[petsc-dev] Enabling single precision for MUMPS and SUPERLU

[Barry Smith]

On Sep 24, 2012, at 12:10 PM, Stefano Zampini <stefano.zampini at gmail.com> wrote:

> Hong: thank you for testing them.
> Barry: it's your call to choose namespacing of mumps call.

  Whatever.


> 
> 2012/9/24 Hong Zhang <hzhang at mcs.anl.gov>
> Stefano :
> I tested the patch with
> single and double precision, double with '--with-scalar-type=complex'
> All passed.
> The patches are pushed to petsc-dev
> (http://petsc.cs.iit.edu/petsc/petsc-dev/)
> 
> Thanks for your contribution!
> Hong
> 
> Sorry Hong, I forgot to pull before doing the patches. Please find attached the patches made with the latest dev
> (changeset 24444 92061a43c7a4)
> 
> Stefano
> 
> 
> 
> 
> 2012/9/21 Hong Zhang <hzhang at mcs.anl.gov>
> Stefano:
> Are you sure these patches are made out of the latest petsc-dev?
> I get
> petsc:/sandbox/hzhang/petsc-dev>patch -Np1 < superlu_patch_single.diff
> patching file config/PETSc/packages/SuperLU.py
> patching file src/mat/impls/aij/seq/superlu/makefile
> patching file src/mat/impls/aij/seq/superlu/superlu.c
> Hunk #1 FAILED at 21.
> Hunk #2 FAILED at 142.
> Hunk #4 FAILED at 311.
> Hunk #5 succeeded at 517 with fuzz 2.
> Hunk #6 FAILED at 569.
> Hunk #8 succeeded at 618 with fuzz 1.
> 4 out of 8 hunks FAILED -- saving rejects to file src/mat/impls/aij/seq/superlu/superlu.c.rej
> 
> Similar error on mumps_patch_single.diff.
> 
> Patching petsc-3.3 is successful though. 
> 
> Barry: do you agree with Stefano's approach?
> 
> Hong
> 
> Hong, attached please find the two patches split.  
> 
> 2012/9/21 Hong Zhang <hzhang at mcs.anl.gov>
> Stefano:
> 
> When Hong has finished testing, I will split the patches and commit them. Please let me know what macro name you'll choose (or the ugly choice of course)
>  
> How about split the patches now?
> Then I'll test one package at a time.
> Hong
> 
> There is also pastix which can be enabled, but the problem is that it requires pt-scotch for partitioning, and in turn pt-scotch will become a possible method to be used in MatPartitioning class. Maybe someone can think about adding a new command line argument in package configure which guaranties you will use a library (pt-scotch) only inside  pastix and not within the MatPartitioning class? (maybe using self.useddirectly in config/BuildSystem/package.py)
> 
> --
> Stefano
> 
> 
> 
> 
> -- 
> Stefano
> 
> 
> 
> 
> -- 
> Stefano
> 
> 
> 
> 
> -- 
> Stefano