[EMAIL PROTECTED] gcc]$ cat /tmp/push-1.c
/* { dg-do compile { target { { i?86-*-* x86_64-*-* } && ilp32 } } } */
/* { dg-options "-w -msse2 -Os" } */
typedef float __m128 __attribute__ ((__vector_size__ (16), __may_alias__));
extern void foo (__m128 x, __m128 y ,__m128 z ,__m128 a, int size);
void
bar (void)
{
__m128 x = { 1.0 };
foo (x, x, x, x, 5);
}
/* { dg-final { scan-assembler-not "movups" } } */
[EMAIL PROTECTED] gcc]$ ./xgcc -B./ -Os -m32 -msse2 -S /tmp/push-1.c
[EMAIL PROTECTED] gcc]$ cat push-1.s
.file "push-1.c"
.text
.globl bar
.type bar, @function
bar:
pushl %ebp
movl %esp, %ebp
subl $20, %esp
movss .LC0, %xmm0
pushl $5
subl $16, %esp
movups %xmm0, (%esp)
movaps %xmm0, %xmm2
movaps %xmm0, %xmm1
call foo
addl $32, %esp
leave
ret
-mno-accumulate-outgoing-args will push argument onto stack.
Since there is no push insn for __m128, ix86_expand_push is
called to emulate push __m128. But it doesn't set proper
alignment. As the result, unaligned move insn is generated
even though memory is aligned. Since ix86_expand_push is
used to push a value onto an aligned location on the stack,
the alignment of destination should be at least aligned to
function argument boundary, which is the minimum alignment
of push destination.
--
Summary: Unaligned move used on aligned destination of push insn
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: hjl dot tools at gmail dot com
GCC target triplet: i686-pc-linux-gnu
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37049
