[PATCH] D105193: [Matrix] Hoist address computation before multiply to enable fusion.

[Florian Hahn via Phabricator via llvm-commits]

fhahn created this revision.
fhahn added a reviewer: anemet.
Herald added subscribers: tschuett, hiraditya.
fhahn requested review of this revision.
Herald added a project: LLVM.

If the store address does not dominate the matrix multiply, try to hoist
address computation instructions without side-effects and/or memory
reads before the multiply, to allow fusion.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D105193

Files:
  llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
  llvm/test/Transforms/LowerMatrixIntrinsics/multiply-fused-dominance.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D105193.355540.patch
Type: text/x-patch
Size: 31776 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210630/9f58183f/attachment.bin>