ERF
Energy Research and Forecasting: An Atmospheric Modeling Code
Functions
ERF_Substep_NS.cpp File Reference
#include <ERF_TI_fast_headers.H>
Include dependency graph for ERF_Substep_NS.cpp:

Functions

void erf_substep_NS (int step, int nrk, int level, int finest_level, Vector< MultiFab > &S_slow_rhs, const Vector< MultiFab > &S_prev, Vector< MultiFab > &S_stage_data, const MultiFab &S_stage_prim, const MultiFab &qt, const MultiFab &pi_stage, const MultiFab &fast_coeffs, Vector< MultiFab > &S_data, MultiFab &lagged_delta_rt, MultiFab &avg_xmom, MultiFab &avg_ymom, MultiFab &avg_zmom, const MultiFab &cc_src, const MultiFab &xmom_src, const MultiFab &ymom_src, const MultiFab &zmom_src, const Geometry geom, const Real gravity, amrex::Gpu::DeviceVector< amrex::Real > &stretched_dz_d, const double dtau_d, const Real beta_s, const Real facinv, Vector< std::unique_ptr< MultiFab >> &mapfac, YAFluxRegister *fr_as_crse, YAFluxRegister *fr_as_fine, bool l_use_moisture, bool l_reflux, bool l_real_bc, const amrex::Real *sinesq_stag_d, const Real l_damp_coef)
 

Function Documentation

◆ erf_substep_NS()

void erf_substep_NS ( int  step,
int  nrk,
int  level,
int  finest_level,
Vector< MultiFab > &  S_slow_rhs,
const Vector< MultiFab > &  S_prev,
Vector< MultiFab > &  S_stage_data,
const MultiFab &  S_stage_prim,
const MultiFab &  qt,
const MultiFab &  pi_stage,
const MultiFab &  fast_coeffs,
Vector< MultiFab > &  S_data,
MultiFab &  lagged_delta_rt,
MultiFab &  avg_xmom,
MultiFab &  avg_ymom,
MultiFab &  avg_zmom,
const MultiFab &  cc_src,
const MultiFab &  xmom_src,
const MultiFab &  ymom_src,
const MultiFab &  zmom_src,
const Geometry  geom,
const Real  gravity,
amrex::Gpu::DeviceVector< amrex::Real > &  stretched_dz_d,
const double  dtau_d,
const Real  beta_s,
const Real  facinv,
Vector< std::unique_ptr< MultiFab >> &  mapfac,
YAFluxRegister *  fr_as_crse,
YAFluxRegister *  fr_as_fine,
bool  l_use_moisture,
bool  l_reflux,
bool  l_real_bc,
const amrex::Real *  sinesq_stag_d,
const Real  l_damp_coef 
)

Function for computing the fast RHS with no terrain and variable vertical spacing

Parameters
[in]stepwhich fast time step within each Runge-Kutta step
[in]nrkwhich Runge-Kutta step
[in]levellevel of resolution
[in]finest_levelfinest level of resolution
[in]S_slow_rhsslow RHS computed in erf_slow_rhs_pre
[in]S_previf step == 0, this is S_old, else the previous fast solution
[in]S_stage_datasolution at previous RK stage
[in]S_stage_primprimitive variables at previous RK stage
[in]pi_stageExner function at previous RK stage
[in]fast_coeffscoefficients for the tridiagonal solve used in the fast integrator
[out]S_datacurrent solution
[in,out]lagged_delta_rt
[in,out]avg_xmomtime-averaged x-momentum to be used for updating slow variables
[in,out]avg_ymomtime-averaged y-momentum to be used for updating slow variables
[in,out]avg_zmomtime-averaged z-momentum to be used for updating slow variables
[in]cc_srcsource terms for conserved variables
[in]xmom_srcsource terms for x-momentum
[in]ymom_srcsource terms for y-momentum
[in]zmom_srcsource terms for z-momentum
[in]geomcontainer for geometric information
[in]gravitymagnitude of gravity
[in]stretched_dz_d
[in]dtaufast time step
[in]beta_sCoefficient which determines how implicit vs explicit the solve is
[in]facinvinverse factor for time-averaging the momenta
[in]mapfacvector of map factors
[in,out]fr_as_crseYAFluxRegister at level l at level l / l+1 interface
[in,out]fr_as_fineYAFluxRegister at level l at level l-1 / l interface
[in]l_use_moisture
[in]l_refluxshould we add fluxes to the FluxRegisters?
73 {
74  //
75  // NOTE: for step > 0, S_data and S_prev point to the same MultiFab data!!
