This is an archive of the discontinued LLVM Phabricator instance.

[AMDGPU] G_IS_FPCLASS lower() support for IEEE fp types
ClosedPublic

Authored by JanekvO on Dec 1 2022, 10:38 AM.

Details

Summary

Simplified globalisel version of sdag's expandIS_FPCLASS.

Diff Detail

Event Timeline

JanekvO created this revision.Dec 1 2022, 10:38 AM
Herald added a project: Restricted Project. · View Herald TranscriptDec 1 2022, 10:38 AM
JanekvO requested review of this revision.Dec 1 2022, 10:38 AM
Herald added a project: Restricted Project. · View Herald TranscriptDec 1 2022, 10:38 AM
JanekvO added reviewers: arsenm, sepavloff, Restricted Project.Dec 1 2022, 10:42 AM
arsenm added a comment.Dec 5 2022, 5:41 PM

LGTM except for some nits

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
7243

We already have getFltSemanticForLLT

7265

I thought this was set already before?

7341

No ISD::

7371

No ISD::

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
992

Ought to have a fallback lower at the end of each too

JanekvO updated this revision to Diff 480580.Dec 6 2022, 12:36 PM
JanekvO marked 5 inline comments as done.
  • Address comments
arsenm accepted this revision.Dec 6 2022, 1:19 PM
arsenm added inline comments.
llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.f16.ll
414–434

We really need to fix this ABI discrepancy for i16/f16 for 16-bit-insts targets

This revision is now accepted and ready to land.Dec 6 2022, 1:19 PM
This revision was automatically updated to reflect the committed changes.