Re: PATCH: PR middle-end/37009: No need to align stack when incomingstack is aligned

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Mon, Aug 04, 2008 at 02:22:28PM +0800, Joey Ye wrote:
> The patch is OK. Suggest to add attached testcase to assert that
> parameter boundary is NOT set to 8 when double is passed.

My patch only records the largest parameter boundary in
parm_stack_boundary. It doesn't sets parameter boundary. My
patch is an optimization. It will remove unnecessary
alignment instructions and won't add new ones.  I am checking it
into stack branch.

BTW, I am not sure what your t.c is testing beyond what we already
have.


H.J.
-----
/* PR middle-end/37009 */
/* { dg-do run { target { { i?86-*-* x86_64-*-* } && ilp32 } } } */
/* { dg-options "-m32 -mincoming-stack-boundary=2 -mpreferred-stack-boundary=2" } */

#include <emmintrin.h>
#include "check.h"

typedef double aligned;

void
bar (char *p, int size)
{
  __builtin_strncpy (p, "good", size);
}

void
__attribute__ ((noinline))
foo (aligned x, aligned y ,aligned z ,aligned a, int size)
{
  char *p = __builtin_alloca (size + 1);
  aligned i;

  bar (p, size);
  if (__builtin_strncmp (p, "good", size) != 0)
    {
#ifdef DEBUG
      p[size] = '\0';
      printf ("Failed: %s != good\n", p);
#endif
     abort ();
    }

  check_int (&i,  __alignof__(i));
}

int
main (void)
{
  aligned x =  1.0 ;
 
  foo (x, x, x, x, 5);

  return 0;
}