76  //
77 
78  BL_PROFILE_REGION("erf_substep_NS()");
79 
80  Real dtau = static_cast<Real>(dtau_d);
81 
82  const Box& domain = geom.Domain();
83  auto const domlo = lbound(domain);
84  auto const domhi = ubound(domain);
85 
86  int ilo = domlo.x;
87  int ihi = domhi.x + 1;
88  int jlo = domlo.y;
89  int jhi = domhi.y + 1;
90 
91  Real beta_1 = myhalf * (one - beta_s); // multiplies explicit terms
92  Real beta_2 = myhalf * (one + beta_s); // multiplies implicit terms
93 
94  // How much do we project forward the (rho theta) that is used in the horizontal momentum equations
95  Real beta_d = Real(0.1);
96 
97  Real RvOverRd = R_v / R_d;
98 
99  bool l_rayleigh_impl_for_w = (sinesq_stag_d != nullptr);
100 
101  const Real* dx = geom.CellSize();
102  const GpuArray<Real, AMREX_SPACEDIM> dxInv = geom.InvCellSizeArray();
103 
104  Real dxi = dxInv[0];
105  Real dyi = dxInv[1];
106 
107  auto dz_ptr = stretched_dz_d.data();
108 
109  const auto& ba = S_stage_data[IntVars::cons].boxArray();
110  const auto& dm = S_stage_data[IntVars::cons].DistributionMap();
111 
112  MultiFab Delta_rho_theta( ba , dm, 1, 1);
113  MultiFab Delta_rho_w (convert(ba,IntVect(0,0,1)), dm, 1, IntVect(1,1,0));
114 
115  MultiFab coeff_A_mf(fast_coeffs, make_alias, 0, 1);
116  MultiFab inv_coeff_B_mf(fast_coeffs, make_alias, 1, 1);
117  MultiFab coeff_C_mf(fast_coeffs, make_alias, 2, 1);
118  MultiFab coeff_P_mf(fast_coeffs, make_alias, 3, 1);
119  MultiFab coeff_Q_mf(fast_coeffs, make_alias, 4, 1);
120 
121  // *************************************************************************
122  // Set gravity as a vector
123  const Array<Real,AMREX_SPACEDIM> grav{zero, zero, -gravity};
124  const GpuArray<Real,AMREX_SPACEDIM> grav_gpu{grav[0], grav[1], grav[2]};
125 
126  // This will hold theta extrapolated forward in time
127  MultiFab extrap(S_data[IntVars::cons].boxArray(),S_data[IntVars::cons].DistributionMap(),1,1);
128 
129  // This will hold the update for (rho) and (rho theta)
130  MultiFab temp_rhs(S_stage_data[IntVars::zmom].boxArray(),S_stage_data[IntVars::zmom].DistributionMap(),2,0);
131 
132  // This will hold the new x- and y-momenta temporarily (so that we don't overwrite values we need when tiling)
133  MultiFab temp_cur_xmom(S_stage_data[IntVars::xmom].boxArray(),S_stage_data[IntVars::xmom].DistributionMap(),1,0);
134  MultiFab temp_cur_ymom(S_stage_data[IntVars::ymom].boxArray(),S_stage_data[IntVars::ymom].DistributionMap(),1,0);
135 
136  // We assume that in the first step (nrk == 0) we are only doing one substep.
137  AMREX_ALWAYS_ASSERT(nrk > 0 || step == 0);
138 
139  // *************************************************************************
140  // First set up some arrays we'll need
141  // *************************************************************************
142 
143 #ifdef _OPENMP
144 #pragma omp parallel if (Gpu::notInLaunchRegion())
145 #endif
146  for ( MFIter mfi(S_stage_data[IntVars::cons],TilingIfNotGPU()); mfi.isValid(); ++mfi)
147  {
148  const Array4<const Real>& prev_cons = S_prev[IntVars::cons].const_array(mfi);
149  const Array4<const Real>& prev_zmom = S_prev[IntVars::zmom].const_array(mfi);
150 
151  const Array4<const Real>& stage_cons = S_stage_data[IntVars::cons].const_array(mfi);
152  const Array4<const Real>& stage_zmom = S_stage_data[IntVars::zmom].const_array(mfi);
153 
154  const Array4<Real>& prev_drho_w = Delta_rho_w.array(mfi);
155  const Array4<Real>& prev_drho_theta = Delta_rho_theta.array(mfi);
156  const Array4<Real>& lagged_arr = lagged_delta_rt.array(mfi);
157  const Array4<Real>& theta_extrap = extrap.array(mfi);
158  const Array4<const Real>& prim = S_stage_prim.const_array(mfi);
159 
160  Box gbx = mfi.growntilebox(1);
161  ParallelFor(gbx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
162  {
163  prev_drho_theta(i,j,k) = prev_cons(i,j,k,RhoTheta_comp) - stage_cons(i,j,k,RhoTheta_comp);
164 
165  if (step == 0) {
166  theta_extrap(i,j,k) = prev_drho_theta(i,j,k);
167  } else {
168  theta_extrap(i,j,k) = prev_drho_theta(i,j,k) + beta_d *
169  ( prev_drho_theta(i,j,k) - lagged_arr(i,j,k) );
170  }
171 
172  // NOTE: qv is not changing over the fast steps so we use the stage data
173  Real qv = (l_use_moisture) ? prim(i,j,k,PrimQ1_comp) : zero;
174  theta_extrap(i,j,k) *= (one + RvOverRd*qv);
175 
176  // We define lagged_delta_rt for our next step as the current delta_rt
177  // (after using it above to extrapolate theta for this step)
178  lagged_arr(i,j,k) = prev_drho_theta(i,j,k);
179  });
180 
181  // NOTE: We must do this here because for step > 0, prev_zmom and cur_zmom both point to the same data,
182  // so by the time we would use prev_zmom to define zflux, it would have already been over-written.
