Automatic Speed-Up of Image Processing Software


Target Software for Parallelization

  • Still Image Processing Software (7000 lines)
  • C code generated from MATLAB/Simulink®
  • Optimizing of the program for maximum parallelism
  • - Renamed array member of structure for data dependency reduction
  • - Automatic parallelization using the OSCARTech® Compiler

Execution Time and Speedup Ratio

  • Evaluated on Intel® Xeon® CPU X5670 (Quad Core, 2.93GHz per core)

Analysis : Visualization of Parallelism in a Program

  • Automatic extraction of 480 macro tasks
  • Increase exploitation of parallelism by optimizing the software before applying OSCARTech® Compiler


Analysis : Visualization of Macro Task Assignment to Cores

  • Automatic assignment of 480 macro tasks to processor cores according to the execution time and data dependencies
  • Visualization of bottlenecks during program execution