design_name>_mcw.sgp depending on which clocking option is selected. This topic shows how multiple System Generator designs can be included as sub-modules in a larger design.

Integration Design Rules

When a System Generator model is to be included into a larger design, the following two design rules must be followed.

Rule 1

: No Gateway or System Generator token should specify an IOB/CLK location constraint. Otherwise, the NGDBuild tool will issue the following warning:

WARNING:NgdBuild:483 - Attribute "LOC" on "clk" is on the wrong type of object. Please see the Constraints Guide for more information on this attribute.

Rule 2

: If there are any I/O ports from the System Generator design that are required to be bubbled up to the top-level design, appropriate buffers should be instantiated in the toplevel HDL code.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

New Integration Flow between System Generator & Project Navigator

The illustration below shows the entire flow of how multiple System Generator designs can be integrated into Project Navigator as lower-level designs. System Generator generates a project file with an extension .sgp that you can add as a System Generator source type in Project Navigator. This file contains all necessary information about the

System Generator design, including file locations and constraint files. Prior to the integration with Project Navigator in Release 10.1, you had to manually consolidate and associate UCF constraints into the top-level design. It is now done automatically during the implementation in Project Navigator as shown in the following figure.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Importing a System Generator Design into a Bigger System

A Step-by-Step Example

In this example, two HDL netlists from System Generator are integrated into a larger

VHDL design. Design #1 is named SPRAM and design #2 is named MAC_FIR. The toplevel VHDL entity combines the two data ports and a control signal from the SPRAM design to create a bidirectional bus. The top-level VHDL also instantiates the MAC_FIR design and supplies a separate clock signal named clk2. A block diagram of this design is shown below.

The files used in this example are located in the System Generator tree at pathname

<sysgen_tree>/examples/projnav/mult_diff_designs

. The following files are provided:

• spram.mdl

- System Generator design #1

• mac_fir.mdl

- System Generator design #2

Files within the sub-directory named top_level:

• top_level.ise

– ProjNav project for compiling top_level design

• top_level.vhd

– Top-level VHDL file

• top_level_testbench.do

– Custom ModelSim .do file

• top_level_testbench.vhd

– Top-level VHDL testbench file

• wave.do

– ModelSim .do file called by top_level_testbench.do to display waveforms

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Generating the HDL Files for the System Generator Designs

The steps used to create the HDL files are as follows:

Open the first design, spram.mdl, in MATLAB. This is a multirate design due to the down sampling block placed after the output of the Single Port RAM.

Double click on the System Generator block; select the HDL Netlist target and press the Generate button. By pressing the Generate button, the HDL file for this design is created in the directory

<sysgen_tree>/examples/projnav/mult_diff_designs/hdl_netlist1

Repeat steps 1 and 2 for the mac_fir.mdl model. The HDL file for this design is created in the directory

<sysgen_tree>/examples/projnav/mult_diff_designs/hdl_netlist2

Note:

You are now finished generating HDL Netlists from System Generator

Switching to Different HDL Libraries

When integrating two or more System Generator designs into a bigger design, you need to rename HDL libraries to prevent name clashes and other undesired behaviors during simulation. System Generator provides a utility that switches library names for all related files in your System Generator design. In addition, it also makes a backup copy in a folder just in case you want to revert back to the original library name. The following is the syntax for this utility:

Syntax

: xlSwitchLibrary(<target_dir_pathname>, <from_lib_name>, <to_lib_name>)

<target_dir_pathname>: location of the design

<from_lib_name>: Original HDL library name

<to_lib_name>: New HDL library name

From the MATLAB Console, enter the following command: xlSwitchLibrary('hdl_netlist1','work','design1_lib')

Next, from the MATLAB Console, enter the following command: xlSwitchLibrary('hdl_netlist2','work','design2_lib')

The transcript should look similar to the following:

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Importing a System Generator Design into a Bigger System

Adding System Generator Source into the Top-Level Design

The next two steps are used to synthesize the top_level design:

Launch ISE and reload the pre-generated top-level design ISE project at

~top_level/top_level.ise.

Note:

At this point, your Project Navigator should look like the figure below. Both spram_cw and mac_fir_cw instances are instantiated at the top_level design. But since they are not located on the same directory as the top-level design, Project Navigator puts a question mark next to each one of them to indicate that it can not find these two instances / modules.

Add the System Generator source: under the Sources tab, right-click on

u_spram_cw -> Add Source

…at

<sysgen_tree>/examples/projnav/mult_diff_designs/hdl_netlist1/s pram_cw.sgp

Repeat item 2 with u_mac_fir at

<sysgen_tree>/examples/projnav/mult_diff_designs/hdl_netlist2/m ac_fir_cw.sgp

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

As shown below, make sure the file top_level is selected, then implement the design by double clicking on Processes tab > Implement Design. Once the implementation is finished, Project Navigator should look like the figure below.

Examine the timing constraints in the PAR report file: Processes tab > Implement

Design > Place & Route > Place & Route Report

Note that in the PAR report the multirate constraints were met:

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Importing a System Generator Design into a Bigger System

Constraints for each System Generator design were created and translated to a UCF (User

Constraint File). These UCF constraint files were then consolidated and associated during

ISE implementation (NGDBUILD). They are briefly described as follows:

A system sample period of 100 ns was set in the System Generator block for both designs

(1 & 2)

•

TS_clk_f488215c2 constraints are from the SRAM design (1)

•

The TS_clk_c4b7e2441 constraints are from the FIR design (2)

•

The ce16_c4b7e244_group_to_ce16_cb47e244_group1 constraint is for all the synchronous elements after the down sampler and it is set to sixteen, the system sample period (3)

•

The down sampling block in the SRAM design performs a down sample by 2. The ce2_f488215c_group_to_ce2_f488215c_group2 constraint is for all the synchronous elements after the down sampler and is set to twice the system sample period (4)

With the new integration between System Generator and Project Navigator, these constraints are automatically associated and consolidated by Project Navigator up to the top-level design. This flow is only available starting with Release 10.1.

Simulating the Entire Design

To perform a behavioral simulation of the top_level design, do the following:

System Generator creates VHDL files and invokes the selected logic synthesis tool to generate the HDL Netlist. These VHDL files are used when simulating the top-level design. The VHDL files generated for a design are named <design>_cw.vhd, and

<design>.vhd

. Open the custom ModelSim do file named “top_level_testbench.do” to see how the VHDL files for both designs are referenced.

Memory initialization (.mif) and coefficient (.coe) files that are used during simulation must be placed in the same directory as the top-level VHDL file. For this example, the mif files are copied from both hdl_netlist1 and hdl_netlist2 subfolders by the following statement in the ModelSim do file (top_level_testbench.do): foreach i [glob ../hdl_netlist1/*.mif] { file copy -force $i .

In a case where there are also coefficient files, you can add a similar statement to the do file to copy the files up to the top-level VHDL file.

In ProjNav > Sources > change from Implementation to Behavioral Simulation option from the pull-down menu. Select the top_level_testbench-

structural(top_level.vhd)

source file. This file is imported into the project as a testbench file, thus allowing you to simulate the design using the Simulator.

In the Processes window, right click on Simulate Behavioral Model > Properties…

You should see a Simulation Properties dialog box as shown below. Note that A

Custom Do File has been specified as shown below (1).

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Importing a System Generator Design into a Bigger System

The previous screen shot shows the ModelSim commands used to compile the VHDL code generated by System Generator. To simulate the top_level design, double left click on the

Simulate Behavioral Model

process. The ModelSim .do file compiles the VHDL code and runs the simulation for 10000 ns. The resulting waveform is shown below.

Summary

This topic has shown you how to import a System Generator Design into a larger system.

There are a few important things to keep in mind during each phase of the process.

While creating a System Generator design:

•

IOB constraints should not be specified on Gateways in the System Generator model; neither should the System Generator block specify a clock pin location.

•

Use the HDL Netlist compilation target in the System Generator block. The HDL

Netlist file that System Generator produces contains both the RTL, EDIF and constraint information for your design.

To instantiate the System Generator design in the top_level HDL:

•

Use a black box and assign the appropriate black box attribute The ce port is not connected to registers within your design. It is provide so that the VHDL file can be imported as a Black Box within System Generator.

For top-level simulation:

•

Create a custom ModelSim .do file in order to compile the VHDL files created by

System Generator. Modify the Project Navigator settings to use this custom .do file

New capabilities:

•

Add System Generator Source type project file (.sgp) into Project Navigator as a submodule design

•

Consolidate and associate System Generator constraints into the top-level design

•

Launch MATLAB and System Generator MDL directly from Project Navigator to perform certain design iterations

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Configurable Subsystems and System Generator

A configurable subsystem is a kind of block that is made available as a standard part of

Simulink. In effect, a configurable subsystem is a block for which you can specify several underlying blocks. Each underlying block is a possible implementation, and you are free to choose which implementation to use. In System Generator you might, for example, specify a general-purpose FIR filter as a configurable subsystem whose underlying blocks are specific FIR filters. Some of the underlying filters might be fast but require much hardware, while others are slow but require less hardware. Switching the choice of the underlying filter allows you to perform experiments that trade hardware cost against speed.

Defining a Configurable Subsystem

A configurable subsystem is defined by creating a Simulink library. The underlying blocks that implement a configurable subsystem are organized in this library. To create such a library, do the following:

•

Make a new empty library.

•

Add the underlying blocks to the library.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Configurable Subsystems and System Generator

•

Drag a template block into the library. (Templates can be found in the Simulink library browser under Simulink/Ports & Subsystems/Configurable Subsystem.)

•

Rename the template block if desired.

•

Save the library.

•

Double click to open the template for the library.

•

In the template GUI, turn on each checkbox corresponding to a block that should be an implementation.

•

Press OK, and then save the library again.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Using a Configurable Subsystem

To use a configurable subsystem in a design, do the following:

•

As described above, create the library that defines the configurable subsystem.

•

Open the library.

•

Drag a copy of the template block from the library to the appropriate part of the design.

•

The copy becomes an instance of the configurable subsystem.

•

Right-click on the instance, and under Block choice select the block that should be used as the underlying implementation for the instance.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Configurable Subsystems and System Generator

Deleting a Block from a Configurable Subsystem

To delete an underlying block from a configurable subsystem, do the following:

•

Open and unlock the library for the subsystem.

•

Double click on the template, and turn off the checkbox associated to the block to be deleted.

•

Press OK, and then delete the block.

•

Save the library.

•

Compile the design by typing Ctrl-d.

•

If necessary, update the choice for each instance of the configurable subsystem.

Adding a Block to a Configurable Subsystem

To add an underlying block to a configurable subsystem, do the following:

•

Open and unlock the library for the subsystem.

•

Drag a block into the library.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

•

Double click on the template, and turn on the checkbox next to the added block.

•

Press OK, and then save the library.

•

Compile the design by typing Ctrl-d.

•

If necessary, update the choice for each instance of the configurable subsystem.

Generating Hardware from Configurable Subsystems

In System Generator, blocks both participate in simulations and produce hardware.

Sometimes, for a configurable subsystem, it is worthwhile to use one underlying block for simulation, but use another for hardware generation. For example, it might make sense to use ordinary System Generator blocks to produce simulation results, but use a black box to supply the corresponding HDL. The System Generator configurable subsystem manager block makes this possible; the ordinary block choice for the configurable subsystem is used when simulating, and the block specified in the manager is used for hardware generation.

To use a configurable subsystem manager, do the following:

•

Open and unlock the library for the configurable subsystem.

•

Select one of the blocks in the library, and double click to open it. (Aside from the template any block will do, provided the block is itself a subsystem. If there is no such subsystem in the library, it is not possible to use a configurable subsystem manager.) www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Configurable Subsystems and System Generator

•

Drag a manager block into the subsystem opened above. (The manager block can be found in Xilinx Blockset/Tools/Configurable Subsystem Manager).

•

Double click to open the GUI on the manager, then select the block that should be used for hardware generation in the configurable subsystem.

•

Press OK, then save the subsystem, and the library.

The Mathworks description of configurable subsystems can be found the following address: http://www.mathworks.com/access/helpdesk/help/toolbox/simulink/slref/configura blesubsystem.shtml

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Notes for Higher Performance FPGA Design

When you use the following design practices, it helps System Generator produce efficient and high performance hardware realizations.

Review the Hardware Notes Included in Block Dialog Boxes

Pay close attention to the Hardware Notes included in the block dialog boxes. Many blocks in the Xilinx Blockset library have notes that explain how to achieve the most hardware efficient implementation. For example, the notes point out that the Scale block costs nothing in hardware. By contrast, the Shift block (which is sometimes used for the same purpose) can use hardware.

Register the Inputs and Outputs of Your Design

Register inputs and outputs of your design. This can be done by placing a Delay block having latency 1 or a Register block after the Gateway In and before Gateway Out blocks.

Selecting any of the Register block features adds hardware.

Double registering the I/Os may also be beneficial. This can be performed by instantiating two separate Register blocks, or by instantiating two Delay blocks, each having latency 1.

This allows one of the registers to be packed into the IOB and the other to be placed next to the logic in the FPGA fabric. A Delay block with latency 2 does not give the same result since this block is implemented using an SRL16 and cannot be packed into an IOB.

Insert Pipeline Registers

Insert pipeline registers wherever possible. Deep pipelines are efficiently implemented with the Delay blocks since the SRL16 primitive is used. If an initial value is needed on a register, the Register block should be used.

Use Saturation Arithmetic and Rounding Only When Necessary

Saturation arithmetic and rounding have area and performance costs. Use only if necessary.

Use the System Generator Timing Analysis Tool

Use System Generator Timing Analysis Tool to Meet Timing Requirements. System

Generator provides a Timing Analysis tool that can help resolve timing related issues. The timing analysis tool shows you the slowest paths and those paths which are failing to meet the timing requirements. For more information refer to topic

Timing Analysis

Compilation

Set the Data Rate Option on All Gateway Blocks

Select the IOB timing constraint option Data Rate on all Gateway In and Gateway Out blocks. When Data Rate is selected, the IOBs are constrained at the data rate at which the

IOBs operate. The rate is determined by the Simulink system period(sec) field in the

System Generator block and the sample rate of the Gateway relative to the other sample periods in the design.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Processing a System Generator Design with FPGA Physical Design Tools

Reduce the Clock Enable (CE) Fanout

An algorithm in the ISE™ Mapper uses register duplication and placement based on recursive partitioning of loads on high fanout nets. This means improved FMAX on

System Generator designs with large CE fanout.

Although this feature is enabled in System Generator by default, the fanout reduction occurs downstream during the ISE mapping operation and the following MAP options must be turned on:

•

Perform Timing-Driven Packing and Placement

: on

•

Map Effort Level

: High

•

Register Duplication

: on

If you are using the ISE Project Navigator flow, these MAP options are also on by default.

However, if you are using a System Generator flow like Bitstream, you must turn on these

MAP options by modifying the bitstream .opt file or by providing you own .opt file. See the topic

XFLOW Option Files for more information.

Processing a System Generator Design with FPGA Physical

Design Tools

HDL Simulation

System Generator creates custom .do files for use with your generated project and a

ModelSim simulator. To use these files, you must have ModelSim. You may run your simulations from the standalone ModelSim tool, or you may associate it with the Xilinx

ISE™ Project Navigator, and run your simulations from within Project Navigator as part of the full software implementation flow.

Compiling Your IP

Before you can simulate your design, you must compile your IP (cores) libraries with

ModelSim.

ModelSim SE

There are multiple ways to compile your IP libraries. Complete instructions for running compxlib can be found in the chapter titled COMPXLIB in the Xilinx Development System

Reference Guide.

From the Windows command line you can compile the necessary HDL libraries using the compxlib program. For example, the following command can be used to compile all the

HDL libraries with ModelSim SE:

compxlib -s mti_se -f all -l all

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Simulation using ModelSim within Project Navigator

Before you can launch ModelSim from Project Navigator you must specify the location of your installed version of ModelSim. To do so, open Project Navigator and choose the main menu Edit > Preferences. This brings up a dialog box. Choose the ISE General >

Integrated Tools

category in the dialog box. Enter the full path to the version of ModelSim on your PC in the Model Tech Simulator edit box. You must include the name of the executable file in this field.

The Project Navigator project is already set up to run simulations at four different stages of implementation. System Generator creates four different ModelSim .do files when the

Create Testbench option is selected on the System Generator block. The ModelSim do files created by System Generator are:

• pn_behavioral.do

- for a behavioral (HDL) simulation on the HDL files in the project, before any synthesis or implementation.

• pn_posttranslate.do

- this file runs a simulation on the output of the Xilinx translation (ngdbuild) step, the first step of implementation.

• pn_postmap.do

- to run a simulation after your design has been mapped. This file also includes a back-annotated simulation on the post-mapped design.

• pn_postpar.do

- to run a simulation after your design has been placed and routed.

This file also includes a back-annotated simulation step.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Processing a System Generator Design with FPGA Physical Design Tools

In the Project Navigator Sources window, use the pull-down menu labeled Sources for to select Behavioral Simulation, Post-Translate Simulation, Post-Map Simulation, or Post-

Route Simulation (corresponding to pn_behavioral.do, pn_posttranslate.do, pn_postmap.do, and pn_postpar.do respectively).

If you select the <your design>_tb.vhd/.v file in the Project Navigator Sources window, the ModelSim Simulator will become available in the Process window. Expand the ModelSim Simulator process by clicking on the plus button to the left of it. A simulation process associated with the ModelSim Simulator will appear (in the image below the process is labeled Simulate Behavioral Model).

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

The Process Properties dialog box shows that the System Generator.do file is already associated as a custom file for this process.

Now if you double-click on the simulation process, the ModelSim console opens, and the associated custom do file is used to compile and run your System Generator testbench. The testbench uses the same input stimuli that was generated in Simulink, and compares the

HDL simulation results with the Simulink results. Provided that your design was error free, ModelSim reports that the simulation finished without errors.

Generating an FPGA Bitstream

Xilinx ISE Project Navigator

During code generation, the System Generator creates several project files for use in Xilinx and partner software tools. One of these project files is for the Xilinx ISE™ Project

Navigator tool. By opening this project file, you can import your System Generator design into the Project Navigator, and from there, you can synthesize, simulate, and implement the design. This file is called <design_name>_cw.ise and it is created in the target directory specified in the System Generator block.

Note:

my_project_cw.ise

is used in the following discussion.

Opening a System Generator Project

You may double-click on your .ise file in Windows Explorer. The Project Navigator file association with .ise causes Project Navigator to launch, opening your my_project_cw.ise

System Generator design project. You may also open the Project

Navigator tool directly, then choose File > Open Project from the top-level pull down menu. Browse to the location of your System Generator my_project_cw.ise and open it.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Processing a System Generator Design with FPGA Physical Design Tools

Customizing your System Generator Project

When first opening your System Generator project, you will see that it has been set up with the synthesis tool, device, package, and speed grade that you specified in the System

Generator block. To change these settings, open the Project Navigator properties dialog, right-click on the device and default package at the top of the sources window, and select

Properties

This brings up the Project Properties window. From this window, you can change your part, package, speed, and synthesis compiler. Note that if you change the device family, the

Xilinx IP cores that were produced by System Generator must be regenerated. In such a case, it is better if you return to the System Generator and re-generate your project.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Implementing Your Design

You have many options within Project Navigator for working on your project. You can open any of the Xilinx software tools such as the Floorplanner, Constraints Editor, report viewers, etc. To implement your design, you can simply instruct Project Navigator to run your design all the way from synthesis to bitstream. In the Sources window, select the top- level HDL module in your design. In our example the top-level HDL module is named my_project_cw - structural

. The Processes window shows the processes that can be run on the top-level HDL module.

In the Processes window, if you right-click on Generate Programming File and select Run, you are instructing Project Navigator to run through whatever processes are necessary to produce a programming file (FPGA bitstream) from the selected HDL source. In the messages console window, you see that Project Navigator is synthesizing, translating, mapping, routing, and generating a bitstream for your design.

Now that you have generated a bitstream for your design, you have access to all the files that were produced on the way to bitstream creation.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Resetting Auto-Generated Clock Enable Logic

System Generator provides a bit and cycle accurate modeling of FPGA hardware in the

Simulink environment. Several clocking options are available including the default option

Clock Enables

. With this option, System Generator uses a single clock accompanied by clock enables (ce) to keep various sample domains in sync. Multirate clocking is described in detail in the topic

Compilation Results

. System Generator models are often included as part of a bigger system design which need dynamic control for specifying the beginning of data path sampling. To allow this control within a bigger framework System Generator block provides an optional ce_clr port in the top-level HDL clock wrapper for resetting the clock enable generation logic. The figure below shows the reset of the CE4 signal generation logic after ce_clr signal is de-asserted.

The effect of ce_clr signal cannot be simulated using the original System Generator design. To model this behavior within Simulink follow the steps below:

Select Provide clock enable clear pin and

NGC Netlist Compilation option on the

System Generator block.

Press the Generate button on the System Generator block.

Run the following command from the MATLAB console to produce the post translate

VHDL netlist. Use “-ofmt verilog” with netgen for generating Verilog netlist:

>> !netgen -ofmt vhdl ./<target_directory>/<design_name>_cw.ngc

Bring in the post translate VHDL/Verilog file as a Black Box within Simulink and use

HDL co-simulation to model the effect of asserting ce_clr signal on your design.

ce_clr and Rate Changing Blocks

The ce_clr signal changes the sampling phase of all the multi-sample data signals. This behavior has the potential of changing the functionality of all rate changing blocks which rely heavily on the ce signal to have a periodic occurrence. The various rate changing blocks and their behavior with regards to the de-assertion of the ce_clr signal is explained in the table below. These blocks were characterized by importing and simulating the post translate HDL model as a black box.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Block Name

Down Sampler with Last Value of frame

Synchronized to

Synchronized ce after ce_clr deasserted to ce_clr

( 1 sample cycle delay)

Behavior after ce_clr is de-asserted and the next ce pulse

Yes N/A The last sampled value is held till the new ce signal arrives.

No No Re-synchronization does not occur after de-assertion of the ce_clr signal.

Down Sampler with First Value of frame

Up Sampler with copy samples

Up Sampler with zeros inserted

Time Division

Multiplexer

Yes

N/A

Yes

In hardware, this block is implemented as a wire.

Time Division

Demultiplexer

Parallel to Serial

Serial to Parallel

Yes

The last value (zero or sample) is held till the next destination ce signal arrives.

The TDM block samples through all the remaining input channels and then sets the output to 0 till the next ce arrives. The new ce signal resynchronizes the output to the new frame definition.

The TDD block holds the output channels to the same value till the next ce signal arrives. The new ce signal re-synchronizes the output to the new frame definition.

The p2s block samples through all the remaining data words and then holds the output to the last sampled word until the next ce arrives. The new ce signal starts the conversion of the parallel data stream to a serial one.

The s2p block holds the output when the ce_clr is asserted. When deasserted, the input is sampled on the last value of the input sample frame, and the output occurs on the first ce pulse corresponding to the output rate.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Resetting Auto-Generated Clock Enable Logic

Block Name

Addressable

Shift Register

(ASR)

Polyphase FIR

Synchronized to

Synchronized ce after ce_clr deasserted to ce_clr

( 1 sample cycle delay)

Behavior after ce_clr is de-asserted and the next ce pulse

Yes

The ASR block will hold the values in the shift register when ce_clr is asserted. When de-asserted, the stored values will be shifted out, and new data will be put into the shift register.

Interpolating or Decimating FIR does not work with the ce_clr signal unless the optional reset port is used to reset the FIR after the ce_clr is de-asserted.

ce_clr Usage Recommendations

•

Based on the above analysis, the ce_clr signal can be used if the following recommendations are adhered to:

•

Replace down sampler blocks with first value of frame behavior with an equivalent circuit using down sampler block with last value of frame selected.

•

Design for N clock cycles of invalid data after ce_clr is de-asserted, where N is the slowest ce associated with the block.

•

Design the model to always use down sampler with last value of frame and up sampler with copy samples.

•

If N cycle invalid data is not desired replace parallel to serial, serial to parallel, time division multiplexer and time division demultiplexer block with an equivalent circuit built out of a counter, mux and up/down sampler blocks. The equivalent design circuit should also have a reset port pulled to the top-level and connected to the same signal driving the ce_clr port.

•

Counters used in performing operations like multiply-accumulate should always be reset using a combination of user reset which is tied to the ce_clr signal and ce signal extracted from the Clock Enable Probe block.

•

Always verify the effect of ce_clr signal on the design by importing and simulating the post translate HDL model as a black box.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator

Design Styles for the DSP48

About the DSP48

Xilinx Virtex™-4 and Virtex™-5 devices offer an efficient building block for DSP applications called the DSP48 (also known as the Xtreme DSP Slice). The DSP48 is available as a System Generator block which is a wrapper for the DSP48 UNISIM primitive.

Architectural and usage information for this primitive can be found in the DSP48 Users

Guide in the Virtex™-4 data book. The DSP48 is available in all Virtex™-4 devices and is featured in the SX series devices which contain up to 512 DSP48 blocks.

B, BCIN

18 18

BCOUT

P, PCOUT

PCIN

to various

control ports

= optional register with optional reset and clock enable

The DSP48 combines an 18-bit by 18-bit signed multiplier with a 48-bit adder and a programmable mux to select the adder's inputs. It implements the basic operation: "p=a*b

+(c+cin);", however other operations can be selected dynamically. Optional input and multiplier pipeline registers are also included and must be used to achieve maximum speed. Also included with the DSP48 are high performance local interconnects between adjacent DSP48 blocks (BCIN-BCOUT and PCIN-PCOUT). The DSP48 also includes support for symmetric rounding. This combination of features enables DSP systems which use the higher-speedDSP48 devices to be clocked at over 500 MHz.

There are three ways to program a DSP48 in System Generator:

•

Use Standard Components

- Map designs to Mult and AddSub blocks or use higherlevel IP such as the MACFIR filter generator blocks. This approach is useful if the design needs to be compatible with V2P or S3 devices or uses a lower-speed clock and the mapping to DSP48s is not required.

•

Use Synthesizable Blocks

- Structure the design to map onto the DSP48's internal architecture and compose the design from synthesizable Mult, AddSub, Mux and

Delay blocks. This approach relies on logic synthesis to infer DSP48 blocks where appropriate. This approach gives the compiler the most freedom and can often achieve full-rate performance.

•

Use DSP48 Blocks

- Use System Generator's DSP48 and DSP48 Macro blocks to directly implement DSP48-based designs. This is the highest performance design technique. Be aware however that obtaining maximum performance and minimum area for designs using DSP48s may require careful mapping of the target algorithm to the DSP48's internal architecture, as well as the physical planning of the design.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Design Styles for the DSP48

Designs Using Standard Components

Designs for Xilinx FPGAs such as Virtex™-II Pro and Spartan™-3 will compile to the

Virtex™-4 devices. Multipliers will be mapped into the DSP48 block, however, logic synthesis tools cannot pack adders and muxes into the DSP48 block since these blocks are delivered as cores which prevents synthesis from optimizing the logic. Place and route tools do place the MULT18x18S and MULT18x18 into the DSP48 block but do not pack the adder, or mux into the DSP48 block. (PAR will however pack the mux into the LUT-based adder).

To obtain the best possible performance, you should set the multiplier latency to 3 and include an input register to cover the delay from the DSP48's output to the adder. The performance of this circuit is in the 200-300 MHz range with V-44-11 devices, and is limited by the adder speed. In Virtex™-4, unlike Virtex™-II Pro and Spartan™-3 devices, the multiply speed in nearly independent of bit width. For medium speed designs, this approach works fine.

An additional way to use the DSP48 is to use IP blocks optimized for the DSP48 such as the

MACFIR block available from coregen, or to use the architecture wizard to generate a custom configured DSP48. Both of these approached require importing the logic containing the DSP48 as a black box into System Generator. Simulation will require

ModelSim HDL cosim.

Designs Using Synthesizable Mult, Mux and AddSub Blocks

Synthesis tools now have the ability to infer DSP48 logic. This enables the tools to pack adders, multipliers and muxes into the DSP48 block, as well as to enable the application of retiming and other synthesis techniques such as register duplication.

If the design is composed of synthesizable blocks, both Synplify Pro and XST have demonstrated the ability to infer DSP48s and to make use of the DSP48's local interconnect buses (PCOUT-PCIN and BCOUT-BCIN). In the above example, three blocks have been built using the MCode blocks which are defined by the following M-functions. function o = xlsynmux2(i0,i1,sel) if (sel==0) o=i0; else o=i1; end function p = xlsynmult(a,b) p=a*b; function s = xlsynadd(a,b)

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator s=a+b;

For synthesis to work, the circuit must be mappable to the DSP48 and signal bitwidths must be less than the equivalent buses in the DPS48.

You should kept in mind that the logic synthesis tools are rapidly evolving and that inferring DPS48 configurations is more of an art than a science. This means that some mappable designs may not be mapped efficiently, or that the mapping results may not be consistent. It will be necessary to inspect the post synthesis netlist using a tool similar to

Synplify Pro's gate-level technology viewer to determine if the design is being correctly mapped. If not, it may be possible to recast it to be correctly inferred. A model of a fully synthesizable FIR filter is located at the follwing pathname in the System Generator software tree:

.../sysgen/examples/dsp48/synth_fir/synth_fir_tb.mdl

Designs that Use DSP48 and DSP48 Macro Blocks

DSP48 Block

The DSP48 block is effectively a wrapper for the DSP48 UNISIM primitive. Because of this, any possible DSP48 design can be implemented. This low-level implementation however requires an 11-bit binary opmode to be routed to the DSP48's control ports in order to configure its function. The Constant block has a special mode enabling it to generate a

DSP48 control field. The DSP48's parameters dialog box is used to configure the pipelining mode of the DSP48 as well as the use of the DSP48's local interconnect buses named

PCOUT-PCIN and BCOUT-BCIN. You can try out the DSP48 block by opening the simulink model that is located at the follwing pathname in the System Generator software tree:

.../sysgen/examples/dsp48/dsp48_primitive.mdl

Dynamic Control of the DSP48

The DSP48 has the unique capability of being able to change its operation on a per cycle basis. This is useful in applications where the DSP48 is used in a 'resource shared' mode such as a FIR filter where multiple taps are implemented by the same multiplier. A simple www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Design Styles for the DSP48

method of generating this type of control pattern is to use a mux to select the DSP48 instruction on a clock by clock basis.

The above example illustrates the use of a DSP48 and Constant blocks to implement a 35bit by 35-bit multiplier over 4 clock cycles. During synthesis, the mux and constant logic is reduced by logic optimization. In the example above, the DSP48 block and the 4:1 mux are reduced to just two 4-LUTs. A Simulink model that illustrates how to implement both parallel and sequential 35*35-bit multipliers using dynamic operation for the sequential mode of operation is located at the follwing pathname in the System Generator software tree:

.../sysgen/examples/dsp48/mult35x35/mult35x35_tb.mdl

DSP48 Macro Block

The DSP48 Macro block is a wrapper for the DSP48 block which makes it simple to implement a sequence of DSP48 instructions (known as dynamic instructions). In addition, it provides support for specifying input and output types. For example, in the model above, a DSP48 Macro block is configured to implement a complex multiplier using a sequence of four different instructions. The instructions are entered in a text window in the

DSP48 Macro's dialog menu. You can try out the DSP48 Macro block by opening the simulink model that is located at the follwing pathname in the System Generator software

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 1: Hardware Design Using System Generator tree:

.../sysgen/examples/dsp48/dsp48_macro.mdl

DSP48 Design Techniques

Designing Filters with the DSP48

The DSP48 is an ideal block to implement FIR filters. You can examine how to use the

DSP48 block for Type 1 and Type 2 FIR filters by opening the simulink model that is located at the follwing pathname in the System Generator software tree:

.../sysgen/examples/dsp48/firs/dsp48_firs_tb.mdl

100

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Design Styles for the DSP48

Design Techniques for Very-High Performance Designs

DSP48-based designs usually require I/O, BRAMS and SLICE logic. Typically, this associated SLICE logic is used to implement delay registers, SRL16s, muxes, counters, and control logic. Since the DSP48 block is expected to operate at speeds greater than 500 MHz, other components will also be required to operate at the same speed. This generally requires special design techniques for the non-DSP48 logic.

At 500 MHz only 2 ns is available in each clock. For V4-11 devices, roughly 300 ps are required for register clock to out and 300 ps for setup. For comparison, a LUT delay is 166 ps. Special inputs and outputs such as clock enables and DSP48 and BRAM signals generally have setup and clock to out times closer to 500 ps. With clock skew and jitter, roughly 1 ns is available for net delays. This restriction will generally allow only 1 net in each path and it must be fairly short.

There are a number of guidelines that can be used to insure the operation at DSP48 speeds.

Some of these guidelines are outlined below.

8) srl16 as control pattern generator

9) dsp48 as counter, adder

5) Limit fanout to 4-8 loads

DSP48

SRL16 BRAM

2) Extra output regs

1) Pipeline regs

BRAM

lut

6) Use input and output regs with

LUTs

7) Limit to 1 level of logic

+ +

3) PCOUT-PCIN b

1) use input and output regs

4) Use extra registers to cover distance greater than 20-40 slices

Always use DSP48, BRAM16, FIFO16 with input, mult and output registers

Use additional FF to buffer DSP48 and BRAM outputs if necessary

Plan out the usage of the PCOUT-PCIN bus to allow DSP48 chaining

Add registers to any path that is greater than 20 - 40 slices long

Limit fanout to 32 loads located within a 20 slice distance

Add output registers to any LUT-based logic

Limit LUTs to 1 level or a 4:1 MUX and insure a local register for input or output

Use RAMs, SRL16 to clock out control patterns instead of state machines

Use DSP48 to implement counters and adders greater than 8-16 bits

10. Use area constraints – "INST ff1* LOC = SLICE_X0Y8:SLICE_X1Y23;"

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

101

Chapter 1: Hardware Design Using System Generator

Physical Planning for DSP48-Based Designs

The DSP48 requires correct placement to achieve dense, high performance designs. While the automatic place and route tools do a good job, the best results may require manual placement of DSP48 and RAM blocks. There are several additional issues with the DSP48s.

102

Cascade Routing Buses

Adjacent DSP48 blocks are connected with two local buses called PCOUT, and BCOUT.

The PCOUT bus is used to pass accumulation data from one DSP48 to the next. The

BCOUT bus is used to pass delayed B input data to the next DSP48. The DSP48 and DSP48

Macro block both support PCOUT and BCOUT buses. The use of the buses is shown in the figure above, which illustrates a pipelined 4-Tap Type 1 FIR filter.

C-Input Sharing

Each pair of DSP48s share a single C input. You should be aware of this when you do resource planning. Since the placer will not always find the most optimal placement to share C inputs, DSP48s should avoid using C inputs if possible.

Adder Trees Planning

Tree-based filter topologies are problematic for efficient DSP48 implementation. An adder tree requires isolated 2-input adders. Two input 36-bit adders can be implemented using a single DSP48, however this requires a C input and precludes the use of the multiplier. In addition, the long signals between DSP48s may require additional pipeline stages. A better approach is to convert the tree into a pipelined cascade.

Placement

Most designs will benefit from some placement of DSP48 and BRAMs. Use of area constraints to constrain LUT fabric logic placement may also be beneficial.

Signal Length Planning

At 500 MHz, signal lengths should be limited to around 20 slices. This means that long signals should have multiple pipeline stages.

Clock Enable Planning

When using the Clock Enables clocking option, the clock enables are often the limiting path at high frequencies. This is partially due to System Generator's use of LUTs to gate clocks at the destination. To avoid clock enables in the critical path, avoid using the System www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Design Styles for the DSP48

Generator upsampled and downsampled clock domains. This requires the manual use of clock enables for logic that runs at less than the system clock rate.

Place and Route Flow

•

Use the command map -timing with effort level high for both map and place

•

Use trce –v 100 to get a good sense of the failing nets and inspect the xflow/design.twr

file to understand the nature of the design's timing.

•

The file bitstream_v4.opt is available in the examples/dsp48 directory. This file can be used with the Bitstream compile target to set the PAR options mentioned above.

Synthesis Flow

•

Use Synplify Pro with retiming and pipelining enabled to avoid having to manually pipeline every LUT and signal.

•

Use Synplify Pro with the fanout limit set around 32 to avoid long net delays.

•

Open compiled projects in Synplify Pro and inspect the generated logic using the

RTL- and Gate-level views to get a good idea of what logic is being generated.

•

The file syn.pl is available in the examples/dsp48 directory. Place this file in

<sysgen_tree>/scripts

directory to modify the synthesis options in System

Generator

Logic Depth Planning

The following rules seem to allow the LUT fabric to run at 450 MHz using a -11 V4 device:

•

Only one net can be allowed in a critical path at 450 MHz. This allows a 4:1 mux to a reg a 4_input LUT to a reg or a net through a LUT directly to a DSP48

•

Counters up to 16-bits can be used, but do not use count limited counters without additional pipelining

•

If accumulators or counters are used, invert the enable line to an active-low condition to prevent a extra LUT from being inserted in the critical path

•

Any adders must have local input registers. It may be necessary to place control counters in the DSP48 to insure speed.

Fanout Planning

Avoid fanouts of more than 32 LUTs or 8 DSP48s or BRAMs. This can be avoided by inserting additional pipeline registers in these signals paths.

Register Retiming

Check retiming on delay blocks to allow them to be used as registers for pipelining. Then use Synplify Pro or XST with retiming enabled to allow the synthesis tool to move registers into optimal positions.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

103

Chapter 1: Hardware Design Using System Generator

Using FDATool in Digital Filter Applications

The following example demonstrates one way of specifying, implementing, and simulating a FIR filter using the FDATool block. The FDATool block is used to define the filter order and coefficients and the Xilinx Blocksets are used to implement a MAC-based

FIR filter using a single MAC (Multiply-ACcumulate) engine. The quality of frequency response is then validated by comparing it to a double-precision Simulink filter model.

104

Although a single MAC engine FIR filter is used for this example, we strongly recommend that you look at the DSP Reference Library provided as a part of the Xilinx Reference

Blockset. The DSP Reference Library consists of multi-MAC, as well as, multi-channel implementation examples with variations on the type of memory used.

A demo included in the System Generator demos library also shows an efficient way to implement a MAC-based interpolation filter. To see the demo, type the following in the

MATLAB command window:

>> demo blockset xilinx then select FIR filtering: Polyphase 1:8 filter using SRL16Es from the list of demo designs.

Design Overview

This design uses the random number source block from the DSP Blockset library to drive two different implementations of a FIR filter:

•

The first filter is the one that could be implemented in a Xilinx device. It is a fixedpoint FIR filter implemented with a dual-port Block memory and a single multiplyaccumulator.

•

The second filter is what is refered to as reference filter. It is a double-precision, directform II transpose filter. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Using FDATool in Digital Filter Applications

The frequency response of each filter is then plotted in a transfer function scope.

Open and Generate the Coefficients for this FIR Filter

From the MATLAB console window, cd into the directory

<sysgen_tree>/sysgen/examples/mac_fir

Open the design model by typing mac_df2t from your MATLAB command window.

For the purpose of this tutorial, the variables coef, coef_width, coef_binpt, data_width

, data_binpt and Fs are not defined. You will first use these variables as mask parameters to the MAC block and then design and assign the filter coefficients using the FDATool. The fully functional model is available in the current directory and is called mac_df2t_soln.mdl

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

105

Chapter 1: Hardware Design Using System Generator

Parameterize the MAC-Based FIR Block

Right Click on the MAC-Based FIR block and select Edit Mask as shown in the figure below.

Add the parameters coef, data_width and data_binpt as shown below.

106

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Using FDATool in Digital Filter Applications

Generate and Assign Coefficients for the FIR Filter

Drag and drop the FDATool block in your model from the DSP Xilinx Blockset Library.

Double-click on the FDATool block and enter the following specifications in the Filter

Design & Analysis Tool for a low-pass filter designed to eliminate high-frequency noise in audio systems:

♦

Sampling frequency: Fs = 44.1 KHz

Passband frequency: Fpass = 6 KHz

Stopband frequency: Fstop = 7.725 KHz

Passband ripple: Apass = 1dB

Stopband ripple: Astop = 48 dB

Click on Design Filter at the bottom of the tool window to find out the filter order and observe the magnitude response.

You can also view the phase response, impulse response, coefficients and more by selecting the appropriate icon at the top-right of the GUI. Based on the FDATool, a 43tap FIR filter is required in order to meet the design specifications listed above.

The filter coefficients can be displayed in the MATLAB workspace by typing:

>> xlfda_numerator('FDATool')

These useful functions help you find the maximum and minimum coefficient value in order to adequately specify the coefficient width and binary point:

>> max(xlfda_numerator('FDATool'))

>> min(xlfda_numerator('FDATool'))

For this tutorial, the coefficient type has been set to be Fix_12_12, which is a 12-bit number with the binary point to the left of the twelfth bit. The result of the max() function above shows that the largest coefficient is 0.3022, which means that the binary point may be positioned to the left of the most significant bit. How do you reason that?

A Fix_12_12 number has a range of -0.5 to 0.4998, meaning the dynamic range is maximized by putting the binary point left of the most significant bit. If you moved the binary point to the right (by using a Fix_12_11 number) you would lose one bit of dynamic range because a Fix_12_11 number has a range of -1 to 0.9995, which is more than you require to represent the coefficients.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

107

Chapter 1: Hardware Design Using System Generator

Enter the parameter values for coef, coef_width, coef_binpt, data_width, data_binpt and Fs as shown below.

In order to be able to preload the coefficients without having to regenerate them with the FDA Tool each time, the mac_df2t.mdl simulink model is opened, you are now going to save it into a file.

