Introduction to Parallel Computation
Exercise 1: Automatic Parallelization

RISC-Linz logo

Write on the Convex SPP parallel programs in Fortran and C solving the "All Pairs Shortest Path" problem. Use the enclosed sequential programs as the starting point and apply the parallelization features of the Convex SPP Fortran and C compilers1.

  1. Compile the sequential programs with -O2. Measure the execution times for matrix dimensions 256 and 512 and take these times as the base times of all your comparisons with the parallel programs.
  2. Compile the programs with -O3 without changes and measure the execution times for both input sizes and 1, 2, 4, 8, 16 processors. Analyze the messages generated by the compiler (-or all) and explain them in detail. Use the CXpa performance profiler (-pa) to analyze the performance of your program using Determine which fraction of execution time was spent in which part (thread/function/loop) of the program, document your findings, and compare the C program with the Fortran program.
  3. Try to improve the programs (wherever appropriate) by rewriting the sequential code and recompiling them with -O3 without explicit pragmas or parallelization constructs. Proceed in the analysis as for the previous case.
  4. Try to improve the programs (wherever appropriate) by using pragmas and/or explicit parallelization constructs and/or compilation flags. recompiling them with -O3. Proceed in the analysis as for the previous case.
  5. Write a report documenting your results including

Deliver your preliminary report by the first deadline shown below; be prepared to present and to discuss it in class. Based on our discussions, overwork your programs and deliver the final report by the second deadline shown below. Both reports will be used for grading.

First Report and Presentation: November 24
Final Report and Presentation: December 9
  • Footnotes

  • Maintained by: Wolfgang Schreiner
    Last Modification: November 4, 1997

    [Up] [RISC-Linz] [University] [Search]