[all-commits] [llvm/llvm-project] dd5f58: [AMDGPU][NFC] Extract the MFMA read-window calcula...

[Michael Halkenhäuser via All-commits]

  Branch: refs/heads/users/mhalk/fix/amdgpu-mfma-hazard-nfc-prep
  Home:   https://github.com/llvm/llvm-project
  Commit: dd5f58e8fd12936d9adc5b390e18eda67f5849f2
      https://github.com/llvm/llvm-project/commit/dd5f58e8fd12936d9adc5b390e18eda67f5849f2
  Author: Michael Halkenhaeuser <MichaelGerald.Halkenhauser at amd.com>
  Date:   2026-09-23 (Wed, 23 Sep 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
    M llvm/lib/Target/AMDGPU/GCNHazardRecognizer.h

  Log Message:
  -----------
  [AMDGPU][NFC] Extract the MFMA read-window calculation

Move the wait states a consumer needs before reading an MFMA result out
of checkMAIHazards90A into getMFMAReadWaitStates, taking the producer as
an argument, so a caller can ask about a specific producer. The partial
srcC overlap half moves into getMFMAOverlappedSrcCWaitStates. The caller
passes the producer the walk recorded, so nothing changes.

AI-assisted.


  Commit: 8cfc65e75341218fcb1562882d5aa0bfceba8793
      https://github.com/llvm/llvm-project/commit/8cfc65e75341218fcb1562882d5aa0bfceba8793
  Author: Michael Halkenhaeuser <MichaelGerald.Halkenhauser at amd.com>
  Date:   2026-09-23 (Wed, 23 Sep 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp

  Log Message:
  -----------
  [AMDGPU][NFC] Avoid duplicate MAI hazard checks

PreEmitNoopsCommon called checkMAIVALUHazards twice whenever padding was
required.

AI-assisted.


Compare: https://github.com/llvm/llvm-project/compare/dd5f58e8fd12%5E...8cfc65e75341

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