projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 785acce) | patch
AMDGPU: Fix v2f16 fneg/fabs pattern
authorMatt Arsenault <Matthew.Arsenault@amd.com>
Tue, 22 May 2018 20:13:34 +0000 (20:13 +0000)
committerMatt Arsenault <Matthew.Arsenault@amd.com>
Tue, 22 May 2018 20:13:34 +0000 (20:13 +0000)
commit606bc315d66e9aa077d04b1076227263b9c971e6
treecc9dbfd6d443e465d2bdf6bcc1d5ae423484c055tree | snapshot
parent785acce51dc3f1d3398222d5bfa0ab2f2dfa8442commit | diff
AMDGPU: Fix v2f16 fneg/fabs pattern

The integer operation convertion for some reason only happens
if the source is a bitcast from an integer, which happens to
always be the situation when the result is loaded. Add
an additional pattern for when the source operation is really
an FP operation.

llvm-svn: 333019
llvm/lib/Target/AMDGPU/SIInstructions.td diff | blob | history
llvm/test/CodeGen/AMDGPU/fneg-fabs.f16.ll diff | blob | history
Domain: System / Toolchain;