[all-commits] [llvm/llvm-project] 25485b: [SLP] Use getRegUsageForType() rather than getNumb...

[Ryan Buchner via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 25485b3c4964ab704e7d4961b67ab2501fedbdbe
      https://github.com/llvm/llvm-project/commit/25485b3c4964ab704e7d4961b67ab2501fedbdbe
  Author: Ryan Buchner <rbuchner at qti.qualcomm.com>
  Date:   2026-09-24 (Thu, 24 Sep 2026)

  Changed paths:
    M llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
    M llvm/lib/Transforms/Vectorize/SLPVectorizer/SLPTypeUtils.cpp
    M llvm/lib/Transforms/Vectorize/SLPVectorizer/SLPTypeUtils.h
    M llvm/test/Transforms/SLPVectorizer/RISCV/wide-load-reduction.ll

  Log Message:
  -----------
  [SLP] Use getRegUsageForType() rather than getNumberOfParts() when determining reduction width (#226277)

`getRegUsageForType()` is more appropriate for comparison to total
number of registers. `getNumberOfParts()` is legalization cost based,
but in this situation, we are looking to make sure we don't create
vectors that are widen than the register file can handle.



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