Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
2 changes: 2 additions & 0 deletions blas/tpls/KokkosBlas1_rot_tpl_spec_avail.hpp
Original file line number Diff line number Diff line change
Expand Up @@ -119,6 +119,7 @@ KOKKOSBLAS1_ROT_TPL_SPEC_AVAIL_CUBLAS(Kokkos::complex<float>,
#endif

// rocBLAS
/*
#ifdef KOKKOSKERNELS_ENABLE_TPL_ROCBLAS
#define KOKKOSBLAS1_ROT_TPL_SPEC_AVAIL_ROCBLAS(SCALAR, LAYOUT, EXECSPACE, \
MEMSPACE) \
Expand All @@ -143,6 +144,7 @@ KOKKOSBLAS1_ROT_TPL_SPEC_AVAIL_ROCBLAS(Kokkos::complex<float>,
Kokkos::LayoutLeft, Kokkos::HIP,
Kokkos::HIPSpace)
#endif
*/

} // namespace Impl
} // namespace KokkosBlas
Expand Down
36 changes: 32 additions & 4 deletions blas/tpls/KokkosBlas1_rotg_tpl_spec_decl.hpp
Original file line number Diff line number Diff line change
Expand Up @@ -521,9 +521,16 @@ namespace Impl {
KokkosBlas::Impl::RocBlasSingleton& singleton = \
KokkosBlas::Impl::RocBlasSingleton::singleton(); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblasSetStream(singleton.handle, space.hip_stream())); \
rocblas_set_stream(singleton.handle, space.hip_stream())); \
rocblas_pointer_mode pointer_mode; \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_get_pointer_mode(singleton.handle, &pointer_mode)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_set_pointer_mode( \
singleton.handle, rocblas_pointer_mode_device)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_drotg( \
singleton.handle, a.data(), b.data(), c.data(), s.data())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(singleton.handle, pointer_mode)); \
Kokkos::Profiling::popRegion(); \
} \
};
Expand Down Expand Up @@ -551,9 +558,16 @@ namespace Impl {
KokkosBlas::Impl::RocBlasSingleton& singleton = \
KokkosBlas::Impl::RocBlasSingleton::singleton(); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblasSetStream(singleton.handle, space.hip_stream())); \
rocblas_set_stream(singleton.handle, space.hip_stream())); \
rocblas_pointer_mode pointer_mode; \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_get_pointer_mode(singleton.handle, &pointer_mode)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_set_pointer_mode( \
singleton.handle, rocblas_pointer_mode_device)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_srotg( \
singleton.handle, a.data(), b.data(), c.data(), s.data())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(singleton.handle, pointer_mode)); \
Kokkos::Profiling::popRegion(); \
} \
};
Expand Down Expand Up @@ -584,12 +598,19 @@ namespace Impl {
KokkosBlas::Impl::RocBlasSingleton& singleton = \
KokkosBlas::Impl::RocBlasSingleton::singleton(); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblasSetStream(singleton.handle, space.hip_stream())); \
rocblas_set_stream(singleton.handle, space.hip_stream())); \
rocblas_pointer_mode pointer_mode; \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_get_pointer_mode(singleton.handle, &pointer_mode)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_set_pointer_mode( \
singleton.handle, rocblas_pointer_mode_device)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_zrotg( \
singleton.handle, \
reinterpret_cast<rocblas_double_complex*>(a.data()), \
reinterpret_cast<rocblas_double_complex*>(b.data()), c.data(), \
reinterpret_cast<rocblas_double_complex*>(s.data()))); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(singleton.handle, pointer_mode)); \
Kokkos::Profiling::popRegion(); \
} \
};
Expand Down Expand Up @@ -619,12 +640,19 @@ namespace Impl {
KokkosBlas::Impl::RocBlasSingleton& singleton = \
KokkosBlas::Impl::RocBlasSingleton::singleton(); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblasSetStream(singleton.handle, space.hip_stream())); \
rocblas_set_stream(singleton.handle, space.hip_stream())); \
rocblas_pointer_mode pointer_mode; \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_get_pointer_mode(singleton.handle, &pointer_mode)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_set_pointer_mode( \
