diff --git a/lib/gpu/lal_lj_sdk.cpp b/lib/gpu/lal_lj_spica.cpp similarity index 81% rename from lib/gpu/lal_lj_sdk.cpp rename to lib/gpu/lal_lj_spica.cpp index 0da094c953..8ba13853a6 100644 --- a/lib/gpu/lal_lj_sdk.cpp +++ b/lib/gpu/lal_lj_spica.cpp @@ -1,9 +1,9 @@ /*************************************************************************** - lj_sdk.cpp + lj_spica.cpp ------------------- W. Michael Brown (ORNL) - Class for acceleration of the lj/sdk/cut pair style + Class for acceleration of the lj/spica/cut pair style __________________________________________________________________________ This file is part of the LAMMPS Accelerator Library (LAMMPS_AL) @@ -14,14 +14,14 @@ ***************************************************************************/ #if defined(USE_OPENCL) -#include "lj_sdk_cl.h" +#include "lj_spica_cl.h" #elif defined(USE_CUDART) -const char *lj_sdk=0; +const char *lj_spica=0; #else -#include "lj_sdk_cubin.h" +#include "lj_spica_cubin.h" #endif -#include "lal_lj_sdk.h" +#include "lal_lj_spica.h" #include namespace LAMMPS_AL { #define CGCMMT CGCMM @@ -53,33 +53,33 @@ int CGCMMT::init(const int ntypes, double **host_cutsq, const double gpu_split, FILE *_screen) { int success; success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split, - _screen,lj_sdk,"k_lj_sdk"); + _screen,lj_spica,"k_lj_spica"); if (success!=0) return success; // If atom type constants fit in shared memory use fast kernel - int sdk_types=ntypes; + int spica_types=ntypes; shared_types=false; int max_shared_types=this->device->max_shared_types(); - if (sdk_types<=max_shared_types && this->_block_size>=max_shared_types) { - sdk_types=max_shared_types; + if (spica_types<=max_shared_types && this->_block_size>=max_shared_types) { + spica_types=max_shared_types; shared_types=true; } - _sdk_types=sdk_types; + _spica_types=spica_types; // Allocate a host write buffer for data initialization - UCL_H_Vec host_write(sdk_types*sdk_types*32,*(this->ucl_device), + UCL_H_Vec host_write(spica_types*spica_types*32,*(this->ucl_device), UCL_WRITE_ONLY); - for (int i=0; iucl_device),UCL_READ_ONLY); - this->atom->type_pack4(ntypes,sdk_types,lj1,host_write,host_cutsq, + lj1.alloc(spica_types*spica_types,*(this->ucl_device),UCL_READ_ONLY); + this->atom->type_pack4(ntypes,spica_types,lj1,host_write,host_cutsq, host_cg_type,host_lj1,host_lj2); - lj3.alloc(sdk_types*sdk_types,*(this->ucl_device),UCL_READ_ONLY); - this->atom->type_pack4(ntypes,sdk_types,lj3,host_write,host_lj3,host_lj4, + lj3.alloc(spica_types*spica_types,*(this->ucl_device),UCL_READ_ONLY); + this->atom->type_pack4(ntypes,spica_types,lj3,host_write,host_lj3,host_lj4, host_offset); UCL_H_Vec dview; @@ -132,7 +132,7 @@ int CGCMMT::loop(const int eflag, const int vflag) { } else { this->k_pair.set_size(GX,BX); this->k_pair.run(&this->atom->x, &lj1, &lj3, - &_sdk_types, &sp_lj, &this->nbor->dev_nbor, + &_spica_types, &sp_lj, &this->nbor->dev_nbor, &this->_nbor_data->begin(), &this->ans->force, &this->ans->engv, &eflag, &vflag, &ainum, &nbor_pitch, &this->_threads_per_atom); diff --git a/lib/gpu/lal_lj_sdk.cu b/lib/gpu/lal_lj_spica.cu similarity index 93% rename from lib/gpu/lal_lj_sdk.cu rename