test release: parallel-netcdf 1.3.0pre1

[Wei-keng Liao]

In that case, the nonblocking APIs are still available but
they do not work as intended. That is, you can still call
the APIs, but they will write the requests independently
and thus will not give you any better performance.

So, the same program calling iput/iget should be able to
compile and run on BGL.

Wei-keng


On Apr 23, 2012, at 4:42 PM, Rob Latham wrote:

> On Mon, Apr 23, 2012 at 03:38:06PM -0600, Jim Edwards wrote:
>> On Mon, Apr 23, 2012 at 3:30 PM, Rob Latham <robl at mcs.anl.gov> wrote:
>> 
>>> On Mon, Apr 23, 2012 at 03:07:33PM -0600, Jim Edwards wrote:
>>>> Oh - and just to confirm - the async capabilities of pnetcdf (eg
>>>> nfmpi_iput_ ) are not supported
>>>> on BGL and BGP - how about BGQ?
>>> 
>>> Sure, let's clarify a bit.  The interface should be supported: IBM's
>>> MPI supports the necessary MPI-2 features.   The API allows you to
>>> describe multiple operations, then when you wait for completion those
>>> operations will get stitched together into one bigger more efficient
>>> request.
>>> 
>>> I do mean non-blocking and when I build on frost I get the following
>> message:
>> 
>> checking if MPI_Offset is size 8... yes
>> checking if MPI_Aint is size 8... no
>> checking if MPI_Aint is size 4... yes
>> checking for MPI_Info_dup... yes
>> checking for MPI_Info_free... yes
>> checking for MPI_Get_address... yes
>> checking for MPI_Request_get_status... yes
>> checking for MPI_Type_dup... yes
>> configure: WARNING: MPI_Offset and MPI_Aint have different sizes: unable to
>> support non-blocking routines
> 
> Oh, crud.  
> 
> OK, thanks for reminding me about that.  Yes, on BlueGene/L, the MPI
> library has a 4 byte MPI_Aint and an 8 byte MPI_Offset.  We got IBM to
> implement 8 byte MPI_Aint types on BlueGene/P, but clearly that never
> made its way to BlueGene/L's version.
> 
> The iput/iget routines do work on BlueGene/P.
> 
> ==rob
> 
> -- 
> Rob Latham
> Mathematics and Computer Science Division
> Argonne National Lab, IL USA