diff options
author | cnugteren <web@cedricnugteren.nl> | 2016-04-14 19:58:26 -0600 |
---|---|---|
committer | cnugteren <web@cedricnugteren.nl> | 2016-04-14 19:58:26 -0600 |
commit | 8be99de82d2ff0634c1289d9b4d1785364a68a44 (patch) | |
tree | 27c16eb24784bed190ca75fe51abf5953e3b0d6a /include | |
parent | e0497807e297e38884efae67a0109a160dc693b7 (diff) |
Added support for the SASUM/DASUM/ScASUM/DzASUM routines
Diffstat (limited to 'include')
-rw-r--r-- | include/internal/routines/level1/xasum.h | 56 | ||||
-rw-r--r-- | include/internal/utilities.h | 2 |
2 files changed, 58 insertions, 0 deletions
diff --git a/include/internal/routines/level1/xasum.h b/include/internal/routines/level1/xasum.h new file mode 100644 index 00000000..b6e5d2cd --- /dev/null +++ b/include/internal/routines/level1/xasum.h @@ -0,0 +1,56 @@ + +// ================================================================================================= +// 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 Xasum routine. The precision is implemented using a template argument. +// +// ================================================================================================= + +#ifndef CLBLAST_ROUTINES_XASUM_H_ +#define CLBLAST_ROUTINES_XASUM_H_ + +#include "internal/routine.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T> +class Xasum: public Routine<T> { + public: + + // Members and methods from the base class + using Routine<T>::db_; + using Routine<T>::source_string_; + using Routine<T>::queue_; + using Routine<T>::event_; + using Routine<T>::context_; + using Routine<T>::GetProgramFromCache; + using Routine<T>::TestVectorX; + using Routine<T>::TestVectorDot; + using Routine<T>::RunKernel; + using Routine<T>::ErrorIn; + + // Constructor + Xasum(Queue &queue, EventPointer event, const std::string &name = "ASUM"); + + // Templated-precision implementation of the routine + StatusCode DoAsum(const size_t n, + const Buffer<T> &asum_buffer, const size_t asum_offset, + const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc); + + private: + // Static variable to get the precision + const static Precision precision_; +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_ROUTINES_XASUM_H_ +#endif diff --git a/include/internal/utilities.h b/include/internal/utilities.h index 6adc1d0a..8ffdc3af 100644 --- a/include/internal/utilities.h +++ b/include/internal/utilities.h @@ -65,6 +65,7 @@ constexpr auto kArgCOffset = "offc"; constexpr auto kArgAPOffset = "offap"; constexpr auto kArgDotOffset = "offdot"; constexpr auto kArgNrm2Offset = "offnrm2"; +constexpr auto kArgAsumOffset = "offasum"; constexpr auto kArgAlpha = "alpha"; constexpr auto kArgBeta = "beta"; @@ -119,6 +120,7 @@ struct Arguments { size_t ap_offset = 0; size_t dot_offset = 0; size_t nrm2_offset = 0; + size_t asum_offset = 0; T alpha = T{1.0}; T beta = T{1.0}; size_t x_size = 1; |