Commit d84ffc53 authored by Stan Moore's avatar Stan Moore
Browse files

Adding dp erfc approximation to pair *coul/long/kk

parent 124bd997

src/KOKKOS/pair_buck_coul_long_kokkos.cpp

+7 −15
Original line number Diff line number Diff line
@@ -30,26 +30,20 @@ 
#include "update.h"

#include "integrate.h"

#include "respa.h"

#include "math_special.h"

#include "math_const.h"

#include "memory.h"

#include "error.h"

#include "atom_masks.h"



using namespace LAMMPS_NS;

using namespace MathSpecial;

using namespace MathConst;



#define KOKKOS_CUDA_MAX_THREADS 256

#define KOKKOS_CUDA_MIN_BLOCKS 8





#define EWALD_F   1.12837917

#define EWALD_P   0.3275911

#define A1        0.254829592

#define A2       -0.284496736

#define A3        1.421413741

#define A4       -1.453152027

#define A5        1.061405429



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



template<class DeviceType>
@@ -235,12 +229,11 @@ compute_fcoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT rinv = 1.0/r;

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]*rinv;

    F_FLOAT forcecoul = prefactor * (erfc + EWALD_F*grij*expm2);

    F_FLOAT forcecoul = prefactor * (erfc + MY_ISPI4*grij*expm2);

    if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;



    return forcecoul*rinv*rinv;
@@ -273,9 +266,8 @@ compute_ecoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]/r;

    F_FLOAT ecoul = prefactor * erfc;

    if (factor_coul < 1.0) ecoul -= (1.0-factor_coul)*prefactor;

src/KOKKOS/pair_coul_long_kokkos.cpp

+7 −15
Original line number Diff line number Diff line
@@ -30,26 +30,20 @@ 
#include "update.h"

#include "integrate.h"

#include "respa.h"

#include "math_special.h"

#include "math_const.h"

#include "memory.h"

#include "error.h"

#include "atom_masks.h"



using namespace LAMMPS_NS;

using namespace MathSpecial;

using namespace MathConst;



#define KOKKOS_CUDA_MAX_THREADS 256

#define KOKKOS_CUDA_MIN_BLOCKS 8





#define EWALD_F   1.12837917

#define EWALD_P   0.3275911

#define A1        0.254829592

#define A2       -0.284496736

#define A3        1.421413741

#define A4       -1.453152027

#define A5        1.061405429



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



template<class DeviceType>
@@ -189,12 +183,11 @@ compute_fcoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT rinv = 1.0/r;

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]*rinv;

    F_FLOAT forcecoul = prefactor * (erfc + EWALD_F*grij*expm2);

    F_FLOAT forcecoul = prefactor * (erfc + MY_ISPI4*grij*expm2);

    if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;



    return forcecoul*rinv*rinv;
@@ -227,9 +220,8 @@ compute_ecoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]/r;

    F_FLOAT ecoul = prefactor * erfc;

    if (factor_coul < 1.0) ecoul -= (1.0-factor_coul)*prefactor;

src/KOKKOS/pair_lj_charmm_coul_long_kokkos.cpp

+7 −15
Original line number Diff line number Diff line
@@ -30,26 +30,20 @@ 
#include "update.h"

#include "integrate.h"

#include "respa.h"

#include "math_special.h"

#include "math_const.h"

#include "memory.h"

#include "error.h"

#include "atom_masks.h"



using namespace LAMMPS_NS;

using namespace MathSpecial;

using namespace MathConst;



#define KOKKOS_CUDA_MAX_THREADS 256

#define KOKKOS_CUDA_MIN_BLOCKS 8





#define EWALD_F   1.12837917

#define EWALD_P   0.3275911

#define A1        0.254829592

#define A2       -0.284496736

#define A3        1.421413741

#define A4       -1.453152027

#define A5        1.061405429



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



template<class DeviceType>
@@ -277,12 +271,11 @@ compute_fcoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT rinv = 1.0/r;

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]*rinv;

