[all-commits] [llvm/llvm-project] 770694: [libc] Work around incorrect fmin/fmax results for...

[Joseph Huber via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 770694539bde0f1b706c70d87dd70777c94f178f
      https://github.com/llvm/llvm-project/commit/770694539bde0f1b706c70d87dd70777c94f178f
  Author: Joseph Huber <huberjn at outlook.com>
  Date:   2024-02-27 (Tue, 27 Feb 2024)

  Changed paths:
    M libc/src/math/amdgpu/fmax.cpp
    M libc/src/math/amdgpu/fmaxf.cpp
    M libc/src/math/amdgpu/fmin.cpp
    M libc/src/math/amdgpu/fminf.cpp
    M libc/src/math/nvptx/fmax.cpp
    M libc/src/math/nvptx/fmaxf.cpp
    M libc/src/math/nvptx/fmin.cpp
    M libc/src/math/nvptx/fminf.cpp
    M libc/test/src/math/smoke/CMakeLists.txt

  Log Message:
  -----------
  [libc] Work around incorrect fmin/fmax results for +/-x (#83158)

Summary:
The IEEE 754 standard as of the 2019 revision states that for fmin -0.0
is always less than 0.0 and for fmax 0.0 is always greater than 0.0.
These are currently not respected by the builtin value and thus cause
the tests to fail. This patch works around it in the implementation for
now by explicitly modifying the sign bit.



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