1 / 17

Center for Embedded Computer Systems University of California, Irvine

SPARK: A High-Level Synthesis Framework for Applying Parallelizing Compiler Transformations. Center for Embedded Computer Systems University of California, Irvine http://www.cecs.uci.edu/~spark. Supported by Semiconductor Research Corporation. M e m o r y. x = a + b c = a < b. If Node. c.

malini
Download Presentation

Center for Embedded Computer Systems University of California, Irvine

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. SPARK: A High-Level Synthesis Framework for Applying Parallelizing Compiler Transformations Center for Embedded Computer Systems University of California, Irvine http://www.cecs.uci.edu/~spark Supported by Semiconductor Research Corporation

  2. M e m o r y x = a + b c = a < b If Node c T F Control d = e - f g = h + i ALU Data path j = d x g l = e + x High Level Synthesis Transform behavioral descriptions to RTL/gate level From C to CDFG to Architecture x = a + b; c = a < b; if (c) then d = e – f; else g = h + i; j = d x g; l = e + x;

  3. Focus of this Paper • Spark High-Level Synthesis Framework • Comprehensive synthesis framework • Coordinated Coarse & Fine-Grain HLS and Compiler optimizations • C Input => RTL VHDL Output • Results in terms of Logic Synthesis Results (Area/Timing) • Large, real-life applications: MPEG, Image Processing • Descriptions with complex and nested conditionals and loops • Objectives: • Improve quality of HLS results by extracting a high degree of parallelization • Reduce impact of control constructs on QOR • Improve controllability of the HLS solutions

  4. Recent Related Work • Code motions in the presence of conditionals • Condition Vector List Scheduling [Wakabayashi 89] • Path Based Scheduling [Camposano 91] • Symbolic Scheduling [Radivojevic 96] • WaveSched Scheduler [Lakshminarayana 98] • Basic Block Control Graph Scheduling [Santos 99] • Early work was on data-intensive DSP algorithms • Pipelining, Algorithmic transformations

  5. SPARK High Level Synthesis Framework • C Input => RTL VHDL Output • VHDL => Logic Synthesis Results • Customizable Scheduler • Modular toolbox of transformations • Heuristics select transformations

  6. The Intermediate Representation • Spark uses Hierarchical Task Graphs (HTG) • Consists of hierarchy of basic blocks and HTG nodes • 3 Types of HTG Nodes: Single, Compound, Loop • Enables application of coarse and fine grain optimizations • Can regenerate C code • Augmented by data dependency graphs

  7. Loop HTG Node

  8. Trailblazing: Code Motion Technique • HTGs enable Hierarchical Operation Moves • Does not visit each node in the graph

  9. Across HTG Across HTG + Speculate Speculate Across HTG Scheduling Heuristic BB 0 a • Get Available Ops • a, b, c, d • Determine Code Motions Required • Assign Cost to each Operation • Cost is based on data dependency chain • Schedule Op with lowest Cost + BB 1 b + BB 2 BB 3 BB 4 BB 5 c + BB 6 BB 7 BB 8 BB 9 + d

  10. Across HTG BB 0 a + c + + BB 1 + d d b + + + BB 2 BB 3 Conditional Speculation BB 4 BB 5 Speculate BB 6 BB 7 BB 8 Across HTG BB 9 + d Scheduling Heuristic BB 0 a + BB 1 b + BB 2 BB 3 BB 4 BB 5 c + BB 6 BB 7 BB 8 BB 9 + d

  11. Experimentation • Experiments for several transformations • Speculative Code Motions • Transformations applied during scheduling: Dynamic CSE • We have used Spark to synthesize designs derived from several industrial designs • MPEG-1 Prediction Block • GIMP Image Processing software • Scheduling Results • Number of States in FSM • Cycles on Longest Path through Design • VHDL: Logic Synthesis • Critical Path Length (ns) • Unit Area

  12. Speculative Code Motions 50 % reduction in delay with 20 % Area increase Code Motions: Logic Synthesis Results Within Basic Blocks & Across Hierar. Blocks + Reverse Speculation & Early Condition Execution + Speculation Condition Speculation

  13. Speculative Code Motions + Dynamic CSE 75 % reduction in delay with No Area increase Dynamic CSE 30 % reduction in delay, 25 % reduction in Area CSE/Dynamic CSE Results All Code Motions Enabled + Only Dyanmic CSE + Only CSE + CSE & Dynamic CSE

  14. Conclusions • Comprehensive High-Level Synthesis framework • Behavioral C to RTL VHDL • Hierarchical IR + Parallelizing Transformations • Toolbox of Transformations guided by Heuristics • Basic compiler transformations: CSE, Copy Propagation • Platform for applying Coarse and Fine-grain Optimizations • Experimentation with large industrial applications

  15. Thank You

  16. Additional Slides

  17. Eliminating Dependencies by Renaming

More Related