ungqr.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/cxx_interface.hpp"
#include "../basic_array.hpp"
#include "../blas/tools.hpp"
#include "../concepts.hpp"
#include "../declarations.hpp"
#include "../macros.hpp"
#include "../mem/address_space.hpp"
#include "../traits.hpp"
 
#include <algorithm>
#include <cmath>
#include <complex>
#include <type_traits>
 
namespace nda::lapack {

  template <BlasArrayCplx<2> A, BlasArrayFor<A, 1> TAU, BlasArrayFor<A, 1> W = vector_value_t<A>>
    requires(has_F_layout<A>)
  int ungqr(A &&a, TAU &&tau, W &&work = vector_value_t<A>{}) { // NOLINT (temporary views are allowed here)
    constexpr bool run_on_device = mem::have_device_compatible_addr_space<A, TAU, W>;
 
    // check the dimensions of the input/output arrays/views
    auto const [m, n] = a.shape();
    auto const k      = tau.size();
    EXPECTS(m >= n);
    EXPECTS(n >= k);
 
    // arrays/views must be LAPACK compatible
    EXPECTS(a.indexmap().min_stride() == 1);
    EXPECTS(tau.indexmap().min_stride() == 1);
 
    // first call to get the optimal buffer size
    auto tmp_lwork = get_value_t<A>{};
    int info       = 0;
    if constexpr (run_on_device) {
      tmp_lwork = device::ungqr_buffer_size(m, n, k, a.data(), get_ld(a), tau.data());
    } else {
      f77::ungqr(m, n, k, a.data(), get_ld(a), tau.data(), &tmp_lwork, -1, info);
    }
    int lwork = static_cast<int>(std::ceil(std::real(tmp_lwork)));
 
    // resize/check work buffer
    resize_or_check_work_buffer(work, lwork);
 
    // perform actual library call
    if constexpr (run_on_device) {
      device::ungqr(m, n, k, a.data(), get_ld(a), tau.data(), work.data(), lwork, info);
    } else {
      f77::ungqr(m, n, k, a.data(), get_ld(a), tau.data(), work.data(), lwork, info);
    }
 
    return info;
  }
 
} // namespace nda::lapack