http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60496
Bug ID: 60496
Summary: ffreep instruction shouldn't be generated when using
i386 instruction set
Product: gcc
Version: 4.8.2
Status: UNCONFIRMED
Severity: minor
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: mikulas at artax dot karlin.mff.cuni.cz
Host: x86_64-linux-gnux32
Target: x86_64-linux-gnux32
Build: x86_64-linux-gnux32
The ffreep instruction isn't supported on all i386-class processors - it is not
supported on the NexGen processor - see this http://www.pagetable.com/?p=16
Consequently - when generating code with -march=i386, ffreep shouldn't be
generated. ffreep should only be generated when generating for i486 or better
instruction set.
This small piece of code shows incorrect ffreep generation when using i386
instruction set and when optimizing for AMD Athlon. Compile this function with
"-m32 -O2 -march=i386 -mtune=athlon"
double f(double a, double b)
{
double c = a + b;
return c != 1 ? c : 0;
}
The resulting assembler:
f:
.LFB0:
.cfi_startproc
fldl 12(%esp)
faddl 4(%esp)
fld1
fxch %st(1)
fucom %st(1)
fnstsw %ax
fstp %st(1)
andb $69, %ah
xorb $64, %ah
je .L5
rep ret
.p2align 4,,7
.p2align 3
.L5:
ffreep %st(0)
fldz
ret
.cfi_endproc
.LFE0:
.size f, .-f
