[PATCH] D95795: [AMDGPU] Add new CostPerUse values for VGPRs

[Christudasan Devadasan via Phabricator via llvm-commits]

cdevadas created this revision.
cdevadas added a reviewer: arsenm.
Herald added subscribers: kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, jvesely, kzhuravl.
cdevadas requested review of this revision.
Herald added subscribers: llvm-commits, wdng.
Herald added a project: LLVM.

We introduced more scratch VGPRs with D76356 <https://reviews.llvm.org/D76356>.
To yield a better occupancy, the CSRs and
the scratch registers were interleaved at a
regular interval.

This patch is to have a balanced allocation of
both CSRs and scratch registers by setting high
cost value for higher indices. This should improve the
cases that ended up with a higher register usage when
one set of VGPRs (either CSR or scratch) is over
allocated than the other.

With D86836 <https://reviews.llvm.org/D86836> targets can support multiple cost models.
The new cost values will be applied only in the presence
of a calling convention. Otherwise, the default
cost model will be used.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D95795

Files:
  llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
  llvm/lib/Target/AMDGPU/SIRegisterInfo.h
  llvm/lib/Target/AMDGPU/SIRegisterInfo.td
  llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
  llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll
  llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll
  llvm/test/CodeGen/AMDGPU/splitkit-copy-live-lanes.mir
  llvm/test/CodeGen/AMDGPU/vgpr-tuple-allocation.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D95795.320501.patch
Type: text/x-patch
Size: 151588 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210201/f9a8c00b/attachment-0001.bin>