[petsc-users] Question about MatGetRow

Sat Feb 4 12:01:04 CST 2023

> PETSC_OWN_POINTER,

indicates that the IS is to take ownership of the memory in indices, hence you are no longer reasonable for that memory and should not call PetscFree() on it. So the code is running correctly.

> On Feb 4, 2023, at 3:24 AM, 김성익 <ksi2443 at gmail.com> wrote:
> 
> Following your comments, that works in multi processes.
> After extracting procedure from 'newmat' which is submatrix.
> 
> PetscInt *indices;
> PetscMalloc1(1, &indices);
> Indices[0] = 0;
> ISCreateGenreral(PETSC_COMM_WORLD, 1, indices, PETSC_OWN_POINTER, &isrow);
> MatCreateSubMatrix(mat,isrow,NULL,MAT_INITIAL_MATRIX,&newmat);
> (extract from newmat)
> 
> I did 'ISDestroy(&isrow); PetscFree(indices);'
> However I got an error 'double free'.
> So I deleted PetscFree. 
> Is this correct response of that error?
> If not, how should I deal with that error??
> 
> Thanks,
> Hyung Kim 
> 
> 
> 
> 
> 
> 
> 
> 2023년 2월 3일 (금) 오후 11:33, Matthew Knepley <knepley at gmail.com <mailto:knepley at gmail.com>>님이 작성:
>> On Fri, Feb 3, 2023 at 9:04 AM 김성익 <ksi2443 at gmail.com <mailto:ksi2443 at gmail.com>> wrote:
>>> Actually in the last mail, below scripts are running in all processes.
>>> 
>>> IS isrow;
>>> PetscInt *indices;
>>> PetscMalloc1(1, &indices);
>>> Indices[0] = 0;
>>> ISCreateGenreral(PETSC_COMM_WORLD, 1, indices, PETSC_OWN_POINTER, &isrow);
>>> MatCreateSubMatrix(mat,isrow,NULL,MAT_INITIAL_MATRIX,&newmat);
>>> (extract from newmat)
>>> 
>>> However, you said it cannot get the values of first row of global matrix.
>>> Please let me know how can I fix this scripts for getting the 1st row of global matrix in all processes.
>> 
>> Did you run and see what you get? If it is on all processes, it should work.
>> 
>>   Thanks,
>> 
>>      Matt
>>  
>>> Hyung Kim
>>> 
>>> 
>>> 
>>> 
>>> 
>>> 
>>> 
>>> 2023년 2월 3일 (금) 오후 10:54, Matthew Knepley <knepley at gmail.com <mailto:knepley at gmail.com>>님이 작성:
>>>> On Fri, Feb 3, 2023 at 8:52 AM 김성익 <ksi2443 at gmail.com <mailto:ksi2443 at gmail.com>> wrote:
>>>>> I want to extract same row values of global matrix in all processes.
>>>>> Then how can I do this??
>>>> 
>>>> Create the same IS on each process.
>>>> 
>>>>   THanks,
>>>> 
>>>>     Matt
>>>>  
>>>>> The case of same problem of vector, I just use vecscattertoall.
>>>>> However, I can't find same function for matrix.
>>>>> 
>>>>> Hyung Kim
>>>>> 
>>>>> 2023년 2월 3일 (금) 오후 10:47, Matthew Knepley <knepley at gmail.com <mailto:knepley at gmail.com>>님이 작성:
>>>>>> On Fri, Feb 3, 2023 at 8:45 AM 김성익 <ksi2443 at gmail.com <mailto:ksi2443 at gmail.com>> wrote:
>>>>>>> Following your comments,
>>>>>>> If I extract first row of below matrix.
>>>>>>> <image.png>
>>>>>>> IS isrow;
>>>>>>> PetscInt *indices;
>>>>>>> PetscMalloc1(1, *indices);
>>>>>> 
>>>>>> That should be &indices.
>>>>>>  
>>>>>>> Indices[0] = 0;
>>>>>>> ISCreateGenreral(PETSC_COMM_WORLD, 1, indices, PETSC_COPY_VALUES, &isrow);
>>>>>> 
>>>>>> You should use PETSC_OWN_POINTER.
>>>>>>  
