Assignment 10
- Ray tracer with parallel I/O. Modify the MPI-parallel ray tracer so that each process writes its finished tiles immediately to a file on the parallel file system, without the master having to receive the data first.
On Emmy, you need the $FASTTMP folder for this to work properly, since this is an MPI-compatible parallel file system. - Hybrid-parallel dMVM. Modify the MPI-parallel (blocking) variant of the dense matrix-vector multiplication code so that each MPI process is multi-threaded. Perform scaling runs up to 80 cores (four Emmy nodes) with problem sizes of 4000x4000 and 10000x10000. Compare with pure MPI: How many threads per process are optimal?
Last modified: Thursday, 21 January 2021, 2:50 PM