[mpich-discuss] MPI_Get error with multiple threads on two nodes
xinzhao3 at illinois.edu
Wed Jul 30 11:10:11 CDT 2014
I looked through Sangmin's attached code and I think it is correct. If it does not work with current mpich RMA code, we should create a ticket for it now. When we replace the current code with the new RMA code, we should run this test again at that time.
From: Seo, Sangmin [sseo at anl.gov]
Sent: Wednesday, July 30, 2014 10:50 AM
To: Zhao, Xin
Subject: Fwd: [mpich-discuss] MPI_Get error with multiple threads on two nodes
Begin forwarded message:
From: Rob Latham <robl at mcs.anl.gov<mailto:robl at mcs.anl.gov>>
Subject: Re: [mpich-discuss] MPI_Get error with multiple threads on two nodes
Date: July 30, 2014 at 10:12:38 AM CDT
To: <discuss at mpich.org<mailto:discuss at mpich.org>>
Reply-To: <discuss at mpich.org<mailto:discuss at mpich.org>>
On 07/28/2014 09:28 AM, Balaji, Pavan wrote:
It seems to work fine with the mpich-dev/new-op-rma branch, which will replace the mpich/master code soon.
To be clear, it replaces mpich/master code *only for ch3*
I haven’t tested it with mpich/master.
Can confirm the RMA code Sangmin sent blows up on Blue Gene, but in a different way:
*** glibc detected *** /gpfs/mira-home/robl/src/rma_get_pthread/./rma_get_pthread: free():
corrupted unsorted chunks: 0x0000001e8349b560 ***
Sangmin's original question remains unanswered: is the test correct? new-op-rma seems to think so, but old ch3 and pamid do not.
Mathematics and Computer Science Division
Argonne National Lab, IL USA
discuss mailing list discuss at mpich.org<mailto:discuss at mpich.org>
To manage subscription options or unsubscribe:
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the discuss