[PATCH] D128342: [LoopVectorize] Disable tail-folding when masked interleaved accesses are unavailable

[David Sherwood via Phabricator via llvm-commits]

david-arm created this revision.
david-arm added reviewers: sdesmalen, fhahn, dmgreen, peterwaller-arm, kmclaughlin, dorit.
Herald added subscribers: ctetreau, rogfer01, bollu, hiraditya, kristof.beyls.
Herald added a project: All.
david-arm requested review of this revision.
Herald added subscribers: llvm-commits, vkmr.
Herald added a project: LLVM.

If we have interleave groups and the target does not support
masked interleaved accesses, then, if permitted, we should fall
back on normal vectorisation with a scalar epilogue. In such
cases with tail-folding we'll almost certainly go on to create vplans
with very high costs for all vector VFs and fall back on VF=1 anyway.
This is likely to be worse than if we'd just used an unpredicated
vector loop.

Tests added here:

  Transforms/LoopVectorize/tail-folding-interleave.ll
  Transforms/LoopVectorize/AArch64/sve-tail-folding-interleave.ll


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D128342

Files:
  llvm/include/llvm/Analysis/VectorUtils.h
  llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
  llvm/test/Transforms/LoopVectorize/AArch64/sve-tail-folding-interleave.ll
  llvm/test/Transforms/LoopVectorize/tail-folding-interleave.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D128342.439017.patch
Type: text/x-patch
Size: 8281 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220622/c9720f69/attachment.bin>