Commit e8a941a0 authored by Vsevak's avatar Vsevak
Browse files

Remove unnecessary data transfer from tip4p/gpu pairstyle init 

Such data transfer is performed at each timestep, so it does not belong to the initialization
parent 337b0fd6

lib/gpu/lal_lj_tip4p_long.cpp

+2 −18
Original line number Diff line number Diff line
@@ -56,9 +56,7 @@ int LJTIP4PLongT::init(const int ntypes, 
    double **host_cut_ljsq,

    const double host_cut_coulsq, const double host_cut_coulsqplus,

    double *host_special_coul, const double qqrd2e,

    const double g_ewald, int* tag,

    int *map_array, int map_size,

    int *sametag, int max_same) {

    const double g_ewald, int map_size, int max_same) {

  int success;

  success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,

                            _screen,lj_tip4p_long,"k_lj_tip4p_long");
@@ -119,21 +117,9 @@ int LJTIP4PLongT::init(const int ntypes, 
  m.alloc(nall,*(this->ucl_device), UCL_READ_WRITE);

  ansO.alloc(nall,*(this->ucl_device), UCL_READ_WRITE);



  // Allocate a host write buffer for data initialization

  UCL_H_Vec<int> host_tag_write(nall,*(this->ucl_device),UCL_READ_WRITE);

  this->tag.alloc(nall,*(this->ucl_device), UCL_READ_ONLY);

  for(int i=0; i<nall; ++i) host_tag_write[i] = tag[i];

  ucl_copy(this->tag, host_tag_write, nall, false);



  //if(max_same>host_tag_write.cols()) host_tag_write.resize(max_same);

  this->atom_sametag.alloc(nall, *(this->ucl_device), UCL_READ_ONLY);

  for(int i=0; i<nall; ++i) host_tag_write[i] = sametag[i];

  ucl_copy(this->atom_sametag, host_tag_write, nall, false);



  host_tag_write.resize_ib(map_size);

  this->atom_sametag.alloc(max_same, *(this->ucl_device), UCL_READ_ONLY);

  this->map_array.alloc(map_size,*(this->ucl_device), UCL_READ_ONLY);

  for(int i=0; i<map_size; ++i) host_tag_write[i] = map_array[i];

  ucl_copy(this->map_array, host_tag_write, map_size, false);



  _allocated=true;

  this->_max_bytes=lj1.row_bytes()+lj3.row_bytes()+cutsq.row_bytes()+
@@ -267,8 +253,6 @@ void LJTIP4PLongT::copy_relations_data(int n, int* tag, int *map_array, 
  this->map_array.resize_ib(map_size);

  for(int i=0; i<map_size; ++i) host_tag_write[i] = map_array[i];

  ucl_copy(this->map_array, host_tag_write, map_size, false);



  host_tag_write.clear();

}

lib/gpu/lal_lj_tip4p_long.h

+1 −3
Original line number Diff line number Diff line
@@ -48,9 +48,7 @@ public: 
           double **host_cut_ljsq,

           const double host_cut_coulsq, const double host_cut_coulsqplus,

           double *host_special_coul, const double qqrd2e,

           const double g_ewald, int* tag,

           int *map_array, int map_size,

           int *sametag, int max_same);

           const double g_ewald, int map_size,int max_same);



  /// Clear all host and device data

  /** \note This is called at the beginning of the init() routine **/

lib/gpu/lal_lj_tip4p_long_ext.cpp

+4 −9
Original line number Diff line number Diff line
@@ -37,9 +37,7 @@ int ljtip4p_long_gpu_init(const int ntypes, double **cutsq, double **host_lj1, 
    double **host_cut_ljsq,

    const double host_cut_coulsq, const double host_cut_coulsqplus,

    double *host_special_coul, const double qqrd2e,

    const double g_ewald, int* tag,

    int *map_array, int map_size,

    int *sametag, int max_same) {

    const double g_ewald, int map_size,int max_same) {

  LJTIP4PLMF.clear();

  gpu_mode=LJTIP4PLMF.device->gpu_mode();

  double gpu_split=LJTIP4PLMF.device->particle_split();
@@ -67,9 +65,7 @@ int ljtip4p_long_gpu_init(const int ntypes, double **cutsq, double **host_lj1, 
        tH, tO, alpha, qdist, nall, 300,

        maxspecial, cell_size, gpu_split, screen,

        host_cut_ljsq, host_cut_coulsq, host_cut_coulsqplus,

        host_special_coul, qqrd2e, g_ewald, tag,

        map_array, map_size,

        sametag, max_same);

        host_special_coul, qqrd2e, g_ewald, map_size, max_same);



  LJTIP4PLMF.device->world_barrier();

  if (message)
@@ -90,9 +86,8 @@ int ljtip4p_long_gpu_init(const int ntypes, double **cutsq, double **host_lj1, 
          tH, tO, alpha, qdist, nall, 300, maxspecial,

          cell_size, gpu_split, screen, host_cut_ljsq,

          host_cut_coulsq, host_cut_coulsqplus,

          host_special_coul, qqrd2e, g_ewald,tag,

          map_array, map_size,

          sametag, max_same);

          host_special_coul, qqrd2e,

          g_ewald, map_size, max_same);



    LJTIP4PLMF.device->gpu_barrier();

    if (message)

src/GPU/pair_lj_cut_tip4p_long_gpu.cpp

+4 −6
Original line number Diff line number Diff line
@@ -58,9 +58,8 @@ int ljtip4p_long_gpu_init(const int ntypes, double **cutsq, double **host_lj1, 
    const double cell_size, int &gpu_mode, FILE *screen,

    double **host_cut_ljsq, const double host_cut_coulsq,

    const double host_cut_coulsqplus, double *host_special_coul,

    const double qqrd2e, const double g_ewald, int* tag,

    int *map_array, int map_size,

    int *sametag, int max_same);

    const double qqrd2e, const double g_ewald,

    int map_size, int max_same);

void ljtip4p_long_gpu_clear();

int ** ljtip4p_long_gpu_compute_n(const int ago, const int inum,

    const int nall, double **host_x, int *host_type,
@@ -227,9 +226,8 @@ void PairLJCutTIP4PLongGPU::init_style() 
                             cell_size, gpu_mode, screen, cut_ljsq,

                             cut_coulsq, cut_coulsqplus,

                             force->special_coul, force->qqrd2e,

                             g_ewald,

                             atom->tag, atom->get_map_array(), atom->get_map_size(),

                             atom->sametag, atom->get_max_same());

                             g_ewald, atom->get_map_size(),

                             atom->get_max_same());

  GPU_EXTRA::check_flag(success,error,world);

  if (gpu_mode == GPU_FORCE) {

    int irequest = neighbor->request(this,instance_me);

CRA Git | Maintained and supported by SUSTech CRA and CCSE