diff options
author | CNugteren <web@cedricnugteren.nl> | 2015-06-21 12:44:03 +0200 |
---|---|---|
committer | CNugteren <web@cedricnugteren.nl> | 2015-06-21 12:44:03 +0200 |
commit | e3829c1067814c0aa83ab440fa431d98837aeeda (patch) | |
tree | 18516e3bc42bc71746312ea9efbcd677d5d71e2b /src/clblast.cc | |
parent | ea7da6a49758af50302be040ab7a97a7a8c0f692 (diff) |
Added prototypes of SYRK and SYR2K
Diffstat (limited to 'src/clblast.cc')
-rw-r--r-- | src/clblast.cc | 122 |
1 files changed, 121 insertions, 1 deletions
diff --git a/src/clblast.cc b/src/clblast.cc index bb0091a3..e0d085a9 100644 --- a/src/clblast.cc +++ b/src/clblast.cc @@ -209,7 +209,7 @@ StatusCode Symm(const Layout layout, const Side side, const Triangle triangle, std::string kernel_source = #include "kernels/xgemm.opencl" auto status = routine.SetUp(common_source1 + common_source2 + common_source3 + common_source4 + - kernel_source); + kernel_source); if (status != StatusCode::kSuccess) { return status; } // Runs the routine @@ -244,4 +244,124 @@ template StatusCode Symm<double2>(const Layout, const Side, const Triangle, cl_command_queue*, cl_event*); // ================================================================================================= + +// SYRK +template <typename T> +StatusCode Syrk(const Layout layout, const Triangle triangle, const Transpose transpose_a, + const size_t n, const size_t k, const T alpha, + const cl_mem a_buffer, const size_t a_offset, const size_t a_ld, const T beta, + cl_mem c_buffer, const size_t c_offset, const size_t c_ld, + cl_command_queue* queue, cl_event* event) { + auto queue_cpp = CommandQueue(*queue); + auto event_cpp = Event(*event); + /* + auto routine = Xsyrk<T>(queue_cpp, event_cpp); + + // Loads the kernel source-code as an include (C++11 raw string literal) + std::string common_source1 = + #include "kernels/copy.opencl" + std::string common_source2 = + #include "kernels/pad.opencl" + std::string common_source3 = + #include "kernels/transpose.opencl" + std::string common_source4 = + #include "kernels/padtranspose.opencl" + std::string kernel_source = + #include "kernels/xgemm.opencl" + auto status = routine.SetUp(common_source1 + common_source2 + common_source3 + common_source4 + + kernel_source); + if (status != StatusCode::kSuccess) { return status; } + + // Runs the routine + return routine.DoSyrk(layout, triangle, transpose_a, n, k, alpha, + Buffer(a_buffer), a_offset, a_ld, beta, + Buffer(c_buffer), c_offset, c_ld); + */ + return StatusCode::kSuccess; +} +template StatusCode Syrk<float>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, const float, + const cl_mem, const size_t, const size_t, const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode Syrk<double>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, const double, + const cl_mem, const size_t, const size_t, const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode Syrk<float2>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, const float2, + const cl_mem, const size_t, const size_t, const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode Syrk<double2>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, const double2, + const cl_mem, const size_t, const size_t, const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); + +// ================================================================================================= + +// SYR2K +template <typename T> +StatusCode Syr2k(const Layout layout, const Triangle triangle, const Transpose transpose_ab, + const size_t n, const size_t k, const T alpha, + const cl_mem a_buffer, const size_t a_offset, const size_t a_ld, + const cl_mem b_buffer, const size_t b_offset, const size_t b_ld, const T beta, + cl_mem c_buffer, const size_t c_offset, const size_t c_ld, + cl_command_queue* queue, cl_event* event) { + auto queue_cpp = CommandQueue(*queue); + auto event_cpp = Event(*event); + /* + auto routine = Xsyr2k<T>(queue_cpp, event_cpp); + + // Loads the kernel source-code as an include (C++11 raw string literal) + std::string common_source1 = + #include "kernels/copy.opencl" + std::string common_source2 = + #include "kernels/pad.opencl" + std::string common_source3 = + #include "kernels/transpose.opencl" + std::string common_source4 = + #include "kernels/padtranspose.opencl" + std::string kernel_source = + #include "kernels/xgemm.opencl" + auto status = routine.SetUp(common_source1 + common_source2 + common_source3 + common_source4 + + kernel_source); + if (status != StatusCode::kSuccess) { return status; } + + // Runs the routine + return routine.DoSyr2k(layout, triangle, transpose_ab, n, k, alpha, + Buffer(a_buffer), a_offset, a_ld, + Buffer(b_buffer), b_offset, b_ld, beta, + Buffer(c_buffer), c_offset, c_ld); + */ + return StatusCode::kSuccess; +} +template StatusCode Syr2k<float>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, const float, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, const float, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode Syr2k<double>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, const double, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, const double, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode Syr2k<float2>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode Syr2k<double2>(const Layout, const Triangle, const Transpose, + const size_t, const size_t, const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); + +// ================================================================================================= } // namespace clblast |