|
| 1 | +/* Copyright 2024 The WarpX Community |
| 2 | + * |
| 3 | + * This file is part of WarpX. |
| 4 | + * |
| 5 | + * License: BSD-3-Clause-LBNL |
| 6 | + */ |
| 7 | + |
| 8 | +#include "ParticleSplitting.H" |
| 9 | +#include "Particles/PhysicalParticleContainer.H" |
| 10 | +#include "Utils/Parser/ParserUtils.H" |
| 11 | +#include "WarpX.H" |
| 12 | +#include <AMReX_REAL.H> |
| 13 | +#include <AMReX_Gpu.H> |
| 14 | +#include <AMReX_Random.H> |
| 15 | +#include <AMReX_GpuQualifiers.H> |
| 16 | +#include <AMReX_ParmParse.H> |
| 17 | + |
| 18 | +using namespace amrex; |
| 19 | +using warpx::fields::FieldType; |
| 20 | + |
| 21 | +ParticleSplitting::ParticleSplitting (const std::string& species_name) |
| 22 | +{ |
| 23 | + using namespace amrex::literals; |
| 24 | + |
| 25 | + const amrex::ParmParse pp_species_name(species_name); |
| 26 | + |
| 27 | + utils::parser::queryWithParser( |
| 28 | + pp_species_name, "resampling_min_ppc", m_min_ppc |
| 29 | + ); |
| 30 | + WARPX_ALWAYS_ASSERT_WITH_MESSAGE( |
| 31 | + m_min_ppc >= 1, |
| 32 | + "Resampling min_ppc should be greater than or equal to 1" |
| 33 | + ); |
| 34 | + |
| 35 | + utils::parser::queryWithParser( |
| 36 | + pp_species_name, "resampling_random_splitting_angle", m_resampling_random_splitting_angle); |
| 37 | + |
| 38 | + pp_species_name.query("resampling_splitting_type", m_splitting_type); |
| 39 | + |
| 40 | + WARPX_ALWAYS_ASSERT_WITH_MESSAGE( |
| 41 | + m_splitting_type == "position_axes_aligned_split" || m_splitting_type == "position_velocity_aligned_split", |
| 42 | + "Invalid resampling_splitting_type specified.\n" |
| 43 | + "Valid options are:\n" |
| 44 | + " - position_axes_aligned_split\n" |
| 45 | + " - position_velocity_aligned_split.\n"); |
| 46 | + |
| 47 | + if (!m_resampling_random_splitting_angle) { |
| 48 | + utils::parser::queryWithParser( |
| 49 | + pp_species_name, "resampling_splitting_angle", m_splitting_angle); |
| 50 | + } |
| 51 | +} |
| 52 | +void ParticleSplitting::operator() ( |
| 53 | + const amrex::Geometry& geom_lev, WarpXParIter& pti, |
| 54 | + const int lev, WarpXParticleContainer * const pc) const |
| 55 | +{ |
| 56 | + using namespace amrex::literals; |
| 57 | + |
| 58 | + auto& ptile = pc->ParticlesAt(lev, pti); |
| 59 | + const auto num_particles_tile = ptile.numParticles(); |
| 60 | + |
| 61 | + if (num_particles_tile == 0) return; |
| 62 | + // Bin particles by cell |
| 63 | + auto bins = ParticleUtils::findParticlesInEachCell(geom_lev, pti, ptile); |
| 64 | + const auto n_cells = static_cast<int>(bins.numBins()); |
| 65 | + auto *const indices = bins.permutationPtr(); |
| 66 | + auto *const cell_offsets = bins.offsetsPtr(); |
| 67 | + |
| 68 | + const std::array<amrex::Real,3>& dx = WarpX::CellSize(lev); |
| 69 | + |
| 70 | + int np_split_per_parent = 2; |
| 71 | + |
| 72 | + if (m_splitting_type == "position_axes_aligned_split") { |
| 73 | +#if defined(WARPX_DIM_1D_Z) |
| 74 | + np_split_per_parent = 2; |
| 75 | +#endif |
| 76 | +#if defined(WARPX_DIM_3D) |
| 77 | + np_split_per_parent = 6; |
| 78 | +#endif |
| 79 | +#if defined(WARPX_DIM_XZ) |
| 80 | + np_split_per_parent = 4; |
| 81 | +#endif |
| 82 | + } |
| 83 | + else if (m_splitting_type == "position_velocity_aligned_split") { |
| 84 | + np_split_per_parent = 2; |
| 85 | + } |
| 86 | + |
| 87 | + const auto min_ppc = m_min_ppc; |
| 88 | + |
| 89 | + amrex::Gpu::DeviceVector<int> n_new_children_per_cell(n_cells); |
| 90 | + int* num_new_children_ptr = n_new_children_per_cell.data(); |
