[PATCH] D145586: [AMDGPU] Tweak PromoteAlloca limits

[Pierre van Houtryve via Phabricator via llvm-commits]

Pierre-vh created this revision.
Pierre-vh added a reviewer: arsenm.
Herald added subscribers: kosarev, foad, kerbowa, hiraditya, tpr, dstuttard, yaxunl, jvesely, kzhuravl.
Herald added a project: All.
Pierre-vh requested review of this revision.
Herald added subscribers: llvm-commits, wdng.
Herald added a project: LLVM.

- Remove the exception for `isEntryFunctionCC` - apparently it was used to work around some issue that has been fixed.
- Increase the budget to 1/2 of available VGPRs if there are more than 64 of them available.

Long-term, PromoteAlloca will need to be made a lot smarter.
This is just a band-aid fix help with abnormally high scratch usage observed in some applications (the current limit was too conservative)


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D145586

Files:
  llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
  llvm/test/CodeGen/AMDGPU/vector-alloca-limits.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D145586.503363.patch
Type: text/x-patch
Size: 9395 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230308/782069ee/attachment.bin>