diff options
author | CNugteren <web@cedricnugteren.nl> | 2015-09-19 18:01:19 +0200 |
---|---|---|
committer | CNugteren <web@cedricnugteren.nl> | 2015-09-19 18:01:19 +0200 |
commit | de6547a92b16bd2220a103aa6606a49eb7231301 (patch) | |
tree | 04de3d0d8a0b1b3c70d8fc37cff266b0e8948ed6 /src/routines/level2 | |
parent | 80da67d28bbcff958071befb48ccacac05ebbe49 (diff) |
Added SBMV and SPMV routines
Diffstat (limited to 'src/routines/level2')
-rw-r--r-- | src/routines/level2/xsbmv.cc | 64 | ||||
-rw-r--r-- | src/routines/level2/xspmv.cc | 64 |
2 files changed, 128 insertions, 0 deletions
diff --git a/src/routines/level2/xsbmv.cc b/src/routines/level2/xsbmv.cc new file mode 100644 index 00000000..457bd762 --- /dev/null +++ b/src/routines/level2/xsbmv.cc @@ -0,0 +1,64 @@ + +// ================================================================================================= +// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This +// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max- +// width of 100 characters per line. +// +// Author(s): +// Cedric Nugteren <www.cedricnugteren.nl> +// +// This file implements the Xsbmv class (see the header for information about the class). +// +// ================================================================================================= + +#include "internal/routines/level2/xsbmv.h" + +#include <string> +#include <vector> + +namespace clblast { +// ================================================================================================= + +// Constructor: forwards to base class constructor +template <typename T> +Xsbmv<T>::Xsbmv(Queue &queue, Event &event, const std::string &name): + Xgemv<T>(queue, event, name) { +} + +// ================================================================================================= + +// The main routine +template <typename T> +StatusCode Xsbmv<T>::DoSbmv(const Layout layout, const Triangle triangle, + const size_t n, const size_t k, + const T alpha, + const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld, + const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc, + const T beta, + const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc) { + + // The data is either in the upper or lower triangle + size_t is_upper = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) || + (triangle == Triangle::kLower && layout == Layout::kRowMajor)); + + // Runs the generic matrix-vector multiplication, disabling the use of fast vectorized kernels. + // The specific symmetric banded matrix-accesses are implemented in the kernel guarded by the + // ROUTINE_SBMV define. + bool fast_kernels = false; + return MatVec(layout, Transpose::kNo, + n, n, alpha, + a_buffer, a_offset, a_ld, + x_buffer, x_offset, x_inc, beta, + y_buffer, y_offset, y_inc, + fast_kernels, fast_kernels, + is_upper, false, k, 0); +} + +// ================================================================================================= + +// Compiles the templated class +template class Xsbmv<float>; +template class Xsbmv<double>; + +// ================================================================================================= +} // namespace clblast diff --git a/src/routines/level2/xspmv.cc b/src/routines/level2/xspmv.cc new file mode 100644 index 00000000..4f1a9c61 --- /dev/null +++ b/src/routines/level2/xspmv.cc @@ -0,0 +1,64 @@ + +// ================================================================================================= +// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This +// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max- +// width of 100 characters per line. +// +// Author(s): +// Cedric Nugteren <www.cedricnugteren.nl> +// +// This file implements the Xspmv class (see the header for information about the class). +// +// ================================================================================================= + +#include "internal/routines/level2/xspmv.h" + +#include <string> +#include <vector> + +namespace clblast { +// ================================================================================================= + +// Constructor: forwards to base class constructor +template <typename T> +Xspmv<T>::Xspmv(Queue &queue, Event &event, const std::string &name): + Xgemv<T>(queue, event, name) { +} + +// ================================================================================================= + +// The main routine +template <typename T> +StatusCode Xspmv<T>::DoSpmv(const Layout layout, const Triangle triangle, + const size_t n, + const T alpha, + const Buffer<T> &ap_buffer, const size_t ap_offset, + const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc, + const T beta, + const Buffer<T> &y_buffer, const size_t y_offset, const size_t y_inc) { + + // The data is either in the upper or lower triangle + size_t is_upper = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) || + (triangle == Triangle::kLower && layout == Layout::kRowMajor)); + + // Runs the generic matrix-vector multiplication, disabling the use of fast vectorized kernels. + // The specific symmetric packed matrix-accesses are implemented in the kernel guarded by the + // ROUTINE_SPMV define. + bool fast_kernels = false; + return MatVec(layout, Transpose::kNo, + n, n, alpha, + ap_buffer, ap_offset, n, + x_buffer, x_offset, x_inc, beta, + y_buffer, y_offset, y_inc, + fast_kernels, fast_kernels, + is_upper, true, 0, 0); +} + +// ================================================================================================= + +// Compiles the templated class +template class Xspmv<float>; +template class Xspmv<double>; + +// ================================================================================================= +} // namespace clblast |