Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 11 Mar 2010 20:05:51 +0100
From: "Magnum, P.I." <>
Subject: Re: Extended MPI support

>> replace MPI_UNSIGNED_LONG_LONG with just MPI_UNSIGNED_LONG on line 154 of
>> status.c and all is golden.
> Thanks. MPI_UNSIGNED_LONG_LONG works fine with MPICH2, are you using 
> OpenMPI? I'm not sure you're safe though. I'm a complete noob at 
> datatypes on different platforms, but I expect MPI_UNSIGNED_LONG to be 
> just 32 bit on some, while LONG_LONG should be 64 bit. Maybe on 64-bit 
> systems MPI_UNSIGNED_LONG will always do?

I just uploaded a third version of the patch. It avoids the problem by 
performing the MPI_Reduce() *after* converting the crypt count to cps. 
This way it's safe to use a 32 bit int.

Tested on OpenMPI as well as MPICH2, seems to work fine now.

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.