field_storage.h

Go to the documentation of this file.

/** @file field_storage.h */
 
#ifndef FIELD_STORAGEH
#define FIELD_STORAGEH
 
#include "plumbing/defs.h"
#include "plumbing/coordinates.h"
#include "plumbing/field.h"
#include "plumbing/backend_vector/vector_types.h"
 
#include "plumbing/has_unary_minus.h"
 
 
template <typename T>
class field_struct;
 
#if defined(CUDA) || defined(HIP)
#define DEVICE __device__
#else
#define DEVICE
#endif
 
/**
 * @brief The field_storage struct contains minimal information for using
 * the field in a loop. It is communicated to CUDA kernels and other
 * accelerator functions.
 */
template <typename T>
class field_storage {
  public:
    // The actual data content of the class: a pointer to the field and a
    // list of neighbour pointers.
    T *RESTRICT fieldbuf = nullptr;
    const unsigned *RESTRICT neighbours[NDIRS];
 
    void allocate_field(const lattice_struct &lattice);
    void free_field();
 
#ifndef VECTORIZED
    // Get an element in a loop
    DEVICE inline auto get(const unsigned i, const unsigned field_alloc_size) const;
 
    // template <typename A>
    DEVICE inline void set(const T &value, const unsigned i, const unsigned field_alloc_size);
 
    // Get a single element outside loops
    auto get_element(const unsigned i, const lattice_struct &lattice) const;
    template <typename A>
    void set_element(A &value, const unsigned i, const lattice_struct &lattice);
 
#else
    inline T get_element(const unsigned i) const;
 
    inline void set_element(const T &value, const unsigned i);
 
    // in vector code, write only 1 element to field at site index idx
    template <typename vecT>
    inline void set_vector(const vecT &val, const unsigned idx);
 
    template <typename vecT>
    inline vecT get_vector(const unsigned idx) const;
 
    void gather_comm_vectors(
        T *RESTRICT buffer, const lattice_struct::comm_node_struct &to_node, Parity par,
        const vectorized_lattice_struct<hila::vector_info<T>::vector_size> *RESTRICT vlat,
        bool antiperiodic) const;
 
    void place_recv_elements(
        const T *RESTRICT buffer, Direction d, Parity par,
        const vectorized_lattice_struct<hila::vector_info<T>::vector_size> *RESTRICT vlat) const;
 
#endif
 
    void gather_comm_elements(T *RESTRICT buffer, const lattice_struct::comm_node_struct &to_node,
                              Parity par, const lattice_struct &lattice, bool antiperiodic) const;
 
    void gather_elements(T *RESTRICT buffer, const unsigned *RESTRICT index_list, int n,
                         const lattice_struct &lattice) const;
 
    void gather_elements_negated(T *RESTRICT buffer, const unsigned *RESTRICT index_list, int n,
                                 const lattice_struct &lattice) const;
 
    /// Place boundary elements from neighbour
    void place_comm_elements(Direction d, Parity par, T *RESTRICT buffer,
                             const lattice_struct::comm_node_struct &from_node,
                             const lattice_struct &lattice);
    void place_elements(T *RESTRICT buffer, const unsigned *RESTRICT index_list, int n,
                        const lattice_struct &lattice);
    /// Place boundary elements from local lattice (used in vectorized version)
    void set_local_boundary_elements(Direction dir, Parity par, const lattice_struct &lattice,
                                     bool antiperiodic);
 
    // Allocate buffers for mpi communication
    T *allocate_mpi_buffer(unsigned n);
    void free_mpi_buffer(T *buffer);
 
    T *RESTRICT get_buffer() {
        return static_cast<T *>(fieldbuf);
    }
};
 
/*
Import backend
*/
 
#ifdef VECTORIZED
 
#include "plumbing/backend_vector/field_storage_backend.h"
 
#elif defined(CUDA) || defined(HIP)
 
#include "plumbing/backend_gpu/field_storage_backend.h"
 
#elif defined(VANILLA)
 
#include "plumbing/backend_cpu/field_storage_backend.h"
 
#else
Something must be defined !
 
#endif
 
#endif // field storage