[mvapich-discuss] Datatypes Error

[sreeram potluri]

Hi Jim,

Thanks for the confirmation. Just to let you know, the patch for this issue
is available as part of the recent MVAPICH2 1.6RC2 version.

Sreeram Potluri

On Wed, Dec 22, 2010 at 12:16 AM, James Dinan <dinan at mcs.anl.gov> wrote:

> Hi Sreeram,
>
> This patch appears to correct the problem.
>
>
> Thanks for your help!
>  ~Jim.
>
> On 12/18/2010 07:53 PM, sreeram potluri wrote:
>
>> Hi Jim,
>>
>> Attached is a patch that fixes the failure with your benchmark when
>> using enable-g=all.
>>
>> The problem was with how MPIDI_Pkt_init is used to set the packet type.
>> This function's behavior changes based on whether enagle-g=meminit is
>> used (clears the packet completely before assignment) or not and hence
>> the behavior we had observed.
>>
>> Can you please confirm if this patch fixes the failure? There are few
>> memory leaks that we are fixing as a separate issue. So, please ignore
>> these.
>>
>> Thanks much
>> Sreeram Potluri
>>
>

>                                On Tue, 14 Dec 2010, James Dinan wrote:
>>
>>                                    Hi,
>>
>>                                    I've run into an error using MPI
>>        datatypes
>>                on MVAPICH-2
>>                                    1.6rc1 on our IB
>>                                    cluster.  I've attached a test case
>>        that uses
>>                        datatypes to do a
>>                                    one-sided gather via an MPI_Get with
>>                datatypes.  The
>>                        program
>>                                    builds two
>>                                    datatypes (origin and target) to
>>        transfer a
>>                patch of
>>                        a 2d
>>                                    array into a
>>                                    local buffer.  For transfers smaller
>>        than ~16kB
>>                                    (SUB_XDIM*SUB_YDIM*sizeof(double)<
>>          16384)
>>                the program
>>                                    works, when
>>                                    larger it fails with no error messages.
>>
>>                                    I tried using the datatype only for the
>>                remote data
>>                                    (included in a
>>                                    comment) and saw the same type of
>>        failure.
>>                  I didn't
>>                        see any
>>                                    errors
>>                                    running this test under the latest
>>        version
>>                of MPICH.
>>
>>                                    Thanks for your help,
>>                                       ~Jim.
>>
>>
>>
>>                            _______________________________________________
>>                            mvapich-discuss mailing list
>>        mvapich-discuss at cse.ohio-state.edu
>>        <mailto:mvapich-discuss at cse.ohio-state.edu>
>>        <mailto:mvapich-discuss at cse.ohio-state.edu
>>        <mailto:mvapich-discuss at cse.ohio-state.edu>>
>>        <mailto:mvapich-discuss at cse.ohio-state.edu
>>        <mailto:mvapich-discuss at cse.ohio-state.edu>
>>        <mailto:mvapich-discuss at cse.ohio-state.edu
>>        <mailto:mvapich-discuss at cse.ohio-state.edu>>>
>>        <mailto:mvapich-discuss at cse.ohio-state.edu
>>        <mailto:mvapich-discuss at cse.ohio-state.edu>
>>        <mailto:mvapich-discuss at cse.ohio-state.edu
>>        <mailto:mvapich-discuss at cse.ohio-state.edu>>
>>        <mailto:mvapich-discuss at cse.ohio-state.edu
>>        <mailto:mvapich-discuss at cse.ohio-state.edu>
>>        <mailto:mvapich-discuss at cse.ohio-state.edu
>>        <mailto:mvapich-discuss at cse.ohio-state.edu>>>>
>>
>>        http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>>
>>
>>
>>
>>
>>
>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20101228/1f8e0753/attachment.html