Browse Through and Understand the Xilinx Filter Block

The following block diagram showing how the MAC-based FIR filter has been implemented for this tutorial.

108

At this point, the MAC filter is set up for a 10-bit signed input data (Fix_10_8), a 12-bit signed coefficient (Fix_12_12), and 43 taps. All these parameters can be modified directly from the MAC block GUI. The coefficients and data need to be stored in a memory system.

For the tutorial, you choose to use a dual-port memory to store the data and coefficients, with the data being captured and read out using a circular RAM buffer. The RAM is used in a mixed-mode configuration: values are written and read from port A (RAM mode), and the coefficients are only read from port B (ROM mode). www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Using FDATool in Digital Filter Applications

The multiplier is set up to use the embedded multiplier resource available in Xilinx

Virtex™-II and Virtex™-II Pro devices as well as three levels of latency in order to achieve the fastest performance possible. The precision required for the multiplier and the accumulator is a function of the filter taps (coefficients) and the number of taps. Since these are fixed at design time, it is possible to tailor the hardware resources to the filter specification. The accumulator need only have sufficient precision to accumulate maximal input against the filter taps, which is calculated asa follows: acc_nbits = ceil(log2(sum(abs(coef*2^coef_width_bp)))) + data_width+ 1;

Upon reset, the accumulator re-initializes to its current input value rather than zero, which allows the MAC engine to stream data without stalling. A capture register is required for streaming operation since the MAC engine reloads its accumulator with an incoming sample after computing the last partial product for an output sample.

Finally, a downsampler reduces the capture register sample period to the output sample period. The block is configured with latency to obtain the most efficient hardware implementation. The downsampling rate is equal to the coefficient array length.

Run the Simulation

When you run the simulation you will get the following message as shown in the figure below. System Generator gets its input sample period from the din Gateway In block which has 1/Fs specified as the data input sample period. As the MAC-based FIR filter is oversampled according to the number of taps, the System Clock Period would always be equal to 1/(Filter Taps * Fs). Click on Update to specify the System Clock Period as

5.273427e-007 = 1/(43 * 44100).

You can now run the simulation and notice that the Xilinx implementation of the MACbased FIR filter meets the original filter specifications and that its frequency response is almost identical to the double precision Simulink models.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

109

Chapter 1: Hardware Design Using System Generator

As you can see, the filter passband response measurement as well as zeros can clearly be seen. Your FIR filter is perfect!

It is possible to increase or decrease the precision of the Xilinx Filter in order to reach the perfect area/performance/quality trade off required by your design specifications.

Stop the simulation and modify the data width to FIX_8_6 and the coefficient width to

FIX_10_10

from the block GUI. Update the model (Ctrl-d) and push into the MAC engine block. You should now notice that the datapath has been automatically updated to only eighteen bits on the output of the multiplier and twenty on the output of the accumulator.

110

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Using FDATool in Digital Filter Applications

Restart the simulation and observe how the frequency response has been affected. The attenuation has indeed degraded (less than 40dB) due to the fixed-wordlength effects.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

111

Chapter 1: Hardware Design Using System Generator

Generating Multiple Cycle-True Islands for Distinct Clocks

System Generator's shared memory interfaces allow you to implement designs that are driven by multiple-clock sources. These multi-clock designs may employ a combination of distinct clocks and derived clock enables to implement advanced clocking strategies completely within a single design environment. This topic describes how to implement multi-clock designs in System Generator through discussions of the following topics:

•

Applications that benefit from multiple clocks;

•

Using hierarchy to partition a System Generator model into two or more clock domains;

•

Using shared memories to cross clock domains;

•

Simulating and netlisting multiple clock designs;

•

Wiring multiple clock domains together using the Xilinx Multiple Subsystem

Generator block.

A step-by-step example is provided to help clarify the topics listed above. Although the example uses two clocks, the concepts presented here can be extended so that System

Generator designs requiring any number of clock sources can be constructed using similar techniques.

Before continuing with the example, you may want to familiarize yourself with standard

System Generator clocking terminology and implementation methodologies. This information is covered in-depth in the topic

Timing and Clocking . In general, System

Generator designs are driven by a single, system clock source. Multirate design portions are handled using clock enables derived from the system clock source. It is possible, however, to use System Generator to implement designs that are driven by distinct clock sources.

Broadly speaking, the approach is the following:

Divide the design into several subsystems, each of which is to be driven by a different clock. In the example, you call these subsystems asynchronous clock islands. Xilinx shared memory blocks should be used as bridges that communicate between these clock islands.

Once the design is partitioned, the Xilinx Multiple Subsystem Generator block may be used to translate the design into hardware that uses multiple distinct clock sources.

Multiple Clock Applications

A common application for multiple clock domains is for interfacing different pieces of external hardware that operate at different clock rates. For example, you may need to provide a set of I/O registers to a microprocessor, and the processor must be able to read and write these registers synchronous to its own clock. You may get data from a clock/data recovery unit and need to re-synchronize the data to your local clock domain. You may need to feed data to a digital-to-analog converter that must be running at a precise sample rate which is different from your system clock.

Another important application for multiple clock domains is in employing a high-speed processing unit. Let us take an example of an interpolating FIR filter. The filter gets symbol data from an external unit, and the filter needs to take the symbols and perform a 4X interpolation that creates four output samples for each input symbol. The output samples are fed to a digital-to-analog converter (DAC) that is clocked at the sample rate.

The FIR filter may be clocked at any of several rates. It may be clocked at the symbol rate, and on each cycle it must create four samples which will then be fed to the DAC at the sample rate. This highly parallel implementation has large hardware resource

112

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Generating Multiple Cycle-True Islands for Distinct Clocks

requirements and would only be employed if the sample rate were very fast. An alternative approach is to clock the FIR filter at the sample rate, creating one sample per cycle. This scenario takes an intermediate amount of hardware and would be used for intermediate sample rates. If the sample rate is slow, the FIR filter may be clocked at a rate several times faster than the sample rate, perhaps by means of a DCM that multiplies the sample-rate clock. In this way the multiplier-accumulator units of the FIR filter may be reused several times during the calculation of each sample output, requiring the least amount of hardware. This last method would use a symbol-rate clock domain, a highspeed processing clock domain, and a sample-rate clock domain.

A good FPGA design practice is to have each resource in the FPGA device operating at the highest possible rate to optimize hardware usage. In general, it is best to use a single clock domain when possible and to use clock enables to gate slower circuitry, creating multicycle paths. The drawback to this technique is that it increases power consumption and may make it difficult to route the high-speed clock enable. As a result, separate domains for high-speed processing are preferable in some instances. Also, it may not be possible to avoid dealing with different clock domains when dealing with asynchronous data inputs and outputs.

Clock Domain Partitioning

Partitioning a multiple-clock design into multiple domains is an important aspect of FPGA design. System Generator uses design hierarchy to support clock domain partitioning.

More specifically, when a design uses multiple clock domains, the logic associated with each distinct clock domain should be grouped together in a Simulink subsystem.

The subsystems, or in this case, synchronous islands, are cycle-true in the sense that the hardware that is generated for an island is faithful to the Simulink behavior of the island model. The notion of bit and cycle accuracy is preserved only within the individual synchronous islands. The end model containing the synchronous islands is not necessarily cycle-true, because it drives the islands with asynchronous clocks. Although System

Generator and Simulink are able to simulate the design using ideal clock sources, the complexities involved with asynchronous clocking systems can result in discrepancies between the software simulations and hardware realizations.

The advantages to partitioning a design using subsystems are manifold:

•

The physical clock lines are abstracted away from the block diagram;

•

Cross-domain transfers are well-defined and can be handled with metastable-safe blocks from the Xilinx Blockset;

•

Because the domains are well-defined, System Generator can accurately produce timing constraints for the synchronous islands.

The abstraction level of System Generator reduces the risk that users will perpetrate some of the more common design errors. These include:

•

Gated Clocks

: because the clocks in System Generator are inferred during hardware generation, it is not possible to connect non-clock lines to clock inputs (i.e., gated clocks).

•

Asynchronous Clears

: because the asynchronous resets in System Generator are inferred during hardware generation, it is not possible to explicitly clear synchronous logic using the asynchronous reset, which often results in timing problems.

•

Inferred Latches

: latches will not be generated from System Generator designs.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

113

Chapter 1: Hardware Design Using System Generator

Crossing Clock Domains

System Generator shared memory blocks should be used whenever it is necessary to cross clock domains. The tool provides several blocks for transferring data across clock domains, each of which is available in the Xilinx Shared Memory library:

•

Shared Memory

•

To FIFO / From FIFO

•

To Register / From Register

When these shared memory blocks are used to cross clock domains, each set should be split into a matched pair.

114

The To FIFO block is put in the domain in which it is to be written. The From FIFO is put in the domain in which it is to be read. The two blocks are linked by the name of the

Shared memory name

parameter. The FIFO is implemented in hardware using the Xilinx

FIFO Generator core. Using FIFO blocks is the safest and easiest-to-use of the three blocks which cross domains and is the best for high-bandwidth, sequential data transfers. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Generating Multiple Cycle-True Islands for Distinct Clocks

A pair of Shared Memory blocks is implemented as embedded Xilinx dual-port block

RAM core. The two blocks are linked by the name of the shared memory object. Each member of the pair resides in a different domain. Because the RAM is a true dual-port, each domain may write to the RAM. Care must be taken, by means of semaphores or other logic, to ensure that two writes or a read and a write to the same address do not happen simultaneously. For example, if domain A writes to a memory location at the same time that domain B is reading from it, the data read may not be valid. The shared memory is implemented as a using Xilinx Dual Port Block Memory core to ensure that large memories are efficiently mapped across multiple BRAMs.

The To Register is put in the domain in which it is to be written, and the From

in the domain from which it is to be read. The two blocks are linked by the name of the shared memory. The To Register may also be read synchronously in its own domain. The register may be of variable width and will synthesize as flip-flops. A 1-bit

To/From Register

pair will synthesize as a single flop.

Note:

Crossing domains in this manner can be unsafe, and requires the use of metastabilityreducing synchronization flops and semaphores for multiple-bit transfers. This technique should only be used when the hardware pitfalls are well-understood.

Netlisting Multiple Clock Designs

Each clock domain should have its own subsystem in a System Generator design. The diagram below shows a two-domain design. The top-level block contains the Multiple

Subsystem Generator block and two subsystems which each comprise a clock domain.

Each subsystem has a System Generator block that sets the system clock period for that clock domain.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

115

Chapter 1: Hardware Design Using System Generator

The diagram below illustrates the concept of putting domain-crossing blocks into their own subsystem. When a multiple-domain design is netlisted, System Generator does the following:

•

Creates an HDL file for Domain 0 (on the left), excluding the To FIFO block, and calls the netlister to create a black-box netlist delivered as an NGC file;

•

Creates an HDL file for Domain 1 (on the right), excluding the From FIFO block, and calls the netlister to create a black-box netlist delivered as an NGC file;

•

Invokes the Xilinx CORE Generator™ to produce a core for the FIFO block (middle);

•

Creates a top-level HDL wrapper that instantiates three block components.

116

Step-by-Step Example

This example shows how design hierarchy can be used to partition a System Generator design into multiple asynchronous clock islands. The example also demonstrates how

Xilinx Shared Memory blocks may be used to communicate between these islands. Lastly, the example describes how the Multiple Subsystem Generator block can be used to netlist the complete multi-clock design.

From the MATLAB window, change directory to the following:

<sysgen_tree>/examples/multiple_clocks/.

Open the two_async_clks model from the MATLAB command window, and save it into a temporary directory of your choosing.

Subsystem hierarchy is used in the example to partition the design into two synchronous clock domains, to which you refer as domains A and B, that are internally synchronous to a single clock, but asynchronous relative to each other. The design includes two subsystems named ss_clk_domainA and ss_clk_domainB, which include logic associated with clock domains, A and B, respectively. The blocks inside the ss_clk_domainA

subsystem operate in clock domain A while all blocks inside the ss_clk_domainB

subsystem operate in a second clock domain, B.

The asynchronous islands in the example communicate with one another via a shared memory interface implemented using a pair of Xilinx Shared Memory blocks. The two

Shared Memory blocks are distributed so that one block resides in domain ss_clk_domainA

and the other resides in domain ss_clk_domainB. Both blocks specify the same shared memory object name, bram_iface. This allows the Shared

Memory blocks to access a common address space during simulation. Note that in the diagram there is no physical connection shown between the two shared memory halves. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Generating Multiple Cycle-True Islands for Distinct Clocks

This is because the connection is implicitly defined by the fact that the two Shared Memory blocks specify the same shared memory object name and therefore, share an address space.

When the two subsystems are wired together and translated into hardware, the shared memory blocks are moved from their respective subsystems and merged into a block RAM core. For more information on how this works, refer to the topic Multiple Subsystem

Generator .

The synchronous islands sample different input sources. Island ss_clk_domainA samples a sinusoid input, while ss_clk_domainB samples a saw-tooth wave input. Each subsystem writes its samples into opposite halves of the shared memory. Once an island has filled its half of memory, it reads samples from the other island's half. You can simulate the design to visualize of the model's behavior.

Press the Simulink Start button to simulate the design.

Open the scope to visualize the output signals.

Also shown in the output scope are the two clocks, clk_A and clk_B. At the default time scale, it is difficult to distinguish the two. Zoom in to get a more detailed view.

Notice that clk_A and clk_B have different periods and are out of phase with one another. Earlier, it was claimed that System Generator uses a single clock source per design. In the scope, you clearly see two different clocks. How is this possible?

The answer is in the hierarchical construction of the design. All blocks are buried in at least one level of hierarchy using subsystems. Because there is no System Generator block at the top level, you can consider each subsystem as a completely separate System Generator design (at least for the time being). In this model, you have effectively defined two clock domains by giving the ss_clk_domainA and ss_clk_domainB subsystems different

Simulink system periods. This is allowed since you are treating these subsystems as separate System Generator designs. The clock probes in the ss_clk_domainA and ss_clk_domainB

subsystems use the Simulink system periods in their respective

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

117

Chapter 1: Hardware Design Using System Generator subsystems to determine their output, hence different system periods yield different system clocks.

Now consider the clocks defined by the System Generator block in the ss_clk_domainA and ss_clk_domainB subsystems.

Open the System Generator block parameter dialog boxes inside the ss_clk_domainA

and ss_clk_domainB subsystems.

118

The System Generator block dialog box in the ss_clk_domainA subsystem defines an

FPGA clock period of 10ns (i.e., a frequency of 100MHz). To simplify the sample period values in the model, the 10 ns clock is normalized to a Simulink system period value of 2 sec. In the ss_clk_domainB subsystem, an FPGA clock period of 15ns (i.e., a frequency

66.7 MHz) is defined. Normalizing this clock period gives us a Simulink system period value of 3 sec.

Because the two subsystems in this example implement multiple, synchronous, System

Generator domains, you will use the Multiple Subsystem Generator block to wire the subsystems together into a single HDL top-level component that exposes two clock ports.

When the Multiple Subsystem Generator translates a design into hardware, it generates each subsystem individually as an NGC netlist file. It also creates a top-level VHDL component or Verilog module that instantiates the subsystem netlist files as black boxes, and wires them together using shared memory cores as clock domain bridges.

You begin by using the Multiple Subsystem Generator block to netlist subsystems ss_clk_domainA

and ss_clk_domainB.

Open the Multiple Subsystem Generator dialog box by double clicking on the Multiple

Subsystem Generator block included in the top-level of the two_async_clks model.

Pick a suitable target directory inside the Multiple Subsystem Generator dialog box.

The default directory is netlist. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Generating Multiple Cycle-True Islands for Distinct Clocks

Press the Generate button. You may leave the Part, Synthesis Tool, and Hardware

Description Language

fields as they are.

Once the Multiple Subsystem Generator block is finished running, it will display a message box indicating that generation is complete. It is worthwhile to take a look at the generated results.

into the design's target directory, netlist.

There are two NGC files in this directory: ss_clk_domaina_cw.ngc and ss_clk_domainb_cw.ngc

. These files store the netlist and constraints information corresponding to the subsystems ss_clk_domaina and ss_clk_domainb. Note that these NGC files include the clock wrapper layer logic associated with each subsystem. This is necessary to ensure that any clock enable logic required by a multirate design is included in netlist file. By using the clock wrapper layer of a design, the corresponding clock driver components are automatically included in the netlist.

Also in this directory is a dual port memory core netlist file named dual_port_block_memory_virtex2_6_1_ef64ec122427b7be.edn

. This core provides the hardware implementation for the Shared Memory blocks used in the original design. The width and depth of the memory are based on values used in the Shared

Memory block configurations.

You will now take a look at the top-level HDL component that the Multiple Subsystem

Generator block produced for the design.

10. Open the two_async_clks.vhd file in a text editor.

This component defines the HDL top-level for the two_async_clks model. entity two_async_clks is port ( din_a: in std_logic_vector(7 downto 0); din_b: in std_logic_vector(7 downto 0); ss_clk_domaina_cw_ce: in std_logic := '1';

ss_clk_domaina_cw_clk: in std_logic;

ss_clk_domainb_cw_ce: in std_logic := '1';

ss_clk_domainb_cw_clk: in std_logic;

dout_a: out std_logic_vector(7 downto 0); dout_b: out std_logic_vector(7 downto 0)

); end two_async_clks;

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

119

Chapter 1: Hardware Design Using System Generator

There are several interesting things to notice about the port interface. First, the component exposes two clock ports (shown in bold text). The two clock ports are named after the subsystems from which they are derived (e.g., ss_clk_domaina), and are wired to their respective subsystem NGC netlist files. Also note that the top-level ports of each subsystem (e.g., din_a and dout_a) appear as top-level ports in the port interface.

The Multiple Subsystem Generator block does not generate circuitry (e.g., a DCM) to generate multiple clock sources. You may modify the top-level HDL component to include the circuitry, or instantiate the top-level HDL as a component in a separate wrapper that includes the clocking circuitry.

Creating a Top-Level Wrapper

If you decide to create a top-level HDL wrapper for your multi-clock System Generator design, it should perform the following tasks at a minimum:

•

Instantiate the System Generator top-level component along with other wrapper logic

(e.g., a DCM);

•

Wire the System generator component to the other logic;

•

Create a new top-level port map which supersedes that from the System Generator component.

The following is an example of making a top-level HDL component to instantiate clocking circuitry. In this example, you take the output created when the example from the previous topic is generated using the Multiple Subsystem Generator block. The resulting System

Generator design is called two_async_clks and the top-level HDL component is called top_wrapper

(for the case of VHDL synthesis).

Because the clock lines and main clock enables are inferred, the names of the clocks and clock enables (with the _ce and _clk suffixes above) are generated automatically by putting suffixes on the subsystem names from which the clocks are inferred. The other port names, such as dout_a, are taken directly from the names given to the gateway blocks in the System Generator design.

An example VHDL top-level wrapper to instantiate the entity two_async_clks, with deletions made for clarity, is provided below. Note that the wrapper uses a DCM component to generate the two clocks required by the System Generator design.

----------------------------------------------------------------------

-------

-- top_wrapper.vhd

-- Example Top Level Wrapper

-- This is an example top-level wrapper for instantiating a System

Generator

-- design along with a DCM. In this example, the DCM connects the two clock

-- inputs of the System Generator block ('two_async_clks') to two buffered

-- outputs of the DCM, namely, CLK0 and CLKFX. CLK0 is the same frequency

-- and phase as the input clock, and CLKFX is configured to be twice the

-- frequency of the input clock.

----------------------------------------------------------------------

--------- library IEEE; library unisim; use IEEE.std_logic_1164.all;

120

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Generating Multiple Cycle-True Islands for Distinct Clocks

use unisim.vcomponents.all; entity top_wrapper is port ( clk : in std_logic; din_a : in std_logic_vector(7 downto 0); din_b : in std_logic_vector(7 downto 0); dout_a : out std_logic_vector(7 downto 0); dout_b : out std_logic_vector(7 downto 0)

); end top_wrapper; architecture structural of top_wrapper is

--------------------------------------

-- SysGen Model Component Declaration

-------------------------------------- component two_async_clks port ( din_a: in std_logic_vector(7 downto 0); din_b: in std_logic_vector(7 downto 0); ss_clk_domaina_cw_ce: in std_logic := '1'; ss_clk_domaina_cw_clk: in std_logic; ss_clk_domainb_cw_ce: in std_logic := '1'; ss_clk_domainb_cw_clk: in std_logic; dout_a: out std_logic_vector(7 downto 0); dout_b: out std_logic_vector(7 downto 0)

);

end component;

component bufg port(i: in std_logic; o: out std_logic); end component;

--------------------------------------

-- DCM Component Declaration

--------------------------------------

component dcm

-- synopsys translate_off generic (clkout_phase_shift : string := "fixed"; dll_frequency_mode : string := "low"; duty_cycle_correction : boolean := true; clkdv_divide : real := 3; clkfx_multiply : integer := 2; clkfx_divide : integer := 1);

-- synopsys translate_on port (clkin : in std_logic; clkfb : in std_logic; dssen : in std_logic; psincdec : in std_logic; psen : in std_logic; psclk : in std_logic; rst : in std_logic; clk0 : out std_logic; clk90 : out std_logic; clk180 : out std_logic; clk270 : out std_logic; clk2x : out std_logic; clk2x180 : out std_logic; clkdv : out std_logic; clkfx : out std_logic; clkfx180 : out std_logic;

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

121

Chapter 1: Hardware Design Using System Generator locked : out std_logic; psdone : out std_logic; status : out std_ulogic_vector(7 downto 0)); end component;

--------------------------------------

-- DCM Attributes

-------------------------------------- attribute dll_frequency_mode : string; attribute duty_cycle_correction : string; attribute startup_wait : string; attribute clkdv_divide : string; attribute clkfx_multiply : string; attribute clkfx_divide : string; attribute clkin_period : string; attribute duty_cycle_correction of dcm0 : label is "true"; attribute startup_wait of dcm0 : label is "false"; attribute dll_frequency_mode of dcm0 : label is "low"; attribute clkdv_divide of dcm0 : label is "3"; attribute clkfx_multiply of dcm0 : label is "2"; attribute clkfx_divide of dcm0 : label is "1"; attribute clkin_period of dcm0 : label is "10"; signal clk0unbuf : std_logic; signal clk0buf : std_logic; signal clkfxbuf : std_logic; signal clk2xunbuf : std_logic; signal clkfxunbuf : std_logic; signal clkdvunbuf : std_logic; signal clkdvbuf : std_logic; signal ff1,ff2,ff3,ff4 : std_logic; signal dcm_rst : std_logic; signal intlock : std_logic;

----------------------------------------------------------------------

---------

-- The top level instantiates the SysGen design, a DCM, and two BUFGs.

-- The DCM generates two clocks of different frequencies.

-- These two clocks are used to drive the two different clock domains

-- in the SysGen block.

----------------------------------------------------------------------

--------- begin dcm0: dcm

-- synopsys translate_off generic map (dll_frequency_mode => frequency_mode, clkdv_divide => clkdv_divide_generic, clkfx_multiply => clkfx_multiply_generic, clkfx_divide => clkfx_divide_generic)

-- synopsys translate_on port map (clkin => clk, clkfb => clk0buf, dssen => '0', psincdec => '0', psen => '0', psclk => '0', rst => dcm_rst, clk0 => clk0unbuf, clk2x => clk2xunbuf,

122

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Generating Multiple Cycle-True Islands for Distinct Clocks

clkfx => clkfxunbuf, clkdv => clkdvunbuf, locked => intlock); bufg_clk0: bufg port map (i => clk0unbuf, o => clk0buf); bufg_clkfx: bufg port map (i => clkfxunbuf, o => clkfxbuf);

--------------------------------------------------------------------

-- This is the DCM reset. It is a four-cycle shift register used to

-- hold the DCM in reset for a few cycles after programming.

-------------------------------------------------------------------flop1: FDS port map (D => '0', C => clk, Q => ff1, S => '0'); flop2: FD port map (D => ff1, C => clk, Q => ff2); flop3: FD port map (D => ff2, C => clk, Q => ff3); flop4: FD port map (D => ff3, C => clk, Q => ff4); dcm_rst <= ff2 or ff3 or ff4;

------------------------------------------------------------

-- SysGen Component Port Mapping

-- One clock input is being connected to clk0 of the DCM,

-- and the other clock is being connected to clkfx.

------------------------------------------------------------ two_async_clks: two_async_clks port map ( din_a => din_a, din_b => din_b, ss_clk_domaina_cw_ce => '1', ss_clk_domaina_cw_clk => clk0buf, ss_clk_domainb_cw_ce => '1', ss_clk_domainb_cw_clk => clkfxbuf, dout_b => dout_b); end structural;

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

123

Chapter 1: Hardware Design Using System Generator

Using ChipScope Pro Analyzer for Real-Time Hardware Debugging

The integration of ChipScope™ Pro in the System Generator flow allows real-time debugging at system speed. By inserting a ChipScope block into your System generator design, you can debug and verify all the internal signals and nodes within your FPGA.

ChipScope Pro Overview

The increasing density of FPGA devices has rendered attaching test probes to these devices impractical. The ChipScope™ Pro tools integrate key logic analyzer hardware components with the target design inside of a Xilinx device. The ChipScope Pro tools communicate with these components during system operation and in effect provide the designer with a logic analyzer for nodes inside the Xilinx FPGA. ChipScope gives you a deep trace memory, fast clock speeds and multiple trigger options, which can vary in complexity. You can easily capture and view signal activity inside your FPGA without having to dedicate critical logic space, come up with complex capture schemes, or allocate additional I/O pins. Data samples are captured based on user-defined trigger conditions and stored in internal block memory. All control and data transfer is done via the JTAG port eliminating the need to drive data off-chip using I/O pins.

Please refer to the following Web page for further details on ChipScope Pro: http://www.xilinx.com/ise/optional_prod/cspro.htm

Tutorial Example: Using ChipScope in System Generator

Note:

This tutorial assumes that you have already installed and configured both the hardware and software required to run an ML506 platform. For installation and configuration information, refer to the

ML506 documents located at the following web address: http://www.xilinx.com/products/boards/ml506/docs.htm

This tutorial shows how to modify a Simulink model to integrate the ChipScope™ block and how to select the data to be captured and viewed for debugging. The steps are as follows:

From the MATLAB console, change the directory to

<sysgen_tree>/examples/chipscope

. The following files are located in this directory:

♦

♦ chip.mdl

– Your working model. chip_soln.mdl

– Solution model, including the ChipScope block.

Open the chip.mdl model from the MATLAB console. This model represents a simple usage model of a DDS Compiler block that will produce sine and cosine output waveforms. Both sine and cosine output waveforms will later be connected to a

Chipscope block, enabling you to debug and verify the Systen Generator block by probing and plotting the waveforms.

124

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Using ChipScope Pro Analyzer for Real-Time Hardware Debugging

The 8-bit Counter is used to trigger ChipScope. The most significant bit is extracted with a slice block and can be used for a variety of purposes such as driving an LED on the ML506 Platform for this exercise.

Simulate the model by clicking on the Start simulation Icon . At this point, without modifying the model, you should be able to see the following plot.

♦

The first plot represents the most significant bit of the 8-bit counter. The MSB becomes 1 when the counter output is within the range of 128 through 255.

The second plot represents the full output of the counter.

The third and forth plots show the output sine and cosine respectively.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

125

Chapter 1: Hardware Design Using System Generator

♦

The fifth plot shows a single-shot pulse that will be used to trigger sine and cosine output data.

Integrate ChipScope into the Simulink model. The ChipScope block can be found in the Simulink Library Browser in the Xilinx Blockset, under the Tools library. While holding down the left mouse button, select the ChipScope block and drag it into the open area in the lower-right corner of the Simulink model.

Double click on the ChipScope block in order to set the following parameters:

♦

Number of trigger ports

: Multiple trigger ports allow a larger range of events to be detected and can reduce the number of values that must be stored. Up to 16 trigger ports can be selected. In this example, only one is used.

♦

Display settings for trigger port

: For each trigger port, the number of match units and the match type need to be set. The pulldown menu displays options for a particular trigger port. For N ports, the display options for trigger port 0 to N-1 can be shown. In this example, there is one Trigger port named Trig0. This option should therefore be set to 0.

♦

Number of match units

: Using multiple match units per trigger port increases the flexibility of event detection. One to four match units can be used in conjunction to test for a trigger event. In this example, this option should be set to 1 since you are only checking for one condition (i.e., the 8-bit counter value). You will set the trigger value at run-time in the ChipScope Pro Analyzer.

♦

Match type

: This option can be set to one of the following six types:

1. Basic: performs = or <> comparisons

2. Basic With Edges: in addition to the basic operations high/low, low/high transitions can also be detected

3. Extended: performs =, <>,>,<, <=, >= comparisons

♦

4. Extended With Edges: in addition to the extended operations, high/low, low/high transitions can also be detected.

5. Range: performs =, <>, >, >=, <, <=, in range, not in range comparisons

6. Range With Edges: in addition to the range operations, high/low, low/high transitions can also be detected. In this example, set the Match Type to Basic with

Edges

Number of data ports

: Up to 256 bits can be captured per sample. This means that the sum over all ports of the bits used per port must be less than or equal to 256.

System Generator propagates the data width automatically; therefore, only the number of data ports needs to be specified. In this example, you want to view the sine and cosine and trig_counter, hence you enter 3.

Depth of capture buffer

: The depth of the capture buffer is a power of 2, up to

16384 samples. In this example, set the depth to 1024 (min value required for V5).

126

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Using ChipScope Pro Analyzer for Real-Time Hardware Debugging

After parameterization the ChipScope™ GUI should look like the following:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

127

Chapter 1: Hardware Design Using System Generator

Connecting the ChipScope Block

The signal used to trigger ChipScope is the counter output. The two buses that you want to probe are the sine and cosine from the Sine/Cosine table. Connect the signals appropriately as shown on the following figure:

Note that the names of the ports on the ChipScope block are specified by names given to the signals connected to the block, e.g. Sine and Cosine.

128

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Using ChipScope Pro Analyzer for Real-Time Hardware Debugging

Location Constraints

Now that the design is fully implemented and simulates correctly, the next step is to prepare it for connection to the hardware target. Although it can work on any hardware platform, the process is described for the ML506.

Two pins need to be locked down in this design: The LED and the clock pin.

♦

LED Pin

: Double click on the Gateway Out1 block, select Specify IOB Location constraints and type in {'AE24'} (note the need for single quotes).

♦

Clock Pin

: Double click on the System Generator block, set the clock period to

10ns and the clock pin location to AH15.

If you are using a different board, the pin locations should be modified appropriately.

System Generator GUI settings

The last two parameters that should be updated before generating a bitstream are the target device and the compilation target.

♦

Double click on the System Generator token and verify the parameter settings as follows:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

129

Chapter 1: Hardware Design Using System Generator

10. Bitstream Generation

Xilinx System Generator software automatically calls both the Core Generator™ and

ChipScope generator to create the netlist and cores. In addition, when the Bitstream target is selected, a configuration bitstream is created.

♦

Create a bitstream by pressing the Generate button.

The Core Generator is automatically called to generate the Sine/Cosine table and

Counter netlists. ChipScope generator is called to create an Integrated Logic

Analyzer (ILA) core and an ICON core to communicate with the ChipScope Pro software via the JTAG port.

Real-Time Debug

The next step is to run the design on the ML506 platform and view the probed outputs with the ChipScope™ Pro Analyzer.

Connect one end of the Parallel Cable IV or Platform USB cable to the General JTAG connector (J1) on the ML506 board. Connect the other end to your computer.

General JTAG (J1) Connector

Launch ChipScope Pro Analyzer

♦

Open the JTAG Chain by clicking on the following icon , or by selecting JTAG

Chain

> Xilinx Platform USE Cable. You should see the following table of the

JTAG Chain Device Order. After observing the order, click OK

130

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Using ChipScope Pro Analyzer for Real-Time Hardware Debugging

Configure the FPGA

♦

Under the New Project Window, right click on Device 4 > Configure > Select

New File

. At this point, you need to look for the bitstream which was generated in step 10 of the previous section ( ./bitstream/chip_cw.bit). After configuration, you should see an INFO message at the bottom of the ChipScope Analyzer window “Found 1 Core Unit in the JTAG chain”.

Import ChipScope Project File

System Generator creates a project file for ChipScope in order to group data signals into buses. A bus is created for each data port so that it can be viewed in the same manner (sign and precision) in which it was viewed in the Simulink environment.

Load this project file by going under File > Import > Select New File, and select

<sysgen_tree>/examples/chipscope/netlist/temp/chip_chipscope.cd

Plot the Sine Waves

♦

In the New Project window, under Device 4, double click on Trigger Setup to bring up the setup window but do not set it yet at this step.

♦

In the New Project window, under Device $ > Unit 0 MyILA0 (ILA), double click on

Bus Plot

A Bus Plot window appears. Select cosine and sine in the Bus Selection section, and then arm the trigger by clicking the button. Since you have not yet set any trigger conditions, values are captured immediately. Both the sine and cosine appear as shown below. You can change the display option to represent the waveforms with points, lines, or both.

Setup Trigger

In the Trigger Setup window, change the current X value with all 1s. A low-to-high pulse is used for this trigger and can be manually triggered by pushing the center PB

SW as shown below. ChipScope starts capturing data when it detects a low-to-high pulse. Earlier, you setup the buffer to 1024 so that up to 1024 data points can be captured and visualized in ChipScope.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

131

Chapter 1: Hardware Design Using System Generator

Re-capture the data by clicking on the button and you should see this screen below indicating that the ChipScope is waiting for a qualified trigger signal before capturing and displaying data.

This method of triggering is useful if you want a full control of when you like to capture the data. This is accomplished by connecting one of the PB switches to a single shot (Rising Edge Detector) circuit. The center PB switch (AJ6, SW14) is used for this exercise.

Center PB Switch

132

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Using ChipScope Pro Analyzer for Real-Time Hardware Debugging

Importing Data Into the MATLAB Workspace From ChipScope

Now you can export the data captured by ChipScope™ back into the MATLAB workspace.

Export data from ChipScope Pro Analyzer

♦

Select File > Export option from within ChipScope Pro Analyzer. Select ASCII format and choose Bus Plot Buses to export. Press the Export button and save the file as sinecos.prn.

Start MATLAB and change the current working directory to the location where you saved sinecos.prn.

♦

Type xlLoadChipScopeData('sinecos.prn'); This loads the data from the

.prn

file into the MATLAB workspace. In the workspace there are two new arrays named Sin and Cos.

You can plot the values using the MATLAB plot function.

♦

Type: plot(1:1024, sine, 1:1024, cosine) and the following plot is generated:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

133

Chapter 1: Hardware Design Using System Generator

134

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Chapter 2 Hardware/Software Co-Design

The Chapter covers topics regarding developing software and hardware in System

Generator.

Hardware/Software Co-Design in System Generator

Integrating a Processor with

Custom Logic

EDK Support

Designing with Embedded

Processors and Microcontrollers

A collection of tutorials that touch on designs with embedded processors.

A collection of tutorials that touch on designs with embedded processors

Documentation of support for the Xilinx Embedded

Development Kit.

A collection of tutorials that touch on designs with embedded processors.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

135

Chapter 2: Hardware/Software Co-Design

Hardware/Software Co-Design in System Generator

System Generator provides three ways for processors to be brought into a model; processors can be imported through a Black Box block, a PicoBlaze Microcontroller block and an EDK Processor block.

Black Box Block

The Black Box approach provides the largest degree of flexibility, at the cost of design complexity. You can interface any processor HDL into a System Generator design in this manner. All ports and buses on the processor can be exposed to the System Generator diagram, and you are free to engineer the required connectivity between the processor and other System Generator blocks. You also have complete control over software compilation

issues. Please refer to the topic Importing HDL Modules for more information.

PicoBlaze Block

The PicoBlaze™ block provides the smallest degree of flexibility but is the least complex to use. The Xilinx PicoBlaze Microcontroller block implements an embedded 8-bit microcontroller using the PicoBlaze macro, and exposes a fixed interface to System

Generator. Ordinarily, a single block ROM containing 1024 or fewer 8 bit words serves as the program store. You can program the PicoBlaze using the PicoBlaze Assembler language. This flow is documented in the topic

Designing PicoBlaze Microcontroller

Applications

EDK Processor Block

The EDK Processor block provides an interface to MicroBlaze™ processors created using the Xilinx Platform Studio (XPS). The EDK Processor block allows System Generator

Shared Memory blocks (i.e., "From/To Register"s, "From/To FIFOs", and "Shared Memory" blocks) to be associated with a processor through an automatically generated memory map interface. Once associated, that memory can be read or written in software running on the

MicroBlaze processor. This flow is documented in the topic Integrating a Processor with

Custom Logic .

The EDK Processor block can import a MicroBlaze specified through an EDK project created using Xilinx Platform Studio and Base System Builder. Alternatively, a System

Generator design with an EDK Processor block can also be exported into an EDK project.

The export process creates a PLB-based or FSL-based pcore, which can be added to any

XPS project and communicate with the MicroBlaze or PowerPC™ processor.

Integrating a Processor with Custom Logic

Integrating a processor with a piece of user-defined logic is typically a fairly involved process. The communications between a processor and a custom piece of hardware often occurs over a shared bus. Additionally, the information conveyed frequently consists of different types of data; for example data for processing, data denoting the status of the hardware or data affecting the mode of operation. Organizing how this data is transferred between the processor and custom logic is a tedious and error prone process that would benefit from automation. Furthermore, connectivity is only half of the problem, writing software to communicate with custom logic can also be challenging.

136

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Integrating a Processor with Custom Logic

The EDK Processor block provides a solution to both these problems through automation.

The EDK Processor block encourages the interface between the processor and the custom logic to be specified via shared-memories. Shared-memories are used to provide storage locations that can be referenced by name. This allows a memory map and the associated software drivers to be generated.

Please refer to the EDK Processor block documentation for information on the use of the block. The topics that follow describe the memory map and software features of the drivers that are generated.

Memory Map Creation

Hardware Generation

Hardware Co-Simulation

Generating Software Drivers

Writing Software for EDK

Processors

Asynchronous Support for EDK

Processors

Explains the memory map generated when shared memories are added to a processor.

Documents the different hardware generation options.

Explains how to create a hardware cosimulation model for the EDK Processor block.

Documents how software drivers are created.

Documents the process of writing software to control hardware created in System Generator.

Documents the capability in System Generator, in both import and export mode, to allow the processor and the System Generator design to run with different clocks.

Memory Map Creation

A System Generator model is shown on the bottom-right of the figure above. The System

Generator model corresponds to custom logic that will be integrated with the

MicroBlaze™ processor. In the construction of the model, shared-memories are used in

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

137

Chapter 2: Hardware/Software Co-Design locations where software access is required. For instance, the status of the hardware might be kept in a register. To make that status information visible in the processor, the register is replaced by a named shared-register. Naming the shared-register "status" gives the name of the memory context that will be useful later on during software development.

The block GUI of the EDK Processor block allows these shared-memories to be added to the memory map of the processor (bottom-left of the figure). The block diagram at the top of the figure above shows the flow of data. When a shared memory is added to the memory map of the processor, the EDK Processor block creates the corresponding matching shared memory. This shared memory is attached to the memory map that is generated for that

EDK Processor block. Next, a bus adaptor is used to connect that memory map to the

MicroBlaze processor.

When hardware is generated, each shared memory pair is implemented with a single physical memory. The implementation for each class of shared memory is documented in the topic

Shared Memory Support , found under the topic

Using Hardware Co-Simulation

Hardware Generation

The EDK Processor block supports two modes of operation: EDK pcore generation and

HDL netlisting. The different modes of operation are illustrated below and can be chosen from a list-box in the EDK Processor block's GUI.

138

EDK pcore Generation Mode

The Xilinx Embedded Development Kit (EDK) allows peripherals to be attached to processors created within the EDK. These peripherals can be packaged as pcores. Each pcore contains a collection of files describing the peripheral's hardware description, software drivers, bus connectivity and documentation.

When set in EDK-pcore-generation mode and used with the

EDK Export Tool

(selected via the System Generator block), System Generator is able to create a pcore from the given

System Generator model. The figure above shows the part of the model that is created as a pcore. When set in this mode, the assumption is that the MicroBlaze™ added to the model is just a place-holder. Its actual implementation will be filled in by the EDK when the peripheral is finally added into an EDK project. As such, the pcore that is created consists of the custom logic, the generated memory map and virtual connections to the custom logic, and the bus adaptor.

HDL Netlist Mode

An EDK processor can also be brought into a System Generator model when HDL netlisting mode is selected. The EDK Processor block can be set to HDL netlisting mode only when an EDK project is supplied to the block. When in HDL netlisting mode, the www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Integrating a Processor with Custom Logic

processor described in the EDK project will be imported into System Generator as a black box. The supplied EDK project is also augmented with the bus interfaces necessary to connect the System Generator memory map to the processor. During netlisting, the

MicroBlaze™ and the generated memory-map hardware are both netlisted into hardware.

Hardware Co-Simulation

Currently the EDK Processor block provides hardware-based simulation through hardware co-simulation. The creation of a hardware co-simulation block follows the

standard co-simulation flow described in the topic Using Hardware Co-Simulation

. The only difference is how top-level ports of the imported XPS project are treated.

When an XPS project is imported into System Generator, the import wizard assumes that all the ports are well constrained and applies that given constraint on the ports during the creation of the HWCosim block. That is to say, if the top-level entity of the XPS system contains ports that connect to pads on the FPGA, when compiling a HWCosim block, these ports will still connect to the pads on the FPGA and will not appear as ports on the

HWCosim block. Similarly, the bitstream flow constraints specified on top-level ports in the imported XPS system will be honored.

