p2z-tests

OpenMP

Compilers: gcc, icc, pgi.

Version 3 is most up to date.

$ make COMPILER=gcc MODE=omp
$ make COMPILER=icc MODE=omp
$ make COMPILER=pgi MODE=omp

OpenACC C++

Compile with pgi

Compile the OpenACC C++ sync version

$ make COMPILER=pgi MODE=acc
$ make COMPILER=gcc MODE=acc

OpenACC C

Compile with pgi, openarc

Compile the OpenACC C sync version

$ make COMPILER=pgi MODE=accc
$ make COMPILER=openarc MODE=accc INCLUDE_DATA=0
$ make COMPILER=gcc MODE=accc

Compile the OpenACC C async version (v3 and v4)

$ make COMPILER=pgi MODE=acccv3
$ make COMPILER=openarc MODE=acccv3 INCLUDE_DATA=0
$ make COMPILER=gcc MODE=acccv3
$ make COMPILER=openarc MODE=acccv4 INCLUDE_DATA=0

OpenACC C for CPU

Compile with pgi, openarc

Compile the OpenACC C sync version for CPU

$ make COMPILER=pgi MODE=accccpu
$ make COMPILER=openarc MODE=accccpu
$ make COMPILER=gcc MODE=accccpu

Compile the OpenACC C async version (v3 and v4) for CPU

$ make COMPILER=pgi MODE=acccv3cpu
$ make COMPILER=openarc MODE=acccv3cpu
$ make COMPILER=gcc MODE=acccv3cpu
$ make COMPILER=openarc MODE=acccv4cpu

OpenMP4 C++

Compile with gcc, llvm, ibm

Compile the OpenMP4 C++ sync version

$ make COMPILER=gcc MODE=omp4
$ make COMPILER=llvm MODE=omp4
$ make COMPILER=ibm MODE=omp4
$ make COMPILER=pgi MODE=omp4 #NVHPC V21.11 fails due to an unsupported feature (Standalone 'omp parallel' in a 'declare target' routine is not supported yet). NVHPC V22.2 also fails.

OpenMP4 C

Compile with openarc, llvm, ibm, and gcc

Compile the OpenMP4 C sync version

$ make COMPILER=openarc MODE=omp4c INCLUDE_DATA=0
$ make COMPILER=llvm MODE=omp4c
$ make COMPILER=ibm MODE=omp4c
$ make COMPILER=gcc MODE=omp4c
$ make COMPILER=pgi MODE=omp4c #NVHPC V21.11 fails due to an unsupported feature (Standalone 'omp parallel' in a 'declare target' routine is not supported yet). NVHPC V22.2 also fails.

Compile the OpenMP4 C async version (v3 and v4)

$ make COMPILER=openarc MODE=omp4cv4 INCLUDE_DATA=0
$ make COMPILER=openarc MODE=omp4cv3 INCLUDE_DATA=0
$ make COMPILER=llvm MODE=omp4cv3
$ make COMPILER=ibm MODE=omp4cv3
$ make COMPILER=gcc MODE=omp4cv3
$ make COMPILER=pgi MODE=omp4cv3 #NVHPC V21.11 fails due to an unsupported feature (Standalone 'omp parallel' in a 'declare target' routine is not supported yet). NVHPC V22.2 also fails.

Tbb

Compilers: gcc, icc

$ make COMPILER=gcc MODE=tbb
$ make COMPILER=icc MODE=tbb

CUDA

Compilers: nvcc

Version 1 uses unified memory. Version 2 and 3 use explicit memory transfers.

Version 3 has the same computation and communiation patterns as OpenACC async version (v3).

