Commit 165fa01a authored by Stan Moore's avatar Stan Moore
Browse files

Cleanup

parent 8f665a5a

doc/src/Speed_kokkos.txt

+1 −1
Original line number Diff line number Diff line
@@ -211,7 +211,7 @@ with the "-DLMP_KOKKOS_USE_ATOMICS" compile switch. Most but not all 
Kokkos-enabled pair_styles support data duplication. Alternatively, full 

neighbor lists avoid the need for duplication or atomics but require 

more compute operations per atom. When using the Kokkos Serial backend 

or the OpenMP backend with a single thread, no duplication or atomics is 

or the OpenMP backend with a single thread, no duplication or atomics are 

used. For CUDA and half neighbor lists, the KOKKOS package always uses 

atomics.

src/KOKKOS/pair_eam_alloy_kokkos.cpp

+0 −1
Original line number Diff line number Diff line
@@ -38,7 +38,6 @@ using namespace LAMMPS_NS; 


// Cannot use virtual inheritance on the GPU, so must duplicate code





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



template<class DeviceType>

src/KOKKOS/pair_eam_kokkos.cpp

+0 −1
Original line number Diff line number Diff line 
/* ----------------------------------------------------------------------

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

   LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator

   http://lammps.sandia.gov, Sandia National Laboratories

src/KOKKOS/pair_reaxc_kokkos.cpp

+28 −51
Original line number Diff line number Diff line
@@ -73,14 +73,6 @@ PairReaxCKokkos<DeviceType>::PairReaxCKokkos(LAMMPS *lmp) : PairReaxC(lmp) 


  k_error_flag = DAT::tdual_int_scalar("pair:error_flag");

  k_nbuf_local = DAT::tdual_int_scalar("pair:nbuf_local");



  static double t1 = 0.0;

  static double t2 = 0.0;

  static double t3 = 0.0;

  static double t4 = 0.0;

  static double t5 = 0.0;

  static double t6 = 0.0;

  static double t7 = 0.0;

}



/* ---------------------------------------------------------------------- */
@@ -678,8 +670,6 @@ void PairReaxCKokkos<DeviceType>::LR_vdW_Coulomb( int i, int j, double r_ij, LR_ 
template<class DeviceType>

void PairReaxCKokkos<DeviceType>::compute(int eflag_in, int vflag_in)

{

  Kokkos::Timer timer;



  copymode = 1;



  bocnt = hbcnt = 0;
@@ -828,7 +818,6 @@ void PairReaxCKokkos<DeviceType>::compute(int eflag_in, int vflag_in) 
    // zero

    Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType, PairReaxZero>(0,nmax),*this);





    if (neighflag == HALF)

      Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType, PairReaxBuildListsHalf<HALF> >(0,ignum),*this);

    else if (neighflag == HALFTHREAD)
@@ -1115,8 +1104,8 @@ template<int NEIGHFLAG, int EVFLAG> 
KOKKOS_INLINE_FUNCTION

void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeLJCoulomb<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {



  // The f array is atomic for Half/Thread neighbor style

  //Kokkos::View<F_FLOAT*[3], typename DAT::t_f_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_f = f;

  // The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial



  auto v_f = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);

  auto a_f = v_f.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -1276,8 +1265,8 @@ template<int NEIGHFLAG, int EVFLAG> 
KOKKOS_INLINE_FUNCTION

void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeTabulatedLJCoulomb<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {



  // The f array is atomic for Half/Thread neighbor style

  //Kokkos::View<F_FLOAT*[3], typename DAT::t_f_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_f = f;

  // The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial



  auto v_f = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);

  auto a_f = v_f.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -1660,10 +1649,9 @@ void PairReaxCKokkos<DeviceType>::operator()(PairReaxBuildListsHalf<NEIGHFLAG>, 
  if (d_resize_bo() || d_resize_hb())

    return;



  //Kokkos::View<F_FLOAT**, typename DAT::t_ffloat_2d_dl::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_dDeltap_self = d_dDeltap_self;

  auto v_dDeltap_self = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_dDeltap_self),decltype(ndup_dDeltap_self)>::get(dup_dDeltap_self,ndup_dDeltap_self);

  auto a_dDeltap_self = v_dDeltap_self.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();

  //Kokkos::View<F_FLOAT*, typename DAT::t_float_1d::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_total_bo = d_total_bo;



  auto v_total_bo = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_total_bo),decltype(ndup_total_bo)>::get(dup_total_bo,ndup_total_bo);

  auto a_total_bo = v_total_bo.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -2356,7 +2344,6 @@ template<int NEIGHFLAG, int EVFLAG> 
KOKKOS_INLINE_FUNCTION

void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeMulti2<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {



  //Kokkos::View<F_FLOAT*, typename DAT::t_float_1d::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_CdDelta = d_CdDelta;

  auto v_CdDelta = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);

  auto a_CdDelta = v_CdDelta.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -2509,12 +2496,10 @@ template<int NEIGHFLAG, int EVFLAG> 
KOKKOS_INLINE_FUNCTION

void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeAngular<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {



  //Kokkos::View<F_FLOAT*[3], typename DAT::t_f_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_f = f;

  auto v_f = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);

  auto a_f = v_f.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();

  Kokkos::View<F_FLOAT**, typename DAT::t_ffloat_2d_dl::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_Cdbo = d_Cdbo;

  //auto a_Cdbo = dup_Cdbo.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();

  //Kokkos::View<F_FLOAT*, typename DAT::t_float_1d::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_CdDelta = d_CdDelta;



  auto v_CdDelta = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);

  auto a_CdDelta = v_CdDelta.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -2758,10 +2743,10 @@ void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeAngular<NEIGHFLAG,EV 


      // Forces



      d_Cdbo(i,j_index) += (CEval1 + CEpen2 + (CEcoa1 - CEcoa4));

      d_Cdbo(j,i_index) += (CEval1 + CEpen2 + (CEcoa1 - CEcoa4));

      d_Cdbo(i,k_index) += (CEval2 + CEpen3 + (CEcoa2 - CEcoa5));

      d_Cdbo(k,i_index) += (CEval2 + CEpen3 + (CEcoa2 - CEcoa5));

      a_Cdbo(i,j_index) += (CEval1 + CEpen2 + (CEcoa1 - CEcoa4));

      a_Cdbo(j,i_index) += (CEval1 + CEpen2 + (CEcoa1 - CEcoa4));

      a_Cdbo(i,k_index) += (CEval2 + CEpen3 + (CEcoa2 - CEcoa5));

      a_Cdbo(k,i_index) += (CEval2 + CEpen3 + (CEcoa2 - CEcoa5));



      CdDelta_i += ((CEval3 + CEval7) + CEpen1 + CEcoa3);

      CdDelta_j += CEcoa4;
@@ -2776,7 +2761,7 @@ void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeAngular<NEIGHFLAG,EV 
        temp = temp_bo_jt * temp_bo_jt * temp_bo_jt;

        pBOjt7 = temp * temp * temp_bo_jt;



        d_Cdbo(i,l_index) += (CEval6 * pBOjt7);

        a_Cdbo(i,l_index) += (CEval6 * pBOjt7);

        d_Cdbopi(i,l_index) += CEval5;

        d_Cdbopi2(i,l_index) += CEval5;

      }
@@ -2823,10 +2808,9 @@ template<int NEIGHFLAG, int EVFLAG> 
KOKKOS_INLINE_FUNCTION

void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeTorsion<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {



  //Kokkos::View<F_FLOAT*[3], typename DAT::t_f_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_f = f;

  auto v_f = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);

  auto a_f = v_f.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();

  //Kokkos::View<F_FLOAT*, typename DAT::t_float_1d::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_CdDelta = d_CdDelta;



  auto v_CdDelta = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);

  auto a_CdDelta = v_CdDelta.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();

  Kokkos::View<F_FLOAT**, typename DAT::t_ffloat_2d_dl::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_Cdbo = d_Cdbo;
@@ -3131,9 +3115,9 @@ void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeTorsion<NEIGHFLAG,EV 
        CdDelta_i += CEtors3;

        CdDelta_j += CEtors3;



        d_Cdbo(i,k_index) += CEtors4 + CEconj1;

        d_Cdbo(i,j_index) += CEtors5 + CEconj2;

        d_Cdbo(j,l_index) += CEtors6 + CEconj3; // trouble

        a_Cdbo(i,k_index) += CEtors4 + CEconj1;

        a_Cdbo(i,j_index) += CEtors5 + CEconj2;

        a_Cdbo(j,l_index) += CEtors6 + CEconj3; // trouble



        // dcos_theta_ijk

        const F_FLOAT coeff74 = CEtors7 + CEconj4;
@@ -3200,7 +3184,6 @@ template<int NEIGHFLAG, int EVFLAG> 
KOKKOS_INLINE_FUNCTION

void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeHydrogen<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {



  //Kokkos::View<F_FLOAT*[3], typename DAT::t_f_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_f = f;

  auto v_f = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);

  auto a_f = v_f.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -3387,9 +3370,9 @@ void PairReaxCKokkos<DeviceType>::operator()(PairReaxUpdateBond<NEIGHFLAG>, cons 
      }



      if (flag) {

        d_Cdbo(j,k_index) += Cdbo_i;

        d_Cdbopi(j,k_index) += Cdbopi_i;

        d_Cdbopi2(j,k_index) += Cdbopi2_i;

        a_Cdbo(j,k_index) += Cdbo_i;

        a_Cdbopi(j,k_index) += Cdbopi_i;

        a_Cdbopi2(j,k_index) += Cdbopi2_i;

      }

    }

  }
@@ -3401,14 +3384,12 @@ template<int NEIGHFLAG, int EVFLAG> 
KOKKOS_INLINE_FUNCTION

void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeBond1<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {



  //Kokkos::View<F_FLOAT*[3], typename DAT::t_f_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_f = f;

  auto v_f = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);

  auto a_f = v_f.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();

  //Kokkos::View<F_FLOAT*, typename DAT::t_ffloat_1d::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_CdDelta = d_CdDelta;



  auto v_CdDelta = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);

  auto a_CdDelta = v_CdDelta.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();





  F_FLOAT delij[3];

  F_FLOAT p_be1, p_be2, De_s, De_p, De_pp, pow_BOs_be2, exp_be12, CEbo, ebond;
@@ -3544,7 +3525,6 @@ template<int NEIGHFLAG, int EVFLAG> 
KOKKOS_INLINE_FUNCTION

void PairReaxCKokkos<DeviceType>::operator()(PairReaxComputeBond2<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {



  //Kokkos::View<F_FLOAT*[3], typename DAT::t_f_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_f = f;

  auto v_f = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);

  auto a_f = v_f.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -3758,11 +3738,11 @@ void PairReaxCKokkos<DeviceType>::ev_tally(EV_FLOAT_REAX &ev, const int &i, cons 
{

  const int VFLAG = vflag_either;



  // The eatom and vatom arrays are atomic for Half/Thread neighbor style

  //Kokkos::View<E_FLOAT*, typename DAT::t_efloat_1d::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_eatom = d_eatom;

  // The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial



  auto v_eatom = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);

  auto a_eatom = v_eatom.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();

  //Kokkos::View<F_FLOAT*[6], typename DAT::t_virial_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_vatom = d_vatom;



  auto v_vatom = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);

  auto a_vatom = v_vatom.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -3827,10 +3807,10 @@ void PairReaxCKokkos<DeviceType>::e_tally(EV_FLOAT_REAX &ev, const int &i, const 
      const F_FLOAT &epair) const

{



  // The eatom array is atomic for Half/Thread neighbor style

  // The eatom array is duplicated for OpenMP, atomic for CUDA, and neither for Serial





  if (eflag_atom) {

    //Kokkos::View<E_FLOAT*, typename DAT::t_efloat_1d::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_eatom = d_eatom;

    auto v_eatom = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);

    auto a_eatom = v_eatom.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -3848,8 +3828,7 @@ KOKKOS_INLINE_FUNCTION 
void PairReaxCKokkos<DeviceType>::e_tally_single(EV_FLOAT_REAX &ev, const int &i,

      const F_FLOAT &epair) const

{

  // The eatom array is atomic for Half/Thread neighbor style

  //Kokkos::View<E_FLOAT*, typename DAT::t_efloat_1d::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_eatom = d_eatom;

  // The eatom array is duplicated for OpenMP, atomic for CUDA, and neither for Serial

  auto v_eatom = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);

  auto a_eatom = v_eatom.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -3884,7 +3863,6 @@ void PairReaxCKokkos<DeviceType>::v_tally(EV_FLOAT_REAX &ev, const int &i, 
  }



  if (vflag_atom) {

    //Kokkos::View<F_FLOAT*[6], typename DAT::t_virial_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_vatom = d_vatom;

    auto v_vatom = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);

    auto a_vatom = v_vatom.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -3902,8 +3880,7 @@ void PairReaxCKokkos<DeviceType>::v_tally3(EV_FLOAT_REAX &ev, const int &i, cons 
  F_FLOAT *fj, F_FLOAT *fk, F_FLOAT *drij, F_FLOAT *drik) const

{



  // The eatom and vatom arrays are atomic for Half/Thread neighbor style

  //Kokkos::View<F_FLOAT*[6], typename DAT::t_virial_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_vatom = d_vatom;

  // The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial

  auto v_vatom = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);

  auto a_vatom = v_vatom.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();
@@ -3945,7 +3922,8 @@ void PairReaxCKokkos<DeviceType>::v_tally4(EV_FLOAT_REAX &ev, const int &i, cons 
  const int &l, F_FLOAT *fi, F_FLOAT *fj, F_FLOAT *fk, F_FLOAT *dril, F_FLOAT *drjl, F_FLOAT *drkl) const

{



  // The vatom array is atomic for Half/Thread neighbor style

  // The vatom array is duplicated for OpenMP, atomic for CUDA, and neither for Serial



  F_FLOAT v[6];



  v[0] = dril[0]*fi[0] + drjl[0]*fj[0] + drkl[0]*fk[0];
@@ -3965,7 +3943,6 @@ void PairReaxCKokkos<DeviceType>::v_tally4(EV_FLOAT_REAX &ev, const int &i, cons 
  }



  if (vflag_atom) {

    //Kokkos::View<F_FLOAT*[6], typename DAT::t_virial_array::array_layout,DeviceType,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_vatom = d_vatom;

    auto v_vatom = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);

    auto a_vatom = v_vatom.template access<AtomicDup<NEIGHFLAG,DeviceType>::value>();

CRA Git | Maintained and supported by SUSTech CRA and CCSE