This is an archive of the discontinued LLVM Phabricator instance.

[TLI] Add mapping for various '__<func>_finite' forms of the math routines to SVML routines
ClosedPublic

Authored by chrischr on Apr 6 2017, 3:00 PM.

Details

Summary

This is the third of a set of patches to followup on the thread in http://lists.llvm.org/pipermail/llvm-dev/2017-March/111323.html. (Builds on the changes within the patch @ https://reviews.llvm.org/D31787)

This extends the mappings used to map some of the math function calls to the Intel SVML library. This handles the case when the functions have been redirected by the -ffast-math flag to call the '__<func>_finite' alternative routine.

Diff Detail

Event Timeline

chrischr created this revision.Apr 6 2017, 3:00 PM
andrew.w.kaylor added inline comments.
lib/Analysis/TargetLibraryInfo.cpp
1377

Is there a reason not to group these with the corresponding non-finite-only versions?

chrischr added inline comments.May 11 2017, 2:56 PM
lib/Analysis/TargetLibraryInfo.cpp
1377

No, and looking back at how the others are ordered/grouped, I think it makes sense to move these to be with those. I'll update it.

chrischr updated this revision to Diff 98769.May 12 2017, 8:14 AM

Updated patch to group the new SVML function mappings to be with the existing function groupings.

This revision is now accepted and ready to land.May 12 2017, 10:17 AM
This revision was automatically updated to reflect the committed changes.