From f4d3d3a2b591ac0dee4d982506ae8e880394a922 Mon Sep 17 00:00:00 2001 From: Trung Nguyen Date: Sat, 2 Oct 2021 00:09:53 -0500 Subject: [PATCH] Gradually cleaned up and removed redundancy in amoeba and hippo --- lib/gpu/lal_base_amoeba.cpp | 156 +++----------------------------- lib/gpu/lal_base_amoeba.h | 20 ++-- lib/gpu/lal_hippo.cpp | 176 +++++++----------------------------- lib/gpu/lal_hippo.cu | 4 +- lib/gpu/lal_hippo.h | 1 - lib/gpu/lal_hippo_ext.cpp | 7 +- src/GPU/pair_hippo_gpu.cpp | 6 +- 7 files changed, 61 insertions(+), 309 deletions(-) diff --git a/lib/gpu/lal_base_amoeba.cpp b/lib/gpu/lal_base_amoeba.cpp index d0631442e0..7cd410b6b8 100644 --- a/lib/gpu/lal_base_amoeba.cpp +++ b/lib/gpu/lal_base_amoeba.cpp @@ -245,8 +245,8 @@ inline int BaseAmoebaT::build_nbor_list(const int inum, const int host_inum, // --------------------------------------------------------------------------- template void BaseAmoebaT::compute_polar_real_host_nbor(const int f_ago, const int inum_full, - const int nall, double **host_x, int *host_type, int *host_amtype, - int *host_amgroup, double **host_rpole, + const int nall, double **host_x, int *host_type, + int *host_amtype, int *host_amgroup, double **host_rpole, double **host_uind, double **host_uinp, int *ilist, int *numj, int **firstneigh, const bool eflag_in, const bool vflag_in, @@ -353,7 +353,6 @@ int** BaseAmoebaT::precompute(const int ago, const int inum_full, const int nall bool &success, double *host_q, double *boxlo, double *prd) { acc_timers(); - //int eflag, vflag; if (eatom) _eflag=2; else if (eflag_in) _eflag=1; else _eflag=0; @@ -401,12 +400,10 @@ int** BaseAmoebaT::precompute(const int ago, const int inum_full, const int nall if (!success) return nullptr; atom->cast_q_data(host_q); - //cast_extra_data(host_amtype, host_amgroup, host_rpole, host_uind, host_uinp, host_pval); hd_balancer.start_timer(); } else { atom->cast_x_data(host_x,host_type); atom->cast_q_data(host_q); - //cast_extra_data(host_amtype, host_amgroup, host_rpole, host_uind, host_uinp, host_pval); hd_balancer.start_timer(); atom->add_x_data(host_x,host_type); } @@ -444,23 +441,6 @@ int** BaseAmoebaT::compute_multipole_real(const int ago, const int inum_full, const double aewald, const double felec, const double off2_mpole, double *host_q, double *boxlo, double *prd, void **tep_ptr) { -/* - acc_timers(); - int eflag, vflag; - if (eatom) eflag=2; - else if (eflag_in) eflag=1; - else eflag=0; - if (vatom) vflag=2; - else if (vflag_in) vflag=1; - else vflag=0; - - #ifdef LAL_NO_BLOCK_REDUCE - if (eflag) eflag=2; - if (vflag) vflag=2; - #endif - - set_kernel(eflag,vflag); -*/ // reallocate per-atom arrays, transfer data from the host // and build the neighbor lists if needed // NOTE: @@ -499,13 +479,7 @@ int** BaseAmoebaT::compute_multipole_real(const int ago, const int inum_full, // copy tep from device to host _tep.update_host(_max_tep_size*4,false); -/* - printf("GPU lib: tep size = %d: max tep size = %d\n", this->_tep.cols(), _max_tep_size); - for (int i = 0; i < 10; i++) { - numtyp4* p = (numtyp4*)(&this->_tep[4*i]); - printf("i = %d; tep = %f %f %f\n", i, p->x, p->y, p->z); - } -*/ + return firstneigh; // nbor->host_jlist.begin()-host_start; } @@ -529,36 +503,11 @@ int** BaseAmoebaT::compute_udirect2b(const int ago, const int inum_full, const double