[petsc-dev] petsc4py, TAO, and complex scalars

[Barry Smith]

On Jul 11, 2014, at 1:29 PM, Satish Balay <balay at mcs.anl.gov> wrote:

> Yeah - I had merged this branch to next to test it out.  Lisandro is
> on travel - so might not be checking e-mails.
> 
> Perhaps tao/complex build should be enabled only for c99/complex..

   Are all the failures only for C++ complex?  How easy/hard would they be too fix. In the long run just getting tao complex clean would make our lives simpler.

   Barry

> 
> Satish
> 
> On Fri, 11 Jul 2014, Barry Smith wrote:
> 
>> 
>>   There are a truck load of complex related errors in tao in next: http://ftp.mcs.anl.gov/pub/petsc/nightlylogs/archive/2014/07/10/filtered-make_next_arch-freebsd-cxx-cmplx-pkgs-dbg_wii.log  see also the other complex builds at http://ftp.mcs.anl.gov/pub/petsc/nightlylogs/archive/2014/07/10/next.html
>> 
>> 
>>  not sure if they are related to your change
>> 
>>   Barry
>> 
>> 
>> On Jul 8, 2014, at 8:57 AM, Lisandro Dalcin <dalcinl at gmail.com> wrote:
>> 
>>> On 8 July 2014 15:16, Matthew Knepley <knepley at gmail.com> wrote:
>>>> On Tue, Jul 8, 2014 at 2:09 PM, Lisandro Dalcin <dalcinl at gmail.com> wrote:
>>>>> 
>>>>> Some time ago I merged tao4py in petsc4py following PETSc. Now I
>>>>> realize that TA0 is not available for --with-scalar-type=complex, all
>>>>> of the TAO API is not built. Adding support for this case in petsc4py
>>>>> will be rather involved, so I want to propose an alternative.
>>>>> 
>>>>> I have a rather small patch that enables the compile of interface TAO
>>>>> APIs.  With this patch, I can build petsc4py, but obviously any
>>>>> attempt of use a specific TAO solver type fails at runtime because the
>>>>> solve implementations are not registered. But at least, petsc4py can
>>>>> be built with complex scalars and all other PETSc components can be
>>>>> used normally.
>>>>> 
>>>>> What do you think? If you accept this proposal, I'll put all this in a
>>>>> branch for review. Additionally, I would like to merge this fix into
>>>>> maint.
>>>> 
>>>> 
>>>> That sounds fine to me. We will fix this for the next release, so it would
>>>> just hang around until then.
>>>> 
>>> 
>>> https://bitbucket.org/petsc/petsc/branch/dalcinl/tao-complex_scalars
>>> 
>>> This hook is a hack, but seems the easiest way to support complex scalars:
>>> https://bitbucket.org/petsc/petsc/commits/a62a097f5bb5c39fb23ca6e1f9f37f7c197bfa6d?at=master#Lsrc/tao/matrix/lmvmmat.cT5
>>> 
>>> Please review,
>>> 
>>> -- 
>>> Lisandro Dalcin
>>> ---------------
>>> CIMEC (UNL/CONICET)
>>> Predio CONICET-Santa Fe
>>> Colectora RN 168 Km 472, Paraje El Pozo
>>> 3000 Santa Fe, Argentina
>>> Tel: +54-342-4511594 (ext 1016)
>>> Tel/Fax: +54-342-4511169
>> 
>> 
>