StarPU documentation for MPI data reduction
Dear StarPU developers and maintainers,
When attempting to develop a distributed data reduction with StarPU, I found it difficult to understand how to do it in the documentation (version 1.3.9, part 7.9).
My principal difficulties were to find out the correct access mode to be given for the reduction handle when calling starpu_mpi_task_insert
(is it STARPU_REDUX
? STARPU_MPI_REDUX
), and also the access mode to be declared for the reduction handle of the reduction codelet.
Also, I was not able to find anything about how to perform an in-place reduction (with or without MPI).
Best regards, Matthieu.