[all-commits] [llvm/llvm-project] 9d08f2: [AMDGPU] Use reductions instead of scans in the at...

[Jay Foad via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 9d08f276d79b59e3d1ad3db3db19077284524ca3
      https://github.com/llvm/llvm-project/commit/9d08f276d79b59e3d1ad3db3db19077284524ca3
  Author: Jay Foad <jay.foad at amd.com>
  Date:   2021-03-26 (Fri, 26 Mar 2021)

  Changed paths:
    M llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
    M llvm/lib/Target/AMDGPU/GCNSubtarget.h
    M llvm/lib/Target/AMDGPU/SIDefines.h
    M llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll

  Log Message:
  -----------
  [AMDGPU] Use reductions instead of scans in the atomic optimizer

If the result of an atomic operation is not used then it can be more
efficient to build a reduction across all lanes instead of a scan. Do
this for GFX10, where the permlanex16 instruction makes it viable. For
wave64 this saves a couple of dpp operations. For wave32 it saves one
readlane (which are generally bad for performance) and one dpp
operation.

Differential Revision: https://reviews.llvm.org/D98953