[petsc-dev] PETSc GPU capabilities

[Jed Brown]

On Thu, Mar 8, 2012 at 13:26, Paul Mullowney <paulm at txcorp.com> wrote:

> If we choose this option, will
> -mat_type cusp
> allow one to use the CUSPARSE TriSolve for ILU or ICC?
>
> Alternatively, if one uses
> -mat_type cusparse
> for SpMV, will one be able to use the CUSP AMG and Poly preconditioners?
>

The preconditioner should make a copy if it needs to. For factorization,
there can be a -pc_factor_mat_solver_package {cusparse,txpetscgpu} to
select which implementation is used (for any matrix type that is used).


>
>  Since I am new to this, what would PETSc developers like to see?
>

I would have -mat_type cusp and -mat_type cusparse, with both
implementations residing in PETSc. But I don't know where you want to go
with txpetscgpu.


>
> One more wrinkle. Given the naming conventions, should the vector types be
> -vec_type thrust
> instead of
> -vec_type cusp
>
> The vectors only rely on Thrust capabilities at the moment.
>

I think so, yes.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120308/604ea682/attachment.html>