[all-commits] [llvm/llvm-project] 740169: [AMDGPU] Add threshold for DPP atomic optimizer on...

[YaFan Tao via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 7401694a38d9126bc910d3edf7e04331a66befa4
      https://github.com/llvm/llvm-project/commit/7401694a38d9126bc910d3edf7e04331a66befa4
  Author: YaFan Tao <765370813 at qq.com>
  Date:   2026-07-14 (Tue, 14 Jul 2026)

  Changed paths:
    M llvm/docs/AMDGPUUsage.rst
    M llvm/lib/IR/Verifier.cpp
    M llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
    A llvm/test/CodeGen/AMDGPU/atomic_optimizations_dpp_lds_expected_active_lanes.ll
    A llvm/test/Verifier/amdgpu-expected-active-lanes.ll

  Log Message:
  -----------
  [AMDGPU] Add threshold for DPP atomic optimizer on LDS atomics (#186762)

Add amdgpu.expected.active.lane metadata which can be applied to operations by a 
compiler front-end.  Do not apply the DPP atomic optimizer to LDS atomics where
less than five active lanes are expected.
This is an empirically derived threshold based on GFX11 and GFX12 testing.

---------

Co-authored-by: YaFan <YaFan.Tao at amd.com>



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