[PATCH] D143731: [AMDGPU] Break-up large PHIs for DAGISel

[Pierre van Houtryve via Phabricator via llvm-commits]

Pierre-vh updated this revision to Diff 501080.
Pierre-vh added a comment.

- Reduce threshold to 32.
  - Lots more test changes. I disabled the transform in some of them to (hopefully) preserve the test's intent better.
  - `loop-live-out-copy-undef-subrange.ll` takes the biggest regression. I think we may not need to do this transform if an incoming value is an arg? Args will be copied anyway so we don't really gain anything in those cases. Thoughts?
- Preserve debug info better


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D143731/new/

https://reviews.llvm.org/D143731

Files:
  llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
  llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
  llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-break-large-phis.ll
  llvm/test/CodeGen/AMDGPU/collapse-endcf.ll
  llvm/test/CodeGen/AMDGPU/cross-block-use-is-not-abi-copy.ll
  llvm/test/CodeGen/AMDGPU/debug-value.ll
  llvm/test/CodeGen/AMDGPU/early-if-convert.ll
  llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll
  llvm/test/CodeGen/AMDGPU/insert_vector_elt.ll
  llvm/test/CodeGen/AMDGPU/loop-live-out-copy-undef-subrange.ll
  llvm/test/CodeGen/AMDGPU/mfma-loop.ll
  llvm/test/CodeGen/AMDGPU/subreg-coalescer-undef-use.ll
  llvm/test/CodeGen/AMDGPU/tuple-allocation-failure.ll
  llvm/test/CodeGen/AMDGPU/v1024.ll
  llvm/test/CodeGen/AMDGPU/wqm.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D143731.501080.patch
Type: text/x-patch
Size: 207276 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230228/00c70c5f/attachment-0001.bin>