Unverified Commit 7e5e741e authored by Axel Kohlmeyer's avatar Axel Kohlmeyer
Browse files

update code for ufm/gpu to match changes in base class where uf4 is removed

parent d271ee65

lib/gpu/lal_ufm.cpp

+15 −16
Original line number Diff line number Diff line
@@ -48,7 +48,7 @@ template <class numtyp, class acctyp> 
int UFMT::init(const int ntypes,

                          double **host_cutsq, double **host_uf1,

                          double **host_uf2, double **host_uf3,

                          double **host_uf4, double **host_offset, 

                          double **host_offset,

                          double *host_special_lj, const int nlocal,

                          const int nall, const int max_nbors,

                          const int maxspecial, const double cell_size,
@@ -81,7 +81,7 @@ int UFMT::init(const int ntypes, 
                         host_cutsq);



  uf3.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);

  this->atom->type_pack4(ntypes,lj_types,uf3,host_write,host_uf3,host_uf4,

  this->atom->type_pack4(ntypes,lj_types,uf3,host_write,host_uf3,host_uf2,

                         host_offset);



  UCL_H_Vec<double> dview;
@@ -96,8 +96,7 @@ int UFMT::init(const int ntypes, 


template <class numtyp, class acctyp>

void UFMT::reinit(const int ntypes, double **host_cutsq, double **host_uf1,

                 double **host_uf2, double **host_uf3,

                 double **host_uf4, double **host_offset) {

                 double **host_uf2, double **host_uf3, double **host_offset) {

  // Allocate a host write buffer for data initialization

  UCL_H_Vec<numtyp> host_write(_lj_types*_lj_types*32,*(this->ucl_device),

                               UCL_WRITE_ONLY);
@@ -107,7 +106,7 @@ void UFMT::reinit(const int ntypes, double **host_cutsq, double **host_uf1, 


  this->atom->type_pack4(ntypes,_lj_types,uf1,host_write,host_uf1,host_uf2,

                         host_cutsq);

  this->atom->type_pack4(ntypes,_lj_types,uf3,host_write,host_uf3,host_uf4,

  this->atom->type_pack4(ntypes,_lj_types,uf3,host_write,host_uf3,host_uf2,

                         host_offset);

}

lib/gpu/lal_ufm.h

+11 −12
Original line number Diff line number Diff line
@@ -41,15 +41,14 @@ class UFM : public BaseAtomic<numtyp, acctyp> { 
    * - -5 Double precision is not supported on card **/

  int init(const int ntypes, double **host_cutsq,

           double **host_uf1, double **host_uf2, double **host_uf3,

           double **host_uf4, double **host_offset, double *host_special_lj,

           double **host_offset, double *host_special_lj,

           const int nlocal, const int nall, const int max_nbors,

           const int maxspecial, const double cell_size,

           const double gpu_split, FILE *screen);



  /// Send updated coeffs from host to device (to be compatible with fix adapt)

  void reinit(const int ntypes, double **host_cutsq,

              double **host_uf1, double **host_uf2, double **host_uf3,

              double **host_uf4, double **host_offset);

  void reinit(const int ntypes, double **host_cutsq, double **host_uf1,

              double **host_uf2, double **host_uf3, double **host_offset);



  /// Clear all host and device data

  /** \note This is called at the beginning of the init() routine **/
@@ -65,7 +64,7 @@ class UFM : public BaseAtomic<numtyp, acctyp> { 


  /// uf1.x = uf1, uf1.y = uf2, uf1.z = cutsq

  UCL_D_Vec<numtyp4> uf1;

  /// uf3.x = uf3, uf3.y = uf4, uf3.z = offset

  /// uf3.x = uf3, uf3.y = uf2, uf3.z = offset

  UCL_D_Vec<numtyp4> uf3;

  /// Special LJ values

  UCL_D_Vec<numtyp> sp_lj;

lib/gpu/lal_ufm_ext.cpp

+16 −17
Original line number Diff line number Diff line
@@ -30,8 +30,8 @@ static UFM<PRECISION,ACC_PRECISION> UFMLMF; 
// Allocate memory on host and device and copy constants to device