>>>>>>> MatCreateSubMatrix(mat,isrow,NULL,MAT_INITIAL_MATRIX,&newmat);
>>>>>>> 
>>>>>>> Then can I get the array about first row of global matrix in all processes?
>>>>>> 
>>>>>> No, just on the process which gives 0. If you do that on every process, every rank with get row 0.
>>>>>> 
>>>>>>   Thanks,
>>>>>> 
>>>>>>      Matt
>>>>>>  
>>>>>>> Hyung Kim
>>>>>>> 
>>>>>>> 2023년 2월 3일 (금) 오후 10:26, Matthew Knepley <knepley at gmail.com <mailto:knepley at gmail.com>>님이 작성:
>>>>>>>> On Fri, Feb 3, 2023 at 8:06 AM 김성익 <ksi2443 at gmail.com <mailto:ksi2443 at gmail.com>> wrote:
>>>>>>>>> Following your comments,
>>>>>>>>> I want to check below things.
>>>>>>>>> For example, the global dense matrix are as below.
>>>>>>>>> <image.png>
>>>>>>>>> If I want to get first row ('1 2 0 0 3 0 0 4') in Proc 1.
>>>>>>>>> Then I should put 'MatCreateSubMatrix <https://petsc.org/main/docs/manualpages/Mat/MatCreateSubMatrix/>(mat, isrow, NULL, MAT_INITIAL_MATRIX, &newmat)'
>>>>>>>>> and isrow will be [0 1 2 3 4 5 6 7].
>>>>>>>>> 
>>>>>>>>> In this case, How can I make isrow?
>>>>>>>>> Actually I can't understand the procedure of handling isrow.
>>>>>>>> 
>>>>>>>> You create an IS object of type ISGENERAL and give it the array of global indices that you want to extract.
>>>>>>>> 
>>>>>>>>   Thanks,
>>>>>>>> 
>>>>>>>>      Matt
>>>>>>>>  
>>>>>>>>> Hyung Kim
>>>>>>>>> 
>>>>>>>>> 2023년 2월 3일 (금) 오후 9:03, Mark Adams <mfadams at lbl.gov <mailto:mfadams at lbl.gov>>님이 작성:
>>>>>>>>>> https://petsc.org/main/docs/manualpages/Mat/MatCreateSubMatrix/
>>>>>>>>>> 
>>>>>>>>>> Note, PETSc lets you give NULL arguments if there is a reasonable default.
>>>>>>>>>> In this case give NULL for the column IS and you will get the whole columns.
>>>>>>>>>> 
>>>>>>>>>> Mark
>>>>>>>>>> 
>>>>>>>>>> On Fri, Feb 3, 2023 at 4:05 AM 김성익 <ksi2443 at gmail.com <mailto:ksi2443 at gmail.com>> wrote:
>>>>>>>>>>> Hello,
>>>>>>>>>>> 
>>>>>>>>>>> 
>>>>>>>>>>> By using MatGetRow, user can get vectors from local matrix (at each process).
>>>>>>>>>>> 
>>>>>>>>>>> However, I need other process's row values.
>>>>>>>>>>> So I have 2 questions.
>>>>>>>>>>> 
>>>>>>>>>>> 1. Is there any function for getting arrays from other process's??
>>>>>>>>>>> 
>>>>>>>>>>> 2. Or is there any function like matrix version of vecscattertoall??
>>>>>>>>>>> 
>>>>>>>>>>> Thanks,
>>>>>>>>>>> Hyung Kim
>>>>>>>> 
>>>>>>>> 
>>>>>>>> -- 
>>>>>>>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
>>>>>>>> -- Norbert Wiener
>>>>>>>> 
>>>>>>>> https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
>>>>>> 
>>>>>> 
>>>>>> -- 
>>>>>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
>>>>>> -- Norbert Wiener
>>>>>> 
>>>>>> https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
>>>> 
>>>> 
>>>> -- 
>>>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
>>>> -- Norbert Wiener
>>>> 
>>>> https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
>> 
>> 
>> -- 
>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
>> -- Norbert Wiener
>> 
>> https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20230204/e18cd78e/attachment.html>