[llvm] [AMDGPU] Code clean up and denorm check removal for bf16 omod folding (PR #219042)
Changpeng Fang via llvm-commits
llvm-commits at lists.llvm.org
Fri Aug 28 12:10:26 PDT 2026
================
@@ -2505,8 +2505,9 @@ SIFoldOperandsImpl::isOMod(const MachineInstr &MI) const {
}
case AMDGPU::V_PK_MUL_BF16: {
// OMOD folding for BF16 packed multiply
- if (MFI->getMode().FP32Denormals.Output != DenormalMode::PreserveSign ||
- MI.mayRaiseFPException())
+ // PK_*_BF16 ignores DENORM controls, and always preserves denorms
----------------
changpeng wrote:
> Why is it checking for "FP32" denormals specifically? Is that the right control for BF16 denormal flushing?
Should use FP64FP16Denormals as the default. Updated, thanks.
https://github.com/llvm/llvm-project/pull/219042
More information about the llvm-commits
mailing list