[PATCH] D49065: AMDGPU: Stop wasting argument registers with v3i32/v3f32

[Matt Arsenault via Phabricator via llvm-commits]

arsenm created this revision.
arsenm added reviewers: tstellar, kzhuravl, rampitec, mareko.
Herald added subscribers: t-tye, tpr, dstuttard, yaxunl, nhaehnle, wdng.

Since v4i32/v4f32 are legal, SelectionDAGBuilder promotes these
to v4i32/v4f32 arguments which consume an additional register.
In addition to wasting argument space, this produces extra
instructions since now it appears the 4th vector component has
a meaningful value to most combines.


https://reviews.llvm.org/D49065

Files:
  lib/Target/AMDGPU/SIISelLowering.cpp
  lib/Target/AMDGPU/SIISelLowering.h
  test/CodeGen/AMDGPU/call-argument-types.ll
  test/CodeGen/AMDGPU/fmaxnum.ll
  test/CodeGen/AMDGPU/fminnum.ll
  test/CodeGen/AMDGPU/function-args.ll
  test/CodeGen/AMDGPU/function-returns.ll
  test/CodeGen/AMDGPU/ret.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D49065.154548.patch
Type: text/x-patch
Size: 15058 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180709/09293a5b/attachment.bin>