diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2016-04-27 18:07:30 +0200 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2016-04-27 18:07:30 +0200 |
commit | d7ddbdeb1f416f56bc469d16c051551207274703 (patch) | |
tree | 2f5ba3abc5a97509b84ecdd1fdf5b449ab543eec /include/internal/routines | |
parent | 13eed1a0f973ff2090062a1ad4485896b22949b0 (diff) |
Added non-absolute counter-parts xSUM and IxMAX of the BLAS routines xASUM and IxAMAX
Diffstat (limited to 'include/internal/routines')
-rw-r--r-- | include/internal/routines/level1/xmax.h | 49 | ||||
-rw-r--r-- | include/internal/routines/level1/xsum.h | 49 |
2 files changed, 98 insertions, 0 deletions
diff --git a/include/internal/routines/level1/xmax.h b/include/internal/routines/level1/xmax.h new file mode 100644 index 00000000..860a043b --- /dev/null +++ b/include/internal/routines/level1/xmax.h @@ -0,0 +1,49 @@ + +// ================================================================================================= +// 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 Xmax routine. The precision is implemented using a template argument. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XMAX_H_ +#define CLBLAST_ROUTINES_XMAX_H_ + +#include "internal/routine.h" +#include "internal/routines/level1/xamax.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T> +class Xmax: public Xamax<T> { + public: + + // Members and methods from the base class + using Xamax<T>::DoAmax; + + // Constructor + Xmax(Queue &queue, EventPointer event, const std::string &name = "MAX"): + Xamax<T>(queue, event, name) { + } + + // Forwards to the regular absolute version. The implementation difference is realised in the + // kernel through a pre-processor macro based on the name of the routine. + StatusCode DoMax(const size_t n, + const Buffer<T> &imax_buffer, const size_t imax_offset, + const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc) { + return DoAmax(n, imax_buffer, imax_offset, x_buffer, x_offset, x_inc); + } +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XMAX_H_ +#endif diff --git a/include/internal/routines/level1/xsum.h b/include/internal/routines/level1/xsum.h new file mode 100644 index 00000000..2f633b52 --- /dev/null +++ b/include/internal/routines/level1/xsum.h @@ -0,0 +1,49 @@ + +// ================================================================================================= +// 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 Xsum routine. The precision is implemented using a template argument. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XSUM_H_ +#define CLBLAST_ROUTINES_XSUM_H_ + +#include "internal/routine.h" +#include "internal/routines/level1/xasum.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T> +class Xsum: public Xasum<T> { + public: + + // Members and methods from the base class + using Xasum<T>::DoAsum; + + // Constructor + Xsum(Queue &queue, EventPointer event, const std::string &name = "SUM"): + Xasum<T>(queue, event, name) { + } + + // Forwards to the regular absolute version. The implementation difference is realised in the + // kernel through a pre-processor macro based on the name of the routine. + StatusCode DoSum(const size_t n, + const Buffer<T> &sum_buffer, const size_t sum_offset, + const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc) { + return DoAsum(n, sum_buffer, sum_offset, x_buffer, x_offset, x_inc); + } +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XSUM_H_ +#endif |