From 16a048f1ac8102ad4bcce9cf9fc320f791323e45 Mon Sep 17 00:00:00 2001 From: cnugteren Date: Wed, 20 Apr 2016 22:12:51 -0600 Subject: Added support for the iSAMAX/iDAMAX/iCAMAX/iZAMAX routines --- test/routines/level1/xamax.h | 139 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 139 insertions(+) create mode 100644 test/routines/level1/xamax.h (limited to 'test/routines') diff --git a/test/routines/level1/xamax.h b/test/routines/level1/xamax.h new file mode 100644 index 00000000..7b404dc3 --- /dev/null +++ b/test/routines/level1/xamax.h @@ -0,0 +1,139 @@ + +// ================================================================================================= +// 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 +// +// This file implements a class with static methods to describe the Xamax 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_XAMAX_H_ +#define CLBLAST_TEST_ROUTINES_XAMAX_H_ + +#include +#include + +#ifdef CLBLAST_REF_CLBLAS + #include "wrapper_clblas.h" +#endif +#ifdef CLBLAST_REF_CBLAS + #include "wrapper_cblas.h" +#endif + +namespace clblast { +// ================================================================================================= + +// See comment at top of file for a description of the class +template +class TestXamax { + public: + + // The BLAS level: 1, 2, or 3 + static size_t BLASLevel() { return 1; } + + // The list of arguments relevant for this routine + static std::vector GetOptions() { + return {kArgN, + kArgXInc, + kArgXOffset, kArgImaxOffset}; + } + + // Describes how to obtain the sizes of the buffers + static size_t GetSizeX(const Arguments &args) { + return args.n * args.x_inc + args.x_offset; + } + static size_t GetSizeImax(const Arguments &args) { + return 1 + args.imax_offset; + } + + // Describes how to set the sizes of all the buffers + static void SetSizes(Arguments &args) { + args.x_size = GetSizeX(args); + args.scalar_size = GetSizeImax(args); + } + + // Describes what the default values of the leading dimensions of the matrices are + static size_t DefaultLDA(const Arguments &) { return 1; } // N/A for this routine + static size_t DefaultLDB(const Arguments &) { return 1; } // N/A for this routine + static size_t DefaultLDC(const Arguments &) { return 1; } // N/A for this routine + + // Describes which transpose options are relevant for this routine + using Transposes = std::vector; + 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 &args, Buffers &buffers, Queue &queue) { + auto queue_plain = queue(); + auto event = cl_event{}; + auto status = Amax(args.n, + buffers.scalar(), args.imax_offset, + buffers.x_vec(), args.x_offset, args.x_inc, + &queue_plain, &event); + clWaitForEvents(1, &event); + return status; + } + + // Describes how to run the clBLAS routine (for correctness/performance comparison) + #ifdef CLBLAST_REF_CLBLAS + static StatusCode RunReference1(const Arguments &args, Buffers &buffers, Queue &queue) { + auto queue_plain = queue(); + auto event = cl_event{}; + auto status = clblasXamax(args.n, + buffers.scalar(), args.imax_offset, + buffers.x_vec(), args.x_offset, args.x_inc, + 1, &queue_plain, 0, nullptr, &event); + clWaitForEvents(1, &event); + return static_cast(status); + } + #endif + + // Describes how to run the CPU BLAS routine (for correctness/performance comparison) + #ifdef CLBLAST_REF_CBLAS + static StatusCode RunReference2(const Arguments &args, Buffers &buffers, Queue &queue) { + std::vector scalar_cpu(args.scalar_size, static_cast(0)); + std::vector x_vec_cpu(args.x_size, static_cast(0)); + buffers.scalar.Read(queue, args.scalar_size, scalar_cpu); + buffers.x_vec.Read(queue, args.x_size, x_vec_cpu); + cblasXamax(args.n, + scalar_cpu, args.imax_offset, + x_vec_cpu, args.x_offset, args.x_inc); + buffers.scalar.Write(queue, args.scalar_size, scalar_cpu); + return StatusCode::kSuccess; + } + #endif + + // Describes how to download the results of the computation (more importantly: which buffer) + static std::vector DownloadResult(const Arguments &args, Buffers &buffers, Queue &queue) { + std::vector result(args.scalar_size, static_cast(0)); + buffers.scalar.Read(queue, args.scalar_size, result); + return result; + } + + // Describes how to compute the indices of the result buffer + static size_t ResultID1(const Arguments &) { return 1; } // N/A for this routine + static size_t ResultID2(const Arguments &) { return 1; } // N/A for this routine + static size_t GetResultIndex(const Arguments &args, const size_t, const size_t) { + return args.imax_offset; + } + + // Describes how to compute performance metrics + static size_t GetFlops(const Arguments &args) { + return args.n; + } + static size_t GetBytes(const Arguments &args) { + return ((args.n) + 1) * sizeof(T); + } +}; + +// ================================================================================================= +} // namespace clblast + +// CLBLAST_TEST_ROUTINES_XAMAX_H_ +#endif -- cgit v1.2.3