Coordination Across Programming Models (OpenMP/MPI) #17
Conversation
| /* register a handler specifically for notifying us if | ||
| * other programming models declare themselves */ | ||
| active = -1; | ||
| PMIx_Register_event_handler(&code, 1, NULL, 0, |
There was a problem hiding this comment.
Is it possible to register muliple event handler for the same event?
There was a problem hiding this comment.
Not at the moment - however, that is being changed per the following RFC: #18
There was a problem hiding this comment.
we might need to have a more flexible way to control when a programming model's callback function is invoked. Looking at section 3.2 of the OpenMP spec there are a number of calls that may lead to the OpenMP runtime adjusting resource requirements etc. that other program models might need to take in to account. I'm thinking, for example, of omp_set_nested.
|
It does look like one of the DOE mini-apps - snap would be a good guinea pig for testing the usefulness of this RFC. It can use nested OpenMP, and also has ports to kokkos - although its unclear if this works. |
…ary integration Refs pmix/RFCs#17 Refs open-mpi/ompi#3448 Signed-off-by: Ralph Castain <rhc@open-mpi.org>
…ary integration Refs pmix/RFCs#17 Refs open-mpi/ompi#3448 Signed-off-by: Ralph Castain <rhc@open-mpi.org>
Hybrid applications (i.e., applications that utilize more than one programming model, such as an MPI application that also uses OpenMP) are growing in popularity, especially as chips with increasingly large numbers of cores and processors proliferate. This RFC offers a potential solution to the problem by providing a pathway for programming models to coordinate their actions. Signed-off-by: Ralph Castain <rhc@open-mpi.org>
|
This has been "seconded" by ORNL. |
Accepted Signed-off-by: Ralph Castain <rhc@open-mpi.org>
Signed-off-by: Ralph Castain rhc@open-mpi.org