Intel® oneAPI HPC Toolkit
Get help with building, analyzing, optimizing, and scaling high-performance computing (HPC) applications.
This community is designed for sharing of public information. Please do not share Intel or third-party confidential information here.
1919 Discussions

ROMIO MPI_File_set_view with overlapped regions for collective read

Hello all,
I am trying to do a collective read using MPI_File_Read_all by setting the view with MPI_File_set_view. THe etype in read all is MPI_SHORT and the filetype is a derived datatype of type contiguous.

In my setup I have ghost layers (1-D decomposition of domain) and the size of offset (displacement) is smaller than size of the derived datatype(filetype). Essentially the displacement for processor n is within the bounds displacement(n-1) + size of filetype (n-1).

I was under the understanding that this is ok in collective read, however, I notice that I am getting wrong results. (The collective read is reading beyond the point where displacement or file offset is greater than the size of the file.

As a sanity check is my notion correct? Is it possible to have displacement for processor n < size of filetype for processor n-1 ?

Thanks in advance,

0 Kudos
0 Replies