[PATCH] D131972: [BasicTTI] Don't unroll vectorized loops (WIP).

[Florian Hahn via Phabricator via llvm-commits]

fhahn created this revision.
Herald added a subscriber: zzheng.
Herald added a project: All.
fhahn requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

Vector loops created by LoopVectorize should not benefit from
additional unrolling, because LV will interleave iterations to increase
parallelism (if profitable on the target) and almost all
canonicalization should have happened earlier.

After eaf48dd9b079c005d92ed9ef858f12bc452e71ef <https://reviews.llvm.org/rGeaf48dd9b079c005d92ed9ef858f12bc452e71ef>, loop unrolling is also
not needed to remove the backedge for vector loops where the VF >= the
vector trip count.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D131972

Files:
  llvm/include/llvm/CodeGen/BasicTTIImpl.h
  llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll
  llvm/test/Transforms/LoopVectorize/X86/metadata-enable.ll
  llvm/test/Transforms/PhaseOrdering/X86/excessive-unrolling.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D131972.453027.patch
Type: text/x-patch
Size: 191229 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220816/5ddabb1d/attachment-0001.bin>