diff options
Diffstat (limited to 'test/routines/level2/xhpr.h')
-rw-r--r-- | test/routines/level2/xhpr.h | 122 |
1 files changed, 122 insertions, 0 deletions
diff --git a/test/routines/level2/xhpr.h b/test/routines/level2/xhpr.h new file mode 100644 index 00000000..03599ddc --- /dev/null +++ b/test/routines/level2/xhpr.h @@ -0,0 +1,122 @@ + +// ================================================================================================= +// 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 a class with static methods to describe the Xhpr routine. Examples of +// such 'descriptions' are how to calculate the size a of buffer or how to run the routine. These +// static methods are used by the correctness tester and the performance tester. +// +// ================================================================================================= + +#ifndef CLBLAST_TEST_ROUTINES_XHPR_H_ +#define CLBLAST_TEST_ROUTINES_XHPR_H_ + +#include <vector> +#include <string> + +#include "wrapper_clblas.h" + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template <typename T, typename U> +class TestXhpr { + public: + + // The BLAS level: 1, 2, or 3 + static size_t BLASLevel() { return 2; } + + // The list of arguments relevant for this routine + static std::vector<std::string> GetOptions() { + return {kArgN, + kArgLayout, kArgTriangle, + kArgXInc, + kArgAPOffset, kArgXOffset, + kArgAlpha}; + } + + // Describes how to obtain the sizes of the buffers + static size_t GetSizeX(const Arguments<U> &args) { + return args.n * args.x_inc + args.x_offset; + } + static size_t GetSizeAP(const Arguments<U> &args) { + return ((args.n*(args.n+1)) / 2) + args.ap_offset; + } + + // Describes how to set the sizes of all the buffers + static void SetSizes(Arguments<U> &args) { + args.ap_size = GetSizeAP(args); + args.x_size = GetSizeX(args); + } + + // Describes what the default values of the leading dimensions of the matrices are + static size_t DefaultLDA(const Arguments<U> &args) { return args.n; } + static size_t DefaultLDB(const Arguments<U> &) { return 1; } // N/A for this routine + static size_t DefaultLDC(const Arguments<U> &) { return 1; } // N/A for this routine + + // Describes which transpose options are relevant for this routine + using Transposes = std::vector<Transpose>; + static Transposes GetATransposes(const Transposes &) { return {}; } // N/A for this routine + static Transposes GetBTransposes(const Transposes &) { return {}; } // N/A for this routine + + // Describes how to run the CLBlast routine + static StatusCode RunRoutine(const Arguments<U> &args, const Buffers<T> &buffers, Queue &queue) { + auto queue_plain = queue(); + auto event = cl_event{}; + auto status = Hpr(args.layout, args.triangle, + args.n, args.alpha, + buffers.x_vec(), args.x_offset, args.x_inc, + buffers.ap_mat(), args.ap_offset, + &queue_plain, &event); + clWaitForEvents(1, &event); + return status; + } + + // Describes how to run the clBLAS routine (for correctness/performance comparison) + static StatusCode RunReference(const Arguments<U> &args, const Buffers<T> &buffers, Queue &queue) { + auto queue_plain = queue(); + auto event = cl_event{}; + auto status = clblasXhpr(static_cast<clblasOrder>(args.layout), + static_cast<clblasUplo>(args.triangle), + args.n, args.alpha, + buffers.x_vec(), args.x_offset, args.x_inc, + buffers.ap_mat(), args.ap_offset, + 1, &queue_plain, 0, nullptr, &event); + clWaitForEvents(1, &event); + return static_cast<StatusCode>(status); + } + + // Describes how to download the results of the computation (more importantly: which buffer) + static std::vector<T> DownloadResult(const Arguments<U> &args, Buffers<T> &buffers, Queue &queue) { + std::vector<T> result(args.ap_size, static_cast<T>(0)); + buffers.ap_mat.Read(queue, args.ap_size, result); + return result; + } + + // Describes how to compute the indices of the result buffer + static size_t ResultID1(const Arguments<U> &args) { return args.ap_size - args.ap_offset; } + static size_t ResultID2(const Arguments<U> &) { return 1; } // N/A for this routine + static size_t GetResultIndex(const Arguments<U> &args, const size_t id1, const size_t) { + return id1 + args.ap_offset; + } + + // Describes how to compute performance metrics + static size_t GetFlops(const Arguments<U> &args) { + return 3 * ((args.n*(args.n+1)) / 2); + } + static size_t GetBytes(const Arguments<U> &args) { + return ((args.n*(args.n+1)) + args.n) * sizeof(T); + } +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_TEST_ROUTINES_XHPR_H_ +#endif |