7f4761ae52
This allows each model to initialise their arrays with a parallel approach, which yields the first touch required for good performance on NUMA architectures.
64 lines
1.5 KiB
C++
64 lines
1.5 KiB
C++
|
|
// Copyright (c) 2015-16 Tom Deakin, Simon McIntosh-Smith,
|
|
// University of Bristol HPC
|
|
//
|
|
// For full license terms please see the LICENSE file distributed with this
|
|
// source code
|
|
|
|
#pragma once
|
|
|
|
#include <iostream>
|
|
#include <sstream>
|
|
#include <stdexcept>
|
|
|
|
#define CL_HPP_ENABLE_EXCEPTIONS
|
|
#define CL_HPP_TARGET_OPENCL_VERSION 120
|
|
#define CL_HPP_MINIMUM_OPENCL_VERSION 120
|
|
|
|
#include "CL/cl2.hpp"
|
|
|
|
#include "Stream.h"
|
|
|
|
#define IMPLEMENTATION_STRING "OpenCL"
|
|
|
|
template <class T>
|
|
class OCLStream : public Stream<T>
|
|
{
|
|
protected:
|
|
// Size of arrays
|
|
unsigned int array_size;
|
|
|
|
// Device side pointers to arrays
|
|
cl::Buffer d_a;
|
|
cl::Buffer d_b;
|
|
cl::Buffer d_c;
|
|
|
|
// OpenCL objects
|
|
cl::Device device;
|
|
cl::Context context;
|
|
cl::CommandQueue queue;
|
|
|
|
cl::KernelFunctor<cl::Buffer, cl::Buffer, cl::Buffer, T, T, T> *init_kernel;
|
|
cl::KernelFunctor<cl::Buffer, cl::Buffer> *copy_kernel;
|
|
cl::KernelFunctor<cl::Buffer, cl::Buffer> * mul_kernel;
|
|
cl::KernelFunctor<cl::Buffer, cl::Buffer, cl::Buffer> *add_kernel;
|
|
cl::KernelFunctor<cl::Buffer, cl::Buffer, cl::Buffer> *triad_kernel;
|
|
|
|
public:
|
|
|
|
OCLStream(const unsigned int, const int);
|
|
~OCLStream();
|
|
|
|
virtual void copy() override;
|
|
virtual void add() override;
|
|
virtual void mul() override;
|
|
virtual void triad() override;
|
|
|
|
virtual void init_arrays(T initA, T initB, T initC) override;
|
|
virtual void read_arrays(std::vector<T>& a, std::vector<T>& b, std::vector<T>& c) override;
|
|
|
|
};
|
|
|
|
// Populate the devices list
|
|
void getDeviceList(void);
|