Skip to content
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

Add CUDA/HIP TPL support for KokkosSparse::spadd #1962

Merged
merged 2 commits into from
Jan 16, 2024
Merged

Add CUDA/HIP TPL support for KokkosSparse::spadd #1962

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
733a090
spadd: change arguments to ctor of SPADDHandle
jczhang07 Jan 9, 2024
c9d03f2
spadd: add cuda/rocm TPL support for spadd_symbolic/numeric
jczhang07 Oct 25, 2023
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
10 changes: 6 additions & 4 deletions perf_test/sparse/KokkosSparse_spadd.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -303,8 +303,8 @@ void run_experiment(int argc, char** argv, CommonInputParams) {
double numericTime = 0;

// Do an untimed warm up symbolic, and preallocate space for C entries/values
spadd_symbolic(&kh, A.graph.row_map, A.graph.entries, B.graph.row_map,
B.graph.entries, row_mapC);
spadd_symbolic(exec_space{}, &kh, A.numRows(), A.numCols(), A.graph.row_map,
A.graph.entries, B.graph.row_map, B.graph.entries, row_mapC);

bool use_kk = !params.use_cusparse && !params.use_mkl;

Expand Down Expand Up @@ -366,7 +366,8 @@ void run_experiment(int argc, char** argv, CommonInputParams) {
for (int sumRep = 0; sumRep < params.repeat; sumRep++) {
timer.reset();
if (use_kk) {
spadd_symbolic(&kh, A.graph.row_map, A.graph.entries, B.graph.row_map,
spadd_symbolic(exec_space{}, &kh, A.numRows(), A.numCols(),
A.graph.row_map, A.graph.entries, B.graph.row_map,
B.graph.entries, row_mapC);
c_nnz = addHandle->get_c_nnz();
} else if (params.use_cusparse) {
Expand Down Expand Up @@ -434,7 +435,8 @@ void run_experiment(int argc, char** argv, CommonInputParams) {
}
#endif
} else {
spadd_numeric(&kh, A.graph.row_map, A.graph.entries, A.values,
spadd_numeric(exec_space{}, &kh, A.numRows(), A.numCols(),
A.graph.row_map, A.graph.entries, A.values,
1.0, // A, alpha
B.graph.row_map, B.graph.entries, B.values,
1.0, // B, beta
Expand Down
22 changes: 14 additions & 8 deletions sparse/impl/KokkosSparse_par_ilut_numeric_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -699,18 +699,24 @@ struct IlutWrap {
multiply_matrices(kh, ih, L_row_map, L_entries, L_values, U_row_map,
U_entries, U_values, LU_row_map, LU_entries, LU_values);

auto addHandle = kh.get_spadd_handle();
KokkosSparse::Experimental::spadd_symbolic(
&kh, A_row_map, A_entries, LU_row_map, LU_entries, R_row_map);
auto addHandle = kh.get_spadd_handle();
typename KHandle::const_nnz_lno_t m = A_row_map.extent(0) - 1,
n = m; // square matrix
// TODO: let compute_residual_norm also take an execution space argument and
// use that for exec!
typename KHandle::HandleExecSpace exec{};
KokkosSparse::Experimental::spadd_symbolic(exec, &kh, m, n, A_row_map,
A_entries, LU_row_map,
LU_entries, R_row_map);

const size_type r_nnz = addHandle->get_c_nnz();
Kokkos::resize(R_entries, r_nnz);
Kokkos::resize(R_values, r_nnz);
Kokkos::resize(exec, R_entries, r_nnz);
Kokkos::resize(exec, R_values, r_nnz);

KokkosSparse::Experimental::spadd_numeric(
&kh, A_row_map, A_entries, A_values, 1., LU_row_map, LU_entries,
LU_values, -1., R_row_map, R_entries, R_values);

exec, &kh, m, n, A_row_map, A_entries, A_values, 1., LU_row_map,
LU_entries, LU_values, -1., R_row_map, R_entries, R_values);
// TODO: how to make this policy use exec?
auto policy = ih.get_default_team_policy();

