Skip to content

Latest commit

 

History

History
35 lines (26 loc) · 1.66 KB

README.rst

File metadata and controls

35 lines (26 loc) · 1.66 KB
Raw

ppcg_polybench_test-benchmark

ppcg meets Polybench in your NVIDIA card, where

  • ppcg is a source-to source compiler
  • Polybench is a collection of structured C code
  • NVIDIA card is a device you got tired of programming directly

By design ppcg_polybench_test-benchmark carries no extra depependancies (except Python).

System requirements:

  1. ppcg tool, Polybench-C test-suite, NVIDIA toolkit, ability to run CUDA executables;
  2. NVIDIA toolchain should include gcc (or at least gcc should be able to link object code created by nvcc);
  3. Python version 2.* (mine is currently 2.7.5).
  4. Core utilities such as which and diff.

What my code does

For every test in Polybench-c suite,

  1. compiles it into ordinary CPU executable;
  2. asks ppcg to automagically convert it;
  3. crunches it here and there so it compiles;
  4. compile the crunched code to object code with nvcc;
  5. link result with gcc (hence requirement 2 is there)

How does it all work? Great, sometimes... 26 tests of 30 produce exactly same result on CPU and GPU... for smallest available datasize DATASIZE=MINI. What happens for the rest 4? One segfault and 3 slightly different results. Well, at least I think the difference is slight. Get .rez files from sample/all.cards.MINI and see for yorself

diff cholesky_sm_21__MINI.rez cholesky_cpu_MINI.rez

Quick start

Compile cuda_cardz.c, place two executables cuda_cardz and ppcg_polybench_benchmark.py, run script sample/all.cards.MINI/runme, modify it if needed.