107 lines
3.3 KiB
C++
107 lines
3.3 KiB
C++
/***************************************************************************
|
|
neighbor_shared.h
|
|
-------------------
|
|
W. Michael Brown (ORNL)
|
|
|
|
Class for management of data shared by all neighbor lists
|
|
|
|
__________________________________________________________________________
|
|
This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
|
|
__________________________________________________________________________
|
|
|
|
begin :
|
|
email : brownw@ornl.gov
|
|
***************************************************************************/
|
|
|
|
#ifndef LAL_NEIGHBOR_SHARED_H
|
|
#define LAL_NEIGHBOR_SHARED_H
|
|
|
|
#if defined(USE_OPENCL)
|
|
#include "geryon/ocl_kernel.h"
|
|
#include "geryon/ocl_texture.h"
|
|
using namespace ucl_opencl;
|
|
#elif defined(USE_CUDART)
|
|
#include "geryon/nvc_kernel.h"
|
|
#include "geryon/nvc_texture.h"
|
|
using namespace ucl_cudart;
|
|
#elif defined(USE_HIP)
|
|
#include "geryon/hip_kernel.h"
|
|
#include "geryon/hip_texture.h"
|
|
using namespace ucl_hip;
|
|
#else
|
|
#include "geryon/nvd_kernel.h"
|
|
#include "geryon/nvd_texture.h"
|
|
using namespace ucl_cudadr;
|
|
#endif
|
|
|
|
namespace LAMMPS_AL {
|
|
|
|
class NeighborShared {
|
|
public:
|
|
NeighborShared() : _compiled(false) {}
|
|
~NeighborShared() { clear(); }
|
|
|
|
/// Free all memory on host and device
|
|
void clear();
|
|
|
|
/// Texture for cached position/type access with CUDA
|
|
UCL_Texture neigh_tex;
|
|
|
|
/// Use a heuristic to approximate best bin size assuming uniform density
|
|
/** This is only called by core LAMMPS for atom sort sizes **/
|
|
inline double update_cell_size(const double subx, const double suby,
|
|
const double subz, const int nlocal,
|
|
const double cut) {
|
|
if (_auto_cell_size==false || subz==0.0) return cut;
|
|
else {
|
|
_cell_size=best_cell_size(subx, suby, subz, nlocal, cut);
|
|
_cached_cell_size=true;
|
|
_cut_sort=cut;
|
|
return _cell_size;
|
|
}
|
|
}
|
|
|
|
/// Use a heuristic to approximate best bin size assuming uniform density
|
|
double best_cell_size(const double subx, const double suby,
|
|
const double subz, const int nlocal,
|
|
const double cut);
|
|
|
|
/// Current cutoff used for cell size determination
|
|
inline double neighbor_cutoff() { return _neighbor_cutoff; }
|
|
|
|
/// Current neighbor cell size
|
|
inline double cell_size() { return _cell_size; }
|
|
|
|
/// Return setting for auto cell size
|
|
inline bool auto_cell_size() { return _auto_cell_size; }
|
|
|
|
inline void setup_auto_cell_size(const bool autosize, const double cut,
|
|
const int simd_size) {
|
|
_auto_cell_size = autosize;
|
|
_cached_cell_size = false;
|
|
_neighbor_cutoff = cut;
|
|
_cell_size = cut;
|
|
_simd_size = simd_size;
|
|
if (_simd_size < 2) _auto_cell_size = false;
|
|
}
|
|
|
|
/// Compile kernels for neighbor lists
|
|
void compile_kernels(UCL_Device &dev, const int gpu_nbor,
|
|
const std::string &flags);
|
|
|
|
// ----------------------------- Kernels
|
|
UCL_Program *nbor_program, *build_program;
|
|
UCL_Kernel k_nbor, k_cell_id, k_cell_counts, k_build_nbor;
|
|
UCL_Kernel k_transpose, k_special;
|
|
|
|
private:
|
|
bool _compiled;
|
|
int _gpu_nbor;
|
|
bool _auto_cell_size, _cached_cell_size;
|
|
double _neighbor_cutoff, _cell_size, _simd_size, _cut_sort;
|
|
};
|
|
|
|
}
|
|
|
|
#endif
|