Seen for cris-elf too...

Confirmed on i686-apple-darwin9, array_memcpy_4.f90.003t.original is

MAIN__ ()
{
  struct t d[5];
  struct t s[5];
  static integer(kind=4) options.0[8] = {68, 255, 0, 0, 0, 1, 0, 1};

  _gfortran_set_options (8, (void *) &options.0);
  (void) __builtin_memcpy ((void *) &d, (void *) &s, 60);
}

and does not depend on the optimization level.

At revision 147065 I get:

MAIN__ ()
{
  struct t d[5];
  struct t s[5];
  static integer(kind=4) options.0[8] = {68, 255, 0, 0, 0, 1, 0, 1};

  _gfortran_set_options (8, (void *) &options.0);
  d = VIEW_CONVERT_EXPR<struct t[5]>(s);
}

This test has been introduced by the patch in http://gcc.gnu.org/ml/fortran/2007-01/msg00425.html.
The tests gfortran.dg/array_memcpy_[1-3].f90 use

! { dg-final { scan-tree-dump-times "memcpy" * "original" } }

So a simple fix would be to replace

! { dg-final { scan-tree-dump-times "d = " 1 "original" } }

with

! { dg-final { scan-tree-dump-times "memcpy" 1 "original" } }

but I don't understand why array_memcpy_4.f90 used to give a different construct and this change could mask a more serious problem.

Subject: Re:  [4.5.0 Regression]  Revision 147083 failed
 gfortran.dg/array_memcpy_4.f90

On Mon, 4 May 2009, dominiq at lps dot ens dot fr wrote:

> ------- Comment #4 from dominiq at lps dot ens dot fr  2009-05-04 08:59 -------
> This test has been introduced by the patch in
> http://gcc.gnu.org/ml/fortran/2007-01/msg00425.html.
> The tests gfortran.dg/array_memcpy_[1-3].f90 use
> 
> ! { dg-final { scan-tree-dump-times "memcpy" * "original" } }
> 
> So a simple fix would be to replace
> 
> ! { dg-final { scan-tree-dump-times "d = " 1 "original" } }
> 
> with
> 
> ! { dg-final { scan-tree-dump-times "memcpy" 1 "original" } }
> 
> but I don't understand why array_memcpy_4.f90 used to give a different
> construct and this change could mask a more serious problem.

I will investigate - we really should get the memcpy removed.

Richard.

I have a patch.

Subject: Bug 40015

Author: rguenth
Date: Mon May  4 11:01:59 2009
New Revision: 147094

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=147094
Log:
2009-05-04  Richard Guenther  <rguenther@suse.de>

	PR middle-end/40015
	* builtins.c (fold_builtin_memory_op): Do not decay to element
	type if the size matches the whole array.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/builtins.c

Fixed.