Skip to content

[X86][clang] 128bit floating-point operations in x86 machines #77401

Closed
#138092
Closed
[X86][clang] 128bit floating-point operations in x86 machines#77401
#138092
Labels
backend:X86floating-pointFloating-point math

Description

@CoTinker
CoTinker
opened on Jan 9, 2024

testcase:

#include <fenv.h>
#include <stdlib.h>
#include <stdio.h>

int main (void)
{
  volatile __float128 a = 0x0.fffp-16382q, b = 0x0.fffp0q, c;
  c = a / b;
  if (fetestexcept (FE_UNDERFLOW | FE_INEXACT))
    printf("Fail1\n");
  if (c != 0x1p-16382q)
    printf("Fail2\n");
  exit (0);
}

compile command:

% clang test.c -o test -lm -m32
% ./test
Fail1
Fail2

compile in x86_64 will return normally, but x86 output Fail.
https://godbolt.org/z/vhdsMhd8d

Activity

  1. github-actions
    added
    clangClang issues not falling into any other category
    on Jan 9, 2024
  2. EugeneZelenko
    added
    backend:X86
    floating-pointFloating-point math
    and removed
    clangClang issues not falling into any other category
    on Jan 9, 2024

  3. llvmbot commented on Jan 9, 2024

    @llvmbot
    llvmbot
    on Jan 9, 2024
    Member

    @llvm/issue-subscribers-backend-x86

    Author: Longsheng Mou (CoTinker)

    Details testcase: 
    #include &lt;fenv.h&gt;
#include &lt;stdlib.h&gt;
#include &lt;stdio.h&gt;

int main (void)
{
  volatile __float128 a = 0x0.fffp-16382q, b = 0x0.fffp0q, c;
  c = a / b;
  if (fetestexcept (FE_UNDERFLOW | FE_INEXACT))
    printf("Fail1\n");
  if (c != 0x1p-16382q)
    printf("Fail2\n");
  exit (0);
}

    compile command:

    % clang test.c -o test -lm -m32
% ./test
Fail1
Fail2

    compile in x86_64 will return normally, but x86 output Fail.
    https://godbolt.org/z/vhdsMhd8d

  4. arsenm commented on Jan 9, 2024

    @arsenm
    arsenm
    on Jan 9, 2024
    Contributor

    Technically you need to enable FENV_ACCESS for this program

  5. CoTinker commented on Jan 9, 2024

    @CoTinker
    CoTinker
    on Jan 9, 2024
    ContributorAuthor

    Technically you need to enable FENV_ACCESS for this program

    #pragma STDC FENV_ACCESS ON
    It seems not work.

  6. CoTinker commented on Jan 15, 2024

    @CoTinker
    CoTinker
    on Jan 15, 2024
    ContributorAuthor

    CC @RKSimon @phoebewang

  7. hstk30-hw commented on Jan 20, 2024

    @hstk30-hw
    hstk30-hw
    on Jan 20, 2024
    Contributor

    @arsenm Hi, enable FENV_ACCESS still fail. Does FENV_ACCESS supported in m32 ?
    https://godbolt.org/z/MsYTe1cPc

  8. hstk30-hw commented on Jan 22, 2024

    @hstk30-hw
    hstk30-hw
    on Jan 22, 2024
    Contributor

    CC @brad0

  9. hstk30-hw commented on Jan 30, 2024

    @hstk30-hw
    hstk30-hw
    on Jan 30, 2024
    Contributor

    https://godbolt.org/z/4KbaWcvsG

    I see the asm, clang x86 with m32 abi is not compatible with libgcc.

  10. hstk30-hw commented on Jan 30, 2024

    @hstk30-hw
    hstk30-hw
    on Jan 30, 2024
    Contributor

    CC @hvdijk

  11. arsenm commented on Feb 1, 2024

    @arsenm
    arsenm
    on Feb 1, 2024
    Contributor

    @arsenm Hi, enable FENV_ACCESS still fail. Does FENV_ACCESS supported in m32 ? https://godbolt.org/z/MsYTe1cPc

    I think it should work the same as x64

  12. hstk30-hw commented on Feb 1, 2024

    @hstk30-hw
    hstk30-hw
    on Feb 1, 2024
    Contributor

    In fact, not work! I guess float128 (long double) not align to 16 in x86-32.
    I'm reading the x86 isel dag->dag pass, maybe lowering mir gets mistake in the step.

  13. hstk30-hw commented on Feb 1, 2024

    @hstk30-hw
    hstk30-hw
    on Feb 1, 2024
    Contributor

    Do you have any idea about this problem?
    It seems a legacy.
    https://lists.llvm.org/pipermail/llvm-dev/2016-December/108033.html
    Seems no body care about x86-32? @arsenm

  14. 16 remaining items

  15. llvm-sync
    added a commit that references this issue on Jul 15, 2025

    Automerge: [X86] Update the fp128/i128 test to show stack alignment (…

    80b50e5
  16. tgross35
    added a commit that references this issue on Jul 16, 2025

    [X86] Align f128 and i128 to 16 bytes when passing on x86-32

    ca1efdd
  17. nikic
    closed this as completedin #138092on Jul 17, 2025
  18. nikic
    added a commit that references this issue on Jul 17, 2025

    [X86] Align f128 and i128 to 16 bytes when passing on x86-32 (#138092)

    a78a0f8
  19. llvm-sync
    added a commit that references this issue on Jul 17, 2025

    Automerge: [X86] Align f128 and i128 to 16 bytes when passing on x86-…

    0fc839d
  20. llvm-sync
    added a commit that references this issue on Jul 29, 2025

    Automerge: [X86] Align f128 and i128 to 16 bytes when passing on x86-…

    20ced06
  21. markrvmurray
    added 2 commits that reference this issue on Jun 14, 2026

    [X86] Update the fp128/i128 test to show stack alignment (NFC) (#148753)

    c5a63d3

    [X86] Align f128 and i128 to 16 bytes when passing on x86-32 (#138092)

    b089f8a
  22. OpenHarmonySCM-noreply
    added 2 commits that reference this issue on Jul 19, 2026

    [X86] Update the fp128/i128 test to show stack alignment (NFC) (#148753)

    5de9c2d

    [X86] Align f128 and i128 to 16 bytes when passing on x86-32 (#138092)

    73d9da1
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    backend:X86floating-pointFloating-point math

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions