| 1 |
// @HEADER |
| 2 |
// ***************************************************************************** |
| 3 |
// Stokhos Package |
| 4 |
// |
| 5 |
// Copyright 2009 NTESS and the Stokhos contributors. |
| 6 |
// SPDX-License-Identifier: BSD-3-Clause |
| 7 |
// ***************************************************************************** |
| 8 |
// @HEADER |
| 9 |
|
| 10 |
#include <iostream> |
| 11 |
|
| 12 |
// Devices |
| 13 |
#include "Kokkos_Core.hpp" |
| 14 |
|
| 15 |
// Utilities |
| 16 |
#include "Teuchos_CommandLineProcessor.hpp" |
| 17 |
#include "Teuchos_StandardCatchMacros.hpp" |
| 18 |
#ifdef KOKKOS_ENABLE_CUDA |
| 19 |
#include "cuda_runtime_api.h" |
| 20 |
#endif |
| 21 |
|
| 22 |
template <typename Scalar, typename Ordinal, typename Device> |
| 23 |
void performance_test_driver( const Ordinal nGrid, |
| 24 |
const Ordinal nIter, |
| 25 |
const Ordinal ensemble_min, |
| 26 |
const Ordinal ensemble_max, |
| 27 |
const Ordinal ensemble_step ); |
| 28 |
|
| 29 |
int main(int argc, char *argv[]) |
| 30 |
{ |
| 31 |
bool success = true; |
| 32 |
bool verbose = false; |
| 33 |
try { |
| 34 |
|
| 35 |
#ifdef KOKKOS_ENABLE_THREADS |
| 36 |
const size_t num_sockets = Kokkos::hwloc::get_available_numa_count(); |
| 37 |
const size_t num_cores_per_socket = |
| 38 |
Kokkos::hwloc::get_available_cores_per_numa(); |
| 39 |
const size_t num_threads_per_core = |
| 40 |
Kokkos::hwloc::get_available_threads_per_core(); |
| 41 |
#endif |
| 42 |
|
| 43 |
// Setup command line options |
| 44 |
Teuchos::CommandLineProcessor CLP; |
| 45 |
CLP.setDocString( |
| 46 |
"This test performance of MP::Vector multiply routines.\n"); |
| 47 |
int nGrid = 32; |
| 48 |
CLP.setOption("n", &nGrid, "Number of mesh points in the each direction"); |
| 49 |
int nIter = 10; |
| 50 |
CLP.setOption("ni", &nIter, "Number of multiply iterations"); |
| 51 |
int ensemble_min = 4; |
| 52 |
CLP.setOption("emin", &ensemble_min, "Staring ensemble size"); |
| 53 |
int ensemble_max = 24; |
| 54 |
CLP.setOption("emax", &ensemble_max, "Stoping ensemble size"); |
| 55 |
int ensemble_step = 4; |
| 56 |
CLP.setOption("estep", &ensemble_step, "Ensemble increment"); |
| 57 |
#ifdef KOKKOS_ENABLE_THREADS |
| 58 |
bool threads = true; |
| 59 |
CLP.setOption("threads", "no-threads", &threads, "Enable Threads device"); |
| 60 |
int num_cores = num_cores_per_socket * num_sockets; |
| 61 |
CLP.setOption("cores", &num_cores, |
| 62 |
"Number of CPU cores to use (defaults to all)"); |
| 63 |
int num_hyper_threads = num_threads_per_core; |
| 64 |
CLP.setOption("hyperthreads", &num_hyper_threads, |
| 65 |
"Number of hyper threads per core to use (defaults to all)"); |
| 66 |
#endif |
| 67 |
#ifdef KOKKOS_ENABLE_CUDA |
| 68 |
bool cuda = true; |
| 69 |
CLP.setOption("cuda", "no-cuda", &cuda, "Enable Cuda device"); |
| 70 |
int device_id = 0; |
| 71 |
CLP.setOption("device", &device_id, "CUDA device ID"); |
| 72 |
#endif |
| 73 |
CLP.parse( argc, argv ); |
| 74 |
|
| 75 |
typedef int Ordinal; |
| 76 |
typedef double Scalar; |
| 77 |
|
| 78 |
#ifdef KOKKOS_ENABLE_THREADS |
| 79 |
if (threads) { |
| 80 |
typedef Kokkos::Threads Device; |
| 81 |
|
| 82 |
Kokkos::InitializationSettings init_args; |
| 83 |
init_args.set_num_threads(num_cores*num_hyper_threads); |
| 84 |
Kokkos::initialize( init_args ); |
| 85 |
|
| 86 |
std::cout << std::endl |
| 87 |
<< "Threads performance with " << num_cores*num_hyper_threads |
| 88 |
<< " threads:" << std::endl; |
| 89 |
|
| 90 |
performance_test_driver<Scalar,Ordinal,Device>( |
| 91 |
nGrid, nIter, ensemble_min, ensemble_max, ensemble_step); |
| 92 |
|
| 93 |
Kokkos::finalize(); |
| 94 |
} |
| 95 |
#endif |
| 96 |
|
| 97 |
#ifdef KOKKOS_ENABLE_CUDA |
| 98 |
if (cuda) { |
| 99 |
typedef Kokkos::Cuda Device; |
| 100 |
|
| 101 |
Kokkos::InitializationSettings init_args; |
| 102 |
init_args.set_device_id(device_id); |
| 103 |
Kokkos::initialize( init_args ); |
| 104 |
|
| 105 |
cudaDeviceProp deviceProp; |
| 106 |
cudaGetDeviceProperties(&deviceProp, device_id); |
| 107 |
std::cout << std::endl |
| 108 |
<< "CUDA performance for device " << device_id << " (" |
| 109 |
<< deviceProp.name << "):" |
| 110 |
<< std::endl; |
| 111 |
|
| 112 |
performance_test_driver<Scalar,Ordinal,Device>( |
| 113 |
nGrid, nIter, ensemble_min, ensemble_max, ensemble_step); |
| 114 |
|
| 115 |
Kokkos::finalize(); |
| 116 |
} |
| 117 |
#endif |
| 118 |
|
| 119 |
} |
| 120 |
TEUCHOS_STANDARD_CATCH_STATEMENTS(verbose, std::cerr, success); |
| 121 |
|
| 122 |
if (success) |
| 123 |
return 0; |
| 124 |
return -1; |
| 125 |
} |