ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update Kokkos library to v2.7.00

Browse Source
This commit is contained in:
Stan Moore
2018-05-25 15:00:53 -06:00
parent ce4a446cea
commit 1422b0413b
1248 changed files with 64103 additions and 5133 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
lib/kokkos/example/tutorial/Hierarchical_Parallelism/03_vectorization/vectorization.cpp
View File

@ -130,30 +130,32 @@ struct SomeCorrelation {
int main(int narg, char* args[]) {
Kokkos::initialize(narg,args);
// Produce some 3D random data (see Algorithms/01_random_numbers for more info)
Kokkos::View<int***,Kokkos::LayoutRight> data("Data",512,512,32);
Kokkos::Random_XorShift64_Pool<> rand_pool64(5374857);
Kokkos::fill_random(data,rand_pool64,100);
{
// Produce some 3D random data (see Algorithms/01_random_numbers for more info)
Kokkos::View<int***,Kokkos::LayoutRight> data("Data",512,512,32);
Kokkos::Random_XorShift64_Pool<> rand_pool64(5374857);
Kokkos::fill_random(data,rand_pool64,100);
// A global value to put the result in
Kokkos::View<int> gsum("Sum");
// A global value to put the result in
Kokkos::View<int> gsum("Sum");
// Each team handles a slice of the data
// Set up TeamPolicy with 512 teams with maximum number of threads per team and 16 vector lanes.
// Kokkos::AUTO will determine the number of threads
// The maximum vector length is hardware dependent but can always be smaller than the hardware allows.
// The vector length must be a power of 2.
// Each team handles a slice of the data
// Set up TeamPolicy with 512 teams with maximum number of threads per team and 16 vector lanes.
// Kokkos::AUTO will determine the number of threads
// The maximum vector length is hardware dependent but can always be smaller than the hardware allows.
// The vector length must be a power of 2.
const Kokkos::TeamPolicy<> policy( 512 , Kokkos::AUTO , 16);
const Kokkos::TeamPolicy<> policy( 512 , Kokkos::AUTO , 16);
Kokkos::parallel_for( policy , SomeCorrelation(data,gsum) );
Kokkos::parallel_for( policy , SomeCorrelation(data,gsum) );
Kokkos::fence();
Kokkos::fence();
// Copy result value back
int sum = 0;
Kokkos::deep_copy(sum,gsum);
printf("Result %i\n",sum);
// Copy result value back
int sum = 0;
Kokkos::deep_copy(sum,gsum);
printf("Result %i\n",sum);
}
Kokkos::finalize();
}

11
lib/kokkos/example/tutorial/Hierarchical_Parallelism/04_team_scan/team_scan.cpp
View File

@ -104,14 +104,15 @@ struct find_2_tuples {
int main(int narg, char* args[]) {
Kokkos::initialize(narg,args);
{
int chunk_size = 1024;
int nchunks = 100000; //1024*1024;
Kokkos::DualView<int*> data("data",nchunks*chunk_size+1);
srand(1231093);
for(int i = 0; i < (int) data.dimension_0(); i++) {
for(int i = 0; i < (int) data.extent(0); i++) {
data.h_view(i) = rand()%TEAM_SIZE;
}
data.modify<Host>();
@ -122,7 +123,10 @@ int main(int narg, char* args[]) {
Kokkos::Timer timer;
// threads/team is automatically limited to maximum supported by the device.
Kokkos::parallel_for( team_policy( nchunks , TEAM_SIZE )
int team_size = TEAM_SIZE;
if( team_size > Device::execution_space::concurrency() )
team_size = Device::execution_space::concurrency();
Kokkos::parallel_for( team_policy( nchunks , team_size )
, find_2_tuples(chunk_size,data,histogram) );
Kokkos::fence();
double time = timer.seconds();
@ -139,6 +143,7 @@ int main(int narg, char* args[]) {
printf("\n");
}
printf("Result: %i %i\n",sum,chunk_size*nchunks);
}
Kokkos::finalize();
}