[all-commits] [llvm/llvm-project] 8ecbad: [AMDGPU] Skip s_delay_alu for gfx1250 WMMA C-reuse...

[Petar Jovanovic via All-commits]

  Branch: refs/heads/users/petar-jovanovic/skip-s_delay_alu
  Home:   https://github.com/llvm/llvm-project
  Commit: 8ecbad03e59743ab6ad302d0668a341a647eb406
      https://github.com/llvm/llvm-project/commit/8ecbad03e59743ab6ad302d0668a341a647eb406
  Author: Petar Jovanovic <pjovanov at amd.com>
  Date:   2026-08-05 (Wed, 05 Aug 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/AMDGPUInsertDelayAlu.cpp
    M llvm/test/CodeGen/AMDGPU/insert-delay-alu-wmma-xdl.mir

  Log Message:
  -----------
  [AMDGPU] Skip s_delay_alu for gfx1250 WMMA C-reuse chains

Consecutive gfx1250 WMMA/SWMMAC ops accumulating into the same matrix C
register reuse the accumulator in place, so the tied srcC read is
omitted and no delay is needed. AMDGPUInsertDelayAlu did not model this
and emitted an s_delay_alu that stalls the reuse chain.

Detect a C-reuse edge (tied srcC overlaps the previous XDL WMMA's dest,
with no intervening instruction) and skip the delay for that operand.
The change is state-neutral and safe, since GFX12+ interlocks the
matrix C hazard in hardware.



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