Kokkos::parallel_reduce(
Expand Down
31 changes: 15 additions & 16 deletions sparse/impl/KokkosSparse_spadd_numeric_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -174,24 +174,23 @@ struct UnsortedNumericSumFunctor {
std::is_same<typename std::remove_const<A>::type, \
typename std::remove_const<B>::type>::value

template <typename KernelHandle, typename alno_row_view_t,
typename alno_nnz_view_t, typename ascalar_t,
typename ascalar_nnz_view_t, typename blno_row_view_t,
typename blno_nnz_view_t, typename bscalar_t,
typename bscalar_nnz_view_t, typename clno_row_view_t,
typename clno_nnz_view_t, typename cscalar_nnz_view_t>
template <
typename execution_space, typename KernelHandle, typename alno_row_view_t,
typename alno_nnz_view_t, typename ascalar_t, typename ascalar_nnz_view_t,
typename blno_row_view_t, typename blno_nnz_view_t, typename bscalar_t,
typename bscalar_nnz_view_t, typename clno_row_view_t,
typename clno_nnz_view_t, typename cscalar_nnz_view_t>
void spadd_numeric_impl(
KernelHandle* kernel_handle, const alno_row_view_t a_rowmap,
const alno_nnz_view_t a_entries, const ascalar_nnz_view_t a_values,
const ascalar_t alpha, const blno_row_view_t b_rowmap,
const blno_nnz_view_t b_entries, const bscalar_nnz_view_t b_values,
const bscalar_t beta, const clno_row_view_t c_rowmap,
clno_nnz_view_t c_entries, cscalar_nnz_view_t c_values) {
const execution_space& exec, KernelHandle* kernel_handle,
const alno_row_view_t a_rowmap, const alno_nnz_view_t a_entries,
const ascalar_nnz_view_t a_values, const ascalar_t alpha,
const blno_row_view_t b_rowmap, const blno_nnz_view_t b_entries,
const bscalar_nnz_view_t b_values, const bscalar_t beta,
const clno_row_view_t c_rowmap, clno_nnz_view_t c_entries,
cscalar_nnz_view_t c_values) {
typedef typename KernelHandle::size_type size_type;
typedef typename KernelHandle::nnz_lno_t ordinal_type;
typedef typename KernelHandle::nnz_scalar_t scalar_type;
typedef
typename KernelHandle::SPADDHandleType::execution_space execution_space;
// Check that A/B/C data types match KernelHandle types, and that C data types
// are nonconst (doesn't matter if A/B types are const)
static_assert(SAME_TYPE(ascalar_t, scalar_type),
Expand Down Expand Up @@ -252,7 +251,7 @@ void spadd_numeric_impl(
sortedNumeric(a_rowmap, b_rowmap, c_rowmap, a_entries, b_entries,
c_entries, a_values, b_values, c_values, alpha, beta);
Kokkos::parallel_for("KokkosSparse::SpAdd:Numeric::InputSorted",
range_type(0, nrows), sortedNumeric);
range_type(exec, 0, nrows), sortedNumeric);
} else {
// use a_pos and b_pos (set in the handle by symbolic) to quickly compute C
// entries and values
Expand All @@ -265,7 +264,7 @@ void spadd_numeric_impl(
c_entries, a_values, b_values, c_values, alpha, beta,
addHandle->get_a_pos(), addHandle->get_b_pos());
Kokkos::parallel_for("KokkosSparse::SpAdd:Numeric::InputNotSorted",
range_type(0, nrows), unsortedNumeric);
range_type(exec, 0, nrows), unsortedNumeric);
}
addHandle->set_call_numeric();
}
Expand Down
62 changes: 35 additions & 27 deletions sparse/impl/KokkosSparse_spadd_numeric_spec.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,10 +28,10 @@
namespace KokkosSparse {
namespace Impl {
// Specialization struct which defines whether a specialization exists
template <class KernelHandle, class a_size_view_t_, class a_lno_view_t,
class a_scalar_view_t, class b_size_view_t_, class b_lno_view_t,
class b_scalar_view_t, class c_size_view_t_, class c_lno_view_t,
class c_scalar_view_t>
template <class ExecSpace, class KernelHandle, class a_size_view_t_,
class a_lno_view_t, class a_scalar_view_t, class b_size_view_t_,
class b_lno_view_t, class b_scalar_view_t, class c_size_view_t_,
class c_lno_view_t, class c_scalar_view_t>
struct spadd_numeric_eti_spec_avail {
enum : bool { value = false };
};
Expand All @@ -44,6 +44,7 @@ struct spadd_numeric_eti_spec_avail {
MEM_SPACE_TYPE) \
template <> \
struct spadd_numeric_eti_spec_avail< \
EXEC_SPACE_TYPE, \
KokkosKernels::Experimental::KokkosKernelsHandle< \
const OFFSET_TYPE, const ORDINAL_TYPE, const SCALAR_TYPE, \
EXEC_SPACE_TYPE, MEM_SPACE_TYPE, MEM_SPACE_TYPE>, \
Expand Down Expand Up @@ -87,20 +88,22 @@ namespace Impl {
// Unification layer
/// \brief Implementation of KokkosBlas::spadd (sparse-sparse matrix addition)

