https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87011
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
CC| |jakub at gcc dot gnu.org
Resolution|--- |INVALID
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
That seems to be intentional and the right thing to me, what GCC 7 did means
unaligned stores.
movq $0, 4(%rdi)
movl $0, 12(%rdi)
movl $3355185, (%rdi)
vs.
pxor %xmm0, %xmm0
movups %xmm0, (%rdi)
movl $3355185, (%rdi)
where the latter, while using unaligned store, is actually at least 64-bit
aligned 128-store. With -O2 -mno-sse it is
movq $0, (%rdi)
movq $0, 8(%rdi)
movl $3355185, (%rdi)
which also looks to be better. With -m32 where &p->a[4] is word-aligned we
still do that, or if you change the testcase to:
struct S { char a[8]; void (*pf)(void); };
void f (struct S *p)
{
__builtin_memset (p, 0, sizeof *p);
__builtin_strcpy (p->a, "1234567");
}
then it is also adjusted.
