Skip to content

CabanaSliceFactory now works with multiple types. #3

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
HazelCullom merged 6 commits into from
Oct 12, 2022
Merged

CabanaSliceFactory now works with multiple types. #3

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
34177d0
Merge pull request #1 from SCOREC/main
HazelCullom Oct 6, 2022
8a49253
trying c++ tuples to extract parameter pack types at index
HazelCullom Oct 6, 2022
7009fba
need to add correct stride length calculation
HazelCullom Oct 6, 2022
9c8249d
multi-slice factory works, only for types that have a size that is a …
HazelCullom Oct 7, 2022
d62a009
now can use cabSlicFactory with any types, regardless of size
HazelCullom Oct 11, 2022
790cedd
no longer requires user to define vecotr length, DefaultExecutionSpac…
HazelCullom Oct 12, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 21 additions & 10 deletions src/SliceWrapper.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,8 +11,8 @@ struct SliceWrapper {
SliceWrapper(SliceType st) : st_(st) {}

KOKKOS_INLINE_FUNCTION
T& access(const int s, const int a, int i) const {
return st_.access(s,a,i);
T& access(const int s, const int a) const {
return st_.access(s,a);
}
int arraySize(int s) {
return st_.arraySize(s);
Expand All @@ -24,23 +24,34 @@ struct SliceWrapper {

using namespace Cabana;

template <class ExecutionSpace, class MemorySpace, class T, int width, int vecLen>
template <class ExecutionSpace, class MemorySpace, class... Ts>
class CabSliceFactory {
static constexpr int vecLen = Impl::PerformanceTraits<ExecutionSpace>::vector_length/8;
Copy link
Contributor

@cwsmith cwsmith Oct 12, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What is the division by 8? Is the vector_length reported in bits?

Can you point me at where Cabana uses this?

Copy link
Contributor Author

@HazelCullom HazelCullom Oct 12, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I divided it by eight because when I changed to this vector length instead of the user defined one, the stride calculation got messed up, being eight times larger than it should have been which gave me errors. I have just revised the calculation to be correct and not divide the vector length by 8.

Cabana uses this here as a default template parameter for the AoSoA
https://github.com/ECP-copa/Cabana/blob/1c44a7fef24b61f73e629ddd2084c4d05b0ec977/core/src/Cabana_AoSoA.hpp#L146:~:text=int%20VectorLength%20%3D%20Impl,DeviceType%3A%3Aexecution_space%3E%3A%3Avector_length

using TypeTuple = std::tuple<Ts...>;
using DeviceType = Kokkos::Device<ExecutionSpace, MemorySpace>;
using DataTypes = Cabana::MemberTypes<T[width]>;
using DataTypes = Cabana::MemberTypes<Ts...>;
using soa_t = SoA<DataTypes, vecLen>;

template <class T, int stride>
using member_slice_t =
Cabana::Slice<T[width], DeviceType,
Cabana::Slice<T, DeviceType,
Cabana::DefaultAccessMemory,
vecLen, width*vecLen>;
using wrapper_slice_t = SliceWrapper<member_slice_t, T>;
vecLen, stride>;

template <class T, int stride>
using wrapper_slice_t = SliceWrapper<member_slice_t<T, stride>, T>;

Cabana::AoSoA<DataTypes, DeviceType, vecLen> aosoa;

public:
wrapper_slice_t makeSliceCab() {
auto slice0 = Cabana::slice<0>(aosoa);
return wrapper_slice_t(std::move(slice0));
template <std::size_t index>
auto makeSliceCab() {
using type = std::tuple_element_t<index, TypeTuple>;
const int stride = (vecLen * sizeof(soa_t)) / (4 * sizeof(type));
auto slice = Cabana::slice<index>(aosoa);
return wrapper_slice_t< type, stride >(std::move(slice));
}

CabSliceFactory(int n) : aosoa("sliceAoSoA", n) {}
};

Expand Down
31 changes: 17 additions & 14 deletions test/SliceWrapper.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,34 +3,37 @@
int main(int argc, char* argv[]) {
// AoSoA parameters
const int vecLen = 4;
Copy link
Contributor

@cwsmith cwsmith Oct 12, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should this still be here?

Copy link
Contributor Author

@HazelCullom HazelCullom Oct 12, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nope. It will be removed in the next version

const int width = 1;
int num_tuples = 10;

Kokkos::ScopeGuard scope_guard(argc, argv);

using member_type = double;
using DataTypes = Cabana::MemberTypes<member_type[width]>;
using ExecutionSpace = Kokkos::Cuda;
using MemorySpace = Kokkos::CudaSpace;

using ExecutionSpace = Kokkos::DefaultExecutionSpace;
using MemorySpace = ExecutionSpace::memory_space;

// Slice Wrapper Factory
CabSliceFactory<ExecutionSpace, MemorySpace,
member_type, width, vecLen> cabSliceFactory(num_tuples);
double, int, float, char> cabSliceFactory(num_tuples);

auto slice_wrapper = cabSliceFactory.makeSliceCab();
auto slice_wrapper0 = cabSliceFactory.makeSliceCab<0>();
auto slice_wrapper1 = cabSliceFactory.makeSliceCab<1>();
auto slice_wrapper2 = cabSliceFactory.makeSliceCab<2>();
auto slice_wrapper3 = cabSliceFactory.makeSliceCab<3>();

// simd_parallel_for setup
Cabana::SimdPolicy<vecLen, ExecutionSpace> simd_policy(0, num_tuples);
Copy link
Contributor

@cwsmith cwsmith Oct 12, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we eliminate the vecLen here?

Copy link
Contributor Author

@HazelCullom HazelCullom Oct 12, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done!


// kernel that reads and writes
auto vector_kernel = KOKKOS_LAMBDA(const int s, const int a) {
for (int i = 0; i < width; i++) {
printf("s: %d, a: %d, i: %d\n", s,a,i);
double x = 42/(s+a+1.3);
slice_wrapper.access(s,a,i) = x;
printf("value: %lf\n", slice_wrapper.access(s,a,i));
}
printf("s: %d, a: %d\n", s,a);
double x = 42/(s+a+1.3);
slice_wrapper0.access(s,a) = x;
slice_wrapper1.access(s,a) = s+a;
slice_wrapper2.access(s,a) = float(x);
slice_wrapper3.access(s,a) = 'a'+s+a;
printf("SW0 value: %lf\n", slice_wrapper0.access(s,a));
printf("SW1 value: %d\n", slice_wrapper1.access(s,a));
printf("SW2 value: %f\n", slice_wrapper2.access(s,a));
printf("SW3 value: %c\n", slice_wrapper3.access(s,a));
};

Cabana::simd_parallel_for(simd_policy, vector_kernel, "parallel_for_cabSliceFactory");
Expand Down