Homec4science
Diffusion akantu (features/node_to_node) Browse

rAKA/src/common/aka_tensor.hh
1db5ba830ccbfeatures/node_to_node

Last Change
Raw File

aka_tensor.hh
View Options

/**
* Copyright (©) 2022-2023 EPFL (Ecole Polytechnique Fédérale de Lausanne)
* Laboratory (LSMS - Laboratoire de Simulation en Mécanique des Solides)
*
* This file is part of Akantu
*
* Akantu is free software: you can redistribute it and/or modify it under the
* terms of the GNU Lesser General Public License as published by the Free
* Software Foundation, either version 3 of the License, or (at your option) any
* later version.
*
* Akantu is distributed in the hope that it will be useful, but WITHOUT ANY
* WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
* A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
* details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with Akantu. If not, see <http://www.gnu.org/licenses/>.
*/
/* -------------------------------------------------------------------------- */
#ifndef AKA_TENSOR_HH_
#define AKA_TENSOR_HH_
namespace akantu {
/* -------------------------------------------------------------------------- */
/* TensorBase */
/* -------------------------------------------------------------------------- */
template <typename T, Int ndim> class TensorBase : public TensorTrait<ndim> {
using RetType = TensorBase<T, ndim>;
static_assert(ndim > 2, "TensorBase cannot by used for dimensions < 3");
protected:
using idx_t = Idx;
template <typename... Args>
using valid_args_t = typename std::enable_if<
aka::conjunction<aka::disjunction<std::is_integral<Args>,
std::is_enum<Args>>...>::value and
ndim == sizeof...(Args),
int>::type;
public:
using proxy = TensorBase<T, ndim>;
using size_type = Idx;
template <Int _ndim = ndim,
std::enable_if_t<_ndim == 1 or _ndim == 2, int> = 0>
TensorBase() {
n.fill(0);
}
TensorBase() { n.fill(0); }
template <typename... Args, valid_args_t<Args...> = 0>
constexpr TensorBase(Args... args)
: n{idx_t(args)...}, _size(detail::product_all(args...)) {}
constexpr TensorBase(const TensorBase & other)
: n(other.n), _size(other._size), values(other.values) {}
constexpr TensorBase(TensorBase && other) noexcept
: n(std::move(other.n)), _size(std::exchange(other._size, 0)),
values(std::exchange(other.values, nullptr)) {}
protected:
template <typename Array, idx_t... I>
constexpr auto check_indices(
const Array & idx,
std::integer_sequence<idx_t, I...> /* for_template_deduction */) const {
bool result = true;
(void)std::initializer_list<int>{(result &= idx[I] < n[I], 0)...};
return result;
}
template <typename... Args> constexpr auto compute_index(Args... args) const {
std::array<idx_t, sizeof...(Args)> idx{idx_t(args)...};
assert(check_indices(
idx, std::make_integer_sequence<idx_t, sizeof...(Args)>{}) &&
"The indexes are out of bound");
idx_t index = 0, i = (sizeof...(Args)) - 1;
for (; i > 0; i--) {
index += idx[i];
if (i > 0) {
index *= n[i - 1];
}
}
return index + idx[0];
}
template <typename S, int... I>
constexpr auto get_slice(idx_t s, std::index_sequence<I...>) {
return S(this->values + s * detail::product_all(n[I]...), n[I]...);
}
template <typename S, std::size_t... I>
constexpr auto get_slice(idx_t s, std::index_sequence<I...>) const {
return S(this->values + s * detail::product_all(n[I]...), n[I]...);
}
public:
template <typename... Args, valid_args_t<Args...> = 0>
inline auto operator()(Args... args) -> T & {
return *(this->values + compute_index(std::move(args)...));
}
template <typename... Args, valid_args_t<Args...> = 0>
inline auto operator()(Args... args) const -> const T & {
return *(this->values + compute_index(std::move(args)...));
}
template <
class R = T, idx_t _ndim = ndim,
std::enable_if_t<(_ndim > 3) and not std::is_const<R>::value> * = nullptr>
inline auto operator()(idx_t s) {
return get_slice<TensorProxy<T, ndim - 1>>(
s, std::make_index_sequence<ndim - 1>());
}
template <idx_t _ndim = ndim, std::enable_if_t<(_ndim > 3)> * = nullptr>
inline auto operator()(idx_t s) const {
return get_slice<TensorProxy<T, ndim - 1>>(
s, std::make_index_sequence<ndim - 1>());
}
template <class R = T, idx_t _ndim = ndim,
std::enable_if_t<(_ndim == 3) and not std::is_const<R>::value> * =
nullptr>
inline auto operator()(idx_t s) {
return get_slice<
Eigen::Map<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>>>(
s, std::make_index_sequence<ndim - 1>());
}
template <idx_t _ndim = ndim, std::enable_if_t<_ndim == 3> * = nullptr>
inline auto operator()(idx_t s) const {
return get_slice<Eigen::Map<const Eigen::Matrix<
std::remove_const_t<T>, Eigen::Dynamic, Eigen::Dynamic>>>(
s, std::make_index_sequence<ndim - 1>());
}
protected:
template <class Operator> auto transform(Operator && op) -> RetType & {