to lib/gpu/lal_lj_spica.cu index 1bd9a93d5e..bae8bd57fa 100644 --- a/lib/gpu/lal_lj_sdk.cu +++ b/lib/gpu/lal_lj_spica.cu @@ -1,9 +1,9 @@ // ************************************************************************** -// lj_sdk.cu +// lj_spica.cu // ------------------- // W. Michael Brown (ORNL) // -// Device code for acceleration of the lj/sdk pair style +// Device code for acceleration of the lj/spica pair style // // __________________________________________________________________________ // This file is part of the LAMMPS Accelerator Library (LAMMPS_AL) @@ -24,7 +24,7 @@ _texture_2d( pos_tex,int4); #define pos_tex x_ #endif -__kernel void k_lj_sdk(const __global numtyp4 *restrict x_, +__kernel void k_lj_spica(const __global numtyp4 *restrict x_, const __global numtyp4 *restrict lj1, const __global numtyp4 *restrict lj3, const int lj_types, @@ -91,6 +91,9 @@ __kernel void k_lj_sdk(const __global numtyp4 *restrict x_, } else if (lj1[mtype].y == 1) { inv2=r2inv*ucl_sqrt(r2inv); inv1=inv2*inv2; + } else if (lj1[mtype].y == 4) { + inv1=r2inv*r2inv*ucl_sqrt(r2inv); + inv2=inv1*r2inv; } else { inv1=r2inv*r2inv*r2inv; inv2=inv1; @@ -119,7 +122,7 @@ __kernel void k_lj_sdk(const __global numtyp4 *restrict x_, ans,engv); } -__kernel void k_lj_sdk_fast(const __global numtyp4 *restrict x_, +__kernel void k_lj_spica_fast(const __global numtyp4 *restrict x_, const __global numtyp4 *restrict lj1_in, const __global numtyp4 *restrict lj3_in, const __global numtyp *restrict sp_lj_in, @@ -192,6 +195,9 @@ __kernel void k_lj_sdk_fast(const __global numtyp4 *restrict x_, } else if (lj1[mtype].y == (numtyp)1) { inv2=r2inv*ucl_sqrt(r2inv); inv1=inv2*inv2; + } else if (lj1[mtype].y == (numtyp)4) { + inv1=r2inv*r2inv*ucl_sqrt(r2inv); + inv2=inv1*r2inv; } else { inv1=r2inv*r2inv*r2inv; inv2=inv1; diff --git a/lib/gpu/lal_lj_sdk.h b/lib/gpu/lal_lj_spica.h similarity index 95% rename from lib/gpu/lal_lj_sdk.h rename to lib/gpu/lal_lj_spica.h index 043bafdda8..ed6e5ab096 100644 --- a/lib/gpu/lal_lj_sdk.h +++ b/lib/gpu/lal_lj_spica.h @@ -1,9 +1,9 @@ /*************************************************************************** - lj_sdk.h + lj_spica.h ------------------- W. Michael Brown (ORNL) - Class for acceleration of the lj/sdk pair style + Class for acceleration of the lj/spica pair style __________________________________________________________________________ This file is part of the LAMMPS Accelerator Library (LAMMPS_AL) @@ -67,7 +67,7 @@ class CGCMM : public BaseAtomic { bool shared_types; /// Number of atom types - int _sdk_types; + int _spica_types; private: bool _allocated; diff --git a/lib/gpu/lal_lj_sdk_ext.cpp b/lib/gpu/lal_lj_spica_ext.cpp similarity index 89% rename from lib/gpu/lal_lj_sdk_ext.cpp rename to lib/gpu/lal_lj_spica_ext.cpp index 4497233861..1467c1806a 100644 --- a/lib/gpu/lal_lj_sdk_ext.cpp +++ b/lib/gpu/lal_lj_spica_ext.cpp @@ -1,9 +1,9 @@ /*************************************************************************** - lj_sdk.h + lj_spica.h ------------------- W. Michael Brown (ORNL) - Functions for LAMMPS access to lj/sdk pair acceleration routines + Functions for LAMMPS access to lj/spica pair acceleration routines __________________________________________________________________________ This file is part of the LAMMPS Accelerator Library (LAMMPS_AL) @@ -17,7 +17,7 @@ #include #include -#include "lal_lj_sdk.h" +#include "lal_lj_spica.h" using namespace std; using namespace LAMMPS_AL; @@ -27,7 +27,7 @@ static CGCMM CMMMF; // --------------------------------------------------------------------------- // Allocate memory on host and device and copy constants to device // --------------------------------------------------------------------------- -int sdk_gpu_init(const int ntypes, double **cutsq, int **cg_types, +int spica_gpu_init(const int ntypes, double **cutsq, int **cg_types, double **host_lj1, double **host_lj2, double **host_lj3, double **host_lj4, double **offset, double *special_lj, const int inum, const int nall, const int max_nbors, @@ -42,7 +42,7 @@ int sdk_gpu_init(const int ntypes, double **cutsq, int **cg_types, int gpu_rank=CMMMF.device->gpu_rank(); int procs_per_gpu=CMMMF.device->procs_per_gpu(); - CMMMF.device->init_message(screen,"lj/sdk",first_gpu,last_gpu); + CMMMF.device->init_message(screen,"lj/spica",first_gpu,last_gpu); bool message=false; if (CMMMF.device->replica_me()==0 && screen) @@ -89,11 +89,11 @@ int sdk_gpu_init(const int ntypes, double **cutsq, int **cg_types, return init_ok; } -void sdk_gpu_clear() { +void spica_gpu_clear() { CMMMF.clear(); } -int** sdk_gpu_compute_n(const int ago, const int inum_full, +int** spica_gpu_compute_n(const int ago, const int inum_full, const int nall, double **host_x, int *host_type, double *sublo, double *subhi, tagint *tag, int **nspecial, tagint **special, const bool eflag, const bool vflag, @@ -105,7 +105,7 @@ int** sdk_gpu_compute_n(const int ago, const int inum_full, vatom, host_start, ilist, jnum, cpu_time, success); } -void sdk_gpu_compute(const int ago, const int inum_full, const int nall, +void spica_gpu_compute(const int ago, const int inum_full, const int nall, double **host_x, int *host_type, int *ilist, int *numj, int **firstneigh, const bool eflag, const bool vflag, const bool eatom, const bool vatom, int &host_start, @@ -114,7 +114,7 @@ void sdk_gpu_compute(const int ago, const int inum_full, const int nall, firstneigh,eflag,vflag,eatom,vatom,host_start,cpu_time,success); } -double sdk_gpu_bytes() { +double spica_gpu_bytes() { return CMMMF.host_memory_usage(); } diff --git a/lib/gpu/lal_lj_sdk_long.cpp b/lib/gpu/lal_lj_spica_long.cpp similarity index 94% rename from lib/gpu/lal_lj_sdk_long.cpp rename to lib/gpu/lal_lj_spica_long.cpp index d78e8d84da..3a2c6c693a 100644 --- a/lib/gpu/lal_lj_sdk_long.cpp +++ b/lib/gpu/lal_lj_spica_long.cpp @@ -1,9 +1,9 @@ /*************************************************************************** - lj_sdk_long.cpp + lj_spica_long.cpp ------------------- W. Michael Brown (ORNL) - Class for acceleration of the lj/sdk/coul/long pair style + Class for acceleration of the lj/spica/coul/long pair style __________________________________________________________________________ This file is part of the LAMMPS Accelerator Library (LAMMPS_AL) @@ -14,14 +14,14 @@ ***************************************************************************/ #if