1. CoDeveloper Overview Updated February 19, 2004

2. 2 Introducing CoDeveloper™  Targeting hardware/software programmable platforms  Target platforms feature a mix of traditional processor elements and programmable hardware elements  Altera, Xilinx, and other key platforms  Emphasizing software-oriented design methods  Compatible with familiar software development frameworks  Including Visual Studio, Codewarrior, GCC, Eclipse, etc.  Providing crucial support for hardware/software interfaces  Tools efficiently map the application to the platform  Programming model supports highly parallel hw/sw systems  Providing more than just C to hardware compilation!

3. 3 CoDeveloper Design Flow CoBuilder™ RTL generator CoBuilder™ RTL generator Impulse C design files FPGA synthesis tool FPGA synthesis tool Generated HDL files Generated HDL files Target cross compiler Target cross compiler Impulse Platform Libraries Impulse Platform Libraries CoBuilder™ architecture generator CoBuilder™ architecture generator CoBuilder™ library generator CoBuilder™ library generator Generated software libraries Generated software libraries Impulse hardware libraries Impulse hardware libraries Impulse software libraries Impulse software libraries Generated HDL files Generated HDL files Complete software/hardware application ready to implement on target platform

4. 4 Accelerating Applications: How? 1.Identify and exploit low-level parallelism in standard C  Examples: pipeline and/or unroll loops, schedule assignments, generate hardware for C processes  Advantage: can accelerate legacy C algorithms, inner code loops, etc. and reduce need for manual optimization 2.Encourage and assist programmers in the use of alternate, more efficient parallel programming techniques  Support the language they are most familiar with (ANSI C)  Provide libraries, tools and examples for parallel programming  Advantage: best way to address system-level optimizations  There is where the biggest performance gains are found!

5. 5 Application Domains  Applications requiring repetitive computations at very high speed  Dataflow-oriented, high degrees of parallelism  Pipelined algorithms (e.g. filters)  For processing streams of data in real time  Imaging  Communications  Digital Signal Processing (DSP)

6. 6 Targeting Programmable Platforms  Standard RISC processor plus FPGA  FPGA with embedded (“soft”) RISC processor RISC FPGA RISC FPGA Impulse C Applications Two categories:

7. 7 Who Makes These Platforms?  Altera: Nios™ processor  Cyclone and Stratix FPGA families  Xilinx: MicroBlaze™, PowerPC™ processors  Virtex and Spartan FPGA families  Quicklogic: QuickMIPS™  Others yet to be announced  This is a HOT area in the FPGA industry

8. 8 Mapping Applications to Platforms S/W process H/W process S/W process H/W process S/W process Embedded processor FPGA hardware resources Bus interface FPGA-based platform This is not a trivial task!

9. 9 Example: Altera Nios S/W process H/W process S/W process H/W process S/W process NIOS processor FPGA hardware resources Avalon interface Cyclone or Stratix FPGA Impulse C application

10. 10 Example: Xilinx MicroBlaze S/W process H/W process S/W process H/W process S/W process MicroBlaze or PowerPC processor FPGA hardware resources OPB and/or FSL interface Virtex or Spartan FPGA Impulse C application

11. 11 How Is This Done Today?  Software engineer writes code (typically C) for embedded processor  High design productivity using high-level languages  But low-performance results (processors are SLOW)  Hardware engineer writes low-level HDL code for FPGA or ASIC portion  Very low design productivity (1/10 th or less of the s/w eng)  But high performance results (hardware runs FAST)  System designer is the hardware/software mediator  Specifies hardware/software interfaces and locks down the design to reflect hardware design lead-times

12. 12 How Does CoDeveloper Help?  Reduces or eliminates the need to re- implement software routines as low-level hardware  Automates the process of creating hardware/software interfaces  Provides a software programming model appropriate for the target platform

13. 13 Impulse for HW/SW Codesign Generate RTL Impulse C design files Altera Quartus, Xilinx ISE, other FPGA tools Altera Quartus, Xilinx ISE, other FPGA tools HDL files HDL files Impulse Platform Libraries Impulse Platform Libraries Generate hardware interfaces Generate hardware interfaces Generate software interfaces Software libraries Software libraries HDL files HDL files Nios compiler, MicroBlaze compiler, others Visual Studio ™ CodeWarrior ™ GCC, etc.

14. 14 From C Application to FPGA Platform S/W process H/W process S/W process H/W process S/W process 3. FPGA hardware is automatically created from C language processes 2. Automatic generation of hardware/software interfaces is optimized for target platforms 1. Platform libraries support existing embedded compiler environments The result? Accelerated software with minimal need for hardware or FPGA design knowledge

15. 15 CoDeveloper™ Tool Flow CoBuilder™ RTL generator CoBuilder™ RTL generator Impulse C design files CoMonitor™ application monitor CoMonitor™ application monitor Host standard C compiler Host standard C compiler FPGA synthesis tool FPGA synthesis tool Host simulation executable Host simulation executable Legacy C algorithms Legacy C algorithms HDL design files HDL design files CoValidator™ HDL simulator CoValidator™ HDL simulator CoWave™ waveform viewer CoWave™ waveform viewer Generated HDL files Generated HDL files Host C debugger Target cross compiler Target cross compiler Impulse Design Assistant Impulse Design Assistant Target download binary Target download binary Impulse Platform Libraries Impulse Platform Libraries Target ISS and/or debugger Target ISS and/or debugger Impulse Simulation Libraries Impulse Simulation Libraries

