diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2018-03-03 16:37:31 +0100 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2018-03-03 16:37:31 +0100 |
commit | a1cedf36e357f0ce19eba67e1e031c3fd2647fae (patch) | |
tree | 629ebcf46797f1e6be7b6deaf8b3776c5a49106e /src/tuning/kernels/copy_fast.hpp | |
parent | 269bddbf34e5cad00f3845d1a68974420997a040 (diff) |
Separate kernel tuners in .cpp with main and .hpp with settings
Diffstat (limited to 'src/tuning/kernels/copy_fast.hpp')
-rw-r--r-- | src/tuning/kernels/copy_fast.hpp | 93 |
1 files changed, 93 insertions, 0 deletions
diff --git a/src/tuning/kernels/copy_fast.hpp b/src/tuning/kernels/copy_fast.hpp new file mode 100644 index 00000000..eab1c7dd --- /dev/null +++ b/src/tuning/kernels/copy_fast.hpp @@ -0,0 +1,93 @@ + +// ================================================================================================= +// 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 uses the auto-tuner to tune the copy OpenCL kernels. +// +// ================================================================================================= + +#include <string> +#include <vector> + +#include "utilities/utilities.hpp" +#include "tuning/tuning.hpp" + +namespace clblast { +// ================================================================================================= + +// Settings for this kernel (default command-line arguments) +TunerDefaults GetTunerDefaults(const int) { + auto settings = TunerDefaults(); + settings.options = {kArgM, kArgN, kArgAlpha}; + settings.default_m = 1024; + settings.default_n = 1024; + return settings; +} + +// Settings for this kernel (general) +template <typename T> +TunerSettings GetTunerSettings(const int, const Arguments<T> &args) { + auto settings = TunerSettings(); + + // Identification of the kernel + settings.kernel_family = "copy"; + settings.kernel_name = "CopyMatrixFast"; + settings.sources = +#include "../src/kernels/level3/level3.opencl" +#include "../src/kernels/level3/copy_fast.opencl" + ; + + // Buffer sizes + settings.size_a = args.m * args.n; + settings.size_b = args.m * args.n; + + // Inputs and outputs IDs (X:0, Y:1, A:2, B:3, C:4, temp:5) + settings.inputs = {2, 3}; + settings.outputs = {3}; + + // Sets the base thread configuration + settings.global_size = {args.m, args.n}; + settings.global_size_ref = settings.global_size; + settings.local_size = {1, 1}; + settings.local_size_ref = {8, 8}; + + // Transforms the thread configuration based on the parameters + settings.mul_local = {{"COPY_DIMX", "COPY_DIMY"}}; + settings.div_global = {{"COPY_VW", "COPY_WPT"}}; + + // Sets the tuning parameters and their possible values + settings.parameters = { + {"COPY_DIMX", {8, 16, 32}}, + {"COPY_DIMY", {8, 16, 32}}, + {"COPY_WPT", {1, 2, 4, 8}}, + {"COPY_VW", {1, 2, 4, 8}}, + }; + + // Describes how to compute the performance metrics + settings.metric_amount = 2 * args.m * args.n * GetBytes(args.precision); + settings.performance_unit = "GB/s"; + + return settings; +} + +// Tests for valid arguments +template <typename T> +void TestValidArguments(const int, const Arguments<T> &) { } +std::vector<Constraint> SetConstraints(const int) { return {}; } + +// Sets the kernel's arguments +template <typename T> +void SetArguments(const int, Kernel &kernel, const Arguments<T> &args, std::vector<Buffer<T>>& buffers) { + kernel.SetArgument(0, static_cast<int>(args.m)); + kernel.SetArgument(1, buffers[2]()); // 2 == A matrix + kernel.SetArgument(2, buffers[3]()); // 3 == B matrix + kernel.SetArgument(3, GetRealArg(args.alpha)); +} + +// ================================================================================================= +} // namespace clblast |