Should there be top-level ports that do not connect to pads, or are not constrained, these ports can be made visible in System Generator by exposing the ports using the Processor

Port Interface table in the Advanced tab of the EDK Processor block. See the topic Exposing

Processor Ports to System Generator

for details.

You may use the EDK's XPS tool to write and compile your software. However before simulation can begin, the Compile and update bitstream button in the co-simulation block's Software tab must be used to put the compiled C-code into the bitstream.

When used in conjunction with a hardware-board supported by network-based hardware co-simulation, it is possible to free up the JTAG port on the FPGA and use that for software debug with XMD.

Generating Software Drivers

In both modes of operation, the software driver templates are automatically created when a memory map is generated. The driver templates are only elaborated during the compilation of software libraries by the EDK. This can be accomplished from the EDK.

Once the software libraries have been compiled, the drivers can be referenced and the software documentation can also be accessed.

When the EDK Processor block is placed in EDK pcore generation mode, the elaboration of the software drivers depend on the instance name of the pcore. For example, a pcore created from System Generator may be called sysgen_fft_sm. This is the name of the peripheral. Since more than one of these peripherals can be added to an EDK processor, each instance of the peripheral needs a unique name. The EDK automatically assigns a numeric postfix to a peripheral's name. So when the peripheral is first added, it might be called sysgen_fft_sm_0, this is the instance name of the peripheral. You can change the instance name of a peripheral. Please refer to the EDK XPS documentation for further information.

When the EDK Processor block is placed in HDL netlisting mode and an EDK Project is imported into the System Generator model, System Generator automatically places a special peripheral into the EDK project. The peripheral provides the connectivity between the MicroBlaze™ and the System Generator model. This peripheral is given the instance

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

139

Chapter 2: Hardware/Software Co-Design name xlsg_iface. Only one instance of this peripheral is allowed; an EDK Project can only be associated with one EDK Processor block.

Writing Software for EDK Processors

Software drivers and documentation for a peripheral is elaborated after all software libraries have been compiled in the EDK. After compilation, the software documentation can be accessed from the EDK.

140

The figure above shows a screen capture of the EDK XPS tool. Locate the System Generator peripheral from the System Assembly view and right click for a pop-up menu. Select

View API Documentation

to bring up the documentation. If this option is not available, the drivers need to be compiled. This can be done in XPS by selecting the menu option

Software > Generate Libraries and BSPs

Please refer to the generated documentation for header file information, driver calls, memory maps and also example code.

The generated software drivers contain four basic functions for accessing sharedmemories. In the table below, <inst> refers to the instance name given to the peripheral. int <inst>_Read (unsigned int memName, unsigned int addr, unsigned int* val); int <inst>_ArrayRead (unsigned int memName, unsigned int startAddr, unsigned int transferLength, unsigned int** valBuf); int <inst>_Write (unsigned int memName, unsigned int addr, unsigned int val); int <inst>_ArrayWrite (unsigned int memName, unsigned int startAddr, unsigned int transferLength, const unsigned int* valBuf); www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Integrating a Processor with Custom Logic

Asynchronous Support for EDK Processors

Asynchronous support for processors allow for the processor and the accelerator hardware hanging off the processor to be clocked with different clocks. This allows the hardware accelerator to run at the fastest possible clock rate, or at a clock rate that is necessary for its correct functioning, for example when it is required to interface with an external peripheral.

This feature is enabled when the Dual Clock check box is selected in the EDK Processor block GUI's Implementation tab. The figure below shows how clocks will be connected for the import and export flow, in the export flow, the MicroBlaze ™(MB) block is not present.

Basically, the custom logic design in System Generator is driven with the clk clock and the processor system is driven with the xps_clk clock. The clock source that drives the PLB bus in the MicroBlaze processor system is extracted to drive the bus adaptor, the memory map, and halves of the shared memories.Shared memories straddle between these two domains

(e.g. the clk domain and the plb_clock domain) and are driven by both these clocks. Shared

Registers are not supported in this flow. In the import flow where an XPS project is imported into System Generator, the PLB bus on the processor must be driven with the same clock as the xps_clk signal.

When Dual Clock is enabled and a design is netlisted for Hardware Co-simulation, a slightly different clock wiring topology is used. This is shown in the figure below. The clock source from the board is bifurcated with one branch going into the Hardware Cosimulation module before being connected to the clk clock (depicted in the figure above).

The other branch is routed through a clock buffer and connected to the xps_clk clock signal.

This topology allows for the custom logic designed in System Generator to be singlestepped, while allowing the MicroBlaze to continue in free-running mode. This allows for clock-sensitive peripherals (such as the RS232 UARTS) to work when the Hardware Co-

Simulation token is set to single-step.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

141

Chapter 2: Hardware/Software Co-Design

In Hardware Co-simulation, the processor subsystem is driven by the board clock directly.

This means that the processor subsystem must be able to meet the requirements set by this clock. In hardware co-simulation, it is possible for users to select different ratios of clock frequencies based of the input board frequency. Note that this hardware co-simulation clock is generated in the hardware co-simulation module and is not available to the processor subsystem.

For exmaple, if the input board frequency is 125MHz, and the hardware co-simulation frequency is set to 33 Mhz, only the custom logic portion of the design will be constrained to 33 MHz, the MicroBlaze must still run at 125 MHz. If the MicroBlaze cannot meet timing at this speed, the user needs to instance a clock generator pheripheral in their XPS project and slow down the clock in that way.

142

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

EDK Support

Importing an EDK Processor

Exposing Processor Ports to

Please refer to the topic EDK Export Tool for more information.

Exporting a pcore

How to import an EDK project into System

Generator using the EDK Import Wizard.

How to route top-level ports in the EDK into

System Generator.

How to export a System Generator design to the

EDK as a pcore.

Importing an EDK Processor

A processor created using the Xilinx Platform Studio (XPS) tool, found in the Xilinx EDK suite of tools, can be imported into a System Generator model using the EDK Import

Wizard.

There are two ways to launch the EDK Import Wizard in the EDK Processor block: (1) press the Import… button, or (2) select HDL netlisting when the EDK project field is empty.

Note:

When you import the EDK Project into System Generator, there are modifications made to the

EDK project. These modifications are described in the following topic.

EDK Import Wizard

When the Wizard starts up, it prompts you for an EDK project file (xmp file).

Clicking the Import... button starts the import process.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

143

Chapter 2: Hardware/Software Co-Design

Note:

The import process will alter your EDK project to work inside System Generator. If you wish to retain an unadulterated version, please make a copy before importing. System Generator automatically backs up the hardware platform specification (i.e., the MHS file) and the software platform specification (the MSS file) of the EDK project to files with the "bak" suffix.

When an EDK project is imported into System Generator, the EDK project is augmented with a pair of FSL, or a PLB46 interface depending on the options made on the EDK

Processor block. A pcore (xlsg_iface for FSL and xlsg_plbiface for PLB) is also added to provide software drivers for the interface.The MHS and MSS files in the EDK project will be altered. Following that, the HDL files that describe the processor will be generated and linked to your System Generator project.

Making Changes to Processor Hardware After an Import

After importing an EDK project, further changes to the hardware inside of the EDK will not be reflected inside of System Generator. In other words, the hardware makeup of the processor is now fixed. If changes to the processor hardware are to be made, the EDK project must be re-imported using the EDK Import Wizard.

It is recommended that you re-import an EDK project when it is changed. The EDK

Processor block can detect the PLB or FSL interfaces and the related pcores that are automatically added by System Generator during a previoius import, and will not include redundant hardware or software to the XPS project.

Limitations

Currently the Wizard can only import single processor projects. Only the MicroBlaze processor is supported. Peripherals added to the processor cannot conflict with the resources used by other System Generator services. For instance, if network-based hardware co-simulation is used, the EDK project cannot make use of the peripherals using the Ethernet MAC.

144

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

EDK Support

Exposing Processor Ports to System Generator

The preferred mechanism for getting data to and from the processor and System Generator is via shared-memories. It is however possible to expose ports on the top-level of the processor to System Generator.

The top-right box in the figure above shows a snippet from an EDK project in XPS. The external port list has among other ports, a user-defined port called myExternalPort.

After importing the EDK project, open up the processor's block GUI in System Generator.

Select the Advanced tab to reveal the processor port interface table.

The port list shows all the top-level ports available on the processor. This port list has been filtered to remove clock ports and also signals used by System Generator to implement the memory-map interface. In this example, the RS232 ports, sys_rst_pin and myexternalport are shown to be ports that can be exposed to the top-level of the System Generator block.

Selecting the expose check box will cause the port to be exposed on the EDK Processor block. As shown in the figure above, the display name of the port can be changed, should the original name be too long.

This mechanism allows ports from the processor to be directly exposed to the System

Generator design without going through the memory map generated by System

Generator. You may choose to do this to expose the reset ports on the processor, or to expose interrupt ports directly to the System Generator diagram.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

145

Chapter 2: Hardware/Software Co-Design

Exporting a pcore

System Generator designs containing an EDK Processor block can be exported as an EDK pcore using the EDK Export Tool compilation target on the System Generator block.

Before exporting to the EDK as a pcore, the EDK Processor block must be configured for

"EDK pcore generation". This can be done by opening the EDK Processor block GUI and selecting the relevant drop down option in the "Configure processor for" parameter.

Designing with Embedded Processors and Microcontrollers

Designing PicoBlaze Microcontroller Applications

The PicoBlaze™ block in System Generator implements an 8-bit microcontroller.

Applications requiring a complex, but non-time critical state machine as well as data processing applications are candidates to employ this block. The microcontroller is fully embedded into the device and requires no external support. Any additional logic can be connected to the microcontroller inside the device providing ultimate flexibility.

PicoBlaze Overview

The following example uses PicoBlaze 3 (hereto referred to simply as PicoBlaze), which is optimized for low resource requirements. A memory block is used as a program store for up to 1024 instructions.

146

Signal Direction

in_port[7:0] Input brk rst instr[17:0] out_port[7:0]

Input

Output

Description

Input Data Port. During an

INPUT operation, data is transferred from the port to a register.

Interrupt. Must be at least two clock cycles in duration.

Reset

Instruction Input.

Output Data Port.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

Signal

port_id[7:0] rs ws addr[9:0] ack

Direction

Output

Description

Port Address.

Read Strobe.

Write Strobe.

Address of the next instruction.

Interrupt Acknowledge.

Architecture Highlights

•

Predictable performance, two clock cycles per instruction

•

43 - 66 MIPS (dependent upon device type and speed grade)

•

Fast interrupt response

•

96 slices, 0.5 to 1 block RAM

•

16 8-bit general-purpose registers

•

64-byte internal RAM

•

Internal 31-location CALL/RETURN stack

•

256 input and 256 output ports

1Kx18

Instruction

PROM

Instruction

Decoder

INTERRUPT

Enable

IN_PORT

64-Byte

Scratchpad RAM

Constants

16 Byte-Wide Registers s0 s4 s1 s5 s2 s6 s8 s9 sA sC sD sE s3 s7 sB sF

PORT_ID

OUT_PORT

Flags

Zero

Carry

Operand 1

ALU

Operand 2

PicoBlaze Instruction Set Architecture

PicoBlaze is a hardware-centric microcontroller, which can be programmed using assembly code. It supports a program length up to 1024 instructions. Requirements for larger program space are typically addressed by using multiple microcontrollers.

16 General Purpose Registers

There are 16 8-bit general-purpose registers specified 's0' to 'sF'.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

147

Chapter 2: Hardware/Software Co-Design

ALU

The Arithmetic Logic Unit (ALU) provides operations such as add, sub, load, and, or, xor,

shift

, rotate, compare, and test. The first operand to each instruction is a register to which the result is stored. Operations requiring a second operand can specify either a second register or an 8-bit constant value.

Flags and Program Control

The result of an ALU operation determines the status of the zero and carry flags. The zero flag is set whenever the result is zero. The carry flag is set when there is an overflow from an arithmetic operation. The status of the flags can be used to determine the execution sequence of the program using conditional program flow control instructions such as

jump

and call.

Input/Output

There are 256 input ports and 256 output ports. The port being accessed is indicated by an

8-bit address value provided on port_id. The port address can be specified in the program as an absolute value or indirectly specified as the contents of a register. During an

input

operation, the value provided to in_port is transferred into any of the 16 registers.

During an output operation, a value is transferred from a register to out_port.

Interrupt

The processor provides a single interrupt input port, brk. When interrupts are enabled, setting brk to 1 causes the program counter to be set to memory location 0x3FF, where a jump vector to the interrupt service routine is stored. At this time, a pulse is generated on the ack port (two clock cycles after brk is asserted), the control flags are preserved and further interrupts are disabled. The return instruction ensures that the end of an interrupt routine restores the status of the control flags and specifies if future interrupts should be enabled.

For extensive details regarding the feature and instruction set, please refer online to the topic PicoBlaze User Resources .

Tutorial Example - Using PicoBlaze in System Generator

In the following example, you modify a PicoBlaze program that alters the output frequency of a Direct Digital Synthesizer (DDS) during an interrupt.

A Simulink model and PicoBlaze assembler code are provided but need modification.

From the MATLAB console, change directory to

<sysgen_tree>/examples/picoblaze

. The following files are located in this directory:

♦

Pico_dds.mdl – An unfinished Simulink model

Pico_code.psm – Unfinished PicoBlaze code

Open Pico_dds.mdl.

Modify the design. a.

Find the PicoBlaze block in the Xilinx Blockset Library under Index or Control

Logic and add it to the model where indicated. The default settings of the block do not give the same number of ports as is expected by the model. This will be corrected in the following step. You may need to resize the block to fit into the space allocated in the design.

148

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

b. Double-click the block and set Version to PicoBlaze 3. Turn off the option to

Display internal state

. Connect the ports to the existing lines in the model.

Find the PicoBlaze Instruction Display block in the Index or Tools Library and add it to the model where indicated. Make sure it is connected properly, as shown in the figure below:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

149

Chapter 2: Hardware/Software Co-Design d. Double-click the PicoBlaze Instruction Display block and set the Version to

PicoBlaze 3

. Check the Disable Display option. Disabling the display option allows the simulation to run without the overhead of updating the block display.

Find the ROM block in the Memory Library and add it to the model where indicated. Flip the block by Right clicking on the block and selecting Format > Flip

Block

. Attach the ports to the existing lines. f.

Change the Single-Step Simulation block to be in continuous mode by double clicking on the block.

Configure the program store. Double click the ROM to do the following.

With the Basic tab selected: a.

The ROM block is used to store the PicoBlaze instructions. The depth of the ROM must be set to 1024. This is because the program uses interrupts and setting brk to

1 causes the program counter to be set to 0x3FF. b. As detailed in step 5, the code is assembled and produces an initialization file for the memory named fill_pico_code_program_store.m. Hence the ROM

Initial Value Vector should be set to fill_pico_code_program_store. c.

To increase the performance for synchronous designs, the Latency should be set to

1

150

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

Click on the Output tab and enter the following: a.

The Word Type should be Unsigned and Number of Bits should be set to 18 with the Binary Point at 0.

Edit the PicoBlaze assembly program.

Open pico_code.psm. b. Add instructions as described in the pico_code.psm file. For detailed information about the PicoBlaze instruction set see the Xilinx Application Note

XAPP627 at http://www.xilinx.com/bvdocs/appnotes/xapp627.pdf

Save the file.

Run the assembler to generate the memory initialization file.

In the MATLAB command window, type:

>> xlpb_as -p pico_code.psm

xlpb_as

stands for Xilinx PicoBlaze Assembler.

A file named fill_pico_code_program_store.m was created.

Simulate the Simulink model.

Run the simulation by clicking on the Start Simulation Icon.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

151

Chapter 2: Hardware/Software Co-Design

Output should look like this:

152

Notice the sine wave frequency increasing proportionally to the phase increment.

Utilize Debug Tools

If the program is not working properly, there are several tools that can be utilized to ease debugging. Deselecting the Disable Display checkbox in the PicoBlaze Instruction

Display block causes the block to be activated, displaying the updated program counter and instruction each clock cycle. In conjunction with enabling the display, the registers and control flag values can be viewed by selecting the Display Internal State in the PicoBlaze

Microcontroller block. Change the Single-Step Simulation block to single-step mode by double clicking on the block. Step through the simulation to debug. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

Designing and Exporting MicroBlaze Processor Peripherals

The Xilinx Platform Studio (XPS) tool suite allows the development of customized

MicroBlaze™ and PowerPC™ processor systems. A hardware peripheral of the processor system is called “pcore”, which consists of a bundle of design files organized according to a specific structure. These design files describe the hardware implementation, the connection interface, and software drivers of the XPS pcore.

The EDK Processor block in conjunction with the EDK Export Tool allows customized

processor hardware peripherals to be designed in System Generator. A System Generator design can be exported as an XPS pcore, which can be included and used in an XPS project.

The following tutorial illustrates the creation of a XPS pcore using System Generator.The files used in this tutorial can be found in: <sysgen_tree>\examples\EDK\rgb2gray, where <sysgen_tree> denotes the System Generator installation directory.

Tutorial Example - Creating MicroBlaze Peripherals in System Generator

Note:

You must have EDK installed to complete this tutorial.

Open the rgb2gray model from pathname

<sysgen_tree>\examples\EDK\rgb2gray

The peripheral contains three inputs, which are 32-bit red, green and blue pixel values.

These values are scaled and summed to produce a result that represents the 32-bit grayscale value. The red, green and blue values are sourced from three shared registers named 'red', 'green' and 'blue'. The result is written back to a shared register called 'result'.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

153

Chapter 2: Hardware/Software Co-Design

Prepare to export the pcore. Drag an EDK processor block into the model. Configure the processor block by double clicking on the block to bring up the block's dialog box, as shown below:

Add all available shared memories in the model to the EDK Processor by verifying/selecting <all> , then click the Add button. As shown above, ensure that the

EDK Processor block has been configured for EDK pcore generation in the Configure processor for drop-down menu. Dismiss the GUI by clicking the OK button. The EDK

Processor will then create a memory map for the shared memories.

Expore the pcore. Double click on the System Generator token to open up the System

Generator dialog box. You will use the EDK Export Tool to create the pcore. Options in

154

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

the

EDK Export Tool

are more fully detailed in the topic

Information on how to create an XPS project can be found in the topic Using XPS . Follow

Compilation Types

As shown above, set the Compilation type to be Export as a pcore to EDK. Click on the

Settings...

button to open up options for the compilation target. Accept the default settings so that the pcore is generated and exported into the model's target directory.

Click on the Generate button to initiate the pcore export process.

Integrate the Exported pcore in the XPS

You will now create an XPS project and integrate the pcore into the XPS project.

the directions there to create an XPS project.

Once the XPS project is created, copy the pcore that is created by System Generator into its local pcore repository. Since System Generator is instructed to place the pcore inside the target directory in the previous step, you should find a directory named pcore inside the target directory. Copy the contents of the directory into the corresponding pcore directory inside your XPS project. If your XPS project does not contain a pcore directory, create one before copying.

In the XPS menu, select Project > Rescan User Repositories. The pcore exported by System

Generator, named rgb2gray_plbw, will appear on the list of EDK Peripherals after the rescan.

Follow the directions in the topic

Using XPS

for information on how to connect up a pcore to the MicroBlaze™ processor in the EDK tool.

After connecting up the pcore, compile the netlist by selecting Hardware > Generate

Netlist

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

155

Chapter 2: Hardware/Software Co-Design

Write Software

Create a new software application in your XPS project. Again, information on how to do

this can be found in the topic Using XPS

. Add the following code to your application and compile the software.

#include "xparameters.h"

#include "stdio.h"

#include "xutil.h"

// header file of System Generator Pcore

#include "rgb2gray_plbw.h" int main (void) {

int i;

uint32_t gray, red, green, blue;

print("-- Entering main() --\n\r");

xc_iface_t *iface;

xc_from_reg_t *fromreg_gray;

xc_to_reg_t *toreg_red, *toreg_green, *toreg_blue;

// initialize the software driver

xc_create(&iface, &RGB2GRAY_PLBW_ConfigTable[0]);

// obtain the memory locations

xc_get_shmem(iface, "result", (void **) &fromreg_gray);

xc_get_shmem(iface, "red", (void **) &toreg_red);

xc_get_shmem(iface, "green", (void **) &toreg_green);

xc_get_shmem(iface, "blue", (void **) &toreg_blue);

for (i=15; i<30; i++){

red = i;

green = i + 10;

blue = i + 20;

// Write RGB value to peripheral

xc_write(iface, toreg_red->din, red);

xc_write(iface, toreg_green->din, green);

xc_write(iface, toreg_blue->din, blue);

xil_printf("R = 0x%x, G = 0x%x, B = 0x%x -- ",

red, green, blue);

xc_read(iface, fromreg_gray->dout, &gray);

xil_printf("Gray = %x \n\r", gray);

}

print("-- Exiting main() --\n\r");

return 0;

}

156

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

There can be multiple instances of a System Generator pcore in an XPS project. Each of the instances is associated with a device ID, which can be found in “xparameter.h”. Assume that the instance of interest has a device ID of 0 based on the following information in

“xparameter.h”.

/* Definitions for driver SG_PLBIFACE */

#define XPAR_SG_PLBIFACE_NUM_INSTANCES 1

/* Definitions for peripheral SG_PLBIFACE_0 */

#define XPAR_SG_PLBIFACE_0_DEVICE_ID 0

Use the device ID of a System Generator pcore instance to select the corresponding item in

RGB2GRAY_PLBW_ConfigTable

, which is then provided to xc_create to retrieve the settings of the specific Systen Generator pcore instance.

The topic

Integrating a Processor with Custom Logic

contains more information on how the hardware is wired up and other software issues.

Running the code will produce the following print out on a RS232 terminal.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

157

Chapter 2: Hardware/Software Co-Design

Tutorial Example - Designing and Simulating MicroBlaze Processor

Systems

This topic shows an example on how to design and simulate a System Generator model containing a MicroBlaze™ processor. A DSP48 co-processor is developed using System

Generator. Using the EDK Processor block, you import a MicroBlaze processor, customized in Xilinx Platform Studio (XPS), into the System Generator model. You then attach the

DSP48 co-processor to the imported MicroBlaze processor through the automatic memory mapping mechanisms provided by the EDK Processor block.

This tutorial uses hardware co-simulation to simulate and verify the design. In this case, the MicroBlaze processor is compiled into hardware, while the DSP48 co-processor model is left in the System Generator diagram for software simulation. In this example, the hardware simulation and software simulation communicate with each other using the point-to-point Ethernet co-simulation technology.

This tutorial example contains the following topics:

•

Create an XPS Project

•

Create a DSP48 Co-Processor Model

•

Import an XPS Project

•

Configure Memory Map Interface

•

Write Software Programs

•

Create a Hardware Co-Simulation Block

•

Create a Testbench Model

•

Update the Co-Simulation Block with Compiled Software

•

Run the Simulation

This example uses the Xilinx Virtex™-4 ML402 Evaluation Platform.

The files used in this tutorial can be found a pathname:

<sysgen_tree>\examples\EDK\DSP48CoProcessor

, where <sysgen_tree> denotes the System Generator installation directory.

158

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

Create an XPS Project

First of all, you will need to create a new XPS project, which contains an PLB-based UART

peripheral. A tutorial on how to create a new XPS project can be found in the topic Using

XPS .

Create a DSP48 Co-Processor Model

Copy the DSP48CoProcessorModel found in the folder

<sysgen_tree>\examples\EDK\DSP48CoProcessor

into a temporary working directory, then open the model.

The model contains a DSP48 block with the a, b, and c ports fed from three shared From

Registers with corresponding names. The op port receives signals from a multiplexer whose select line is sourced from a shared From FIFO named instr.

The output port p of the DSP48 block is sliced and fed into another two shared To Resigers: the top 16 bits into the overflow register and the bottom 32 bits into the result register.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

159

Chapter 2: Hardware/Software Co-Design

Import the XPS Project

In this step, you import an XPS project that contains a MicroBlaze processor into the DSP48

Co-Processor model. Double click on the subsystem called Processor Subsystem and look into it. In that subsystem, you will find a System Generator token and a blue text box as the place holder for an EDK Processor block. Open the System Generator block set and drag an

EDK Processor block from the Index library into the Processor Subsystem. Your augmented subsystem should look like the one shown below:

160

You will now configure the EDK Processor block to import the XPS project. The import process will make changes to the XPS project. Thus, ensure that the XPS project is not currently opened by Xilinx Platform Studio before importing.

Double click on the processor block to bring up the block dialog box. In the Configure

processor for

drop-down menu, select HDL netlisting. The Import… button is enabled as a result of the selection.

Note that the Import… button is disabled when the processor is configured for EDK pcore

generation. In EDK pcore generation mode, it is expected that you will create a pcore in

System Generator and export it to be used in another XPS project. In this case, the processor is not instanced inside the EDK Processor block. In HDL netlisting mode, it is expected that you import an XPS project into the System Generator model and netlist it with other System Generator blocks.

If no XPS project is ever imported, configuring the processor for HDL netlisting will automatically trigger the launching of the XPS Import Wizard. The XPS Import Wizard can be launched manually by pressing the Import… button.

In the pop-up file selection dialog, browse to the XPS project created in earlier steps. The import process starts once a XPS project file (xmp file) is selected. The import process copies necessary files into the XPS project and changes the project accordingly to allow the

MicroBlaze processor to communicate with the System Generator model.

Note that if there is any software applications contained by the imported XPS project, they are not compiled during import. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

Configure Memory Map Interface

Re-open the dialog box of the EDK Processor block. Add all the shared memories in the model to the processor's memory map by selecting <all> in the Available memories pulldown menu and then press the Add button. The EDK Processor block dialog box should look like the following screenshot. Dismiss the dialog box by clicking the OK button at the bottom.

Write Software Programs

You will write software programs running on MicroBlaze to read from and write to the shared memories. Re-open the XPS project in Xilinx Platform Studio. Create a new software application called MyProject. Make sure that the MyProject software application is marked for download into BRAM while the other software applications are unmarked.

Refer to the topic Using XPS

on how to add a new software application to an EDK project.

Create a new source code file MyProject.c for MyProject and open it in the XPS code editor.

The above figure shows a portion of the System Assembly View of the XPS project in Xilinx

Platform Studio. A sg_plbiface peripheral is automatically added to an XPS project after it is successfully imported into System Generator. The sg_plbiface peripheral connects the PLB bus attached to the imported MicroBlaze processor to the System Generator model through a memory-mapped interface, and to capture information on how to generate the

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

161

Chapter 2: Hardware/Software Co-Design corresponding device software drivers. Right click on sg_plbiface in the System Assembly

View to see its API documentation.

Follow the instructions in the API documentation to include the following header file and initialize the software driver in MyProject.c.

#include "sg_plbiface.h" xc_iface_t *iface;

// initialize the software driver xc_create(&iface, &SG_PLBIFACE_ConfigTable[0]);

Before reviewing the code to run on the processor, first consider how to write data to the a register on the model. Look at the DSP48 Co-Processor model. Recall that the a port of the

DSP48 block is driven by the output of a shared register by the same name. You want to write a value to that shared register from with-in MicroBlaze code. By referring to the driver API, you can see that the shared memory called a is a “To Register” memory type with xc_to_reg_t access data type, which contains the following data fields: typedef struct {

xc_w_addr_t din;

uint32_t n_bits;

uint32_t bin_pt;

} xc_to_reg_t;

Once the software driver is initialized, din stores the memory-mapped address of the din port of the shared memory a, while n_bits and bin_pt store the number of bits and binary point information.

162

So in order to write a value to the a shared register, you need to first obtain its settings through xc_get_shmem and thus:

xc_to_reg_t *toreg_a;

xc_get_shmem(iface, "a", (void **) &toreg_a);

Note:

Calling xc_get_shmem is expensive. You should cache the returned toreg_a for later use and avoid calling xc_get_shmem multiple times in a program.

You can then use the following single-word write access function to write to the a shared register:

// -- Set the a port register to 2

xc_write(iface, toreg_a->din, 2); www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

The full code of MyProject.c is shown below:

#include "xparameters.h"

#include "stdio.h"

#include "xutil.h"

// header file of System Generator Pcore

#include "sg_plbiface.h" int main (void) {

uint32_t result, overflow;

char c;

xc_iface_t *iface;

xc_from_reg_t *fromreg_result, *fromreg_overflow;

xc_to_reg_t *toreg_a, *toreg_b, *toreg_c, *toreg_instr;

// initialize the software driver

xc_create(&iface, &SG_PLBIFACE_ConfigTable[0]);

print("Performing p=2*4\r\n");

xc_get_shmem(iface, "result", (void **) &fromreg_result);

xc_get_shmem(iface, "overflow", (void **) &fromreg_overflow);

xc_get_shmem(iface, "a", (void **) &toreg_a);

xc_get_shmem(iface, "b", (void **) &toreg_b);

xc_get_shmem(iface, "c", (void **) &toreg_c);

xc_get_shmem(iface, "instr", (void **) &toreg_instr);

// -- Set the a port register to 2

xc_write(iface, toreg_a->din, 2);

// -- Set the b port register to 4

xc_write(iface, toreg_b->din, 4);

// -- Set the instr register to 2: p=a*b

xc_write(iface, toreg_instr->din, 2);

print("Press a key to read back result\n\r");

c = inbyte();

// -- Read back the result register

xc_read(iface, fromreg_result->dout, &result);

// -- Read back the overflow register

xc_read(iface, fromreg_overflow->dout, &overflow);

xil_printf("Read back: result=%d, overflow=%d\n\r",

result, overflow);

print("Performing p=c+(a:b)\r\n");

// -- Set the a port register to 262140

xc_write(iface, toreg_a->din, 262140);

// -- Set the b port register to 262140

xc_write(iface, toreg_b->din, 262140);

// -- Set the cport register to 1

xc_write(iface, toreg_c->din, 1);

// -- Set the instr register to 3: p=c+(a:b)

xc_write(iface, toreg_instr->din, 3);

print("Press a key to read back result\n\r");

c = inbyte();

// -- Read back the result register

xc_read(iface, fromreg_result->dout, &result);

// -- Read back the overflow register

xc_read(iface, fromreg_overflow->dout, &overflow);

xil_printf("Read back: result=%d, overflow=%d\n\r",

result, overflow);

return 0;

}

Copy and paste the above code into MyProject.c.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

163

Chapter 2: Hardware/Software Co-Design

Create a Hardware Co-Simulation Block

The complete Simulink model can be simulated through hardware co-simulation. Make sure that the shared memories are added into the Memory Maps window and the EDK

Processor block is configured for HDL netlisting. These are required for hardware cosimulation.

Open the dialog box of the System Generator token in the same subsystem as the EDK

Processor block. You generate the hardware co-simulation block from this level of the model so that only the imported MicroBlaze processor runs in hardware while the rest of the design is kept in Simulink for software simulation.

Under the Compilation menu select Hardware Co-simulation > ML402 > Ethernet >

Point-to-point

. Next, press the Generate button to begin the compilation process. This may take some time. Upon completion, a hardware co-simulation block is created that contains a MicroBlaze processor.

164

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

Create a Testbench Model

A testbench model will be created to use the co-simulation block created in the previous step. Open the DSP48CoProcessor model and delete the Processor Subsystem. Copy the

Processor Subsystem hwcosim block you just generated into the DSP48CoProcessor model. Save the model as DSP48CoProcessor_testbench.mdl.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

165

Chapter 2: Hardware/Software Co-Design

Update the Co-Simulation Block with Compiled Software

166

Return to the testbench model. Double click on the Processor Subsystem hwcosim block to bring up the dialog box shown above. To compile the software contained in the XPS project listed in the Software tab and load it into the hardware co-simulation bitstream, click the button labeled Compile and update bitstream.

Since Point-to-point Ethernet co-simulation is chosen, you need to configure the Ethernet interface and also the Configuration interface of the Processor Subsystem hwcosim block.

Select a valid Host interface for your Ethernet communications, and set the configure interface to Point-to-point Ethernet. Refer to the topic

Using Hardware Co-Simulation

for more usage information of the hardware co-simulation block.

Run the Simulation

Before starting the simulation, you need to set up a terminal connected to the COM port of your computer. This allows for text inputs and outputs to be read from and written to the

MicroBlaze through the RS232 port.

Open up your favorite terminal program. Windows comes with a Hyperterminal application, which can be found in Start > All Programs > Accessories > Communications

> Hyperterminal. Set up the terminal program to listen to the COM port that you have wired your RS232 to.

Configure your terminal as follows:

•Baud rate = 115200

•Data = 8 bits

•Parity = none

•Stop = 1 bit www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

•Flow control = none

Set the simulation time of the testbench model to inf, allowing enough simulation time for the MicroBlaze processor to wake up and respond.

Using XPS

This topic provides a quick tutorial on several aspects of the Xilinx Embedded

Development Kit (EDK). Please refer to the EDK documentation for more in depth explanations and tutorials.

Tutorial Example - Creating a New XPS Project

The Base System Builder is an EDK Wizard to help you construct a fully configured EDK project. This topic walks you through creating an EDK project configured with a

MicroBlaze™ Processor, running on a Xilinx ML402 hardware development board.

Launch Xilinx Platform Studio from the Windows start menu

When XPS launches, the following dialog should appear. Select Base System Builder

wizard (recommended)

, then click OK.

Next, tell Base System Builder that you wish to create a new design by selecting the I

would like to create a new design

radio button. Click Next.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

167

Chapter 2: Hardware/Software Co-Design

Base System Builder – Select Board. Select the kind of board you want to make use of, then click Next.

Base System Builder – Select a processor. Next, you are asked to select a processor.

Select MicroBlaze, then click Next.

168

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

Base System Builder – Configure MicroBlaze. You will now configure the MicroBlaze.

The Configure MicroBlase dialog box is shown below.:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

169

Chapter 2: Hardware/Software Co-Design

Base System Builder – Configure IO Interfaces. You only want the RS232 interface. Set the parameters of the RS232_Uart as shown below, then un-select the other IO interfaces

170

Base System Builder – Configure Additional IO Interfaces. The next dialog contains additional IO interfaces supported by this board. Depending on the board, there may be multiple dialog pages of "Additional IO Interfaces". At this time, you have no need for any additional IO interfaces, so un-select them all, then click Next

Base System Builder – Add Internal Peripherals. You have no internal peripherals to add. Click Next

10. Base System Builder – Software Setup. In this dialog box, you can select the input and output of STDIN and STDOUT. By default they should be routed to the RS232_Uart.

You may also select to have example code produced in the project for you. Click Next.

11. Base System Builder – System Created. This dialog shows a summary of the options selected. Click on Generate to cause the EDK project to be created

12. Base System Builder – Finished. This last dialog box shows the files generated. Click

Finish

to end the Base System Builder and allow you to continue the EDK flow using

XPS. At this point you have created a basic EDK system.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Designing with Embedded Processors and Microcontrollers

Adding a New Software Application

To add a new software application to an EDK Project, first open the EDK project in the

EDK.

In the Project Information Area, click on the Applications tab to reveal the Software

Projects page.

The first item on this page is Add Software Application Project … Double click on this to bring up the Add Software Application Project dialog box. Type in a project name, then click OK.

By default, the project is created and not set to be initialized into BRAMS. Make sure to initialized the project into BRAMS; otherwise, the software code will not be compiled and added to the bitstream. Also, if you have more than one application, ensure that all other applications have Marked to Initialize BRAM unchecked.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

171

Chapter 2: Hardware/Software Co-Design

Next, create Source or header files. Double click on the Sources branch of a project tree to cause a File Open Dialog to pop-up. The dialog is rooted at the base location of your

EDK project. It is good to create a directory named after your project and keep your source and header files there; in this case, MyProject. Create the directory in the same directory as your EDK.xmp file

Adding a pcore to an EDK Project

Pcores in an EDK project must be in the user repository, or in a directory named pcores, at the same directory level as the EDK project file

To ensure that the pcore has been loaded, from XPS, select Project > Rescan User

Repositories

Pcores in System Generator are currently FSL-based, so you may use the Configure

Co-processor

tool. The tool can be launched from XPS by selecting Hardware >

Configure Coprocessor...

172

Available FSL-based pcores are listed on the right hand window. Select the relevant pcore, then click on the Add button. The Configure Coprocessor tool takes care of connecting the clock and reset signals for the FSL bus, however, any user signals must be wired up by you.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Chapter 3 Using Hardware Co-Simulation

Introduction

System Generator provides hardware co-simulation, making it possible to incorporate a design running in an FPGA directly into a Simulink simulation. "Hardware Co-

Simulation" compilation targets automatically create a bitstream and associate it to a block.

When the design is simulated in Simulink, results for the compiled portion are calculated in hardware. This allows the compiled portion to be tested in actual hardware and can speed up simulation dramatically.

M-Code Access to Hardware Co-Simulation

It is possible to programmatically control the hardware created through the System

Generator hardware co-simulation flow using MATLAB M-code (M-Hwcosim). The M-

Hwcosim interfaces allow for MATLAB objects that correspond to the hardware to be created in pure M-code, independent of the Simulink framework. These objects can then be used to read and write data into hardware. This capability is useful for providing a scripting interface to hardware co-simulation, allowing for the hardware to be used in a scripted test-bench or deployed as hardware acceleration in M-code.

For more information of this subject, refer to the topic M-Code Access to Hardware Co-

Simulation in the section Programmatic Access.

Installing Your Hardware Platform

The first step in performing hardware co-simulation is to install and setup your hardware platform. The following topics provide Specific installation and setup instructions for

Xilinx supported platforms:

Ethernet-Based Hardware Co-Simulation

Installing an ML402 Platform for Ethernet Hardware Co-Simulation

Installing an ML506 Platform for Ethernet Hardware Co-Simulation

Installing a Spartan-3A DSP 1800A Starter Platform for Ethernet Hardware Co-Simulation

Installing a Spartan-3A DSP 3400A Development Platform for Ethernet Hardware Co-

Simulation

Note:

If installation instructions for your particular platform are not provided here, please refer to the installation instuctions that come with your Platform Kit.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

173

Chapter 3: Using Hardware Co-Simulation

JTAG-Based Hardware Co-Simulation

Installing an ML402 Board for JTAG Hardware Co-Simulation

Third-Party Hardware Co-Simulation

As part of the Xilinx XtremeDSP™ Initiative, Xilinx works with distributors and many

OEMs to provide a variety of DSP prototyping and development platforms. Please refer to the following Xilinx web site page for more information on available platforms: http://www.xilinx.com/technology/dsp/thirdparty_devboards.htm

174

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Compiling a Model for Hardware Co-Simulation

Once your hardware platform is installed, the starting point for hardware co-simulation is the System Generator model or subsystem you would like to run in hardware. A model can be co-simulated, provided it meets the requirements of the underlying hardware platform.

This model must include a System Generator block; this block defines how the model should be compiled into hardware. The first step in the flow is to open the System

Generator block dialog box and select a compilation type under Compilation.

For information on how to use the System Generator block, see

Compiling and Simulating

Using the System Generator Block

Choosing a Compilation Target

You may choose the hardware co-simulation platform by selecting an appropriate compilation type in the System Generator block dialog box. Hardware co-simulation targets are organized under the Hardware Co-Simulation submenu in the Compilation dialog box field.

When a compilation target is selected, the fields on the System Generator block dialog box are automatically configured with settings appropriate for the selected compilation target.

System Generator remembers the dialog box settings for each compilation target. These settings are saved when a new target is selected, and restored when the target is recalled.

Invoking the Code Generator

The code generator is invoked by pressing the Generate button in the System Generator block dialog box.

The code generator produces a FPGA configuration bitstream for your design that is suitable for hardware co-simulation. System Generator not only generates the HDL and netlist files for your model during the compilation process, but it also runs the downstream tools necessary to produce an FPGA configuration file.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

175

Chapter 3: Using Hardware Co-Simulation

Note:

A status dialog box (shown below) will appear after you press the Generate button. During compilation, the status box provides a Cancel and Show Details button. Pressing the Cancel button will stop compilation. Pressing the Show Details button exposes details about each phase of compilation as it is run. It is possible to hide the compilation details by pressing the Hide Details button on the status dialog box.

The configuration bitstream contains the hardware associated with your model, and also contains additional interfacing logic that allows System Generator to communicate with your design using a physical interface between the platform and the PC. This logic includes a memory map interface over which System Generator can read and write values to the input and output ports on your design. It also includes any platform-specific circuitry (e.g., DCMs, external component wiring) that is required for the target FPGA platform to function correctly.

Hardware Co-Simulation Blocks

System Generator automatically creates a new hardware co-simulation block once it has finished compiling your design into an FPGA bitstream. A Simulink library is also created in order to store the hardware co-simulation block. At this point, you can copy the block

176

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Hardware Co-Simulation Blocks

out of the library and use it in your System Generator design as you would other Simulink and System Generator blocks.

The hardware co-simulation block assumes the external interface of the model or subsystem from which it is derived. The port names on the hardware co-simulation block match the ports names on the original subsystem. The port types and rates also match the original design.

Hardware co-simulation blocks are used in a Simulink design the same way other blocks are used. During simulation, a hardware co-simulation block interacts with the underlying

FPGA platform, automating tasks such as device configuration, data transfers, and clocking. A hardware co-simulation block consumes and produces the same types of

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

177

Chapter 3: Using Hardware Co-Simulation signals that other System Generator blocks use. When a value is written to one of the block's input ports, the block sends the corresponding data to the appropriate location in hardware. Similarly, the block retrieves data from hardware when there is an event on an output port.

