PETSc acceleration on novel architectures

[Ahmed El Zein]

On Wed, 2009-04-08 at 10:52 -0500, Matthew Knepley wrote:
> On Tue, Apr 7, 2009 at 9:51 PM, Ahmed El Zein <ahmed at azein.com> wrote:
>         On Tue, 2009-04-07 at 10:39 -0500, Matthew Knepley wrote:
>         > On Tue, Apr 7, 2009 at 9:07 AM, Stephen Ball
>         > <Stephen.R.Ball at awe.co.uk> wrote:
>         >         Hi
>         >
>         >         We are keen to start investigating whether or not
>         PETSc is
>         >         suitable for
>         >         acceleration on novel architectures like GPUs, Cell
>         >         processors, etc.
>         >
>         >         I would very much like to get your opinions on this.
>         >
>         >         Do you think such an endeavour is at all feasible
>         with PETSc?
>         >         If so,
>         >         what areas of PETSc do you think our efforts would
>         best be
>         >         spent?
>         >
>         >         Should we for example focus on matrix operations, or
>         on
>         >         specific
>         >         preconditioners or solvers? Where would be a good
>         place to
>         >         start?
>         >
>         >         Can you suggest some specific routines/functions in
>         PETSc that
>         >         are
>         >         potential candidates for acceleration?
>         >
>         > We are actually already working on this, and I plan on
>         having a
>         > PETSc-GPU
>         > come out at the end of the year.
>         
>         
>         What language are you using? I would have thought that OpenCL
>         would be
>         the best solution. Maybe even rewriting the whole of PETSc in
>         OpenCL,
>         targeting both multicore CPUs and a few novel architectures at
>         the same
>         time. AMD and NVIDIA are both going to support OpenCL for
>         their GPUs and
>         I believe that it will be supported on Intel's larrabee and
>         the Cell
>         processor.
> 
> That is probably a mistake. OpenCL is not mature and only a few
> operations in
> PETSc would really benefit.
I would still be interested in what language you are using for PETSc-GPU
and what PETSc-GPU is?

Ahmed
> 
>   Matt
>  
>         
>         Ahmed
>         
>         
>         > Therefore, I suggest working on PCs that
>         > are specific to your problems. People are already doing good
>         work on
>         > sparse
>         > matrices in general, and solvers will see no speedup at all,
>         since
>         > they are all
>         > logic.
>         >
>         >    Matt
>         >
>         >
>         >         Regards
>         >
>         >         Stephen R. Ball
>         >         Advanced Technologies
>         >         HPC
>         >         DRAS
>         >         Rm: G17
>         >         Bldg: E1.1
>         >         AWE(A)
>         >         Aldermaston
>         >         Reading
>         >         Berkshire
>         >         ENGLAND
>         >         RG7 4PR
>         >         Tel: +44 (0)118 982 4528
>         >         e-mail: stephen.r.ball at awe.co.uk
>         >
>         >         This e-mail and any attachments may contain
>         confidential and
>         >         privileged information. If you are not the intended
>         recipient,
>         >         please notify the sender immediately by return
>         e-mail, delete
>         >         this
>         >         e-mail and destroy any copies. Any dissemination or
>         use of
>         >         this
>         >         information by a person other than the intended
>         recipient is
>         >         unauthorized and may be illegal.
>         >
>         >
>         >
>         > --
>         > What most experimenters take for granted before they begin
>         their
>         > experiments is infinitely more interesting than any results
>         to which
>         > their experiments lead.
>         > -- Norbert Wiener
>         
>         
> 
> 
> 
> -- 
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which
> their experiments lead.
> -- Norbert Wiener