[petsc-dev] ISLocalToGlobalMappingGetInfo

[Stefano Zampini]

> On Aug 9, 2019, at 10:06 AM, Pierre Jolivet via petsc-dev <petsc-dev at mcs.anl.gov> wrote:
> 
> Nevermind, the first component of the arrays are too frequently used in PCBDDC.
> So, I guess, the documentation should be updated by someone who knows what is stored there.
> 

The documentation is wrong. You can find better information on ISLocalToGlobalMappingGetInfo here
https://bitbucket.org/petsc/petsc/src/4dca3327524fa2afc3caa7721b48fed838417d43/src/ksp/pc/impls/is/pcis.h#lines-58 <https://bitbucket.org/petsc/petsc/src/4dca3327524fa2afc3caa7721b48fed838417d43/src/ksp/pc/impls/is/pcis.h#lines-58>


> Also, it’s not clear to me what would happened with a domain decomposition with only non-connected subdomains.
> If there are more than a single subdomain, is nproc = 1 or 0 like when there is a single subdomain? Cf. https://www.mcs.anl.gov/petsc/petsc-current/src/vec/is/utils/isltog.c.html#line1012 <https://www.mcs.anl.gov/petsc/petsc-current/src/vec/is/utils/isltog.c.html#line1012>.
> 

The ISL2G code does not use any information about connectivity. It only returns the sharing information of local indices.

BTW,  I think returning 0 when sizeofcomm is 1 is an inconsistency (it should return 1)

> Thanks anyway,
> Pierre
> 
>> On 9 Aug 2019, at 9:35 AM, Pierre Jolivet via petsc-dev <petsc-dev at mcs.anl.gov <mailto:petsc-dev at mcs.anl.gov>> wrote:
>> 
>> Actually, I don’t think it is a bug, but rather an undocumented feature.
>> Someone clearly knew what they were doing when writing this: https://www.mcs.anl.gov/petsc/petsc-current/src/vec/is/utils/isltog.c.html#line1379 <https://www.mcs.anl.gov/petsc/petsc-current/src/vec/is/utils/isltog.c.html#line1379>
>> Now, the question becomes: is this really needed? Can’t the values associated to "rank" be stripped from the arrays from the get go?
>> I’m guessing this also explains why some loop start at 1 and not 0, e.g., https://www.mcs.anl.gov/petsc/petsc-current/src/vec/is/utils/isltog.c.html#line1583 <https://www.mcs.anl.gov/petsc/petsc-current/src/vec/is/utils/isltog.c.html#line1583>
>> 
>> I’ll try to take a crack at it, anyway, now I know that the information I’m looking for is basically all the arrays shifted by 1 and nproc = nproc - 1.
>> 
>> Thanks,
>> Pierre
>> 
>>> On 9 Aug 2019, at 8:52 AM, Smith, Barry F. <bsmith at mcs.anl.gov <mailto:bsmith at mcs.anl.gov>> wrote:
>>> 
>>> 
>>>  The documents and the code indicate that nproc should be 1 in this case. So there is a bug, if you are unable to determine the cause I can try to look at it but have stones of things to do.
>>> 
>>>  Barry
>>> 
>>> 
>>>> On Aug 9, 2019, at 12:37 AM, Pierre Jolivet <pierre.jolivet at enseeiht.fr <mailto:pierre.jolivet at enseeiht.fr>> wrote:
>>>> 
>>>> Again, with only 2 processes, where nproc is set to 2, I get the following numprocs and indices arrays on rank 0:
>>>> 0	110 111 112 113 114 115 116 117 … 
>>>> 1	110 111 112 113 114 115 116 117 …
>>>> Are these the expected values? If so, what is the true definition of nproc, please?
>>>> 
>>>> Thanks,
>>>> Pierre
>>>> 
>>>>> On 9 Aug 2019, at 3:45 AM, Smith, Barry F. <bsmith at mcs.anl.gov <mailto:bsmith at mcs.anl.gov>> wrote:
>>>>> 
>>>>> 
>>>>> There could be a bug. Perhaps check the entries for that "extra" connection, are they all actual meaningful connections. The code that fills up these data structures is somewhat involved.
>>>>> 
>>>>> Barry
>>>>> 
>>>>> 
>>>>>> On Aug 8, 2019, at 9:39 AM, Pierre Jolivet via petsc-dev <petsc-dev at mcs.anl.gov <mailto:petsc-dev at mcs.anl.gov>> wrote:
>>>>>> 
>>>>>> Hello,
>>>>>> When I use ISLocalToGlobalMappingGetInfo, e.g., in src/ksp/ksp/examples/tutorials/ex71.c ISLocalToGlobalMappingGetInfo(map, &nproc, &procs, &numprocs, &indices), I get the following value for nproc:
>>>>>> - with 1 process: 0
>>>>>> - with 2 processes: 2
>>>>>> In the source code, it says that nproc is the “number of processors that are connected to this one”.
>>>>>> How come this value is equal to the size of the global communicator when using 2 processes?
>>>>>> According to the comment in the source code, shouldn’t the returned value be 1?
>>>>>> 
>>>>>> Thanks,
>>>>>> Pierre
>>>>> 
>>>> 
>>> 
>> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20190809/35ebe349/attachment.html>