183  Box gtbz = mfi.nodaltilebox(2);
184  gtbz.grow(IntVect(1,1,0));
185  ParallelFor(gtbz, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
186  prev_drho_w(i,j,k) = prev_zmom(i,j,k) - stage_zmom(i,j,k);
187  });
188  } // mfi
189 
190  // *************************************************************************
191  // Define updates in the current RK stage
192  // *************************************************************************
193 
194 #ifdef _OPENMP
195 #pragma omp parallel if (Gpu::notInLaunchRegion())
196 #endif
197  for ( MFIter mfi(S_stage_data[IntVars::cons],TilingIfNotGPU()); mfi.isValid(); ++mfi)
198  {
199  Box tbx = mfi.nodaltilebox(0);
200  Box tby = mfi.nodaltilebox(1);
201 
202  const Array4<Real const>& xmom_src_arr = xmom_src.const_array(mfi);
203  const Array4<Real const>& ymom_src_arr = ymom_src.const_array(mfi);
204 
205  const Array4<const Real> & stage_xmom = S_stage_data[IntVars::xmom].const_array(mfi);
206  const Array4<const Real> & stage_ymom = S_stage_data[IntVars::ymom].const_array(mfi);
207  const Array4<const Real> & qt_arr = qt.const_array(mfi);
208 
209  const Array4<const Real>& slow_rhs_rho_u = S_slow_rhs[IntVars::xmom].const_array(mfi);
210  const Array4<const Real>& slow_rhs_rho_v = S_slow_rhs[IntVars::ymom].const_array(mfi);
211 
212  const Array4<Real>& temp_cur_xmom_arr = temp_cur_xmom.array(mfi);
213  const Array4<Real>& temp_cur_ymom_arr = temp_cur_ymom.array(mfi);
214 
215  const Array4<const Real>& prev_xmom = S_prev[IntVars::xmom].const_array(mfi);
216  const Array4<const Real>& prev_ymom = S_prev[IntVars::ymom].const_array(mfi);
217 
218  // These store the advection momenta which we will use to update the slow variables
219  const Array4< Real>& avg_xmom_arr = avg_xmom.array(mfi);
220  const Array4< Real>& avg_ymom_arr = avg_ymom.array(mfi);
221 
222  const Array4<const Real>& pi_stage_ca = pi_stage.const_array(mfi);
223 
224  const Array4<Real>& theta_extrap = extrap.array(mfi);
225 
226  // Map factors
227  const Array4<const Real>& mf_ux = mapfac[MapFacType::u_x]->const_array(mfi);
228  const Array4<const Real>& mf_vy = mapfac[MapFacType::v_y]->const_array(mfi);
229 
230  // *********************************************************************
231  // Define updates in the RHS of {x, y, z}-momentum equations
232  // *********************************************************************
233  if (nrk == 0 and step == 0) { // prev == stage
234  ParallelFor(tbx, tby,
235  [=] AMREX_GPU_DEVICE (int i, int j, int k)
236  {
237  Real new_drho_u = dtau * slow_rhs_rho_u(i,j,k) + dtau * xmom_src_arr(i,j,k);;
238  avg_xmom_arr(i,j,k) += facinv*new_drho_u;
239  temp_cur_xmom_arr(i,j,k) = stage_xmom(i,j,k) + new_drho_u;
240  },
241  [=] AMREX_GPU_DEVICE (int i, int j, int k)
242  {
243  Real new_drho_v = dtau * slow_rhs_rho_v(i,j,k) + dtau * ymom_src_arr(i,j,k);
244  avg_ymom_arr(i,j,k) += facinv*new_drho_v;
245  temp_cur_ymom_arr(i,j,k) = stage_ymom(i,j,k) + new_drho_v;
246  });
247  } else {
248  ParallelFor(tbx, tby,
249  [=] AMREX_GPU_DEVICE (int i, int j, int k)
250  {
251  // Add (negative) gradient of (rho theta) multiplied by lagged "pi"
252  Real gpx = (l_real_bc && (level==0) && (i==ilo || i==ihi)) ? Real(0.) :
253  (theta_extrap(i,j,k) - theta_extrap(i-1,j,k))*dxi;
254  gpx *= mf_ux(i,j,0);
255 
256  Real q = (l_use_moisture) ? myhalf * (qt_arr(i,j,k) + qt_arr(i-1,j,k)) : zero;
257 
258  Real pi_c = myhalf * (pi_stage_ca(i-1,j,k,0) + pi_stage_ca(i,j,k,0));
259  Real fast_rhs_rho_u = -Gamma * R_d * pi_c * gpx / (one + q);
260 
261  Real new_drho_u = prev_xmom(i,j,k) - stage_xmom(i,j,k)
