This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug target/51408] New: Miscompilation in arm.md:*minmax_arithsi


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51408

             Bug #: 51408
           Summary: Miscompilation in arm.md:*minmax_arithsi
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: kazu@gcc.gnu.org


The following testcase, reduced from efgcvt_r.c:fcvt_r in glibc, gets
miscompiled:

extern void abort (void);

int __attribute__((noinline))
foo (int a, int b)
{
  int max = (b > 0) ? b : 0;
  return max - a;
}

int
main (void)
{
  if (foo (3, -1) != -3)
    abort ();
  return 0;
}

arm-none-eabi-gcc -O1 generates:

foo:
    @ Function supports interworking.
    @ args = 0, pretend = 0, frame = 0
    @ frame_needed = 0, uses_anonymous_args = 0
    @ link register save eliminated.
    cmp    r1, #0
    rsbge    r0, r0, r1
    bx    lr

This would be equivalent to:

  return b >= 0 ? b - a : a;

which is different from:

  return b >= 0 ? b - a : -a;

That is, in assembly code, we should have an "else" clause like so:

    cmp    r1, #0
    rsbge    r0, r0, r1  <- then clause
    rsblt    r0, r0, #0  <- else clause
    bx    lr

This bug comes from gcc/config/arm/arm.md:*minmax_arithsi, not
outputting an else clause on the MINUS case.

I've tested a patch and will post it shortly.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]