Hardware co-simulation blocks may be driven by Xilinx fixed-point signal types, Simulink fixed-point signal types, or Simulink doubles. Output ports assume a signal type that is appropriate for the block they drive. If an output port connects to a System Generator block, the output port produces a Xilinx fixed-point signal. Alternatively, the port produces a Simulink data type when the port drives a Simulink block directly.

Note:

When Simulink data types are used as the block signal type, quantization of the input data is handled by rounding, and overflow is handled by saturation.

Like other System Generator blocks, hardware co-simulation blocks provide parameter dialog boxes that allow them to be configured with different settings. The parameters that a hardware co-simulation block provides depend on the FPGA platform the block is implemented for (i.e., different FPGA platforms provide their own customized hardware co-simulation blocks).

178

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Hardware Co-Simulation Clocking

Selecting the Target Clock Frequency

If you are using a Xilinx ML402 or ML506 platform, System Generator allows you to choose a clock frequency for the target design that is equal to or less than the system clock frequency. The following table outlines the frequencies that are available:

Platform

Xilinx ML402

Interface

System Clock

Frequency

100 MHz

Xilinx ML506

JTAG,

Point-to-point Ethernet,

Network-based Ethernet

Point-to-point Ethernet,

Network-based Ethernet

200 MHz

Available

Frequencies

100 MHz

66.7 MHz

50 MHz

33.3 MHz

100 MHz

66.7 MHz

50 MHz

33.3 MHz

As shown below, you set the target clock frequency at compilation time, by clicking the

Settings

button on the System Generator block dialog box, then select the frequency in the pulldown menu.

1. Click

2. Select

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

179

Chapter 3: Using Hardware Co-Simulation

Clocking Modes

There are several ways in which a System Generator hardware co-simulation block can be synchronized with its associated FPGA hardware. In single-step mode, the FPGA is in effect clocked from Simulink, whereas in free-running clock mode, the FPGA runs off an internal clock, and is sampled asynchronously when Simulink wakes up the hardware cosimulation block.

Single-Step Clock

In single-step clock mode, the hardware is kept in lock step with the software simulation.

This is achieved by providing a single clock pulse (or some number of clock pulses if the

FPGA is over-clocked with respect to the input/output rates) to the hardware for each simulation cycle. In this mode, the hardware co-simulation block is bit-true and cycle-true to the original model.

Because the hardware co-simulation block is in effect producing the clock signal for the

FPGA hardware only when Simulink awakes it, the overhead associated with the rest of the Simulink model's simulation, and the communication overhead (e.g. bus latency) between Simulink and the FPGA platform can significantly limit the performance achieved by the hardware. As a general rule of thumb, as long as the amount of computation inside the FPGA is significant with respect to the communication overhead

(e.g. the amount of logic is large, or the hardware is significantly over-clocked), the hardware will provide significant simulation speed-up.

Free-Running Clock

In free-running clock mode, the hardware runs asynchronously relative to the software simulation. Unlike the single-step clock mode, where Simulink effectively generates the

FPGA clock, in free-running mode, the hardware clock runs continuously inside the FPGA itself.

In this mode, simulation is not bit and cycle true to the original model, because Simulink is only sampling the internal state of the hardware at the times when Simulink awakes the hardware co-simulation block. The FPGA port I/O is no longer synchronized with events in Simulink. When an event occurs on a Simulink port, the value is either read from or written to the corresponding port in hardware at that time. However, since an unknown number of clock cycles have elapsed in hardware between port events, the current state of the hardware cannot be reconciled to the original System Generator model. For many streaming applications, this is in fact highly desirable, as it allows the FPGA to work at full speed, synchronizing only periodically to Simulink.

In free-running mode, you must build explicit synchronization mechanisms into the

System Generator model. A simple example is a status register, exposed as an output port on the hardware co-simulation block, which is set in hardware when a condition is met.

The rest of the System Generator model can poll the status register to determine the state of the hardware.

Selecting the Clock Mode

Not every hardware platform supports a free running clock. However, for those that do, the parameters dialog box for the hardware co-simulation block provides a means to select the desired clocking mode. You may change the co-simulation clocking mode before simulation starts by selecting either the Single stepped or Free running radio button under the Clocking etch box.

180

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Board-Specific I/O Ports

Note:

The clocking options available to a hardware co-simulation block depend on the FPGA platform being used (i.e., some platforms may not support a free-running clock source, in which case it is not available as a dialog box parameter).

Board-Specific I/O Ports

FPGA platforms often include a variety of on-board devices (e.g., external memory, analog to digital converters, etc.) that the FPGA can communicate with. For a variety of reasons, it may be useful to form connections to these components in your System Generator models, and to use these components during hardware co-simulation. For example, if your board includes external memory, you may want to define the control and interface logic to this memory in your System Generator design, and use the physical memory during hardware co-simulation.

You can interface to these types of components by including board-specific I/O ports in your System Generator models. A board-specific port is a port that is wired to an FPGA pad when the model is compiled for hardware co-simulation. Note that this type of port differs from standard co-simulation ports that are controlled by a corresponding port on a hardware co-simulation block.

A board-specific I/O port is implemented using special non-memory mapped gateway blocks that tell System Generator to wire the signals to the appropriate FPGA pins when the model is compiled into hardware. To connect a System Generator signal to a board-specific port, connect the appropriate wire to the special gateway (in the same way as is done for a traditional gateway).

Non-memory mapped gateways that are common to a specific device are often packaged together in a Simulink subsystem or library. The XtremeDSP Development Kit, for example, provides a library of external device interface subsystems, including analog to digital converters, digital to analog converters, LEDs, and external memory. The interface subsystems are constructed using Gateways that specify board-specific port connections.

These subsystems are treated like other System Generator subsystems during simulation

(i.e., they perform double precision to Xilinx fixed-type conversions). When System

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

181

Chapter 3: Using Hardware Co-Simulation

Generator compiles the design into hardware, it connects the signals that are associated with the Gateways to the appropriate external devices they signify in hardware.

I/O Ports in Hardware Co-simulation

A hardware co-simulation block does not include board-specific ports on its external interface. This means that if a model includes a gateway that corresponds to a boardspecific port, the corresponding port is connected to the simulation model instead of the actual hardware when the design is compiled for hardware co-sim. To leave the port connected to a real port, use a non-memory mapped gateway instead. See the topic on nonmemory-mapped ports

Supporting New Platforms .

Ethernet Hardware Co-Simulation

System Generator provides hardware co-simulation interfaces that facilitate highthroughput communication with an FPGA platform over an Ethernet connection. These interfaces eliminate the communication range limitation imposed by programming cable solutions, while also offering superior bandwidth for real-time applications. By supporting device configuration over Ethernet, there is no need for a separate programming cable.

Two flavors of Ethernet hardware co-simulation are supported by the tool. Point-to-point

Ethernet co-simulation provides a straightforward high-performance co-simulation environment using a direct, point-to-point Ethernet connection between a PC and FPGA platform. Network-based Ethernet Co-Simulation allows communication with a remote

FPGA through the widely deployed IPv4 network infrastructure.

182

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Ethernet Hardware Co-Simulation

Point-to-Point Ethernet Hardware Co-Simulation

Point-to-point Ethernet Hardware Co-simulation provides a co-simulation interface using a raw Ethernet connection. The raw Ethernet connection refers to a Layer 2 (a.k.a. Data-

Link Layer) Ethernet connection, between a supported FPGA development board and a host PC, with no network routing equipment along the path. By taking the advantage of the ubiquity and advancement of Ethernet technologies, the interface facilitates a convenient and high-bandwidth co-simulation to an external FPGA device.

Interface Features

The interface supports 10/100/1000 Mbps half/full duplex modes. Jumbo Frame is also supported on a Gigabit Ethernet, provided it is enabled by the underlying connection. For

FPGA device configuration, the interface supports either JTAG-based configuration over a

Xilinx Parallel Cable IV or a Xilinx Platform USB cable, or Ethernet-based configuration over the same Point-to-point Ethernet connection for co-simulation.

Note:

This co-simulation interface utilizes an evaluation version of the Ethernet MAC core. Because this is an evaluation version of the core, it will become dysfunctional after continuous, prolonged operation (e.g., around 7 hours) in the target FPGA. Operation of the core will restart with a new simulation. For more information about obtaining the full version of the core, please visit the product page at http://www.xilinx.com/xlnx/xebiz/designResources/ip_product_details.jsp?key=TEMAC .

Supported FPGA Development Platforms

Development platforms that support point-to-point ethernet hardware co-simulation are listed in the topic

Ethernet-Based Hardware Co-Simulation

. Links to the appropriate platform installation instructions are also provided.

Configuring Co-Simulation Block Parameters

There are several block parameters specific to the Point-to-point Ethernet co-simulation interface. The rest of this topic describes step-by-step how to configure the parameters of the Point-to-point Ethernet co-simulation block. Refer to the topic Point-to-point Ethernet

Co-Simulation in the Xilinx Block section for details of all the block parameters.

Use the Basic tab to select the appropriate clock source for the co-simulation.

Select a Clock

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

183

Chapter 3: Using Hardware Co-Simulation

Use the Configuration tab to select the Configuration Method:

♦

For the Download cable panel, choose Point-to-point Ethernet.

For JTAG-based download cables (Parallel Cable IV or Platform USB), change the cable speed if the default value is not suitable for the cable in use.

Change the Configuration timeout (ms) value only when necessary. The default value should suffice in most cases. A larger value is needed when it takes a considerable amount of time to re-establish a network connection with the FPGA platform after device configuration completes.

If there is a Video I/O daughter card attached to the ML402 board, select Video

I/O Daughter Card (VIODC)

from the Configuration profile pulldown menu

184

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Ethernet Hardware Co-Simulation

Use the Ethernet tab to configure the Ethernet Interface Settings:

Select an Interface

♦

From the Host interface panel, use the pulldown list to select the appropriate network interface for co-simulation.

Note:

The pull down list only shows those Ethernet-compatible network interfaces installed on the host, which support 10/100/1000 Mbps, and are currently enabled and attached to an active Ethernet segment. If the target interface is not listed as expected, examine the connection and click the Refresh button to update the list.

♦

The information box beneath the pull-down list provides the details about the selected interface. Examine the information to ensure the appropriate interface is chosen, and adjust the network settings in the operating system when necessary.

Depending on which configuration method is chosen, the MAC address in the FPGA

interface

panel may need to be changed.

For Point-to-point Ethernet-based configuration:

Observe the MAC address displayed on the LCD screen of the target board when the configuration boot-loader is running. Change the FPGA MAC address in the co-

simulation block if the default value does not match the target board. Refer to Optional

Step to set the Ethernet MAC Address and the IPv4 Address

for details about assigning the MAC address on a ML402 board.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

185

Chapter 3: Using Hardware Co-Simulation

Note:

The MAC address must be specified using six pairs of two-digit hexadecimal number separated by colons (for example, 00:0a:35:11:22:33).

Co-Simulating the Design

After setting the block parameters appropriately, you may begin co-simulation by pressing the Simulink Start button. System Generator automates the device configuration process and transfers the design under test (DUT) into the FPGA device for co-simulation. A dialog box will be shown to describe the status of the process.

The final configuration file is first generated based on the input bitstream specified in the block parameters.

The final configuration file is then transferred to the target board using the selected download cable, and used to configure the FPGA device. The progress of configuration is shown in the dialog box when the configuration is performed over a

Point-to-point Ethernet connection.

Upon the completion of device configuration, the co-simulation engine re-establishes the connection to the target board, and starts co-simulating the design.

186

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Ethernet Hardware Co-Simulation

Known Issues

•

If you encounter problems transmitting data over a point-to-point Ethernet connection or experience instability issues, please disable the Hyper-Threading option in the BIOS on an Intel platform.

•

IP fragmentation is not supported by the network-based Ethernet configuration.

Please ensure the connection established between the host and the target FPGA platform can handle a maximum transmission unit (MTU) size of at least 1300 bytes without fragmentation.

Network-Based Ethernet Hardware Co-Simulation

Interface Features

The interface supports operations in 10/100/1000 Mbps half/full duplex modes. For

FPGA device configuration, the interface supports Ethernet-based configuration over the same network connection for co-simulation. This means that a separate programming cable (e.g., Parallel Cable IV) is not required.

Note:

Supported FPGA Development Boards

The Xilinx ML402 and ML506 development platform is currently supported for the network-based Ethernet co-simulation.

Setup Procedures

Network-based Ethernet co-simulation performs device configuration over the network configuration. Before using network configuration, you must ensure the IP address, MAC address, and configuration server are properly setup on the System

ACE™ CompactFlash. Refer to the topic Optional Step to set the Ethernet MAC

Address and the IPv4 Address

for information on how to do this.

The target FPGA listens on the UDP port 9999. Please ensure the underlying network does not block the associated traffic.

Known Issues

IP fragmentation is not supported by the network-based Ethernet co-simulation interface.

Please ensure the connection established between the host and the target FPGA platform can handle a maximum transmission unit (MTU) size of at least 1300 bytes without fragmentation.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

187

Chapter 3: Using Hardware Co-Simulation

Shared Memory Support

System Generator's hardware co-simulation interfaces allow shared memory blocks and shared memory block derivatives (e.g., Shared FIFO and Shared Registers) to be compiled and co-simulated in FPGA hardware. These interfaces make it possible for hardware-based shared memory resources to map transparently to common address spaces on the host PC.

When applied to System Generator co-simulation hardware, shared memories can help facilitate high-speed data transfers between the host PC and FPGA, and further bolster the tool's real-time hardware co-simulation capabilities. This topic describes how shared memories can be used within the context of System Generator's hardware co-simulation framework.

Compiling Shared Memories for

Hardware Co-Simulation

Co-Simulating Unprotected

Shared Memories

Describes how to compile a System Generator design for hardware co-simulation when the design contains shared memory blocks.

Describes how shared memory blocks configured with unprotected access mode behave during hardware co-simulation.

Co-Simulating Lockable Shared

Memories

Describes how shared memory blocks configured with lockable access mode behave during hardware co-simulation.

Co-Simulating Shared Registers

Describes how to compile a System Generator design for hardware co-simulation when the design contains shared Registers.

Co-Simulating Shared FIFOs

Describes how to compile a System Generator design for hardware co-simulation when the design contains shared FIFOs.

Restrictions on Shared Memories

Lists the restrictions that are imposed when using shared memory blocks with hardware cosimulation.

188

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Shared Memory Support

Compiling Shared Memories for Hardware Co-Simulation

System Generator allows shared memory and shared memory derivative (e.g., shared

FIFO and shared register) blocks to be compiled for hardware co-simulation. Designs that include shared memories are compiled for hardware co-simulation the same way traditional System Generator designs are compiled – by selecting a compilation target and pressing the Generate button on the System Generator dialog box. A design containing shared memory blocks may be compiled for hardware co-simulation provided the requirements described in the topic

Restrictions on Shared Memories are satisfied.

When a shared memory is compiled for hardware co-simulation, it is implemented in hardware either by a core or HDL component. The table below shows how shared memory blocks are mapped to hardware implementations

To Block From Block

Shared Memory Shared Memory

To FIFO

To Register

To FIFO

To Register

Hardware Implementation

Dual Port Block Memory 6.1

Fifo Generator 2.1

synth_reg_w_init.(vhd,v)

There are two ways in which shared memories are compiled for hardware co-simulation.

The type of compilation depends on whether the shared memory name is unique in the design, or if the shared memory has a partner who shares the same name. The following topics describe the two types of compilation behavior.

Single Shared Memory Compilation

A shared memory block is considered "single" if it has a unique name within a design.

When a single shared memory is compiled for hardware co-simulation, System Generator builds the other half of the memory and automatically wires it into the resulting netlist.

Additional interfacing logic is attached to the memory that allows it to communicate with the PC. When you co-simulate the shared memory, one half of the memory is used by the logic in your System Generator design. The other half communicates with the PC interfacing logic as shown in the figure below. In this manner, it is possible to communicate with the shared memory embedded inside the FPGA while a simulation is running.

The shared memory hardware and interface logic are completely encapsulated by the hardware co-simulation block that is generated for the design. By co-simulating a hardware co-simulation block that contains a shared memory, it is possible for your design logic and host PC software to share a common address space on the FPGA.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

189

Chapter 3: Using Hardware Co-Simulation

Note:

The name of the hardware shared memory is the same as the shared memory name used by the original shared memory block. For example, if a shared memory block uses "my_memory," the hardware implementation of the block can be accessed using the "my_memory" name.

All shared memories embedded inside the FPGA are automatically created and initialized before the start of a simulation by their respective co-simulation blocks. This means that any other shared memory objects that wish to access the hardware shared memory must specify Ownership and initialization parameter as Owned and initialized elsewhere.

Doing so causes the software-based shared memories to attach automatically to the shared memories that reside inside the FPGA.

Compiling Shared Memory Pairs

It is also possible to compile a shared memory pair (i.e., two shared memories that specify the same name) for hardware co-simulation. In this case, the two shared memory halves are merged into a single hardware implementation during compilation. Unlike single shared memories, both sides of a shared memory pair connect to System Generator user design logic. For example, the figure below shows the hardware implementation for a To /

From FIFO shared memory pair.

190

Note that because both sides of the shared memory connect to user design logic, it is not possible to communicate with these shared memories directly from the host PC. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Shared Memory Support

Viewing Shared Memory Information

Hardware co-simulation blocks allow you to view information about the shared memories that were compiled as part of the design. A hardware co-simulation block that contains shared memories will have an enabled Shared Memories tab in the block configuration dialog box shown below. Clicking on this tab exposes a table of information about each shared memory in the design.

The shared memory information table describes the type, bit width, and depth of each shared memory in the design. For Shared Memory blocks, it also specifies the Access

Protection

mode. Clicking on the [+] or [-] symbol next to the shared memory icon expands or collapses the shared memory table, respectively.

The icons associated with each shared memory type are shown in the table below.

Memory Type Icon

Shared Memory

Shared FIFO

Shared Register

Co-Simulating Unprotected Shared Memories

Unprotected shared memory blocks may be written to or read from at any time – this type of memory has no notion of mutually exclusive access. Data transfers to and from an unprotected hardware shared memory occur a single-word at a time, unlike the highspeed data transfer mode used by lockable shared memories. To ensure data coherency between software and hardware, a single image of the shared memory data is shared between hardware and software. This image is stored in the FPGA using dual port memory. System Generator allows both hardware design logic and other software-based shared memory objects on the host PC to access the shared memory data concurrently.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

191

Chapter 3: Using Hardware Co-Simulation

When software shared memory objects read or write data to the shared memory, a proxy seamlessly handles communication with the hardware memory resource.

The following figure shows an example of unprotected shared memory implemented in the FPGA that is communicating with three shared memory objects running on the host

PC. In this example, the software shared memory objects access the hardware shared memory by specifying the same shared memory name, my_mem. From the perspective of the software shared memories, the implementation of the shared memory resource is irrelevant; the hardware shared memory is treated as any another shared memory object.

Read and writes to the shared memory are handled by the shared memory API.

Note:

Not all shared memory objects need to be created or executed in the Simulink environment.

The C++ application in the figure below is just one example of an external application that may communicate with the hardware shared memory data using the shared memory API.

192

Co-Simulating Lockable Shared Memories

In lockable access mode, the System Generator co-simulation hardware must acquire lock over the shared memory object before it may access its contents. When the hardware acquires (releases) lock of the shared memory, the memory contents are transferred to

(from) the FPGA using a high-speed data transfer. Using this methodology, it is possible to implement System Generator hardware co-simulation designs with high memory bandwidth requirements. For more information on how to do this, refer to the tutorial entitled

Real-Time Signal Processing using Hardware Co-Simulation

Unlike unprotected shared memories, two images of the shared memory data are used when a lockable shared memory is co-simulated. One memory image is stored using dual port memory in the FPGA. This image is accessed by the System Generator hardware cosimulation design and co-simulation interfacing logic. The other image is implemented as a shared memory object on the host PC. This software shared memory image is accessed by any software shared memory objects used in a design.

In lockable mode, a software process or hardware circuit that wishes to access the shared memory must first obtain the lock. If the hardware has lock of the memory, no software objects may access the memory contents. Likewise, if a software object controls the memory, the hardware cannot read or write to the memory. Note that lockable hardware www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Shared Memory Support

shared memories include additional logic to handle the mutual exclusion. The interaction between hardware and software lockable shared memories is shown in the figure below: .

The red circle in the figure above represents a lock token. This token may be passed to any shared memory object, regardless of whether it is implemented in hardware or software.

The dashed circle represents lock placeholders and signifies that lock can be passed to the block it is associated with. The diamond in the figure above represents a modifiable token.

This token illustrates that when hardware has lock of the memory, the hardware shared memory image may be modified. Likewise, when a software shared memory object has lock, the software shared memory image may be modified.

Having two shared memory images requires synchronization between software and hardware to ensure the images are coherent. This synchronization is accomplished by transferring the memory image between software and hardware upon lock transfer.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

193

Chapter 3: Using Hardware Co-Simulation

System Generator performs high speed data transfers between the host PC and FPGA. The semantics associated with these transactions are shown in the figure below. .

194

Co-Simulating Shared Registers

A To Register , From Register or shared register pair may be generated and co-simulated in

FPGA hardware. Here and throughout this topic, a shared register pair is defined as a To

Register block and From Register block that specify the same name (e.g., 'Bar'). In hardware, a shared register is implemented using a synthesizable register component (for

VHDL) or a module (for Verilog). This topic explains how single shared registers and shared register pairs behave during hardware co-simulation.

When a design that includes a shared register pair is compiled for hardware co-simulation, the pair is replaced by a single register instance. Both sides of the register attach to user design logic; that is, logic that originated from the original System Generator model.

Unlike designs compiled using the Multiple Subsystem Generator block, all ports on the hardware register attach to signals in the same clock domain. In this case, control of the register is not shared between the PC and FPGA hardware since all register ports are attached to user design logic. Compiling a shared register pair into hardware is equivalent to compiling a System Generator Register or Delay block.

Compiling a single To Register or From Register block for hardware co-simulation results in a different type of implementation. A single register is still created to replace the To or

From Register block. Only in this case, the register connects to both the PC interface and

FPGA logic. The side of the register in the original model remains connected to user design logic. The other side of the register attaches to data and control ports that interface with the

PC.

For example, in the following figure, when a From Register block is compiled for hardware co-simulation, the dout register port remains attached to the user design. The din, ce, and

clk

System Generator for DSP

Release 10.1.3 September, 2008

Shared Memory Support

is possible for the PC to write to the register using System Generator's hardware cosimulation interfaces.

When a To Register block is compiled for hardware co-simulation, as shown in the figure below, the input ports are wired to user logic while the output port is wired to PC interface logic. You may access a shared register during hardware co-simulation using the other half of the shared register (i.e., using a To or From Register block), a C program or executable

(System Generator API), or a MATLAB program.

For designs that use hardware co-simulation, shared register pairs are typically distributed between software and FPGA hardware. In other words, one half of the pair is implemented in the FPGA while the other half is simulated in software using a To or From Register block. When data is written to a software To Register block, the hardware register is updated to with the same data. Similarly, when data is written into the hardware register, the same data is read by the From Register software block. A software shared register may connect to a hardware shared register simply by specifying the name of the shared register as it was compiled for hardware co-simulation.

Note:

You may find the names of all shared memories embedded inside an FPGA co-simulation design by viewing the Shared Memories tab on a hardware co-simulation block.

When a software / hardware shared memory pair is co-simulated, System Generator transparently manages the interaction between the PC and FPGA hardware. This means that a shared register pair simulated in software should behave the same as a shared register pair distributed between the PC and FPGA hardware.

Co-Simulating Shared FIFOs

A To FIFO , From FIFO or shared FIFO pair may be generated and co-simulated in hardware.

Here and throughout this topic, a shared FIFO pair is defined as a To FIFO block and From

FIFO block that specify the same name (e.g., 'Bar'). In hardware, a shared FIFO is implemented using the FIFO Generator core. The core is configured to use independent

(asynchronous) clocks, and block memory for data storage. This topic explains why cosimulating shared FIFOs is useful, and also how these blocks behave in hardware.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

195

Chapter 3: Using Hardware Co-Simulation

Asynchronous FIFOs are typically used in multi-clock applications to safely cross clock domain boundaries. When a

Free-Running Clock

mode is used for hardware cosimulation, the FPGA operates asynchronously relative to the Simulink simulation. That is, the FPGA is not kept in lockstep with the simulation. Using the Free-Running Clock mode effectively establishes two clock domains: the Simulink simulation clock domain and the

FPGA free-running clock domain. In these designs, Shared FIFOs provide a reliable and safe way to transfer data between the host PC and FPGA platform.

Shared FIFOs may also be used to support burst transfers during co-simulation. It is possible to create vectors or frames of data, and transfer the data to the FPGA in a single transaction with the hardware. These interfaces can be used to further accelerate simulation speeds beyond what is typically possible with hardware co-simulation. For more information on how this is accomplished, refer to the topic

Frame-Based Acceleration using Hardware Co-Simulation

When a shared FIFO pair is generated for co-simulation, a single asynchronous FIFO core replaces the two software shared FIFO blocks. As shown in the figure below, the read / write FIFO sides are attached to user design logic (i.e., logic derived from the original

System Generator model) that attached to the From FIFO and To FIFO blocks. Because both

FIFO sides attach to user logic in hardware, the PC does not share control of the FIFO with the design. Instead, the FIFO behavior is similar to a System Generator design that includes a traditional FIFO block.

196

Note that even though the FIFO exposes independent clock ports, the same co-simulation clock drives both ports when a FIFO pair is compiled. This is different from compiling a shared FIFO pair using the Multiple Subsystem Generator block, where the clocks are from distinct clock domains.

Single shared FIFO blocks are treated differently than shared FIFO pairs. A single To FIFO or From FIFO block is replaced by an asynchronous FIFO core when it is compiled for hardware co-simulation. One side of the FIFO (i.e., the unused shared FIFO half in System

Generator) is connected to PC interface logic. The other side is connected to user design logic that attached to the original To or From FIFO block. In this manner, control over the

FIFO is distributed between the PC and FPGA design.

As shown in the following figure, when a To FIFO block is compiled for hardware cosimulation, the write side of the FIFO is connected to the same logic that attached to To www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Shared Memory Support

FIFO block in user design. The read side of the FIFO is connected to PC interface logic that allows the PC to read data from the FIFO during simulation.

In the figure below, the opposite wiring approach is used when a From FIFO block is compiled for hardware co-simulation. In this case, the write side of the FIFO is connected to PC interface logic, while the read side is connected to the user design logic. The host PC writes data into the FIFO and the design logic may read data from the FIFO.

For designs that use hardware co-simulation, shared FIFO pairs are typically distributed between software and FPGA hardware. In other words, one half of the pair is implemented in the FPGA while the other half is simulated in software using a To or From FIFO block.

Together, the software and hardware portions form a fully functional asynchronous FIFO.

When a software / hardware shared FIFO pair is co-simulated, System Generator transparently manages the necessary transactions between the PC and FPGA hardware.

When data is written to a software To FIFO block during simulation, the same data is written to the FIFO in hardware. The design in hardware may then retrieve this data by reading from the FIFO. Similarly, when data is written into the hardware FIFO by design logic, the data may be read by the From FIFO software block. Note that the empty, full, read and write count ports on the shared FIFO blocks pessimistically reflect the state of the hardware FIFO counterpart. A software shared FIFO may connect to a hardware shared

FIFO simply by specifying the name of the shared FIFO as it was compiled for hardware co-simulation.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

197

Chapter 3: Using Hardware Co-Simulation

Note:

You may find the names of all shared memories embedded inside an FPGA co-simulation design by viewing the Shared Memories tab on a hardware co-simulation block.

Restrictions on Shared Memories

The following restrictions apply to System Generator designs that use shared memory, register, or FIFO blocks in conjunction with hardware co-simulation:

•

The access protection mode of a shared memory may not be modified once it has been compiled for hardware co-simulation.

•

Shared memory address port widths are limited to 24-bits (or less), allowing an address space of 16,777,216 words;

•

Shared memory, register, and FIFO data port widths are currently limited to 32-bits or less.

•

Shared memories and FIFOs are implemented in hardware using block memories; neither distributed nor external memory implementations are currently supported.

•

No more than two shared memories with the same shared memory name may be compiled for hardware co-simulation.

•

Two or more hardware co-simulation blocks that have shared memory names in common may not concurrently be used in the same design.

Specifying Xilinx Tool Flow Settings

When a design is compiled for System Generator hardware co-simulation, the command line tool, XFLOW, is used to implement and configure your design for the selected FPGA platform. XFLOW defines various flows that determine the sequence of programs that should be run on your design during compilation. There are typically multiple flows that must be run in order to achieve the desired output results, which in the case of hardware co-simulation targets, is a configuration bitstream.

System Generator uses two flows, implementation and configuration, in order to produce a configuration bitstream. The implementation flow is responsible for compiling the synthesis tool netlist output (e.g., EDIF or NGC) into a placed and routed NCD file. To accomplish this, it runs the Xilinx tools NGDBuild, MAP, and PAR. The implementation flow can also execute TRACE (for timing analysis purposes), although this program is typically omitted in order to expedite the compilation process. The configuration flow runs the tools necessary to create an FPGA bitstream, using the fully elaborated NCD file as input.

The implementation and configuration flow types have separate XFLOW options files associated with them. An XFLOW options file declares the programs that should be run for a particular flow, and defines the command line options that are used by these tools. Each hardware co-simulation compilation target provides options files that define the default configuration options for these tools. Sometimes you may want to use options files that use settings that differ (e.g., to specify a higher placer effort level in PAR) from the default options provided by the target. In this case, you may create your own options files, or edit the default options files to include your desired settings.

198

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Specifying Xilinx Tool Flow Settings

The Hardware Co-Simulation Settings dialog box, shown below, allows you to specify options files other than the default options files provided by the compilation target.

Parameters available on the Hardware Co-Simulation Settings GUI are:

•

Implementation Flow

: Specifies the options file that is used by the implement flow type. By default, System Generator will use the implement options file that is specified by the compilation target.

•

Configuration Flow

: Specifies the options file that is used by the config flow type. By default, System Generator will use the config options file that is specified by the compilation target.

The Xilinx ISE™ software includes several example XFLOW options files. From the base directory of your Xilinx ISE software tree, these files are located under the directory xilinx\data

. Three commonly used implementation options files include:

• balanced.opt

• fast_runtime.opt

• high_effort.opt

Note:

It is possible to define options files that may cause errors in the System Generator hardware co-simulation flow. As a result, it is typically a good idea to make backup copies of the default options files before modifying them. In addition, the configuration options file should be edited with caution, as most FPGA hardware platforms have specific configuration parameter requirements.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

199

Chapter 3: Using Hardware Co-Simulation

Frame-Based Acceleration using Hardware Co-Simulation

With the tremendous growth in programmable device size and computational power, lengthy simulation times have become an expected, yet undesirable part of life for most engineers. Depending on the design size and complexity, the required simulation time can be quite large, sometimes on the order of days to run to completion. This problem is exacerbated by the fact that most systems must be simulated many times before the design is considered functional and ready for deployment. Fortunately, System Generator for DSP provides hardware co-simulation interfaces that allow you to dramatically accelerate simulation speeds of your FPGA designs.

There are several factors that influence exactly how much acceleration can be gained by using hardware co-simulation. These considerations include the size of the design, the number of ports on the model, and the hardware over-sampling rate. Under normal operation, the PC communicates with the FPGA during each Simulink simulation cycle.

These software / hardware transactions often involve significant overhead and can end up being the limiting factor in simulation performance. Also of importance is the cosimulation interface being used. Some interfaces (e.g., PCI) are faster than others (e.g.,

JTAG). For a reasonably large design, the typical simulation is accelerated by an order of magnitude when co-simulated in hardware.

Keeping the above points in mind, there are ways to further bolster simulation performance. Remember that every time the PC interacts with hardware, there is an overhead cost that impacts simulation performance. One of the ways the number of FPGA transactions can be mitigated is by utilizing Simulink vector and frame signal types. Here and throughout the rest of this tutorial, FPGA transactions involving Simulink vector and frame signals as simply referred to as vector transfers. This idea is straightforward – bundle as many input data samples together as possible and have the FPGA process the data in a single transaction. Fewer transactions with the FPGA results in better simulation performance.

In this tutorial, Simulink vector and frame signals are used to increase simulation performance beyond what is traditionally possible with hardware co-simulation. A stepby-step example filter design is presented to help illustrate these concepts..

Before diving into the details, it is worth exploring exactly what you are trying to accomplish from a high-level perspective. In summary, you will do the following during a

Simulink simulation cycle:

•

Buffer a series of scalar input data values into a Simulink vector;

•

Transfer the vector data to a buffer residing on the FPGA using a burst transfer;

•

Use the FPGA, in free-running clock mode, to sequentially process the entire input buffer;

•

Use the FPGA to write the data into an output buffer;

•

Transfer the contents of the output buffer back into Simulink and reconstruct the data as a Simulink vector;

•

Unbuffer the vector into a series of output scalar values.

Shared Memories

Before a System Generator design can support vector transfers, it must be augmented with appropriate input and output buffers. In hardware, these buffers are implemented using internal memory (e.g., BRAMs) and are used to store vectors of simulation data that are written to and read from the FPGA by the PC. This means that the maximum size of the

200

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Frame-Based Acceleration using Hardware Co-Simulation

buffers is limited by the amount of internal memory available on the target device. In

System Generator, shared memory blocks provide interfaces that implement such buffers.

A question that quickly comes to mind is why not use standard FIFO or memory blocks?

The buffers required for hardware co-simulation differ from traditional FIFOs and memories in that they must be controllable by both the PC and FPGA user design logic.

The standard FIFO and memory blocks provided by System Generator can only interface with user design logic.

There are two types of shared memories that provide this control: lockable shared memories and shared FIFOs. These blocks provide different buffering styles; each with their own handshaking protocols that determine when and how burst transactions with the FPGA occur. In this tutorial, primary attention is focused on shared FIFO buffers. For an example on how to use lockable shared memories, please refer to the tutorial entitled

Real-Time Signal Processing using Hardware Co-Simulation . You may find the lockable

shared memory and FIFO blocks in the Shared Memory library of the Xilinx Blockset.

Because shared FIFOs play a central role in enabling vector transfers, it is worth a brief aside to discuss their behavior. A shared FIFO pair is comprised of a To FIFO block and a

From FIFO block that specify the same name (e.g., Bar in the figure above). The To FIFO block provides the "write side" control signals, while the From FIFO block provides the

"read side" control signals. When used together, a shared FIFO pair is conceptually the same thing as a single FIFO – only the control signals for the two sides are graphically disjoint. This means that a shared FIFO pair shares the same FIFO memory space. For example, if you write data into a To FIFO block, you may retrieve the same data by reading from the From FIFO block. The connection between these two blocks is implicit; shared

FIFOs are associated with one another by name and not by explicit Simulink wires.

Shared FIFOs and shared memories in general may be compiled for hardware cosimulation. Note that although this tutorial touches briefly on how shared FIFOs are cosimulated, it is useful to refer to the topic titled

Co-Simulating Shared FIFOs

for more indepth information. When one-half of a shared FIFO block is compiled for hardware cosimulation, a full FIFO block is embedded in the FPGA using the FIFO Generator core. One side of the FIFO connects to user design logic (i.e., the System Generator logic that connected to the shared FIFO block). The other half connects to interface logic that allows it to be controlled by the PC. This side of the FIFO may be controlled by other System

Generator software model logic (e.g., the half of the shared FIFO), by a C program or software executable, or by a MATLAB program. By compiling shared FIFOs for hardware

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

201

Chapter 3: Using Hardware Co-Simulation co-simulation, you create embedded FIFO-style buffers in the FPGA that can be controlled directly by a PC.

202

There are several ways to communicate with a shared FIFO that is embedded inside the

FPGA. The most common approach is to include the other half of the shared FIFO in the

System Generator design. It is also possible to communicate with the shared FIFO using a

C program or MATLAB program. System Generator provides additional blocks that support vector transfers to and from the FIFO. These blocks will be touched on later in the tutorial as they play a key role in supporting burst transfers to and from the FPGA.

Adding Buffers to a Design

Having gained an understanding of how shared FIFOs work in hardware, you will now turn you attention towards building designs that can utilize these buffers for high-speed vector processing in the FPGA.

Consider the scenario in which you have an FPGA data path that you would like to accelerate using vector transfers. You need to include input buffer storage in the FPGA that can store data input samples that are written by the PC. An output buffer is also required so that the processed data values can be stored while the FPGA waits for the PC to retrieve them. With these requirements in mind, a From FIFO block is used to implement the input data buffer and a To FIFO block is used to implement the output data buffer. In the model shown below, data is written into the data path as soon as it shows up in the input FIFO.

Note that the data path block contains new data (nd) and data valid (vld) flow control ports. These ports support a simple flow control scheme that determines when new data enters and valid data leaves the data path. The nd signal is asserted whenever there is data www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Frame-Based Acceleration using Hardware Co-Simulation

available in the input FIFO. Conversely, data is written into the output FIFO whenever valid data is present on the data path.

To gain a better understanding of how the Shared FIFOs are used, you will now take a look at an example design that uses vector transfers to accelerate a MAC filter design.

From the MATLAB console, change directory to

<sysgen_tree>/examples/shared_memory/hardware_cosim/frame_acc

Open macfir_sw_w_fifos.mdl from the MATLAB console.

The example design implements a 32-tap MAC FIR filter that removes additive white noise from a sinusoid input source. The amount of white noise can be adjusted interactively by moving the Slider Gain control bar before or during simulation. An output scope compares the filtered output data against the unfiltered input data. The MAC filter itself is contained inside a subsystem named hw_cosim. This subsystem contains all of the logic that will be

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

203

Chapter 3: Using Hardware Co-Simulation compiled into the FPGA for hardware co-simulation. You consider everything else in the design (i.e., all blocks in the top-level) as the design testbench.

204

Pushing into the hw_cosim subsystem, you have an n-tap MAC FIR Filter block that implements the design data path. Wrapping the filter are From FIFO and To FIFO blocks that provide the input and output buffers, respectively. The MAC filter in the example design is a modified version of the n-tap MAC filter available in the System Generator DSP

Reference Blockset library. In the example, the filter is modified to include valid in and valid out ports in order to support the FIFO flow control scheme.

In total, there are four shared memory blocks in the design that define the CA and VA shared FIFO pairs. In truth, you only need the shared FIFO blocks contained inside the hw_cosim subsystem to successfully compile the design for hardware co-simulation.

Because you would like to simulate the complete design, including FPGA hardware, you include a CA To FIFO block and VA From FIFO block in the testbench logic. These shared

FIFO blocks are responsible for writing and reading test data from the shared FIFOs in the hw_cosim subsystem.

Unfiltered data from the din Gateway In block is written into the CA To FIFO block. At this point, the CA From FIFO block in the hw_cosim subsystem reads data from the FIFO and writes it into the MAC filter. The MAC filter in turn processes the data and writes it into the output buffer, represented by the VA To FIFO block. Lastly, the VA From FIFO block in the top-level reads the data and sends it to the Scope block for visualization.

For this example, you have chosen a maximum buffer size of 4K. This parameter is set by specifying 4K for the Depth parameter on the CA From FIFO and VA To FIFO block dialog boxes. Note that because shared FIFOs are implemented using asynchronous FIFO www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Frame-Based Acceleration using Hardware Co-Simulation

Generator cores, the actual depth of the hardware FIFO is n-1 words, where n is the depth specified on the dialog box.

You will now have a chance to simulate the design to see how fast it runs in software.

Press the Simulink Start button to simulate the design in software.

Record the time required to simulate the design for 10000 cycles. To get an accurate measurement, it is preferable to leave the scope block closed since the graphic updates may affect simulation performance.

You may adjust the Slider Gain bar during simulation to see how the presence of additional noise affects the filter performance. You may view the filtered and unfiltered data in the output scope block. The top axis shows the unfiltered input data. The bottom axis shows the filtered data results.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

205

Chapter 3: Using Hardware Co-Simulation

Compiling for Hardware Co-simulation

You will now compile the design for hardware co-simulation. Before performing the following steps, ensure that you have an appropriate hardware co-simulation platform installed in System Generator and attached to your PC. In this example, you only want to compile the portion of the design that resides inside the hw_cosim subsystem. This is because you want the CA To FIFO and VA From FIFO blocks to remain in software as part of the design testbench (while their partner shared FIFOs are compiled into FPGA logic).

Double-click on the System Generator block in the hw_cosim subsystem to open the

System Generator dialog box.

From the Compilation submenu, choose an appropriate hardware co-simulation target. Note that although you use the Point-to-point Ethernet hardware co-simulation interface in this example, any installed hardware co-simulation platform (e.g., a board that supports JTAG co-simulation) will suffice.

206

Press the Generate button on the System Generator dialog box to generate the design.

A new hardware co-simulation library and block are created once System Generator finishes compiling the design. Note that the new hardware co-simulation block does not have any input or output ports. This is because the subsystem that was compiled did not contain gateway blocks or Simulink ports. Instead, all connections to other Simulink blocks are handled implicitly through shared memories that were compiled into the FPGA.

Because you left the To FIFO and From FIFO blocks as part of the software testbench, the software FIFOs will automatically attach to the FIFOs in hardware at the beginning of simulation.

It is often necessary to examine the type and configuration of a shared memory that was compiled for hardware co-simulation. The information about each shared memory is available in a Shared Memories tab on the hardware co-simulation block dialog box. This tab contains a tree view of information about each shared memory embedded in the design.

Double-click on the hardware co-simulation block to open the parameters dialog box.

Select the Shared Memories tab in the hardware co-simulation block dialog box.