std::transform(this->values, this->values + this->_size, this->values,
std::forward<Operator>(op));
return *(static_cast<RetType *>(this));
}
template <class Other, class Operator>
auto transform(Other && other, Operator && op) -> RetType & {
AKANTU_DEBUG_ASSERT(_size == other.size(),
"The two tensors do not have the same size "
<< this->_size << " != " << other._size);
std::transform(this->values, this->values + this->_size, other.values,
this->values, std::forward<Operator>(op));
return *(static_cast<RetType *>(this));
}
template <class Operator> auto accumulate(T init, Operator && op) -> T {
return std::accumulate(this->values, this->values + this->_size,
std::move(init), std::forward<Operator>(op));
}
template <class Other, class Init, class Accumulate, class Operator>
auto transform_reduce(Other && other, T init, Accumulate && acc,
Operator && op) -> T {
return std::inner_product(
this->values, this->values + this->_size, other.data(), std::move(init),
std::forward<Accumulate>(acc), std::forward<Operator>(op));
}
// element wise arithmetic operators -----------------------------------------
public:
inline decltype(auto) operator+=(const TensorBase & other) {
return transform(other, [](auto && a, auto && b) { return a + b; });
}
/* ------------------------------------------------------------------------ */
inline auto operator-=(const TensorBase & other) -> TensorBase & {
return transform(other, [](auto && a, auto && b) { return a - b; });
}
/* ------------------------------------------------------------------------ */
inline auto operator+=(const T & x) -> TensorBase & {
return transform([&x](auto && a) { return a + x; });
}
/* ------------------------------------------------------------------------ */
inline auto operator-=(const T & x) -> TensorBase & {
return transform([&x](auto && a) { return a - x; });
}
/* ------------------------------------------------------------------------ */
inline auto operator*=(const T & x) -> TensorBase & {
return transform([&x](auto && a) { return a * x; });
}
/* ---------------------------------------------------------------------- */
inline auto operator/=(const T & x) -> TensorBase & {
return transform([&x](auto && a) { return a / x; });
}
/// Y = \alpha X + Y
inline auto aXplusY(const TensorBase & other, const T alpha = 1.)
-> TensorBase & {
return transform(other,
[&alpha](auto && a, auto && b) { return alpha * a + b; });
}
/* ------------------------------------------------------------------------ */
auto data() -> T * { return values; }
auto data() const -> const T * { return values; }
// clang-format off
[[deprecated("use data instead to be stl compatible")]]
auto storage() -> T*{
return values;
}
[[deprecated("use data instead to be stl compatible")]]
auto storage() const -> const T * {
return values;
}
// clang-format on
auto size() const { return _size; }
auto size(idx_t i) const {
AKANTU_DEBUG_ASSERT(i < ndim, "This tensor has only " << ndim
<< " dimensions, not "
<< (i + 1));
return n[i];
};
inline void set(const T & t) { std::fill_n(values, _size, t); };
inline void zero() { set(T()); };
public:
/// "Entrywise" norm norm<L_p> @f[ \|\boldsymbol{T}\|_p = \left(
/// \sum_i^{n[0]}\sum_j^{n[1]}\sum_k^{n[2]} |T_{ijk}|^p \right)^{\frac{1}{p}}
/// @f]
template <Int norm_type,
std::enable_if_t<norm_type == Eigen::Infinity> * = nullptr>
auto lpNorm() const -> T {
return accumulate(
T(), [](auto && init, auto && a) { return init + std::abs(a); });
}
template <Int norm_type, std::enable_if_t<norm_type == 1> * = nullptr>
auto lpNorm() const -> T {
return accumulate(T(), [](auto && init, auto && a) {
return std::max(init, std::abs(a));
});
}
template <Int norm_type, std::enable_if_t<norm_type == 2> * = nullptr>
auto norm() const -> T {
return std::sqrt(
accumulate(T(), [](auto && init, auto && a) { return init + a * a; }));
}
template <Int norm_type, std::enable_if_t<(norm_type > 2)> * = nullptr>
auto norm() const -> T {
return std::pow(accumulate(T(),
[](auto && init, auto && a) {
return init + std::pow(a, norm_type);
}),
1. / norm_type);
}
auto norm() const -> T { return lpNorm<2>(); }
protected:
template <Int N, typename... Args,
std::enable_if_t<(sizeof...(Args) == ndim), int> = 0>
void serialize(std::ostream & stream, Args... args) const {
stream << this->operator()(std::move(args)...);
}
template <Int N, typename... Args,
std::enable_if_t<(sizeof...(Args) < ndim), int> = 0>
void serialize(std::ostream & stream, Args... args) const {
stream << "[";
for (idx_t i = 0; i < n[N]; ++i) {
if (i != 0) {
stream << ",";
}
serialize<N + 1>(stream, std::move(args)..., i);
}
stream << "]";
}
public:
void printself(std::ostream & stream) const { serialize<0>(stream); };
protected:
template <std::size_t... I>
