On 10/29/2014 04:10 AM, Sean Delaney wrote:

Hi,

There seems to be a bug in ‘mpi/romio/adio’, relating to ‘buf_idx[]’ and associated variables, which are 32-bit signed integer arrays (or pointers to suitable allocations). As mentioned in the code comments, this type limits the size of the user’s arrays to 2GB.

Recent fixes in other areas of mpich (e.g. trac issue 2010) have facilitated individual IO calls accessing >2GB of data. (In my case, this involved a large ‘MPI_Type_indexed’ file view.) Unfortunately, those fixes have exposed this limitation in adio.

I briefly tried fixing this, but the changes grew beyond my available time.

Regards, and thanks for the excellent work to date,

this is in the latest 3.1.3 MPICH? I thought I bumped up the size of buf_idx and related fields, though my test cases are meager at best. Do you have a test case? Which file system? ==rob

Sean

______________________________________________________________________ This email has been scanned by the Symantec Email Security.cloud service. For more information please visit http://www.symanteccloud.com ______________________________________________________________________ _______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

-- Rob Latham Mathematics and Computer Science Division Argonne National Lab, IL USA _______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss