[llvm-bugs] [Bug 38833] New: __builtin_isinf[_sign] yield incorrect results on X86 when using x87 math

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=38833

            Bug ID: 38833
           Summary: __builtin_isinf[_sign] yield incorrect results on X86
                    when using x87 math
           Product: clang
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: -New Bugs
          Assignee: unassignedclangbugs at nondot.org
          Reporter: andres at anarazel.de
                CC: llvm-bugs at lists.llvm.org

Hi,

(thanks to Andrew Gierth for the test program and help in narrowing this down)

$ cat ~/tmp/flttst.c

#include <stdio.h>
#include <math.h>

int main(int argc, char **argv)
{
    double d1 = (argc ? 1e180 : 0);
    double d2 = (argv ? 1e200 : 0);
    int r2 = __builtin_isinf_sign(d1 * d2);
    int r1 = isinf(d1 * d2);
    printf("r1 = %d, r2 = %d\n", r1, r2);
    return 0;
}

$ clang-8 -std=c99 -march=i386 -O2 -m32 ~/tmp/flttst.c -o ~/tmp/flttst && 
~/tmp/flttst
r1 = 1, r2 = 0

$ clang-8 -std=c99 -march=pentium4 -O2 -m32 ~/tmp/flttst.c -o ~/tmp/flttst &&
~/tmp/flttst
r1 = 1, r2 = 1

The issue appears to be that the use of x87 math, with its 80bit registers,
avoids the overflow.   This appears to be a C99 violation - note that gcc
enables its -fexcess-precision=standard automatically in C99 mode.


We only noticed this because clang on freebsd (and other bsds) defaults to
targetting 486, whereas linux targets pentium4 (which has non x87 math). That
caused postgres' tests to fail on 32bit x86 freebsd, after starting to use the
intrinsic, whereas linux was fine.

Regards,

Andres Freund

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20180904/3a7c632b/attachment.html>