[all-commits] [llvm/llvm-project] b3eab3: [mlir][vector] Fix mask neutral value for masked f...

[lorenzo chelini via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: b3eab3a932118c3a462741000d2118428f5bd268
      https://github.com/llvm/llvm-project/commit/b3eab3a932118c3a462741000d2118428f5bd268
  Author: lorenzo chelini <l.chelini at icloud.com>
  Date:   2026-09-10 (Thu, 10 Sep 2026)

  Changed paths:
    M mlir/include/mlir/Dialect/LLVMIR/LLVMIntrinsicOps.td
    M mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp
    M mlir/test/Conversion/VectorToLLVM/vector-reduction-to-llvm.mlir
    A mlir/test/Integration/Dialect/Vector/CPU/reductions-masked-minmax-f32.mlir
    M mlir/test/Target/LLVMIR/Import/intrinsic.ll
    M mlir/test/Target/LLVMIR/llvmir-intrinsics.mlir

  Log Message:
  -----------
  [mlir][vector] Fix mask neutral value for masked fmaximum/fminimum reductions (#222275)

Masked maximumf/minimumf reductions were lowered by replacing inactive
lanes with a neutral value before applying an unmasked reduction. The
maximumf neutral was incorrectly set to the negative subnormal closest
to zero, causing it to beat any negative active value. Similarly,
minimumf used FLT_MAX, which loses to an active +Inf.
This change adds LLVM dialect support for
llvm.vp.reduce.fmaximum/fminimum and lowers masked reductions directly
to these predicated intrinsics.



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