[PATCH] D79763: [mlir][Linalg] Add canonicalization pattern to drop unit-trip count loops.

Mahesh Ravishankar via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon May 11 23:11:30 PDT 2020


mravishankar created this revision.
Herald added subscribers: llvm-commits, Kayjukh, frgossen, grosul1, Joonsoo, stephenneuendorffer, liufengdb, aartbik, lucyrfox, mgester, arpith-jacob, nicolasvasilache, antiagainst, shauheen, jpienaar, rriddle, mehdi_amini.
Herald added a reviewer: nicolasvasilache.
Herald added a project: LLVM.
mravishankar added a child revision: D79764: [mlir][Linalg] Allow reshapes to collapse to a zero-rank tensor..

Adding unit-trip count loops adds complications to down-stream
progressive lowering. For example when tiling the Linalg operation for
distributing to GPUs unit-trip count loops affect the way the lopo is
distributed to workgroups/threadblocks. This is first of the set of
patterns that canonicalize the use of unit-extent dimensions for
specifying broadcasts.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D79763

Files:
  mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td
  mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOpsInterface.td
  mlir/include/mlir/Dialect/Linalg/IR/LinalgTraits.h
  mlir/lib/Dialect/Linalg/IR/LinalgOps.cpp
  mlir/lib/Dialect/Linalg/Transforms/Fusion.cpp
  mlir/test/Dialect/Linalg/canonicalize.mlir

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D79763.263348.patch
Type: text/x-patch
Size: 13020 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200512/806fd27c/attachment-0001.bin>


More information about the llvm-commits mailing list