This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU: Fix unnecessary ands when packing f16 vectors
ClosedPublic

Authored by arsenm on Feb 27 2017, 7:09 PM.

Details

Reviewers
kzhuravl
Summary

computeKnownBits didn't handle fp_to_fp16 to report
the high bits as 0. ARM maps the generic node to an instruction
that does not modify the high bits of the register, so introduce
a target node where the high bits are known 0.

Diff Detail

Event Timeline

arsenm created this revision.Feb 27 2017, 7:09 PM
kzhuravl accepted this revision.Mar 15 2017, 9:58 AM

LGTM.

lib/Target/AMDGPU/AMDGPUISelLowering.cpp
2116

Extra new line.

2119

Extra new line.

This revision is now accepted and ready to land.Mar 15 2017, 9:58 AM
arsenm closed this revision.Mar 15 2017, 2:02 PM

r297873