defined(USE_OPENCL) -#include "lj_sdk_long_cl.h" +#include "lj_spica_long_cl.h" #elif defined(USE_CUDART) -const char *lj_sdk_long=0; +const char *lj_spica_long=0; #else -#include "lj_sdk_long_cubin.h" +#include "lj_spica_long_cubin.h" #endif -#include "lal_lj_sdk_long.h" +#include "lal_lj_spica_long.h" #include namespace LAMMPS_AL { #define CGCMMLongT CGCMMLong @@ -58,7 +58,7 @@ int CGCMMLongT::init(const int ntypes, double **host_cutsq, const double g_ewald) { int success; success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split, - _screen,lj_sdk_long,"k_lj_sdk_long"); + _screen,lj_spica_long,"k_lj_spica_long"); if (success!=0) return success; diff --git a/lib/gpu/lal_lj_sdk_long.cu b/lib/gpu/lal_lj_spica_long.cu similarity index 94% rename from lib/gpu/lal_lj_sdk_long.cu rename to lib/gpu/lal_lj_spica_long.cu index 3972ed2076..12da967f2e 100644 --- a/lib/gpu/lal_lj_sdk_long.cu +++ b/lib/gpu/lal_lj_spica_long.cu @@ -1,9 +1,9 @@ // ************************************************************************** -// lj_sdk_long.cu +// lj_spica_long.cu // ------------------- // W. Michael Brown (ORNL) // -// Device code for acceleration of the lj/sdk/coul/long pair style +// Device code for acceleration of the lj/spica/coul/long pair style // // __________________________________________________________________________ // This file is part of the LAMMPS Accelerator Library (LAMMPS_AL) @@ -29,7 +29,7 @@ _texture( q_tex,int2); #define q_tex q_ #endif -__kernel void k_lj_sdk_long(const __global numtyp4 *restrict x_, +__kernel void k_lj_spica_long(const __global numtyp4 *restrict x_, const __global numtyp4 *restrict lj1, const __global numtyp4 *restrict lj3, const int lj_types, @@ -107,6 +107,9 @@ __kernel void k_lj_sdk_long(const __global numtyp4 *restrict x_, } else if (lj3[mtype].x == (numtyp)1) { inv2=r2inv*ucl_rsqrt(rsq); inv1=inv2*inv2; + } else if (lj3[mtype].x == (numtyp)4) { + inv1=r2inv*r2inv*ucl_rsqrt(rsq); + inv2=inv1*r2inv; } else { inv1=r2inv*r2inv*r2inv; inv2=inv1; @@ -157,7 +160,7 @@ __kernel void k_lj_sdk_long(const __global numtyp4 *restrict x_, vflag,ans,engv); } -__kernel void k_lj_sdk_long_fast(const __global numtyp4 *restrict x_, +__kernel void k_lj_spica_long_fast(const __global numtyp4 *restrict x_, const __global numtyp4 *restrict lj1_in, const __global numtyp4 *restrict lj3_in, const __global numtyp *restrict sp_lj_in, @@ -236,6 +239,9 @@ __kernel void k_lj_sdk_long_fast(const __global numtyp4 *restrict x_, } else if (lj3[mtype].x == (numtyp)1) { inv2=r2inv*ucl_rsqrt(rsq); inv1=inv2*inv2; + } else if (lj3[mtype].x == (numtyp)4) { + inv1=r2inv*r2inv*ucl_rsqrt(rsq); + inv2=inv1*r2inv; } else { inv1=r2inv*r2inv*r2inv; inv2=inv1; diff --git a/lib/gpu/lal_lj_sdk_long.h b/lib/gpu/lal_lj_spica_long.h similarity index 96% rename from lib/gpu/lal_lj_sdk_long.h rename to lib/gpu/lal_lj_spica_long.h index 102b007b59..2b598168c1 100644 --- a/lib/gpu/lal_lj_sdk_long.h +++ b/lib/gpu/lal_lj_spica_long.h @@ -1,9 +1,9 @@ /*************************************************************************** - lj_sdk_long.h + lj_spica_long.h ------------------- W. Michael Brown (ORNL) - Class for acceleration of the lj/sdk/coul/long