The tree-view contains information about the CA and VA shared FIFO blocks that were compiled. If your co-simulation design contains other shared memory blocks, information about these blocks will also be displayed here. You may expand or collapse shared www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Frame-Based Acceleration using Hardware Co-Simulation

memory information by clicking on the (+) or (-) icons located adjacent to the shared memory icons.

10. Close the parameters dialog box.

You are now ready to insert the hardware co-simulation block in the original design. Before continuing on with the next steps, it is worthwhile to either rename the design or create a backup of the original since you will be making modifications.

11. Remove the hw_cosim subsystem from the design.

12. Insert the hardware co-simulation block in place of the hw_cosim subsystem.

13. Configure the hardware co-simulation block with any settings necessary to cosimulate using single-step clock mode.

14. Press the Simulink Start button to start the design.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

207

Chapter 3: Using Hardware Co-Simulation

15. Record the amount of time required to simulate the design for 10000 cycles.

16. Close the design, but leave the hardware co-simulation library open since you will need it in the next topic.

In the simulation above, hardware co-simulation uses single word transfers. That is, whenever there is a new simulation value to be read or written to the hardware cosimulation, the PC initiates a transaction with the FPGA. The next topic describes how vector transfers may be used to increase simulation speed by making more efficient use of the available hardware co-simulation bandwidth.

Using Vector Transfers

The System Generate Shared Memory Read and Write blocks allow you to use vector transfers with hardware co-simulation. These blocks may be found in the Shared Memory library in the Xilinx Blockset.

The Shared Memory Write block accepts a Simulink scalar, vector, matrix or frame data type and writes the data sequentially into a shared memory. The complete contents of the

Simulink signal are written into the shared memory in a single simulation cycle. As is the case with all shared memory blocks, an association is made between a Shared Memory

Read or Write block and another shared memory by specifying the same shared memory name.

Matrix types are treated as having a column-major order. That is, when data is written sequentially into a shared memory, the elements in a column are written first before advancing to the next column. For example, assume you have the matrix of data shown below. During simulation, this matrix data is written into the FIFO (or shared memory) in the following order:

208

Using these blocks, it is possible to read or write full vector, frame, or matrix signals into shared memories, provided the following conditions are met:

•

The input signal driven to a shared memory write block is an 8-bit, 16-bit, or 32-bit signed or unsigned integer;

•

The number of elements in the vector or matrix does not exceed the depth of the shared memory or FIFO.

•

The data width of the Shared Memory Read or Write block (i.e., the bitwidth of the scalar, or vector or matrix element) equals the shared memory or FIFO data width.

You can use these blocks in the example design to read and write vectors of data samples to the MAC filter in a single software / hardware transaction.

17. Open macfir_hw_w_frames_tb.mdl from the MATLAB console. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Frame-Based Acceleration using Hardware Co-Simulation

This design is a very similar to the previous design, with a few modifications made to support the Shared Memory Read and Write blocks. Before simulating the design, you consider each of these modifications. Most importantly, Shared Memory Read and Write blocks have been substituted in place of the To and From FIFO testbench blocks in the previous design. By specifying CA and VA as the Write and Read shared memory names, respectively, an association is automatically made to the input and output FIFO buffers in the FPGA hardware during simulation.

A Simulink Buffer block builds a frame of scalar input samples by sequentially buffering the unfiltered input data. A simple analogy is that the Buffer block is performing a serial to parallel conversion. Recalling that you compiled the FIFO buffers with a depth of 4K, you choose a frame size of 4095.

Note that the buffer block introduces a sample rate change in the design. For every 4095 inputs, there is only one output. Thus if the data input sample period is 1, the buffer data output sample period is 4095. This means that the Shared Memory Write block need only send a new frame of data to the FPGA on every 4095th simulation cycle (which is considerably more efficient than initiating a hardware transaction during every simulation cycle).

Because the Buffer block introduces a rate change, you must adjust the downstream blocks to accommodate the slower sample period. You begin by telling the Shared Memory Read block to read a frame of data every 4095th simulation cycle.

18. Double-click on the Shared Memory Read block to open its parameters dialog box.

On the Type field under the Basic tab, you have configured the block to use shared FIFOs.

To ensure a new frame is read at the appropriate time, you configure the Shared Memory

Read block with a Sample time value of 4095.

The Shared Memory Read block allows you to specify the output data type and dimensions.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

209

Chapter 3: Using Hardware Co-Simulation

19. On the parameters dialog box, switch to the Output Type tab.

There are several things of interest on this tab. First, you set the output data type as an int32 to match the filter data path output width of 32-bits. Note the design will not simulate unless these widths match. Secondly, you choose an output dimension that is 4095 words deep in the Output dimensions field. Finally, you tell the block to generate frame-based output since frame data types are required by the downstream Unbuffer block.

20. Close the parameters dialog box.

The Simulink Unbuffer block takes the frame data from the Shared Memory Read block and deserializes it into sequential scalar values. The Simulink Unbuffer block also introduces a sample rate change in the diagram. Because the input sample period to the block is 4095, and the frame size is 4095 words, the Unbuffer block output sample period is

1. This works out nicely since you have data moving through the overall system at an effective sample period of 1.

Because the Shared Memory Write and Read block operate on integer values, you must insert Simulink type conversion blocks into the diagram so that the data is interpreted correctly in various portions of the model. The in_data_conv subsystem converts the

Simulink doubles into 16-bit integer values that can be interpreted appropriately by the

FPGA hardware. On the output side, the out_data_conv subsystem converts the 32-bit integers into 32-bit Simulink fixed-precision values.

Before simulating the design, you must add the hardware co-simulation block you created from the previous design.

210

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Frame-Based Acceleration using Hardware Co-Simulation

21. Add the hardware co-simulation block to the design as shown below.

As mentioned before, the Shared Memory Write block writes a new input frame of 4095 words to the FPGA on every 4095th clock cycle. Likewise, the Shared Memory Read block reads an output frame of 4095 words from the FPGA on every 4095th clock cycle. This means that the FPGA must process the entire frame in a single-cycle. How exactly is this accomplished?

The first step is to configure the FPGA in free-running clock mode. In doing so, you allow the FPGA to process data considerably faster than if it were otherwise kept in lockstep with the Simulink simulation. Whereas in single-step mode the FPGA can only process one data per Simulink cycle, the FPGA processing speed is limited only by the system clock frequency when operating in free-running clock mode. Even so, if the buffer is large enough, the FPGA may not have time to process the complete buffer before the next block in the design is woken up. You still need a way to stall the rest of the simulation while the

FPGA processes the entire buffer.

The Shared Memory Read block checks the number of FIFO words available in the output buffer before trying to read a frame. If the number of words in the buffer is insufficient, the

Read block waits for a small amount of time, and then checks again to determine if the words have become available. It only reads the frame once all of the words are available in the output buffer, in this case 4095. In this manner, the Shared Memory Read block can stall the simulation until the complete frame has been processed by the FPGA.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

211

Chapter 3: Using Hardware Co-Simulation

The simulation flow of data through the diagram is shown below.

Two steps necessary to run the simulation using Simulink frames signals are provided below:

22. Double-click on the hardware co-simulation block to bring up the parameters dialog box.

23. Select Free running clock mode as shown below.

212

24. Configure the hardware co-simulation block with any additional settings necessary for simulation according to the requirements of your co-simulation platform.

25. Press the Simulink Start button to start the design.

26. Record the amount of time required to simulate the design for 10000 cycles.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Real-Time Signal Processing using Hardware Co-Simulation

The shared memory interfaces available in System Generator allow signal processing designs with high bandwidth and memory requirements to be co-simulated using FPGA hardware. When used in conjunction with the Xilinx Shared Memory Read and Write blocks, it is possible for hardware co-simulation designs to process complete Simulink vector and matrix signals in a single simulation cycle. These large data transactions between Simulink and the FPGA are realized using burst transfers, and depending on the co-simulation interface, often provide sufficient throughput for real-time signal processing applications.

There are two types of System Generator interfaces that support burst transfers when compiled into FPGA hardware. These interfaces include lockable shared memories and shared FIFO blocks. Both blocks provide different handshaking protocols that determine how and when transactions between the FPGA and host PC occur. Before using these blocks, it is useful to understand how they work in relation to hardware co-simulation. For more information, please refer to the following topics:

Co-Simulating Lockable Shared Memories

Co-Simulating Shared FIFOs

In this document, a high-speed co-simulation buffering interface implemented as a System

Generator model is presented. The example interface uses lockable-shared memories to implement the required buffer storage. Note that it is relatively straightforward to modify the flow control logic so that shared FIFOs may be used in place of the shared memories.

The high-speed buffering interface is discussed first, followed by an example in which the interface is used to support real-time processing of a video stream using a 5x5 filter kernel.

Described last is how an additional unprotected shared memory is applied to the system to support dynamic reloading of the image kernel during co-simulation.

Shared Memory I/O Buffering Example

When a lockable shared memory is compiled for hardware co-simulation, additional circuitry is included in the FPGA to the handle the mutual exclusion. Part of this circuitry includes logic to enable high-speed transfers of the memory image when the FPGA acquires or releases lock of the memory. It takes advantage of the lockable shared memory mutual exclusion semantics to implement a high-speed I/O buffering interface for hardware co-simulation. This topic describes this interface, which is included as an example model in your System Generator software installation.

From the MATLAB console, change directory to

<sysgen_tree>/examples/shared_memory/hardware_cosim/io_bufferin

Open highspeed_iobuf_ex.mdl from the MATLAB console.

The I/O buffering interface allows you to easily buffer and stream data through a System

Generator signal processing data path during hardware co-simulation. The example design is comprised of two subsystems that implement input and output buffer storage, named Input Buffer and Output Buffer, respectively. The turquoise block in the center of the diagram is a placeholder for the signal processing data path which you will substitute into the design.

At the heart of each buffering subsystem is a lockable shared memory block that provides the buffer storage. Each shared memory is wrapped by logic that controls the flow of data

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

213

Chapter 3: Using Hardware Co-Simulation from the host PC, through the interface, and back to the host PC. Operation of the I/O buffering interface is shown in the flow chart below:

214

Notice that the buffering interface design includes several data valid ports. These ports are used for data flow control. A "true" output from the Input Buffer dout_valid port indicates new data is ready to be processed by the data path. Likewise, when the data path is finished processing the data, it should drive the Output Buffer subsystem's din_valid

port to "true" to indicate valid output data (the din_valid port is analogous to a write enable control signal). www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Real-Time Signal Processing using Hardware Co-Simulation

The example includes a placeholder that should be replaced by a System Generator data path. You may insert any data path in the buffer interface provided that it works within the valid signal semantics described above.

Note:

The output buffer shared memory does not release lock until the output buffer is full. To avoid deadlock, the number of valid assertions by the data path should equal the output memory buffer size for a given processing cycle.

Applying a 5x5 Filter Kernel Data Path

You will now apply a data path to the I/O buffering interface to demonstrate a complete system capable of processing a 128x128 8-bit grayscale video stream in real-time. You have chosen to use a 5x5 image processing kernel to implement the data path portion of the high-speed buffering interface. For more information about the filter kernel, refer to the

System Generator demo entitled sysgenConv5x5. You begin by considering various aspects of the design implementation.

From the MATLAB console, change directory to

$SYSGEN/examples/shared_memory/hardware_cosim/conv5x5_video

Open conv5x5_video_ex.mdl from the MATLAB console.

Buffer and Data Path Configuration

With the frame and pixel constraints in mind, the input and output buffer parameter dialog boxes are configured with a depth of 128x128 (16K) words and a word width of 8bits. This depth allows the interface to process a complete frame in a single simulation cycle. Note that these configuration parameters are propagated automatically to the lockable shared memories that implement the buffer storage.

The data path uses line buffers to properly align data samples in the filter kernel. The size of these line buffers can be parameterized to accommodate different frame sizes. In this example, the line buffers are implemented in the Virtex2 5 Line Buffer block in the conv5x5_video_ex/5x5_filter

subsystem, and are pre-configured with a line size of

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

215

Chapter 3: Using Hardware Co-Simulation

128. If you decide to process a different size frame, the Line Size parameter should be updated accordingly.

216

Valid Bit Generation

The data path includes a subsystem named valid_generator that is responsible for driving the din_valid port of the output buffer block. The subsystem has two inputs, valid_in

and offset. The valid_in port is driven by the dout_valid signal from the input buffer block, which is delayed by a variable number of cycles before it is driven to the valid_out

port. The logic associated with the valid_generator subsystem is shown below.

An addressable shift register block (ASR) is used to delay the valid bit. The offset port is used to control the address of the ASR block, which in turn controls the amount of latency the valid bit incurs. By simply delaying the valid bit generated by the input buffer block,

You ensure the number of words written to the output buffer is always equal to the buffer size. Note that when the design is run in hardware, a change in the offset value will cause the vertical alignment of the filtered images to change.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Real-Time Signal Processing using Hardware Co-Simulation

Support for Coefficient Reloading

An interesting characteristic of the kernel data path is that its coefficients can be dynamically reloaded at run-time. The 5x5 filter block includes Load and Coef control ports, which are driven by the coefficient_memory subsystem.

The coefficient_memory contains a copy of the most recently loaded filter coefficients, which are stored in an unprotected shared memory named coef_buffer. During runtime, the subsystem monitors the shared memory contents, and initiates a reload sequence if detects a change. By co-simulating the unprotected shared memory, any process on the host PC may write new kernel coefficients simply by writing to a shared memory object named coef_buffer. This interface is convenient, as communication with the FPGA hardware is completely abstracted through the Shared Memory API.

Compiling for Hardware Co-simulation

The full filter kernel design must be compiled for hardware co-simulation before it can be simulated.

Double click on the System Generator block located at the top of the conv5x5_video_ex

model.

Select an appropriate hardware co-simulation target.

Press the Generate button to compile the design for hardware co-simulation.

A hardware co-simulation block is created once the design finishes compiling.

Hardware co-simulation blocks include information about any shared memories, registers, or FIFOs that were compiled as part of the design. You may view this information by double-clicking on the hardware co-simulation block to open the parameters dialog box.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

217

Chapter 3: Using Hardware Co-Simulation

Once the dialog box is open, selecting the Shared Memories tab reveals information about each shared memory in the compiled design.

218

Go ahead and leave the hardware co-simulation library open. In the next topic you will include the hardware co-simulation block in a video processing testbench design.

5x5 Filter Kernel Test Bench

Included with the example files is a Simulink test bench model that uses the hardware cosimulation block to filter a looped video sequence.

From the

$SYGEN/examples/shared_memory/hardware_cosim/conv5x5_video directory, open conv5x5_video_testbench.mdl.

The testbench model uses a From Workspace block to produce the looped video sequence.

Each frame of the video sequence is represented as a 128x128 uint8 Simulink matrix (a pre-load function loads and initializes the video sequence automatically when the model is opened). Video frames are written into the FPGA Processing subsystem where they are filtered at the rate of one frame per simulation cycle. The filtered output is then written to a Matrix Viewer block for analysis.

The FPGA Processing subsystem contains a stub for the hardware co-simulation block, as well as Shared Memory Read and Write blocks. In this example, the Shared Memory

Read and Write blocks are responsible for managing video frame I/O to and from the shared memories operating inside the FPGA. The operation of these blocks is described below: a.

The Shared Memory Write block wakes up and requests lock of the input buffer lockable shared memory Foo. Once lock is granted, the block writes the video frame data input into the lockable shared memory and releases lock. b. The hardware co-simulation block wakes up and requests lock of the input and output buffer shared memories Foo and Bar. The host PC shared memory images are transferred to the FPGA and lock is granted. The FPGA processes the input buffer data and writes the output into the output buffer. Lastly the FPGA releases www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Real-Time Signal Processing using Hardware Co-Simulation

lock of Foo and Bar, causing the FPGA shared memory images to be transferred back to the host PC. c.

The Shared Memory Read block wakes up and requests lock of the output buffer lockable shared memory Bar. The block reads a video from the output buffer and drives its output port with the processed video frame data.

Note that the three steps listed above assume a specific sequencing of the hardware cosimulation and Shared Memory Read and Write blocks. To ensure these blocks are properly sequenced, you can set block priorities, where a lower priority block is woken up first during simulation.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

219

Chapter 3: Using Hardware Co-Simulation

Add the hardware co-simulation block to the testbench model in place of the turquoise placeholder residing in the FPGA Processing subsystem.

The Shared Memory Write block in the testbench is pre-configured with a priority of 1, and the Shared Memory Read block is pre-configured with a priority of 3. Since you want the hardware co-simulation block to wake up second in the simulation sequence, you must set the hardware co-simulation block priority to 2.

10. Right-click on the hardware co-simulation block, and select Block Properties.

220

11. Specify a Priority of 2 in the Block Properties dialog box.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Real-Time Signal Processing using Hardware Co-Simulation

For high-speed processing applications, the hardware co-simulation block should be configured to operate in Free Running clock mode. When this mode is used, the synchronization between the FPGA and Simulink are handled entirely by the lockable shared memories. By running the FPGA in free-running mode, you allow it to run fast enough to process a complete video frame in a single Simulink cycle. Keep in mind that the hardware co-simulation block circuitry waits to acquire lock before processing data. Since the lock cannot be granted until the hardware co-simulation block is woken up, the FPGA sits idle until new data is presented in the input buffer.

12. Double-click on the hardware co-simulation block and choose a Free Running

Clock

under the Basic Tab.

You are now ready to simulate the design.

13. Press the Simulink Start button to start simulation.

Two windows will appear showing the original and filtered video streams.

The left image is the original video frame. The image on the right is the same frame that has been processed using the "smooth" filter kernel. Note that the smoothing filter is just one of several filters that can be applied to the video source.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

221

Chapter 3: Using Hardware Co-Simulation

Reloading the Kernel

The filter data path is designed so that the filter kernel can be reloaded dynamically while hardware co-simulation is running. Once the simulation is running, you may use the xlReloadFilterCoef

function to load a new kernel. The function accepts a string kernel identifier (e.g., sobelxy) as an input parameter. A list of available filter kernels can be viewed by typing help xlReloadFilterCoef in the MATLAB console. The function is supplied as a MATLAB source file and can be found in the

$SYGEN/examples/shared_memory/hardware_cosim/conv5x5_video directory

Note:

Once you have reloaded the filter, you may choose to adjust the coefficient gain. The gain can be adjusted using the Coefficient Adjust slider control at the top-level of the testbench model. This also demonstrates how System Generator's traditional, port-based, hardware co-simulation interfacing can be used in conjunction with the shared memory hardware co-simulation interfaces.

It is worthwhile to note that System Generator provides a MATLAB object interface to shared memory objects. The xlReloadFilterCoef function uses this object interface to write new coefficients into the unprotected shared memory named coef_buffer running in the FPGA. The function is fully annotated with comments that explain how the shared memory object is created, written to, and released when the operation is complete.

Note:

The source code for the MATLAB object interface is supplied with the System Generator software installation, and can be found in the $SYGEN/examples/shared_memory/mex_function directory. Also included in this directory is MATLAB M-code that demonstrates how the mex-function source code was built.

14. After ensuring the testbench design is running, load the SobelXY filter kernel into the

FPGA by typing xlReloadFilterCoef('sobelxy') from the MATLAB command window.

You will now see the video output generated using the SobelX-Y kernel.

222

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Note:

If installation instructions for your particular platform are not provided here, please refer to the installation instuctions that come with your Platform Kit.

Installing an ML402 Platform for Ethernet Hardware Co-Simulation

The following procedure describes how to install and configure the hardware and software required to run Ethernet Hardware Co-Simulation on an ML402 board.

Assemble the Required Hardware

Xilinx Virtex™-4 SX ML402 Platform which includes the following: a.

Virtex™-4 ML402 board b. 5V Power Supply bundled with the ML402 Kit c.

CompactFlash Card

You also need the following items on hand: a.

Ethernet network Interface Card (NIC) for the host PC.

b. Ethernet RJ45 Male/Male cable. (May be a Network or Crossover cable.) c.

CompactFlash Reader for the PC.

Install the Software on the Host PC

Make sure the following software is installed on your PC:

•

System Generator version as specified in the current System Generator Release Notes

•

Xilinx ISE™ Software version as specified in the current System Generator Release

Notes

•

WinPcap version 4.0, which may be installed through the System Generator installer or obtained from the website at http://www.winpcap.org

Setup the Local Area Network on the PC

You are required to have a 10/100 Fast Ethernet or a Gigabit Ethernet Adapter on you PC.

To configure the settings do the following:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

223

Chapter 3: Using Hardware Co-Simulation

As shown below, from the Start menu, select Control Panel, then right-click on Local

Area Connection

, then select Properties.

As shown below, select Internet Protocol (TCP/IP), then click on the Properties button and set the IP Address 192.168.8.2 and Subnet mask to 255.255.255.0. (The last digit of the IP Address must be something other than 1, because 192.168.8.1 is the default IP

address for the ML402. See the topic Load the Sysgen ML402 HW Co-Sim

Configuration Files

below for further details.)

224

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Click on the Configure button, select the Advanced tab, select Flow Control, then select Auto.

Set Speed & Duplex to Auto, then click out using the OK button.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

225

Chapter 3: Using Hardware Co-Simulation

Load the Sysgen ML402 HW Co-Sim Configuration Files

System Generator comes with HW Co-Sim configuration files that first need to be loaded into the ML402 CompactFlash card with a CompactFlash Reader.

Optionally Backup the ML402 Demo Files

The ML402 CompactFlash card comes with a series of demo files that you might want to re-load and exercise later. a.

Connect the CompactFlash Reader to the PC. This is usually done through a USB port.

b. Insert the CompactFlash card into a CompactFlash Reader. c.

Click on the MyComputer icon, then select the Removable Disk drive that represents the CompactFlash Reader.

d. Create or open a backup folder on the PC and copy the content of the

CompactFlash card to that folder for later use.

Note:

For the following steps, 'e:' is assumed to be the drive name associated with the

CompactFlash reader.

Re-Format the CompactFlash Card

The card needs to be re-formatted to a FAT16 file system before the System Generator files can be transferred. You use the mkdosfs utility to format the card. a.

Download the mkdosfs program from the Xilinx URL address: http://www.xilinx.com/products/boards/ml310/current/utilities/mkdosfs.zip.

b. Extract to folder C:/mkdosfs c.

Open a Windows shell by selecting Start > Run..., then type cmd in the Run dialog box and click OK.

d. In the shell, move to the mkdosfs folder: cd C:\mkdosfs

Caution!

In the following step, make sure the drive name (e.g., 'e:' in this case) is specified correctly for the Compact Flash Removable Disk. Otherwise, the information on the mistakenly targeted drive will be erased and the drive will be re-formatted.

Type the following mkdosfs command after the Windows command prompt: mkdosfs -v -F 16 e:

The content of the Compact Flash card should be wiped clean and re-formatted.

Copy the Sysgen configuration files to the Compact Flash card

Note:

For reference, the Sysgen files to be copied are located at the following pathname:

...<sysgen_tree>\plugins\bin\ML402_sysace_cf.zip

Invoke MATLAB on the PC, then enter the following command on the MATLAB

Command Line: unzip(fullfile(xlFindSysgenRoot,'plugins/bin/ML402_sysace_cf.zip'),'e:/')

226

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

The following files and folder should now be listed on the CompactFlash drive:

Optional Step to set the Ethernet MAC Address and the IPv4 Address

Note:

The following step may be necessary if the default MAC and IP addresses conflict with your default network settings, or if you wish to co-simulate two or more ML402 boards concurrently. If not, proceed to the next topic.

After writing the data to the card, you will find two files, mac.dat and ip.dat, in the card root directory. The mac.dat and ip.dat files specify the Ethernet MAC address and IPv4 address associated with the board, respectively. These addresses are used to uniquely identify a target board during Ethernet hardware co-simulation. a.

Open mac.dat in a text editor and change the Ethernet MAC address. The MAC address must be specified as a six pair of two-digit hexadecimal separated by colons (e.g. 00:0a:35:11:22:33). All-zeros, broadcast, or multicast MAC addresses are not supported. b. Open ip.dat in a text editor and change the IP address. The IP address must be specified in IPv4 dotted decimal notation (e.g. 192.168.8.1). All-zeros, broadcast, multicast, or loop-back IP address are not supported. After changing the IP address for ML402 board, update the IP address for the network connection

on the PC accordingly as mentioned in topic Setup the Local Area Network on the

PC . For direct connection, the ML402 and the PC must be on the same subnet.

Otherwise, the ML402 IP address should be reachable from the PC and vice versa.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

227

Chapter 3: Using Hardware Co-Simulation

Setup the ML402 Board

The figure below illustrates the ML402 components of interest in this setup procedure:

Position the ML402 board so the Virtex™-4 and Xilinx logos are oriented near the top edge of the board.

Make sure the power switch, located in the upper-right corner of the board, is in the

OFF

position.

As shown below, Eject the CompactFlash card from the CompactFlash Reader.

228

Remove the CompactFlash card from the CompactFlash Reader.

Locate the CompactFlash card slot (on the back side of the ML402 board), and carefully insert the CompactFlash card with its front label facing away from the board. The figure below shows the back side of the board with the ConpactFlash card properly inserted.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Note:

The CompactFlash card provided with your board might differ.

Caution!

Be careful when inserting or removing the CompactFlash card from the slot. Do not force it.

Connect the AC power cord to the power supply brick. Plug the power supply adapter cable into the ML402 board. Plug in the power supply to AC power.

Caution!

Make sure you use an appropriate power supply with corrrect voltage and power ratings.

Using the RJ45 Male/Male Ethernet Cable, connect the Ethernet connector on the

ML402 board directly to the Ethernet connector on the host PC.

Set the Configuration Address DIP Switches.

As shown below, set the Configuration Address DIP Switches as follows: 1:on, 2:off,

3:off, 4:on, 5:off, 6:on

]

Set the Configuration Source Selector Switch.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

229

Chapter 3: Using Hardware Co-Simulation

As shown below, set the Configuration Source Selector Switch to SYS ACE

230

10. Verify the Configuration Settings a.

Turn the target board Power switch ON. b. Check the on-board status LEDs to ensure the FPGA is configured. If the configuration succeeded, the DONE LED should be on and all error LEDs should be off.

As shown below, check the information displayed on the 16-character 2-line LCD screen of the board. If no error occurred, the Ethernet MAC address (without colons) should appear on the first line of the display and the IPv4 address should appear on the second line.

d. If the LCD display does not show the information correctly, press the System

ACE™ Reset button to reconfigure the FPGA. e.

Check the status LEDs again to ensure the configuration sequence completed successfully.

11. Verify the Ethernet Interface and Connection Status a.

Connect the Ethernet interface of the board to a network connection, or directly to a host. b. Check the on-board Ethernet status LEDs to make sure the Ethernet interface is attached to an active Ethernet segment. The LEDs should reflect the link speed and the duplex mode at which the interface is operating. The TX and RX leds should flash on and off occasionally depending on the network traffic. If no LED is on, press the CPU Reset button to reset the FPGA, and also examine whether the

Ethernet segment is active.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

To ensure the board is reachable by the host, issue ICMP ping from the host to check the connectivity. For example, type "ping 192.168.8.1" on a console to test the connectivity to a board with IP address 192.168.8.1.

d. The target FPGA listens on the UDP port 9999. Please ensure the underlying network does not block the associated traffic when network-based Ethernet configuration is used. This does not affect point-to-point Ethernet configuration.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

231

Chapter 3: Using Hardware Co-Simulation

Installing an ML506 Platform for Ethernet Hardware Co-Simulation

The following procedure describes how to install and configure the hardware and software required to run an ML506 Board Point-to-Point Ethernet Hardware Co-Simulation.

Assemble the Required Hardware

Xilinx Virtex™-5 SX ML506 Platform which includes the following: a.

Virtex-5 ML506 board b. 5V Power Supply bundled with the ML506 kit c.

CompactFlash Card

You also need the following items on hand: a.

Ethernet network Interface Card (NIC) for the host PC.

b. Ethernet RJ45 Male/Male cable. (May be a Network or Crossover cable.) c.

CompactFlash Reader for the PC.

Install the Software on the Host PC

Make sure the following software is installed on your PC:

•

System Generator version as specified in the current System Generator Release Notes.

•

Xilinx ISE™ Software version as specified in the current System Generator Release

Notes.

•

WinPcap version 4.0, which may be installed through the System Generator installer or obtained from the website at http://www.winpcap.org

Setup the Local Area Network on the PC

You are required to have a 10/100 Fast Ethernet or a Gigabit Ethernet Adapter on you PC.

To configure the settings do the following:

As shown below, from the Start menu, select Control Panel, then right-click on Local

Area Connection

, then select Properties.

232

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

As shown below, select Internet Protocol (TCP/IP), then click on the Properties button and set the IP address 192.168.8.2 and the Subnet mask to 255.255.255.0. (The last digit of the IP Address must be something other than 1 because 192,168.8.1 is the default IP

address fo ML506. See Load the Sysgen ML506 HW Co-Sim Configuration Files

for further details.).

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

233

Chapter 3: Using Hardware Co-Simulation

Click on the Configure button, select the Advanced tab, select Flow Control, then select Auto.

Set Speed & Duplex to Auto, then click out using the OK button.

234

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Load the Sysgen ML506 HW Co-Sim Configuration Files

System Generator comes with HW Co-Sim configuration files that first need to be loaded into the ML506 CompactFlash card with a CompactFlash Reader.

Optionally Backup the ML506 Demo Files

The ML506 CompactFlash card comes with a series of demo files that you might want to re-load and exercise later. a.

Connect the CompactFlash Reader to the PC. This is usually done through a USB port.

b. Insert the CompactFlash card into a CompactFlash Reader. c.

Click on the MyComputer icon, then select the Removable Disk drive that represents the CompactFlash Reader..

d. Create or open a backup folder on the PC and copy the content of the

CompactFlash card to that folder for later use.

Note:

For the following steps, 'e:' is assumed to be the drive name associated with the

CompactFlash reader.

Re-Format the CompactFlash Card

The card needs to be re-formatted to a FAT16 file system before the System Generator files can be transferred. You use the mkdosfs utility to format the card. a.

Download the mkdosfs program from the Xilinx URL address: http://www.xilinx.com/products/boards/ml310/current/utilities/mkdosfs.zip.

b. Extract to folder C:/mkdosfs c.

Open a Windows shell by selecting Start > Run..., then type cmd in the Run dialog box and click OK.

d. In the shell, move to the mkdosfs folder: cd C:\mkdosfs

Caution!

Type the following mkdosfs command after the Windows command prompt: mkdosfs -v -F 16 e:

The content of the Compact Flash card should be wiped clean and re-formatted.

Copy the Sysgen configuration files to the Compact Flash card

Note:

For reference, the Sysgen files to be copied are located at the following pathname:

...<sysgen_tree>\plugins\bin\ML506_sysace_cf.zip

Invoke MATLAB on the PC, then enter the following command on the MATLAB

Command Line: unzip(fullfile(xlFindSysgenRoot,'plugins/bin/ML506_sysace_cf.zip'),'e:/')

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

235

Chapter 3: Using Hardware Co-Simulation

The following files and folder should now be listed on the CompactFlash drive:

Optional Step to set the Ethernet MAC Address and the IPv4 Address

Note:

The following step may be necessary if the default MAC and IP addresses conflict with your default network settings, or if you wish to co-simulate two or more ML506 boards concurrently. If not, proceed to the next topic.

Open ip.dat in a text editor and change the IP address. The IP address must be specified in IPv4 dotted decimal notation (e.g. 192.168.8.1). All-zeros, broadcast, multicast, or loop-back IP address are not supported. After changing the IP address for the ML506 board, update the IP address for the network connection on the PC accordingly as mentioned in the topic

Setup the Local Area

Network on the PC

. For direct connection, the ML506 and the PC must be on the same subnet. Otherwise, the ML506 IP address should be reachable from the PC and vice versa.

236

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Setup the ML506 Board

The figure below illustrates the ML506 components of interest in this setup procedure:

Configuration Address

DIP Switches (SW3)

Power Connector

Power Switch

Ethernet

Ethernet Mode Select jumpers (J22 & J23)

Ethernet Status LEDs

LCD

Position the ML506 board so the Xilinx logo is oriented near the lower-left corner.

Make sure the power switch, located in the upper-right corner of the board, is in the

OFF

position.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

237

Chapter 3: Using Hardware Co-Simulation

As shown below, Eject the CompactFlash card from the CompactFlash Reader.

Remove the CompactFlash card from the CompactFlash Reader.

Locate the CompactFlash card slot (on the back side of the ML506 board), and carefully insert the CompactFlash card with its front label facing away from the board. The figure below shows the back side of the board with the ConpactFlash card properly inserted.

Note:

The CompactFlash card provided with your board might differ.

Caution!

Be careful when inserting or removing the CompactFlash card from the slot. Do not force it.

238

Connect the AC power cord to the power supply brick. Plug the 5V power supply adapter cable into the ML506 board. Plug in the power supply to AC power.

Caution!

Make sure you use an appropriate power supply with correct voltage and power ratings.

Using the RJ45 Male/Male Ethernet Cable, connect the Ethernet connector on the

ML506 board directly to the Ethernet connector on the host PC. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Set the SW3 Configuration Address DIP Switches.

SW3 Configuration

Address

DIP Switches

(not yet configured)

Set the Configuration Address DIP Switches as follows:

1:on, 2:off, 3:off, 4:on, 5:off, 6:on, 7:off, 8:on

Set the Ethernet Mode Select jumpers

As shown below, connect pin 1 and 2 on both the Ethernet Mode Select jumpers (J22 and J23)

Ethernet Mode Select jumpers (J22 & J23)

10. Verify the Configuration Settings a.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

239

Chapter 3: Using Hardware Co-Simulation d. If the LCD display does not show the information correctly, press the System

ACE™ Reset button to reconfigure the FPGA.

System ACE

™

Reset

Check the status LEDs again to ensure the configuration sequence completed successfully.

11. Verify the Ethernet Interface and Connection Status a.

Ethernet segment is active.

240

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Installing a Spartan-3A DSP 1800A Starter Platform for Ethernet

Hardware Co-Simulation

The following procedure describes how to install and setup the hardware and software required to run Hardware Co-Simulation on a Spartan-3A DSP 1800A Starter Platform.

This platform uses a JTAG cable instead of System ACE™ to download the configurtion bitstream.

Assemble the Required Hardware

Xilinx Spartan-3A DSP 1800A Starter Platform which includes the following: a.

Spartan-3A DSP 1800A Starter Platform b. 5V Power Supply bundled with the development kit

You also need the following items on hand: a.

Ethernet network Interface Card (NIC) for the host PC.

b. Ethernet RJ45 Male/Male cable. (May be a Network or Crossover cable.) c.

Xilinx Parallel Cable IV with associated Power Jack splitter cable or a Xilinx

Platform USB Cable and a 14-pin ribbon cable.

Install the Software on the Host PC

Make sure the following software is installed on your PC:

•

System Generator version as specified in the current System Generator Release Notes.

•

Xilinx ISE™ Software version as specified in the current System Generator Release

Notes.

•

WinPcap version 4.0, which may be installed through the System Generator installer or obtained from the website at http://www.winpcap.org

Setup the Local Area Network on the PC

You are required to have a 10/100 Fast Ethernet or a Gigabit Ethernet Adapter on you PC.

To configure the settings do the following:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

241

Chapter 3: Using Hardware Co-Simulation

As shown below, from the Start menu, select Control Panel, then right-click on Local

Area Connection

, then select Properties.

242

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Click on the Configure button, select the Advanced tab, select Flow Control, then select Auto.

Set Speed & Duplex to Auto, then click out using the OK button.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

243

Chapter 3: Using Hardware Co-Simulation

Setup the Spartan-3A DSP 1800A Starter Platform

Position the Spartan-3A DSP 1800A Starter Platform so the Xilinx logo is oriented rightside up and located in the lower-right quadrant of the platform.

Make sure the power switch, located in the upper-right corner of the platform, is in the

OFF

position.

If you are using a Xilinx Parallel Cable IV, follow steps 3a through 3d.

Connect the DB25 Plug Connector on the Xilinx Parallel Cable IV to the IEEE-1284 compliant PC Parallel (Printer) Port Connector .

b. Using the narrow (14 pin) 6” High Performance Ribbon cable, connect the pod end of the Xilinx Parallel Cable IV to the JTAG Port (J2) on the Starter Platform.

Connect the attached Power Jack cable to the Keyboard/Mouse connector on the

PC.

d. If necessary, connect the male end of the Keyboard/Mouse cable to the associated female connector on the Xilinx Power Jack cable (splitter cable) .

If you are using a Xilinx Platform Cable USB, follow step 4a and 4b.

Connect the Xilinx Platform Cable USB to a USB port on the PC.

b. Using the narrow (14 pin) 6” High Performance Ribbon cable, connect the pod end of the Xilinx Platform Cable USB to the JTAG Port (J2) on the Starter Platform.

Connect the AC power cord to the power supply brick. Plug the 5V power supply adapter cable into the 5V DC ONLY connector (J5) on the Starter Board. Plug the power supply cord into AC power.

Caution!

Make sure you use an appropriate power supply with correct voltage and power ratings.

Turn the Spartan-3A DSP 1800A Starter Platform POWER switch ON.

Installing a Spartan-3A DSP 3400A Development Platform for Ethernet

Hardware Co-Simulation

The following procedure describes how to install and configure the hardware and software required to run an Spartan-3A DSP 3400A Development Platform Point-to-Point Ethernet

Hardware Co-Simulation.

Assemble the Required Hardware

Xilinx Spartan-3A DSP 3400A Development Platform Kit which includes the following: a.

Spartan-3A DSP 3400A Development board b. +12V Power Supply bundled with Board LYR178-101C (Rev C) or

+5 V Power Supply bundled with Board LYR178-101D (Rev D) c.

CompactFlash Card

You also need the following items on hand: a.

Ethernet network Interface Card (NIC) for the host PC.

b. Ethernet RJ45 Male/Male cable. (May be a Network or Crossover cable.) c.

CompactFlash Reader for the PC.

244

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Install the Software on the Host PC

Make sure the following software is installed on your PC:

•

System Generator version as specified in the current System Generator Release Notes.

•

Xilinx ISE Software version as specified in the current System Generator Release

Notes.

•

WinPcap version 4.0, which may be installed through the System Generator installer or obtained from the website at http://www.winpcap.org

Setup the Local Area Network on the PC

You are required to have a 10/100 Fast Ethernet or a Gigabit Ethernet Adapter on you PC.

To configure the settings do the following:

As shown below, from the Start menu, select Control Panel, then right-click on Local

Area Connection

, then select Properties.

address fo ML506. See Load the Sysgen ML506 HW Co-Sim Configuration Files

for further details.).

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

245

Chapter 3: Using Hardware Co-Simulation

246

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Click on the Configure button, select the Advanced tab, select Flow Control, then select Auto.

Set Speed & Duplex to Auto, then click out using the OK button.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

247

Chapter 3: Using Hardware Co-Simulation

Load the Sysgen Spartan-3A DSP 3400A HW Co-Sim Configuration Files

System Generator comes with HW Co-Sim configuration files that first need to be loaded into the Spartan-3A DSP 3400A CompactFlash card with a CompactFlash Reader.

Optionally Backup the Spartan-3A DSP 3400A Demo Files

The Spartan-3A DSP 3400A CompactFlash card comes with a series of demo files that you might want to re-load and exercise later. a.

Connect the CompactFlash Reader to the PC. This is usually done through a USB port.

b. Insert the CompactFlash card into a CompactFlash Reader. c.

Click on the MyComputer icon, then select the Removable Disk drive that represents the CompactFlash Reader..

d. Create or open a backup folder on the PC and copy the content of the

CompactFlash card to that folder for later use.

Note:

For the following steps, 'e:' is assumed to be the drive name associated with the

CompactFlash reader.

Re-Format the CompactFlash Card

The card needs to be re-formatted to a FAT16 file system before the System Generator files can be transferred. You use the mkdosfs utility to format the card. a.

Download the mkdosfs program from the Xilinx URL address: http://www.xilinx.com/products/boards/ml310/current/utilities/mkdosfs.zip.

b. Extract to folder C:/mkdosfs c.

Open a Windows shell by selecting Start > Run..., then type cmd in the Run dialog box and click OK.

d. In the shell, move to the mkdosfs folder: cd C:\mkdosfs

Caution!

Type the following mkdosfs command after the Windows command prompt: mkdosfs -v -F 16 e:

The content of the Compact Flash card should be wiped clean and re-formatted.

Copy the Sysgen configuration files to the Compact Flash card

Note:

For reference, the Sysgen files to be copied are located at the following pathname:

...<sysgen_tree>\plugins\bin\S3ADSP_DB_sysace_cf.zip

Invoke MATLAB on the PC, then enter the following command on the MATLAB

Command Line: unzip(fullfile(xlFindSysgenRoot,'plugins/bin/S3ADSP_DB_sysace_cf.zip'),'e

:/')

Optional Step to set the Ethernet MAC Address and the IPv4 Address

Note:

248

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Setup the Local Area

Network on the PC

. For direct connection, the ML506 and the PC must be on the same subnet. Otherwise, the ML506 IP address should be reachable from the PC and vice versa.

Setup the Spartan-3A 3400A Development Board

The figure below illustrates the Spartan-3A 3400A Board (Rev C) components of interest in this setup procedure:

Ethernet Mode Select jumper (JP2)

Power Switch

+12V Power Connector

Ethernet Port

LYR178-101C (Rev C)

Configuration

Address

DIP Switches (S2)

System ACE

™

Reset Button

System Generator for DSP

Release 10.1.3 September, 2008

Compact Flash Card

LCD

www.xilinx.com

249

Chapter 3: Using Hardware Co-Simulation

The figure below illustrates the Spartan-3A 3400A Board (Rev D) components of interest in this setup procedure:

Ethernet Mode Select

Ethernet Port

LYR178-101D (Rev D)

Configuration

Address

DIP Switches (S2)

System ACE

™

Reset Button

LCD

Compact Flash Card

+5V Power Connector

Power Switch

Position the Spartan-3A 3400A Development Board as shown above with the LCD display at the bottom.