| 91 | + |
| 92 | + amrex::ParallelFor(n_cells, |
| 93 | + [=] AMREX_GPU_DEVICE (int i_cell) noexcept |
| 94 | + { |
| 95 | + const auto cell_start = static_cast<int>(cell_offsets[i_cell]); |
| 96 | + const auto cell_stop = static_cast<int>(cell_offsets[i_cell+1]); |
| 97 | + const auto cell_numparts = cell_stop - cell_start; |
| 98 | + |
| 99 | + // Skip cells with enough particles or empty cells |
| 100 | + if (cell_numparts == 0 || cell_numparts >= min_ppc) { |
| 101 | + num_new_children_ptr[i_cell] = 0; |
| 102 | + return; |
| 103 | + } |
| 104 | + |
| 105 | + // Calculate how many particles need to be split per cell |
| 106 | + const int deficit = min_ppc - cell_numparts; |
| 107 | + const int particles_to_split = amrex::min( |
| 108 | + cell_numparts, |
| 109 | + static_cast<int>(amrex::Math::ceil(static_cast<amrex::Real>(deficit) / (np_split_per_parent - 1.0_prt))) |
| 110 | + ); |
| 111 | + |
| 112 | + // Each parent that splits creates np_split_per_parent children per cell, or particles_to_split * np_split_per_parent per tile |
| 113 | + num_new_children_ptr[i_cell] = particles_to_split * np_split_per_parent; |
| 114 | + } |
| 115 | + ); |
| 116 | + |
| 117 | + amrex::Gpu::DeviceVector<int> offsets(n_cells); |
| 118 | + int* offset_ptr = offsets.data(); |
| 119 | + |
| 120 | + int num_new_children_tile = amrex::Scan::ExclusiveSum(n_cells, num_new_children_ptr, offset_ptr); |
| 121 | + |
| 122 | + if (num_new_children_tile == 0) return; |
| 123 | + |
| 124 | + ptile.resize(num_particles_tile + num_new_children_tile); |
| 125 | + |
| 126 | + auto& soa = ptile.GetStructOfArrays(); |
| 127 | + |
| 128 | +#if !defined(WARPX_DIM_1D_Z) |
| 129 | + auto * const AMREX_RESTRICT x = soa.GetRealData(PIdx::x).data(); |
| 130 | +#endif |
| 131 | +#if defined(WARPX_DIM_3D) |
| 132 | + auto * const AMREX_RESTRICT y = soa.GetRealData(PIdx::y).data(); |
| 133 | +#endif |
| 134 | +#if defined(WARPX_ZINDEX) |
| 135 | + auto * const AMREX_RESTRICT z = soa.GetRealData(PIdx::z).data(); |
| 136 | +#endif |
| 137 | + |
| 138 | + auto * const AMREX_RESTRICT ux = soa.GetRealData(PIdx::ux).data(); |
| 139 | + auto * const AMREX_RESTRICT uy = soa.GetRealData(PIdx::uy).data(); |
| 140 | + auto * const AMREX_RESTRICT uz = soa.GetRealData(PIdx::uz).data(); |
| 141 | + auto * const AMREX_RESTRICT w = soa.GetRealData(PIdx::w).data(); |
| 142 | + auto * const AMREX_RESTRICT idcpu = soa.GetIdCPUData().data(); |
| 143 | + |
| 144 | + amrex::ParallelForRNG(n_cells, |
| 145 | + [=] AMREX_GPU_DEVICE (int i_cell, amrex::RandomEngine const& engine) noexcept |
| 146 | + { |
| 147 | + const auto cell_start = static_cast<int>(cell_offsets[i_cell]); |
| 148 | + const auto cell_stop = static_cast<int>(cell_offsets[i_cell+1]); |
| 149 | + const auto cell_numparts = cell_stop - cell_start; |
| 150 | + |
| 151 | + // Skip cells with enough particles or empty cells |
| 152 | + if (cell_numparts == 0 || cell_numparts >= min_ppc) return; |
| 153 | + |
| 154 | + // Calculate how many particles need to be split |
| 155 | + const int deficit = min_ppc - cell_numparts; |
| 156 | + const int particles_to_split = amrex::min( |
| 157 | + cell_numparts, |
| 158 | + static_cast<int>(amrex::Math::ceil(static_cast<amrex::Real>(deficit) / (np_split_per_parent - 1.0_prt))) |
| 159 | + ); |
| 160 | + |
| 161 | + // Calculate cell-dependent position offset |
| 162 | + const amrex::Real inv_split = 1.0_prt / (2.0_prt * np_split_per_parent * particles_to_split); |
| 163 | +#if !defined(WARPX_DIM_1D_Z) |
| 164 | + amrex::ParticleReal offset_x = dx[0] * inv_split; |
| 165 | +#endif |
| 166 | +#if defined(WARPX_DIM_3D) |
| 167 | + amrex::ParticleReal offset_y = dx[1] * inv_split; |
| 168 | +#endif |
| 169 | +#if defined(WARPX_ZINDEX) |
| 170 | + amrex::ParticleReal offset_z = dx[2] * inv_split; |
| 171 | +#endif |
| 172 | + |
| 173 | + const auto resampling_random_splitting_angle = m_resampling_random_splitting_angle; |
| 174 | + const amrex::Real splitting_angle = resampling_random_splitting_angle ? |
| 175 | + amrex::Random(engine) * 2.0_rt * MathConst::pi : m_splitting_angle; |
| 176 | + |
| 177 | + // Starting index for new children particles for i_cell |
| 178 | + const int new_particle_start = num_particles_tile + offset_ptr[i_cell]; |
| 179 | + |
| 180 | + int split_count = 0; |
| 181 | + for (int i = cell_start; i < cell_stop && split_count < particles_to_split; ++i) { |
| 182 | + const int parent_idx = indices[i]; |
| 183 | + |
| 184 | +#if !defined(WARPX_DIM_1D_Z) |
| 185 | + amrex::ParticleReal xp = x[parent_idx]; |
| 186 | +#endif |
| 187 | +#if defined(WARPX_DIM_3D) |
| 188 | + amrex::ParticleReal yp = y[parent_idx]; |
| 189 | +#endif |
| 190 | +#if defined(WARPX_ZINDEX) |
| 191 | + amrex::ParticleReal zp = z[parent_idx]; |
| 192 | +#endif |
| 193 | + // Get parent particle properties |
| 194 | + const amrex::Real parent_weight = w[parent_idx]; |
| 195 | + const amrex::Real child_weight = parent_weight / static_cast<amrex::Real>(np_split_per_parent); |
| 196 | + const int child_base = new_particle_start + split_count * np_split_per_parent; |
| 197 | + |
| 198 | + if (m_splitting_type == "position_axes_aligned_split") { |
| 199 | +#if defined(WARPX_DIM_1D_Z) |
| 200 | + amrex::Print() << "Splitting particle along z axis (WARPX_DIM_1D_Z)" << std::endl; |
| 201 | + // Split particle in 2 along z axis |
| 202 | + for (int k = 0; k < 2; ++k) { |
| 203 | + const int sign_offset = (k == 0) ? -1 : 1; |
| 204 | + const int idx = child_base + k; |
| 205 | + z[idx] = zp + sign_offset * offset_z; |
| 206 | + ux[idx] = ux[parent_idx]; |
| 207 | + uy[idx] = uy[parent_idx]; |
| 208 | + uz[idx] = uz[parent_idx]; |
| 209 | + w[idx] = child_weight; |
| 210 | + idcpu[idx] = amrex::SetParticleIDandCPU( |
| 211 | + amrex::LongParticleIds::NoSplitParticleID, |
| 212 | + amrex::ParallelDescriptor::MyProc() |
| 213 | + ); |
| 214 | + } |
| 215 | +#elif defined(WARPX_DIM_XZ) |
| 216 | + // Split particle in 4 particles: split in the x–z plane with a rotation by splitting_angle around the z-axis. |
| 217 | + for (int k = 0; k < 4; ++k) { |
| 218 | + const int idx = child_base + k; |
| 219 | + const int sign_offset = (k % 2 == 0) ? -1 : 1; |
| 220 | + if (k < 2) { |
| 221 | + // Split 2 of 4 particles |
| 222 | + x[idx] = xp + std::cos(splitting_angle) * sign_offset * offset_x; |
| 223 | + z[idx] = zp - std::sin(splitting_angle) * sign_offset * offset_z; |
| 224 | + } else { |
| 225 | + // Split other 2 of 4 particles |
| 226 | + x[idx] = xp - std::sin(splitting_angle) * sign_offset * offset_x; |
| 227 | + z[idx] = zp + std::cos(splitting_angle) * sign_offset * offset_z; |
| 228 | + } |
| 229 | + ux[idx] = ux[parent_idx]; |
| 230 | + uy[idx] = uy[parent_idx]; |
| 231 | + uz[idx] = uz[parent_idx]; |
| 232 | + w[idx] = child_weight; |
| 233 | + idcpu[idx] = amrex::SetParticleIDandCPU( |
| 234 | + amrex::LongParticleIds::NoSplitParticleID, |
| 235 | + amrex::ParallelDescriptor::MyProc() |
| 236 | + ); |
| 237 | + } |