template <class KernelHandle, class a_size_view_t, class a_lno_view_t,
class a_scalar_view_t, class b_size_view_t, class b_lno_view_t,
class b_scalar_view_t, class c_size_view_t, class c_lno_view_t,
class c_scalar_view_t,
template <class ExecSpace, class KernelHandle, class a_size_view_t,
class a_lno_view_t, class a_scalar_view_t, class b_size_view_t,
class b_lno_view_t, class b_scalar_view_t, class c_size_view_t,
class c_lno_view_t, class c_scalar_view_t,
bool tpl_spec_avail = spadd_numeric_tpl_spec_avail<
KernelHandle, a_size_view_t, a_lno_view_t, a_scalar_view_t,
b_size_view_t, b_lno_view_t, b_scalar_view_t, c_size_view_t,
c_lno_view_t, c_scalar_view_t>::value,
ExecSpace, KernelHandle, a_size_view_t, a_lno_view_t,
a_scalar_view_t, b_size_view_t, b_lno_view_t, b_scalar_view_t,
c_size_view_t, c_lno_view_t, c_scalar_view_t>::value,
bool eti_spec_avail = spadd_numeric_eti_spec_avail<
KernelHandle, a_size_view_t, a_lno_view_t, a_scalar_view_t,
b_size_view_t, b_lno_view_t, b_scalar_view_t, c_size_view_t,
c_lno_view_t, c_scalar_view_t>::value>
ExecSpace, KernelHandle, a_size_view_t, a_lno_view_t,
a_scalar_view_t, b_size_view_t, b_lno_view_t, b_scalar_view_t,
c_size_view_t, c_lno_view_t, c_scalar_view_t>::value>
struct SPADD_NUMERIC {
static void spadd_numeric(KernelHandle *handle,
static void spadd_numeric(const ExecSpace &exec, KernelHandle *handle,
typename KernelHandle::const_nnz_lno_t m,
typename KernelHandle::const_nnz_lno_t n,
typename a_scalar_view_t::const_value_type alpha,
a_size_view_t row_mapA, a_lno_view_t entriesA,
a_scalar_view_t valuesA,
Expand All @@ -112,24 +115,27 @@ struct SPADD_NUMERIC {

