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

Kokkos lib update

Browse Source
This commit is contained in:
Steve Plimpton
2016-09-08 13:56:18 -06:00
parent 0252347d43
commit 236ebf7fab
212 changed files with 18902 additions and 13466 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
lib/kokkos/example/tutorial/01_hello_world/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/01_hello_world_lambda/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/02_simple_reduce/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/02_simple_reduce_lambda/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/03_simple_view/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/03_simple_view_lambda/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/04_simple_memoryspaces/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/05_simple_atomics/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/Advanced_Views/01_data_layouts/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

4
lib/kokkos/example/tutorial/Advanced_Views/01_data_layouts/data_layouts.cpp
View File

@ -142,7 +142,7 @@ int main (int narg, char* arg[]) {
// Measure time to execute the contraction kernel when giving it a
// LayoutLeft view for v1 and a LayoutRight view for v2. This should be
// fast on GPUs and slow on CPUs
Kokkos::Impl::Timer time1;
Kokkos::Timer time1;
Kokkos::parallel_for(size,contraction<left_type,right_type>(a,l,r));
Kokkos::fence();
double sec1 = time1.seconds();
@ -154,7 +154,7 @@ int main (int narg, char* arg[]) {
// Measure time to execute the contraction kernel when giving it a
// LayoutRight view for v1 and a LayoutLeft view for v2. This should be
// fast on CPUs and slow on GPUs
Kokkos::Impl::Timer time2;
Kokkos::Timer time2;
Kokkos::parallel_for(size,contraction<right_type,left_type>(a,r,l));
Kokkos::fence();
double sec2 = time2.seconds();

2
lib/kokkos/example/tutorial/Advanced_Views/02_memory_traits/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

4
lib/kokkos/example/tutorial/Advanced_Views/02_memory_traits/memory_traits.cpp
View File

@ -124,12 +124,12 @@ int main(int narg, char* arg[]) {
// Run the localsum functor using the RandomAccess trait. On CPUs there should
// not be any different in performance to not using the RandomAccess trait.
// On GPUs where can be a dramatic difference
Kokkos::Impl::Timer time1;
Kokkos::Timer time1;
Kokkos::parallel_for(size,localsum<view_type,view_type_rnd>(idx,dest,src));
Kokkos::fence();
double sec1 = time1.seconds();
Kokkos::Impl::Timer time2;
Kokkos::Timer time2;
Kokkos::parallel_for(size,localsum<view_type,view_type>(idx,dest,src));
Kokkos::fence();
double sec2 = time2.seconds();

2
lib/kokkos/example/tutorial/Advanced_Views/03_subviews/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/Advanced_Views/04_dualviews/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

10
lib/kokkos/example/tutorial/Advanced_Views/04_dualviews/dual_view.cpp
View File

@ -87,9 +87,9 @@ struct localsum {
// For example, the const_data_type version of double** is const
// double**.
Kokkos::View<idx_type::const_data_type, idx_type::array_layout, memory_space> idx;
// "array_intrinsic_type" is a typedef in ViewTraits (and DualView) which is the
// "scalar_array_type" is a typedef in ViewTraits (and DualView) which is the
// array version of the value(s) stored in the View.
Kokkos::View<view_type::array_intrinsic_type, view_type::array_layout, memory_space> dest;
Kokkos::View<view_type::scalar_array_type, view_type::array_layout, memory_space> dest;
Kokkos::View<view_type::const_data_type, view_type::array_layout,
memory_space, Kokkos::MemoryRandomAccess> src;
@ -150,6 +150,9 @@ protected:
int main (int narg, char* arg[]) {
Kokkos::initialize (narg, arg);
// If View is non-trivial constructible type then add braces so it is out of scope
// before Kokkos::finalize() call
{
ParticleTypes test("Test");
Kokkos::fence();
test.h_view(0) = ParticleType(-1e4,1);
@ -182,7 +185,7 @@ int main (int narg, char* arg[]) {
// Run on the device. This will cause a sync of idx to the device,
// since it was marked as modified on the host.
Kokkos::Impl::Timer timer;
Kokkos::Timer timer;
Kokkos::parallel_for(size,localsum<view_type::execution_space>(idx,dest,src));
Kokkos::fence();
double sec1_dev = timer.seconds();
@ -208,6 +211,7 @@ int main (int narg, char* arg[]) {
printf("Device Time with Sync: %f without Sync: %f \n",sec1_dev,sec2_dev);
printf("Host Time with Sync: %f without Sync: %f \n",sec1_host,sec2_host);
}
Kokkos::finalize();
}

2
lib/kokkos/example/tutorial/Advanced_Views/05_NVIDIA_UVM/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/Advanced_Views/05_NVIDIA_UVM/uvm_example.cpp
View File

