projects / oota-llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 175c683) | patch
AMDGPU: Stop reserving v[254:255]
authorMatt Arsenault <Matthew.Arsenault@amd.com>
Tue, 20 Oct 2015 03:59:58 +0000 (03:59 +0000)
committerMatt Arsenault <Matthew.Arsenault@amd.com>
Tue, 20 Oct 2015 03:59:58 +0000 (03:59 +0000)
commit08e1ec066dc01294e34e7adbe877a7e06d4f00a7
tree6f501e4c55e927b3a01d23d569f3fa68a8de393etree | snapshot
parent175c683f63a4f1c3c06205996a4ce395349cd313commit | diff
AMDGPU: Stop reserving v[254:255]

This wasn't doing anything useful. They weren't explicitly used
anywhere, and the RegScavenger ignores reserved registers.

This for some reason caused a random scheduling change in the test.
Getting the check lines to pass is too frustrating, and there's probably
not too much value in checking the vector case's operands N times.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250794 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/AMDGPU/SIRegisterInfo.cpp diff | blob | history
test/CodeGen/AMDGPU/udivrem.ll diff | blob | history
C/C++ LLVM-based compilers that forbids OOTA behaviors