not_equal.hpp

//=======================================================================
// Copyright (c) 2014-2023 Baptiste Wicht
// Distributed under the terms of the MIT License.
// (See accompanying file LICENSE or copy at
//  http://opensource.org/licenses/MIT)
//=======================================================================

#pragma once

#include "etl/impl/egblas/not_equal.hpp"

namespace etl {

template <typename T>
struct not_equal_binary_op {
    static constexpr bool linear      = true;  
    static constexpr bool thread_safe = true;  
    static constexpr bool desc_func   = false; 

    template <vector_mode_t V>
    static constexpr bool vectorizable = false;

    template <typename L, typename R>
    static constexpr bool gpu_computable = (all_single_precision<L, R> && impl::egblas::has_snot_equal)
                                           || (all_double_precision<L, R> && impl::egblas::has_dnot_equal)
                                           || (all_complex_single_precision<L, R> && impl::egblas::has_cnot_equal)
                                           || (all_complex_double_precision<L, R> && impl::egblas::has_znot_equal);

    static constexpr int complexity() {
        return 1;
    }

    static constexpr bool apply(const T& lhs, const T& rhs) noexcept {
        return lhs != rhs;
    }

    template <typename X, typename Y, typename YY>
    static auto gpu_compute_hint(const X& x, const Y& y, YY& yy) noexcept {
        decltype(auto) t1 = smart_gpu_compute_hint(x, yy);
        decltype(auto) t2 = smart_gpu_compute_hint(y, yy);

        constexpr size_t inca = gpu_inc<decltype(x)>;
        constexpr size_t incb = gpu_inc<decltype(y)>;

        auto t3 = force_temporary_gpu_dim_only_t<bool>(t1);

        impl::egblas::not_equal(etl::size(yy), t1.gpu_memory(), inca, t2.gpu_memory(), incb, t3.gpu_memory(), 1);

        return t3;
    }

    template <typename X, typename Y, typename YY>
    static YY& gpu_compute(const X& x, const Y& y, YY& yy) noexcept {
        decltype(auto) t1 = smart_gpu_compute_hint(x, yy);
        decltype(auto) t2 = smart_gpu_compute_hint(y, yy);

        constexpr size_t inca = gpu_inc<decltype(x)>;
        constexpr size_t incb = gpu_inc<decltype(y)>;

        impl::egblas::not_equal(etl::size(yy), t1.gpu_memory(), inca, t2.gpu_memory(), incb, yy.gpu_memory(), 1);

        yy.validate_gpu();
        yy.invalidate_cpu();

        return yy;
    }

    static std::string desc() noexcept {
        return "!=";
    }
};

} //end of namespace etl