@ -97,7 +97,7 @@ int main(int narg, char* arg[]) {
Kokkos::fence();
// Run on the device
// This will cause a sync of idx to the device since it was modified on the host
Kokkos::Impl::Timer timer;
Kokkos::Timer timer;
Kokkos::parallel_for(size,localsum<view_type::execution_space>(idx,dest,src));
Kokkos::fence();
double sec1_dev = timer.seconds();

2
lib/kokkos/example/tutorial/Advanced_Views/06_AtomicViews/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/Advanced_Views/07_Overlapping_DeepCopy/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3 --default-stream per-thread
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/Advanced_Views/07_Overlapping_DeepCopy/overlapping_deepcopy.cpp
View File

@ -116,7 +116,7 @@ int main(int argc, char * argv[]) {
Kokkos::parallel_for(Kokkos::RangePolicy<Kokkos::Cuda>(0,size),FillDevice(0.0,d_a));
Kokkos::parallel_for(Kokkos::RangePolicy<Kokkos::Cuda>(0,size),FillDevice(1.3513,d_b));
Kokkos::fence();
Kokkos::Impl::Timer timer;
Kokkos::Timer timer;
Kokkos::parallel_for(Kokkos::RangePolicy<Kokkos::Cuda>(0,size),ComputeADevice(20,d_a,d_b));
if(synch==1)

2
lib/kokkos/example/tutorial/Algorithms/01_random_numbers/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/Algorithms/01_random_numbers/random_numbers.cpp
View File

@ -122,7 +122,7 @@ int main(int argc, char* args[]) {
Kokkos::DualView<uint64_t*> vals("Vals",size*samples);
// Run some performance comparisons
Kokkos::Impl::Timer timer;
Kokkos::Timer timer;
Kokkos::parallel_for(size,generate_random<Kokkos::Random_XorShift64_Pool<> >(vals.d_view,rand_pool64,samples));
Kokkos::fence();

2
lib/kokkos/example/tutorial/Hierarchical_Parallelism/01_thread_teams/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

3
lib/kokkos/example/tutorial/Hierarchical_Parallelism/01_thread_teams_lambda/Makefile
View File

@ -5,13 +5,14 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =
EXE = $(SRC:.cpp=.cuda)
KOKKOS_DEVICES = "Cuda,OpenMP"
KOKKOS_ARCH = "SNB,Kepler35"
KOKKOS_CUDA_OPTIONS = "enable_lambda"
else
CXX = g++
CXXFLAGS = -O3

3
lib/kokkos/example/tutorial/Hierarchical_Parallelism/01_thread_teams_lambda/thread_teams_lambda.cpp
View File

@ -62,7 +62,8 @@ int main (int narg, char* args[]) {
// Set up a policy that launches 12 teams, with the maximum number
// of threads per team.
const team_policy policy (12, team_policy::team_size_max ( [=]{} ));
const team_policy policy (12, Kokkos::AUTO);
// This is a reduction with a team policy. The team policy changes
// the first argument of the lambda. Rather than an integer index

2
lib/kokkos/example/tutorial/Hierarchical_Parallelism/02_nested_parallel_for/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/Hierarchical_Parallelism/03_vectorization/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

6
lib/kokkos/example/tutorial/Hierarchical_Parallelism/03_vectorization/vectorization.cpp
View File

@ -141,11 +141,11 @@ int main(int narg, char* args[]) {
// 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.
// The team_size_max function will determine the maximum number of threads taking into account
// shared memory requirements of the Functor.
// 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::TeamPolicy<>::team_size_max(SomeCorrelation(data,gsum)) , 16);
const Kokkos::TeamPolicy<> policy( 512 , Kokkos::AUTO , 16);
Kokkos::parallel_for( policy , SomeCorrelation(data,gsum) );

2
lib/kokkos/example/tutorial/Hierarchical_Parallelism/04_team_scan/Makefile
View File

@ -5,7 +5,7 @@ default: build
echo "Start Build"
ifneq (,$(findstring Cuda,$(KOKKOS_DEVICES)))
CXX = nvcc_wrapper
CXX = ../../../../config/nvcc_wrapper
CXXFLAGS = -O3
LINK = ${CXX}
LINKFLAGS =

2
lib/kokkos/example/tutorial/Hierarchical_Parallelism/04_team_scan/team_scan.cpp
View File

@ -117,7 +117,7 @@ int main(int narg, char* args[]) {
Kokkos::DualView<int**> histogram("histogram",TEAM_SIZE,TEAM_SIZE);
Kokkos::Impl::Timer timer;
Kokkos::Timer timer;
// threads/team is automatically limited to maximum supported by the device.
Kokkos::parallel_for( team_policy( nchunks , TEAM_SIZE )
, find_2_tuples(chunk_size,data,histogram) );