singleton.handle, rocblas_pointer_mode_device)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_crotg( \
singleton.handle, \
reinterpret_cast<rocblas_float_complex*>(a.data()), \
reinterpret_cast<rocblas_float_complex*>(b.data()), c.data(), \
reinterpret_cast<rocblas_float_complex*>(s.data()))); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(singleton.handle, pointer_mode)); \
Kokkos::Profiling::popRegion(); \
} \
};
Expand Down
71 changes: 39 additions & 32 deletions blas/tpls/KokkosBlas1_rotm_tpl_spec_decl.hpp
Original file line number Diff line number Diff line change
Expand Up @@ -256,34 +256,37 @@ namespace Impl {
template <> \
struct Rotm< \
EXEC_SPACE, \
Kokkos::View<double, LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::View<double*, LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>, \
Kokkos::View<double const, LAYOUT, \
Kokkos::View<const double[5], LAYOUT, \
Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>, \
Kokkos::View<double[5], LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>, \
true, ETI_SPEC_AVAIL> { \
using DXView = \
Kokkos::View<double, LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
using VectorView = \
Kokkos::View<double*, LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>; \
using YView = Kokkos::View<double const, LAYOUT, \
using PView = Kokkos::View<const double[5], LAYOUT, \
Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>; \
using PView = \
Kokkos::View<double[5], LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>; \
\
static void rotm(EXEC_SPACE const& space, DXView const& d1, \
DXView const& d2, DXView const& x1, YView const& y1, \
PView const& param) { \
Kokkos::Profiling::pushRegion("KokkosBlas::nrm1[TPL_ROCBLAS,double]"); \
static void rotm(EXEC_SPACE const& space, VectorView const& X, \
VectorView const& Y, PView const& param) { \
Kokkos::Profiling::pushRegion("KokkosBlas::rotm[TPL_ROCBLAS,double]"); \
rotm_print_specialization<double>(); \
KokkosBlas::Impl::RocBlasSingleton& s = \
KokkosBlas::Impl::RocBlasSingleton::singleton(); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblasSetStream(s.handle, space.hip_stream())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_drotm(s.handle, &a, &b, &c, &s)); \
rocblas_set_stream(s.handle, space.hip_stream())); \
rocblas_pointer_mode pointer_mode; \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_get_pointer_mode(s.handle, &pointer_mode)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(s.handle, rocblas_pointer_mode_device)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_drotm(s.handle, static_cast<int>(X.extent(0)), X.data(), 1, \
Y.data(), 1, param.data())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(s.handle, pointer_mode)); \
Kokkos::Profiling::popRegion(); \
} \
};
Expand All @@ -302,33 +305,37 @@ KOKKOSBLAS1_DROTM_TPL_SPEC_DECL_ROCBLAS(Kokkos::LayoutRight, Kokkos::HIP,
template <> \
struct Rotm< \
EXEC_SPACE, \
Kokkos::View<float, LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>, \
Kokkos::View<float const, LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::View<float*, LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>, \
Kokkos::View<float[5], LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::View<const float[5], LAYOUT, \
Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>, \
true, ETI_SPEC_AVAIL> { \
using DXView = \
Kokkos::View<float, LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
using VectorView = \
Kokkos::View<float*, LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>; \
using YView = Kokkos::View<float const, LAYOUT, \
using PView = Kokkos::View<const float[5], LAYOUT, \
Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>; \
using PView = \
Kokkos::View<float[5], LAYOUT, Kokkos::Device<EXEC_SPACE, MEM_SPACE>, \
