[all-commits] [llvm/llvm-project] 2e2655: [AMDGPU] Require flushed FP16 denormals for the ma...

[Dmitry Sidorov via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 2e2655f10d1a05bb585e172083b448a25a5b7434
      https://github.com/llvm/llvm-project/commit/2e2655f10d1a05bb585e172083b448a25a5b7434
  Author: Dmitry Sidorov <Dmitry.Sidorov at amd.com>
  Date:   2026-09-21 (Mon, 21 Sep 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/VOP3PInstructions.td
    M llvm/test/CodeGen/AMDGPU/mad-mix-fptrunc-rounding.ll
    M llvm/test/CodeGen/AMDGPU/mad-mix-hi.ll
    M llvm/test/CodeGen/AMDGPU/mad-mix-lo.ll

  Log Message:
  -----------
  [AMDGPU] Require flushed FP16 denormals for the mad-mix f16 results (#224911)

v_mad_mixlo_f16 and v_mad_mixhi_f16 are the unfused gfx900 forms and flush 16-bit denormals, so a denormal half result is written as zero even when the FP16 mode asks for it to be kept, while the patterns only required the FP32 mode to flush and that is the one a HIP compile turns off on its own.

Assisted-by: Claude Code Opus 5



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications