GitHub - ugovaretto-accel/cuda-mpi-scratch: Scratchpad for testing cuda-mpi interoperability

ugovaretto-accel / cuda-mpi-scratch Public

Notifications You must be signed in to change notification settings
Fork 1
Star 3

Scratchpad for testing cuda-mpi interoperability

Notifications

Name		Name	Last commit message	Last commit date
Latest commit History 49 Commits
stencil2d		stencil2d
test-benchmark		test-benchmark
.gitignore		.gitignore
README		README
cuda_error_handler.h		cuda_error_handler.h
host_allocator.h		host_allocator.h
mpi-complex-types.cpp		mpi-complex-types.cpp
mpi1.cpp		mpi1.cpp
mpi10.cpp		mpi10.cpp
mpi2.cpp		mpi2.cpp
mpi3.cpp		mpi3.cpp
mpi4.cpp		mpi4.cpp
mpi5.cpp		mpi5.cpp
mpi6.cpp		mpi6.cpp
mpi7.cpp		mpi7.cpp
mpi8.cpp		mpi8.cpp
mpi9.cpp		mpi9.cpp
mpi_cuda_pbs_ref.sh		mpi_cuda_pbs_ref.sh
mpi_pbs_sample.sh		mpi_pbs_sample.sh
mpicuda2.cpp		mpicuda2.cpp
mpicuda2.cu		mpicuda2.cu
mpicuda3.cu		mpicuda3.cu
mpicuda4.cu		mpicuda4.cu
mpierr.h		mpierr.h
ref_parallel-dot-product-atomics.cu		ref_parallel-dot-product-atomics.cu

Repository files navigation

CUDA-MPI scratchpad.

Environment:
* mvapich2 >= 1.7 & CRAY MPI
* CUDA >= 3.2
* PBS and SLURM
* Linux clusters and Cray XK6
* Tesla M2090 and X2090

Compilation
-----------

The best option is to use CMake since it works nicely with CUDA and MPI and will automatically have nvcc use the proper mpi compiler wrapper.

Another option is to simply specify all the include and library paths as well as all the libraries on the nvcc command line.