On Jul 28, 2011, at 3:37 PM, Mark F. Adams wrote:

> I could not fine PETSC_HAVE_MPI_EXSCAN in the tags file.
> 
> Am I missing something?

   Why would it ever be in the tags file until someone uses it?   Like I said 
below you need to use it (where needed) and make sure it gets defined by adding 
code to MPI.py

   Barry

> 
> Mark
> 
> On Jul 28, 2011, at 3:31 PM, Barry Smith wrote:
> 
>> 
>> Ideally code like
>> 
>> #if defined(PETSC_HAVE_MPI_EXSCAN)
>>    cool C code that uses MPI_EXSCAN
>> #else 
>>    less cool C code that uses something else in MPI that is in MPI 1
>> #endif
>> 
>> less ideal but ok code
>> 
>> #if defined(PETSC_HAVE_MPI_EXSCAN)
>>    cool C code that uses MPI_EXSCAN
>> #else 
>>    SETERRQ(..., "Sorry but this code requires MPI_EXSCAN that doesn't exist 
>> on your machine's version of MPI, install a MPI2 with PETSc to get this 
>> functionality")
>> #endif
>> 
>> 
>>   Then to make sure PETSC_HAVE_MPI_EXSCAN gets defined when it can add  a 
>> test for MPI_Exscan() to bottom of config/BuildSystem/config/packages/MPI.py 
>> with the other tests for MPI to functions
>> 
>>   PETSc only requires MPI 1.1 support so any 2.0 or later stuff used needs 
>> this kind of protection.
>> 
>>  Barry
>> 
>> 
>> On Jul 28, 2011, at 2:24 PM, Mark F. Adams wrote:
>> 
>>> Dumb question: How do you protect it?
>>> Mark
>>> 
>>> On Jul 28, 2011, at 3:05 PM, Barry Smith wrote:
>>> 
>>>> 
>>>> On Jul 28, 2011, at 11:55 AM, Jose E. Roman wrote:
>>>> 
>>>>> We are getting an undefined reference `MPI_Exscan'. This function is used 
>>>>> in src/mat/impls/nest/matnest.c.
>>>>> If I am not wrong, MPI_Exscan was introduced in MPI-2 but was not 
>>>>> available in the previous standard. Apparently, the machine where we are 
>>>>> compiling is very old and has MPI-1 only.
>>>>> 
>>>>> I guess there should be a PETSC_HAVE_MPI_EXSCAN test in configure.
>>>> 
>>>> Correct, whoever used that MPI 2 function in PETSc is obligated to protect 
>>>> it.
>>>> 
>>>> Barry
>>>> 
>>>> 
>>>>> 
>>>>> Jose
>>>>> 
>>>> 
>>>> 
>>> 
>> 
>> 
> 


Reply via email to