llvm-project

fab6b85e - [AMDGPU] Fold fmul with zero checks to fmul.legacy

[AMDGPU] Fold fmul with zero checks to fmul.legacy Recognize f32 multiplication with zero checks in AMDGPUCodeGenPrepare, including fabs/fneg wrappers. Example: ``` (x == 0 || y == 0) ? +0.0 : x * y ==> llvm.amdgcn.fmul.legacy(x, y) ```