| 238 | +#elif defined(WARPX_DIM_3D) || defined(WARPX_DIM_RZ) |
| 239 | + // Split parent particle in 6 particles |
| 240 | + for (int k = 0; k < 6; ++k) { |
| 241 | + const int idx = child_base + k; |
| 242 | + const int sign_offset = (k % 2 == 0) ? -1 : 1; |
| 243 | + |
| 244 | + if (k < 4) { |
| 245 | + // Split 4 of 6 particles in the x–y plane with a rotation by splitting_angle around the z-axis. |
| 246 | + if (k < 2) { |
| 247 | + x[idx] = xp + std::cos(splitting_angle) * sign_offset * offset_x; |
| 248 | + y[idx] = yp + std::sin(splitting_angle) * sign_offset * offset_y; |
| 249 | + } else { |
| 250 | + x[idx] = xp - std::sin(splitting_angle) * sign_offset * offset_x; |
| 251 | + y[idx] = yp + std::cos(splitting_angle) * sign_offset * offset_y; |
| 252 | + } |
| 253 | + z[idx] = zp; |
| 254 | + } else { |
| 255 | + // Split 2 of 6 particles along z-axis |
| 256 | + x[idx] = xp; |
| 257 | + y[idx] = yp; |
| 258 | + z[idx] = zp + sign_offset * offset_z; |
| 259 | + } |
| 260 | + ux[idx] = ux[parent_idx]; |
| 261 | + uy[idx] = uy[parent_idx]; |
| 262 | + uz[idx] = uz[parent_idx]; |
| 263 | + w[idx] = child_weight; |
| 264 | + idcpu[idx] = amrex::SetParticleIDandCPU( |
| 265 | + amrex::LongParticleIds::NoSplitParticleID, |
| 266 | + amrex::ParallelDescriptor::MyProc() |
| 267 | + ); |
| 268 | + } |
| 269 | +#elif defined(WARPX_DIM_RCYLINDER) || defined(WARPX_DIM_RSPHERE) |
| 270 | + // Split particle in 2 along x axis |
| 271 | + for (int k = 0; k < 2; ++k) { |
| 272 | + const int sign_offset = (k == 0) ? -1 : 1; |
| 273 | + const int idx = child_base + k; |
| 274 | + x[idx] = xp + sign_offset * offset_x; |
| 275 | + ux[idx] = ux[parent_idx]; |
| 276 | + uy[idx] = uy[parent_idx]; |
| 277 | + uz[idx] = uz[parent_idx]; |
| 278 | + w[idx] = child_weight; |
| 279 | + idcpu[idx] = amrex::SetParticleIDandCPU( |
| 280 | + amrex::LongParticleIds::NoSplitParticleID, |
| 281 | + amrex::ParallelDescriptor::MyProc() |
| 282 | + ); |
| 283 | + } |
| 284 | +#endif |
| 285 | + } |
| 286 | + else if (m_splitting_type == "position_velocity_aligned_split") { |
| 287 | + // Split particle in 2 along the velocity direction |
| 288 | + const amrex::Real u2 = ux[parent_idx] * ux[parent_idx] + |
| 289 | + uy[parent_idx] * uy[parent_idx] + |
| 290 | + uz[parent_idx] * uz[parent_idx]; |
| 291 | + const amrex::Real u_norm = (std::sqrt(u2) > 0.0_rt) ? std::sqrt(u2) : 1.0_rt; |
| 292 | + |
| 293 | + for (int k = 0; k < 2; ++k) { |
| 294 | + const int sign_offset = (k == 0) ? -1 : 1; |
| 295 | + const int idx = child_base + k; |
| 296 | +#if !defined(WARPX_DIM_1D_Z) |
| 297 | + x[idx] = xp + sign_offset * offset_x * ux[parent_idx] / u_norm; |
| 298 | +#endif |
| 299 | +#if defined(WARPX_DIM_3D) |
| 300 | + y[idx] = yp + sign_offset * offset_y * uy[parent_idx] / u_norm; |
| 301 | +#endif |
| 302 | +#if defined(WARPX_ZINDEX) |
| 303 | + z[idx] = zp + sign_offset * offset_z * uz[parent_idx] / u_norm; |
| 304 | +#endif |
| 305 | + ux[idx] = ux[parent_idx]; |
| 306 | + uy[idx] = uy[parent_idx]; |
| 307 | + uz[idx] = uz[parent_idx]; |
| 308 | + w[idx] = child_weight; |
| 309 | + idcpu[idx] = amrex::SetParticleIDandCPU( |
| 310 | + amrex::LongParticleIds::NoSplitParticleID, |
| 311 | + amrex::ParallelDescriptor::MyProc() |
| 312 | + ); |
| 313 | + } |
| 314 | + } |
| 315 | + |
| 316 | + // Mark parent particles as invalid |
| 317 | + idcpu[parent_idx] = amrex::ParticleIdCpus::Invalid; |
| 318 | + ++split_count; |
| 319 | + } |
| 320 | + } |
| 321 | + ); |
| 322 | +} |
0 commit comments