$ make COMPILER=nvcc MODE=cuda #default behaviors: asynchronous execution; measure both memory transfer times and compute times
$ make COMPILER=nvcc MODE=cuda INCLUDE_DATA=1 #measure both memory transfer times and compute times (default)
$ make COMPILER=nvcc MODE=cuda INCLUDE_DATA=0 #measure compute times only
$ make COMPILER=nvcc MODE=cudav1 USE_ASYNC=1 #for asynchronous execution (default)
$ make COMPILER=nvcc MODE=cudav1 USE_ASYNC=0 #for synchronous execution
$ make COMPILER=nvcc MODE=cudav1 INCLUDE_DATA=1 #measure both memory transfer times and compute times (default)
$ make COMPILER=nvcc MODE=cudav1 INCLUDE_DATA=0 #measure compute times only
$ make COMPILER=nvcc MODE=cudav2 INCLUDE_DATA=1 #measure both memory transfer times and compute times (default)
$ make COMPILER=nvcc MODE=cudav2 INCLUDE_DATA=0 #measure compute times only
$ make COMPILER=nvcc MODE=cudav3 INCLUDE_DATA=1 #measure both memory transfer times and compute times (default)
$ make COMPILER=nvcc MODE=cudav3 INCLUDE_DATA=0 #measure compute times only
$ make COMPILER=nvcc MODE=cudav4 USE_ASYNC=1 #for asynchronous execution (default)
$ make COMPILER=nvcc MODE=cudav4 USE_ASYNC=0 #for synchronous execution
$ make COMPILER=nvcc MODE=cudav4 INCLUDE_DATA=1 #measure both memory transfer times and compute times (default)
$ make COMPILER=nvcc MODE=cudav4 INCLUDE_DATA=0 #measure compute times only
$ make COMPILER=nvcc MODE=cudauvm INCLUDE_DATA=1 #measure both memory transfer times and compute times (default)
$ make COMPILER=nvcc MODE=cudauvm INCLUDE_DATA=0 #measure compute times only

Eigen

Compilers: gcc, icc, nvcc

$ make COMPILER=gcc MODE=eigen
$ make COMPILER=icc MODE=eigen
$ make COMPILER=nvcc MODE=eigen

Alpaka

Compilers: gcc, nvcc

$ make COMPILER=gcc MODE=alpaka
$ make COMPILER=nvcc MODE=alpaka//not yet functional

Kokkos

Multiple Kokkos versions exist in the src directory: src/kokkos_src_v1, ..., src/kokkos_src_v6, each of which contains the p2z code along with a Makefile and README to help the user make use of the Kokkos library.

Here is the brief information on each version:

src/kokkos_src_v1: target CUDA GPU with unified memory; has user data layouts different from those in the manual CUDA versions.

src/kokkos_src_v2: target CUDA GPU without unified memory; has the same user data layouts as src/kokkos_src_v1 but with explicit memory transfers at each iteration of the outermost itr loop, as the manual CUDA versions do.

src/kokkos_src_v3: target CUDA GPU without unified memory; has the same user data layouts and memory transfer patterns as the manual CUDA_v3 version (propagate-toz-test_CUDA_v3.cu) but using a single device instance.

src/kokkos_src_v4: has the same user data layouts and compute patterns as `src/kokkos_src_v3, but use multiple asynchronous device instances (multiple CUDA streams)

src/kokkos_src_v5: target CUDA GPU without unified memory; has the same user data layouts and memory transfer patterns as the manual CUDA_v4 version (propagate-toz-test_CUDA_v4.cu) but using a single device instance.

src/kokkos_src_v6: has the same user data layouts and compute patterns as `src/kokkos_src_v5, but use multiple asynchronous device instances (multiple CUDA streams); the best performing version.

Here we have basic instructions.

Getting started

1. clone https://github.com/kokkos/kokkos to ${KOKKOS_ROOT}
2. in the src/kokkos_src_v6 direcory edit the Makefile setting to match your needs
3. run make to build
4. The executable can be found under bin with the others

Makefile setting

• KOKKOS_DEVICES = OpenMP or OpenMP,Cuda
• KOKKOS_ARCH = [GPU arch],[CPU arch] i.e. "BDW,Volta70"