[all-commits] [llvm/llvm-project] 9ce06f: [AMDGPU] Track LDS DMA VGPR source reads in expert...

[Alexander Weinrauch via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 9ce06fd29838777aaa89f133fe4b9f764be11d35
      https://github.com/llvm/llvm-project/commit/9ce06fd29838777aaa89f133fe4b9f764be11d35
  Author: Alexander Weinrauch <alexander.weinrauch at amd.com>
  Date:   2026-06-15 (Mon, 15 Jun 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
    M llvm/test/CodeGen/AMDGPU/expert_scheduling_gfx1250.mir

  Log Message:
  -----------
  [AMDGPU] Track LDS DMA VGPR source reads in expert scheduling (#203292)

LDS DMA instructions are marked as VALU but we need to treat them as
FLAT instructions for expert mode waitcnt insertion. This meant we were
missing a `vm_vsrc` wait for WAR and RAW on input registers of the LDS
DMA instruction.

I added lit tests for other memory ops as well to catch future
regressions.



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