262  + dtau * fast_rhs_rho_u + dtau * slow_rhs_rho_u(i,j,k)
263  + dtau * xmom_src_arr(i,j,k);
264 
265  avg_xmom_arr(i,j,k) += facinv*new_drho_u;
266 
267  temp_cur_xmom_arr(i,j,k) = stage_xmom(i,j,k) + new_drho_u;
268  },
269  [=] AMREX_GPU_DEVICE (int i, int j, int k)
270  {
271  // Add (negative) gradient of (rho theta) multiplied by lagged "pi"
272  Real gpy = (l_real_bc && (level==0) && (j==jlo || j==jhi)) ? Real(0.) :
273  (theta_extrap(i,j,k) - theta_extrap(i,j-1,k))*dyi;
274  gpy *= mf_vy(i,j,0);
275 
276  Real q = (l_use_moisture) ? myhalf * (qt_arr(i,j,k) + qt_arr(i,j-1,k)) : zero;
277 
278  Real pi_c = myhalf * (pi_stage_ca(i,j-1,k,0) + pi_stage_ca(i,j,k,0));
279  Real fast_rhs_rho_v = -Gamma * R_d * pi_c * gpy / (one + q);
280 
281  Real new_drho_v = prev_ymom(i,j,k) - stage_ymom(i,j,k)
282  + dtau * fast_rhs_rho_v + dtau * slow_rhs_rho_v(i,j,k)
283  + dtau * ymom_src_arr(i,j,k);
284 
285  avg_ymom_arr(i,j,k) += facinv*new_drho_v;
286 
287  temp_cur_ymom_arr(i,j,k) = stage_ymom(i,j,k) + new_drho_v;
288  });
289  } // nrk > 0 and/or step > 0
290  } //mfi
291 
292 #ifdef _OPENMP
293 #pragma omp parallel if (Gpu::notInLaunchRegion())
294 #endif
295  {
296  std::array<FArrayBox,AMREX_SPACEDIM> flux;
297  for ( MFIter mfi(S_stage_data[IntVars::cons],TileNoZ()); mfi.isValid(); ++mfi)
298  {
299  Box bx = mfi.tilebox();
300  Box tbz = surroundingNodes(bx,2);
301 
302  Box vbx = mfi.validbox();
303  const auto& vbx_hi = ubound(vbx);
304 
305  const Array4<Real const>& zmom_src_arr = zmom_src.const_array(mfi);
306 
307  const Array4<const Real>& stage_xmom = S_stage_data[IntVars::xmom].const_array(mfi);
308  const Array4<const Real>& stage_ymom = S_stage_data[IntVars::ymom].const_array(mfi);
309  const Array4<const Real>& stage_zmom = S_stage_data[IntVars::zmom].const_array(mfi);
310  const Array4<const Real> & prim = S_stage_prim.const_array(mfi);
311 
312  const Array4<const Real>& prev_drho_theta = Delta_rho_theta.array(mfi);
313 
314  const Array4<const Real>& prev_cons = S_prev[IntVars::cons].const_array(mfi);
315  const Array4<const Real>& stage_cons = S_stage_data[IntVars::cons].const_array(mfi);
316 
317  const Array4<const Real>& slow_rhs_cons = S_slow_rhs[IntVars::cons].const_array(mfi);
318  const Array4<const Real>& slow_rhs_rho_w = S_slow_rhs[IntVars::zmom].const_array(mfi);
319 
320  const Array4<const Real>& prev_zmom = S_prev[IntVars::zmom].const_array(mfi);
321  const Array4< Real>& cur_zmom = S_data[IntVars::zmom].array(mfi);
322 
323  const Array4<Real>& temp_cur_xmom_arr = temp_cur_xmom.array(mfi);
324  const Array4<Real>& temp_cur_ymom_arr = temp_cur_ymom.array(mfi);
325 
326  // These store the advection momenta which we will use to update the slow variables
327  const Array4< Real>& avg_zmom_arr = avg_zmom.array(mfi);
328 
329  // Map factors
330  const Array4<const Real>& mf_mx = mapfac[MapFacType::m_x]->const_array(mfi);
331  const Array4<const Real>& mf_my = mapfac[MapFacType::m_y]->const_array(mfi);
332  const Array4<const Real>& mf_uy = mapfac[MapFacType::u_y]->const_array(mfi);
333  const Array4<const Real>& mf_vx = mapfac[MapFacType::v_x]->const_array(mfi);
334 
335  FArrayBox RHS_fab;
336  RHS_fab.resize(tbz,1, The_Async_Arena());
337 
338  FArrayBox soln_fab;
339  soln_fab.resize(tbz,1, The_Async_Arena());
340 
341  auto const& RHS_a = RHS_fab.array();
342  auto const& soln_a = soln_fab.array();
343 
344  auto const& temp_rhs_arr = temp_rhs.array(mfi);
345 
346  auto const& coeffA_a = coeff_A_mf.array(mfi);
347  auto const& inv_coeffB_a = inv_coeff_B_mf.array(mfi);
348  auto const& coeffC_a = coeff_C_mf.array(mfi);
