https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113243
Bug ID: 113243
Summary: mips: Wrong code for pr91323.c
Product: gcc
Version: 14.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: xry111 at gcc dot gnu.org
Target Milestone: ---
In pr91323.c:
int
__attribute__ ((noinline, noclone))
f3 (float a, float b)
{
return a < b || a > b;
}
is compiled to:
test(float, float):
c.ueq.s $fcc5,$f12,$f14
li $2,1
jr $31
movt $2,$0,$fcc5
for mips64r2, and
test(float, float):
cmp.ne.s $f12,$f12,$f14
bc1nez $f12,$L4
li $2,1
move $2,$0
$L4:
jrc $31
for mips64r6. Both are incorrect because both cmp.ne.s and c.ueq.s do not
raise INVALID exception with qNaN inputs, but in C "<" and ">" should raise
INVALID for qNaN inputs.