ElementwiseUnaryOperation.h

浏览该文件的文档.

/*
 * Copyright (c) The Shogun Machine Learning Toolbox
 * Written (w) 2015 Soumyajit De
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *
 * 1. Redistributions of source code must retain the above copyright notice, this
 *    list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright notice,
 *    this list of conditions and the following disclaimer in the documentation
 *    and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 * The views and conclusions contained in the software and documentation are those
 * of the authors and should not be interpreted as representing official policies,
 * either expressed or implied, of the Shogun Development Team.
 */

#ifndef ELEMENTWISE_OPERATION_H_
#define ELEMENTWISE_OPERATION_H_

#include <shogun/lib/config.h>
#include <shogun/lib/SGMatrix.h>

#ifdef HAVE_VIENNACL
#include <shogun/lib/GPUMatrix.h>
#include <shogun/mathematics/linalg/internal/opencl_config.h>
#include <shogun/mathematics/linalg/internal/opencl_util.h>
#include <shogun/mathematics/linalg/internal/implementation/operations/opencl_operation.h>
#endif // HAVE_VIENNACL

#include <algorithm>
#include <type_traits>

namespace shogun
{

namespace linalg
{

namespace implementation
{

template <enum Backend, class Operand, class ReturnType, class UnaryOp>
struct elementwise_unary_operation
{
};

template <class Operand, class ReturnType, class UnaryOp>
struct elementwise_unary_operation<Backend::NATIVE, Operand, ReturnType, UnaryOp>
{
    using T = typename Operand::Scalar;

    using ST = typename ReturnType::Scalar;

#ifdef HAVE_VIENNACL

    static_assert(std::is_same<SGMatrix<T>, Operand>::value
            || std::is_same<SGVector<T>, Operand>::value,
            "NATIVE backend not allowed for GPU operands! Use SGMatrix/SGVector "
            "in order to use NATIVE or use VIENNACL backend instead.\n");
#endif // HAVE_VIENNACL

    static void compute(Operand operand, ReturnType result, UnaryOp unary_op)
    {
        static_assert(std::is_same<ST,decltype(unary_op(operand.data()[0]))>::value,
            "The return type of the unary operator and the scalar types of the "
            "result must be the same!\n");

#pragma omp parallel for
        for (decltype(operand.size()) i=0; i<operand.size(); ++i)
            result.data()[i]=unary_op(operand.data()[i]);
    }
};

#ifdef HAVE_EIGEN3

template <class Operand, class ReturnType, class UnaryOp>
struct elementwise_unary_operation<Backend::EIGEN3, Operand, ReturnType, UnaryOp>
{
    using T = typename Operand::Scalar;

    using ST = typename UnaryOp::return_type;

#ifdef HAVE_VIENNACL

    static_assert(std::is_same<SGMatrix<T>, Operand>::value
            || std::is_same<SGVector<T>, Operand>::value,
            "NATIVE backend not allowed for GPU operands! Use SGMatrix/SGVector "
            "in order to use NATIVE or use VIENNACL backend instead.\n");
#endif // HAVE_VIENNACL

    static void compute(Operand operand, ReturnType result, UnaryOp unary_op)
    {
        auto eigen_result=unary_op.compute_using_eigen3(operand);
        std::copy(eigen_result.data(), eigen_result.data()+eigen_result.size(), result.data());
    }
};
#endif // HAVE_EIGEN3

#ifdef HAVE_VIENNACL

template <class Operand, class UnaryOp>
struct elementwise_unary_operation<Backend::VIENNACL, Operand, Operand, UnaryOp>
{
    using T = typename Operand::Scalar;

    static_assert(!std::is_same<T,complex128_t>::value,
            "Complex numbers not supported!\n");

    static_assert(std::is_same<CGPUMatrix<T>, Operand>::value ||
            std::is_same<CGPUVector<T>, Operand>::value,
            "VIENNACL backend not allowed for CPU operands! Use CGPUMatrix/CGPUVector "
            "in order to use VIENNACL or use NATIVE/EIGEN3 backend instead.\n");

    static void compute(Operand operand, Operand result, operations::ocl_operation unary_op)
    {
        const std::string operation=unary_op.get_operation();
        std::hash<std::string> hash_fn;
        const std::string hash=std::to_string(hash_fn(operation));
        const std::string kernel_name="kernel_"+hash+"_"+ocl::get_type_string<T>();

        viennacl::ocl::kernel& kernel=
            ocl::generate_single_arg_elementwise_kernel<T>(kernel_name, operation);

        kernel.global_work_size(0, ocl::align_to_multiple_1d(operand.size()));

        viennacl::ocl::enqueue(kernel(operand.data(),
            cl_int(operand.size()), cl_int(operand.offset),
            result.data(), cl_int(result.offset)));
    }
};
#endif // HAVE_VIENNACL

}

}

}
#endif // ELEMENTWISE_OPERATION_H_