constexpr decltype(auto) begin(std::index_sequence<I...>) {
return view_iterator<ViewIteratorHelper_t<sizeof...(I), T>>(values,
n[I]...);
}
template <std::size_t... I>
constexpr decltype(auto) end(std::index_sequence<I...>) {
return view_iterator<ViewIteratorHelper_t<sizeof...(I), T>>(values + _size,
n[I]...);
}
template <std::size_t... I>
constexpr decltype(auto) begin(std::index_sequence<I...>) const {
return const_view_iterator<ViewIteratorHelper_t<sizeof...(I), T>>(values,
n[I]...);
}
template <std::size_t... I>
constexpr decltype(auto) end(std::index_sequence<I...>) const {
return const_view_iterator<ViewIteratorHelper_t<sizeof...(I), T>>(
values + _size, n[I]...);
}
public:
decltype(auto) begin() { return begin(std::make_index_sequence<ndim - 1>{}); }
decltype(auto) end() { return end(std::make_index_sequence<ndim - 1>{}); }
decltype(auto) begin() const {
return begin(std::make_index_sequence<ndim - 1>{});
}
decltype(auto) end() const {
return end(std::make_index_sequence<ndim - 1>{});
}
protected:
// size per dimension
std::array<idx_t, ndim> n;
// total storage size
idx_t _size{0};
// actual data location
T * values{nullptr};
};
/* -------------------------------------------------------------------------- */
/* TensorProxy */
/* -------------------------------------------------------------------------- */
template <typename T, Int ndim> class TensorProxy : public TensorBase<T, ndim> {
private:
using parent = TensorBase<T, ndim>;
public:
// proxy constructor
template <typename... Args>
constexpr TensorProxy(T * data = reinterpret_cast<T *>(0xdeadbeef),
Args... args)
: parent(args...) {
this->values = data;
}
constexpr TensorProxy(const TensorProxy<T, ndim> & other) : parent(other) {
this->values = other.values;
}
constexpr TensorProxy(const Tensor<T, ndim> & other) : parent(other) {
this->values = other.values;
}
// move constructors ---------------------------------------------------------
// proxy -> proxy
TensorProxy(TensorProxy && other) noexcept : parent(other) {}
auto operator=(const TensorBase<T, ndim> & other) -> TensorProxy & {
AKANTU_DEBUG_ASSERT(
other.size() == this->size(),
"You are trying to copy too a tensors proxy with the wrong size "
<< this->_size << " != " << other._size);
static_assert(std::is_trivially_copyable<T>{},
"Cannot copy a tensor on non trivial types");
std::copy(other.values, other.values + this->_size, this->values);
return *this;
}
};
/* -------------------------------------------------------------------------- */
/* Tensor */
/* -------------------------------------------------------------------------- */
template <typename T, Int ndim> class Tensor : public TensorBase<T, ndim> {
private:
using parent = TensorBase<T, ndim>;
public:
template <typename... Args> constexpr Tensor(Args... args) : parent(args...) {
static_assert(
std::is_trivially_constructible<T>{},
"Cannot create a tensor on non trivially constructible types");
this->values = new T[this->_size];
}
/* ------------------------------------------------------------------------ */
virtual ~Tensor() { delete[] this->values; }
// copy constructors ---------------------------------------------------------
constexpr Tensor(const Tensor & other) : parent(other) {
this->values = new T[this->_size];
std::copy(other.values, other.values + this->_size, this->values);
}
constexpr explicit Tensor(const TensorProxy<T, ndim> & other)
: parent(other) {
// static_assert(false, "Copying data are you sure");
this->values = new T[this->_size];
std::copy(other.values, other.values + this->_size, this->values);
}
// move constructors ---------------------------------------------------------
// proxy -> proxy, non proxy -> non proxy
Tensor(Tensor && other) noexcept : parent(other) {}
// copy operator -------------------------------------------------------------
/// operator= copy-and-swap
auto operator=(const TensorBase<T, ndim> & other) -> Tensor & {
if (&other == this)
return *this;
std::cout << "Warning: operator= delete data" << std::endl;
delete[] this->values;
this->n = other.n;
this->_size = other._size;
static_assert(
std::is_trivially_constructible<T>{},
"Cannot create a tensor on non trivially constructible types");
this->values = new T[this->_size];
static_assert(std::is_trivially_copyable<T>{},
"Cannot copy a tensor on non trivial types");
std::copy(other.values, other.values + this->_size, this->values);
return *this;
}
};
/* -------------------------------------------------------------------------- */
template <typename T> using Tensor3 = Tensor<T, 3>;
template <typename T> using Tensor3Proxy = TensorProxy<T, 3>;
template <typename T> using Tensor3Base = TensorBase<T, 3>;
} // namespace akantu
#endif // AKA_TENSOR_HH_
c4science · Help