349  auto const& coeffP_a = coeff_P_mf.array(mfi);
350  auto const& coeffQ_a = coeff_Q_mf.array(mfi);
351 
352  // *************************************************************************
353  // Define flux arrays for use in advection
354  // *************************************************************************
355  for (int dir = 0; dir < AMREX_SPACEDIM; ++dir) {
356  flux[dir].resize(surroundingNodes(bx,dir),2,The_Async_Arena());
357  flux[dir].setVal<RunOn::Device>(0);
358  }
359  const GpuArray<const Array4<Real>, AMREX_SPACEDIM>
360  flx_arr{{AMREX_D_DECL(flux[0].array(), flux[1].array(), flux[2].array())}};
361 
362  // *********************************************************************
363  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
364  Real xflux_lo = (temp_cur_xmom_arr(i ,j,k) - stage_xmom(i ,j,k)) / mf_uy(i ,j,0);
365  Real xflux_hi = (temp_cur_xmom_arr(i+1,j,k) - stage_xmom(i+1,j,k)) / mf_uy(i+1,j,0);
366  Real yflux_lo = (temp_cur_ymom_arr(i,j ,k) - stage_ymom(i,j ,k)) / mf_vx(i,j ,0);
367  Real yflux_hi = (temp_cur_ymom_arr(i,j+1,k) - stage_ymom(i,j+1,k)) / mf_vx(i,j+1,0);
368 
369  Real mfsq = mf_mx(i,j,0) * mf_my(i,j,0);
370 
371  temp_rhs_arr(i,j,k,Rho_comp ) = ( xflux_hi - xflux_lo ) * dxi * mfsq
372  + ( yflux_hi - yflux_lo ) * dyi * mfsq;
373  temp_rhs_arr(i,j,k,RhoTheta_comp) = (( xflux_hi * (prim(i,j,k,0) + prim(i+1,j,k,0)) -
374  xflux_lo * (prim(i,j,k,0) + prim(i-1,j,k,0)) ) * dxi * mfsq +
375  ( yflux_hi * (prim(i,j,k,0) + prim(i,j+1,k,0)) -
376  yflux_lo * (prim(i,j,k,0) + prim(i,j-1,k,0)) ) * dyi * mfsq) * myhalf;
377 
378  if (l_reflux) {
379  (flx_arr[0])(i,j,k,0) = xflux_lo;
380  (flx_arr[0])(i,j,k,1) = (flx_arr[0])(i ,j,k,0) * myhalf * (prim(i,j,k,0) + prim(i-1,j,k,0));
381 
382  (flx_arr[1])(i,j,k,0) = yflux_lo;
383  (flx_arr[1])(i,j,k,1) = (flx_arr[1])(i,j ,k,0) * myhalf * (prim(i,j,k,0) + prim(i,j-1,k,0));
384 
385  if (i == vbx_hi.x) {
386  (flx_arr[0])(i+1,j,k,0) = xflux_hi;
387  (flx_arr[0])(i+1,j,k,1) = (flx_arr[0])(i+1,j,k,0) * myhalf * (prim(i,j,k,0) + prim(i+1,j,k,0));
388  }
389  if (j == vbx_hi.y) {
390  (flx_arr[1])(i,j+1,k,0) = yflux_hi;
391  (flx_arr[1])(i,j+1,k,1) = (flx_arr[1])(i,j+1,k,0) * myhalf * (prim(i,j,k,0) + prim(i,j+1,k,0));
392  }
393  }
394  });
395 
396  Box bx_shrunk_in_k = bx;
397  int klo = tbz.smallEnd(2);
398  int khi = tbz.bigEnd(2);
399  bx_shrunk_in_k.setSmall(2,klo+1);
400  bx_shrunk_in_k.setBig(2,khi-1);
401 
402  // Note that the notes use "g" to mean the magnitude of gravity, so it is positive
403  // We set grav_gpu[2] to be the vector component which is negative
404  // We define halfg to match the notes (which is why we take the absolute value)
405  Real halfg = std::abs(myhalf * grav_gpu[2]);
406 
407  // *********************************************************************
408  // fast_loop_on_shrunk
409  // *********************************************************************
410  //Note we don't act on the bottom or top boundaries of the domain
411  ParallelFor(bx_shrunk_in_k, [=] AMREX_GPU_DEVICE (int i, int j, int k)
412  {
413  Real coeff_P = coeffP_a(i,j,k);
414  Real coeff_Q = coeffQ_a(i,j,k);
415 
416  Real theta_t_lo = myhalf * ( prim(i,j,k-2,PrimTheta_comp) + prim(i,j,k-1,PrimTheta_comp) );
417  Real theta_t_mid = myhalf * ( prim(i,j,k-1,PrimTheta_comp) + prim(i,j,k ,PrimTheta_comp) );
418  Real theta_t_hi = myhalf * ( prim(i,j,k ,PrimTheta_comp) + prim(i,j,k+1,PrimTheta_comp) );
419 
420  Real Omega_kp1 = prev_zmom(i,j,k+1) - stage_zmom(i,j,k+1);
421  Real Omega_k = prev_zmom(i,j,k ) - stage_zmom(i,j,k );
422  Real Omega_km1 = prev_zmom(i,j,k-1) - stage_zmom(i,j,k-1);
