elementwise_trinary.hpp

Go to the documentation of this file.

// Copyright (c) 2024--present, The Simons Foundation
// This file is part of TRIQS/nda and is licensed under the Apache License, Version 2.0.
// SPDX-License-Identifier: Apache-2.0
// See LICENSE in the root of this distribution for details.

 
#pragma once
 
#include "./interface/cutensor_interface.hpp"
#include "./tools.hpp"
#include "../exceptions.hpp"
#include "../mem/address_space.hpp"
#include "../traits.hpp"
 
#include <string_view>
#include <utility>
 
namespace nda::tensor {


  template <BlasArrayOrConj A, BlasArrayOrConjFor<A> B, BlasArrayFor<A> C>
  void elementwise_trinary(get_value_t<A> alpha, A const &a, std::string_view idx_a, get_value_t<A> beta, B const &b, std::string_view idx_b,
                           get_value_t<A> gamma, C &&c, std::string_view idx_c, binary_op op_AB = binary_op::SUM, // NOLINT
                           binary_op op_ABC = binary_op::SUM) {
    // compile-time checks
    constexpr bool run_on_device = mem::have_device_compatible_addr_space<A, B, C>;
    static_assert(!run_on_device || have_cutensor, "nda::tensor::elementwise_trinary: cuTENSOR support is required");
    static_assert(run_on_device || (get_rank<A> == get_rank<B> && get_rank<A> == get_rank<C>),
                  "nda::tensor::elementwise_trinary: host fallback requires identical ranks");
 
    // dispatch to backends
    if constexpr (run_on_device) {
      device::elementwise_trinary(alpha, a, idx_a, beta, b, idx_b, gamma, c, idx_c, c, op_AB, op_ABC);
    } else {
      require_equal_indices(idx_a, idx_b, get_rank<A>, "elementwise_trinary");
      require_equal_indices(idx_b, idx_c, get_rank<A>, "elementwise_trinary");
      c = nda::map([alpha, beta, gamma, op_AB, op_ABC](auto x, auto y, auto z) {
        return detail::apply_binary(op_ABC, detail::apply_binary(op_AB, alpha * x, beta * y), gamma * z);
      })(a, b, c);
    }
  }

  template <BlasArrayOrConj A, BlasArrayOrConjFor<A> B, BlasArrayFor<A> C>
  void elementwise_trinary(A const &a, std::string_view idx_a, B const &b, std::string_view idx_b, C &&c, std::string_view idx_c, // NOLINT
                           binary_op op_AB = binary_op::SUM, binary_op op_ABC = binary_op::SUM) {
    elementwise_trinary(get_value_t<A>{1}, a, idx_a, get_value_t<A>{1}, b, idx_b, get_value_t<A>{0}, std::forward<C>(c), idx_c, op_AB, op_ABC);
  }

 
} // namespace nda::tensor