Re: icc bug on test-frexp-nolibm

[Eric Blake]

On 11/04/2010 11:43 AM, Eric Blake wrote:
> On 11/03/2010 05:10 PM, Eric Blake wrote:
>> I'm running out of time today to get to the bottom of this, so if anyone
>> else wants to help out, great.  If not, I'll see if I can resume testing
>> tomorrow and get a minimal test case that exposes the compiler bug, as
>> well as try and isolate ways to work around it.
>>
>> I'm on a Linux x86_64 machine with icc (ICC) 10.0 20070426.
>>
>> I hate it when a heisenbug disappears because of the mere act of adding
>> debugging hooks. :(
> 
> I've reduced it to a minimal testcase:

icc also botches this test-case, which was used by our strtod
replacement.  In an odd twist of fate, the miscompilation in strtod was
masked by the miscompilation of test-strtod, so the testsuite wasn't
picking up on this until my cleanups to the testsuite yesterday.

$ cat foo.c
#define _GNU_SOURCE 1
#include <math.h>
#include <assert.h>
double foo (double x, int i)
{
  return i ? -x : x;
}
double zero;
int main (void)
{
  assert (!signbit (foo (zero, 0)));
  assert (signbit (foo (zero, 1)));
  return 0;
}
$ icc -o foo foo.c
$ ./foo
foo: foo.c:12: main: Assertion `(( sizeof( foo (zero, 1) ) > sizeof(
double )) ? __signbitl( (long double)(foo (zero, 1)) ) : (( sizeof( foo
(zero, 1) ) == sizeof( float )) ? __signbitf( (float)(foo (zero, 1)) ) :
__signbit( (double)(foo (zero, 1)) )))' failed.
Aborted (core dumped)
$ icc -o foo foo.c -g
$ ./foo

with -g:
(gdb) disas foo
Dump of assembler code for function foo:
   0x00000000004005f8 <+0>:     push   %rbp
   0x00000000004005f9 <+1>:     mov    %rsp,%rbp
   0x00000000004005fc <+4>:     sub    $0x30,%rsp
   0x0000000000400600 <+8>:     movsd  %xmm0,-0x20(%rbp)
   0x0000000000400605 <+13>:    mov    %edi,-0x10(%rbp)
   0x0000000000400608 <+16>:    mov    -0x10(%rbp),%eax
   0x000000000040060b <+19>:    test   %eax,%eax
   0x000000000040060d <+21>:    jne    0x400617 <foo+31>
   0x000000000040060f <+23>:    fldl   -0x20(%rbp)
   0x0000000000400612 <+26>:    fstpt  -0x30(%rbp)
   0x0000000000400615 <+29>:    jmp    0x40061f <foo+39>
   0x0000000000400617 <+31>:    fldl   -0x20(%rbp)
   0x000000000040061a <+34>:    fchs
   0x000000000040061c <+36>:    fstpt  -0x30(%rbp)
   0x000000000040061f <+39>:    fldt   -0x30(%rbp)
   0x0000000000400622 <+42>:    fstpl  -0x18(%rbp)
   0x0000000000400625 <+45>:    movsd  -0x18(%rbp),%xmm0
   0x000000000040062a <+50>:    leaveq
   0x000000000040062b <+51>:    retq

without -g:
(gdb) disas foo
Dump of assembler code for function foo:
   0x0000000000400b3a <+0>:     pxor   %xmm1,%xmm1
   0x0000000000400b3e <+4>:     test   %edi,%edi
   0x0000000000400b40 <+6>:     subsd  %xmm0,%xmm1
   0x0000000000400b44 <+10>:    jne    0x400b49 <foo+15>
   0x0000000000400b46 <+12>:    movaps %xmm0,%xmm1
   0x0000000000400b49 <+15>:    movaps %xmm1,%xmm0
   0x0000000000400b4c <+18>:    retq
   0x0000000000400b4d <+19>:    nop

So any code that wants to conditionally negate floating-point must
special case 0 due to the icc bug. :(

-- 
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature