Re: [Starpu-devel] StarPU feature suggestion on a collective MPI operation

[Henrik Henriksson <henhe459@student.liu.se>]

I'll try to implement your suggestion to use starpu_mpi_recv_submit
along with the partitioning, that should work out rather well in my use
case. I missed that idea when looking through the API. Having a helper
function would be nice though.


Thanks for your help and sorry for the late reply!

// Henrik Henriksson

On 4/2/19 4:17 PM, Samuel Thibault wrote:
> Samuel Thibault, le mar. 02 avril 2019 15:25:27 +0200, a ecrit:
>>> void starpu_mpi_allgather_copy(starpu_data_handle_t *source_data_handles, 
>>> int handle_count, starpu_data_handle_t destination_handle, MPI_Comm comm);
>>> void starpu_mpi_gather_copy(starpu_data_handle_t *source_data_handles, 
>>> int handle_count, starpu_data_handle_t destination_handle, int root, 
>>> MPI_Comm comm);
>> But then the question is how the data should be combined, which is
>> exactly what data partitioning is about :)
>>
>> What you can do is on the receiving node to register the big vector,
>> partition it into pieces, run starpu_mpi_recv_submit() for each
>> piece (and the send on the sending nodes), and unpartition the big
>> vector. Then you have your whole vector.  This can even be done
>> completely asynchronously with the partitioning planning.
>>
>> Probably it'd be useful if you could contribute an example that we would
>> include in StarPU :)
> Thinking a bit more about it, it could make sense to have some
> starpu_mpi helper which would do all of this for you. Additionally
> to what you have written, you'd have to provide the struct
> starpu_data_filter for the partitioning details.
>
> Samuel