EIC code displayed by LXR master/​include/​Acts/​Utilities/​detail/​Subspace.hpp	Source navigation Diff markup Identifier search General search
	Version: master test Revert   Change

File indexing completed on 2025-01-18 09:27:55

 // This file is part of the Acts project.
 //
 // Copyright (C) 2020 CERN for the benefit of the Acts project
 //
 // This Source Code Form is subject to the terms of the Mozilla Public
 // License, v. 2.0. If a copy of the MPL was not distributed with this
 // file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #pragma once
 
 #include "Acts/Definitions/Algebra.hpp"
 
 #include <array>
 #include <cassert>
 #include <cstddef>
 #include <cstdint>
 
 namespace Acts::detail {
 
 /// @defgroup subspace Linear subspace definitions
 ///
 /// All types in this group define a linear subspace of a larger vector space
 /// with the following properties: the subspace is defined by a set of axis
 /// indices in the full space, i.e. there is no need for a non-trivial
 /// projection matrix, all subspace axis indices are unique and there are no
 /// duplicated ones, and the set of axis indices must be ordered, i.e. the axis
 /// order in the subspace is the same as in the full space. The last requirement
 /// is not strictly required by the implementation, but is still added to
 /// simplify reasoning.
 ///
 /// Only the size of the subspace are defined by the types at compile time.
 /// Which axes comprise the subspace is defined at runtime. This allows to use
 /// fixed-size computations (selected at compile time) but avoids the
 /// combinatorial explosion of also having to handle all possible combination of
 /// axes at compile time. This was tried previously and resulted in sizable
 /// resource consumption at compile time without runtime benefits.
 ///
 /// All types are intentionally using `std::size_t` as their template values,
 /// instead of the more specific index enums, to reduce the number of templates.
 /// This is fully compatible as the index enums are required to be convertible
 /// to `std::size_t`.
 ///
 /// All types intentionally only define the subspace but not how vectors
 /// and matrices are stored to avoid unnecessary coupling between components,
 /// i.e here between the pure definition and the storage.
 ///
 /// All types provide `.projectVector(...)` and `.exandVector(...)` methods to
 /// convert to/from the subspace. They also provide `.projector()` and
 /// `.expander()` methods to create projection and expansion matrices if they
 /// are required explicitly. For the specific subspace requirements listed
 /// above, the projection and expansion matrices are transpose to each other. In
 /// the general case, this does not have to be the case and users are encouraged
 /// to use `.projector()` and `.expander()` instead of e.g.
 /// `.projector().transpose()`.
 ///
 /// @{
 
 /// Fixed-size subspace representation.
 ///
 /// @tparam kFullSize Size of the full vector space
 /// @tparam kSize Size of the subspace
 template <std::size_t kFullSize, std::size_t kSize>
 class FixedSizeSubspace {
   static_assert(kFullSize <= static_cast<std::size_t>(UINT8_MAX),
                 "Full vector space size is larger than the supported range");
   static_assert(1u <= kSize, "Subspace size must be at least 1");
   static_assert(kSize <= kFullSize,
                 "Subspace can only be as large as the full space");
 
   template <typename source_t>
   using SubspaceVectorFor = Eigen::Matrix<typename source_t::Scalar, kSize, 1>;
   template <typename source_t>
   using FullspaceVectorFor =
       Eigen::Matrix<typename source_t::Scalar, kFullSize, 1>;
   template <typename scalar_t>
   using ProjectionMatrix = Eigen::Matrix<scalar_t, kSize, kFullSize>;
   template <typename scalar_t>
   using ExpansionMatrix = Eigen::Matrix<scalar_t, kFullSize, kSize>;
 