Make sure the power switch is in the OFF position.

As shown below, Eject the CompactFlash card from the CompactFlash Reader.

250

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Remove the CompactFlash card from the CompactFlash Reader.

Locate the CompactFlash card slot (on the back side of the Spartan-3A 3400A board) and carefully insert the CompactFlash card with its front label facing away from the board. The figure below shows the back side of a board with the ConpactFlash card properly inserted.

Note:

The CompactFlash card provided with your board might differ.

Caution!

Be careful when inserting or removing the CompactFlash card from the slot. Do not force it.

If you are using a “Rev C” 3400A Development Platform, plug the +12V power supply adapter cable into the power connector. Plug in the power supply into AC power.

If you are using a “Rev D” 3400A Development Platform, plug the +5V power supply adapter cable into the power connector. Plug in the power supply into AC power.

Caution!

Make sure you use an appropriate power supply with the correct voltage and power ratings.

Using the RJ45 Male/Male Ethernet Cable, connect the Ethernet connector on the

Spartan-3A 3400A board directly to the Ethernet connector on the host PC.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

251

Chapter 3: Using Hardware Co-Simulation

Set the S2 Configuration Address DIP Switches as follows:

1:off, 2:on, 3:off, 4:on, 5:off, 6:on, 7:on, 8:off

Set the Ethernet Mode Select jumper JP2 to pin 1 and pin 2 (the default GMII).

10. Verify the Configuration Settings a.

Turn the target board Power switch ON. b. As shown below, check the information displayed on the 16-character 2-line LCD screen of the board. If no error occurred, the Ethernet MAC address (without colons) should appear on the first line of the display and the IPv4 address should appear on the second line.

If the LCD display does not show the information correctly, press the System

ACE™ Reset button to reconfigure the FPGA.

11. Verify the Ethernet Interface and Connection Status a.

252

b. The target FPGA listens on the UDP port 9999. Please ensure the underlying network does not block the associated traffic when network-based Ethernet configuration is used. This does not affect point-to-point Ethernet configuration.

For indepth reference information on the Spartan-3A 3400A Development Platform, plese refer to the following online manual: http://www.xilinx.com/bvdocs/ipcenter/user_guide_user_manual/s3a-dsp-3400auserguide.pdf

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Installing Your Hardware Co-Simulation Board

Installing an ML402 Board for JTAG Hardware Co-Simulation

The following procedure describes how to install and setup the hardware and software required to run JTAG Hardware Co-Simulation on an ML402 board.

Assemble the Required Hardware

Xilinx Virtex™-4 SX ML402 Platform which includes the following: a.

Virtex-4 ML402 board b. 5V Power Supply bundled with the ML402 kit c.

CompactFlash Card

You also need the following items on hand: a.

Xilinx Parallel Cable IV with associated Power Jack splitter cable or Xilinx

Platform USB Cable and a 14-pin ribbon cable.

b. CompactFlash Reader for the PC.

Install the Software on the Host PC

•

System Generator version as specified in the current System Generator Release Notes.

•

Xilinx ISE™ Software version as specified in the current System Generator Release

Notes.

Setup the ML402 Board

The figure below illustrates the ML402 components of interest in this JTAG setup procedure:

FPGA & CPU

Debug Port

Position the ML402 board so the Virtex™-4 and Xilinx logos are oriented near the top edge of the board.

Make sure the power switch, located in the upper-right corner of the board, is in the

OFF

position.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

253

Chapter 3: Using Hardware Co-Simulation

If you are using a Xilinx Parallel Cable IV, follow steps 3a through 3d.

Connect the DB25 Plug Connector on the Xilinx Parallel Cable IV to the IEEE-1284 compliant PC Parallel (Printer) Port Connector .

b. Using the narrow (14 pin) 6” High Performance Ribbon cable, connect the pod end of the Xilinx Parallel Cable IV to the FPGA & CPU Debug Port (shown above) on the ML402 Board.

Connect the attached Power Jack cable to the Keyboard/Mouse connector on the

PC.

d. If necessary, connect the male end of the Keyboard/Mouse cable to the associated female connector on the Xilinx Power Jack cable (splitter cable) .

If you are using a Xilinx Platform Cable USB, follow step 4a and 4b..

Connect the Xilinx Platform Cable USB to a USB port on the PC.

b. Using the narrow (14 pin) 6” High Performance Ribbon cable, connect the pod end of the Xilinx Platform Cable USB to the FPGA & CPU Debug Port (shown above) on the ML402 Board.

Connect the AC power cord to the power supply brick. Plug the power supply adapter cable into the ML402 board. Plug in the power supply to AC power.

Caution!

Make sure you use an appropriate power supply with correct voltage and power ratings.

Turn the ML402 Board Power switch ON.

Supporting New Platforms through JTAG Hardware Co-Simulation

System Generator provides a generic interface that uses JTAG and a Xilinx programming cable (e.g., Parallel Cable IV or Platform Cable USB) to communicate with FPGA hardware. This takes advantage of the ability of JTAG to extend System Generator's hardware in the simulation loop capability to numerous other FPGA platforms.

Hardware Requirements

An FPGA platform can support the JTAG hardware co-simulation interface, provided it includes the following hardware components:

•

A Xilinx FPGA part that is available in System Generator as a supported device (i.e., a device that can be chosen in the Part field of the System Generator block dialog box);

•

An on-board oscillator that supplies the FPGA with a free-running clock source;

•

A JTAG header that provides access to the FPGA.

Supporting New Platforms

Although the JTAG hardware co-simulation interface is generic, an FPGA platform must provide its own board support package before it can be supported in System Generator. A board support package is comprised of four files that provide information about the board, or platform. A number of FPGA platforms already have board support packages available.

See Hardware Co-Simulation Installation for more information on how to download these files.

You may have an FPGA platform that does not have a hardware co-simulation board support package. In this case, you can create your own, assuming your platform meets the

specified Hardware Requirements . Creating a new board support package for a platform is

254

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Supporting New Platforms through JTAG Hardware Co-Simulation

a straightforward process. System Generator provides a utility, called the System

Generator Board Description Builder (SBDBuilder), that allows you to create new board support packages in a graphical environment. It is also possible to define board support packages manually by editing a series of template files that are included in the System

Generator software tree.

SBDBuilder can be launched by using the command xlSBDBuilder

in the MATLAB console. Alternatively, SBDBuilder can also be launched from the System Generator Token by double-clicking on the System Generator token; under Compilation select Hardware

Co-Simulation > New Compilation Target ...

SBDBuilder Dialog Box

After invoking SBDBuilder, the main dialog box will appear as shown below:

Once the main dialog box is open, you may create a board support package by filling in the required fields described below:

Board Name

: Tells a descriptive name of the board. This is the name that will be listed in

System Generator when selecting your JTAG hardware co-simulation platform for compilation.

System Clock

: JTAG hardware co-simulation requires an on-board clock to drive the

System Generator design. The fields described below specify information about the board's system clock:

•

Frequency (MHz)

: Specifies the frequency of the on-board system clock in MHz.

•

Pin Location

: Specifies the FPGA input pin to which the system clock is connected.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

255

Chapter 3: Using Hardware Co-Simulation

JTAG Options

: System Generator needs to know several things about the FPGA board's

JTAG chain to be able to program the FPGA for hardware co-simulation. The topic

Obtaining Platform Information describes how and where to find the information required

for these fields. If you are unsure of the specifications of your board, please refer to the manufacturer's documentation. The fields specific to JTAG Options are described below:

•

Boundary Scan Position

: Specifies the position of the target FPGA on the JTAG chain.

This value should be indexed from 1. (e.g. the first device in the chain has an index of

1, the second device has an index of 2, etc.)

•

IR Lengths

: Specifies the lengths of the instruction registers for all of the devices on the JTAG chain. This list may be delimited by spaces, commas, or semicolons.

•

Detect

: This action attempts to identify the IR Lengths automatically by querying the

FPGA board. The board must be powered and connected to a Parallel Cable IV for this to function properly. Any unknown devices on the JTAG chain will be represented with a "?" in the list, and must be specified manually.

Targetable Devices

: This table displays a list of available FPGAs on the board for programming. This is not a description of all of the devices on the JTAG chain, but rather a description of the possible devices that may exist at the aforementioned boundary scan position. For most boards, only one device needs to be specified, but some boards may have alternate, e.g., a choice between an xcv1000 or an xcv2000 in the same socket. Use the

Add

and Delete buttons described below to build the device list:

•

Add

: Brings up a menu to select a new device for the board. As shown in the figure below, devices are organized by family, then part name, then speed, and finally the package type.

•

Delete

: Remove the selected device from the list.

256

Non-Memory-Mapped Ports

: You can add support for your own board-specific ports when creating a board support package. Board-specific ports are useful when you have onboard components (e.g., external memories, DACs, or ADCs) that you would like the

FPGA to interface to during hardware co-simulation. Board specific ports are also referred to as non-memory-mapped because when the design is compiled for hardware cosimulation, these ports will be mapped to their physical locations, rather than creating

Simulink ports. See

Specifying Non-Memory Mapped Ports

for more information. The

Add

, Edit, and Delete buttons provide the controls needed for configuring non-memory mapped ports.

•

Add

: Brings up the dialog to enter information about the new port.

•

Edit

: Make changes to the selected port.

•

Delete

: Remove the selected port from the list.

Help

: Displays this documentation. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Supporting New Platforms through JTAG Hardware Co-Simulation

Load

: Fill in the form with values stored in an SBDBuilder Saved Description XML file.

This file is automatically saved with every plugin that you create, so it is useful for reloading old plugin files for easy modification.

Save Zip

: Prompts you for a filename and a target pathname. This will create a zip file with all of the plugin files for System Generator. The zip will be in a suitable format for passing to the System Generator xlInstallPlugin function.

Exit

: Quit the application.

Specifying Non-Memory Mapped Ports

You may use SBDBuilder to specify the non-memory mapped ports for your FPGA platform. When you choose to Add or Edit a non-memory mapped port from the main dialog, the port editor dialog will come up as shown below.

The port editor dialog presents the following controls for port configuration:

Port Options

: Specifies the options that will affect the entire port.

•

Port Name

: This is the name that will describe the port in System Generator. It should be a MATLAB-compatible name (begins with a letter, followed by only letters, numbers, and underscores).

•

Input/Output

: Specifies the direction of the port.

New Pin

: This is the entry point to add pins to a port. Ports may consist of a single pin for a Boolean value, or multiple pins for a vector or bus.

•

Pin LOC

: Defines the absolute placement of the pin within the FPGA by specifying a location constraint. It is necessary to define this for every pin to make sure that the

FPGA programming corresponds to the actual hardware connections.

•

PULLUP

: A constraint that can be applied to each pin. It guarantees a logic High level to allow 3-stated nets to avoid floating when not being driven.

•

FAST

: A constraint that can be applied to each pin. It increases the speed of an IOB output. FAST produces a faster output but may increase noise and power consumption.

•

Add Pin

: Add a pin to the port. Note that the pin is not part of the port until this button is selected.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

257

Chapter 3: Using Hardware Co-Simulation

Note:

Pressing 'enter' while the cursor is in the Pin LOC field is equivalent to pressing this button.

Pin List

•

Index

: (Cannot edit directly) Since a port can be more than one bit, it is represented as a vector of pins. The index indicates which bit position a particular pin represents in the port. Zero is the least-significant bit.

•

Move Up/Down

: Move the selected pin up or down in the pin list. This is useful to correct the vector bit-ordering of the port.

•

Delete Pin

: Removes the selected pin from the list.

Save and Start New

: Save the port to the board support package. The form will then be cleared so that you may enter a new port.

Save and Close

: Save the port to the board support package and return to the main screen.

Cancel

: Discard changes to the current port and return to the main screen.

When you are finished entering a port, it will look similar to the dialog box shown below:

258

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Supporting New Platforms through JTAG Hardware Co-Simulation

Saving Plugin Files

Once you have filled out the dialog box with information about your platform, it should resemble the dialog box shown below:

At this point, you can save the board support package into a System Generator plugin zip

file or as the raw board support package files described in the topic Board Support Package

Files , plus the additional SBDBuilder files described below:

•

yourboard.xml

: This is the SBDBuilder Saved Description, which allows SBDBuilder to reload plugins you have previously created. The name you select for this file

('yourboard') will propagate into the names of the other files as well.

•

yourboard_libgen.m

: Automates the process of creating the gateways for the nonmemory-mapped ports on this device. Running this script results in the creation a library like that shown below:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

259

Chapter 3: Using Hardware Co-Simulation

Board Support Package Files

An FPGA platform that supports JTAG hardware co-simulation is defined in System

Generator by its board support package. This package tells System Generator useful information about the platform, such as the appropriate part settings and additional information related to the JTAG and Boundary Scan interface provided by the platform. A board support package is comprised of the files listed below:

Note:

In this document, three of the filenames are prefixed by the name 'yourboard'. This prefix should be replaced with a moniker suitable for 'your board' (e.g., xtremedspkit, mblazedemo).

xltarget.m

– Tells System Generator that your FPGA platform is a compilation target.

There is a unique xltarget.m file for each compilation target. This function tells the tool the name of the compilation target (this name is shown in the compilation target field of the System Generator block dialog box) and also the name of the function where it can look for information about the particular board.

yourboard_target.m

– Configures the System Generator block dialog box with information about the FPGA platform, including device and part information, clock frequency, and the location of the clock pin.

yourboard_postgeneration.m

– Tells System Generator the scripts to run after HDL netlist generation in order to produce an FPGA configuration file that is suitable for your platform. It also specifies non-System Generator block dialog box related information, including the position of the device in the platform's Boundary Scan chain and the instruction register lengths of each device. This function is referred to as a post-generation function.

yourboard.ucf

– User Constraints File (UCF) for the FPGA platform. Specifies clock pin location and frequency, and optionally constrains any board-specific ports.

Included in the System Generator software tree are templates for the files listed above. If you would like to manually support a new board, you may customize each of the four template files with information that is specific to your platform. You must also rename the files by substituting a suitable name in place of the 'yourboard' prefix.

Each template is fully annotated with step-by-step instructions that indicate which fields should be modified, and the types of values that should be given to these fields. The fields that must be modified are underlined using "~~~" notation. The template files can be found in the sysgen/hwcosim/jtag/templates directory of your System Generator install tree.

260

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Supporting New Platforms through JTAG Hardware Co-Simulation

Obtaining Platform Information

SBDBuilder (or alternatively, the board support package template files) require certain information about your FPGA platform. The table below lists the information you need.

Information

Clock pin location

Clock period

Device position in the Boundary Scan

Chain

Instruction register lengths

Description

Pin location constraint for the FPGA system clock source.

Period constraint for the FPGA system clock source.

Tells the position of the target FPGA in the platform's Boundary

Scan Chain. Indexing begins at 1, with device 1 being the first device in the chain.

Instruction register length of every device in the Boundary Scan

Chain.

You may obtain the clock pin location and period from any number of possible sources, including the vendor documentation, existing constraints files, or vendor online documentation/support.

If you do not know which devices are in your platform's boundary scan chain, you may use iMPACT to assist you in finding this information. iMPACT is a tool that is included with the Xilinx ISE™ software that allows you to perform device configuration and file generation functions. When the tool is invoked, it automatically detects the contents of your platform's boundary scan chain, and displays these contents graphically, as shown below.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

261

Chapter 3: Using Hardware Co-Simulation

Once you have determined which devices are in the Boundary Scan Chain, you must determine the instruction register lengths for each device. The table below specifies the instruction register lengths for various Xilinx families. You may use the auto detection capability of SBDBuilder to determine the instruction register lengths. If this utility does not work, you may use the following table to find the instruction register lengths for a particular part family.

Family

XC9500 / XC9500XL / XC9500XV

XC1800 / XC18V00

XC4000XL/XLA

Spartan™-XL

System_ACE ™ -CF

Virtex ™ / Virtex ™ -E(EM)

Spartan

™

-II / Spartan

™

-IIE

Virtex

™

-II

Spartan ™ -3

Spartan

™

-3E

Spartan

™

-3A/Spartan

™

-3AN

Spartan

™

-3A DSP

Virtex ™ -II Pro 2, 4, 7

Virtex ™ -II Pro 20, 30, 40, 50, 70, 100

Virtex

™

-II Pro 125

Virtex

™

-4 LX

Virtex ™ -4 SX

Virtex ™ -4 FX 12, 20

Virtex

™

-4 FX 40, 60, 100, 140

Virtex

™

-5 LX

XCR3000XL

XCR3000A / XCR3128

XCR3320 / XCR3960

XCR5128 / XCR5032C / XCR5064C / XCR5128C

CoolRunner ™ -II

Platform Flash XCFxxS

Platform Flash XCFxxP

IR Length

262

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Supporting New Platforms through JTAG Hardware Co-Simulation

Manual Specification of Board-Specific Ports

You can manually specify your own board-specific ports when creating a board support package. To define board-specific ports for your FPGA platform, you must do the following:

•

Add all board-specific ports to the yourboard.ucf template file. Each constraint should be accompanied by a special comment, <port> contingent, where <port> is the name of the board specific port. When System Generator compiles a model for hardware, it creates a custom UCF file. Constraints associated with signals that aren't used in the model are removed from the custom UCF file.

Example constraints for ports adc1_d(0) and adc1_d(1): net adc1_d(0) loc = af20; # adc1_d contingent net adc1_d(1) loc = ad18; # adc1_d contingent

•

Declare all board-specific ports in the yourboard_postgeneration.m function.

Port Name

(Must match UCF) Width

non_mm_ports.('adc1_d') = ('in',14);

Direction

Note:

Bi-directional ports are currently not supported.

Include this line in yourboard_postgeneration.m function: params.('non_memory_mapped_ports') = non_mm_ports;

•

Customize a gateway with the board-specific port information:

♦

Create a library and add a gateway

Name the Gateway with the name of your board specific port (this name must match the port name used in the post-generation function and UCF file)

♦

Select the Gateway by clicking on it

In the MATLAB command window, type the following

> xlSetNonMemMap(gcb, 'Xilinx', 'jtaghwcosim')

♦

Save the library

You are now ready to use your board-specific gateway in System Generator. When you include the gateway in your model, you must make sure the signals that drive (or are driven by) the gateway have widths that match the widths of the ports in hardware. You can force the width of a signal driving a gateway out by preceding it with a convert block.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

263

Chapter 3: Using Hardware Co-Simulation

Note:

A subsystem (as shown below) is a convenient place to store the gateway out and convert block pairs.

264

Providing Your Own Top Level

When a model is compiled for JTAG hardware co-simulation, System Generator produces a generic top-level HDL entity for the design. This entity instantiates the logic required by the model and the interfacing logic required for JTAG hardware co-simulation.

Sometimes your platform may have a special requirement that precludes you from using this generic top level. For example, your platform may have components that rely on clocks that are generated by a DCM that resides in the platform's FPGA. In these situations,

System Generator allows you to use your own top-level netlist when it compiles the model into hardware.

Note:

If you choose to use your own top-level component, you must provide a previously synthesized version (.ngc, .edf, .edn) to System Generator.

Note:

Your top-level component must instantiate the generic JTAG hardware co-simulation top-level component. The component instantiation must include the required clocking signals, plus any boardspecific I/O ports your board may support. An example component instantiation is provided below: component jtagcosim_top port (

-- required clocking ports sys_clk : in std_logic; cosim_clk : out std_logic; sys_clk_buf : out std_logic;

-- board specific ports adc1_d : in std_logic_vector(13 downto 0); dac1_d : out std_logic_vector(13 downto 0); dac1_div0 : out std_logic; dac1_div1 : out std_logic; dac1_mod0 : out std_logic; dac1_mod1 : out std_logic; dac1_reset : out std_logic

); end component;

You may specify your own top-level netlist in yourboard_postgeneration.m as follows: params.vendor_toplevel = 'yourboard_toplevel'; www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Supporting New Platforms through JTAG Hardware Co-Simulation

Here yourboard_toplevel is the name of the pre-compiled, top-level netlist component you would like System Generator to use for the top level. You must also tell System

Generator the netlist file names that are associated with the top-level component. These files are specified as shown below in yourboard_postgeneration.m. params.vendor_netlists = {'yourboard_toplevel.ngc','foo.edf'};

Installing Board-Support Packages

SBDBuilder can generate a plugin zip file for your board support package that may be installed automatically using the xlInstallPlugin utility provided with System Generator.

You may manually install the board support package files if an appropriate plugin zip file is not provided. This topic describes how to install the files manually in your System

Generator software tree.

Plugins Directory

The System Generator software provides a special directory in which the board support package files for new compilation targets can be added. This directory, plugins/compilation

, provides a repository for System Generator compilation target plugins, and has unique properties that are discussed later in this topic. Your System

Generator software tree should resemble the tree hierarchy shown below.

The board support package files for you platform should be saved in a subdirectory, or series of subdirectories, under the plugins/compilation directory.

Note:

All configuration files associated with a board support package must be saved in the same directory.

System Generator searches this directory (and subdirectories) for compilation targets.

Recall that the xltarget.m file tells System Generator the platform should be used as a compilation target. When the tool searches the plugins/compilation directory, it adds a compilation target to the System Generator block dialog box for every xltarget.m file that it encounters.

The System Generator block dialog box Compilation submenus mirror the directory structure under the plugins/compilation directory. When you create a new directory,

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

265

Chapter 3: Using Hardware Co-Simulation or directory hierarchy, for a board support package, the names of the directories define the taxonomy of the compilation target submenus.

Detecting New Packages

In order for System Generator to recognize the new target, you must tell it to search for new compilation targets by entering the following command in the MATLAB command window:

xlrehash_xltarget_cache

You can now select the FPGA platform from the list of compilation targets in the System

Generator block dialog box.

Note:

If you have a System Generator block dialog box open when you enter this command, it will not show up until you close and re-open the dialog box.

266

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Chapter 4 Importing HDL Modules

Sometimes it is important to add one or more existing HDL modules to a System Generator design. The System Generator Black Box block allows VHDL, Verilog, and EDIF to be brought into a design. The Black Box block behaves like other System Generator blocks - it is wired into the design, participates in simulations, and is compiled into hardware. When

System Generator compiles a Black Box block, it automatically wires the imported module and associated files into the surrounding netlist.

The Black Box Interface

Black Box HDL Requirements and Restrictions

Details the requirements and restrictions for VHDL,

Verilog, and EDIF associated with black boxes.

Black Box Configuration Wizard

Describes how to use the Black Box Configuration

Wizard.

Black Box Configuration M-

Function

Describes how to create a black box configuration Mfunction.

HDL Co-Simulation

Configuring the HDL Simulator

Explains how to configure ISE ™ Software or

ModelSim to co-simulate the HDL in the Black Box block.

Co-Simulating Multiple Black

Boxes

Describes how to co-simulate several Black Box blocks in a single HDL simulator session.

Black Box Tutorial Example 1:

Module that Satisfies Black Box

Black Box Tutorial Example 2:

Wrapper to Satisfy Black Box

Module that Needs a VHDL

Black Box Tutorial Example 3:

Importing a VHDL Module

Describes an approach that uses the System Generator

Black Box Configuration Wizard.

Describes an approach that requires that you to provide a VHDL core wrapper. Simulation issues are also addressed.

Describes how to use the Black Box block to import

VHDL into a System Generator design and how to use

ModelSim to co-simulate.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

267

Chapter 4: Importing HDL Modules

Black Box Tutorial Example 4:

Importing a Verilog Module

Black Box Tutorial Example 5:

Dynamic Black Boxes

Black Box Tutorial Example 6:

Simulating Several Black Boxes

Simultaneously

Black Box Tutorial Exercise 7:

Advanced Black Box Example

Using ModelSim

Demonstrates how Verilog black boxes can be used in

System Generator and co-simulated using ModelSim.

Demonstrates dynamic black boxes using a transpose

FIR filter black box that dynamically adjusts to changes in the widths of its inputs.

Demonstrates how several System Generator Black

Box Blocks can be co-simulated simultaneously, using only one ModelSim license while doing so.

Describes how to design a Black Box block with a dynamic port interface and how to configure a black box using mask parameters. Also, describes how to assign generic values based on input port data types and how to save black box blocks in Simulink libraries for later reuse. How to specify custom scripts for

ModelSim HDL co-simulation is also covered.

Black Box HDL Requirements and Restrictions

An HDL component associated with a black box must adhere to the following System

Generator requirements and restrictions:

•

The entity name must not collide with any other entity name in the design.

•

Bi-directional ports are supported in HDL black boxes, however they will not be displayed in the System Generator as ports; they only appear in the generated HDL after netlisting.

•

For Verilog black boxes, the module and port names must be lower case and must follow standard VHDL naming conventions.

•

Any port that is a clock or clock enable must be of type std_logic. (For Verilog black boxes, ports must be of non-vector inputs, e.g., input clk.)

•

Clock and clock enable ports in black box HDL should be expressed as follows: Clock and clock enables must appear as pairs (i.e., for every clock, there is a corresponding clock enable, and vice-versa). Although a black box may have more than one clock port, a single clock source is used to drive each clock port. Only the clock enable rates differ.

•

Each clock name (respectively, clock enable name) must contain the substring clk, for example my_clk_1 and my_ce_1.

•

The name of a clock enable must be the same as that for the corresponding clock, but with ce substituted for clk. For example, if the clock is named src_clk_1, then the clock enable must be named src_ce_1.

•

Falling-edge triggered output data cannot be used.

Black Box Configuration Wizard

System Generator provides a configuration wizard that makes it easy to associate a VHDL or Verilog module to a Black Box block. The Configuration Wizard parses the VHDL or

Verilog module that you are trying to import, and automatically constructs a configuration

M-function based on its findings. It then associates the configuration M-function it produces to the Black Box block in your model. Whether or not you can use the configuration M-function as is depends on the complexity of the HDL you are importing.

268

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Configuration Wizard

Sometimes the configuration M-function must be customized by hand to specify details the configuration wizard misses. Details on the construction of the configuration Mfunction can be found in the topic

Black Box Configuration M-Function

Using the Configuration Wizard

The Black Box Configuration Wizard opens automatically when a new black box block is added to a model.

Note:

Before running the Configuration Wizard, ensure the VHDL or Verilog you are importing meets the specified

Black Box HDL Requirements and Restrictions

For the Configuration Wizard to find your module, the model must be saved in the same directory as the module you are trying to import. This means, in particular, that the model must be saved to same directory.

Note:

The wizard only searches for .vhd and .v files in the same directory as the .mdl file. If the wizard does not find any files it issues a warning and the black box is not automatically configured.

The warning looks like the following:

After searching the model's directory for .vhd and .v files, the Configuration Wizard opens a new window that lists the possible files that can be imported. An example screenshot is shown below:

You can select the file you would like to import by selecting the file, and then pressing the

Open

button. At this point, the configuration wizard generates a configuration M-function and associates it with the black box block.

Note:

The configuration M-function is saved in the model's directory as <module>_config.m, where <module> is the name of the module that you are importing.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

269

Chapter 4: Importing HDL Modules

Configuration Wizard Fine Points

The configuration wizard automatically extracts certain information from the imported module when it is run, but some things must be specified by hand. These things are described below:

Note:

The configuration function is annotated with comments that instruct you where to make these changes.

•

If your model has a combinational path, you must call the tagAsCombinational method of the block's SysgenBlockDescriptor object.

•

The Configuration Wizard only knows about the top-level entity that is being imported. There are typically other files that go along with this entity. These files must be added manually in the configuration M-function by invoking the addFile method for each additional file.

•

The Configuration Wizard creates a single-rate black box. This means that every port on the black box runs at the same rate. In most cases, this is acceptable. You may want to explicitly set port rates, which can result in a faster simulation time.

Black Box Configuration M-Function

An imported module is represented in System Generator by a Black Box block. Information about the imported module is conveyed to the black box by a configuration M-function.

This function defines the interface, implementation, and the simulation behavior of the black box block it is associated with. More specifically, the information a configuration Mfunction defines includes the following:

•

Name of the top-level entity for the module;

•

VHDL or Verilog language selection;

•

Port descriptions;

•

Generics required by the module;

•

Clocking and sample rates;

•

Files associated with the module;

•

Whether the module has any combinational paths.

The name of the configuration M-function associated with a black box is specified as a parameter in the black box parameters dialog box (parity_block_config.m in the example shown below).

270

Configuration M-functions use an object-based interface to specify black box information.

This interface defines two objects, SysgenBlockDescriptor and SysgenPortDescriptor.

When System Generator invokes a configuration M-function, it passes the function a block descriptor: www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Configuration M-Function

function sample_block_config(this_block)

A SysgenBlockDescriptor object provides methods for specifying information about the black box. Ports on a block descriptor are defined separately using port descriptors.

Language Selection

The black box can import VHDL and Verilog modules. SysgenBlockDescriptor provides a method, setTopLevelLanguage, that tells the black box what type of module you are importing. This method should be invoked once in the configuration M-function. The following code shows how to select between the VHDL and Verilog languages.

VHDL Module: this_block.setTopLevelLanguage('VHDL');

Verilog Module: this_block.setTopLevelLanguage('Verilog');

Note:

The Configuration Wizard automatically selects the appropriate language when it generates a configuration M-function.

Specifying the Top-Level Entity

You must tell the black box the name of the top-level entity that is associated with it.

SysgenBlockDescriptor provides a method, setEntityName, which allows you to specify the name of the top-level entity.

Note:

Use lower case text to specify the entity name.

For example, the following code specifies a top-level entity named foo. this_block.setEntityName('foo');

Note:

The Configuration Wizard automatically sets the name of the top-level entity when it generates a configuration M-function.

Defining Block Ports

The port interface of a black box is defined by the block's configuration M-function. Recall that black box ports are defined using port descriptors. A port descriptor provides methods for configuring various port attributes, including port width, data type, binary point, and sample rate.

Adding New Ports

When defining a black box port interface, it is necessary to add input and output ports to the block descriptor. These ports correspond to the ports on the module you are importing.

In your model, the black box block port interface is determined by the port names that are declared on the block descriptor object. SysgenBlockDescriptor provides methods for adding input and output ports:

Adding an input port: this_block.addSimulinkInport('din');

Adding an output port: this_block.addSimulinkOutport('dout');

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

271

Chapter 4: Importing HDL Modules

The string parameter passed to methods addSimulinkInport and addSimulinkOutport specifies the port name. These names should match the corresponding port names in the imported module.

Note:

Use lower case text to specify port names.

Adding a bidirectional port: config_phase = this_block.getConfigPhaseString; if (strcmpi(config_phase,'config_netlist_interface')) this_block.addInoutport('bidi');

% Rate and type info should be added here as well end

Bi-directional ports are supported only during the netlisting of a design and will not appear on the System Generator diagram; they only appear in the generated HDL. As such, it is important to only add the bi-drirectional ports when System Generator is generating the HDL. The if-end conditional statement is guarding the execution of the code to add-in the bi-directional port.

It is also possible to define both the input and output ports using a single method call. The setSimulinkPorts method accepts two parameters. The first parameter is a cell array of strings that define the input port names for the block. The second parameter is a cell array of strings that define the output port names for the block.

Note:

The Configuration Wizard automatically sets the port names when it generates a configuration M-function

Obtaining a Port Object

Once a port has been added to a block descriptor, it is often necessary to configure individual attributes on the port. Before configuring the port, you must obtain a descriptor for the port you would like to configure. SysgenBlockDescriptor provides methods for accessing the port objects that are associated with it. For example, the following method retrieves the port named din on the this_block descriptor:

Accessing a SysgenPortDescriptor object: din = this_block.port('din');

In the above code, an object din is created and assigned to the descriptor returned by the port

function call.

SysgenBlockDescriptor also provides methods, inport and outport, that return a port object given a port index. A port index is the index of the port (in the order shown on the block interface) and is some value between 1 and the number of input/output ports on the block. These methods are useful when you need to iterate through the block's ports (e.g., for error checking).

Configuring Port Types

SysgenPortDescriptor provides methods for configuring individual ports. For example, assume port dout is unsigned, 12 bits, with binary point at position 8. The code below shows one way in which this type can be defined.

dout = this_block.port('dout'); dout.setWidth(12); dout.setBinPt(8); dout.makeUnsigned();

The following also works: dout = this_block.port('dout');

272

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Configuration M-Function

dout.setType('Ufix_12_8');

The first code segment sets the port attributes using individual method calls. The second code segment defines the signal type by specifying the signal type as a string. Both code segments are functionally equivalent.

The black box supports HDL modules with 1-bit ports that are declared using either single bit port (e.g., std_logic) or vectors (e.g., std_logic_vector(0 downto 0)) notation. By default,

System Generator assumes ports to be declared as vectors. You may change the default behavior using the useHDLVector method of the descriptor. Setting this method to true tells System Generator to interpret the port as a vector. A false value tells System

Generator to interpret the port as single bit. dout.useHDLVector(true); % std_logic_vector dout.useHDLVector(false); % std_logic

Note:

The Configuration Wizard automatically sets the port types when it generates a configuration

M-function.

Configuring Bi-Directional Ports for Simulation

Bi-directional ports (or inout ports) are supported only during the generation of the HDL netlist, that is, bi-directional ports will not show up in the System Generator diagram. By default, bi-directional ports will be driven with 'X' during simulation. It is possible to overwrite this behavior by associating a data file to the port. Be sure to guard this code since bi-directional ports can only be added to a block during the config_netlist_interface phase.

(strcmpi(this_block.getConfigPhaseString,'config_netlist_interface'))

bidi_port = this_block.port('bidi');

bidi_port.setGatewayFileName('bidi.dat'); end

In the above example, a text file "bidi.dat" is used during simulation to provide stimulation to the port. The data file should be a text file, where each line represents the signal driven on the port at each simulation cycle. For example, a 3-bit bi-directional port that is simulated for 4 cycles might have the following data file:

ZZZ

110

011

XXX

Simulation will return with an error if the specified data file cannot be found.

Configuring Port Sample Rates

The black box block supports ports that have different sample rates. By default, the sample rate of an output port is the sample rate inherited from the input port (or ports, if the inputs run at the same sample rate). Sometimes it is necessary to explicitly specify the sample rate of a port (e.g., if the output port rate is different than the block's input sample rate).

Note:

When the inputs to a black box have different sample rates, you must specify the sample rates of every output port.

SysgenPortDescriptor provides a method, setRate, which allows you to explicitly set the rate of a port.

Note:

The rate parameter passed to the setRate method is not necessarily the Simulink sample rate of that the port runs at. Instead, it is a positive Integer value that defines the ratio between the desired

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

273

Chapter 4: Importing HDL Modules port sample period and the Simulink system clock period defined by the System Generator block dialog box.

Assume you have a model in which the Simulink system period value for the model is defined as 2 sec. Also assume, the example dout port is assigned a rate of 3 by invoking the setRate method as follows: dout.setRate(3);

A rate of 3 means that a new sample is generated on the dout port every 3 Simulink system periods. Since the Simulink system period is 2 sec, this means the Simulink sample rate of the port is 3 x 2 = 6 sec.

Note:

If your port is a non-sampled constant, you may define it as so in the configuration M-function using the setConstant method of SysgenPortDescriptor. You can also define a constant by passing

Inf to the setRate method.

Dynamic Output Ports

A useful feature of the black box is its ability to support dynamic output port types and rates. For example, it is often necessary to set an output port width based on the width of an input port. SysgenPortDescriptor provides member variables that allow you to determine the configuration of a port. You can set the type or rate of an output port by examining these member variables on the block's input ports.

For example, you can obtain the width and rate of a port (in this case din) as follows: input_width = this_block.port('din').width; input_rate = this_block.port('din').rate;

Note:

A black box's configuration M-function is invoked at several different times when a model is compiled. The configuration function may be invoked before the data types and rates have been propagated to the black box.

The SysgenBlockDescriptor object provides Boolean member variables inputTypesKnown

and inputRatesKnown that tell whether the port types and rates have been propagated to the block. If you are setting dynamic output port types or rates based on input port configurations, the configuration calls should be nested inside conditional statements that check that values of inputTypesKnown and inputRatesKnown

The following code shows how to set the width of a dynamic output port dout to have the same width as input port din: if (this_block.inputTypesKnown) dout.setWidth(this_block.port('din').width); end

Setting dynamic rates works in a similar manner. The code below sets the sample rate of output port dout to be twice as slow as the sample rate of input port din: if (this_block.inputRatesKnown) dout.setRate(this_block.port('din').rate*2); end

Black Box Clocking

In order to import a multirate module, you must tell System Generator information about the module's clocking in the configuration M-function. System Generator treats clock and clock enables differently than other types of ports. A clock port on an imported module must always be accompanied by a clock enable port (and vice versa). In other words, clock

274

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Configuration M-Function

and clock enables must be defined as a pair, and exist as a pair in the imported module.

This is true for both single rate and multirate designs.

Note:

Although clock and clock enables must exist as pairs, System Generator drives all clock ports on your imported module with the FPGA system clock. The clock enable ports are driven by clock enable signals derived from the FPGA system clock.

SysgenBlockDescriptor provides a method, addClkCEPair, which allows you to define clock and clock enable information for a black box. This method accepts three parameters.

The first parameter defines the name of the clock port (as it appears in the module). The second parameter defines the name of the clock enable port (also as it appears in the module).

The port names of a clock and clock enable pair must follow the naming conventions provided below:

•

The clock port must contain the substring clk

•

The clock enable must contain the substring ce

•

The strings containing the substrings clk and ce must be the same (e.g., my_clk_1 and my_ce_1).

The third parameter defines the rate relationship between the clock and the clock enable port. The rate parameter should not be thought of as a Simulink sample rate. Instead, this parameter tells System Generator the relationship between the clock sample period, and the desired clock enable sample period. The rate parameter is an integer value that defines the ratio between the clock rate and the corresponding clock enable rate.

For example, assume you have a clock enable port named ce_3 that would like to have a period three times larger than the system clock period. The following function call establishes this clock enable port: addClkCEPair('clk_3','ce_3',3);

When System Generator compiles a black box into hardware, it produces the appropriate clock enable signals for your module, and automatically wires them up to the appropriate clock enable ports.

Combinational Paths

If the module you are importing has at least one combinational path (i.e., a change on any input can effect an output port without a clock event), you must indicate this in the configuration M-function. SysgenBlockDescriptor object provides a tagAsCombinational method that indicates your module has a combinational path. It should be invoked as follows in the configuration M-function: this_block.tagAsCombinational;

Specifying VHDL Generics and Verilog Parameters

You may specify a list of generics that get passed to the module when System Generator compiles the model into HDL. Values assigned to these generics can be extracted from mask parameters and from propagated port information (e.g., port width, type, and rate).

This flexible means of generic assignment allows you to support highly parametric modules that are customized based on the Simulink environment surrounding the black box.

The addGeneric method allows you to define the generics that should be passed to your module when the design is compiled into hardware. The following code shows how to set a VHDL Integer generic, dout_width, to a value of 12.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

275

Chapter 4: Importing HDL Modules addGeneric('dout_width','Integer','12');

It is also possible to set generic values based on port on propagated input port information

(e.g., a generic specifying the width of a dynamic output port).

Because a black box's configuration M-function is invoked at several different times when a model is compiled, the configuration function may be invoked before the data types (or rates) have been propagated to the black box. If you are setting generic values based on input port types or rates, the addGeneric calls should be nested inside a conditional statement that checks the value of the inputTypesKnown or inputRatesKnown variables. For example, the width of the dout port can be set based on the value of din as follows: if (this_block.inputTypesKnown)

% set generics that depend on input port types this_block.addGeneric('dout_width', ... this_block.port('din').width); end

Generic values can be configured based on mask parameters associated with a block box.

SysgenBlockDescriptor provides a member variable, blockName, which is a string representation of the black box's name in Simulink. You may use this variable to gain access the black box associated with the particular configuration M-function. For example, assume a black box defines a parameter named init_value. A generic with name init_value

can be set as follows: simulink_block = this_block.blockName; init_value = get_param(simulink_block,'init_value'); this_block.addGeneric('init_value', 'String', init_value);

Note:

You can add your own parameters (e.g., values that specify generic values) to the black box by doing the following:

•

Copy a black box into a Simulink library or model;

•

Break the link on the black box;

•

Add the desired parameters to the black box dialog box.

Error Checking

It is often necessary to perform error checking on the port types, rates, and mask parameters of a black box. SysgenBlockDescriptor provides a method, setError, which allows you to specify an error message that is reported to the user. The string parameter passed to setError is the error message that is seen by user.

276

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Configuration M-Function

Black Box API

SysgenBlockDescriptor Member Variables

Type

String

Integer

Member

entityName blockName numSimulinkInports numSimulinkOutports

Boolean

Array of

Doubles

Boolean

Cell Array of

Strings inputTypesKnown inputRatesKnown inputRates error errorMessages

Description

Name of the entity or module.

Name of the black box block.

Number of input ports on black box.

Number of output ports on the black box.

true if all input types are defined, and false otherwise.

true if all input rates are defined, and false otherwise.

Array of sample periods for the input ports (indexed as in inport(indx)).

Sample period values are expressed as integer multiples of the Simulink

System Period value specified by the master System Generator block true if an error has been detected, and false otherwise.