aewald, const double off2_polar, double *host_q, double *boxlo, double *prd, void** fieldp_ptr) { -/* - acc_timers(); - int eflag, vflag; - if (eatom) eflag=2; - else if (eflag_in) eflag=1; - else eflag=0; - if (vatom) vflag=2; - else if (vflag_in) vflag=1; - else vflag=0; - - #ifdef LAL_NO_BLOCK_REDUCE - if (eflag) eflag=2; - if (vflag) vflag=2; - #endif - - set_kernel(eflag,vflag); -*/ // reallocate per-atom arrays, transfer data from the host // and build the neighbor lists if needed int** firstneigh = nullptr; -/* - firstneigh = precompute(ago, inum_full, nall, host_x, host_type, - host_amtype, host_amgroup, host_rpole, - host_uind, host_uinp, nullptr, sublo, subhi, tag, - nspecial, special, nspecial15, special15, - eflag_in, vflag_in, eatom, vatom, - host_start, ilist, jnum, cpu_time, - success, host_q, boxlo, prd); -*/ + cast_extra_data(host_amtype, host_amgroup, host_rpole, host_uind, host_uinp, host_pval); atom->add_extra_data(); @@ -577,14 +526,7 @@ int** BaseAmoebaT::compute_udirect2b(const int ago, const int inum_full, // copy field and fieldp from device to host (_fieldp store both arrays, one after another) _fieldp.update_host(_max_fieldp_size*8,false); -/* - printf("GPU lib: _fieldp size = %d: max fieldp size = %d\n", - this->_fieldp.cols(), _max_fieldp_size); - for (int i = 0; i < 10; i++) { - numtyp4* p = (numtyp4*)(&this->_fieldp[4*i]); - printf("i = %d; field = %f %f %f\n", i, p->x, p->y, p->z); - } -*/ + return firstneigh; //nbor->host_jlist.begin()-host_start; } @@ -608,36 +550,11 @@ int** BaseAmoebaT::compute_umutual2b(const int ago, const int inum_full, const double aewald, const double off2_polar, double *host_q, double *boxlo, double *prd, void** fieldp_ptr) { -/* - acc_timers(); - int eflag, vflag; - if (eatom) eflag=2; - else if (eflag_in) eflag=1; - else eflag=0; - if (vatom) vflag=2; - else if (vflag_in) vflag=1; - else vflag=0; - - #ifdef LAL_NO_BLOCK_REDUCE - if (eflag) eflag=2; - if (vflag) vflag=2; - #endif - - set_kernel(eflag,vflag); -*/ // reallocate per-atom arrays, transfer extra data from the host // and build the neighbor lists if needed int** firstneigh = nullptr; -/* - firstneigh = precompute(ago, inum_full, nall, host_x, host_type, - host_amtype, host_amgroup, host_rpole, - host_uind, host_uinp, nullptr, sublo, subhi, tag, - nspecial, special, nspecial15, special15, - eflag_in, vflag_in, eatom, vatom, - host_start, ilist, jnum, cpu_time, - success, host_q, boxlo, prd); -*/ + cast_extra_data(host_amtype, host_amgroup, host_rpole, host_uind, host_uinp, host_pval); atom->add_extra_data(); @@ -656,14 +573,7 @@ int** BaseAmoebaT::compute_umutual2b(const int ago, const int inum_full, // copy field and fieldp from device to host (_fieldp store both arrays, one after another) _fieldp.update_host(_max_fieldp_size*8,false); -/* - printf("GPU lib: _fieldp size = %d: max fieldp size = %d\n", - this->_fieldp.cols(), _max_fieldp_size); - for (int i = 0; i < 10; i++) { - numtyp4* p = (numtyp4*)(&this->_fieldp[4*i]); - printf("i = %d; field = %f %f %f\n", i, p->x, p->y, p->z); - } -*/ + return firstneigh; //nbor->host_jlist.begin()-host_start; } @@ -686,44 +596,9 @@ int** BaseAmoebaT::compute_polar_real(const int ago, const int inum_full, const double aewald, const double felec, const double off2_polar, double *host_q, double *boxlo, double *prd, void **tep_ptr) { -/* - acc_timers(); - int eflag, vflag; - if (eatom) eflag=2; - else if (eflag_in) eflag=1; - else eflag=0; - if (vatom) vflag=2; - else if (vflag_in) vflag=1; - else vflag=0; - - #ifdef LAL_NO_BLOCK_REDUCE - if (eflag) eflag=2; - if (vflag) vflag=2; - #endif - - set_kernel(eflag,vflag); -*/ - // reallocate per-atom arrays, transfer data from the host - // and build the neighbor lists if needed - // NOTE: - // For now we invoke precompute() again here, - // to be able to turn on/off the udirect2b kernel (which comes before this) - // Once all the kernels are ready, precompute() is needed only once - // in the first kernel in a time step. - // We only need to cast uind and uinp from host to device here - // if the neighbor lists are rebuilt and other per-atom arrays - // (x, type, amtype, amgroup, rpole) are ready on the device. int** firstneigh = nullptr; -/* - firstneigh = precompute(ago, inum_full, nall, host_x, host_type, - host_amtype, host_amgroup, host_rpole, - host_uind, host_uinp, nullptr, sublo, subhi, tag, - nspecial, special, nspecial15, special15, - eflag_in, vflag_in, eatom, vatom, - host_start, ilist, jnum, cpu_time, - success, host_q, boxlo, prd); -*/ + cast_extra_data(host_amtype, host_amgroup, host_rpole, host_uind, host_uinp, host_pval); atom->add_extra_data(); @@ -750,13 +625,7 @@ int** BaseAmoebaT::compute_polar_real(const int ago, const int inum_full, // copy tep from device to host _tep.update_host(_max_tep_size*4,false); -/* - printf("GPU lib: tep size = %d: max tep size = %d\n", this->_tep.cols(), _max_tep_size); - for (int i = 0; i < 10; i++) { - numtyp4* p = (numtyp4*)(&this->_tep[4*i]); - printf("i = %d; tep = %f %f %f\n", i, p->x, p->y, p->z); - } -*/ + return firstneigh; // nbor->host_jlist.begin()-host_start; } @@ -826,7 +695,6 @@ void BaseAmoebaT::cast_extra_data(int* amtype, int* amgroup, double** rpole, n += nstride*_nall; if (pval) { - for (int i = 0; i < _nall; i++) { int idx = n+i*nstride; pextra[idx] = pval[i]; @@ -889,9 +757,9 @@ int BaseAmoebaT::add_onefive_neighbors() { k_special15.set_size(GX,BX); k_special15.run(&nbor->dev_nbor, &_nbor_data->begin(), - &atom->dev_tag, &dev_nspecial15, &dev_special15, - &ainum, &_nall, &nbor_pitch, - &_threads_per_atom); + &atom->dev_tag, &dev_nspecial15, &dev_special15, + &ainum, &_nall, &nbor_pitch, + &_threads_per_atom); return GX; } diff --git a/lib/gpu/lal_base_amoeba.h b/lib/gpu/lal_base_amoeba.h index cb040c630d..dc3467f692 100644 --- a/lib/gpu/lal_base_amoeba.h +++ b/lib/gpu/lal_base_amoeba.h @@ -145,14 +145,14 @@ class BaseAmoeba { /// Compute multipole real-space with device neighboring virtual int** compute_multipole_real(const int ago, const int inum_full, const int nall, double **host_x, int *host_type, int *host_amtype, - int *host_amgroup, double **host_rpole, double *host_pval, double *sublo, double *subhi, - tagint *tag, int **nspecial, tagint **special, - int *nspecial15, tagint **special15, - const bool eflag, const bool vflag, - const bool eatom, const bool vatom, int &host_start, - int **ilist, int **numj, const double cpu_time, bool &success, - const double aewald, const double felec, const double off2_mpole, double *charge, - double *boxlo, double *prd, void **tep_ptr); + int *host_amgroup, double **host_rpole, double *host_pval, + double *sublo, double *subhi, tagint *tag, + int **nspecial, tagint **special, int *nspecial15, tagint **special15, + const bool eflag, const bool vflag, const bool eatom, const bool vatom, + int &host_start, int **ilist, int **numj, const double cpu_time, + bool &success, const double aewald, const double felec, + const double off2_mpole, double *charge, double *boxlo, + double *prd, void **tep_ptr); /// Compute the real space part of the permanent field (udirect2b) with device neighboring virtual int** compute_udirect2b(const int ago, const int inum_full, const int nall, @@ -165,8 +165,8 @@ class BaseAmoeba { const bool eflag, const bool vflag, const bool eatom, const bool vatom, int &host_start, int **ilist, int **numj, const double cpu_time, bool &success, - const double aewald, const double off2_polar, double *charge, - double *boxlo, double *prd, void **fieldp_ptr); + const double aewald, const double off2_polar, + double *charge, double *boxlo, double *prd, void **fieldp_ptr); /// Compute the real space part of the induced field (umutual2b) with device neighboring virtual int** compute_umutual2b(const int ago, const int inum_full, const int nall, diff --git a/lib/gpu/lal_hippo.cpp b/lib/gpu/lal_hippo.cpp index 9a86be8f42..a5e3be5974 100644 --- a/lib/gpu/lal_hippo.cpp +++ b/lib/gpu/lal_hippo.cpp @@ -48,22 +48,20 @@ int HippoT::bytes_per_atom(const int max_nbors) const { template int HippoT::init(const int ntypes, const int max_amtype, const int max_amclass, - const double *host_pdamp, const double *host_thole, - const double *host_dirdamp, const int *host_amtype2class, - const double *host_special_hal, - const double *host_special_repel, - const double *host_special_disp, - const double *host_special_mpole, - const double *host_special_polar_wscale, - const double *host_special_polar_piscale, - const double *host_special_polar_pscale, - const double *host_sizpr, const double *host_dmppr, const double *host_elepr, - const double *host_csix, const double *host_adisp, - const double *host_pcore, const double *host_palpha, - const int nlocal, const int nall, const int max_nbors, - const int maxspecial, const int maxspecial15, - const double cell_size, const double gpu_split, FILE *_screen, - const double polar_dscale, const double polar_uscale) { + const double *host_pdamp, const double *host_thole, + const double *host_dirdamp, const int *host_amtype2class, + const double *host_special_repel, const double *host_special_disp, + const double *host_special_mpole, + const double *host_special_polar_wscale, + const double *host_special_polar_piscale, + const double *host_special_polar_pscale, + const double *host_sizpr, const double *host_dmppr, const double *host_elepr, + const double *host_csix, const double *host_adisp, + const double *host_pcore, const double *host_palpha, + const int nlocal, const int nall, const int max_nbors, + const int maxspecial, const int maxspecial15, + const double cell_size, const double gpu_split, FILE *_screen, + const double polar_dscale, const double polar_uscale) { int success; success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,maxspecial15, cell_size,gpu_split,_screen,hippo, @@ -133,9 +131,9 @@ int HippoT::init(const int ntypes, const int max_amtype, const int max_amclass, sp_nonpolar.alloc(5,*(this->ucl_device),UCL_READ_ONLY); for (int