[PATCH] D75976: [AMDGPU] Optimize AtomicOptimizer

[Sebastian Neubauer via Phabricator via llvm-commits]

Flakebi created this revision.
Flakebi added reviewers: foad, arsenm, nhaehnle.
Herald added subscribers: llvm-commits, kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, wdng, jvesely, kzhuravl.
Herald added a project: LLVM.
Flakebi added a parent revision: D65088: [AMDGPU][RFC] New llvm.amdgcn.ballot intrinsic.

Mark the ctpop as convergent so it does not get moved into the
single-lane basic block. This saves us currently one instruction.
Another way to save this instruction is reusing the saved exec register
from the inserted control flow (output of s_saveexec). This is
currently hard to do though it might work when GlobalISel gets used.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D75976

Files:
  llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
  llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll
  llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
  llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll
  llvm/test/CodeGen/AMDGPU/atomic_optimizations_pixelshader.ll
  llvm/test/CodeGen/AMDGPU/atomic_optimizations_raw_buffer.ll
  llvm/test/CodeGen/AMDGPU/atomic_optimizations_struct_buffer.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D75976.249564.patch
Type: text/x-patch
Size: 67453 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200311/85ea0048/attachment.bin>