[prev in list] [next in list] [prev in thread] [next in thread] 

List:       gcc-patches
Subject:    Re: [PATCH][RFC] Optimize type-punning union accesses to VIEW_CONVERT_EXPRs
From:       "Daniel Berlin" <dberlin () dberlin ! org>
Date:       2008-02-29 18:01:37
Message-ID: 4aca3dc20802291001j61ce5331wb777b88173396b7e () mail ! gmail ! com
[Download RAW message or body]

On Fri, Feb 29, 2008 at 5:44 AM, Richard Guenther <rguenther@suse.de> wrote:
> On Thu, 28 Feb 2008, Daniel Berlin wrote:
>
>  > On 2/28/08, Richard Guenther <rguenther@suse.de> wrote:
>  > >
>  > > Does this make sense?
>  > > Richard.
>  >
>  > Looks like the right approach to me.
>  > You may want to go further and eliminate storing of fields entirely,
>  > and just changing the tuples for reference ops to be <access size,
>  > access offset, access type> for everything, not just unions.
>
>  Sure - I'll leave that to pull off after we get the MEM_REF stuff
>  (or decide to not do that).  I am experimenting with re-structuring
>  the fixup by keeping some extra dummy ssa vars with value-numbers
>  (but that seems to complicate things somehow...).
Works for me.

FWIW, I tried to do the same thing with global vars in order to get
them PRE'd (IE represent gvar, statement pairs with dummy vars), and
it turned out pretty ugly, sadly.
>
>  Richard.
>
[prev in list] [next in list] [prev in thread] [next in thread]