i=0; i<5; i++) { - dview[i].x=host_special_hal[i]; - dview[i].y=host_special_repel[i]; - dview[i].z=host_special_disp[i]; + dview[i].x=host_special_repel[i]; + dview[i].y=host_special_disp[i]; + dview[i].z=(numtyp)0; dview[i].w=(numtyp)0; } ucl_copy(sp_nonpolar,dview,5,false); @@ -211,7 +209,7 @@ int** HippoT::compute_repulsion(const int ago, const int inum_full, // to be able to turn on/off the udirect2b kernel (which comes before this) // Once all the kernels are ready, precompute() is needed only once // in the first kernel in a time step. - // We only need to cast uind and uinp from host to device here + // We only need to cast the necessary from host to device here // if the neighbor lists are rebuilt and other per-atom arrays // (x, type, amtype, amgroup, rpole) are ready on the device. @@ -240,7 +238,7 @@ int** HippoT::compute_repulsion(const int ago, const int inum_full, _c3 = c3; _c4 = c4; _c5 = c5; - const int red_blocks=repulsion(eflag,vflag); + const int red_blocks=repulsion(this->_eflag,this->_vflag); // only copy them back if this is the last kernel // otherwise, commenting out these two lines to leave the answers @@ -316,32 +314,14 @@ int** HippoT::compute_dispersion_real(const int ago, const int inum_full, const double cpu_time, bool &success, const double aewald, const double off2_disp, double *host_q, double *boxlo, double *prd) { - this->acc_timers(); - int eflag, vflag; - if (eatom) eflag=2; - else if (eflag_in) eflag=1; - else eflag=0; - if (vatom) vflag=2; - else if (vflag_in) vflag=1; - else vflag=0; - - #ifdef LAL_NO_BLOCK_REDUCE - if (eflag) eflag=2; - if (vflag) vflag=2; - #endif - - this->set_kernel(eflag,vflag); - // reallocate per-atom arrays, transfer data from the host // and build the neighbor lists if needed // NOTE: // For now we invoke precompute() again here, // to be able to turn on/off the udirect2b kernel (which comes before this) - // Once all the kernels are ready, precompute() is needed only once - // in the first kernel in a time step. - // We only need to cast uind and uinp from host to device here - // if the neighbor lists are rebuilt and other per-atom arrays - // (x, type, amtype, amgroup, rpole) are ready on the device. + // We only need to cast necesary data arrays from host to device here + // because the neighbor lists are rebuilt and other per-atom arrays + // (x, type) are ready on the device. int** firstneigh = nullptr; firstneigh = this->precompute(ago, inum_full, nall, host_x, host_type, @@ -350,11 +330,11 @@ int** HippoT::compute_dispersion_real(const int ago, const int inum_full, nspecial, special, nspecial15, special15, eflag_in, vflag_in, eatom, vatom, host_start, ilist, jnum, cpu_time, - success, host_q, boxlo, prd); + success, host_q, boxlo, prd); this->_off2_disp = off2_disp; this->_aewald = aewald; - const int red_blocks=dispersion_real(eflag,vflag); + const int red_blocks=dispersion_real(this->_eflag,this->_vflag); // only copy them back if this is the last kernel // otherwise, commenting out these two lines to leave the answers @@ -427,22 +407,6 @@ int** HippoT::compute_multipole_real(const int ago, const int inum_full, const double aewald, const double felec, const double off2_mpole, double *host_q, double *boxlo, double *prd, void **tep_ptr) { - this->acc_timers(); - int eflag, vflag; - if (eatom) eflag=2; - else if (eflag_in) eflag=1; - else eflag=0; - if (vatom) vflag=2; - else if (vflag_in) vflag=1; - else vflag=0; - - #ifdef LAL_NO_BLOCK_REDUCE - if (eflag) eflag=2; - if (vflag) vflag=2; - #endif - - this->set_kernel(eflag,vflag); - // reallocate per-atom arrays, transfer data from the host // and build the neighbor lists if needed // NOTE: @@ -474,7 +438,7 @@ int** HippoT::compute_multipole_real(const int ago, const int inum_full, this->_off2_mpole = off2_mpole; this->_felec = felec; this->_aewald = aewald; - const int red_blocks=multipole_real(eflag,vflag); + const int red_blocks=multipole_real(this->_eflag,this->_vflag); // leave the answers (forces, energies and virial) on the device, // only copy them back in the last kernel (this one, or polar_real once done) @@ -486,13 +450,7 @@ int** HippoT::compute_multipole_real(const int ago, const int inum_full, // copy tep from device to host this->_tep.update_host(this->_max_tep_size*4,false); -/* - printf("GPU lib: tep size = %d: max tep size = %d\n", this->_tep.cols(), _max_tep_size); - for (int i = 0; i < 10; i++) { - numtyp4* p = (numtyp4*)(&this->_tep[4*i]); - printf("i = %d; tep = %f %f %f\n", i, p->x, p->y, p->z); - } -*/ + return firstneigh; // nbor->host_jlist.begin()-host_start; } @@ -558,22 +516,6 @@ int** HippoT::compute_udirect2b(const int ago, const int inum_full, const double aewald, const double off2_polar, double *host_q, double *boxlo, double *prd, void** fieldp_ptr) { - this->acc_timers(); - int eflag, vflag; - if (eatom) eflag=2; - else if (eflag_in) eflag=1; - else eflag=0; - if (vatom) vflag=2; - else if (vflag_in) vflag=1; - else vflag=0; - - #ifdef LAL_NO_BLOCK_REDUCE - if (eflag) eflag=2; - if (vflag) vflag=2; - #endif - - this->set_kernel(eflag,vflag); - // reallocate per-atom arrays, transfer data from the host // and build the neighbor lists if needed @@ -596,19 +538,12 @@ int** HippoT::compute_udirect2b(const int ago, const int inum_full, this->_off2_polar = off2_polar; this->_aewald = aewald; - const int red_blocks=udirect2b(eflag,vflag); + const int red_blocks=udirect2b(this->_eflag,this->_vflag); // copy field and fieldp from device to host (_fieldp store both arrays, one after another) this->_fieldp.update_host(this->_max_fieldp_size*8,false); -/* - printf("GPU lib: _fieldp size = %d: max fieldp size = %d\n", - this->_fieldp.cols(), _max_fieldp_size); - for (int i = 0; i < 10; i++) { - numtyp4* p = (numtyp4*)(&this->_fieldp[4*i]); - printf("i = %d; field = %f %f %f\n", i, p->x, p->y, p->z); - } -*/ + return firstneigh; //nbor->host_jlist.begin()-host_start; } @@ -673,22 +608,6 @@ int** HippoT::compute_umutual2b(const int ago, const int inum_full, const double aewald, const double off2_polar, double *host_q, double *boxlo, double *prd, void** fieldp_ptr) { - this->acc_timers(); - int eflag, vflag; - if (eatom) eflag=2; - else if (eflag_in) eflag=1; - else eflag=0; - if (vatom) vflag=2; - else if (vflag_in) vflag=1; - else vflag=0; - - #ifdef LAL_NO_BLOCK_REDUCE - if (eflag) eflag=2; - if (vflag) vflag=2; - #endif - - this->set_kernel(eflag,vflag); - // reallocate per-atom arrays, transfer extra data from the host // and build the neighbor lists if needed @@ -711,19 +630,12 @@ int** HippoT::compute_umutual2b(const int ago, const int inum_full, this->_off2_polar = off2_polar; this->_aewald = aewald; - const int red_blocks=umutual2b(eflag,vflag); + const int red_blocks=umutual2b(this->_eflag,this->_vflag); // copy field and fieldp from device to host (_fieldp store both arrays, one after another) this->_fieldp.update_host(this->_max_fieldp_size*8,false); -/* - printf("GPU lib: _fieldp size = %d: max fieldp size = %d\n", - this->_fieldp.cols(), _max_fieldp_size); - for (int i = 0; i < 10; i++) { - numtyp4* p = (numtyp4*)(&this->_fieldp[4*i]); - printf("i = %d; field = %f %f %f\n", i, p->x, p->y, p->z); - } -*/ + return firstneigh; //nbor->host_jlist.begin()-host_start; } @@ -786,29 +698,11 @@ int** HippoT::compute_polar_real(const int ago, const int inum_full, const double aewald, const double felec, const double off2_polar, double *host_q, double *boxlo, double *prd, void **tep_ptr) { - this->acc_timers(); - int eflag, vflag; - if (eatom) eflag=2; - else if (eflag_in) eflag=1; - else eflag=0; - if (vatom) vflag=2; - else if (vflag_in) vflag=1; - else vflag=0; - - #ifdef LAL_NO_BLOCK_REDUCE - if (eflag) eflag=2; - if (vflag) vflag=2; - #endif - - this->set_kernel(eflag,vflag); - // reallocate per-atom arrays, transfer data from the host // and build the neighbor lists if needed // NOTE: // For now we invoke precompute() again here, // to be able to turn on/off the udirect2b kernel (which comes before this) - // Once all the kernels are ready, precompute() is needed only once - // in the first kernel in a time step. // We only need to cast uind and uinp from host to device here // if the neighbor lists are rebuilt and other per-atom arrays // (x, type, amtype, amgroup, rpole) are ready on the device. @@ -833,7 +727,7 @@ int** HippoT::compute_polar_real(const int ago, const int inum_full, this->_off2_polar = off2_polar; this->_felec = felec; this->_aewald = aewald; - const int red_blocks=polar_real(eflag,vflag); + const int red_blocks=polar_real(this->_eflag,this->_vflag); // only copy answers (forces, energies and virial) back from the device // in the last kernel (which is polar_real here) @@ -845,13 +739,7 @@ int** HippoT::compute_polar_real(const int ago, const int inum_full, // copy tep from device to host this->_tep.update_host(this->_max_tep_size*4,false); -/* - printf("GPU lib: tep size = %d: max tep size = %d\n", this->_tep.cols(), _max_tep_size); - for (int i = 0; i < 10; i++) { - numtyp4* p = (numtyp4*)(&this->_tep[4*i]); - printf("i = %d; tep = %f %f %f\n", i, p->x, p->y, p->z); - } -*/ + return firstneigh; // nbor->host_jlist.begin()-host_start; } diff --git a/lib/gpu/lal_hippo.cu b/lib/gpu/lal_hippo.cu index cb11bd4022..f38a9f4ac0 100644 --- a/lib/gpu/lal_hippo.cu +++ b/lib/gpu/lal_hippo.cu @@ -520,7 +520,7 @@ __kernel void k_hippo_repulsion(const __global numtyp4 *restrict x_, numtyp valk = coeff_rep[jtype].z; // elepr[jtype]; const numtyp4 sp_nonpol = sp_nonpolar[sbmask15(jextra)]; - numtyp factor_repel = sp_nonpol.y; // factor_repel = special_repel[sbmask15(j)]; + numtyp factor_repel = sp_nonpol.x; // factor_repel = special_repel[sbmask15(j)]; if (factor_repel == (numtyp)0) continue; // intermediates involving moments and separation distance @@ -830,7 +830,7 @@ __kernel void k_hippo_dispersion(const __global numtyp4 *restrict x_, numtyp damp3,damp5; numtyp ddamp; const numtyp4 sp_nonpol = sp_nonpolar[sbmask15(jextra)]; - numtyp factor_disp = sp_nonpol.z; // factor_disp = special_disp[sbmask15(j)]; + numtyp factor_disp = sp_nonpol.y; // factor_disp = special_disp[sbmask15(j)]; if (ai != ak) { ai2 = ai * ai; diff --git a/lib/gpu/lal_hippo.h b/lib/gpu/lal_hippo.h index ceab20d17b..9941460bff 100644 --- a/lib/gpu/lal_hippo.h +++ b/lib/gpu/lal_hippo.h @@ -41,7 +41,6 @@ class Hippo : public BaseAmoeba { const double *host_pdamp, const double *host_thole, const double *host_dirdamp, const int *host_amtype2class, const double *host_special_mpole, - const double *host_special_hal, const double *host_special_repel, const double *host_special_disp, const double *host_special_polar_wscale, diff --git a/lib/gpu/lal_hippo_ext.cpp b/lib/gpu/lal_hippo_ext.cpp index 15cb53cdb1..4152833320 100644 --- a/lib/gpu/lal_hippo_ext.cpp +++ b/lib/gpu/lal_hippo_ext.cpp @@ -30,7 +30,6 @@ static Hippo HIPPOMF; int hippo_gpu_init(const int ntypes, const int max_amtype, const int max_amclass, const double *host_pdamp, const double *host_thole, const double *host_dirdamp, const int *host_amtype2class, - const double *host_special_hal, const double *host_special_repel, const double *host_special_disp, const double *host_special_mpole, @@ -71,8 +70,7 @@ int hippo_gpu_init(const int ntypes, const int max_amtype, const int max_amclass if (world_me==0) init_ok=HIPPOMF.init(ntypes, max_amtype, max_amclass, host_pdamp, host_thole, host_dirdamp, - host_amtype2class, host_special_hal, - host_special_repel, host_special_disp, + host_amtype2class, host_special_repel, host_special_disp, host_special_mpole, host_special_polar_wscale, host_special_polar_piscale, host_special_polar_pscale, host_sizpr, host_dmppr, host_elepr, @@ -97,8 +95,7 @@ int hippo_gpu_init(const int ntypes, const int max_amtype, const int max_amclass if (gpu_rank==i && world_me!=0) init_ok=HIPPOMF.init(ntypes, max_amtype, max_amclass, host_pdamp, host_thole, host_dirdamp, - host_amtype2class, host_special_hal, - host_special_repel, host_special_disp, + host_amtype2class, host_special_repel, host_special_disp, host_special_mpole, host_special_polar_wscale, host_special_polar_piscale, host_special_polar_pscale, host_sizpr, host_dmppr, host_elepr, diff --git a/src/GPU/pair_hippo_gpu.cpp b/src/GPU/pair_hippo_gpu.cpp index 014b14471e..dcdac836bd 100644 --- a/src/GPU/pair_hippo_gpu.cpp +++ b/src/GPU/pair_hippo_gpu.cpp @@ -53,8 +53,8 @@ enum{GORDON1,GORDON2}; int hippo_gpu_init(const int ntypes, const int max_amtype, const int max_amclass, const double *host_pdamp, const double *host_thole, const double *host_dirdamp, const int* host_amtype2class, - const double *host_special_hal, const double *host_special_repel, - const double *host_special_disp, const double *host_special_mpole, + const double *host_special_repel, const double *host_special_disp, + const double *host_special_mpole, const double *host_special_polar_wscale, const double *host_special_polar_piscale, const double *host_special_polar_pscale, @@ -203,7 +203,7 @@ void PairHippoGPU::init_style() int tq_size; int mnf = 5e-2 * neighbor->oneatom; int success = hippo_gpu_init(atom->ntypes+1, max_amtype, max_amclass, - pdamp, thole, dirdamp, amtype2class, special_hal, + pdamp, thole, dirdamp, amtype2class, special_repel, special_disp, special_mpole, special_polar_wscale, special_polar_piscale, special_polar_pscale, sizpr, dmppr, elepr,