diff options
| author | Matt Arsenault <Matthew.Arsenault@amd.com> | 2021-06-12 11:21:57 -0400 |
|---|---|---|
| committer | Matt Arsenault <Matthew.Arsenault@amd.com> | 2021-06-18 19:09:03 -0400 |
| commit | d6467e00dfccf10aa91e5a1d55d981a9b13642a3 (patch) | |
| tree | ac4d8b3102e0beb896629ccf80c77d9fcda265d2 /libcxx/include/__algorithm/replace.h | |
| parent | fb32de9e97af0921242a021e30020ffacf7aa6e2 (diff) | |
| download | llvm-d6467e00dfccf10aa91e5a1d55d981a9b13642a3.zip llvm-d6467e00dfccf10aa91e5a1d55d981a9b13642a3.tar.gz llvm-d6467e00dfccf10aa91e5a1d55d981a9b13642a3.tar.bz2 | |
AMDGPU: Fix infinite loop in DAG combine with fneg + fma
We were not reporting isFNegFree for v2f32, although it is effectively
free after legalization. The generic combine was pulling fneg out of
the fma source operands, and the AMDGPU combine was doing the
opposite.
Diffstat (limited to 'libcxx/include/__algorithm/replace.h')
0 files changed, 0 insertions, 0 deletions