[mpich2-dev] regression in datatype decoding

[Dave Goodell]

This is now fixed in the trunk, and should make it into the 1.5rc3 release that I'll likely cut later today.  Thanks again for the report and reproducer.

https://trac.mcs.anl.gov/projects/mpich2/changeset/10311

-Dave

On Oct 2, 2012, at 1:27 PM CDT, Dave Goodell wrote:

> Thanks for letting us know, Lisandro.  I'll take a look at it right now and see if I can find the bug.  Nice catch :)
> 
> -Dave
> 
> On Oct 2, 2012, at 12:41 PM CDT, Lisandro Dalcin wrote:
> 
>> This morning I've svn updated mpich2-dev and rebuilt. I think I've
>> found a regression running mpi4py testsuite (I believe last time I
>> tried all my tests passed, but I could be confused).
>> 
>> To help you debug the issue, see the C program at the end. On decoding
>> a type created with MPI_Type_create_hindexed_block(), I expect to get
>> displacements in the array of addresses, however I'm getting the
>> displacements in the array of integers. This is my ouput for the C
>> code below:
>> 
>> $ mpicc decode.c
>> $ ./a.out
>> integers:  4 2 0 8 16 24
>> addresses:
>> datatypes: MPI_INT,
>> 
>> This smells as copy&paste oversight from MPI_Type_create_indexed_block().
>> 
>> #include <mpi.h>
>> #include <stdlib.h>
>> #include <stdio.h>
>> 
>> int main(int argc, char* argv[])
>> {
>> MPI_Datatype t;
>> int count = 4;
>> int blocklength = 2;
>> MPI_Aint displacements[] ={0, 8, 16, 24};
>> 
>> MPI_Init(0,0);
>> 
>> MPIX_Type_create_hindexed_block(count, blocklength,
>>                                 displacements, MPI_INT,
>>                                 &t);
>> MPI_Type_commit(&t);
>> {
>>   int ni, na, nd, combiner;
>>   int i[1024];
>>   MPI_Aint a[1024];
>>   MPI_Datatype d[1024];
>>   int k;
>>   MPI_Type_get_envelope(t, &ni, &na, &nd, &combiner);
>>   MPI_Type_get_contents(t, ni, na, nd, i, a, d);
>> 
>>   printf("integers:  ");
>>   for (k=0;k<ni;k++) printf("%d ",i[k]);
>>   printf("\n");
>> 
>>   printf("addresses: ");
>>   for (k=0;k<na;k++) printf("%d ",a[k]);
>>   printf("\n");
>> 
>>   printf("datatypes: ");
>>   for (k=0;k<nd;k++) {
>>     char name[MPI_MAX_OBJECT_NAME]; int n;
>>     MPI_Type_get_name(d[k],name,&n);
>>     printf("%s, ",name);
>>   }
>>   printf("\n");
>> 
>> }
>> 
>> MPI_Type_free(&t);
>> MPI_Finalize();
>> }
>> 
>> 
>> 
>> -- 
>> Lisandro Dalcin
>> ---------------
>> CIMEC (INTEC/CONICET-UNL)
>> Predio CONICET-Santa Fe
>> Colectora RN 168 Km 472, Paraje El Pozo
>> 3000 Santa Fe, Argentina
>> Tel: +54-342-4511594 (ext 1011)
>> Tel/Fax: +54-342-4511169
>