pair style + Class for acceleration of the lj/spica/coul/long pair style __________________________________________________________________________ This file is part of the LAMMPS Accelerator Library (LAMMPS_AL) diff --git a/lib/gpu/lal_lj_sdk_long_ext.cpp b/lib/gpu/lal_lj_spica_long_ext.cpp similarity index 90% rename from lib/gpu/lal_lj_sdk_long_ext.cpp rename to lib/gpu/lal_lj_spica_long_ext.cpp index 3170ac8b52..293801391f 100644 --- a/lib/gpu/lal_lj_sdk_long_ext.cpp +++ b/lib/gpu/lal_lj_spica_long_ext.cpp @@ -1,9 +1,9 @@ /*************************************************************************** - lj_sdk_long.h + lj_spica_long.h ------------------- W. Michael Brown (ORNL) - Functions for LAMMPS access to lj/sdk/coul/long acceleration functions + Functions for LAMMPS access to lj/spica/coul/long acceleration functions __________________________________________________________________________ This file is part of the LAMMPS Accelerator Library (LAMMPS_AL) @@ -17,7 +17,7 @@ #include #include -#include "lal_lj_sdk_long.h" +#include "lal_lj_spica_long.h" using namespace std; using namespace LAMMPS_AL; @@ -27,7 +27,7 @@ static CGCMMLong CMMLMF; // --------------------------------------------------------------------------- // Allocate memory on host and device and copy constants to device // --------------------------------------------------------------------------- -int sdkl_gpu_init(const int ntypes, double **cutsq, int **cg_type, +int spical_gpu_init(const int ntypes, double **cutsq, int **cg_type, double **host_lj1, double **host_lj2, double **host_lj3, double **host_lj4, double **offset, double *special_lj, const int inum, const int nall, const int max_nbors, @@ -44,7 +44,7 @@ int sdkl_gpu_init(const int ntypes, double **cutsq, int **cg_type, int gpu_rank=CMMLMF.device->gpu_rank(); int procs_per_gpu=CMMLMF.device->procs_per_gpu(); - CMMLMF.device->init_message(screen,"lj/sdk/coul/long",first_gpu,last_gpu); + CMMLMF.device->init_message(screen,"lj/spica/coul/long",first_gpu,last_gpu); bool message=false; if (CMMLMF.device->replica_me()==0 && screen) @@ -93,11 +93,11 @@ int sdkl_gpu_init(const int ntypes, double **cutsq, int **cg_type, return init_ok; } -void sdkl_gpu_clear() { +void spical_gpu_clear() { CMMLMF.clear(); } -int** sdkl_gpu_compute_n(const int ago, const int inum_full, +int** spical_gpu_compute_n(const int ago, const int inum_full, const int nall, double **host_x, int *host_type, double *sublo, double *subhi, tagint *tag, int **nspecial, tagint **special, const bool eflag, const bool vflag, @@ -111,7 +111,7 @@ int** sdkl_gpu_compute_n(const int ago, const int inum_full, host_q,boxlo,prd); } -void sdkl_gpu_compute(const int ago, const int inum_full, const int nall, +void spical_gpu_compute(const int ago, const int inum_full, const int nall, double **host_x, int *host_type, int *ilist, int *numj, int **firstneigh, const bool eflag, const bool vflag, const bool eatom, const bool vatom, int &host_start, @@ -122,7 +122,7 @@ void sdkl_gpu_compute(const int ago, const int inum_full, const int nall, host_q,nlocal,boxlo,prd); } -double sdkl_gpu_bytes() { +double spical_gpu_bytes() { return CMMLMF.host_memory_usage(); }