projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: e016fe8) | patch
[AMDGPU] Avoid RegScavenger::forward in copyPhysReg/indirectCopyToAGPR
authorJay Foad <jay.foad@amd.com>
Mon, 15 May 2023 13:42:14 +0000 (14:42 +0100)
committerJay Foad <jay.foad@amd.com>
Tue, 16 May 2023 14:51:31 +0000 (15:51 +0100)
commit64c938e8e31d311829c525b77e05dfd798c8d88f
tree519cee19761d781ae6a0add8a7a6ec8c1ff65abatree | snapshot
parente016fe817c8e4774c3996314b9519f8d856cc616commit | diff
[AMDGPU] Avoid RegScavenger::forward in copyPhysReg/indirectCopyToAGPR

RegScavenger::backward is preferred because it does not rely on accurate
kill flags.

Differential Revision: https://reviews.llvm.org/D150571
llvm/lib/Target/AMDGPU/SIInstrInfo.cpp diff | blob | history
Domain: System / Toolchain;