   // the functionality could also be implemented using a std::bitset where each
   // bit corresponds to an axis in the fullspace and set bits indicate which
   // bits make up the subspace. this would be a more compact representation but
   // complicates the implementation since we can not easily iterate over the
   // indices of the subspace. storing the subspace indices directly requires a
   // bit more memory but is easier to work with. for our typical use cases with
   // n<=8, this still takes only 64bit of memory.
   std::array<uint8_t, kSize> m_axes;
 
  public:
   /// Construct from a container of axis indices.
   ///
   /// @tparam index_t Input index type, must be convertible to uint8_t
   /// @param indices Unique, ordered indices
   template <typename index_t>
   constexpr FixedSizeSubspace(const std::array<index_t, kSize>& indices) {
     for (std::size_t i = 0u; i < kSize; ++i) {
       assert((indices[i] < kFullSize) &&
              "Axis indices must be within the full space");
       if (0u < i) {
         assert((indices[i - 1u] < indices[i]) &&
                "Axis indices must be unique and ordered");
       }
     }
     for (std::size_t i = 0; i < kSize; ++i) {
       m_axes[i] = static_cast<uint8_t>(indices[i]);
     }
   }
   // The subset can not be constructed w/o defining its axis indices.
   FixedSizeSubspace() = delete;
   FixedSizeSubspace(const FixedSizeSubspace&) = default;
   FixedSizeSubspace(FixedSizeSubspace&&) = default;
   FixedSizeSubspace& operator=(const FixedSizeSubspace&) = default;
   FixedSizeSubspace& operator=(FixedSizeSubspace&&) = default;
 
   /// Size of the subspace.
   static constexpr std::size_t size() { return kSize; }
   /// Size of the full vector space.
   static constexpr std::size_t fullSize() { return kFullSize; }
 
   /// Axis indices that comprise the subspace.
   ///
   /// The specific container and index type should be considered an
   /// implementation detail. Users should treat the return type as a generic
   /// container whose elements are convertible to `std::size_t`.
   constexpr const std::array<uint8_t, kSize>& indices() const { return m_axes; }
 
   /// Check if the given axis index in the full space is part of the subspace.
   constexpr bool contains(std::size_t index) const {
     bool isContained = false;
     // always iterate over all elements to avoid branching and hope the compiler
     // can optimise this for us.
     for (auto a : m_axes) {
       isContained = (isContained || (a == index));
     }
     return isContained;
   }
 
   /// Project a full vector into the subspace.
   ///
   /// @tparam fullspace_vector_t Vector type in the full space
   /// @param full Vector in the full space
   /// @return Subspace vector w/ just the configured axis components
   ///
   /// @note Always returns a column vector regardless of the input
   template <typename fullspace_vector_t>
   auto projectVector(const Eigen::MatrixBase<fullspace_vector_t>& full) const
       -> SubspaceVectorFor<fullspace_vector_t> {
     EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(fullspace_vector_t, kFullSize);
 
     SubspaceVectorFor<fullspace_vector_t> sub;
     for (auto i = 0u; i < kSize; ++i) {
       sub[i] = full[m_axes[i]];
     }
     return sub;
   }
 
   /// Expand a subspace vector into the full space.
   ///
   /// @tparam subspace_vector_t Subspace vector type
   /// @param sub Subspace vector
   /// @return Vector in the full space w/ zeros for non-subspace components
   ///
   /// @note Always returns a column vector regardless of the input
   template <typename subspace_vector_t>
   auto expandVector(const Eigen::MatrixBase<subspace_vector_t>& sub) const
       -> FullspaceVectorFor<subspace_vector_t> {
     EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(subspace_vector_t, kSize);
 
     FullspaceVectorFor<subspace_vector_t> full;
     full.setZero();
     for (auto i = 0u; i < kSize; ++i) {
       full[m_axes[i]] = sub[i];
     }
     return full;
   }
 
   /// Projection matrix that maps from the full space into the subspace.
   ///
   /// @tparam scalar_t Scalar type for the projection matrix
   template <typename scalar_t>
   auto projector() const -> ProjectionMatrix<scalar_t> {
     ProjectionMatrix<scalar_t> proj;
     proj.setZero();
     for (auto i = 0u; i < kSize; ++i) {
       proj(i, m_axes[i]) = 1;
     }
     return proj;
   }
 
   /// Expansion matrix that maps from the subspace into the full space.
   ///
   /// @tparam scalar_t Scalar type of the generated expansion matrix
   template <typename scalar_t>
   auto expander() const -> ExpansionMatrix<scalar_t> {
     ExpansionMatrix<scalar_t> expn;
     expn.setZero();
     for (auto i = 0u; i < kSize; ++i) {
       expn(m_axes[i], i) = 1;
     }
     return expn;
   }
 };
 
 /// @}
 
 }  // namespace Acts::detail

This page was automatically generated by the 2.3.7 LXR engine. The LXR team