Loop dependence analysis!!

Loop dependence analysis!!

[Ritu Sabharwal]

   We are using gcc-3.0 for modulo scheduling, the loop dependence 
analysis is done in dependene.c file. This file is not used anywhere in 
the source code but we need to use it to find the distance and direction 
vectors for finding dependencies for finding II(initiation interval).
   We have used the functions from this file but this is not working 
properly, giving segmentation fault. Is something missing in that file?

regards,
Ritu.

Re: Loop dependence analysis!!

[Andreas Jaeger]

Ritu Sabharwal <ritu@csa.iisc.ernet.in> writes:

>    We are using gcc-3.0 for modulo scheduling, the loop dependence 
> analysis is done in dependene.c file. This file is not used anywhere in 
> the source code but we need to use it to find the distance and direction 
> vectors for finding dependencies for finding II(initiation interval).
>    We have used the functions from this file but this is not working 
> properly, giving segmentation fault. Is something missing in that file?

It seems that this file is nowhere used in any GCC version.  Since
it's not used anywhere, I'm not surprised that you get problems using
it.

It was added with this ChangeLog and afterwards nothing happened:
2000-08-22  Stan Cox  <scox@redhat.com>

        * Makefile.in (OBJS): Add dependence.o.
        * dependence.c: New file.

Looking at the gcc-patches archive, the patch contains a follow-on
patch that was never committed:

http://gcc.gnu.org/ml/gcc-patches/2000-07/msg00788.html
http://gcc.gnu.org/ml/gcc-patches/2000-08/msg00117.html
http://gcc.gnu.org/ml/gcc-patches/2000-08/msg00187.html

Stan, what's going on here?  Should we remove dependence.c completly?
Or will the rest get added at some point?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj

Re: Loop dependence analysis!!

[Ritu Sabharwal]

  We have made the changes as given by you in the link but it gives 
seg. fault.Is there a way through which we can get the gcc source code 
after these modifications in which loop dependency analysis is enabled or 
there is any other way through whcih we can fine the direction and 
distance vectors for constructiing data dependence graph.


regards,
Ritu.

 
On Wed, 16 Oct 2002, Andreas Jaeger wrote:

> Ritu Sabharwal <ritu@csa.iisc.ernet.in> writes:
> 
> >    We are using gcc-3.0 for modulo scheduling, the loop dependence 
> > analysis is done in dependene.c file. This file is not used anywhere in 
> > the source code but we need to use it to find the distance and direction 
> > vectors for finding dependencies for finding II(initiation interval).
> >    We have used the functions from this file but this is not working 
> > properly, giving segmentation fault. Is something missing in that file?
> 
> It seems that this file is nowhere used in any GCC version.  Since
> it's not used anywhere, I'm not surprised that you get problems using
> it.
> 
> It was added with this ChangeLog and afterwards nothing happened:
> 2000-08-22  Stan Cox  <scox@redhat.com>
> 
>         * Makefile.in (OBJS): Add dependence.o.
>         * dependence.c: New file.
> 
> Looking at the gcc-patches archive, the patch contains a follow-on
> patch that was never committed:
> 
> http://gcc.gnu.org/ml/gcc-patches/2000-07/msg00788.html
> http://gcc.gnu.org/ml/gcc-patches/2000-08/msg00117.html
> http://gcc.gnu.org/ml/gcc-patches/2000-08/msg00187.html
> 
> Stan, what's going on here?  Should we remove dependence.c completly?
> Or will the rest get added at some point?
> 
> Andreas
> 

Re: Loop dependence analysis!!

[Diego Novillo]

On Wed, 16 Oct 2002, Pop Sébastian wrote:

> Is it possible to keep dependence.c in tree-ssa and free some bits from 
> the mainline?
>
Sure.  I don't see why not.


Diego.

Re: Loop dependence analysis!!

[Pop Sébastian]

On Wed, Oct 16, 2002 at 02:53:47PM -0400, Diego Novillo wrote:
> tree-ssa will eventually use this as well.  We're not doing any
> dependence analysis at the moment, but that is in the medium term
> plan.
> 

Is it possible to keep dependence.c in tree-ssa and free some bits from 
the mainline?

Re: Loop dependence analysis!!

[Diego Novillo]

On Wed, 16 Oct 2002, Andreas Jaeger wrote:

> What kind of infrastructure do you speak about?  The new loop
> optimizer done by Zdenek, etc?   In that case it might make sense to
> add your code to the rtlopt branch...
> 
tree-ssa will eventually use this as well.  We're not doing any
dependence analysis at the moment, but that is in the medium term
plan.


Diego.

Re: Loop dependence analysis!!

[Andreas Jaeger]

scox@redhat.com writes:

>> Or will the rest get added at some point?
>
> The front end didn't have the infrastructure to recognize and analyze
> loops, so it was easier to wait until that was available.  Now that it
> is becoming available it will be possible to hook that in and make it
> useful.  It pretty much implements the GCD test.

What kind of infrastructure do you speak about?  The new loop
optimizer done by Zdenek, etc?   In that case it might make sense to
add your code to the rtlopt branch...

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj

Re: Loop dependence analysis!!

[scox]

> Or will the rest get added at some point?

The front end didn't have the infrastructure to recognize and analyze
loops, so it was easier to wait until that was available.  Now that it
is becoming available it will be possible to hook that in and make it
useful.  It pretty much implements the GCD test.