423 
424  // line 2 last two terms (order dtau)
425  Real old_drho_k = prev_cons(i,j,k ,Rho_comp) - stage_cons(i,j,k ,Rho_comp);
426  Real old_drho_km1 = prev_cons(i,j,k-1,Rho_comp) - stage_cons(i,j,k-1,Rho_comp);
427  Real R0_tmp = coeff_P * prev_drho_theta(i,j,k) + coeff_Q * prev_drho_theta(i,j,k-1)
428  - halfg * ( old_drho_k + old_drho_km1 );
429 
430  // lines 3-5 residuals (order dtau^2) one <-> beta_2
431  Real R1_tmp = halfg * (-slow_rhs_cons(i,j,k ,Rho_comp) - slow_rhs_cons(i,j,k-1,Rho_comp)
432  + temp_rhs_arr(i,j,k ,Rho_comp) + temp_rhs_arr(i,j,k-1,Rho_comp) )
433  + ( coeff_P * (slow_rhs_cons(i,j,k ,RhoTheta_comp) - temp_rhs_arr(i,j,k ,RhoTheta_comp)) +
434  coeff_Q * (slow_rhs_cons(i,j,k-1,RhoTheta_comp) - temp_rhs_arr(i,j,k-1,RhoTheta_comp)) );
435 
436  // lines 6&7 consolidated (reuse Omega & metrics) (order dtau^2)
437  R1_tmp += beta_1 * ( ( (Omega_kp1 - Omega_k) / dz_ptr[k] + (Omega_k - Omega_km1) / dz_ptr[k-1] ) * halfg
438  +(-(Omega_kp1*theta_t_hi - Omega_k *theta_t_mid) * coeff_P / dz_ptr[k]
439  -(Omega_k *theta_t_mid - Omega_km1*theta_t_lo ) * coeff_Q / dz_ptr[k-1]) );
440 
441  // line 1
442  RHS_a(i,j,k) = Omega_k + dtau * (slow_rhs_rho_w(i,j,k) + R0_tmp + dtau * beta_2 * R1_tmp + zmom_src_arr(i,j,k));
443 
444  }); // bx_shrunk_in_k
445 
446  Box b2d = tbz; // Copy constructor
447  b2d.setRange(2,0);
448 
449  auto const lo = lbound(bx);
450  auto const hi = ubound(bx);
451 
452  ParallelFor(b2d, [=] AMREX_GPU_DEVICE (int i, int j, int)
453  {
454  // w at bottom boundary of grid is 0 if at domain boundary, otherwise w = w_old + dtau * slow_rhs
455  RHS_a (i,j,lo.z) = prev_zmom(i,j,lo.z) - stage_zmom(i,j,lo.z)
456  + dtau * slow_rhs_rho_w(i,j,lo.z)
457  + dtau * zmom_src_arr(i,j,lo.z);
458 
459  // w at top boundary of grid is 0 if at domain boundary, otherwise w = w_old + dtau * slow_rhs
460  RHS_a (i,j,hi.z+1) = prev_zmom(i,j,hi.z+1) - stage_zmom(i,j,hi.z+1)
461  + dtau * slow_rhs_rho_w(i,j,hi.z+1)
462  + dtau * zmom_src_arr(i,j,hi.z+1);
463  }); // b2d
464 
465 #ifdef AMREX_USE_GPU
466  ParallelFor(b2d, [=] AMREX_GPU_DEVICE (int i, int j, int)
467  {
468  // w = specified Dirichlet value at k = lo.z
469  soln_a(i,j,lo.z) = RHS_a(i,j,lo.z) * inv_coeffB_a(i,j,lo.z);
470  cur_zmom(i,j,lo.z) = stage_zmom(i,j,lo.z) + soln_a(i,j,lo.z);
471 
472  for (int k = lo.z+1; k <= hi.z+1; k++) {
473  soln_a(i,j,k) = (RHS_a(i,j,k)-coeffA_a(i,j,k)*soln_a(i,j,k-1)) * inv_coeffB_a(i,j,k);
474  }
475 
476  cur_zmom(i,j,hi.z+1) = stage_zmom(i,j,hi.z+1) + soln_a(i,j,hi.z+1);
477 
478  for (int k = hi.z; k >= lo.z; k--) {
479  soln_a(i,j,k) -= ( coeffC_a(i,j,k) * inv_coeffB_a(i,j,k) ) *soln_a(i,j,k+1);
480  cur_zmom(i,j,k) = stage_zmom(i,j,k) + soln_a(i,j,k);
481  }
482  }); // b2d
483 #else
484  for (int j = lo.y; j <= hi.y; ++j) {
485  AMREX_PRAGMA_SIMD
486  for (int i = lo.x; i <= hi.x; ++i) {
487  soln_a(i,j,lo.z) = RHS_a(i,j,lo.z) * inv_coeffB_a(i,j,lo.z);
488  }
489  }
490  for (int k = lo.z+1; k <= hi.z+1; ++k) {
491  for (int j = lo.y; j <= hi.y; ++j) {
492  AMREX_PRAGMA_SIMD
493  for (int i = lo.x; i <= hi.x; ++i) {
494  soln_a(i,j,k) = (RHS_a(i,j,k)-coeffA_a(i,j,k)*soln_a(i,j,k-1)) * inv_coeffB_a(i,j,k);
495  }
496  }
497  }
498  for (int j = lo.y; j <= hi.y; ++j) {
499  AMREX_PRAGMA_SIMD
500  for (int i = lo.x; i <= hi.x; ++i) {
501  cur_zmom(i,j,hi.z+1) = stage_zmom(i,j,hi.z+1) + soln_a(i,j,hi.z+1);
502  }
503  }
504  for (int k = hi.z; k >= lo.z; --k) {
505  for (int j = lo.y; j <= hi.y; ++j) {
506  AMREX_PRAGMA_SIMD
507  for (int i = lo.x; i <= hi.x; ++i) {
508  soln_a(i,j,k) -= ( coeffC_a(i,j,k) * inv_coeffB_a(i,j,k) ) * soln_a(i,j,k+1);
