XGCa
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Pages
species.hpp
Go to the documentation of this file.
1 #ifndef SPECIES_HPP
2 #define SPECIES_HPP
3 #include <Cabana_AoSoA.hpp>
4 #include <Cabana_DeepCopy.hpp>
5 #include <Kokkos_Core.hpp>
6 #include "NamelistReader.hpp"
7 #include "timer_macro.hpp"
8 #include "magnetic_field.hpp"
9 #include "grid.hpp"
10 #include "domain_decomposition.hpp"
11 #include "particles.hpp"
12 #include "space_settings.hpp"
13 #include "distribution.hpp"
14 #include "profile.hpp"
15 #include "gyro_avg_mat.hpp"
17 #include "basic_physics.hpp"
18 #include "memory_prediction.hpp"
19 
20 extern "C" void set_spall_num_and_ptr(int idx, int n_ptl, int n_vecs, VecParticles* ptl);
21 extern "C" void set_min_max_num(int isp, int n_ptl);
22 extern "C" void adjust_n_ptl_for_core_ptl(int* n_ptl);
23 
28  return false;
29 }
30 
35  return false;
36 }
37 
38 // Used for Cabana slices (getting one particle property at a time)
39 namespace PtlSlice{
40 #ifdef ESC_PTL
41 enum{Ph=0,Ct,Gid,Flag};
42 #else
43 enum{Ph=0,Ct,Gid};
44 #endif
45 }
46 
47 struct PtlMvmt{
48  // Options for particle location management when looping over particles
49  enum SendOpt{
50  NoSend=0,
53  };
54 
55  enum ReturnOpt{
59  };
60 
63 
64  PtlMvmt(SendOpt send_opt, ReturnOpt return_opt) : send_opt(send_opt), return_opt(return_opt){}
65 };
66 
70 };
71 
72 // Species class
73 template<class Device>
74 class Species{
75  public:
76 
77  int idx;
78  bool is_electron;
79  bool is_adiabatic;
82  double mass;
83  double charge;
84  double charge_eu;
85  double c_m;
86  double c2_2m;
87 
88  bool is_deltaf;
89 
90  int ncycles;
92 
94  int n_ptl;
95  Cabana::AoSoA<ParticleDataTypes,HostType,VEC_LEN> particles;
96 
97  // Device particles
98  Cabana::AoSoA<ParticleDataTypes,Device,VEC_LEN> particles_d;
99 
101 
104 
105  /*** Could be its own class inside species? ***/
109 
110  // phase0 (for ion restoration)
111  Cabana::AoSoA<PhaseDataTypes,HostType,VEC_LEN> phase0;
112  Cabana::AoSoA<PhaseDataTypes,Device,VEC_LEN> phase0_d;
113 
114  // For electron restoration
115  Cabana::AoSoA<ParticleDataTypes,HostType,VEC_LEN> backup_particles;
116  int n_backup_particles; // Number of particles stored in backup_particles (can't be deduced from its size due to buffer)
117  /****/
118 
120 
122 
123  Eq::Profile<Device> eq_temp; // Equilibrium temperature
124  Eq::Profile<Device> eq_den; // Equilibrium density
125  Eq::Profile<Device> eq_flow; // Equilibrium flow
126  int eq_flow_type; // Type of Equilibirum flow
127 
129 
130  /*** Constructors ***/
131 
132  Species(int idx_in, int nonadiabatic_idx_in, bool is_electron_in, bool is_adiabatic_in, KinType kintype_in, double mass_in, double charge_in, double charge_eu_in, bool is_deltaf_in,
133  int ncycles_in);
134 
135  Species(NLReader::NamelistReader& nlr, const Grid<DeviceType> &grid, const MagneticField<DeviceType> &magnetic_field, const DomainDecomposition<DeviceType>& pol_decomp, int idx_in, int nonadiabatic_idx_in);
136 
137  // Electron or ion default constructor
138  Species(SpeciesType sp_type, int n_ptl)
139  : idx(sp_type==ELECTRON ? 0 : 1),
140  is_electron(sp_type==ELECTRON),
141  mass(is_electron ? 3.344e-30 : PROTON_MASS),
143  charge_eu(is_electron ? -1.0 : 1.0),
144  is_deltaf(true),
145  is_adiabatic(false),
146  nonadiabatic_idx(idx), // Since is_adiabatic is false above
148  ncycles(is_electron ? 70 : 1),
149  c_m(charge/mass),
150  c2_2m(0.5*charge*charge/mass),
152  n_ptl(n_ptl),
153  backup_particles("backup_particles", 0),
154  particles("particles", add_vec_buffer(n_ptl)),
157  eq_temp(1.0e3,-0.1),
158  eq_den(1.0e19,-0.1),
159  eq_flow_type(2),
160  owns_particles_d(false),
164 
165  // Special constructor for tests that involve tracking particles in memory
166  // The idea is to use these particles to test functions that reorder particles,
167  // e.g. sort, shift, and cleaning
168  Species(int n_ptl_in)
169  : n_ptl(n_ptl_in),
170  is_electron(true),
171  is_adiabatic(false),
172  particles("particles", add_vec_buffer(n_ptl_in)),
174  owns_particles_d(false),
179  {
180 
181  // Slice particle properties
182  auto ph = Cabana::slice<PtlSlice::Ph>(particles);
183  auto ct = Cabana::slice<PtlSlice::Ct>(particles);
184  auto gid = Cabana::slice<PtlSlice::Gid>(particles);
185 #ifdef ESC_PTL
186  auto flag = Cabana::slice<PtlSlice::Flag>(particles);
187 #endif
188 
189  // Offset gid if using MPI
190 #ifdef USE_MPI
191  long long int gid_offset = n_ptl*SML_COMM_RANK;
192 #else
193  long long int gid_offset = 0;
194 #endif
195 
196  // Assign trackable values
197  for (int i=0;i<n_ptl;i++){
198  // Set GID in order
199  gid(i) = gid_offset + i+1; // 1-indexed
200 
201  // Value of properties is gid + 0.1*(property index)
202  // First particle: (1.0, 1.1, ... 1.8)
203  // Second particle: (2.0, 2.1, ... 2.8)
204  for (int j=0;j<6;j++) ph(i, j) = gid(i) + (j)*0.1;
205  for (int j=0;j<3;j++) ct(i, j) = gid(i) + (j+6)*0.1;
206  }
207 
208  // Buffer particles: same but with gid = -1
209  if(n_ptl>0){
210  for (int i=n_ptl;i<add_vec_buffer(n_ptl);i++){
211  gid(i) = -1;
212  for (int j=0;j<6;j++) ph(i, j) = gid(i) + (j)*0.1;
213  for (int j=0;j<3;j++) ct(i, j) = gid(i) + (j+6)*0.1;
214  }
215  }
216  }
217 
218  static std::vector<MemoryPrediction> estimate_memory_usage(NLReader::NamelistReader& nlr, const Grid<DeviceType> &grid, const DomainDecomposition<DeviceType>& pol_decomp, int species_idx);
219 
220  static int get_initial_n_ptl(NLReader::NamelistReader& nlr, const Grid<DeviceType> &grid, const DomainDecomposition<DeviceType>& pol_decomp, int sml_special, int species_idx, bool verbose);
221 
222  void resize_particles(int new_n_ptl){
223  n_ptl = new_n_ptl;
224 
225 #ifndef USE_GPU
226  // If CPU-only, particles_d points to the same location as particles. If particles is resized, then Cabana will not deallocate the first allocation
227  // since it is still used by particles_d. So, reset particles_d before resize, and point it back to particles only afterwards
228  particles_d = Cabana::AoSoA<ParticleDataTypes,Device,VEC_LEN>();
229 #endif
230 
231  particles.reserve(minimum_ptl_reservation); // Can only raise reservation (no-op if AoSoA is already larger)
232  particles.resize(add_vec_buffer(n_ptl));
233 
234 #ifndef USE_GPU
235  // Point particles_d back to particles after resize
237 #endif
238 
240  }
241 
243 #ifdef USE_GPU
244  particles.reserve(minimum_ptl_reservation); // Can only raise reservation (no-op if AoSoA is already larger)
245  particles.resize(particles_d.size());
246 #else
247  // Point particles back to particles_d after resize
249 #endif
250 
252  }
253 
254  /* If using CPU-only, then "device" particles are a shallow copy of host particles so that
255  * there is no unnecessary duplication. When "device" particles are resized, Cabana will keep the
256  * original allocation if there is a second reference (i.e. host particles).
257  * To resolve this, we free the host particles here so that there is no second reference
258  * */
260  particles = Cabana::AoSoA<ParticleDataTypes,HostType,VEC_LEN>();
261  }
262 
263  /* Resizes device particles if on GPU, or just creates a shallow copy if CPU only
264  * */
266  if(!owns_particles_d) exit_XGC("\nSpecies tried to resize device particles, but doesn't own the device array.");
267 
268 #ifdef USE_GPU
269  particles_d.reserve(minimum_ptl_reservation); // Can only raise reservation (no-op if AoSoA is already larger)
270  // Resize device particles to match n particles
272 #else
273  // If kernels are on CPU, do shallow copy
275 #endif
276  }
277 
278  /* Resizes device particles
279  * */
280  void resize_device_particles(int new_n_ptl){
281  if(!owns_particles_d) exit_XGC("\nSpecies tried to resize device particles, but doesn't own the device array.");
282 
283  n_ptl = new_n_ptl;
284 
285  particles_d.reserve(minimum_ptl_reservation); // Can only raise reservation (no-op if AoSoA is already larger)
286 
287  // Resize device particles to match host particles
289  }
290 
291  /* Copies particles to device - deep copy if using GPU, otherwise shallow copy
292  * Also takes the opportunity to set the buffer particles to realistic values
293  * */
295  if(!owns_particles_d) exit_XGC("\nSpecies tried to copy particles to device, but doesn't own the device array.");
296 
297 #ifdef USE_GPU
298  // Copy to device
299  Cabana::deep_copy(particles_d, particles);
300 #else
301  // No operation required if CPU-only
302 #endif
303 
304  // Copy last particle to fill remainder of trailing vector in AoSoA
306  }
307 
308  /* Copies particles from device - deep copy if using GPU, otherwise no copy is necessary
309  * */
311  if(!owns_particles_d) exit_XGC("\nSpecies tried to copy particles from device, but doesn't own the device array.");
312 
313 #ifdef USE_GPU
314  // Copy particles to host
315  Cabana::deep_copy(particles, particles_d);
316 #else
317  // No operation required if CPU-only
318 #endif
319  }
320 
321  /* Copies particles to device if they are resident on the device
322  * */
325  }
326 
327  /* Copies particles from device if they are resident on the device
328  * */
331  }
332 
333  /* Copies particles to device if they are NOT resident on the device
334  * */
337  }
338 
339  /* Copies particles from device if they are NOT resident on the device
340  * */
343  }
344 
353  if (n_ptl>0){
354  int last_ptl_index = n_ptl - 1;
355  auto ph = Cabana::slice<PtlSlice::Ph>(particles_d);
356  auto ct = Cabana::slice<PtlSlice::Ct>(particles_d);
357  auto gid = Cabana::slice<PtlSlice::Gid>(particles_d);
358 #ifdef ESC_PTL
359  auto flag = Cabana::slice<PtlSlice::Flag>(particles_d);
360 #endif
361 
362  Kokkos::parallel_for("set_buffer_particles_d", Kokkos::RangePolicy<ExSpace>( n_ptl, add_vec_buffer(n_ptl) ), KOKKOS_LAMBDA( const int i ){
363  // Buffer particles: same as last particle, gid = -1
364  for (int j=0;j<6;j++) ph(i, j) = ph(last_ptl_index, j);
365  for (int j=0;j<3;j++) ct(i, j) = ct(last_ptl_index, j);
366  gid(i) = -1;
367 #ifdef ESC_PTL
368  flag(i) = flag(last_ptl_index);
369 #endif
370  });
371  }
372  }
373 
382  if (n_ptl>0){
383  int last_ptl_index = n_ptl - 1;
384  auto ph = Cabana::slice<PtlSlice::Ph>(phase0_d);
385 
386  Kokkos::parallel_for("set_buffer_phase0", Kokkos::RangePolicy<ExSpace>( n_ptl, add_vec_buffer(n_ptl) ), KOKKOS_LAMBDA( const int i ){
387  // copy final real particle
388  for (int j=0;j<6;j++) ph(i, j) = ph(last_ptl_index, j);
389  });
390  }
391  }
392 
393  // Options for custom launch bounds since kokkos defaults are suboptimal for electron push kernel
394  enum class LaunchBounds{
395  Default,
396  Custom
397  };
398 
404  template<typename F>
405  inline void for_all_particles(const std::string label, F lambda_func) const {
406  Kokkos::RangePolicy<ExSpace> particle_range_policy( 0, p_range<DeviceType>(n_ptl) );
407  Kokkos::parallel_for(label, Opt::require(particle_range_policy, Async), lambda_func);
408  }
409 
410  inline void back_up_SoA(Cabana::AoSoA<ParticleDataTypes,Device,VEC_LEN>& backup_SoA, int offset, int n) const{
411  auto ph_b = Cabana::slice<PtlSlice::Ph>(backup_SoA);
412  auto ct_b = Cabana::slice<PtlSlice::Ct>(backup_SoA);
413  auto gid_b = Cabana::slice<PtlSlice::Gid>(backup_SoA);
414 #ifdef ESC_PTL
415  auto flag_b = Cabana::slice<PtlSlice::Flag>(backup_SoA);
416 #endif
417 
418  auto ph = Cabana::slice<PtlSlice::Ph>(particles_d);
419  auto ct = Cabana::slice<PtlSlice::Ct>(particles_d);
420  auto gid = Cabana::slice<PtlSlice::Gid>(particles_d);
421 #ifdef ESC_PTL
422  auto flag = Cabana::slice<PtlSlice::Flag>(particles_d);
423 #endif
424 
425  Kokkos::parallel_for("backup_first_soa", Kokkos::RangePolicy<ExSpace>( 0, n ), KOKKOS_LAMBDA( const int i ){
426  int i_offset = i + offset;
427  // Make backup copy
428  for (int j=0;j<6;j++) ph_b(i, j) = ph(i_offset, j);
429  for (int j=0;j<3;j++) ct_b(i, j) = ct(i_offset, j);
430  gid_b(i) = gid(i_offset);
431 #ifdef ESC_PTL
432  flag_b(i) = flag(i_offset);
433 #endif
434  // Deactivate
435  gid(i_offset) = -1;
436  });
437  }
438 
439  inline void restore_backup_SoA(Cabana::AoSoA<ParticleDataTypes,Device,VEC_LEN>& backup_SoA, int offset, int n) const{
440  auto ph_b = Cabana::slice<PtlSlice::Ph>(backup_SoA);
441  auto ct_b = Cabana::slice<PtlSlice::Ct>(backup_SoA);
442  auto gid_b = Cabana::slice<PtlSlice::Gid>(backup_SoA);
443 #ifdef ESC_PTL
444  auto flag_b = Cabana::slice<PtlSlice::Flag>(backup_SoA);
445 #endif
446 
447  auto ph = Cabana::slice<PtlSlice::Ph>(particles_d);
448  auto ct = Cabana::slice<PtlSlice::Ct>(particles_d);
449  auto gid = Cabana::slice<PtlSlice::Gid>(particles_d);
450 #ifdef ESC_PTL
451  auto flag = Cabana::slice<PtlSlice::Flag>(particles_d);
452 #endif
453 
454  Kokkos::parallel_for("backup_first_soa", Kokkos::RangePolicy<ExSpace>( 0, n ), KOKKOS_LAMBDA( const int i ){
455  int i_offset = i + offset;
456  // Restore from backup copy
457  for (int j=0;j<6;j++) ph(i_offset, j) = ph_b(i, j);
458  for (int j=0;j<3;j++) ct(i_offset, j) = ct_b(i, j);
459  gid(i_offset) = gid_b(i);
460 #ifdef ESC_PTL
461  flag(i_offset) = flag_b(i);
462 #endif
463  });
464  }
465 
471  template<typename F>
472  inline void for_particle_range(int begin_idx, int end_idx, const std::string label, F lambda_func) const {
473  if(end_idx <= begin_idx) return; // Return if range is 0 or less
474 
475  // Still need the subset to line up with the AoSoA vector length
476  int first_soa = begin_idx/VEC_LEN;
477  int n_other_ptl_in_first_soa = begin_idx - first_soa*VEC_LEN;
478  bool first_soa_is_partial = (n_other_ptl_in_first_soa>0);
479  int last_soa = (end_idx-1)/VEC_LEN;
480  int n_other_ptl_in_last_soa = (last_soa+1)*VEC_LEN - end_idx;
481  bool last_soa_is_partial = (n_other_ptl_in_last_soa>0);
482 
483 #ifdef USE_GPU
484  int first_item_in_shifted_range = first_soa*VEC_LEN;
485 #else
486  int first_item_in_shifted_range = first_soa;
487 #endif
488 
489  Cabana::AoSoA<ParticleDataTypes,Device,VEC_LEN> ptl_first_soa;
490  Cabana::AoSoA<ParticleDataTypes,Device,VEC_LEN> ptl_last_soa;
491  if(first_soa_is_partial){
492  // Make a backup of the first SoA and set the particles_d GIDs to -1
493  ptl_first_soa = Cabana::AoSoA<ParticleDataTypes,Device,VEC_LEN>("ptl_first_soa", n_other_ptl_in_first_soa);
494  back_up_SoA(ptl_first_soa, first_soa*VEC_LEN, n_other_ptl_in_first_soa);
495  }
496  if(last_soa_is_partial){
497  // Make a backup of the last SoA and set the particles_d GIDs to -1
498  ptl_last_soa = Cabana::AoSoA<ParticleDataTypes,Device,VEC_LEN>("ptl_last_soa", n_other_ptl_in_last_soa);
499  back_up_SoA(ptl_last_soa, end_idx, n_other_ptl_in_last_soa);
500  }
501 
502  // Finally, do the parallel_for
503  Kokkos::RangePolicy<ExSpace> particle_range_policy( first_item_in_shifted_range, p_range<DeviceType>(end_idx) );
504  Kokkos::parallel_for(label, Opt::require(particle_range_policy, Async), lambda_func);
505 
506  // Restore from backup
507  if(first_soa_is_partial){
508  restore_backup_SoA(ptl_first_soa, first_soa*VEC_LEN, n_other_ptl_in_first_soa);
509  }
510  if(last_soa_is_partial){
511  restore_backup_SoA(ptl_last_soa, end_idx, n_other_ptl_in_last_soa);
512  }
513  }
514 
522  template<typename F>
523  inline void for_all_particles(const std::string label, F lambda_func,
524  const PtlMvmt mvmt, LaunchBounds launch_bounds=LaunchBounds::Default) {
525  if(!owns_particles_d) exit_XGC("\nSpecies tried to loop over particles on device, but doesn't own the device array.");
526 
527  bool use_streaming = stream_particles;
528 #ifndef USE_STREAMS
529  use_streaming = false; // Just to be safe, turn streams off here
530 #endif
531 
532  bool send_ptl = ( mvmt.send_opt==PtlMvmt::Send ||
534  bool return_ptl = ( mvmt.return_opt==PtlMvmt::Return ||
536 
537  // Don't need to stream if particles are already present and don't need to be returned
538  if((!send_ptl) && (!return_ptl)) use_streaming = false;
539 
540  if(use_streaming){
541 #ifdef USE_STREAMS
542  Streamed::Option stream_option = Streamed::Normal; // Send to device and back
543  if(!send_ptl) stream_option = Streamed::NoSend;
544  if(!return_ptl) stream_option = Streamed::NoReturn;
545 
546  // Execute streaming parallel_for
547  Streamed::parallel_for(label, n_ptl, lambda_func, stream_option, particles, particles_d);
548 #endif
549  }else{
550  if(send_ptl) TIMER("copy_ptl_to_device",copy_particles_to_device() );
551 
552  if (launch_bounds==LaunchBounds::Custom) {
553 #ifdef USE_EPUSH_LAUNCH_BOUNDS
554 # if !defined(PUSH_MAX_THREADS_PER_BLOCK) || !defined(PUSH_MIN_WARPS_PER_EU)
555 # error "USE_EPUSH_LAUNCH_BOUNDS requires PUSH_MAX_THREADS_PER_BLOCK and PUSH_MIN_WARPS_PER_EU to be defined"
556 # endif
557  Kokkos::RangePolicy<ExSpace, Kokkos::LaunchBounds<PUSH_MAX_THREADS_PER_BLOCK, PUSH_MIN_WARPS_PER_EU>>
558  particle_range_policy( 0, p_range<DeviceType>(n_ptl) );
559  Kokkos::parallel_for(label, Opt::require(particle_range_policy, Async), lambda_func);
560 #else
561  exit_XGC("\nERROR: LaunchBounds::Custom specified, but USE_EPUSH_LAUNCH_BOUNDS is not defined\n");
562 #endif
563  } else {
564  Kokkos::RangePolicy<ExSpace>
565  particle_range_policy( 0, p_range<DeviceType>(n_ptl) );
566  Kokkos::parallel_for(label, Opt::require(particle_range_policy, Async), lambda_func);
567  }
568 
569  if(return_ptl) TIMER("copy_ptl_from_device", copy_particles_from_device() );
570  }
571  }
572 
573  KOKKOS_INLINE_FUNCTION VecParticles* ptl() const{
574  return (VecParticles*)(&particles_d.access(0));
575  }
576 
577  KOKKOS_INLINE_FUNCTION VecPhase* ph0() const{
578  return (VecPhase*)(&phase0_d.access(0));
579  }
580 
582  for_all_particles("copy_to_phase0", KOKKOS_LAMBDA( const int idx ){
583  AoSoAIndices<DeviceType> inds(idx);
584  VecParticles* ptl_loc = species.ptl();
585  VecPhase* ph0_loc = species.ph0();
586  for (int i_simd = 0; i_simd<SIMD_SIZE; i_simd++){
587  int p_vec = inds.a + i_simd;
588  ph0_loc[inds.s].r[p_vec] = ptl_loc[inds.s].ph.r[p_vec];
589  ph0_loc[inds.s].z[p_vec] = ptl_loc[inds.s].ph.z[p_vec];
590  ph0_loc[inds.s].phi[p_vec] = ptl_loc[inds.s].ph.phi[p_vec];
591  ph0_loc[inds.s].rho[p_vec] = ptl_loc[inds.s].ph.rho[p_vec];
592  ph0_loc[inds.s].w1[p_vec] = ptl_loc[inds.s].ph.w1[p_vec];
593  ph0_loc[inds.s].w2[p_vec] = ptl_loc[inds.s].ph.w2[p_vec];
594  }
595  });
596  }
597 
600  // If particles are resident on the device, then use the host particle allocation as the backup
601  // Otherwise, resize the backup_particle array
604  }else{
605  backup_particles.resize(particles_d.size());
606  }
607 
608  // Copy particle data to backup
609  Cabana::deep_copy(backup_particles, particles_d);
611  } else {
612  // For ions, save phase to phase0
613  phase0_d.resize(particles_d.size());
614  copy_to_phase0(*this); // Separate function to avoid implicit copy into lambda
615  }
617  }
618 
621  // If particles are resident on device, don't need to resize host particles since they are already used as the backup.
622  // If not, resize host particle array to fit backup particles
626  }else{
627  // If particles are not resident on device, resize host particle array to fit backup particles
629  }
630 
631  // Resize device particle array to fit backed up particles
633 
634  // Restore particle data from backup
635  Cabana::deep_copy(particles_d, backup_particles);
636 
638  // If the backup particles are simply pointing to the host particles, then
639  // point them to their own 0-sized allocation when finished using them
640  // so that they don't make a copy when host particles get resized
641  backup_particles = Cabana::AoSoA<ParticleDataTypes,HostType,VEC_LEN>("backup_particles", 0);
642  }
643  } else {
644  // When ipc==2, copy phase0 to device
645  // First, resize device phase0 and reset pointer
646  phase0_d.resize(phase0.size());
647 
648  // Next copy data to phase0 on device
649  Cabana::deep_copy(phase0_d, phase0);
650 
651  // Fill buffer with realistic ptl data
653  }
654  particles_are_backed_up = false;
655  }
656 
657  KOKKOS_INLINE_FUNCTION void restore_phase_from_phase0(const AoSoAIndices<Device>& inds, SimdParticles& part_one) const {
658  VecPhase* ph0_loc = ph0();
659  for (int i_simd = 0; i_simd<SIMD_SIZE; i_simd++){
660  int p_vec = inds.a + i_simd;
661  part_one.ph.r[i_simd] = ph0_loc[inds.s].r[p_vec];
662  part_one.ph.z[i_simd] = ph0_loc[inds.s].z[p_vec];
663  part_one.ph.phi[i_simd] = ph0_loc[inds.s].phi[p_vec];
664  part_one.ph.rho[i_simd] = ph0_loc[inds.s].rho[p_vec];
665  part_one.ph.w1[i_simd] = ph0_loc[inds.s].w1[p_vec];
666  part_one.ph.w2[i_simd] = ph0_loc[inds.s].w2[p_vec];
667  }
668  }
669 
670  long long int get_total_n_ptl(){
671 #ifdef USE_MPI
672  long long int tmp_n_ptl = n_ptl;
673  long long int out_n_ptl = 0;
674  MPI_Allreduce(&tmp_n_ptl, &out_n_ptl, 1, MPI_LONG_LONG_INT, MPI_SUM, SML_COMM_WORLD);
675  return out_n_ptl;
676 #else
677  return (long long int)(n_ptl);
678 #endif
679  }
680 
682 #ifdef USE_MPI
683  int tmp_n_ptl = n_ptl;
684  int out_n_ptl = 0;
685  MPI_Allreduce(&tmp_n_ptl, &out_n_ptl, 1, MPI_INT, MPI_MAX, SML_COMM_WORLD);
686  return out_n_ptl;
687 #else
688  return n_ptl;
689 #endif
690  }
691 
692  // Gets the gyro_radius of a species based on equilibrium temperature
693  // inode is the LOCAL (poloidally decomposed) grid node index to get temperature
694  // smu_n is the normalized sqrt(mu)
695  // bfield is the magnetic field at inode
696  KOKKOS_INLINE_FUNCTION double get_f0_eq_gyro_radius(int inode, double smu_n, double bfield) const{
697  // Should replace UNIT_CHARGE*charge_eu with charge(?)
698  return smu_n*sqrt(mass*f0.temp_ev(inode)*EV_2_J) / (UNIT_CHARGE*charge_eu*bfield);
699  }
700 
701  // Gets the equilibrium thermal velocity of a species based on f0 temperature
702  // inode is the GLOBAL node index to get temperature
703  KOKKOS_INLINE_FUNCTION double get_f0_eq_thermal_velocity(int inode) const{
704  return thermal_velocity(mass, f0.temp_global(inode));
705  }
706 
707  // Gets the equilibrium thermal velocity of a species based on f0 temperature, on device
708  // inode is the local node index to get temperature
709  KOKKOS_INLINE_FUNCTION double get_f0_eq_thermal_velocity_lnode(int inode) const{
710  return thermal_velocity(mass, f0.temp_ev(inode));
711  }
712 
713  // Gets the equilibrium thermal velocity of a species based on f0 temperature, on host
714  // inode is the local node index to get temperature
715  KOKKOS_INLINE_FUNCTION double get_f0_eq_thermal_velocity_lnode_h(int inode) const{
716  return thermal_velocity(mass, f0.temp_ev_h(inode));
717  }
718 
719  // Get species velocity
720  KOKKOS_INLINE_FUNCTION void get_particle_velocity_and_nearest_node(const Grid<DeviceType>& grid, const MagneticField<DeviceType>& magnetic_field, const DomainDecomposition<DeviceType>& pol_decomp, SimdParticles& part, Simd<double>& smu, Simd<double>& vp, Simd<int>& nearest_node, Simd<bool>& not_in_triangle, Simd<bool>& not_in_poloidal_domain) const{
721 
722  // This modulo surely doesnt need to be here (at least, should be elsewhere).
723  // Modulo phi coordinate
724  grid.wedge_modulo_phi(part.ph.phi);
725 
727  grid.get_grid_weights(magnetic_field, part.ph.v(), grid_wts0);
728 
729  // Output argument
730  for (int i_simd = 0; i_simd<SIMD_SIZE; i_simd++){
731  not_in_triangle[i_simd] = !grid_wts0.is_valid(i_simd);
732  }
733 
734  Simd<double> bmag;
735  magnetic_field.bmag_interpol(part.ph.v(), bmag);
736 
737  for (int i_simd = 0; i_simd<SIMD_SIZE; i_simd++){
738  if(!grid_wts0.is_valid(i_simd)) continue;
739 
740  nearest_node[i_simd]=grid_wts0.node[i_simd] - pol_decomp.node_offset;
741  not_in_poloidal_domain[i_simd] = (nearest_node[i_simd]<0 || nearest_node[i_simd]>=pol_decomp.nnodes);
742 
743  double temp_ev_norm = not_in_poloidal_domain[i_simd] ? f0.temp_ev(0) : f0.temp_ev(nearest_node[i_simd]);
744 
745  // get vp and smu
746  const double& B = bmag[i_simd];
747  vp[i_simd] = normalized_v_para(c_m, mass, B, temp_ev_norm, part.ph.rho[i_simd]);
748  smu[i_simd] = normalized_sqrt_mu(B, temp_ev_norm, part.ct.mu[i_simd]);
749  }
750  }
751 };
752 
753 #include "species.tpp"
754 #endif
Cabana::AoSoA< PhaseDataTypes, HostType, VEC_LEN > phase0
Definition: species.hpp:111
bool stream_particles
Whether to stream particles between host and device if possible.
Definition: species.hpp:103
Definition: globals.hpp:84
KOKKOS_INLINE_FUNCTION VecPhase * ph0() const
Definition: species.hpp:577
KOKKOS_INLINE_FUNCTION int divide_and_round_up(int a, int b)
Definition: globals.hpp:179
bool owns_particles_d
Whether the species owns the device particle allocation right now.
Definition: species.hpp:100
void back_up_SoA(Cabana::AoSoA< ParticleDataTypes, Device, VEC_LEN > &backup_SoA, int offset, int n) const
Definition: species.hpp:410
KOKKOS_INLINE_FUNCTION double normalized_v_para(double c_m, double mass, double B, double temp_ev, double rho)
Definition: basic_physics.hpp:80
KOKKOS_INLINE_FUNCTION double get_f0_eq_gyro_radius(int inode, double smu_n, double bfield) const
Definition: species.hpp:696
subroutine adjust_n_ptl_for_core_ptl(n_ptl)
Definition: load.F90:473
void set_spall_num_and_ptr(int idx, int n_ptl, int n_vecs, VecParticles *ptl)
void for_particle_range(int begin_idx, int end_idx, const std::string label, F lambda_func) const
Definition: species.hpp:472
constexpr double PROTON_MASS
Definition: constants.hpp:7
Distribution< Device > f0
Species distribution in velocity space on local mesh nodes.
Definition: species.hpp:119
MPI_Comm SML_COMM_WORLD
Definition: my_mpi.cpp:4
Cabana::AoSoA< ParticleDataTypes, HostType, VEC_LEN > backup_particles
Copy of particles to be restored for RK2.
Definition: species.hpp:115
bool is_electron
Whether this species is the electrons.
Definition: species.hpp:78
void for_all_particles(const std::string label, F lambda_func, const PtlMvmt mvmt, LaunchBounds launch_bounds=LaunchBounds::Default)
Definition: species.hpp:523
void save_backup_particles()
Definition: species.hpp:598
double c2_2m
c2/2m
Definition: species.hpp:86
double rho[VEC_LEN]
Definition: particles.hpp:96
void copy_to_phase0(Species< Device > &species)
Definition: species.hpp:581
Definition: species.hpp:57
Simd< double > w1
Definition: particles.hpp:22
double c_m
c/m
Definition: species.hpp:85
Definition: species.hpp:56
constexpr double EV_2_J
Conversion rate ev to J.
Definition: constants.hpp:5
bool default_streaming_option()
Definition: species.hpp:27
Eq::Profile< Device > eq_den
Definition: species.hpp:124
Definition: globals.hpp:89
KOKKOS_INLINE_FUNCTION VecParticles * ptl() const
Definition: species.hpp:573
Definition: grid_weights.hpp:51
KOKKOS_INLINE_FUNCTION double thermal_velocity(double mass, double temp_ev)
Definition: basic_physics.hpp:58
Definition: NamelistReader.hpp:193
KinType kintype
Whether the species is gyrokinetic or drift kinetic.
Definition: species.hpp:81
Definition: magnetic_field.hpp:12
int add_vec_buffer(int n_ptl)
Definition: particles.hpp:194
int idx
Index in all_species.
Definition: species.hpp:77
Definition: particles.hpp:92
int a
The index in the inner array of the AoSoA.
Definition: particles.hpp:150
Definition: particles.hpp:109
KOKKOS_INLINE_FUNCTION void bmag_interpol(const SimdVector &v, Simd< double > &bmag) const
Definition: magnetic_field.tpp:257
bool particles_are_backed_up
Whether particles are currently backed up.
Definition: species.hpp:108
int nonadiabatic_idx
Index of species skipping adiabatic species (for compatibility with fortran arrays) ...
Definition: species.hpp:80
bool default_residence_option()
Definition: species.hpp:34
int n_ptl
Number of particles.
Definition: species.hpp:94
KOKKOS_INLINE_FUNCTION void get_particle_velocity_and_nearest_node(const Grid< DeviceType > &grid, const MagneticField< DeviceType > &magnetic_field, const DomainDecomposition< DeviceType > &pol_decomp, SimdParticles &part, Simd< double > &smu, Simd< double > &vp, Simd< int > &nearest_node, Simd< bool > &not_in_triangle, Simd< bool > &not_in_poloidal_domain) const
Definition: species.hpp:720
Definition: streamed_parallel_for.hpp:16
int node_offset
Offset of first mesh node belonging to this MPI rank.
Definition: domain_decomposition.hpp:41
Definition: streamed_parallel_for.hpp:14
void set_buffer_phase0_d()
Definition: species.hpp:381
long long int get_total_n_ptl()
Definition: species.hpp:670
void set_buffer_particles_d()
Definition: species.hpp:352
Simd< double > rho
Definition: particles.hpp:21
Definition: species.hpp:58
int p_range< DeviceType >(int num_particle)
Definition: particles.hpp:187
int eq_flow_type
Definition: species.hpp:126
double charge_eu
Particle charge in eu.
Definition: species.hpp:84
Definition: species.hpp:50
int nnodes
Number of nodes belonging to this MPI rank.
Definition: domain_decomposition.hpp:42
void resize_particles(int new_n_ptl)
Definition: species.hpp:222
double mass
Particle mass.
Definition: species.hpp:82
KOKKOS_INLINE_FUNCTION double get_f0_eq_thermal_velocity(int inode) const
Definition: species.hpp:703
Species(int idx_in, int nonadiabatic_idx_in, bool is_electron_in, bool is_adiabatic_in, KinType kintype_in, double mass_in, double charge_in, double charge_eu_in, bool is_deltaf_in, int ncycles_in)
Definition: species.tpp:23
void for_all_particles(const std::string label, F lambda_func) const
Definition: species.hpp:405
Cabana::AoSoA< ParticleDataTypes, Device, VEC_LEN > particles_d
Particles on device.
Definition: species.hpp:98
Definition: species.hpp:47
double w2[VEC_LEN]
Definition: particles.hpp:98
#define TIMER(N, F)
Definition: timer_macro.hpp:24
RKRestorationMethod
Definition: species.hpp:67
idx
Definition: diag_f0_df_port1.hpp:32
void copy_particles_to_device_if_not_resident()
Definition: species.hpp:335
RKRestorationMethod RK_restoration_method
Currently, electrons must use first method and ions must use second.
Definition: species.hpp:106
Simd< double > r
Definition: particles.hpp:18
void resize_host_particles_to_match_device()
Definition: species.hpp:242
Definition: species.hpp:68
KOKKOS_INLINE_FUNCTION SimdVector & v()
Definition: particles.hpp:39
ReturnOpt return_opt
Definition: species.hpp:62
ReturnOpt
Definition: species.hpp:55
Option
Definition: streamed_parallel_for.hpp:13
void restore_particles_from_backup()
Definition: species.hpp:619
Definition: globals.hpp:90
static std::vector< MemoryPrediction > estimate_memory_usage(NLReader::NamelistReader &nlr, const Grid< DeviceType > &grid, const DomainDecomposition< DeviceType > &pol_decomp, int species_idx)
Definition: species.tpp:58
SendOpt send_opt
Definition: species.hpp:61
double charge
Particle charge.
Definition: species.hpp:83
SimdPhase ph
Definition: particles.hpp:59
void copy_particles_from_device()
Definition: species.hpp:310
KOKKOS_INLINE_FUNCTION double get_f0_eq_thermal_velocity_lnode(int inode) const
Definition: species.hpp:709
void copy_particles_from_device_if_not_resident()
Definition: species.hpp:341
KOKKOS_INLINE_FUNCTION void wedge_modulo_phi(Simd< double > &phi_mod) const
Definition: grid.tpp:767
void unassign_host_particles()
Definition: species.hpp:259
int ncycles_between_sorts
Number of subcycles between sorts.
Definition: species.hpp:91
Definition: particles.hpp:58
Cabana::AoSoA< PhaseDataTypes, Device, VEC_LEN > phase0_d
Definition: species.hpp:112
int SML_COMM_RANK
Definition: my_mpi.cpp:5
KinType
Definition: globals.hpp:88
Species(SpeciesType sp_type, int n_ptl)
Definition: species.hpp:138
bool is_deltaf
Whether this species is deltaf.
Definition: species.hpp:88
VecPhase ph
Definition: particles.hpp:110
Definition: species.hpp:43
Definition: species.hpp:69
void set_min_max_num(int isp, int n_ptl)
int minimum_ptl_reservation
The minimum reservation size for particles.
Definition: species.hpp:93
int s
The index in the outer array of the AoSoA.
Definition: particles.hpp:149
KOKKOS_INLINE_FUNCTION double normalized_sqrt_mu(double B, double temp_ev, double mu)
Definition: basic_physics.hpp:90
Simd< double > z
Definition: particles.hpp:19
void copy_particles_to_device_if_resident()
Definition: species.hpp:323
Definition: species.hpp:51
void resize_device_particles(int new_n_ptl)
Definition: species.hpp:280
Definition: species.hpp:52
constexpr double UNIT_CHARGE
Charge of an electron (C)
Definition: constants.hpp:4
KOKKOS_INLINE_FUNCTION double get_f0_eq_thermal_velocity_lnode_h(int inode) const
Definition: species.hpp:715
void exit_XGC(std::string msg)
Definition: globals.hpp:37
void copy_particles_from_device_if_resident()
Definition: species.hpp:329
bool is_adiabatic
Whether this species is adiabatic.
Definition: species.hpp:79
Simd< double > phi
Definition: particles.hpp:20
Definition: magnetic_field.F90:1
static constexpr const Kokkos::Experimental::WorkItemProperty::HintLightWeight_t Async
Definition: space_settings.hpp:82
int n_backup_particles
Definition: species.hpp:116
Eq::Profile< Device > eq_flow
Definition: species.hpp:125
Definition: streamed_parallel_for.hpp:15
SendOpt
Definition: species.hpp:49
SimdConstants ct
Definition: particles.hpp:60
void copy_particles_to_device()
Definition: species.hpp:294
KOKKOS_INLINE_FUNCTION void restore_phase_from_phase0(const AoSoAIndices< Device > &inds, SimdParticles &part_one) const
Definition: species.hpp:657
Species(int n_ptl_in)
Definition: species.hpp:168
double phi[VEC_LEN]
Definition: particles.hpp:95
Simd< double > w2
Definition: particles.hpp:23
double r[VEC_LEN]
Definition: particles.hpp:93
GyroAverageMatrices< HostType > gyro_avg_matrices
Definition: species.hpp:128
Definition: species.hpp:74
void parallel_for(const std::string name, int n_ptl, Function func, Option option, HostAoSoA aosoa_h, DeviceAoSoA aosoa_d)
Definition: streamed_parallel_for.hpp:252
KOKKOS_INLINE_FUNCTION void get_grid_weights(const MagneticField< Device > &magnetic_field, const SimdVector &v, const Simd< double > &psi, SimdVector2D &xff, SimdGridWeights< Order::One, PIT > &grid_wts) const
Definition: grid.tpp:682
Definition: species.hpp:43
Eq::Profile< Device > eq_temp
Definition: species.hpp:123
bool particles_resident_on_device
Whether the particles can reside on device.
Definition: species.hpp:102
Simd< double > mu
Definition: particles.hpp:52
PtlMvmt(SendOpt send_opt, ReturnOpt return_opt)
Definition: species.hpp:64
static int get_initial_n_ptl(NLReader::NamelistReader &nlr, const Grid< DeviceType > &grid, const DomainDecomposition< DeviceType > &pol_decomp, int sml_special, int species_idx, bool verbose)
Definition: species.tpp:115
int ncycles
Number of subcycles.
Definition: species.hpp:90
Definition: profile.hpp:171
int collision_grid_index
Which collision grid to use.
Definition: species.hpp:121
Definition: particles.hpp:148
SpeciesType
Definition: globals.hpp:83
double z[VEC_LEN]
Definition: particles.hpp:94
void restore_backup_SoA(Cabana::AoSoA< ParticleDataTypes, Device, VEC_LEN > &backup_SoA, int offset, int n) const
Definition: species.hpp:439
Definition: distribution.hpp:10
void resize_device_particles()
Definition: species.hpp:265
int get_max_n_ptl()
Definition: species.hpp:681
LaunchBounds
Definition: species.hpp:394
Cabana::AoSoA< ParticleDataTypes, HostType, VEC_LEN > particles
Particles.
Definition: species.hpp:95
double w1[VEC_LEN]
Definition: particles.hpp:97
Definition: species.hpp:43