[gSoc] [graphite] general plan for Automatic parallelization in Graphite

[gSoc] [graphite] general plan for Automatic parallelization in Graphite

[Li Feng]

Hi,

It's nice that the proposal 'Automatic parallelization in Graphite'
is accepted. Which means I will be working with great Graphtie
developers this summer, and trying to implement the project .

I have set up a blog for this project, which will mainly about this
project: 1. plans 2. what I have done 3. related Graphite internals
You can subscribe to it if you like:
http://summergraphite.blogspot.com/

Here is a general plan for this project, keep you in loop,
and feel free to comment :)

 1. Mark the innermost loop parallel  [done]

 2. Try to schedule autopar pass after Graphite, and enable
     code generation if flag_graphite_force_parallel is set
         - There should be some discussion with Razya about
           her plan about the autopar part
         - But before that, I'll try to schedule autopar first

 3. I may try to write testcases for the loops that should be
     parallel, from simple to hard, and check autopar's code
     generation part, make sure this works correctly as we
     expected.
         - The testcases is important. There should be some
           detailed discussion maybe with Sebastian and Konrad.
           To see what kind of loop we can/decide to handle.
         - Check autopar's code generation with
           flag_graphite_force_parallel set with these testcases,
           report bugs if it goes wrong.

 4. Try to write code for deciding if a loop can be parallel
     with data dependency test under this polyhedral model.
        - Try to understand the interface of data dependency test
        - Write code, if data dependency success, mark the loop parallel

Cheers,
Li

Re: [gSoc] [graphite] general plan for Automatic parallelization in Graphite

[Tobias Grosser]

Hi Alex,

On Wed, 2009-04-22 at 08:19 -0700, Alex Turjan wrote:
> Are there any plans to move the partial unrolling phase from
> RTL to Tree-SSA?
> The move would benefit from better (or easier to implement) Tree-SSA alias analysis. 

We thought about loop unrolling in graphite.
It seems to be an interesting project and definitely possible in
general.
But at the moment I do not know of anybody who is working on this. So it
is an open project. If you want to work on this we will definitely
support you. ;-)

Tobi

> 
> Alex
> > 
> > 
> > 
> > 
> > --- On Wed, 4/22/09, Li Feng <nemokingdom@gmail.com>
> > wrote:
> > 
> > > From: Li Feng <nemokingdom@gmail.com>
> > > Subject: [gSoc] [graphite] general plan for Automatic
> > parallelization in   Graphite
> > > To: "GCC" <gcc@gcc.gnu.org>
> > > Cc: "Tobias Grosser"
> > <tobi-grosser@web.de>, "Sebastian"
> > <sebpop@gmail.com>, "Razya Ladelsky"
> > <RAZYA@il.ibm.com>, konrad.trifunovic@gmail.com
> > > Date: Wednesday, April 22, 2009, 5:10 PM
> > > Hi,
> > > 
> > > It's nice that the proposal 'Automatic
> > > parallelization in Graphite'
> > > is accepted. Which means I will be working with great
> > > Graphtie
> > > developers this summer, and trying to implement the
> > project
> > > .
> > > 
> > > I have set up a blog for this project, which will
> > mainly
> > > about this
> > > project: 1. plans 2. what I have done 3. related
> > Graphite
> > > internals
> > > You can subscribe to it if you like:
> > > http://summergraphite.blogspot.com/
> > > 
> > > Here is a general plan for this project, keep you in
> > loop,
> > > and feel free to comment :)
> > > 
> > >  1. Mark the innermost loop parallel  [done]
> > > 
> > >  2. Try to schedule autopar pass after Graphite, and
> > enable
> > >      code generation if flag_graphite_force_parallel
> > is set
> > >          - There should be some discussion with Razya
> > about
> > >            her plan about the autopar part
> > >          - But before that, I'll try to schedule
> > > autopar first
> > > 
> > >  3. I may try to write testcases for the loops that
> > should
> > > be
> > >      parallel, from simple to hard, and check
> > autopar's
> > > code
> > >      generation part, make sure this works correctly
> > as we
> > >      expected.
> > >          - The testcases is important. There should be
> > some
> > >            detailed discussion maybe with Sebastian
> > and
> > > Konrad.
> > >            To see what kind of loop we can/decide to
> > > handle.
> > >          - Check autopar's code generation with
> > >            flag_graphite_force_parallel set with these
> > > testcases,
> > >            report bugs if it goes wrong.
> > > 
> > >  4. Try to write code for deciding if a loop can be
> > > parallel
> > >      with data dependency test under this polyhedral
> > model.
> > >         - Try to understand the interface of data
> > > dependency test
> > >         - Write code, if data dependency success, mark
> > the
> > > loop parallel
> > > 
> > > Cheers,
> > > Li
> 
> 
>       

Re: [gSoc] [graphite] general plan for Automatic parallelization in Graphite

[Alex Turjan]

Are there any plans to move the partial unrolling phase from
RTL to Tree-SSA?
The move would benefit from better (or easier to implement) Tree-SSA alias analysis. 

Alex
> 
> 
> 
> 
> --- On Wed, 4/22/09, Li Feng <nemokingdom@gmail.com>
> wrote:
> 
> > From: Li Feng <nemokingdom@gmail.com>
> > Subject: [gSoc] [graphite] general plan for Automatic
> parallelization in   Graphite
> > To: "GCC" <gcc@gcc.gnu.org>
> > Cc: "Tobias Grosser"
> <tobi-grosser@web.de>, "Sebastian"
> <sebpop@gmail.com>, "Razya Ladelsky"
> <RAZYA@il.ibm.com>, konrad.trifunovic@gmail.com
> > Date: Wednesday, April 22, 2009, 5:10 PM
> > Hi,
> > 
> > It's nice that the proposal 'Automatic
> > parallelization in Graphite'
> > is accepted. Which means I will be working with great
> > Graphtie
> > developers this summer, and trying to implement the
> project
> > .
> > 
> > I have set up a blog for this project, which will
> mainly
> > about this
> > project: 1. plans 2. what I have done 3. related
> Graphite
> > internals
> > You can subscribe to it if you like:
> > http://summergraphite.blogspot.com/
> > 
> > Here is a general plan for this project, keep you in
> loop,
> > and feel free to comment :)
> > 
> >  1. Mark the innermost loop parallel  [done]
> > 
> >  2. Try to schedule autopar pass after Graphite, and
> enable
> >      code generation if flag_graphite_force_parallel
> is set
> >          - There should be some discussion with Razya
> about
> >            her plan about the autopar part
> >          - But before that, I'll try to schedule
> > autopar first
> > 
> >  3. I may try to write testcases for the loops that
> should
> > be
> >      parallel, from simple to hard, and check
> autopar's
> > code
> >      generation part, make sure this works correctly
> as we
> >      expected.
> >          - The testcases is important. There should be
> some
> >            detailed discussion maybe with Sebastian
> and
> > Konrad.
> >            To see what kind of loop we can/decide to
> > handle.
> >          - Check autopar's code generation with
> >            flag_graphite_force_parallel set with these
> > testcases,
> >            report bugs if it goes wrong.
> > 
> >  4. Try to write code for deciding if a loop can be
> > parallel
> >      with data dependency test under this polyhedral
> model.
> >         - Try to understand the interface of data
> > dependency test
> >         - Write code, if data dependency success, mark
> the
> > loop parallel
> > 
> > Cheers,
> > Li