509  cur_zmom(i,j,k) = stage_zmom(i,j,k) + soln_a(i,j,k);
510  }
511  }
512  }
513 #endif
514  if (l_rayleigh_impl_for_w) {
515  ParallelFor(bx_shrunk_in_k, [=] AMREX_GPU_DEVICE (int i, int j, int k)
516  {
517  Real damping_coeff = l_damp_coef * dtau * sinesq_stag_d[k];
518  cur_zmom(i,j,k) /= (one + damping_coeff);
519  });
520  }
521 
522  // **************************************************************************
523  // Define updates in the RHS of rho and (rho theta)
524  // **************************************************************************
525  const Array4<Real>& prev_drho_w = Delta_rho_w.array(mfi);
526  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
527  {
528  Real zflux_lo = beta_2 * soln_a(i,j,k ) + beta_1 * prev_drho_w(i,j,k );
529  Real zflux_hi = beta_2 * soln_a(i,j,k+1) + beta_1 * prev_drho_w(i,j,k+1);
530 
531  avg_zmom_arr(i,j,k) += facinv*zflux_lo / (mf_mx(i,j,0) * mf_my(i,j,0));
532  if (l_reflux) {
533  (flx_arr[2])(i,j,k,0) = zflux_lo / (mf_mx(i,j,0) * mf_my(i,j,0));
534  (flx_arr[2])(i,j,k,1) = (flx_arr[2])(i,j,k,0) * myhalf * (prim(i,j,k) + prim(i,j,k-1));
535  }
536 
537  if (k == vbx_hi.z) {
538  avg_zmom_arr(i,j,k+1) += facinv * zflux_hi / (mf_mx(i,j,0) * mf_my(i,j,0));
539  if (l_reflux) {
540  (flx_arr[2])(i,j,k+1,0) = zflux_hi / (mf_mx(i,j,0) * mf_my(i,j,0));
541  (flx_arr[2])(i,j,k+1,1) = (flx_arr[2])(i,j,k+1,0) * myhalf * (prim(i,j,k) + prim(i,j,k+1));
542  }
543  }
544 
545  Real dz_inv = one / dz_ptr[k];
546  temp_rhs_arr(i,j,k,Rho_comp ) += dz_inv * ( zflux_hi - zflux_lo );
547  temp_rhs_arr(i,j,k,RhoTheta_comp) += myhalf * dz_inv * ( zflux_hi * (prim(i,j,k) + prim(i,j,k+1))
548  - zflux_lo * (prim(i,j,k) + prim(i,j,k-1)) );
549  });
550 
551  // We only add to the flux registers in the final RK step
552  if (l_reflux) {
553  int strt_comp_reflux = 0;
554  // For now we don't reflux (rho theta) because it seems to create issues at c/f boundaries
555  int num_comp_reflux = 1;
556  if (level < finest_level) {
557  fr_as_crse->CrseAdd(mfi,
558  {{AMREX_D_DECL(&(flux[0]), &(flux[1]), &(flux[2]))}},
559  dx, dtau, strt_comp_reflux, strt_comp_reflux, num_comp_reflux, RunOn::Device);
560  }
561  if (level > 0) {
562  fr_as_fine->FineAdd(mfi,
563  {{AMREX_D_DECL(&(flux[0]), &(flux[1]), &(flux[2]))}},
564  dx, dtau, strt_comp_reflux, strt_comp_reflux, num_comp_reflux, RunOn::Device);
565  }
566 
567  // This is necessary here so we don't go on to the next FArrayBox without
568  // having finished copying the fluxes into the FluxRegisters (since the fluxes
569  // are stored in temporary FArrayBox's)
570  Gpu::streamSynchronize();
571 
572  } // two-way coupling
573  } // mfi
574  } // OMP
575 
576 #ifdef _OPENMP
577 #pragma omp parallel if (Gpu::notInLaunchRegion())
578 #endif
579  for ( MFIter mfi(S_stage_data[IntVars::cons],TilingIfNotGPU()); mfi.isValid(); ++mfi)
580  {
581  const Box& bx = mfi.tilebox();
582 
583  const Array4< Real>& cur_cons = S_data[IntVars::cons].array(mfi);
584  const Array4<const Real>& prev_cons = S_prev[IntVars::cons].const_array(mfi);
585  auto const& temp_rhs_arr = temp_rhs.const_array(mfi);
586  auto const& slow_rhs_cons = S_slow_rhs[IntVars::cons].const_array(mfi);
587  const Array4<Real const>& cc_src_arr = cc_src.const_array(mfi);
588 
589  if (step == 0) {
590  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
591  {
592  cur_cons(i,j,k,Rho_comp) = prev_cons(i,j,k,Rho_comp) +
593  dtau * (slow_rhs_cons(i,j,k,Rho_comp) - temp_rhs_arr(i,j,k,Rho_comp));
594  cur_cons(i,j,k,RhoTheta_comp) = prev_cons(i,j,k,RhoTheta_comp) +
595  dtau * (slow_rhs_cons(i,j,k,RhoTheta_comp) - temp_rhs_arr(i,j,k,RhoTheta_comp));