    F_FLOAT forcecoul = prefactor * (erfc + EWALD_F*grij*expm2);

    F_FLOAT forcecoul = prefactor * (erfc + MY_ISPI4*grij*expm2);

    if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;



    return forcecoul*rinv*rinv;
@@ -314,9 +307,8 @@ compute_ecoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]/r;

    F_FLOAT ecoul = prefactor * erfc;

    if (factor_coul < 1.0) ecoul -= (1.0-factor_coul)*prefactor;

src/KOKKOS/pair_lj_class2_coul_long_kokkos.cpp

+7 −15
Original line number Diff line number Diff line
@@ -26,26 +26,20 @@ 
#include "update.h"

#include "integrate.h"

#include "respa.h"

#include "math_special.h"

#include "math_const.h"

#include "memory.h"

#include "error.h"

#include "atom_masks.h"



using namespace LAMMPS_NS;

using namespace MathSpecial;

using namespace MathConst;



#define KOKKOS_CUDA_MAX_THREADS 256

#define KOKKOS_CUDA_MIN_BLOCKS 8





#define EWALD_F   1.12837917

#define EWALD_P   0.3275911

#define A1        0.254829592

#define A2       -0.284496736

#define A3        1.421413741

#define A4       -1.453152027

#define A5        1.061405429



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



template<class DeviceType>
@@ -205,12 +199,11 @@ compute_fcoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT rinv = 1.0/r;

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]*rinv;

    F_FLOAT forcecoul = prefactor * (erfc + EWALD_F*grij*expm2);

    F_FLOAT forcecoul = prefactor * (erfc + MY_ISPI4*grij*expm2);

    if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;



    return forcecoul*rinv*rinv;
@@ -263,9 +256,8 @@ compute_ecoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]/r;

    F_FLOAT ecoul = prefactor * erfc;

    if (factor_coul < 1.0) ecoul -= (1.0-factor_coul)*prefactor;

src/KOKKOS/pair_lj_cut_coul_long_kokkos.cpp

+7 −15
Original line number Diff line number Diff line
@@ -26,26 +26,20 @@ 
#include "update.h"

#include "integrate.h"

#include "respa.h"

#include "math_special.h"

#include "math_const.h"

#include "memory.h"

#include "error.h"

#include "atom_masks.h"



using namespace LAMMPS_NS;

using namespace MathSpecial;

using namespace MathConst;



#define KOKKOS_CUDA_MAX_THREADS 256

#define KOKKOS_CUDA_MIN_BLOCKS 8





#define EWALD_F   1.12837917

#define EWALD_P   0.3275911

#define A1        0.254829592

#define A2       -0.284496736

#define A3        1.421413741

#define A4       -1.453152027

#define A5        1.061405429



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



template<class DeviceType>
@@ -226,12 +220,11 @@ compute_fcoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT rinv = 1.0/r;

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]*rinv;

    F_FLOAT forcecoul = prefactor * (erfc + EWALD_F*grij*expm2);

    F_FLOAT forcecoul = prefactor * (erfc + MY_ISPI4*grij*expm2);

    if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;



    return forcecoul*rinv*rinv;
@@ -282,9 +275,8 @@ compute_ecoul(const F_FLOAT& rsq, const int& i, const int&j, 
  } else {

    const F_FLOAT r = sqrt(rsq);

    const F_FLOAT grij = g_ewald * r;

    const F_FLOAT expm2 = exp(-grij*grij);

    const F_FLOAT t = 1.0 / (1.0 + EWALD_P*grij);

    const F_FLOAT erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;

    const F_FLOAT expm2 = expmsq(grij);

    const F_FLOAT erfc = my_erfcx(grij) * expm2;

    const F_FLOAT prefactor = qqrd2e * qtmp*q[j]/r;

    F_FLOAT ecoul = prefactor * erfc;

    if (factor_coul < 1.0) ecoul -= (1.0-factor_coul)*prefactor;

CRA Git | Maintained and supported by SUSTech CRA and CCSE