This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU/GlobalISel: Selection for fminnum/fmaxnum
ClosedPublic

Authored by arsenm on Jul 10 2019, 7:05 AM.

Details

Summary

v2f16 case doesn't work yet because the VOP3P complex patterns haven't
been ported yet.

Diff Detail

Event Timeline

arsenm created this revision.Jul 10 2019, 7:05 AM

Why are you removing the testcases for the intrinsics?

Oh, is that because the new node causes the intrinsics to be lowered to G_FMINNUM etc.? Why doesn't this affect any other targets?

Oh, is that because the new node causes the intrinsics to be lowered to G_FMINNUM etc.? Why doesn't this affect any other targets?

I think Tom just added these here as sample intrinsics. The handling wasn't correct anyway. Other targets didn't try to do this sort of thing. I don't think any generic intrinsic should be directly handled this way, and they should all route through a G_* instruction to be legalized

nhaehnle accepted this revision.Jul 18 2019, 7:11 AM

That seems reasonable to me.

This revision is now accepted and ready to land.Jul 18 2019, 7:11 AM
arsenm closed this revision.Jul 19 2019, 7:42 AM

r366585

test/CodeGen/AMDGPU/GlobalISel/inst-select-fminnum.mir