On Wed, Jan 13, 2016 at 10:00:08PM -0700, Dave Johansen wrote:
I was working on packaging rr [1] and one of the tests [2] fails to
build
when optimizations are turned on. I've reduced it to the following and
still been able to reproduce the issue:
static const float xmm0 = 10;
int main() {
__asm__ __volatile__(
#if __i386__
"movss xmm0, %xmm0\n\t"
#elif __x86_64__
"movss xmm0(%rip), %xmm0\n\t"
#else
#error unexpected architecture
#endif
);
return 0;
}
Here's the output on F23 x86_64:
$ gcc fxregs.c -O0
$ gcc fxregs.c -O1
/tmp/cccdze3O.o: In function `main':
fxregs.c:(.text+0x4): undefined reference to `xmm0'
collect2: error: ld returned 1 exit status
Is this a gcc bug or is there something that I need to do in the build to
get the tests to build without error?
I think it's a bug in your test. GCC is just optimizing away the xmm0
variable because it cannot see "inside" the asm to know that it is
being used.
This works for me. I believe you don't need the #if conditional
either since this should work on any x86 machine:
__asm__ __volatile__ ("movss %0, %%xmm0\n\t" : : "m" (xmm0) :
"%xmm0");
Rich.
--
Richard Jones, Virtualization Group, Red Hat
http://people.redhat.com/~rjones
Read my programming and virtualization blog:
http://rwmj.wordpress.com
virt-p2v converts physical machines to virtual machines. Boot with a
live CD or over the network (PXE) and turn machines into KVM guests.
http://libguestfs.org/virt-v2v