Array of all error messages for this block.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

277

Chapter 4: Importing HDL Modules

SysgenBlockDescriptor Methods

Method

setTopLevelLanguage(language) setEntityName(name) addSimulinkInport(pname) addSimulinkOutport(pname) setSimulinkPorts(in,out) addInoutport(pname) tagAsCombinational() addClkCEPair(clkPname, cePname, rate) port(name) inport(indx) outport(indx)

Description

Declares language for the top-level entity (or module) of the black box. language should be

'VHDL' or 'Verilog'.

Sets name of the entity or module.

Adds an input port to the black box. pname tells the name the port should have.

Adds an output port to the black box. pname tells the name the port should have.

Adds input and output ports to the black box. in

(respectively, out) is a cell array whose element tell the names to use for the input (resp., output) ports.

Adds a bi-directional port to the black box. pname spcefies the name the port should have. Bi- directional ports can only be added during the

'config_netlist_interface' phase of configuration.

Indicate that the block has a combinational path (i.e., direct feedthrough) from an input port to an output port.

Defines a clock/clock enable port pair for the block. clkPname and cePname tell the names for the clock and clock enable ports respectively. rate, a double, tells the rate at which the port pair runs. The rate must be a positive integer. Note the clock

(respectively, clock enable) name must contain the substring clk (resp., ce). The names must be parallel in the sense that the clock enable name is obtained from the clock name by replacing clk with ce.

Returns the SysgenPortDescriptor that describes a given input port. indx tells the index of the port to look for, and should be between 1 and numInputPorts.

Returns the SysgenPortDescriptor that describes a given output port. indx tells the index of the port to look for, and should be between 1 and numOutputPorts.

278

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Configuration M-Function

Method

addGeneric(identifier, value) addGeneric(identifier, type, value) addFile(fn) getDeviceFamilyName() getConfigPhaseString setSimulatorCompilationScript

(script) setError(message)

Description

Defines a generic (or parameter if using Verilog) for the block. identifier is a string that tells the name of the generic. value can be a double or a string. The type of the generic is inferred from value's type. If value is an integral double, e.g., 4.0, the type of the generic is set to integer. For a non-integral double, the type is set to real. When value is a string containing only zeros and ones, e.g., `0101', the type is set to bit_vector. For any other string value the type is set to string.

Explicitly specifies the name, type, and value for a generic (or parameter if using Verilog) for the block.

All three arguments are strings. identifier tells the name, type tells the type, and value tells the value.

Adds a file name to the list of files associated to this black box. fn is the file name. Ordinarily, HDL files are associated to black boxes, but any sorts of files are acceptable. VHDL (respectively, Verilog) file names should end in .vhd (resp., .v). The order in which file names are added is preserved, and becomes the order in which HDL files are compiled.

File names can be absolute or relative. Relative file names are interpreted with respect to the location of the .mdl or library .mdl for the design.

Gets the name of the FPGA device corresponding to the Blackbox.

Returns the current configuration phase as a string.

A valid return string includes: config_interface, config_rate_and_type, config_post_rate_and_type, config_simulation, config_netlist_interface and config_netlist.

Overrides the default HDL co-simulation compilation script that the black box generates. script tells the name of the script to use. This method can, for example, be used to short-circuit the compilation phase for repeated simulations where the HDL for the black box remains unchanged.

Indicates that an error has occurred, and records the error message. message gives the error message.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

279

Chapter 4: Importing HDL Modules

SysgenPortDescriptor Member Variables

Type

String

Integer

Member

name simulinkPortNumber

Boolean

String

Boolean

Integer

Boolean

Double typeKnown type isBool isSigned isConstant width binpt rateKnown rate

Description

Tells the name of the port.

Tells the index of this port in Simulink.

Indexing starts with 1 (as in Simulink).

True if this port's type is known, and false otherwise.

Type of the port, e.g., UFix_<n>_<b>,

Fix_<n>_<b>, or Bool

True if port type is Bool, and false otherwise.

True if type is signed, and false otherwise.

True if port is constant, and false otherwise.

Tells the port width.

Tells the binary point position, which must be an integer in the range

0..width.

True if the rate is known, and false otherwise.

Tells the port sample time. Rates are positive integers expressed as MATLAB doubles. A rate can also be infinity, indicating that the port outputs a constant.

SysgenPortDescriptor Methods

Method

setName(name) setSimulinkPortNumber(num) setType(typeName) setWidth(w) setBinpt(bp) makeBool() makeSigned() makeUnsigned()

Description

Sets the HDL name to be used for this port.

Sets the index associated with this port in Simulink. num tells the index to assign. Indexing starts with 1

(as in Simulink).

Sets the type of this port to type. Type must be one of

Bool, UFix_<n>_<b> , Fix_<n>_<b> , signed or unsigned. The last two choices leave the width and binary point position unchanged.

Sets the width of this port to w.

Sets the binary point position of this port to bp.

Makes this port Boolean.

Makes this port signed.

Makes this port unsigned.

280

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

HDL Co-Simulation

Method

setConstant() setGatewayFileName(filename) setRate(rate) useHDLVector(s)

HDLTypeIsVector()

Description

Makes this port constant

Sets the dat file name that will be used in simulations and test-bench generation for this port. This function is only meant for use with bi-directional ports so that a hand written data file can be used during simulation. Setting this parameter for input or output ports is invalid and will be ignored.

Assigns the rate for this port. rate must be a positive integer expressed as a MATLAB double or Inf for constants.

Tells whether a 1-bit port is represented as single-bit

(ex: std_logic) or vector (ex: std_logic_vector(0 downto 0)).

Sets representation of the 1-bit port to std_logic_vector(0 downto 0).

HDL Co-Simulation

Introduction

This topic describes how a mixed language/mixed flow design that includes Xilinx blocks,

HDL modules, and a Simulink block diagram can be simulated in its entirety.

System Generator simulates black boxes by automatically launching an HDL simulator, generating additional HDL as needed (analogous to an HDL testbench), compiling HDL, scheduling simulation events, and handling the exchange of data between the Simulink and the HDL simulator. This is called HDL co-simulation.

Configuring the HDL Simulator

Black box HDL can be co-simulated with Simulink using the System Generator interface to either ISE™ Simulator or the ModelSim simulation software from Model Technology, Inc.

ISE Simulator

To use the ISE™ Simulator for co-simulating the HDL associated with the black box, select

ISE Simulator

as the option for the Simulation mode parameter on the black box as shown in the following figure. The model is then ready to be simulated and the HDL cosimulation takes place automatically.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

281

Chapter 4: Importing HDL Modules

282

ModelSim Simulator

To use the ModelSim simulator by Model Technology, Inc., you must first add the

ModelSim block that appears in the Tools library of the Xilinx Blockset to your Simulink diagram.

For each black box that you wish to have co-simulated using ModelSim simulator, you need to open its block parameterization dialog and set it to use the ModelSim session represented by the black box that was just added. You do this by making the following two settings: www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

HDL Co-Simulation

1. Change the Simulation Mode field from Inactive to External co-simulator.

2. Enter the name of the ModelSim block (e.g., ModelSim) in the HDL Co-Simulator to use field.

The block parameter dialog for the ModelSim block includes some parameters that you can use to control various options for the ModelSim session. See the Modelsim block help pages for details. The model is then ready to be simulated with these options, and the HDL co-simulation takes place automatically.

Co-Simulating Multiple Black Boxes

System Generator allows many black boxes to share a common ModelSim co-simulation session. I.e., many black boxes can be set to "use" the same ModelSim block. In this case,

System Generator automatically combines all black box HDL components into a single shared top-level co-simulation component. This is transparent to the user. It does mean, however, that only one ModelSim simulation license is needed to co-simulate several black boxes in the Simulink simulation.

For an example of how to do this, see

Simulating Several Black Boxes Simultaneously .

Multiple black boxes can also be co-simulated with ISE™ Simulator by just selecting ISE

Simulator as the option for Simulation mode on each black box.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

283

Chapter 4: Importing HDL Modules

Black Box Examples

Black Box Tutorial Example 1:

Module that Satisfies Black Box

Black Box Tutorial Example 2:

Wrapper to Satisfy Black Box

Module that Needs a VHDL

Black Box Tutorial Example 3:

Importing a VHDL Module

Describes an approach that uses the System Generator

Black Box Configuration Wizard.

Describes an approach that requires that you to provide a VHDL core wrapper. Simulation issues are also addressed.

Black Box Tutorial Example 4:

Importing a Verilog Module

Black Box Tutorial Example 5:

Dynamic Black Boxes

Black Box Tutorial Example 6:

Simulating Several Black Boxes

Simultaneously

Black Box Tutorial Exercise 7:

Advanced Black Box Example

Using ModelSim

Describes how to use the Black Box block to import

VHDL into a System Generator design and how to use

ModelSim to co-simulate.

Demonstrates how Verilog black boxes can be used in

System Generator and co-simulated using ModelSim.

Demonstrates dynamic black boxes using a transpose

FIR filter black box that dynamically adjusts to changes in the widths of its inputs.

Demonstrates how several System Generator Black

Box Blocks can be co-simulated simultaneously, using only one ModelSim license while doing so.

ModelSim HDL co-simulation is also covered.

Importing a Xilinx Core Generator Module

This topic describes two different ways of importing Xilinx CORE Generator™ modules, as black boxes, into System Generator. The first example shows how to import blocks

which satisfy Black Box HDL Requirements and Restrictions

. The second example shows

284

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

how to write a VHDL wrapper to import CORE Generator™ modules as black boxes. The flow graph below illustrates the process of importing CORE generator modules.

Black Box Tutorial Example 1: Importing a Core Generator Module that

Satisfies Black Box HDL Requirements

Start Core Generator and open the the following Core Generator project file:

<sysgen_tree>/examples/coregen_import/example1/coregen_import_e xample1.cgp

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

285

Chapter 4: Importing HDL Modules

Double click the CORDIC 3.0 icon to launch the customization GUI

Parameterize and generate the CORDIC 3.0 core with component name cordic_sincos

, a functional Selection of Sin and Cos and the remaining options set to be the default values as shown below:

286

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

Click Generate. Core Generator produces the following files after generation:

♦

♦ cordic_sincos.edn

cordic_sincos.vhd

cordic_sincos.vho

cordic_sincos.xco

: Implementation netlist

: VHDL wrapper for behavioral simulation

: Core instantiation template

: Parameters selected for core generation

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

287

Chapter 4: Importing HDL Modules

Start Simulink and open the design file

(<sysgen_tree>/examples/coregen_import/example1/coregen_import_ example1.mdl

)

Drag and drop the black box from the "Basic Elements" library into the model coregen_import_example1.mdl

. Select cordic_sincos.vhd for the top-level

HDL file.

Connect the input and output ports of the black box to the open wires.

Open the cordic_sincos_config.m file, and add the EDIF netlist to the black box file list as shown below. This file will get included as part of the System Generator netlist for the design when it is netlisted.

288

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

Open the black box parameterization GUI and select ISE Simulator for the simulation mode.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

289

Chapter 4: Importing HDL Modules

10. Press the Simulate button to compile and co-simulate the CORDIC core using the ISE simulator. The simulation results are as shown below.

290

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

Black Box Tutorial Example 2: Importing a Core Generator Module that

Needs a VHDL Wrapper to Satisfy Black Box HDL Requirements

Start Core Generator and open the Core Generator following project file:

<sysgen_tree>/examples/coregen_import/example2/coregen_import_e xample2.cgp

Double click the MAC FIR Filter 5.1 icon to launch the customization GUI.

Customize and generate the MAC FIR Filter 5.1 core with the following parameters:

♦

Component Name: mac_fir_8tap

Number of Taps: 8

Impulse Response: Symmetric

Load Coefficients: mac_fir_8tap.coe file located in sysgen directory

System Clock Rate: 100 Mhz, Input rate: 25 Mhz

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

291

Chapter 4: Importing HDL Modules

♦

Other parameters are set to default values.

292

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

293

Chapter 4: Importing HDL Modules

294

Core Generator produces the following files after generation:

♦

♦ mac_fir_8tap.edn

mac_fir_8tap.vhd

mac_fir_8tap.vho

mac_fir_8tap.xco

mac_fir_8tap.mif

: Implementation netlist

: VHDL wrapper for behavioral simulation

: Core instantiation template

: Parameters selected for core generation

DATA_COEF_BUFFER_B1.mif: Memory initialization files for functional simulation

, DATA_COEF_BUFFER_A1.mif,

Since the MACFIR core does not have a ce port and the System Generator blackbox requires a clk, ce pair, you will need to specify a core wrapper to add a ce port to the top level.

Open the following empty template wrapper file:

<sysgen_tree>/examples/coregen_import/example2/ mac_fir_8tap_wrapper.vhd

This file contains an empty entity declaration.

Modify the template wrapper according to the instructions below:

♦

Open the mac_fir_8tap.vho file.

Copy the component declaration from mac_fir_8tap.vho and paste it in mac_fir_8tap_wrapper.vhd

in the component declaration area.

(after -- Add Component Declaration from VHO file ------)

♦

Copy the core instantiation template from mac_fir_8tap.vho and paste it in mac_fir_8tap_wrapper.vhd

in the architecture body.

(after ------------- ADD INSTANTIATION Template -----) www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

♦

Copy the port declaration for the component mac_fir_8tap and paste it for the mac_fir_8tap

entity declaration

(after ---- Add Port declaration for entity ----)

Add the ce port to the top-level entity declaration, and change the case of the CLK port to clk.

Start Simulink and open the following design file:

<sysgen_tree>/examples/coregen_import/example2/coregen_import_e xample2.mdl

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

295

Chapter 4: Importing HDL Modules

Drag and drop the black box from the "Basic Elements" library in the coregen_import_example2.mdl

. Select mac_fir_8tap_wrapper.vhd for the top-level HDL file.

10. Connect up the black box to the open wires.

11. Open the mac_fir_8tap_wrapper_config.m file, and add the VHDL file, EDIF netlist and MIF files to the black box file list as shown below. These files get included as part of the System Generator netlist for the design when it is generated.

296

Note:

The order in which the files are added in the configuration function is the order in which they get compiled during synthesis and simulation.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

12. Open the black box parameterization GUI and select the ISE Simulator for simulation mode.

13. Press the Simulate button to compile and co-simulate the FIR core using the ISE simulator. The simulation results are as shown below.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

297

Chapter 4: Importing HDL Modules

Importing a VHDL Module

Black Box Tutorial Example 3: Importing a VHDL Module

This topic explains how to use the black box to import VHDL into a System Generator design and how to use ModelSim to co-simulate the VHDL module.

From the MATLAB console, change the directory to

<sysgen_tree>/examples/black_box/intro

The following files are located in this directory:

♦ black_box_intro.mdl

- A Simulink model containing an example black box.

♦ transpose_fir.vhd

- Top-level VHDL for a transpose form FIR filter. This file is the VHDL that is associated with the black box.

♦ mac.vhd

– Multiply and add component used to build the transpose FIR filter.

Open the black_box_intro model from the MATLAB command window by typing

>> black_box_intro

Open the subsystem named Transpose FIR Filter Black Box. At this point, the subsystem contains two inports and one outport. The black box subsystem is shown below:

298

Go to the Simulink Library Browser and add a black box block to this subsystem. The black box is located in the Xilinx Blockset's Basic Elements library. The Black Box

Configuration Wizard is automatically invoked when a new black box is added to the subsystem. A browser window appears that lists the VHDL source files that can be www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

associated with the black box. From this window, select the top -level VHDL file transpose_fir.vhd

. This is illustrated in the figure below:

Note:

The wizard will only run if the black box is added to a model that has been saved to a file. If the model has not been saved, the wizard does not know where to search for files and System

Generator will instead display a warning that looks like the following:

The wizard parses the VHDL to generate a configuration M-function for the black box.

This is a MATLAB script that, among other things, associates the black box to the

VHDL and creates black box ports. Once the function has run, the ports on the black box match those in the top-level VHDL entity (not including clock and clock enable ports). This is illustrated below:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

299

Chapter 4: Importing HDL Modules

Be aware of the following rules when working this example:

♦

A synchronous HDL design that is associated with a black box must have one or more clock and clock enable ports. These ports must occur in pairs, one clock for each clock enable, and vice-versa. Each of these ports must be of type std_logic.

The name of the clock port must contain the substring clk. The name of the clock enable port must be the same as the name of the clock port, but with ce substituted for clk.

♦

The clock enable port has a specific meaning to System Generator and is not a general purpose user enable for the block. Refer to the topic

Black Box HDL

Requirements and Restrictions

for details.

Double click on the black box block. The dialog box shown below appears:

300

The following are the fields in the dialog box:

♦

Block configuration M-function

- This specifies the name of the configuration Mfunction for the black box. In this example, the field contains the name of the function that was generated by the Configuration Wizard. By default, the black box uses the function the wizard produces. You can, however, substitute one you produce yourself. For more information on the configuration M-function, refer to

the topic Black Box Configuration M-Function .

Simulation mode

- There are three simulation modes:

Inactive

- When the mode is Inactive, the black box participates in the simulation by ignoring its inputs and producing zeros. This setting is typically used when a separate simulation model is available for the black box, and the model is wired in parallel with the black box using a simulation multiplexer.

Black Box Tutorial Example 1: Importing a Core Generator

Module that Satisfies Black Box HDL Requirements

shows how this is accomplished.

ISE Simulator

- When the mode is ISE Simulator, simulation results for the black box are produced using co-simulation on the HDL associated to the black box. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

♦

External co-simulator

- When the mode is External co-simulator, it is necessary to add a ModelSim HDL co-simulation block to the design, and to specify the name of the ModelSim block in the field labeled HDL co-

simulator to use

. In this mode, the black box is simulated using HDL cosimulation.

FPGA Area Estimation

- The numbers entered in this field are estimates of how much of the FPGA is used by the HDL for the black box. These numbers must be entered by hand. The numbers are only needed if you would like to use the resource estimating utilities supplied with System Generator. For more

information, see Resource Estimation .

To continue the tutorial, leave the parameters set as they currently are.

Wire the black box's ports to the corresponding subsystem ports.

Run the simulation by clicking the Simulation Play button and then double click on the scope block. Notice the black box output shown in the Output Signal scope is zero.

This is expected as the black box is configured to be inactive during simulation.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

301

Chapter 4: Importing HDL Modules

Go to the Simulink Library Browser and add a ModelSim block to this subsystem. The

ModelSim block is located in the Xilinx Blockset /Tools library. This block enables the black box to communicate with a ModelSim simulator. Double click on the ModelSim block to open the dialog box shown below:

302

10. Make sure the parameters match those shown in the preceding figure. Close the dialog box.

11. From the Simulink menu, select Port Data Types from the Format menu to display the port types for the black box. Compile the model (Ctrl-d) to ensure the port data types are up to date. Notice that the black box port output type is UFix_26_0. This means it is unsigned, 26 bits wide and has a binary point 0 positions to the left of the least significant bit.

12. Open the configuration M-function transpose_fir_config.m and change the output type from UFix_26_0 to Fix_26_12. The modified line should read: dout_port.setType('Fix_26_12');

13. Edit the configuration M-function to associate an additional HDL file with the black box. Locate the line: this_block.addFile('transpose_fir.vhd');

Immediately above this line, add the following: this_block.addFile('mac.vhd'); www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

14. Save the changes to the configuration M-function and recompile the model (Ctrl-d).

Your subsystem should appear as follows:

15. From the black box block parameter dialog box, change the Simulation mode field from Inactive to External co-simulator. Enter ModelSim in the HDL co-simulator to

use

field. The name in this field corresponds to the name of the ModelSim block that you added to the model. The black box dialog box should appear as follows:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

303

Chapter 4: Importing HDL Modules

16. Run the simulation. A ModelSim command window and waveform viewer opens.

ModelSim simulates the VHDL while Simulink controls the overall simulation. The resulting waveform looks something like the following:

The following warnings received in ModelSim can safely be ignored.

# ** Warning: There is an 'U'|'X'|'W'|'Z'|'-' in an arithmetic operand, the result will be 'X'(es).

# Time: 0 ps Iteration: 0 Instance: / xlcosim_black_box_ex1_down_converter_transpose_fir_filter_bl ack_box_modelsim/ black_box_ex1_down_converter_transpose_fir_filter_black_box_ black_box/g0__22/g_last/m2

They are caused by the black box VHDL not specifying initial values at the start of simulation.

304

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

17. Examine the scope output after the simulation has completed. When the Simulation

Mode was set to Inactive, the Output Signal scope displayed constant zero. Notice the waveform is no longer zero. Instead, Output Signal shows the results from the

ModelSim simulation.

Importing a Verilog Module

This example demonstrates how Verilog black boxes can be used in System Generator and co-simulated using ModelSim. Verilog modules are imported the same way VHDL modules are imported. For more information on how this is done, seethe topics

Black Box

Configuration Wizard

and

Black Box Configuration M-Function

. System Generator provides all of the code that is needed to incorporate Verilog black boxes, both to generate hardware and to co-simulate HDL. System Generator also allows Verilog black boxes to be parameterized. This example demonstrates all of these capabilities. The files for this example are contained in the following directory:

<sysgen_tree>/examples/black_box/example4.

The files are:

• black_box_ex4.mdl

– A Simulink model with two black boxes, one using VHDL and the other using Verilog.

• word_parity_block.vhd

– The VHDL for the combinational portion of the state machine seen in word parity example presented above. This is a purely combinational

(stateless) block that computes the parity of each input word and outputs the parity bit. It has been parameterized with a generic so that it can accept any input type (see the description of dynamic black boxes for a discussion of generics).

• word_parity_block_config.m

– The configuration M-function for the VHDL black box, including the generic setting. The M-function tags this block as combinational so that it simulates correctly in Simulink.

• shutter.v

– The Verilog for a simple synchronous latch. The code has been parameterized so that the input port din can have arbitrary width.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

305

Chapter 4: Importing HDL Modules

• shutter_config.m

– The configuration M-function for the Verilog black box, including the parameter setting. The configuration M-function uses methods referring to VHDL syntax even for configuring Verilog black boxes. Thus for this black box, you have the lines: this_block.setEntityName('shutter'); this_block.addGeneric('din_width', dwidth);

Black Box Tutorial Example 4: Importing a Verilog Module

Navigate into the example4 directory and open the example model.

This is a simple design with two black boxes, one VHDL and the other Verilog. The

VHDL black box computes the parity of each input word, and the Verilog black box latches the words that have odd parity. No Simulink model is used to compute the behavior of the black boxes; instead, HDL co-simulation is used. The example model is shown in the figure below.

306

You must have a license for mixed-mode ModelSim simulation to run this example. If you do and you run the simulation, you will see a ModelSim waveform window that looks like the one captured below. The behavior of both black boxes is shown. You can browse the design structure in ModelSim to see how System Generator has combined the two black boxes. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

Change the input type to an arbitrary type and rerun the simulation. Both black boxes adjust in the appropriate way to the change.

Dynamic Black Boxes

This example extends the transpose FIR filter black box so that it is dynamic, i.e., able to adjust to changes in the widths of its inputs. The example is contained in the directory

<sysgen_tree>/examples/black_box/example3

. For this example to run correctly, you must change your directory (cd within the MATLAB command window) to this directory before launching the example model.

The files contained in this directory are:

• black_box_ex3.mdl

- A Simulink model containing a dynamic black box.

• transpose_fir_parametric.vhd

– The VHDL for the transpose FIR filter.

• mac.vhd

– Multiply and add component used to build the transpose FIR filter.

• transpose_fir_parametric_config.m

– The configuration M-function for the black box.

Black Box Tutorial Example 5: Dynamic Black Boxes

Open the model by typing black_box_ex3 at the MATLAB command prompt.

Run the simulation from the top-level model, and view the results displayed in the scopes.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

307

Chapter 4: Importing HDL Modules

Reduce the number of bits on the gateway Din Gateway In from 16 bits down to 12 and the binary point from 14 to 10, then run the simulation again. Note that both the input and output widths on the black box adjust automatically. The black box subsystem and simulation results should look like those shown below.

308

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

The black box is able to adjust to changes in input width because of its configuration

M-function. To make this work, the M-function must be augmented by hand. Open the

M-function file transpose_fir_parametric.m. The important points are described below.

•

Obtaining data input width: input_bitwidth = this_block.port('din').width;

•

Calculating output width: output_bitwidth = ceil(log2(2^(input_bitwidth-1)*2^(coef_bitwidth-1) * number_of_coef));

•

Setting output data type: dout_port.makeSigned; dout_port.width = output_bitwidth; dout_port.binpt = 12;

•

Passing input and output bit widths to VHDL as generics: this_block.addGeneric('input_bitwidth',this_block.port('din').width); this_block.addGeneric('output_bitwidth',output_bitwidth);

For details concerning the black box configuration M-function, seethe topic Black Box

Configuration M-Function .

If you examine the black box VHDL file transpose_fir_parametric.vhd you see generics input_bitwidth

and output_bitwidth that specify input and output width. These are passed to lower-level VHDL components.

Simulating Several Black Boxes Simultaneously

Several System Generator black boxes can co-simulate simultaneously, using only one

ModelSim license while doing so. The example shown below illustrates this. The files for the example are contained in the directory

<sysgen_tree>/examples/black_box/example2.

The files contained in this directory are:

• black_box_ex2.mdl

: A Simulink model containing two black boxes.

• parity_block.vhd

: VHDL for a simple state machine that tracks the running parity of an 8-bit input word.

• parity_block_config.m

: The configuration M-function for the black boxes. The code has barely been changed from what was produced by the Configuration Wizard: the line that tagged the block as having a combinational feed-through path

(this_block.tagAsCombinational) has been removed.

Black Box Tutorial Example 6: Simulating Several Black Boxes

Simultaneously

Navigate into the example2 directory and open the example model. This is a simple model with two identical black boxes, each implementing a state machine. The state machines compute the running parity of their inputs. One black box is fed the input stream of the model and the other is fed the input stream after it has been serialized and deserialized. Notice that no simulation model is provided for either state machine. Instead,

HDL co-simulation is used to produce simulation results. The ModelSim block provides

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

309

Chapter 4: Importing HDL Modules the connection between the black boxes and ModelSim. The example model is shown in the figure below.

If you run the simulation, you will see a Simulink scope and ModelSim waveform window that look like the figures below. The scope shows that the black boxes produce matching parity results (as expected), but with one delayed from the other by one clock cycle. The waveform window shows the same results, but viewed in ModelSim and expressed in binary. System Generator automatically configures the waveform viewer to display the input and output signals of each black box. You can also browse the design structure in

ModelSim to see how System Generator has elaborated the design to combine the two black boxes.

310

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

Advanced Black Box Example Using ModelSim

The following topics are discussed in this example:

•

How to design a black box with a dynamic port interface;

•

How to configure a black box using mask parameters;

•

How to assign generic values based on input port data types;

•

Saving black box blocks in Simulink libraries for later reuse;

•

How to specify custom scripts for ModelSim HDL co-simulation.

This example also shows a way to view signals coming from a black box. In Simulink, waveforms are typically viewed with a scope. The Simulink scope block serves this purpose and the System Generator WaveScope block is available in versions 8.1 and later.

The waveform viewer in the ModelSim simulator may also be used to view waveforms. In this example, a black box is configured as a specialized ModelSim waveform scope for

Xilinx fixed-point signals. When a model that uses the black box scope is simulated, the signals that drive the black box are displayed in ModelSim.

The files for this example are contained in the directory

<sysgen_tree>/examples/black_box/example5.

The files contained in this directory are:

• black_box_ex5.mdl

: A Simulink model containing a black box scope.

• scope_lib.mdl

: A Simulink library containing the black box waveform viewer.

• scope_config.m

: The configuration M-function for the black box waveform viewer.

• scope1.vhd, scope2.vhd

, scope3.vhd, scope4.vhd: Black box VHDL for the signal scope that accept one, two three, and four input signals, respectively.

• waveform.do

– A script that instructs ModelSim how to display signals during simulation.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

311

Chapter 4: Importing HDL Modules

Black Box Tutorial Exercise 7: Advanced Black Box Example Using

ModelSim

Navigate into the example5 directory and open the example black_box_ex5.mdl file.

The model includes an adder that is driven by two input gateways. The gateways are configured to produce signed 8-bit values, each with six bits to the right of the binary point. Sine wave generators drive the gateways. The model also includes a black box named waveform scope. This is driven by three signals. The first input is driven by the adder. The other two are driven by the inputs to the adder. The ModelSim block enables HDL co-simulation. The example model is shown below.

Simulate the black_box_ex5 model. A ModelSim window opens and ModelSim compiles the files necessary for simulation. After the compilation is complete, both

MATLAB and ModelSim simulations begin. A ModelSim waveform viewer opens and displays four signals. The first input to the block, sig1, is driven by the adder. This

312

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

signal is represented in two ways in the ModelSim viewer – binary and analog. The

ModelSim waveforms for the black_box_ex5 simulation are shown below.

Double click on the Simulink scope in the model. The output is shown below and resembles the analog signal in the ModelSim waveform viewer.

The black box in this example is configured using mask parameters. There are many situations in which this is useful. In this case, the number of black box input ports, i.e., the number of scope inputs, is determined by a mask parameter.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

313

Chapter 4: Importing HDL Modules

Double click on the waveform scope black box. Notice a Number of Input Ports field is included in the block dialog box and is unique to this black box instance. The dialog box is shown below:

314

Change the number of input ports from 3 to 4 and apply the changes. The black box now has an additional input port labeled sig4 and should look like the following:

Every black box has a standard list of mask parameters. The black box in this example has an additional mask parameter nports that stores the number of input ports selected by the user. To change a black box mask it is necessary to disable the link to the library. When a black box is changed in this way, it is best to save the black box in a library. (See the Simulink documentation on libraries for details.) The tutorial library scope_lib.mdl

contains the modified signal scope black box used in this example.

When a black box configuration M-function adds an HDL file, the path to the file can be relative to the directory in which the library is saved. This eliminates the need to copy the HDL into the same directory as the model.

The black box's configuration M-function is invoked whenever the block parameter dialog box is modified. This allows the M-function to check the mask parameters and configure the black box accordingly. In this example, the M-function adjusts the number of block input ports based on the nports parameter specified in the mask.

Open the file scope_config.m that defines the configuration M-function for the example black box. Locate the line: simulink_block = this_block.blockName;

This obtains the Simulink name of the black box and assigns it to the variable simulink_block

. The name is useful because it is the handle that MATLAB functions need to manipulate the block.

Locate the line: nports = eval(get_param(simulink_block,'nports'));

The value of the nports mask parameter is obtained by the get_param command.

The get_param returns a string containing the number of ports. An eval encloses the get_param

and converts the string into an integer that is assigned to the nports variable. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Black Box Examples

Once the number of input ports is determined, the M-function adds the input ports to the black box. The code that does this is shown below. for i=1:nports this_block.addSimulinkInport(sprintf('sig%d',i)); end

There are four VHDL files, named scope1.vhd, scope2.vhd, scope3.vhd, and scope4.vhd

, which the black box in this example can use. The black box associates itself to the one that declares an appropriate number of ports.

The configuration M-function selects the appropriate VHDL file for the black box.

Locate the following line in scope_config.m: entityName = sprintf('scope%d',nports);

The HDL entity name for the black box is constructed by appending the value of nports

to scope. The VHDL is associated with the black box in the following line: this_block.addFile(['vhdl/' entityName '.vhd']);

10. The input port widths for each VHDL entity are assigned using generics. The generic name identifies the input port to which the width is assigned. For example, the width3 generic specifies the width of the third input. In scope_config.m, the generic names and values are set as follows:

% ----------------------------- if (this_block.inputTypesKnown) for i=1:nports width = this_block.inport(i).width; this_block.addGeneric(sprintf('width%d',i),width); end end % if(inputTypesKnown)

% -----------------------------

11. You can change the way ModelSim displays the signal waveforms during simulation by using custom tcl scripts in the ModelSim block. Double click on the ModelSim block in the black_box_ex5 model. The following dialog box appears:

Custom scripts are defined by selecting the Add Custom Scripts checkbox. In this case, a script named waveform.do is specified in the Script to Run after vsim field.

This script contains the ModelSim commands necessary to display the adder output as an analog waveform.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

315

Chapter 4: Importing HDL Modules

316

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Chapter 5 System Generator Compilation Types

There are different ways in which System Generator can compile your design into an equivalent, often lower-level, representation. The way in which a design is compiled depends on settings in the System Generator dialog box. The support of different compilation types provides you the freedom to choose a suitable representation for your design’s environment. For example, an HDL or NGC netlist is an appropriate representation when your design is used as a component in a larger system. If, on the other hand, the complete system is modeled inside System Generator, you may choose to compile your design into an FPGA configuration bitstream. Sometimes you may want to compile your design into an equivalent high-level module that performs a specific function in applications external to System Generator (e.g., ModelSim hardware cosimulation).

HDL Netlist Compilation

NGC Netlist Compilation

Bitstream Compilation

EDK Export Tool

Hardware Co-Simulation

Compilation

Timing Analysis Compilation

Creating Compilation Targets

System Generator uses the HDL Netlist compilation type as the default generation target. More details regarding the HDL Netlist compilation flow can be

found in the topic Compilation Results .

Describes how System Generator can be configured to compile your design into a standalone NGC file.

Describes how System Generator can be configured to compile your design into an FPGA configuration bitstream.

Describes how System Generator can be configured to compile your design into an FPGA configuration bitstream that is appropriate for the selected part.

Describes how System Generator can be configured to compile your design into FPGA hardware that can be used by Simulink and ModelSim.

Describes how to use the System Generator Timing

Analysis tool compilation target.

Describes how to add custom compilation targets to the System Generator block.

317 System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

Chapter 5: System Generator Compilation Types

HDL Netlist Compilation

System Generator uses the HDL Netlist compilation type as the default generation target.

More details regarding the HDL Netlist compilation flow can be found in the sub-topic titled

Compilation Results

As shown below, you may select HDL netlist compilation by left-clicking the Compilation submenu control on the System Generator block dialog box, and select the HDL Netlist target.

NGC Netlist Compilation

The NGC Netlist compilation target allows you to compile your design into a standalone

Xilinx NGC binary netlist file. The NGC netlist file that System Generator produces contains the logical and optional constraint information for your design. This means the

HDL, cores, and constraints file information corresponding to a System Generator design are self-contained within a single file.

If you have chosen to include clock wrapper logic in your design, the netlist file is saved as

<design>_cw.ngc

. Otherwise, the file is saved as <design>.ngc. Here <design> is derived from the portion of the design being compiled. This file can be used as a module in a larger design, or as input to NGDBuild when the netlist constitutes the complete design.

For an example showing how a System Generator design can be used as a component in a larger design, refer to the topic titled

Importing a System Generator Design into a Bigger

System

The NGC compilation target generates an HDL component instantiation template that makes it easy to include your System Generator design as a component in a larger design.

For VHDL compilation, the template is saved as <design>_cw.vho when the clock wrapper is included. Otherwise it is saved as <design>.vho . Alternatively, a .veo extension is used for Verilog compilation. The instantiation template is saved in the design's target directory.

System Generator produces the NGC netlist file by performing the following steps during compilation:

Generates an HDL netlist for the design;

Runs the selected synthesis tool to produce a lower-level netlist. The type of netlist

(e.g., EDIF for Synplify or Synplify Pro, NGC for XST) depends on which synthesis tool is chosen for compilation.

Note:

Note: IO buffers are not inserted in the design during synthesis.

318

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Bitstream Compilation

Combines synthesis results, core netlists, black box netlists, and optionally the constraints files into a single NGC file.

As shown below, you may select the NGC compilation target by left-clicking the

Compilation

submenu control on the System Generator block dialog box, and selecting the

NGC Netlist

target.

You may access additional compilation settings specific to NGC Netlist compilation by clicking on the Settings... button when NGC Netlist is selected as the compilation type in the System Generator block dialog box. Parameters specific to the NGC Netlist Settings dialog box include:

•

Include Clock Wrapper

: Selecting this checkbox tells System Generator whether the clock wrapper portion of your design should be included in the NGC netlist file. Refer

to the topic Compilation Results for more information on the clock wrapper.

Note:

If you exclude the clock wrapper from multirate designs, you will need to drive the clock enable ports with appropriate signals from your own top-level design.

•

Include Constraints File

: Selecting this checkbox tells System Generator whether the constraints file associated with the design should be included in the NGC netlist file.

Note:

When the constraints file is excluded, you should supply your own constraints to ensure the multi-cycle paths in the System Generator design are appropriately constrained.

Bitstream Compilation

The Bitstream compilation type allows you to compile your design into a Xilinx configuration bitstream file that is suitable for the FPGA part that is selected in the System

Generator dialog box. The bitstream file is named <design>_cw.bit and is placed in the design's target directory, where <design> is derived from the portion of the design being compiled.

System Generator produces the bitstream file by performing the following steps during compilation:

Generates an HDL netlist for the design;

Runs the selected synthesis tool to produce a lower-level netlist. The type of netlist

(e.g., EDIF for Synplify Pro, NGC for XST) depends on which synthesis tool is chosen for compilation.

Runs XFLOW to produce a configuration bitstream.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

319

Chapter 5: System Generator Compilation Types

As shown below, you may select the Bitstream compilation by left-clicking the

Compilation

submenu control on the System Generator block dialog box, and selecting the

Bitstream

target.

320

System Generator uses XFLOW to run the tools necessary to produce the configuration bitstream. Execution of XFLOW is broken into two flows, implementation and configuration.

The implementation flow is responsible for compiling the synthesis tool netlist output

(e.g., EDIF or NGC) into a placed and routed NCD file. In summary, the implementation flow performs the following tasks:

Combines synthesis results, core netlists, black box netlists, and constraints files using

NGDBuild.

Runs MAP, PAR, and Trace on the design (in that particular order).

The configuration flow type runs the tools (e.g., BitGen) necessary to create an FPGA BIT file, using the fully elaborated NCD file as input.

XFLOW Option Files

Xilinx ISE™ software includes several example XFLOW options files. From the base directory of your Xilinx ISE software tree, these files are located under the xilinx\data directory. Three commonly used implementation options files include:

• balanced.opt;

• fast_runtime.opt;

• high_effort.opt.

Note:

By default, System Generator uses the balanced.opt file for the implementation flow, and bitgen.opt

file for the configuration flow.

Sometimes you may want to use options files that use settings that differ (e.g., to specify a higher placer effort level in PAR) from the default options provided by the target. In this case, you may create your own options files, or edit the default options files to include your desired settings. The Bitstream settings dialog box allows you to specify options files other than the default files.

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Bitstream Compilation

Additional Settings

You may access additional compilation settings specific to Bitstream compilation by clicking on the Settings... button when Bitstream is selected as the compilation type in the

System Generator block dialog box. Parameters specific to the Bitstream Settings dialog box include:

•

Import Top-level Netlist: Allows you to specify your own top-level netlist into which the System Generator portion of the design is included as a module. You may choose to import your own top-level netlist if you have a larger design that instantiates the

System Generator clock wrapper level as a component. Refer to the

Compilation

Results

topic for more information on the clock wrapper level. This top-level netlist is included in the bitstream file that is generated during compilation. Selecting this checkbox enables the edit fields Top-level Netlist File (EDIF or NGC) and Search Path for Additional Netlist and Constraint Files.

♦

Top-level Netlist File (EDIF or NGC)

: Specifies the name and location of the toplevel netlist file to include during compilation. Note that any HDL components that are used by your top-level (including the top-level itself) must have been previously synthesized into netlist files.

♦

Search Path for Additional Netlist and Constraint Files

: Specifies the directory where System Generator should look for additional netlist and constraint files that go along with the top-level netlist file. System Generator copies all netlist

(e.g., .edn, .edf, .ngc) and constraints files (e.g., .ucf, .xcf, .ncf) into the implementation directory when this directory is specified. If you do not specify a directory, System Generator will only copy the netlist file specified in the Top-

level Netlist File

field.

•

Specify Alternate Clock Wrapper

: Allows you to substitute your own clock wrapper logic in place of the clock wrapper HDL System Generator produces. The clock wrapper level is the top-level HDL file that is created for a System Generator design, and is responsible for driving the clock and clock enable signals in that design.

Sometimes you may want to supply your own clock wrapper, for example, if your design uses multiple clock signals, or if you have a board-specific hardware you would like your design to interface to.

Note:

The name of the alternate clock wrapper file must be named <design>_cw.vhd or

<design>_cw.v

or it will not be used during bitstream generation.

•

XFLOW Option Files

: When a design is compiled for System Generator hardware cosimulation, the command line tool, XFLOW, is used to implement and configure your design for the selected FPGA platform. XFLOW defines various flows that determine the sequence of programs that should be run on your design during compilation.

There are typically multiple flows that must be run in order to achieve the desired output results, which in the case of hardware co-simulation targets, is a configuration bitstream.

♦

Implementation Phase (NBDBuild, MAP, PAR, TRACE)

: Specifies the options file that is used by the implement flow type. By default, System Generator will use the implement options file that is specified by the compilation target.

♦

Configuration Phase (BitGen)

: Specifies the options file that is used by the configuration flow type. By default, System Generator will use the configuration options file that is specified by the compilation target.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

321

Chapter 5: System Generator Compilation Types

Re-Compiling EDK Processor Block Software Programs in Bitstreams

When you perform bitstream compilation on a System Generator design with an EDK

Processor block, the imported EDK project and the shared memories sitting between the

System Generator design and MicroBlaze™ processor are netlisted and included in the resulting bitstream.

System Generator also tries to compile any active software programs inside the imported

EDK project. If the compilation of active software programs succeeds, System Generator invokes the data2bram utility to include the compiled software programs into the resulting bitstream.

Note:

No error or warning message is issued when System Generator encounters failures during software program compilation or when System Generator updates the resulting bitstream with the compiled software programs.

