Bug in MPI_File_read_all MPI_File_write_all soubroutines

[Mochalskyy__Serhiy]

Dear Intel support team,

Maybe I was not clear enough in my previous topic https://software.intel.com/en-us/comment/1907393#comment-1907393 where two different problems were discussed. Here I would like to concentrate on the issue that was not answered in that topic.

According to the MPI 3.0 standard MPI_File_read_all takes two following parameters:

INTEGER, INTENT(IN) :: count ;       TYPE(MPI_Datatype),INTENT(IN) :: datatype

count is 4 bytes integer that can have value of 2147483647. Therefore, It should be possible to Read or write such amount of array elements with the type that I want, which we can provide using the second variable datatype. However, it does not work. For example, if I want to read count=2147483647 elements; datatype=MPI_CHARACTER (1byte) - everything is fine. However, if I would like to read count=2147483647 elements; and datatype=MPI_INTEGER (4 bytes) the subroutine crushes. In the standard it is written that such suboutine should be able to read  count=2147483647 elements of any data type but in Intel Implementation it does not work reducing the number of elements that can be read by a factor of the datatype. For example if I will make direved datatype of size 2147483647 bytes - I will be able only to read count=1 element. Again the standard says that I can read 2147483647 elements of any size  (not only of 1 bytes elements). 

I hope I was clear enough this time. Could you confirm this bug?

Regards, Serhiy

[Mochalskyy__Serhiy]

Dear Intel support team,

I understand that it is summer holiday now and most of people in vocation.

However, could someone say when should I expect the answer, because my work and code development strongly depends on the issue reported above.

Regards,

Dr. Serhiy Mochalskyy

Max Planck Institute, IPP

[James_S]

Hi mochalskyy_serhiy, I would suggest you to submit a ticket via Online Service Center (supporttickets.intel.com/) if you are urgent in this issue, thanks.