1 /*************************************************************************** 2 gauss.h 3 ------------------- 4 Trung Dac Nguyen (ORNL) 5 6 Class for acceleration of the gauss pair style. 7 8 __________________________________________________________________________ 9 This file is part of the LAMMPS Accelerator Library (LAMMPS_AL) 10 __________________________________________________________________________ 11 12 begin : 13 email : nguyentd@ornl.gov 14 ***************************************************************************/ 15 16 #ifndef LAL_GAUSS_H 17 #define LAL_GAYSS_H 18 19 #include "lal_base_atomic.h" 20 21 namespace LAMMPS_AL { 22 23 template <class numtyp, class acctyp> 24 class Gauss : public BaseAtomic<numtyp, acctyp> { 25 public: 26 Gauss(); 27 ~Gauss(); 28 29 /// Clear any previous data and set up for a new LAMMPS run 30 /** \param max_nbors initial number of rows in the neighbor matrix 31 * \param cell_size cutoff + skin 32 * \param gpu_split fraction of particles handled by device 33 * 34 * Returns: 35 * - 0 if successfull 36 * - -1 if fix gpu not found 37 * - -3 if there is an out of memory error 38 * - -4 if the GPU library was not compiled for GPU 39 * - -5 Double precision is not supported on card **/ 40 int init(const int ntypes, double **host_cutsq, 41 double **host_a, double **host_b, double **host_offset, 42 double *host_special_lj, 43 const int nlocal, const int nall, const int max_nbors, 44 const int maxspecial, const double cell_size, 45 const double gpu_split, FILE *screen); 46 47 /// Clear all host and device data 48 /** \note This is called at the beginning of the init() routine **/ 49 void clear(); 50 51 /// Returns memory usage on device per atom 52 int bytes_per_atom(const int max_nbors) const; 53 54 /// Total host memory used by library for pair style 55 double host_memory_usage() const; 56 57 // --------------------------- TYPE DATA -------------------------- 58 59 /// gauss1.x = a, gauss1.y = b, gauss1.z = cutsq, gauss1.w = offset 60 UCL_D_Vec<numtyp4> gauss1; 61 /// Special LJ values 62 UCL_D_Vec<numtyp> sp_lj; 63 64 /// If atom type constants fit in shared memory, use fast kernels 65 bool shared_types; 66 67 /// Number of atom types 68 int _lj_types; 69 70 private: 71 bool _allocated; 72 void loop(const bool _eflag, const bool _vflag); 73 }; 74 75 } 76 77 #endif 78