http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47914
Summary: __builtin_isunordered(Inf, NaN) doesn't work on
linux-m68k/aranym
Product: gcc
Version: 4.6.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: mi...@it.uu.se
> cat compare-fp-1.c
/* derived from gcc.c-torture/execute/ieee/compare-fp-1.c */
float pinf;
float NaN;
void iun (void)
{
float x = pinf;
float y = NaN;
if (!__builtin_isunordered (x, y))
__builtin_abort();
}
int main (void)
{
pinf = __builtin_inf ();
NaN = __builtin_nan ("");
iun ();
return 0;
}
> gcc -O1 compare-fp-1.c ; ./a.out
Abort
> gcc -O1 -S compare-fp-1.c ; cat compare-fp-1.s
#NO_APP
.file "compare-fp-1.c"
.text
.align 2
.globl iun
.type iun, @function
iun:
fmove.s pinf,%fp0
fcmp.s NaN,%fp0
fjun .L1
jsr abort
.L1:
rts
.size iun, .-iun
.align 2
.globl main
.type main, @function
main:
move.l #0x7f800000,pinf
move.l #0x7fffffff,NaN
jsr iun
clr.l %d0
rts
.size main, .-main
.comm NaN,4,2
.comm pinf,4,2
.ident "GCC: (GNU) 4.6.0 20110226 (experimental)"
.section .note.GNU-stack,"",@progbits
I'm not an expert on m68k FPU instructions, but this looks correct. The bug
occurs with gcc-4.4/4.5/4.6 at all optimization levels. In the gcc testsuite
it causes gcc.c-torture/execute/ieee/compare-fp-1.c to fail. There are lots of
other FP test failures too, that may or may not be related to this one.
However, I'm running this on aranym (aranym-0.9.10-1.1.x86_64.rpm), so it's
possible that aranym is broken. A gdb session seems to support that:
> gdb ./a.out
...
(gdb) break iun
Breakpoint 1 at 0x800003f8
(gdb) display/x $fpstatus
(gdb) display/i $pc
(gdb) run
Starting program: /home/mikpe/a.out
(no debugging symbols found)
(no debugging symbols found)
Breakpoint 1, 0x800003f8 in iun ()
2: x/i $pc
0x800003f8 <iun>: fmoves 0x80002670 <pinf>,%fp0
1: /x $fpstatus = 0x4000000
(gdb) stepi
0x80000400 in iun ()
2: x/i $pc
0x80000400 <iun+8>: fcmps 0x80002674 <NaN>,%fp0
1: /x $fpstatus = 0x2000000
(gdb)
0x80000408 in iun ()
2: x/i $pc
0x80000408 <iun+16>: fbun 0x80000412 <iun+26>
1: /x $fpstatus = 0x2000000
*** Shouldn't fcmp on a NaN operand set the NAN bit (bit 24) of fpstatus?
(gdb)
0x8000040c in iun ()
2: x/i $pc
0x8000040c <iun+20>: jsr 0x8000030c <abort@plt>
1: /x $fpstatus = 0x0
Could someone please test this on a real m68k with an FPU?