// ---------------------------------------------------------------------------

int ufml_gpu_init(const int ntypes, double **cutsq, double **host_uf1,

                 double **host_uf2, double **host_uf3, double **host_uf4,

                 double **offset, double *special_lj, const int inum, const int nall,

                 double **host_uf2, double **host_uf3, double **offset,

                 double *special_lj, const int inum, const int nall,

                 const int max_nbors,  const int maxspecial, const double cell_size,

                 int &gpu_mode, FILE *screen) {

  UFMLMF.clear();
@@ -57,7 +57,7 @@ int ufml_gpu_init(const int ntypes, double **cutsq, double **host_uf1, 
  int init_ok=0;

  if (world_me==0)

    init_ok=UFMLMF.init(ntypes, cutsq, host_uf1, host_uf2, host_uf3,

                       host_uf4, offset, special_lj, inum, nall, 300,

                        offset, special_lj, inum, nall, 300,

                        maxspecial, cell_size, gpu_split, screen);



  UFMLMF.device->world_barrier();
@@ -74,7 +74,7 @@ int ufml_gpu_init(const int ntypes, double **cutsq, double **host_uf1, 
      fflush(screen);

    }

    if (gpu_rank==i && world_me!=0)

      init_ok=UFMLMF.init(ntypes, cutsq, host_uf1, host_uf2, host_uf3, host_uf4,

      init_ok=UFMLMF.init(ntypes, cutsq, host_uf1, host_uf2, host_uf3,

                         offset, special_lj, inum, nall, 300, maxspecial,

                         cell_size, gpu_split, screen);
@@ -94,19 +94,18 @@ int ufml_gpu_init(const int ntypes, double **cutsq, double **host_uf1, 
// Copy updated coeffs from host to device

// ---------------------------------------------------------------------------

void ufml_gpu_reinit(const int ntypes, double **cutsq, double **host_uf1,

                    double **host_uf2, double **host_uf3, double **host_uf4,

                    double **offset) {

                    double **host_uf2, double **host_uf3, double **offset) {

  int world_me=UFMLMF.device->world_me();

  int gpu_rank=UFMLMF.device->gpu_rank();

  int procs_per_gpu=UFMLMF.device->procs_per_gpu();



  if (world_me==0)

    UFMLMF.reinit(ntypes, cutsq, host_uf1, host_uf2, host_uf3, host_uf4, offset);

    UFMLMF.reinit(ntypes, cutsq, host_uf1, host_uf2, host_uf3, offset);

  UFMLMF.device->world_barrier();



  for (int i=0; i<procs_per_gpu; i++) {

    if (gpu_rank==i && world_me!=0)

      UFMLMF.reinit(ntypes, cutsq, host_uf1, host_uf2, host_uf3, host_uf4, offset);

      UFMLMF.reinit(ntypes, cutsq, host_uf1, host_uf2, host_uf3, offset);

    UFMLMF.device->gpu_barrier();

  }

}

src/GPU/pair_ufm_gpu.cpp

+4 −4
Original line number Diff line number Diff line
@@ -43,13 +43,13 @@ using namespace LAMMPS_NS; 
// External functions from cuda library for atom decomposition



int ufml_gpu_init(const int ntypes, double **cutsq, double **host_uf1,

                 double **host_uf2, double **host_uf3, double **host_uf4,

                 double **host_uf2, double **host_uf3,

                 double **offset, double *special_lj, const int nlocal,

                 const int nall, const int max_nbors, const int maxspecial,

                 const double cell_size, int &gpu_mode, FILE *screen);



int ufml_gpu_reinit(const int ntypes, double **cutsq, double **host_uf1,

                   double **host_uf2, double **host_uf3, double **host_uf4,

                   double **host_uf2, double **host_uf3,

                   double **offset);



void ufml_gpu_clear();
@@ -166,7 +166,7 @@ void PairUFMGPU::init_style() 
  int maxspecial=0;

  if (atom->molecular)

    maxspecial=atom->maxspecial;

  int success = ufml_gpu_init(atom->ntypes+1, cutsq, uf1, uf2, uf3, uf4,

  int success = ufml_gpu_init(atom->ntypes+1, cutsq, uf1, uf2, uf3,

                             offset, force->special_lj, atom->nlocal,

                             atom->nlocal+atom->nghost, 300, maxspecial,

                             cell_size, gpu_mode, screen);
@@ -185,7 +185,7 @@ void PairUFMGPU::reinit() 
{

  Pair::reinit();



  ufml_gpu_reinit(atom->ntypes+1, cutsq, uf1, uf2, uf3, uf4, offset);

  ufml_gpu_reinit(atom->ntypes+1, cutsq, uf1, uf2, uf3, offset);

}



/* ---------------------------------------------------------------------- */

CRA Git | Maintained and supported by SUSTech CRA and CCSE