projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: cff73a2) | patch
AMDGPU: Ignore any_extend in mul24 combine
authorMatt Arsenault <Matthew.Arsenault@amd.com>
Wed, 9 May 2018 21:11:35 +0000 (21:11 +0000)
committerMatt Arsenault <Matthew.Arsenault@amd.com>
Wed, 9 May 2018 21:11:35 +0000 (21:11 +0000)
commiteac81b244843423a58ed6301c5fb8ca25ffa2308
tree40d150b89a1c43be306dae5cb633c20c628ce7b1tree | snapshot
parentcff73a211802895f4d525ffc37cf53f4b7df1f15commit | diff
AMDGPU: Ignore any_extend in mul24 combine

If a multiply is truncated, SimplifyDemandedBits
sometimes turns a zero_extend of the inputs into an
any_extend, which makes the known bits computation unhelpful.
Ignore these and compute known bits for the underlying value,
since we insert the correct extend type after.

llvm-svn: 331919
llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp diff | blob | history
llvm/test/CodeGen/AMDGPU/mul_uint24-amdgcn.ll diff | blob | history
Domain: System / Toolchain;