[all-commits] [llvm/llvm-project] 1baa89: [AMDGPU] Optimize LDS DMA soft waitcnt

[Austin Kerbow via All-commits]

  Branch: refs/heads/users/kerbowa/direct-lds-load-memory-model-waits
  Home:   https://github.com/llvm/llvm-project
  Commit: 1baa89691a472afe572beabc4df59bc7dfc43adc
      https://github.com/llvm/llvm-project/commit/1baa89691a472afe572beabc4df59bc7dfc43adc
  Author: Austin Kerbow <Austin.Kerbow at amd.com>
  Date:   2025-05-06 (Tue, 06 May 2025)

  Changed paths:
    M llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
    M llvm/lib/Target/AMDGPU/SIInstrInfo.h
    M llvm/lib/Target/AMDGPU/SOPInstructions.td
    M llvm/test/CodeGen/AMDGPU/lds-dma-waitcnt.mir

  Log Message:
  -----------
  [AMDGPU] Optimize LDS DMA soft waitcnt

This patch adds support for optimizing `S_WAITCNT_VMCNT_LDS_DMA_soft`
pseudo instructions by analyzing whether they can be removed based on
the absence of LDS DMA operations.

These optimizations are a precursor to a dependent patch where these
waitcnt pseudos will actually be emitted by the memory legalizer. Adding
the waitcnt in the memory model first without any optimization would be
too painful of a performance penalty.



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