[all-commits] [llvm/llvm-project] aaa414: [Valuetracking] Use all FPClasses ordering informa...

[Niklas Ulvinge via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: aaa414e4e053a6195a33e82012b8e1141536b7f4
      https://github.com/llvm/llvm-project/commit/aaa414e4e053a6195a33e82012b8e1141536b7f4
  Author: Niklas Ulvinge <niklas.ulvinge at arm.com>
  Date:   2026-06-15 (Mon, 15 Jun 2026)

  Changed paths:
    M llvm/include/llvm/ADT/FloatingPointMode.h
    M llvm/lib/Support/FloatingPointMode.cpp
    M llvm/lib/Support/KnownFPClass.cpp
    M llvm/test/Transforms/Attributor/nofpclass-minimum-maximum.ll
    M llvm/test/Transforms/Attributor/nofpclass-minimumnum-maximumnum.ll
    M llvm/test/Transforms/Attributor/nofpclass-minnum-maxnum.ll
    M llvm/test/Transforms/InstSimplify/known-never-infinity.ll

  Log Message:
  -----------
  [Valuetracking] Use all FPClasses ordering information for min/max (#199651)

Min/Max functions can exclude more FPClasses than
OrderedLessThanZeroMask/OrderedGreaterThanZeroMask. Now it excludes all
analyzable FPClasses, of which +/-Inf are the most useful.

This enhances analysis for transforms which need to exclude Inf.
Here is a simplified example: 0*y -> 0 is only correct if y cannot be
Inf or NaN, otherwise it may be NaN.



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