[mpich-discuss] Bug: Adio 2GB Read/Write Size Limits

[Rob Latham]

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     discuss at mpich.org
> 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     discuss at mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss