Commit 6dd8efd0 authored by Stan Moore's avatar Stan Moore
Browse files

Add if statements

parent dbc308f3

src/KOKKOS/fix_qeq_reax_kokkos.cpp

+32 −14
Original line number Diff line number Diff line
@@ -249,8 +249,12 @@ void FixQEqReaxKokkos<DeviceType>::pre_force(int vflag) 
  k_t.template modify<LMPHostType>();

  k_t.template sync<DeviceType>();



  if (neighflag != FULL)

    red_o = Kokkos::Experimental::create_scatter_view<> (d_o); // allocate duplicated memory

  need_dup = lmp->kokkos->need_dup<DeviceType>();



  if (need_dup)

    dup_o = Kokkos::Experimental::create_scatter_view<Kokkos::Experimental::ScatterSum, Kokkos::Experimental::ScatterDuplicated> (d_o); // allocate duplicated memory

  else

    ndup_o = Kokkos::Experimental::create_scatter_view<Kokkos::Experimental::ScatterSum, Kokkos::Experimental::ScatterNonDuplicated> (d_o);



  // 1st cg solve over b_s, s

  cg_solve1();
@@ -258,8 +262,8 @@ void FixQEqReaxKokkos<DeviceType>::pre_force(int vflag) 
  // 2nd cg solve over b_t, t

  cg_solve2();



  if (neighflag != FULL)

    red_o = decltype(red_o)(); // free duplicated memory

  if (need_dup)

    dup_o = decltype(dup_o)(); // free duplicated memory



  // calculate_Q();

  calculate_q();
@@ -489,7 +493,9 @@ void FixQEqReaxKokkos<DeviceType>::cg_solve1() 
    Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagZeroQGhosts>(nlocal,nlocal+atom->nghost),*this);

    FixQEqReaxKokkosSparse13Functor<DeviceType,HALFTHREAD> sparse13_functor(this);

    Kokkos::parallel_for(inum,sparse13_functor);

    Kokkos::Experimental::contribute(d_o, red_o);

    if (need_dup)

        if (need_dup)

    Kokkos::Experimental::contribute(d_o, dup_o);

  } else {

    Kokkos::parallel_for(Kokkos::TeamPolicy <DeviceType, TagSparseMatvec1> (inum, teamsize), *this);

  }
@@ -537,10 +543,13 @@ void FixQEqReaxKokkos<DeviceType>::cg_solve1() 
    Kokkos::parallel_for(inum,sparse22_functor);

    if (neighflag != FULL) {

      Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagZeroQGhosts>(nlocal,nlocal+atom->nghost),*this);

      red_o.reset_except(d_o);

      if (need_dup)

        dup_o.reset_except(d_o);

      FixQEqReaxKokkosSparse23Functor<DeviceType,HALFTHREAD> sparse23_functor(this);

      Kokkos::parallel_for(inum,sparse23_functor);

      Kokkos::Experimental::contribute(d_o, red_o);

      if (need_dup)

          if (need_dup)

    Kokkos::Experimental::contribute(d_o, dup_o);

    } else {

      Kokkos::parallel_for(Kokkos::TeamPolicy <DeviceType, TagSparseMatvec2> (inum, teamsize), *this);

    }
@@ -610,10 +619,13 @@ void FixQEqReaxKokkos<DeviceType>::cg_solve2() 
  Kokkos::parallel_for(inum,sparse32_functor);

  if (neighflag != FULL) {

    Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagZeroQGhosts>(nlocal,nlocal+atom->nghost),*this);

    red_o.reset_except(d_o);

    if (need_dup)

      dup_o.reset_except(d_o);

    FixQEqReaxKokkosSparse33Functor<DeviceType,HALFTHREAD> sparse33_functor(this);

    Kokkos::parallel_for(inum,sparse33_functor);

    Kokkos::Experimental::contribute(d_o, red_o);

    if (need_dup)

        if (need_dup)

    Kokkos::Experimental::contribute(d_o, dup_o);

  } else {

    Kokkos::parallel_for(Kokkos::TeamPolicy <DeviceType, TagSparseMatvec3> (inum, teamsize), *this);

  }
@@ -661,10 +673,13 @@ void FixQEqReaxKokkos<DeviceType>::cg_solve2() 
    Kokkos::parallel_for(inum,sparse22_functor);

    if (neighflag != FULL) {

      Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagZeroQGhosts>(nlocal,nlocal+atom->nghost),*this);

      red_o.reset_except(d_o);

      if (need_dup)

        dup_o.reset_except(d_o);

      FixQEqReaxKokkosSparse23Functor<DeviceType,HALFTHREAD> sparse23_functor(this);

      Kokkos::parallel_for(inum,sparse23_functor);

      Kokkos::Experimental::contribute(d_o, red_o);

      if (need_dup)

          if (need_dup)

    Kokkos::Experimental::contribute(d_o, dup_o);

    } else {

      Kokkos::parallel_for(Kokkos::TeamPolicy <DeviceType, TagSparseMatvec2> (inum, teamsize), *this);

    }
@@ -777,7 +792,8 @@ KOKKOS_INLINE_FUNCTION 
void FixQEqReaxKokkos<DeviceType>::sparse13_item(int ii) const

{

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

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

  auto v_o = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);

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



  const int i = d_ilist[ii];

  if (mask[i] & groupbit) {
@@ -829,7 +845,8 @@ KOKKOS_INLINE_FUNCTION 
void FixQEqReaxKokkos<DeviceType>::sparse23_item(int ii) const

{

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

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

  auto v_o = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);

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



  const int i = d_ilist[ii];

  if (mask[i] & groupbit) {
@@ -888,7 +905,8 @@ KOKKOS_INLINE_FUNCTION 
void FixQEqReaxKokkos<DeviceType>::sparse33_item(int ii) const

{

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

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

  auto v_o = ScatterViewHelper<NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);

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



  const int i = d_ilist[ii];

  if (mask[i] & groupbit) {

src/KOKKOS/fix_qeq_reax_kokkos.h

+3 −1
Original line number Diff line number Diff line
@@ -149,6 +149,7 @@ class FixQEqReaxKokkos : public FixQEqReax { 
 private:

  int inum;

  int allocated_flag;

  int need_dup;



  typedef Kokkos::DualView<int***,DeviceType> tdual_int_1d;

  Kokkos::DualView<params_qeq*,Kokkos::LayoutRight,DeviceType> k_params;
@@ -193,7 +194,8 @@ class FixQEqReaxKokkos : public FixQEqReax { 
  HAT::t_ffloat_2d h_s_hist, h_t_hist;

  typename AT::t_ffloat_2d_randomread r_s_hist, r_t_hist;



  Kokkos::Experimental::ScatterView<F_FLOAT*, typename AT::t_ffloat_1d::array_layout, DeviceType> red_o;

  Kokkos::Experimental::ScatterView<F_FLOAT*, typename AT::t_ffloat_1d::array_layout, DeviceType, Kokkos::Experimental::ScatterSum, Kokkos::Experimental::ScatterDuplicated> dup_o;

  Kokkos::Experimental::ScatterView<F_FLOAT*, typename AT::t_ffloat_1d::array_layout, DeviceType, Kokkos::Experimental::ScatterSum, Kokkos::Experimental::ScatterNonDuplicated> ndup_o;



  void init_shielding_k();

  void init_hist();

src/KOKKOS/kokkos.h

+13 −0
Original line number Diff line number Diff line
@@ -16,6 +16,7 @@ 


#include "pointers.h"

#include "kokkos_type.h"

#include "pair_kokkos.h"



namespace LAMMPS_NS {
@@ -39,6 +40,18 @@ class KokkosLMP : protected Pointers { 
  ~KokkosLMP();

  void accelerator(int, char **);

  int neigh_count(int);



  template<class DeviceType>

  int need_dup()

  {

    int value = 0;

  

    if (neighflag == HALFTHREAD)

      value = NeedDup<HALFTHREAD,DeviceType>::value;

  

    return value;

  };



 private:

  static void my_signal_handler(int);

};

src/KOKKOS/pair_kokkos.h

+25 −4
Original line number Diff line number Diff line
@@ -95,8 +95,8 @@ struct AtomicDup<HALFTHREAD,Kokkos::Threads> { 


// Determine duplication traits for force array

// Use duplication when running threaded and not using atomics

template<class DeviceType>

struct DupF {

template<int NEIGHFLAG, class DeviceType>

struct NeedDup {

  enum {value = Kokkos::Experimental::ScatterNonDuplicated};

};
@@ -104,20 +104,41 @@ struct DupF { 


#ifdef KOKKOS_ENABLE_OPENMP

template<>

struct DupF<Kokkos::OpenMP> {

struct NeedDup<HALFTHREAD,Kokkos::OpenMP> {

  enum {value = Kokkos::Experimental::ScatterDuplicated};

};

#endif



#ifdef KOKKOS_ENABLE_THREADS

template<>

struct DupF<Kokkos::Threads> {

struct NeedDup<HALFTHREAD,Kokkos::Threads> {

  enum {value = Kokkos::Experimental::ScatterDuplicated};

};

#endif



#endif



template<int value, typename T1, typename T2>

class ScatterViewHelper {};



template<typename T1, typename T2>

class ScatterViewHelper<Kokkos::Experimental::ScatterDuplicated,T1,T2> {

public:

  KOKKOS_INLINE_FUNCTION

  static T1 get(const T1 &dup, const T2 &nondup) {

    return dup;

  }

};



template<typename T1, typename T2>

class ScatterViewHelper<Kokkos::Experimental::ScatterNonDuplicated,T1,T2> {

public:

  KOKKOS_INLINE_FUNCTION

  static T2 get(const T1 &dup, const T2 &nondup) {

    return nondup;

  }

};





//Specialisation for Neighborlist types Half, HalfThread, Full

template <class PairStyle, int NEIGHFLAG, bool STACKPARAMS, class Specialisation = void>

src/KOKKOS/pair_reaxc_kokkos.cpp

+134 −78

File changed.

Preview size limit exceeded, changes collapsed.

CRA Git | Maintained and supported by SUSTech CRA and CCSE