diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2016-06-16 18:07:46 +0200 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2016-06-16 18:07:46 +0200 |
commit | 52ccaf5b25e14c9ce032315e5e96b1f27886d481 (patch) | |
tree | 087288b7aebf2a06ffc4e7dcbcd4353f7a3be6a7 /src/clblast.cc | |
parent | 39b7dbc5e37829abfbcfb77852b9138b31540b42 (diff) |
Added XOMATCOPY routines to perform out-of-place matrix scaling, copying, and/or transposing
Diffstat (limited to 'src/clblast.cc')
-rw-r--r-- | src/clblast.cc | 56 |
1 files changed, 56 insertions, 0 deletions
diff --git a/src/clblast.cc b/src/clblast.cc index 07322327..e3df6ede 100644 --- a/src/clblast.cc +++ b/src/clblast.cc @@ -68,6 +68,9 @@ #include "internal/routines/level3/xher2k.h" #include "internal/routines/level3/xtrmm.h" +// Extra includes (level-x) +#include "internal/routines/levelx/xomatcopy.h" + namespace clblast { // ================================================================================================= @@ -2062,6 +2065,59 @@ template StatusCode PUBLIC_API Trsm<half>(const Layout, const Side, const Triang cl_command_queue*, cl_event*); // ================================================================================================= +// Extra non-BLAS routines (level-X) +// ================================================================================================= + +// Scaling and out-place transpose/copy (non-BLAS function): SOMATCOPY/DOMATCOPY/COMATCOPY/ZOMATCOPY/HOMATCOPY +template <typename T> +StatusCode Omatcopy(const Layout layout, const Transpose a_transpose, + const size_t m, const size_t n, + const T alpha, + const cl_mem a_buffer, const size_t a_offset, const size_t a_ld, + cl_mem b_buffer, const size_t b_offset, const size_t b_ld, + cl_command_queue* queue, cl_event* event) { + auto queue_cpp = Queue(*queue); + auto routine = Xomatcopy<T>(queue_cpp, event); + auto status = routine.SetUp(); + if (status != StatusCode::kSuccess) { return status; } + return routine.DoOmatcopy(layout, a_transpose, + m, n, + alpha, + Buffer<T>(a_buffer), a_offset, a_ld, + Buffer<T>(b_buffer), b_offset, b_ld); +} +template StatusCode PUBLIC_API Omatcopy<float>(const Layout, const Transpose, + const size_t, const size_t, + const float, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Omatcopy<double>(const Layout, const Transpose, + const size_t, const size_t, + const double, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Omatcopy<float2>(const Layout, const Transpose, + const size_t, const size_t, + const float2, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Omatcopy<double2>(const Layout, const Transpose, + const size_t, const size_t, + const double2, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode PUBLIC_API Omatcopy<half>(const Layout, const Transpose, + const size_t, const size_t, + const half, + const cl_mem, const size_t, const size_t, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); + +// ================================================================================================= // Clears the cache of stored binaries StatusCode ClearCache() { return cache::ClearCache(); } |