[all-commits] [llvm/llvm-project] afeee2: [VectorCombine] Use TCK_CodeSize for size-optimize...

[陈子昂 via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: afeee229f5160fecf427571141165a4d97c93f00
      https://github.com/llvm/llvm-project/commit/afeee229f5160fecf427571141165a4d97c93f00
  Author: 陈子昂 <2802328816 at qq.com>
  Date:   2026-06-13 (Sat, 13 Jun 2026)

  Changed paths:
    M llvm/lib/Transforms/Vectorize/VectorCombine.cpp
    A llvm/test/Transforms/VectorCombine/X86/fold-signbit-reduction-cmp-codesize.ll

  Log Message:
  -----------
  [VectorCombine] Use TCK_CodeSize for size-optimized functions (#202207)

VectorCombine currently uses `TCK_RecipThroughput` for all functions,
including functions optimized for size.

Select `TCK_CodeSize` when `Function::hasOptSize()` is true, covering
both `-Os` (`optsize`) and `-Oz` (`minsize`), while retaining
  `TCK_RecipThroughput` for the default optimization mode.

The X86 regression test demonstrates a sign-bit reduction where the
throughput cost model folds an `or` reduction into a `umax` reduction.
The code-size model preserves the smaller form for `optsize` and
`minsize` functions, while the default function retains the existing
throughput-oriented transformation.

  Fixes #153375.



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