#if !defined(KOKKOSKERNELS_ETI_ONLY) || KOKKOSKERNELS_IMPL_COMPILE_LIBRARY

template <class KernelHandle, class a_size_view_t, class a_lno_view_t,
class a_scalar_view_t, class b_size_view_t, class b_lno_view_t,
class b_scalar_view_t, class c_size_view_t, class c_lno_view_t,
class c_scalar_view_t>
struct SPADD_NUMERIC<KernelHandle, a_size_view_t, a_lno_view_t, a_scalar_view_t,
b_size_view_t, b_lno_view_t, b_scalar_view_t,
c_size_view_t, c_lno_view_t, c_scalar_view_t, false,
KOKKOSKERNELS_IMPL_COMPILE_LIBRARY> {
static void spadd_numeric(KernelHandle *handle,
template <class ExecSpace, class KernelHandle, class a_size_view_t,
class a_lno_view_t, class a_scalar_view_t, class b_size_view_t,
class b_lno_view_t, class b_scalar_view_t, class c_size_view_t,
class c_lno_view_t, class c_scalar_view_t>
struct SPADD_NUMERIC<
ExecSpace, KernelHandle, a_size_view_t, a_lno_view_t, a_scalar_view_t,
b_size_view_t, b_lno_view_t, b_scalar_view_t, c_size_view_t, c_lno_view_t,
c_scalar_view_t, false, KOKKOSKERNELS_IMPL_COMPILE_LIBRARY> {
static void spadd_numeric(const ExecSpace &exec, KernelHandle *handle,
typename KernelHandle::const_nnz_lno_t /* m */,
typename KernelHandle::const_nnz_lno_t /* n */,
typename a_scalar_view_t::const_value_type alpha,
a_size_view_t row_mapA, a_lno_view_t entriesA,
a_scalar_view_t valuesA,
typename b_scalar_view_t::const_value_type beta,
b_size_view_t row_mapB, b_lno_view_t entriesB,
b_scalar_view_t valuesB, c_size_view_t row_mapC,
c_lno_view_t entriesC, c_scalar_view_t valuesC) {
spadd_numeric_impl(handle, row_mapA, entriesA, valuesA, alpha, row_mapB,
entriesB, valuesB, beta, row_mapC, entriesC, valuesC);
spadd_numeric_impl(exec, handle, row_mapA, entriesA, valuesA, alpha,
row_mapB, entriesB, valuesB, beta, row_mapC, entriesC,
valuesC);
}
};

Expand All @@ -142,6 +148,7 @@ struct SPADD_NUMERIC<KernelHandle, a_size_view_t, a_lno_view_t, a_scalar_view_t,
SCALAR_TYPE, ORDINAL_TYPE, OFFSET_TYPE, LAYOUT_TYPE, EXEC_SPACE_TYPE, \
MEM_SPACE_TYPE) \
extern template struct SPADD_NUMERIC< \
EXEC_SPACE_TYPE, \
typename KokkosKernels::Experimental::KokkosKernelsHandle< \
const OFFSET_TYPE, const ORDINAL_TYPE, const SCALAR_TYPE, \
EXEC_SPACE_TYPE, MEM_SPACE_TYPE, MEM_SPACE_TYPE>, \
Expand Down Expand Up @@ -178,6 +185,7 @@ struct SPADD_NUMERIC<KernelHandle, a_size_view_t, a_lno_view_t, a_scalar_view_t,
SCALAR_TYPE, ORDINAL_TYPE, OFFSET_TYPE, LAYOUT_TYPE, EXEC_SPACE_TYPE, \
MEM_SPACE_TYPE) \
template struct SPADD_NUMERIC< \
EXEC_SPACE_TYPE, \
KokkosKernels::Experimental::KokkosKernelsHandle< \
const OFFSET_TYPE, const ORDINAL_TYPE, const SCALAR_TYPE, \
EXEC_SPACE_TYPE, MEM_SPACE_TYPE, MEM_SPACE_TYPE>, \
Expand Down Expand Up @@ -210,6 +218,6 @@ struct SPADD_NUMERIC<KernelHandle, a_size_view_t, a_lno_view_t, a_scalar_view_t,
Kokkos::MemoryTraits<Kokkos::Unmanaged> >, \
false, true>;

#include <KokkosSparse_spadd_tpl_spec_decl.hpp>
#include <KokkosSparse_spadd_numeric_tpl_spec_decl.hpp>

#endif
Loading