This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [trans-mem] fix memory optimization bug with loops


On Mon, Oct 12, 2009 at 11:52 AM, Aldy Hernandez <aldyh@redhat.com> wrote:
> Dan keeps harassing me over not initializing the sets to all 1's, which
> would probably require major surgery.

I'm just gently trying to guide you away from a path we've gone down
before and discovered monsters in ;)

> ?How about this...
>
> If we never seed the OUT sets with an uninitialized block, we basically
> get an all 1's scenario. ?I've added a VISITED bit and keep it updated
> throughout. ?(I'm using the same bit for both the AVAIL and ANTIC sets.
> I could use two bits and avoid calling tm_memopt_clear_visited() at all
> if preferred.)
It's only equivalent if you make one extra addition to your logic:
Never intersect with non-visited blocks either (IE they are simply
skipped while intersecting).
This is how PRE does it (see compute_antic_aux).


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]