C for Process Networks

Size: px

Start display at page:

Download "C for Process Networks"

Pauline Rich
7 years ago
Views:

Mapping of Applications to MPSoCs, Schloss Rheinfels,

1 C for Process Networks Stefan Schürmans, Weihua Sheng, Anastasia Stulova, Jeronimo Castrillon 3 rd Workshop on Mapping of Applications to MPSoCs, Schloss Rheinfels, June 29 th 2010 Institute for Integrated Signal Processing Systems

2 Outline Introduction MPSoC Application Programming Studio C for Process Networks Generated Code Targets & Applications Summary & Outlook 2

3 Embedded Systems use MPSoCs Embedded Systems Requirements High Computational Performance High Energy Efficiency Fujitsu, MPSOC 2009 SoC Consumer Portable Design Complexity Trends (ITRS 2007) Parallel Hardware: MPSoCs 3

4 Programming Languages for Embedded Systems 100% 80% 60% 40% 20% 0% C C++ Assembly others Source: Embedded.com (EE Times) Dominant Embedded Systems Programming Language: C Existing legacy software New projects Programmers use language they know 4

5 Mismatch between HW and SW Parallel Hardware Sequential Software RISC VLIW ASIP DSP Mismatch MPSoC C Possible Solutions: Partition sequential software Write parallel software e.g. Process Networks (PN) Tool Support required 5

6 Outline Introduction MPSoC Application Programming Studio C for Process Networks Generated Code Targets & Applications Summary & Outlook 6

Specification Support both sequential C or

Partitioning High-level transformations to

Parallelism extraction MPSoC Backends Code

7 MAPS (MPSoC Application Programming Studio) Approach Applications Specification Support both sequential C or parallel models (process networks) Multi-App Scenarios (Real-Time constraints, interference) Application Partitioning High-level transformations to expose more parallelism Semi-automatic Parallelism extraction MPSoC Backends Code generation for MPSoCs Native Hardware/Virtual Platform System Simulation Efficient Mapping/Scheduling Spatial/Temporal Mapping Efficient/Fast Multi-task Run-time processing 7

8 MAPS Tool Flow Arch. Model Applications PNs Seq. Multi-applications App. concurrency graph (ACG) Real-time constraints PE class preferences MVP simulator Analysis Parsing Profiling & Tracing Seq. Partitioning Mapping & Scheduling Mapping Scheduling Performance Estimation Sequential Application (C) Process Network Application (Language?) Multi-app. Analysis Code Generation Execution Real Silicon / Virtual Platforms Run-time Info OS Primitives Low level APIs 8

9 Outline Introduction MPSoC Application Programming Studio C for Process Networks Generated Code Targets & Applications Summary & Outlook 9

10 C for Process Networks Process Network (PN) C code inside processes Merge C + PN into a single language C KPN, SDF C 2 C C for Process Networks (CPN) Minimal language extension to C New keywords Processes Channels Channel Accesses C 1 10

11 CPN Example: Fibonacci Numbers PNsdf Add PNin(int u, int v) PNout(int sum) { PNloop { sum = u + v; } } PNkpn Delay PNin(int i) PNout(int o) PNparam(int first) { PNout(o) { o = first; } while (1) { PNin(i) PNout(o) { o = i; } } } PNsdf Print PNin(int wnd : 2) { PNloop { printf( %d %d\n, wnd[0], wnd[1]); } } + a z -1 z -1 c Pr. b PNchannel int a, b, c; PNprocess add = Add PNin(b, c) PNout(a); PNprocess delay1 = Delay PNin(a) PNout(b) PNparam(1); PNprocess delay2 = Delay PNin(b) PNout(c) PNparam(0); PNprocess print = Print PNin(c); 11

12 Outline Introduction MPSoC Application Programming Studio C for Process Networks Generated Code Targets & Applications Summary & Outlook 12

13 Translation of CPN to Message Passing CPN Code void fft(int i[64], int o[64]); PNkpn FFT PNin(int i : 64) PNout(int o : 64) { PNin(i) PNout(o) { fft(i, o); { C Code for Message Passing int i[64], o[64]; recv(ch_i, i, 64); fft(i, o); } } } send(ch_o, o, 64); ch_i recv fft( i, o ) send ch_o 13

14 Translation of CPN for Shared Memory Systems FIFO Channel in global Shared Memory Cyclic buffer Part filled with data, free part Windowed Access (WA) Parts being written and read Multiple Readers (MR) Replication of reader information free read data write free write free free free read data read data data read data Writer Reader 2 nd Reader 14

15 Translation of CPN for Shared Memory Systems CPN Code void fft(int i[64], int o[64]); PNkpn FFT PNin(int i : 64) PNout(int o : 64) { PNin(i) PNout(o) { fft(i, o); { C Code for Shared Memory FIFO int * p_i = rd_begin(ch_i, 64); int * p_o = wr_begin(ch_o, 64); fft(p_i, p_o); } } } wr_end(ch_o, 64); rd_end(ch_i, 64); ch_i ch_o fft( p_i, p_o ) 15

16 Outline Introduction MPSoC Application Programming Studio C for Process Networks Generated Code Targets & Applications Summary & Outlook 16

Targets Pthreads Execution of application on host MAPS Virtual Platform (MVP) Early Exploration of Mapping & Scheduling CoWare/Synopsys ATM Abstract Virtual Platform Based on

17 Targets Pthreads Execution of application on host MAPS Virtual Platform (MVP) Early Exploration of Mapping & Scheduling CoWare/Synopsys ATM Abstract Virtual Platform Based on Virtual Processing Unit (VPU) Target Shared Memory Process Channel Pthreads Yes Host Thread Sh. Mem. (WA + MR) MVP Yes MVP Task Sh. Mem. (WA + MR) ATM No VPU Task System C FIFO 17

18 JEPG Encoder / Decoder Task Graph Source Parameters Sink R2B R2B R2B B2R B2R B2R DCT DCT DCT IDCT IDCT IDCT Q Q Q IQ IQ IQ Difference (amplified 16x) ZZE ZZD RLE Encoder RLD Decoder Input Output 18

19 MMSE Processing in MIMO Receiver P1 P3 P2 Task Graph Input P7 P9 Output P4 P5 P6 P8 P10 Input Test Vectors Reference Expected Results = * Output Results *) on all 3 Targets 19

20 Outline Introduction MPSoC Application Programming Studio C for Process Networks Generated Code Targets & Applications Summary & Outlook 20

21 Summary & Outlook MPSoC programming requires tool support MAPS: MPSoC Application Programming Studio CPN to write PN applications Minimal extension to C Concise and expressive Code Generation Portable to different targets Use special features of target Efficiency Outlook: Code generation for real hardware platforms Evaluate additional applications 21

22 Thank You! Questions? 22

MPSoC Virtual Platforms

CASTNESS 2007 Workshop MPSoC Virtual Platforms Rainer Leupers Software for Systems on Silicon (SSS) RWTH Aachen University Institute for Integrated Signal Processing Systems Why focus on virtual platforms?