596 
597  // add in source terms for cell-centered conserved variables
598  cur_cons(i,j,k,Rho_comp) += dtau * cc_src_arr(i,j,k,Rho_comp);
599  cur_cons(i,j,k,RhoTheta_comp) += dtau * cc_src_arr(i,j,k,RhoTheta_comp);
600  });
601  } else {
602  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
603  {
604  //
605  // We didn't need to set cur_cons = prev_cons above because they point to the same data for step > 0
606  //
607  cur_cons(i,j,k,Rho_comp) += dtau * (slow_rhs_cons(i,j,k,Rho_comp) - temp_rhs_arr(i,j,k,Rho_comp));
608  cur_cons(i,j,k,RhoTheta_comp) += dtau * (slow_rhs_cons(i,j,k,RhoTheta_comp) - temp_rhs_arr(i,j,k,RhoTheta_comp));
609 
610  // add in source terms for cell-centered conserved variables
611  cur_cons(i,j,k,Rho_comp) += dtau * cc_src_arr(i,j,k,Rho_comp);
612  cur_cons(i,j,k,RhoTheta_comp) += dtau * cc_src_arr(i,j,k,RhoTheta_comp);
613  });
614  } // step = 0
615 
616  const Array4<Real>& cur_xmom = S_data[IntVars::xmom].array(mfi);
617  const Array4<Real>& cur_ymom = S_data[IntVars::ymom].array(mfi);
618 
619  const Array4<Real const>& temp_cur_xmom_arr = temp_cur_xmom.const_array(mfi);
620  const Array4<Real const>& temp_cur_ymom_arr = temp_cur_ymom.const_array(mfi);
621 
622  Box tbx = surroundingNodes(bx,0);
623  Box tby = surroundingNodes(bx,1);
624 
625  ParallelFor(tbx, tby,
626  [=] AMREX_GPU_DEVICE (int i, int j, int k)
627  {
628  cur_xmom(i,j,k) = temp_cur_xmom_arr(i,j,k);
629  },
630  [=] AMREX_GPU_DEVICE (int i, int j, int k)
631  {
632  cur_ymom(i,j,k) = temp_cur_ymom_arr(i,j,k);
633  });
634 
635  } // mfi
636 }
R_v
constexpr amrex::Real R_v
Definition: ERF_Constants.H:48
one
constexpr amrex::Real one
Definition: ERF_Constants.H:9
zero
constexpr amrex::Real zero
Definition: ERF_Constants.H:8
myhalf
constexpr amrex::Real myhalf
Definition: ERF_Constants.H:13
R_d
constexpr amrex::Real R_d
Definition: ERF_Constants.H:47
Gamma
constexpr amrex::Real Gamma
Definition: ERF_Constants.H:62
v_x
@ v_x
Definition: ERF_DataStruct.H:24
u_y
@ u_y
Definition: ERF_DataStruct.H:25
v_y
@ v_y
Definition: ERF_DataStruct.H:25
m_y
@ m_y
Definition: ERF_DataStruct.H:25
u_x
@ u_x
Definition: ERF_DataStruct.H:24
m_x
@ m_x
Definition: ERF_DataStruct.H:24
PrimQ1_comp
#define PrimQ1_comp
Definition: ERF_IndexDefines.H:58
Rho_comp
#define Rho_comp
Definition: ERF_IndexDefines.H:36
RhoTheta_comp
#define RhoTheta_comp
Definition: ERF_IndexDefines.H:37
PrimTheta_comp
#define PrimTheta_comp
Definition: ERF_IndexDefines.H:55
dxInv
amrex::GpuArray< Real, AMREX_SPACEDIM > dxInv
Definition: ERF_InitCustomPertVels_ParticleTests.H:17
dx
const Real dx
Definition: ERF_InitCustomPert_ABL.H:23
khi
const int khi
Definition: ERF_InitCustomPert_Bubble.H:21
AMREX_ALWAYS_ASSERT
AMREX_ALWAYS_ASSERT(bx.length()[2]==khi+1)
ParallelFor
ParallelFor(grown_box, [=] AMREX_GPU_DEVICE(int i, int j, int k) { qrcuten_arr(i, j, k)=Real(0);qscuten_arr(i, j, k)=Real(0);qicuten_arr(i, j, k)=Real(0);})
Real
amrex::Real Real
Definition: ERF_ShocInterface.H:19
TileNoZ
AMREX_FORCE_INLINE amrex::IntVect TileNoZ()
Definition: ERF_TileNoZ.H:11
GpVars::gpy
@ gpy
Definition: ERF_IndexDefines.H:186
GpVars::gpx
@ gpx
Definition: ERF_IndexDefines.H:185
IntVars::ymom
@ ymom
Definition: ERF_IndexDefines.H:195
IntVars::cons
@ cons
Definition: ERF_IndexDefines.H:193
IntVars::zmom
@ zmom
Definition: ERF_IndexDefines.H:196
IntVars::xmom
@ xmom
Definition: ERF_IndexDefines.H:194
MicVar_Kess::qt
@ qt
Definition: ERF_Kessler.H:29
MicVar_Kess::qv
@ qv
Definition: ERF_Kessler.H:30
WSM6Ind::q
@ q
Definition: ERF_WSM6.H:182
Here is the call graph for this function: