heat-hybrid

Heat equation solver with hybrid MPI + OpenMP

In this exercise, you should parallelize the two dimensional heat equation with hybrid MPI + OpenMP parallelization. Starting point is a working MPI code parallelized over columns (in Fortran) or rows (in C/C++). See Code description for some theory and more details about the code.

Fine-grained OpenMP parallelization

Here, parallel regions are only within the computational routines and all the MPI communication is handled by the master process. Your tasks are thus:

1. Initialize MPI with appropriate thread safety level
2. Determine the number of threads in the main routine (cpp/main.cpp or fortran/main.F90)
3. Parallelize the generation of initial temperature in the routine generate_field() (in fortran/setup.F90) or in the generate() method (in cpp/heat.cpp
4. Parallelize the main computational routine evolve() in cpp/core.cpp or fortran/core.F90

Coarse-grained OpenMP parallelization

Now, there is only one parallel region (within the main routine), however, MPI communication in the halo exchange is still done only by one thread (but not necessarily master).

1. Initialize MPI with appropriate thread safety level
2. Insert apprioriate OpenMP directives thoughtout the code

To build the code, please use the provided Makefile (by typing make). By default, GNU compiler is used, in order to use Intel compiler type make COMP=intel.