[all-commits] [llvm/llvm-project] be6a2b: [AArch64] Prohibit vectorizing non-native [de]inte...

[Utpal Bora via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: be6a2b4fddb62515049789e292fa166790a5707b
      https://github.com/llvm/llvm-project/commit/be6a2b4fddb62515049789e292fa166790a5707b
  Author: Utpal Bora <utpal.bora at cl.cam.ac.uk>
  Date:   2026-09-14 (Mon, 14 Sep 2026)

  Changed paths:
    M llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
    A llvm/test/Transforms/LoopVectorize/AArch64/sve-non-native-interleaved-accesses.ll

  Log Message:
  -----------
  [AArch64] Prohibit vectorizing non-native [de]interleaved pattern for fixed-length SVE (#219244)

This patch prevents LoopVectorize from using non-native interleaved
memory access groups with wider fixed-length SVE.

For wider fixed-length SVE, non-native factors produce awkward wide
shuffles that lower poorly with extra permutes, scalar lane moves, and
sometimes stack traffic. Native interleave factors and NEON-sized fixed
vectors remain allowed.



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications