[PATCH] D36626: [test-suite]: Adding RSBench proxy-app.

[Hal Finkel via Phabricator via llvm-commits]

hfinkel added a comment.

This test does not have a numerical verification output, but nevertheless, I'd prefer that we not depend explicitly on libc's rand. Different libc implementations have different implementations of rand, and I'd rather not have that be a cause of any latter difficultly tracking down performance differences between systems/configurations.

The easiest way to deal with this is probably to substitute our own rand implementation. I wrote one of these for the original XSBench patch (see glibc_compat_rand.{c.h} in https://reviews.llvm.org/D27311). Can you grab those files from there and add them to this patch and then, in rsbench.h, add something like:

  #ifndef NO_GLIBC_COMPAT_RAND
  #define rand glibc_compat_rand
  #define srand glibc_compat_srand
  #endif


https://reviews.llvm.org/D36626