Practical work in mixing MPI and GPU programming
Contents: Adress multiple GPUs with your own code
Tasks:
- Take the code from skalar
and run it
- Take your own OpenACC code and do a similar MPI parallelization
Material:
Dense LinAlg libraries for multi/many core (overview):
- BLAS: the
classical LinAlg package
- ATLAS: shm
parallelization
- Intel-MKL:
shm parallelization
- BLACS
(incl. in ubuntu repository) and PBLAS
for distributed memory
- Lapack: the classical
solver package
- ScaLapack:
distributed memory parallelization
- Intel-MKL:
shm parallelization
- PLASMA:
shm parallelization
- MAGMA:
(multiple) GPU parallelization [also Intel Xeon Phi]
- Paralution (GPU, Xeon Phi)
- PaStiX
.