projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: dab7bda) | patch
AMDGPU/GlobalISel: Avoid handling registers twice in waterfall loops
authorMatt Arsenault <Matthew.Arsenault@amd.com>
Sat, 1 Feb 2020 16:29:41 +0000 (11:29 -0500)
committerMatt Arsenault <Matthew.Arsenault@amd.com>
Thu, 6 Feb 2020 17:38:24 +0000 (09:38 -0800)
commit5a8c0f552b382cbf4397aff9cc3b0d9418bdf3ff
treee7af52612f591cc587c091165140a255b495b45etree | snapshot
parentdab7bdad049221113ccf8c8ed9f7ead98809cef1commit | diff
AMDGPU/GlobalISel: Avoid handling registers twice in waterfall loops

When multiple instructions are moved into a waterfall loop, it's
possible some of them re-use the same operands. Avoid creating
multiple sequences of readfirstlanes for them. None of the current
uses will hit this, but will be used in a future patch.
llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp diff | blob | history
Domain: System / Toolchain;