You can modify the software programs in the imported EDK project and use the following command to compile the software programs, and update the System Generator bitstream with the compiled software programs: xlProcBlockCallbacks('updatebitstream', [], xmp_file, bit_file, bmm_file); where

xmp_file

is the pathname to the imported EDK project file

bit_file

is the pathname to the Sysgen bitstream file

bmm_file

is the pathname of the back-annotated BMM file produced by Sysgen during bitstream compilation

If the imported EDK project contains a BMM file named imported_edk_project.bmm,

System Generator creates a back-annotated BMM file named imported_edk_project_bd.bmm

. You should provide the later back-annotated BMM file to the above command in order to update the bitstream properly.

322

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

EDK Export Tool

The EDK Export Tool allows a System Generator design to be exported to a Xilinx

Embedded Development Kit (EDK) project. The EDK Export Tool simplifies the process of creating a peripheral by automatically generating the files required by the EDK.

The EDK Export Tool can be accessed from the System Generator block GUI under the

Compilation pull-down menu – the figure below shows this being done. After the EDK

Export Tool is selected, the Settings… button will be enabled.

Clicking on the Settings… button brings up the EDK export settings dialog.

Pcore options

allow you to do the following:

•

Assign a version number to your pcore

•

Select Pcore under development

This feature works for both FSL- and PLB-based pcore export. When a pcore is marked as Pcore under development, XPS will not cache the HDL produced for this pcore.

This is useful when you are developing pcores in System Generator and testing them out in XPS. You can just enable this checkbox, make changes in System Generator and compiled in XPS. XPS always compiles the generated pcore, so you don’t have to empty the XPS cache which may contain caches of other peripherals, thus slowing down the compile of the final bitstream.

•

Select Enable custom bus interfaces

This feature works for both FSL- and PLB-based pcore export and allows you to create custom bus interfaces that will be understood in XPS.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

323

Chapter 5: System Generator Compilation Types

Creating a Custom Bus Interface for Pcore Export

Consider the following example. In the model below, you have one design that you are going to export as a pcore to XPS. This design has the output ports Pixel Enable, Y, Cr, and

Cb

. You want to group these signals into a bus to simplify the connection in XPS.

2. Select

3. Click

1. Double Click

4. Select

5. Click

6. Enter Data

You follow the sequence in the figure above to bring up the Bus Interface dialog box. In this dialog box, you define a new Bus Interface called vid_out that is marked as a

myVideoBus

Bus Standard and is Bus Type INITIATOR. (Other supported Bus Types include: Target, Master, Slave, Master-slave, Monitor.) Next, in the Port-Bus Mapping table, you list all the gateways that you want in the bus, then give each a Bus Interface

Name. You then Netlist the design as a pcore. Remember that you marked this pcore bus as

INITIATOR

since it contains outputs.

324

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

EDK Export Tool

In another model (shown below), you create corresponding input gateways. You set this up as a TARGET bus giving the bus interface the same Bus Standard myVideoBus. XPS will use the Bus Standard name to match different bus interfaces. XPS will then connect the outputs to the inputs with the same Bus Interface Names.

You export this pcore to the XPS project. When these two pcores are used in the same XPS project, XPS will detect that they have compatible buses and will allow you to connect them if you wish.

Export as Pcore to EDK

When a System Generator design is exported to the EDK, the name of the pcore (processor core) has the postfix "_plbw" appended to the model name if a PLB v6.4 bus is specified.

For example, when a model called mul_accumulate is exported to the EDK, it will be called mul_accumulate_plbw on the EDK side. If Fast Simplex Link is specified, the postfix

“_sm” is appended to the model name.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

325

Chapter 5: System Generator Compilation Types

The following table shows subdirectory structure of the pcore that is generated by System

Generator:

pcore

Subdirectory

Description

data doc hdl netlist src

The data directory contains four files: BBD, PAO, MPD and

TCL.

• The BBD (black-box definition) file tells the EDK what EDN or NGC files are used in the design.

• The PAO (peripheral analyze order) file tells the EDK the analyze order of the HDL files.

• The MPD (Microprocessor Peripheral Description) file tells the EDK how the peripheral will connect to the processor.

• The TCL file is used by LibGen when elaborating software drivers for this peripheral.

Documentation files in HTML format.

The hdl directory contains the hdl files produced by System

Generator.

The netlist directory contains the EDN and NGC files listed by the BBD file

Source files for the software drivers.

System Generator Ports as Top-Level Ports in EDK

Input and output ports created in System Generator are made available to the EDK tool as ports on the peripheral. You may pull these ports to the top-level of the EDK design. This is useful for instance when the System Generator design has ports that go to the input/output pads on the FPGA device.

Supported Processors and Current Limitations

Currently, PLB v4.6 memory-map links and FSL memory-map links to the MicroBlaze™ processor are exported with the EDK Export Tool. There can only be one instance of an

EDK Processor block.

Hardware Co-Simulation Compilation

System Generator can compile designs into FPGA hardware that can be used in the loop with Simulink simulations. This capability is discussed in the topic

Using Hardware Co-

Simulation .

You may select a hardware co-simulation target by left-clicking the Compilation submenu control on the System Generator dialog box, and selecting the desired hardware cosimulation platform. The list of available co-simulation platforms depends on which hardware co-simulation plugins are installed on your system.

Note:

If you have an FPGA platform that is not listed as a compilation target, you may create a new

System Generator compilation target that uses JTAG to communicate with the FPGA hardware. Refer

to the Supporting New Platforms

for more information on how to do this.

Timing Analysis Compilation

Sometimes the hardware created by System Generator may not meet the requested timing requirements. System Generator provides a Timing Analysis tool that can help you resolve timing related issues. The timing analysis tool shows you, both in graphical and textual formats, the slowest system paths and those paths that are failing to meet the timing requirements. This allows you to concentrate on methods of speeding up those paths.

Methods for doing so will be discussed. Underlying the System Generator Timing Analysis tool is Trace, a software application delivered as part of the ISE™ software used to analyze timing paths.

As shown below, you invoke the Timing Analyzer by double-clicking on the System

Generator block and selecting the Timing Analysis option from the Compilation submenu. Specify the Part as you desired. It is important to choose the exact part you wish to target as the size and speed of the part will affect the path delays. Result files will be put in the Target Directory. The value in the FPGA Clock Period box is the value that will be used during place & route:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

327

Chapter 5: System Generator Compilation Types

After filling out the dialog box, click the Generate button and System Generator will perform the following steps:

The design is compiled using Simulink.

The design is netlisted by System Generator into HDL source.

The HDL Synthesis Tool is called to turn the HDL into an EDIF (Synplify/Synplify

Pro) or NGC (XST) netlist.

NGD Build is called to turn the netlist into an NGD netlist file.

The ISE Mapper software is called to map elements of logic together into slices; this creates an NCD file.

The ISE Place & Route software is called to place the slices and other elements on the

Xilinx die and to route the connections between the slices. This creates another NCD file.

The ISE Trace software is called to analyze the NCD and to find the paths with the worst slack. This creates a trace report.

The System Generator Timing Analyzer tool appears, displaying the data from the trace report.

Note:

If timing data is generated using this method and you wish to view it again at a later time, then enter the following command at the MATLAB command line:

>>xlTimingAnalysis('timing') where 'timing' is the name of the target directory in which a prior analysis was carried out.

Timing Analysis Concepts Review

This brief topic is intended for those with little or no knowledge of logic path analysis.

Period and Slack

A timing failure usually means there is a setup time violation in the design. A setup time violation means that a particular signal cannot get from the output of one synchronous element to the input of another synchronous element within the requested clock period and subject to the second synchronous element's setup time requirement. A typical path is shown in this Synplify schematic:

328

The path shown is from the Q output of the register on the left (register3) to the D input of the register on the right (parity_reg). The path goes through two LUTs (lookup tables) that are configured as 4-input XOR gates. This path has two levels of logic. That means that it goes through two separate combinational elements (the two LUTs).

The requested period for this path is 10ns. This path easily meets timing. The second of the two red comma-separated numbers above each logic elements shows the slack for the path. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Timing Analysis Compilation

The slack is the amount of time by which the path 'meets timing'. In this case the slack is

7.79ns. That means that the path could be 7.79ns slower and still meet the 10ns period requirement. A negative slack value indicates that the path does not meet timing and has a setup (or hold) time violation.

Path Analysis Example

Let us examine this path in more detail. The first value on the top of register3 is 0.35ns. This means that the clk-to-out time of the register is 0.35ns, so the data will appear on the Q output 0.35ns after the rising edge of the clock signal. (The clock signal, not shown, drives the C inputs of both registers.)

The input of the LUT y_4[0] shows two numbers on each input. The first is the arrival time of the signal. This value is 0.98ns. This means that the signal arrives at the input 0.98ns after the rising edge of the clock. Therefore the net delay is (0.98ns-0.35ns)=0.63ns. Any path delay is divided into net delays and logic delays. In an FPGA, the net delays are normally the predominant type of delay. This is because the configurable routing fabric of the FPGA requires that a net traverse many delay-inducing switchboxes in order to reach its destination.

The path leaves y_4[0] and travels along another net to y[0]. The first of the two values at the output of y[0] shows the arrival time of the signal at the output of that LUT. This value is 1.62ns. The signal travels along the final net, incurring a net delay of 0.26ns to arrive at the D input of parity_reg at 1.88ns after the clock edge. This register has a required setup

time. The setup time for this register is 0.33ns. This means that the signal must arrive at the

D input 0.33ns before the rising edge of the next clock. Therefore the total path requires

(1.88ns+0.33ns)=2.21ns. Subtracted from 10ns, this yields the 7.79ns slack value.

Clock Skew and Jitter

The net delay values shown here are estimates provided by Synplify. The synthesizer doesn't know the actual net delay values because these are not determined until after the place & route process. An actual path contains other variables which must be accounted for, including clock skew and clock jitter. Clock skew is the amount of time between clock arrival at the source and destination synchronous elements. Clock jitter is a variation of the clock period from cycle to cycle. Jitter is created by the DCMs (digital clock managers) and by other means. The timing analysis is carried out with worst-case values for the given part's delay values, jitter, skew, and temperature derating.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

329

Chapter 5: System Generator Compilation Types

Timing Analyzer Features

Observing the Slow Paths

Clicking on the Slow Paths icon displays the paths with the least slack for each timing constraint. An example is shown below:

330

The top section of the display shows a list of slow paths, while the bottom section of the display shows details of the path that is selected. The elements of this display are explained here:

•

Timing Constraint

: You may opt to view the paths from all timing constraints or just a single constraint. A typical System Generator design has but a single timing constraint which defines the period of the system clock. This is the constraint shown in this example. TS_clk_a5c9593d is the name of the constraint; the (sometimes confusing) suffix is a hash meant to make the identifier unique when multiple System

Generator designs are used as components inside a larger design. The timing group clk_a5c9593 is a group of synchronous logic, again with a hash suffix. The group in this case contains all the synchronous elements in the design. The period of the clock here is 10ns with a 50% duty cycle.

•

Source

: The System Generator block that drives the path.

•

Destination

: This is the System Generator block that is the terminus of the path.

•

Slack

: The slack for this particular path. See the topic entitled Period and Slack for

more details.

•

Delay (Path)

: The delay of the entire path, including the setup time requirement.

•

% Route Delay

: This is the percentage of the path that is consumed by routing (net) delay. The remainder portion of the path is consumed by logic delay.

•

Levels of Logic

: The number of levels of combinatorial logic in the path. The combinatorial logic typically comprises LUTs, F5 muxes, and carry chain muxes.

•

Path Element

: This shows the logic and net elements in the highlighted path.

•

Delay (Element)

: This shows the delay through the logic and net elements in the highlighted path.

•

Type of Delay

: This is the kind of delay incurred by the given path element. These values are defined in the Xilinx part's data sheet. In the example shown above, Tcko is the clk-to-out time of a flip-flop; net is a net delay; Tilo is the delay through a LUT, and

Tas is the setup time of a flip-flop. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Timing Analysis Compilation

You may click on the column headings to reorder the paths or elements according to delay, slack, path name, or other column headings. Failing paths are highlighted in red/pink.

Name Unmunging and Displaying Low-Level Names

Part of the magic of the timing analyzer lies in its ability to perform the un-glorious task of

name unmunging, the task of automatically correlating System Generator components with the low-level component names produced by the Xilinx implementation tools. The names of these components often differ considerably. In fact, the logic blocks and wires that appear in a System Generator diagram may have only a loose relation to the actual logic that gets generated during the synthesis process. The System Generator timing analyzer must correlate the names of logic elements and nets in the trace report to blocks and wires in the System Generator diagram.

The timing analyzer cannot always perform this un-munging process. In the path shown in the screen capture above, path elements #2 and #5 have a question mark displayed in the name field. This means that the timing analyzer could not un-munge the name from the trace report and correlate it to a System Generator block.

To see the actual names from the trace report, check the Display low-level names box. This will show the trace report names. You may be able to correlate them to System Generator elements by observation.

Cross-Probing

Highlighting a path in the Slow Paths view will highlight the blocks in the path in the

System Generator diagram. The path's source and destination blocks, as well as combinational blocks through which the path passes, will be highlighted in red. The diagram below shows how the model appears when the path that has Registerc as its source and parity_reg as its destination is highlighted. The blocks xor_1b, xor_2a, and xor_3a are also highlighted because they are part of the path.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

331

Chapter 5: System Generator Compilation Types

Histogram Charts

Clicking on the Charts icon displays a histogram of the slow paths. This histogram is a useful metric in analyzing the design. You may know that the design will only run at, for example, 99MHz in your part when you wish it to run at 100MHz. But how close is the design to meeting timing and how much work is involved in meeting this requirement?

The histogram will quickly give you an estimate of the work involved. For example, look at the histogram of the results of a simple design below:

This shows that most of the slow paths are concentrated about 1.5ns. The slowest path is about 2.35ns. The numbers at the tops of the bins show the number of paths in each bin.

There is only one path in the bin which encompasses the time range 2.31ns-2.39ns. The bins to the right of it are empty. This shows that the slowest path is an outlier and that if your timing requirement were for a period of, for example, 2ns, you would need only to speed up this single path to meet your timing requirements.

332

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Timing Analysis Compilation

Histogram Detail

The slider bar allows you to adjust the width of the bins in the histogram. This allows you to get more detail about the paths if desired. The display below shows the results of a different design with a larger number of bins than the diagram above:

This diagram shows the paths grouped into three regions, with each forming a rough bell curve distribution. These groups are probably from different portions of the circuit or from different timing constraints that are from different clock regions. If you wish to analyze the paths from a single timing constraint, you may select a single constraint for viewing from the Timing constraint pulldown menu at the top of the display.

Note the bins and portions thereof shown in red. These are the paths that have negative slack; i.e., they do not meet the timing constraint. In this example you can see that some paths have failed but not by a large margin so it seems reasonable that with some work this design could be reworked to meet timing.

Statistics

Clicking on the Statistics icon displays several design statistics, including the number of constraints, paths analyzed, and maximum frequency of the design.

Trace Report

Clicking on the Trace icon shows the raw text report from the Trace program. This file gives considerable detail about the paths analyzed. Each path analyzed contains information

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

333

Chapter 5: System Generator Compilation Types about every net and logic delay, clock skew, and clock uncertainty. The box at the bottom left of this display shows the path name of the timing report.

334

Improving Failing Paths

"Now I have information about my failing paths; but what do I do now?" you may ask yourself. This is the trick for which there is no simple answer, and this is where you may need to delve into the lower-level aspects of FPGA design.

In general, steps that may be taken to meet timing are, in this order:

Change the source design

. Just about any timing problem can be solved by changing the source design and this is the easiest way to speed up the circuit. Unfortunately, this is often the last step taken by designers, who often look for a quick solution such as using a faster part. The source design may be changed in several ways: a.

Pipelining

. This is the surest way to improve speed, but may also be tricky.

Adding pipelining registers increases latency. For designs with feedback, this may require great care since portions of the design may require pipeline rebalancing.

See the later example for more details on pipelining. b. Parallelization. This is probably the second most-important improvement you can make. Do you have a FIR filter that won't operate at the correct speed? You can use two FIR filters in parallel, each operating at half-rate, and interleave the outputs.

This is the classic speed/area tradeoff. c.

Retiming

. This involves taking existing registers and moving them to different points within the combinational logic to rob from Peter to pay Paul, so to speak.

This works if, to stretch the maxim, Paul is bereft of slack, while Peter has a surfeit.

Some synthesis tools can perform a degree of retiming automatically. d. Replication. Replication of registers or buffers increases the amount of logic but reduces the fanout on the replicated objects. This decreases the capacitance of the net and reduces net delay. The replicated registers may also be floorplanned to place them closer to the logic groups they drive. Replication is often performed automatically by the tools and manual replication is not a common practice in a high-level design environment like System Generator. e.

Shannon Expansion

. This method involves replicating the faster logic in a critical path in order to remove dependencies on slower logic. This is sometimes done automatically by the synthesizer. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Timing Analysis Compilation

Using Hard Cores

. Are you using a ROM that is implemented in distributed RAM when it would operate much faster in a block memory hard core? Do you have a wide adder that would benefit from being put in a DSP48 block, which can operate at 500MHz? Take advantage of the embedded hard cores. g. New Paradigms. Do you need to create a large delay? Instead of using a counter with a long carry chain, why not build a delay out of cascaded Johnson rings using

SRL16s? Or how about using an LFSR? Neither requires a carry chain and can operate much faster. Sometimes you have to rethink certain design elements completely.

Eliminate overconstraints

. Ensure that elements of your design that only need to be operated at a subsampled rate are designed that way by using the downsample and upsample blocks in System Generator. If these blocks are not used, then the timing analyzer is not aware that these sections of the circuit are subsampled, and the design is overconstrainted.

Change the constraints

. Is it possible to run the design at a lower clock speed? If so, this is an easy way to meet your requirements. Unfortunately, this is rarely possible due to design requirements.

Increase PAR effort levels

. The mapper and place & route tools (PAR) in ISE take effort levels as arguments. When using ISE (from the Project Navigator GUI), try the – timing option in MAP. You may also increase the PAR effort levels which will increase the PAR execution time but may also result in a faster design.

Multipass PAR

. PAR is an iterative process and is somewhat chaotic in that the initial conditions can vastly influence the final result. PAR uses a seed value to determine the initial conditions. This is referred to as the cost table value. You may change this value in the Project Navigator by hand. Even better, you may perform a multipass PAR process which runs PAR multiple times with different cost table values. This is timeconsuming but often effective.

Floorplanning

. This step should be avoided if possible, but can yield huge improvements. The automatic placer in PAR can be improved upon by human intervention. Floorplanning places critical elements close to each other on the Xilinx die, reducing net delays. The PACE and Floorplanner tools in ISE may be used for this.

A more advanced tool, PlanAhead™ software, is also available separately from Xilinx to aid in this task.

Use a faster part

. This is often the first solution seized upon, but is also expensive. If you are using an old Xilinx part, porting your design to a newer, faster Xilinx part may often save money because the new parts may be cheaper on account of Moore's Law.

However, moving to a faster part in the same family incurs significant extra costs, and often isn't necessary if the previous steps are followed.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

335

Chapter 5: System Generator Compilation Types

Tutorial Example: Using the Timing Analyzer

Sometimes the hardware created by System Generator may not meet the requested timing requirements. This is typically due to a setup time violation in the design. A setup time violation means that a particular signal cannot get from the output of one synchronous element to the input of another synchronous element within the requested clock period and subject to the second synchronous element's setup time requirement.

Let us use an example to show how we would use the timing analyzer to improve circuit performance. Our example will be a parity calculator that will find the parity of a byte by using an 8-input XOR. The design can be found at:

<sysgen_tree>/examples/timing_analysis/parity_test.mdl

336

The design has eight one-bit gateway inputs that are registered by one-bit registers. These are processed by seven 2-input XOR blocks. These have a latency of zero and thus are purely combinational. The final register, parity_reg, registers the final result (the parity) which is connected to an output gateway. The design appears to have three levels of logic, because each path fanning in to parity_reg goes through three XOR blocks. www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Timing Analysis Compilation

Generate the Example Design

We'll generate the design using the Timing Analysis target and a requested period of 1.4ns

(714MHz). This is admittedly a very high clock frequency, but we wish some paths to fail for demonstration purposes. We set these parameters in the System Generator token:

Examine the Slow Paths

After clicking on Generate, after a time, the timing analyzer window will appear as shown below:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

337

Chapter 5: System Generator Compilation Types

There are two failing paths, normally highlighted in red/pink. (The top path is gray because it is selected.) The negative slack values are shown in boldface. The worst of the two fails by 96ps.

Note that there are two levels of logic in the path shown. How can this be? The System

Generator diagram shows three levels of logic in all paths. The reason is that the implemented design does not correlate exactly to the System Generator diagram. In this case, the synthesizer has compressed some of the 2-bit XOR blocks into 4-input LUTs and created the 8-input XOR using only two levels of logic as shown in this Synplify Pro schematic:

LUT4_6996

LUT2_L_6 registerh_q_net[0:0] registerg_q_net[0:0] registerf_q_net[0:0] registere_q_net[0:0] xor_3a_y_net[0:0]

y[0]

y_5[0]

LUT4_6996 registerd_q_net[0:0] registerc_q_net[0:0] registerb_q_net[0:0] registera_q_net[0:0] y_4[0]

Note how the net and block names have all been munged, requiring the magic unmunging capabilities of the timing analyzer.

Also note the details of the selected path. The logic delays cannot be reduced. One of the net delays is 813ps. This could possibly be reduced by means of floorplanning, multipass

PAR, or simply by increasing the PAR effort level.

Rescue the Design

Instead, let us attempt a more robust solution to fix the path by changing the source design.

There are no feedback paths in this design, so let us assume we can add a cycle of latency and pipeline the design. There are two levels of logic in the failing paths. Any design can theoretically be re-implemented with only a single level of logic. We will do this now.

To add a pipeline stage, we will merely add latency to selected XOR blocks. By clicking on an XOR block, you may change its latency from zero to one like so:

338

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Timing Analysis Compilation

This will add a register to the end of the XOR gate. We will change the latency on blocks xor_2a

and xor_2b. We know from examining the Synplify Pro schematic that the outputs of these blocks form the output of the first level of logic in the synthesized design.

The modified System Generator looks very similar with the exception of the z

-1

on the labels of the two modified XOR blocks, indicating their new latency.

We generate this design as before and examine the slow paths:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

339

Chapter 5: System Generator Compilation Types

Excellent! No more failing paths! The design has been rescued, all in record time and without using a more expensive part. Surely raises and promotions shall follow you for all your days.

Note that all paths now have but a single level of logic. What exactly has happened here?

Let us examine the Synplify Pro schematic to see how the modified circuit was synthesized:

xor_2b_f0fd230c1d

LUT4_L_6996

1.18,0.45

0.98,0.45

1.18,0.45

0.35,-0.06

latency_pipe_5_26_0_ fully_2_1_bit[0]

xor_3a_7e73f4292b

LUT2_L_6

1.18,-0.06

0.98,-0.06

1.18,-0.06

y[0] parity_reg_3123b0b42f

1.43,-0.06

Q op_mem_19_20_0_

xor_3a

parity_reg

xor_2b

xor_2a_8d6908ffef

LUT4_L_6996

1.18,0.45

0.98,0.45

1.18,0.45

0.35,-0.06

latency_pipe_5_26_0_ fully_2_1_bit[0]

xor_2a

See that there is an extra set of registers (highlighted in red) in between the two levels of logic. The circuit functions the same as before but with an additional cycle of latency.

Use Retiming to Rescue the Design

If a cycle of latency had to be eliminated to match the latency of the original design, it might be possible to remove the final output register or the input registers. This would increase the constraints upon the paths outside the Xilinx chip (i.e., the copper paths on the

PCB), but it may be feasible depending upon board-level path delays. This would be an example of retiming, because the latency is the same but the registers have been moved into the logic "cloud".

Creating Compilation Targets

The HDL and netlist files that System Generator produces when it compiles a design into hardware must be run through additional tools in order to produce a configuration bitstream file that is suitable for your FPGA. A typical flow that allows you to generate an

FPGA configuration file is ProjectNavigator. There are other ways in which a bitstream can be generated for your model. For example, it is possible to configure System Generator to automatically run the tools necessary to produce a configuration file when it compiles a design. This is advantageous since the complete bitstream generation process is accomplished inside the tool. Moreover, you can have System Generator run different tools

(e.g., ChipScope™ Pro Analyzer and iMPACT) once the configuration file is generated for a model.

The way in which System Generator compiles a model into hardware depends on the compilation target that is chosen for the design. The HDL Netlist compilation target is most common, and generates an HDL netlist of your design plus any cores that go along

340

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Creating Compilation Targets

with it. New compilation targets can be created that extend the HDL Netlist target so that additional tools can be applied to the resulting HDL netlist files.

This topic explains how you can create new compilation targets that extend the HDL

Netlist target in order to produce and configure FPGA hardware. More specifically, it describes how to configure System Generator to produce a bitstream for a model, and how to invoke various tools once the bitstream is created.

Defining New Compilation Targets

You can create new compilation targets to run tools that process the output files associated with HDL Netlist compilation. A compilation target is defined by a minimum of two

MATLAB functions. The first function, xltarget.m, tells System Generator to support the target (i.e., make it selectable from the System Generator block dialog box), and specifies the MATLAB function where more information about the target can be found.

This function is called a "target info" function. A target info function defines information about the target, and can take any name, provided it is specified correctly in the target's xltarget.m

function. In some cases, a target info function defines a post-generation function. A post-generation function is responsible for invoking tools or scripts after normal HDL netlist compilation is complete. These functions are discussed in more detail in the topics that follow.

The xltarget Function

An xltarget function specifies one or more compilation targets that should be supported by

System Generator. It also provides entry points through which System Generator can find out more information about these targets.

Note:

System Generator determines which compilation targets to support by searching the plugins/compilation (and its subdirectories) of your System Generator software install tree for xltarget.m

files.

Although an xltarget function can specify multiple targets, it is not uncommon for each compilation target to have its own xltarget function. The directories these functions are saved in distinguish the targets. This means that each xltarget.m file must be saved in its own subdirectory under the plugins/compilation directory.

An xltarget function returns a cell array of target information. Different elements in this cell array define different compilation targets. The elements in this cell array are MATLAB structs that define two parameters:

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

341

Chapter 5: System Generator Compilation Types

The name of the compilation target as it should appear in the Compilation field of the

System Generator parameters dialog box;

The name of the MATLAB function it should invoke to find out more information (e.g.,

System Generator dialog box parameters, which post-generation function to use, if any) about the target.

The following code shows how to define three compilation targets named Standalone

Bitstream, iMPACT, and ChipScope™ Pro Analyzer: function s = xltarget s = {}; target_1.('name') = 'Standalone Bitstream'; target_1.('target_info') = 'xltools_target'; target_2.('name') = 'iMPACT'; target_2.('target_info') = 'xltools_target'; target_3.('name') = 'ChipScope Pro Analyzer'; target_3.('target_info') = 'xltools_target'; s = {target_1, target_2, target_3};

The name field in the code shown above specifies the name of the compilation target, as it should appear in the Compilation field of the System Generator dialog box: target_1.('name') = 'Standalone Bitstream';

The target_info field tells System Generator the target info function it should call to find out more information about the target. This function can have any name provided it is saved in the same directory as the corresponding xltarget.m file, or it is saved somewhere in the MATLAB path. target_1.('target_info') = 'xltools_target';

Note:

An example xltarget function is included in the examples/comp_targets directory of your

System Generator install tree. You can modify this function to define your own bitstream-related compilation targets.

Target Info Functions

A target info function (specified by the target_info field in the code above) is responsible for two things:

•

It defines the available and default settings for the target in the System Generator block dialog box;

•

It specifies the functions System Generator should call before and after the standard code generation process.

Note:

An example target info function, xltools_target.m, is included in the examples/comp_targets directory of your System Generator install tree.

One such function that is particularly useful to compilation targets is the post-generation function. A post-generation function is run after standard code generation. The code below shows how a post-generation function is specified in a target info function: settings.('postgeneration_fcn') = 'xltools_postgeneration';

Post-generation Functions

One way to extend System Generator compilation is by defining a new variety of compilation that specifies a post-generation function. A post-generation function is a

MATLAB function that tells System Generator how to process the HDL and netlist files once they are generated. This function is run after System Generator finishes the normal code generation steps involved with HDL Netlist compilation (i.e., producing an HDL

342

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Creating Compilation Targets

description of the design, running CORE Generator™, etc). For example, a hardware cosimulation target defines a post-generation function that in turn runs the tools necessary to produce hardware that can be used in the Simulink simulation loop.

Note:

Two post-generation functions xlBitstreamPostGeneration.m and xltools_postgeneration.m

, are included in the examples/comp_targets directory of your

System Generator install tree.

xlBitstreamPostGeneration.m

This example post-generation function compiles your model into a configuration bitstream that is appropriate for the settings (e.g., FPGA part, clock frequency, clock pin location) given in the System Generator dialog box of your design.

It then uses an XFLOW-based flow to invoke the Xilinx tools necessary to produce an

FPGA configuration bitstream.

It is possible to configure the tools and configurations for each tool invoked by XFLOW.

For more information on how to do this, refer to the topic in this example entitled Using

XFLOW

xltools_postgeneration.m

Sometimes you may want to run tools that configure and run the FPGA after a configuration bitstream has been generated (e.g., iMPACT, ChipScope™ Pro Analyzer).

The xltools_postgeneration function first calls the xlBitstreamGeneration function to generate the bitstream. It then invokes the appropriate tool (or tools) depending on the compilation target that is selected.

For example, you may want a compilation target that invokes iMPACT after the bitstream is generated. This can be done as follows (assuming iMPACT is in your system path): if (strcmp(params.compilation, 'iMPACT')) dos('impact'); end;

The first line checks the name of the compilation target. The second line sets up a DOS command that invokes iMPACT. ChipScope Pro Analyzer can be invoked similarly to the code above: if (strcmp(params.compilation, 'ChipScope Pro Analyzer')) xlCallChipScopeAnalyzer; end;

Note:

xlCallChipScopeAnalyzer is a MATLAB function provided by System Generator to invoke

ChipScope.

Configuring and Installing the Compilation Target

Listed below are the steps necessary to configure and install new bitstream compilation targets.

Copy the xltarget.m, xltools_postgeneration.m, and xltools_target.m files from examples/comp_targets into a temporary directory.

Change the permissions of the above files so they can be modified.

Add the desired compilation targets (e.g., iMPACT, ChipScope Analyzer Pro) to the xltarget.m

file.

Add the desired tool invocations to the xltools_postgeneration.m file.

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

343

Chapter 5: System Generator Compilation Types

Create a new directory (e.g., Bitstream) under the plugins/compilation directory of your System Generator software install tree. Copy the xltarget.m, xltools_postgeneration.m

, and xltools_target.m files into this directory.

Note:

The System Generator Compilation submenus mirror the directory structure under the plugins/compilation directory. When you create a new directory, or directory hierarchy, for the compilation target files, the names of the directories define the taxonomy of the compilation target submenus.

Copy the xlBitstreamPostGeneration.m, xlToolsMakebit.pl, balanced_xltools.opt

and bitgen_xltools.opt files from the examples/comp_targets

directory into a directory that is in your MATLAB path.

These files must be in a common directory.

In the MATLAB command window, type the following:

>> rehash toolboxcache

>> xlrehash_xltarget_cache

You can now access the newly installed compilation target from the System Generator graphical interface.

Using XFLOW

The post-generation scripting included with this example uses XFLOW to produce a configuration file for your FPGA. XFLOW allows you to automate the process of design synthesis, implementation, and simulation using a command line interface. XFLOW uses command files to tell it which tools to run, and how they should be run.

This example contains two XFLOW options files, balanced_xltools.opt and bitgen_xltools.opt

. These files are associated with the implementation and configuration flows of XFLOW, respectively. The balanced_xltools.opt options files runs the Xilinx NGDBUILD, MAP, and PAR tools. The settings for each tool are specified in the options files . The bitgen_xltools.opt file runs BITGEN to produce a configuration file for your FPGA. You may modify these files as desired (e.g., to run the timing analyzer after PAR).

344

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

Index

A

Addressable Shift Register block

Algorithm Exploration

ASR block

Asynchronous Clocking

Auto-Generated Clock Enable Logic resetting in System Generator

Automatic Code Generation

B

Bit-Accurate

Bitstream Compilation

319

Bit-True Modeling

Black Box

Configuration M-Function adding new ports

black box API

black box clocking

combinational paths

configuring port sample rates

273

configuring port types

defining block ports

dynamic output ports

error checking

language selection

obtaining a port object

272

specifying the top-level entity

271

specifying Verilog parameters

275

specifying VHDL Generics

275

SysgenBlockDescriptor Member Variables

277

SysgenBlockDescriptor methods

278

SysgenPortDescriptor Member

Variables

280

SysgenPortDescriptor methods

280

Examples

284

advanced black box example using ModelSim

311

dynamic black boxes

307

importing a Core Generator module

285

importing a Core Generator module that needs a VHDL wrapper

291

importing a Verilog module

305

importing a VHDL module

298

importing a Xilinx Core Generator module

284

simulating several black boxes simultaneously

309

HDL Co-Sim configuring the HDL simulator

281

co-simulating multiple black boxes

283

Black Box Configuration

M-function

270

Black Box Configuration Wizard

268

Block Masks

Blockset

Xilinx

C

ChipScope Pro Analyzer

Clock Domain Partitioning

113

Clock Enable

Fanout Reduction

Clock Frequency selecting for Hardware Co-Sim

179

Clocking and timing

asynchronous

synchronous

Clocking Options

Clock Enable

Expose Clock Ports

Hybrid DCM-CE

Code Generation automatic

Color Shading blocks by signal rate

Compilation Type using XFLOW

344

Compilation Types

Bitstream Compilation

319

configuring and installing the Compilation Target

343

creating new compilation targets

340

EDK Export Tool

Hardware Co-Simulation Compilation

HDL Netlist Compilation

NGC Netlist Compilation

Compiling for bitstream generation

319

EDK Export

Hardware Co-Simulation

NGC Netlist generation

Compiling for HDL Netlist generation

Compiling MATLAB complex multiplier with latency

disp function

finite state machines

FIR example

into an FPGA

optional input ports

parameterizable accumulator

passing parameters into the MCode block

RPN calculator

shift operation

simple arithmetic operation

simple selector

Compiling Shared Memories for HW Co-Sim

189

Configurable Subsystems and System

Generator

Configuring and Installing the Compilation Target

343

Constraints File

System Generator

Controls hierarchical

Creating Compilation Targets

340

Crossing Clock Domains

114

Custom Bus Interfaces for exported pcore

324

Cycle-Accurate

Cycle-True Clock Islands

112

Cycle-True Modeling

D

DCM locked pin

DCM reset pin

Debugging using ChipScope Pro

Defining New Compilation Targets

341

Target Info functions

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

345

xltools_target

342

the xltarget Function

341

Discrete Time Systems

Distinct Clocks generating multiple cycle-true islands

112

DSP48 design styles for

design techniques

100

mapping from the DSP48 block

mapping standard components to

mapping to from logic synthesis tools

physical planning for

102

DSP48 Macro block

E

EDK generating software drivers

139

support from System Generator

143

writing software drivers

EDK Export Tool

exporting a pcore

EDK Import Wizard

EDK Processor exposing processor ports

145

importing

143

Ethernet-based HW Co-Sim

241

Export pcore enable Custom Bus Interfaces

324

Exporting a pcore

a System Generator model as a pcore

Expose Clock Ports Option tutorial

F

Fanout Reduction for Clock Enable

FDATool using in digital filter applications

104

FPGA a brief introduction

generating a bitstream

notes for higher performance

Frame-Based Acceleration using Hardware Co-Sim

200

FSL-based pcore

136

Full Precision signal type

G

Generating an FPGA bitstream

EDK software drivers

139

Generating an FPGA Bitstream

H

Hardware oversampling

Hardware Co-Sim

173

blocks

176

choosing a compilation target

175

compiling shared memories

189

co-simulating lockable shared memories

192

co-simulating shared FIFOs

195

co-simulating shared registers

194

co-simulating unprotected shared memories

191

invoking the code generator

175

JTAG hardware requirements

254

Network-Based Ethernet

187

Point-to-Point Ethernet

183

processor integration

139

restrictions on shared memories

198

selecting the target clock frequency

179

shared memory support

188

using for frame-based acceleration

200

using for real-time signal processing

213

Xilinx tool flow settings

198

Hardware Co-Simulation Compilation

Hardware Debugging using ChipScope Pro

Hardware Generation

Hardware Generation Mode

EDK pcore

HDL netlist

Hardware/Software Co-Design

136

Examples creating MicroBlaze Peripherals in System Generator

153

designing and simulating MicroBlaze Processor Systems

158

using EDK

167

using PicoBlase in System Generator

148

HDL Co-Sim configuring the HDL simulator

281

co-simulating multiple black boxes

283

HDL Netlist Compilation

HDL Testbench

Hierarchical Controls

Histogram Charts from Timing Analyzer

332

334

Hybrid DCM-CE Option locked pin

reset pin

tutorial

I

Implementing a complete design

part of a design

Importing a System Generator design

an EDK processor

143

an EDK project

Importing a System Generator Design

integration design rules

integration flow with Project Navigator

step-by-step example

Installation

Installing a Spartan-3A DSP 1800A

Starter Platform for Hardware Co-

Sim

241

Installing am ML402 Board for JTAG

Hardware Co-Sim

253

Introduction to FPGAs

J

JTAG Hardware Co-Sim board support package files

260

Detecting New Board Packages

266

installing board-support packages

265

manually specifying board-specific ports

263

obtaining platform information

261

providing your own top-level

264

supporting new platforms

254

JTAG-based HW Co-Sim

253

346

www.xilinx.com

System Generator for DSP

Release 10.1.3 September, 2008

L

Locked pin

Hybrid DCM-CE Option

produced by System Generator

Oversampling

M

MATLAB compiling into an FPGA

complex multiplier with latency

disp function

finite state machines

FIR example

optional input ports

parameterizable accumulator

passing parameters into the MCode block

RPN calculator

simple arithmetic operation

simple selector

simple shift operation

Memory Map Creation for processor integration

137

M-Function black box configuration

270

MicroBlaze in System Generator tutorial

153

System Design and Simulation

158

ML402 Board

Installation for JTAG HW Co-Sim

253

Modeling bit-true and cycle-true

Multiple Clock Applications

112

Multirate Designs color shading by signal rate

Multirate Models

N

Netlisting multiple clock designs

115

Network-Based Ethernet Hardware Co-

Sim

187

NGC Netlist Compilation

Notes for higher performance FPGA design

O

OutputFiles

P

Parameter Passing

Pcore pcore export as under development

exporting

exporting a System Generator model as a peripheral

PicoBlaze designing within System Generator

in System Generator tutorial

148

overview

PLB-based pcore

136

Point-to-Point Ethernet HW Co-Sim

183

Processor Integration

Hardware Co-Sim

hardware generation

memory map creation

using custom logic

Project Navigator integration flow with System Generator

R

Rate-Changing Blocks

Real-Time Signal Processing using Hardware Co-Sim

213

Reducing

Clock Enable Fannout

Reference Blockset

Xilinx

Reset pin

Hybrid DCM-CE Option

Resource Estimation

S

SBD Builder saving plugin files

259

specifying board-specific I/O ports

257

Shared Memory Support for HW Co-Sim

188

Signal Types

displaying data types

full precision

gateway blocks

user-specified precision

Simulink System Period

Spartan-3A DSP 1800A Starter Platform

Installation for Ethernet HW Co-Sim

241

Synchronization Mechanisms indeterminate data

valid ports

Synchronous Clocking

Clock Enable option

Expose Clock Ports option

Hybrid DCM-CE option

System Generator adding a block to a Configurable

Subsystem

and Configurable Subsystems

blocksets

defining a Configurable Subsystem

deleting a block from a Configurable

Subsystem

generating hardware from Configurable Subsystems

output files

processing a design with physical design tools

resetting auto-generated Clock Enable logic

system-level modeling

using a Configurable Subsystem

System Generator block compiling and simulating

System Generator Constraints constraints file

example

IOB timing and placement

multicycle path

system clock period

System Generator Design Flows algorithm exploration

implementing a complete design

implementing part of a larger design

System-Level Modeling

T

Tapped Delay Lines

TDM data streams

Testbench

HDL

Time-Division Multiplexed

System Generator for DSP

Release 10.1.3 September, 2008 www.xilinx.com

347

Timing Analysis clock skew and jitter

329

compilation type

Compiling for

timing analysis

concepts review

328

cross-probing

331

displaying low-level names

331

histogram charts

332

334

improving failing paths

334

observing slow paths

330

path analysis example

period and slack

statistics

trace report

tutorial

Timing Analyzer invoking on previously-generated data

328

Timing and Clocking

Trace Report timing analysis

333

Tutorials

Black Box

Dynamic Black Boxes

307

Importing a Core Generator

Module

285

Importing a Core Generator

Module that Needs a VHDL

Wrapper

291

Importing a Verilog Module

306

Importing a VHDL Module

298

Simulating Several Black Boxes

Simultaneously

309

ChipScope

Using ChipScope in System

Generator

Clocking

Using the Clock Generator(DCM) Option

Using the Expose Clock Ports

Option

Hardware/Software Co-Design

Creating a New XPS Project

167

Creating MicroBlaze Peripherals in System Generator

153

Designing and Simulating MicroBlaze Processor Systems

158

Using PicoBlaze in System Generator

148

Timing Analysis

Using the Timing Analyzer

336

U

Underdevelopment export pcore as