<div class="gmail_extra">Oh - thanks for that clarification.<br><br><div class="gmail_quote">On Mon, Apr 23, 2012 at 3:52 PM, Wei-keng Liao <span dir="ltr"><<a href="mailto:wkliao@ece.northwestern.edu" target="_blank">wkliao@ece.northwestern.edu</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><br>
In that case, the nonblocking APIs are still available but<br>
they do not work as intended. That is, you can still call<br>
the APIs, but they will write the requests independently<br>
and thus will not give you any better performance.<br>
<br>
So, the same program calling iput/iget should be able to<br>
compile and run on BGL.<br>
<span class="HOEnZb"><font color="#888888"><br>
Wei-keng<br>
</font></span><div class="HOEnZb"><div class="h5"><br>
<br>
On Apr 23, 2012, at 4:42 PM, Rob Latham wrote:<br>
<br>
> On Mon, Apr 23, 2012 at 03:38:06PM -0600, Jim Edwards wrote:<br>
>> On Mon, Apr 23, 2012 at 3:30 PM, Rob Latham <<a href="mailto:robl@mcs.anl.gov">robl@mcs.anl.gov</a>> wrote:<br>
>><br>
>>> On Mon, Apr 23, 2012 at 03:07:33PM -0600, Jim Edwards wrote:<br>
>>>> Oh - and just to confirm - the async capabilities of pnetcdf (eg<br>
>>>> nfmpi_iput_ ) are not supported<br>
>>>> on BGL and BGP - how about BGQ?<br>
>>><br>
>>> Sure, let's clarify a bit. The interface should be supported: IBM's<br>
>>> MPI supports the necessary MPI-2 features. The API allows you to<br>
>>> describe multiple operations, then when you wait for completion those<br>
>>> operations will get stitched together into one bigger more efficient<br>
>>> request.<br>
>>><br>
>>> I do mean non-blocking and when I build on frost I get the following<br>
>> message:<br>
>><br>
>> checking if MPI_Offset is size 8... yes<br>
>> checking if MPI_Aint is size 8... no<br>
>> checking if MPI_Aint is size 4... yes<br>
>> checking for MPI_Info_dup... yes<br>
>> checking for MPI_Info_free... yes<br>
>> checking for MPI_Get_address... yes<br>
>> checking for MPI_Request_get_status... yes<br>
>> checking for MPI_Type_dup... yes<br>
>> configure: WARNING: MPI_Offset and MPI_Aint have different sizes: unable to<br>
>> support non-blocking routines<br>
><br>
> Oh, crud.<br>
><br>
> OK, thanks for reminding me about that. Yes, on BlueGene/L, the MPI<br>
> library has a 4 byte MPI_Aint and an 8 byte MPI_Offset. We got IBM to<br>
> implement 8 byte MPI_Aint types on BlueGene/P, but clearly that never<br>
> made its way to BlueGene/L's version.<br>
><br>
> The iput/iget routines do work on BlueGene/P.<br>
><br>
> ==rob<br>
><br>
> --<br>
> Rob Latham<br>
> Mathematics and Computer Science Division<br>
> Argonne National Lab, IL USA<br>
<br>
</div></div></blockquote></div><br><br clear="all"><br>-- <br><pre>Jim Edwards<br><br><br></pre><br>
</div>