Kokkos::MemoryTraits<Kokkos::Unmanaged>>; \
\
static void rotm(EXEC_SPACE const& space, DXView const& d1, \
DXView const& d2, DXView const& x1, YView const& y1, \
PView const& param) { \
Kokkos::Profiling::pushRegion("KokkosBlas::nrm1[TPL_ROCBLAS,float]"); \
static void rotm(EXEC_SPACE const& space, VectorView const& X, \
VectorView const& Y, PView const& param) { \
Kokkos::Profiling::pushRegion("KokkosBlas::rotm[TPL_ROCBLAS,float]"); \
rotm_print_specialization<float>(); \
KokkosBlas::Impl::RocBlasSingleton& s = \
KokkosBlas::Impl::RocBlasSingleton::singleton(); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblasSetStream(s.handle, space.hip_stream())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_srotm(s.handle, &a, &b, &c, &s)); \
rocblas_set_stream(s.handle, space.hip_stream())); \
rocblas_pointer_mode pointer_mode; \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_get_pointer_mode(s.handle, &pointer_mode)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(s.handle, rocblas_pointer_mode_device)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_srotm(s.handle, static_cast<int>(X.extent(0)), X.data(), 1, \
Y.data(), 1, param.data())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(s.handle, pointer_mode)); \
Kokkos::Profiling::popRegion(); \
} \
};
Expand Down
30 changes: 24 additions & 6 deletions blas/tpls/KokkosBlas1_rotmg_tpl_spec_decl.hpp
Original file line number Diff line number Diff line change
Expand Up @@ -296,13 +296,22 @@ namespace Impl {
static void rotmg(EXEC_SPACE const& space, DXView const& d1, \
DXView const& d2, DXView const& x1, YView const& y1, \
PView const& param) { \
Kokkos::Profiling::pushRegion("KokkosBlas::nrm1[TPL_ROCBLAS,double]"); \
Kokkos::Profiling::pushRegion("KokkosBlas::rotmg[TPL_ROCBLAS,double]"); \
rotmg_print_specialization<double>(); \
KokkosBlas::Impl::RocBlasSingleton& s = \
KokkosBlas::Impl::RocBlasSingleton::singleton(); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblasSetStream(s.handle, space.hip_stream())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_drotmg(s.handle, &a, &b, &c, &s)); \
rocblas_set_stream(s.handle, space.hip_stream())); \
rocblas_pointer_mode pointer_mode; \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_get_pointer_mode(s.handle, &pointer_mode)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(s.handle, rocblas_pointer_mode_device)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_drotmg(s.handle, d1.data(), \
d2.data(), x1.data(), \
y1.data(), param.data())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(s.handle, pointer_mode)); \
Kokkos::Profiling::popRegion(); \
} \
};
Expand Down Expand Up @@ -341,13 +350,22 @@ KOKKOSBLAS1_DROTMG_TPL_SPEC_DECL_ROCBLAS(Kokkos::LayoutRight, Kokkos::HIP,
static void rotmg(EXEC_SPACE const& space, DXView const& d1, \
DXView const& d2, DXView const& x1, YView const& y1, \
PView const& param) { \
Kokkos::Profiling::pushRegion("KokkosBlas::nrm1[TPL_ROCBLAS,float]"); \
Kokkos::Profiling::pushRegion("KokkosBlas::rotmg[TPL_ROCBLAS,float]"); \
rotmg_print_specialization<float>(); \
KokkosBlas::Impl::RocBlasSingleton& s = \
KokkosBlas::Impl::RocBlasSingleton::singleton(); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblasSetStream(s.handle, space.hip_stream())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_srotmg(s.handle, &a, &b, &c, &s)); \
rocblas_set_stream(s.handle, space.hip_stream())); \
rocblas_pointer_mode pointer_mode; \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_get_pointer_mode(s.handle, &pointer_mode)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(s.handle, rocblas_pointer_mode_device)); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL(rocblas_srotmg(s.handle, d1.data(), \
d2.data(), x1.data(), \
y1.data(), param.data())); \
KOKKOS_ROCBLAS_SAFE_CALL_IMPL( \
rocblas_set_pointer_mode(s.handle, pointer_mode)); \
Kokkos::Profiling::popRegion(); \
} \
};
Expand Down