[all-commits] [llvm/llvm-project] 070711: Class for checking floating-point properties

[Serge Pavlov via All-commits]

  Branch: refs/heads/users/spavloff/fpopt
  Home:   https://github.com/llvm/llvm-project
  Commit: 07071121f2740cb170d94828b83c76830f8d3339
      https://github.com/llvm/llvm-project/commit/07071121f2740cb170d94828b83c76830f8d3339
  Author: Serge Pavlov <sepavloff at gmail.com>
  Date:   2026-06-22 (Mon, 22 Jun 2026)

  Changed paths:
    M llvm/include/llvm/Analysis/InstSimplifyFolder.h
    M llvm/include/llvm/Analysis/InstructionSimplify.h
    M llvm/include/llvm/Analysis/TargetFolder.h
    M llvm/include/llvm/IR/ConstantFolder.h
    M llvm/include/llvm/IR/FMF.h
    M llvm/include/llvm/IR/FPEnv.h
    A llvm/include/llvm/IR/FPTransformChecker.h
    M llvm/include/llvm/IR/IRBuilderFolder.h
    M llvm/include/llvm/IR/NoFolder.h
    M llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
    M llvm/lib/Analysis/InlineCost.cpp
    M llvm/lib/Analysis/InstructionSimplify.cpp
    M llvm/lib/IR/CMakeLists.txt
    A llvm/lib/IR/FPTransformChecker.cpp
    M llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
    M llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
    M llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
    M llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
    M llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
    M llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp
    M llvm/lib/Transforms/Vectorize/VectorCombine.cpp
    M llvm/test/Transforms/InstCombine/fdim.ll
    M llvm/test/Transforms/InstCombine/fma.ll
    M llvm/test/Transforms/NewGVN/commute.ll

  Log Message:
  -----------
  Class for checking floating-point properties

Applicability of a particular transformation to floating-point
expressions may depend on many parameters. Previously these parameters
were mainly represented by fast-math flags, and in some cases they were
supplemented by rounding mode and exception behavior, and all were
passed as separate arguments to the relevant functions. Implementing
more elaborated support of floating-point requires access to other
properties, including function attributes (such as denormal mode or sNaN
support). Passing these parameters as separate arguments is
inconvenient. To address this problem a special class that collects all
the necessary information is introduced in this commit.

The class is a wrapper over an integer value, where various bit fields
represent properties, that transformations may depend on. This allows
instances of the class to be passed by value. It incorporates fast-math
flags and borrows part of the FastMathFlags interface.

This change was intended to be NFC, but in some cases, the behavior
differs, because the floating-point operation information could
previously be lost when traveling throughout function call stack.



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