I have a specific use case that I want to describe, and I'm brand new to
MPI. It's rather complex, so I want to be careful that I design it so that
there are no race conditions.
Pool A is a buffer (of type 1) handle manager, that feeds buffer handles
into thread set 1, and receives old handles from thread set 2. There is
only one thread running pool A.
Thread set 1 is a set of N threads that fill up buffers of type 1 with
regular data. It pushes full buffers to thread set 2.
Thread set 2 is a set of M threads that compress buffers of type 1 into
data buffers of type 2. Once the buffer of type 1 has be used up, it is
cleared and sent back to Pool A. It pulls fresh data buffers of type 2 from
Pool B, and sends freshly filled compressed buffers to Pool C.
Pool C does something with the buffer, clears it and hands it to Pool B.
Can this be done safely using an implementation like what I see here:
Note 1: -= denotes a 1-to-N connect and =- a N-to-1 connect
RR = Request reply relationship
Questions: does MPI work with intel PIN tools? If so, then how can I apply
it to this? (I need specific modules and some examples, the OpenMPI manual
just appears to be a bunch of references with no textual guides...).