[MPICH] Problem with MPI_TYPE_CREATE_RESIZED

[Trach-Minh Tran]

Hi Pavan,
The error is gone now! Thanks.
-Minh.

On 02/11/2008 09:42 AM, Pavan Balaji wrote:
> 
> Can you try the attached patch and see if the problem still exists? The 
> patch is against mpich2-1.0.6p1:
> 
>  $ cd mpich2-1.0.6p1 && patch -p0 < dataloop.patch
> 
> Thanks.
> 
>  -- Pavan
> 
> On 02/10/2008 11:28 AM, Rajeev Thakur wrote:
>> I can reproduce the problem with Intel 9.1 compilers but not with gcc 
>> (3.4).
>> We will look into it.
>>
>> Rajeev
>>
>>> -----Original Message-----
>>> From: Trach-Minh Tran [mailto:trach-minh.tran at epfl.ch] Sent: Sunday, 
>>> February 10, 2008 11:10 AM
>>> To: Rajeev Thakur
>>> Cc: 'MPICH Discuss'
>>> Subject: Re: [MPICH] Problem with MPI_TYPE_CREATE_RESIZED
>>>
>>> I have rebuild mpich2-1.0.6 with gfortran/gcc/g++ (GCC-4.1.2) and ... 
>>> the error is gone! However with the C version of my example, the same 
>>> error reappears!
>>>
>>> #include "mpi.h"
>>> #include <stdio.h>
>>> MPI_Datatype LINE_TYPE,  LINES_TYPE;
>>> MPI_Aint lb, extent;
>>>
>>> int main(int argc,char *argv[])
>>> {
>>>   MPI_Init(&argc,&argv);
>>>
>>>   MPI_Type_vector(10, 1, 10, MPI_INT, &LINE_TYPE);
>>>   MPI_Type_commit(&LINE_TYPE);
>>>   MPI_Type_get_extent(LINE_TYPE, &lb, &extent);
>>>   printf("LINE_TYPE: lb = %d, extent = %d\n", lb, extent);
>>>
>>>   MPI_Type_get_extent(MPI_INT, &lb, &extent);
>>>   MPI_Type_create_resized(LINE_TYPE, lb, extent, &LINES_TYPE);
>>>   MPI_Type_commit(&LINES_TYPE);
>>>   MPI_Type_get_extent(LINES_TYPE, &lb, &extent);
>>>   printf("LINES_TYPE: lb = %d, extent = %d\n", lb, extent);
>>>
>>>   MPI_Finalize();
>>>   return 0;
>>> }
>>>
>>> $ mpiexec -n 1 ./a.out
>>> LINE_TYPE: lb = 0, extent = 364
>>> Assertion failed in file dataloop.c at line 554: old_loop_sz > 0
>>> [cli_0]: aborting job:
>>> internal ABORT - process 0
>>> rank 0 in job 5  gimel3_45420   caused collective abort of all ranks
>>>   exit status of rank 0: return code 1
>>>
>>>
>>> Best regards,   -Minh.
>>>
>>>
>>> On 02/10/2008 03:28 PM, Rajeev Thakur wrote:
>>>> I can't reproduce the error. What compiler are you using? 
>>> Also, it may
>>>> not matter but it should be integer(kind=MPI_ADDRESS_KIND).
>>>>
>>>> Rajeev
>>>>
>>>>> -----Original Message-----
>>>>> From: owner-mpich-discuss at mcs.anl.gov 
>>>>> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of 
>>> Trach-Minh Tran
>>>>> Sent: Sunday, February 10, 2008 2:16 AM
>>>>> To: MPICH Discuss
>>>>> Subject: [MPICH] Problem with MPI_TYPE_CREATE_RESIZED
>>>>>
>>>>> Hello,
>>>>>
>>>>> Using mpich2-1.0.6p1, MPI_TYPE_CREATE_RESIZED produces an error. 
>>>>> Running the following fortran program aborts with the error:
>>>>>
>>>>> Assertion failed in file dataloop.c at line 554: old_loop_sz > 0
>>>>>
>>>>> With versions prior to 1.0.5, this programs runs without any error.
>>>>>
>>>>> Regards,   -Minh.
>>>>>
>>>>> PROGRAM main
>>>>>   USE mpi
>>>>>   IMPLICIT NONE
>>>>>   INTEGER :: ierr, LINE_TYPE, LINES_TYPE
>>>>>   INTEGER(MPI_ADDRESS_KIND) :: extent, lb !
>>>>>   CALL MPI_INIT(ierr)
>>>>> !
>>>>>   CALL MPI_TYPE_VECTOR(10, 1, 10, MPI_INTEGER, LINE_TYPE, ierr)
>>>>>   CALL MPI_TYPE_COMMIT(LINE_TYPE, ierr) !
>>>>>   CALL MPI_TYPE_GET_EXTENT(MPI_INTEGER, lb, extent, ierr)
>>>>>   CALL MPI_TYPE_CREATE_RESIZED(LINE_TYPE, lb, extent, LINES_TYPE, 
>>>>> ierr)
>>>>>   CALL MPI_TYPE_COMMIT(LINES_TYPE, ierr) !
>>>>>   CALL MPI_TYPE_GET_EXTENT(LINES_TYPE, lb, extent, ierr)
>>>>>   PRINT*, 'LINES_TYPE: lb, extent', lb, extent !
>>>>>   CALL MPI_FINALIZE(ierr)
>>>>> END PROGRAM main
>>>>>
>>>>>
>>> -- 
>>> Dr. Trach-Minh Tran
>>> CRPP/SB/EPFL http://crppwww.epfl.ch
>>> Swiss Federal Institute of Technology
>>> CH-1015 Lausanne
>>> Office: http://plan.epfl.ch/?room=PPB111
>>> Tel: +41 21 693 4569
>>> Fax: +41 21 693 5176
>>>
>>>
>>
> 


-- 
Dr. Trach-Minh Tran
CRPP/SB/EPFL http://crppwww.epfl.ch
Swiss Federal Institute of Technology
CH-1015 Lausanne
Office: http://plan.epfl.ch/?room=PPB111
Tel: +41 21 693 4569
Fax: +41 21 693 5176