Re: [Ada] Redo handling of consistency check for Optimize_Alignment

[Laurent GUERBY <laurent at guerby dot net>]

On Sun, 2008-04-20 at 12:33 +0200, Eric Botcazou wrote:
> > The attached patch is the Gigi support for Optimize_Alignment applied to
> > stand-alone objects: when not optimizing alignment for space, the alignment
> > of some of these objects is promoted to the smallest power of 2 greater (or
> > equal) than their size.
> 
> The attached patch extends this to constant objects as well, minus exceptions.
> 
> Tested on i586-suse-linux, applied on the mainline.  This introduces an ACATS 
> failure at -O (cc50a01) which I think is a latent problem with the optimizer.

FWIW in my records I have about this ACATS:

25738: (run) cc50a01 fails with struct aliasing enabled

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25738

<<
Comment #1 From Richard Guenther 2007-10-11 12:20 [reply]
With

2007-06-21  Eric Botcazou  <ebotcazou@adacore.com>

        PR tree-optimization/25737
        * misc.c (gnat_post_options): Do not force flag_tree_salias to 0.

this has to be no longer true.  Indeed, the test now passes on x86_64.
>>

Laurent