16. 16 Impulse C  Based on Streams-C from Los Alamos National Labs  Dataflow-oriented C language extensions for highly parallel systems  HDL generation for FPGA elements  C language or other code generation for processor elements  Impulse improvements include:  Language redesign (standard ANSI C with Impulse C library functions)  Redesigned scheduling and code generation for FPGA targets  Application Manager, Application Monitor and full compatibility with Visual Studio and other standard development/debugging environments

17. 17  Modified Communicating Sequential Process (CSP) model  Buffered communication channels (FIFOs) to implement streams  Supports dataflow and message-based communications between functional units and local or shared memories  Supports parallelism at the application level and at the level of individual processes (via automated scheduling/pipelining) Impulse C Programming Model S/W process H/W process S/W process

18. 18 Impulse C Processes Software processes set up data and perform non time- critical functions Hardware processes are independently synchronized and perform most of the work C language process C language process Shared memory reads/writes Stream inputs Stream outputs Signal inputs Signal outputs App Monitor outputs

19. 19 CoBuilder Compilation Process CoBuilder™ RTL generator CoBuilder™ RTL generator Impulse C design files FPGA synthesis tool FPGA synthesis tool Generated HDL files Generated HDL files Target cross compiler Target cross compiler Impulse Platform Libraries Impulse Platform Libraries CoBuilder™ architecture generator CoBuilder™ architecture generator CoBuilder™ library generator CoBuilder™ library generator Generated software libraries Generated software libraries Impulse hardware libraries Impulse hardware libraries Impulse software libraries Impulse software libraries Generated HDL files Generated HDL files Complete software/hardware application ready to implement on target platform

20. 20 Impulse C H/W-S/W interfaces Impulse C application, compiled using standard C cross-compiler and related tools Impulse C h/w processes, compiled to HDL by CoBuilder and synthesized using standard tools Library of h/w interface components Bus interface wrapper automatically generated by CoBuilder for the selected platform Impulse C function library for stream and signal h/w-s/w interfaces Interface routines (drivers) automatically generated by CoBuilder for the selected platform Standard bus specified for target platform Impulse C runtime library FPGA platform bus Processor/FPGA interface (s/w driver generated by CoBuilder) FPGA hardware wrapper (generated by CoBuilder as HDL) Impulse C hardware library Application software (embedded processor application) Impulse C hardware processes (generated by CoBuilder as HDL)

21. Impulse C Case Study 3DES Encryption (Xilinx Microblaze platform)

22. 22 Case Study: 3DES Encryption  Use publicly available 3DES source code  Written by Phil Karn (www.ka9q.net)www.ka9q.net  Not written to take advantage of parallel programming techniques (optimized for typical processor targets)  Make minimal changes in support of streams-based communication  Use stream reads/writes for data blocks and keys  Simulate legacy and Impulse C versions using same data  Build one simulation executable containing both versions  Compile to hardware, compare results again  Legacy version runs in uP, Impulse C version runs in hardware

23. 23 Application Structure (HW Test) Reference/prototype board Producer (random data) H/W encrypt S/W encrypt (legacy C) Consumer (compare results and speed) Embedded processor FPGA gates “Platform” FPGA

24. 24 3DES Desktop Simulation  Complete 3DES application compiled using standard tools  Legacy C and Impulse C compiled as one application  Verified using both Visual Studio.NET and GCC/GDB  Application Monitor used to debug and improve stream data movement  Example: identified throughput increase possible through simple increase in buffer sizes (double-buffer inputs)

25. 25 3DES Desktop Simulation Producer (from file) H/W encrypt (Impulse C) S/W encrypt (legacy C) Consumer (display results) S/W decrypt (legacy C) H/W decrypt (Impulse C)

26. 26 Desktop Simulation CoDeveloper Application Monitor

27. 27 Compiler Flow Generate RTL Impulse C design files Altera Quartus, Xilinx ISE, other FPGA tools Altera Quartus, Xilinx ISE, other FPGA tools HDL files HDL files Impulse Platform Libraries Impulse Platform Libraries Generate hardware interfaces Generate hardware interfaces Generate software interfaces Software libraries Software libraries HDL files HDL files Nios compiler, MicroBlaze compiler, others Visual Studio ™ CodeWarrior ™ GCC, etc.

28. 28 Results (Virtex II FPGA)  MicroBlaze clock running at 100MHz  3DES algorithm clocking at 24MHz  Performance (1000 blocks):  Clock cycles to process one block: 149 (not including I/O overhead)  36X speedup over software-only version 3DES in MicroBlaze3DES in FPGA 252 ms6.9 ms* *Includes I/O overhead: 11%

29. 29 Summary  CoDeveloper and Impulse C provide:  Programming model and tools supporting hardware / software codesign for programmable hardware targets  Compiler technologies (CoBuilder) allowing software processes to be compiled directly to hardware.  CoDeveloper is a software-oriented solution  Designed (and priced) to appeal to software developers, not just leading edge FPGA experts  Designed to operate in conjunction with existing software development tools and tool flows  Designed to accelerate hardware/software prototypes and products