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

Functions

void erf_fast_rhs_MT (int step, int, int level, int finest_level, Vector< MultiFab > &S_slow_rhs, const Vector< MultiFab > &S_prev, Vector< MultiFab > &S_stg_data, const MultiFab &S_stg_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, const bool use_lagged_delta_rt, std::unique_ptr< MultiFab > &z_t_rk, const MultiFab *z_t_pert, std::unique_ptr< MultiFab > &z_phys_nd_old, std::unique_ptr< MultiFab > &z_phys_nd_new, std::unique_ptr< MultiFab > &z_phys_nd_stg, std::unique_ptr< MultiFab > &detJ_cc_old, std::unique_ptr< MultiFab > &detJ_cc_new, std::unique_ptr< MultiFab > &detJ_cc_stg, const Real dtau, 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)
 

Function Documentation

◆ erf_fast_rhs_MT()

void erf_fast_rhs_MT ( int  step,
int  ,
int  level,
int  finest_level,
Vector< MultiFab > &  S_slow_rhs,
const Vector< MultiFab > &  S_prev,
Vector< MultiFab > &  S_stg_data,
const MultiFab &  S_stg_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,
const bool  use_lagged_delta_rt,
std::unique_ptr< MultiFab > &  z_t_rk,
const MultiFab *  z_t_pert,
std::unique_ptr< MultiFab > &  z_phys_nd_old,
std::unique_ptr< MultiFab > &  z_phys_nd_new,
std::unique_ptr< MultiFab > &  z_phys_nd_stg,
std::unique_ptr< MultiFab > &  detJ_cc_old,
std::unique_ptr< MultiFab > &  detJ_cc_new,
std::unique_ptr< MultiFab > &  detJ_cc_stg,
const Real  dtau,
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   
)

Function for computing the fast RHS with moving terrain

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_prevprevious solution
[in]S_stg_datasolution at previous RK stage
[in]S_stg_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]use_lagged_delta_rtdefine lagged_delta_rt for our next step
[in]z_t_rkrate of change of grid height – only relevant for moving terrain
[in]z_t_pertrate of change of grid height – interpolated between RK stages
[in]z_phys_nd_oldheight coordinate at nodes at old time
[in]z_phys_nd_newheight coordinate at nodes at new time
[in]z_phys_nd_stgheight coordinate at nodes at previous stage
[in]detJ_cc_oldJacobian of the metric transformation at old time
[in]detJ_cc_newJacobian of the metric transformation at new time
[in]detJ_cc_stgJacobian of the metric transformation at previous stage
[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?
[in]l_implicit_substepping
87 {
88  BL_PROFILE_REGION("erf_fast_rhs_MT()");
89 
90  Real beta_1 = 0.5 * (1.0 - beta_s); // multiplies explicit terms
91  Real beta_2 = 0.5 * (1.0 + beta_s); // multiplies implicit terms
92 
93  // How much do we project forward the (rho theta) that is used in the horizontal momentum equations
94  Real beta_d = 0.1;
95 
96  Real RvOverRd = R_v / R_d;
97 
98  const Real* dx = geom.CellSize();
99  const GpuArray<Real, AMREX_SPACEDIM> dxInv = geom.InvCellSizeArray();
100 
101  Real dxi = dxInv[0];
102  Real dyi = dxInv[1];
103  Real dzi = dxInv[2];
104 
105  MultiFab coeff_A_mf(fast_coeffs, make_alias, 0, 1);
106  MultiFab inv_coeff_B_mf(fast_coeffs, make_alias, 1, 1);
107  MultiFab coeff_C_mf(fast_coeffs, make_alias, 2, 1);
108  MultiFab coeff_P_mf(fast_coeffs, make_alias, 3, 1);
109  MultiFab coeff_Q_mf(fast_coeffs, make_alias, 4, 1);
110 
111  // *************************************************************************
112  // Set gravity as a vector
113  const Array<Real,AMREX_SPACEDIM> grav{0.0, 0.0, -gravity};
114  const GpuArray<Real,AMREX_SPACEDIM> grav_gpu{grav[0], grav[1], grav[2]};
115 
116  MultiFab extrap(S_data[IntVars::cons].boxArray(),S_data[IntVars::cons].DistributionMap(),1,1);
117 
118  MultiFab Omega(S_data[IntVars::zmom].boxArray(), S_data[IntVars::zmom].DistributionMap(), 1, 1);
119 
120  // *************************************************************************
121  // Define updates in the current RK stg
122  // *************************************************************************
123 #ifdef _OPENMP
124 #pragma omp parallel if (Gpu::notInLaunchRegion())
125 #endif
126  {
127  FArrayBox temp_rhs_fab;
128 
129  FArrayBox RHS_fab;
130  FArrayBox soln_fab;
131 
132  std::array<FArrayBox,AMREX_SPACEDIM> flux;
133 
134  // NOTE: we leave tiling off here for efficiency -- to make this loop work with tiling
135  // will require additional changes
136  for ( MFIter mfi(S_stg_data[IntVars::cons],false); mfi.isValid(); ++mfi)
137  {
138  Box bx = mfi.tilebox();
139  Box tbx = surroundingNodes(bx,0);
140  Box tby = surroundingNodes(bx,1);
141  Box tbz = surroundingNodes(bx,2);
142 
143  Box vbx = mfi.validbox();
144  const auto& vbx_hi = ubound(vbx);
145 
146  const Array4<Real const>& xmom_src_arr = xmom_src.const_array(mfi);
147  const Array4<Real const>& ymom_src_arr = ymom_src.const_array(mfi);
148  const Array4<Real const>& zmom_src_arr = zmom_src.const_array(mfi);
149  const Array4<Real const>& cc_src_arr = cc_src.const_array(mfi);
150 
151  const Array4<const Real> & stg_cons = S_stg_data[IntVars::cons].const_array(mfi);
152  const Array4<const Real> & stg_xmom = S_stg_data[IntVars::xmom].const_array(mfi);
153  const Array4<const Real> & stg_ymom = S_stg_data[IntVars::ymom].const_array(mfi);
154  const Array4<const Real> & stg_zmom = S_stg_data[IntVars::zmom].const_array(mfi);
155  const Array4<const Real> & prim = S_stg_prim.const_array(mfi);
156  const Array4<const Real> & qt_arr = qt.const_array(mfi);
157 
158  const Array4<const Real>& slow_rhs_cons = S_slow_rhs[IntVars::cons].const_array(mfi);
159  const Array4<const Real>& slow_rhs_rho_u = S_slow_rhs[IntVars::xmom].const_array(mfi);
160  const Array4<const Real>& slow_rhs_rho_v = S_slow_rhs[IntVars::ymom].const_array(mfi);
161  const Array4<const Real>& slow_rhs_rho_w = S_slow_rhs[IntVars::zmom].const_array(mfi);
162 
163  const Array4<Real>& cur_cons = S_data[IntVars::cons].array(mfi);
164  const Array4<Real>& cur_xmom = S_data[IntVars::xmom].array(mfi);
165  const Array4<Real>& cur_ymom = S_data[IntVars::ymom].array(mfi);
166  const Array4<Real>& cur_zmom = S_data[IntVars::zmom].array(mfi);
167 
168  const Array4<Real>& lagged = lagged_delta_rt.array(mfi);
169 
170  const Array4<const Real>& prev_cons = S_prev[IntVars::cons].const_array(mfi);
171  const Array4<const Real>& prev_xmom = S_prev[IntVars::xmom].const_array(mfi);
172  const Array4<const Real>& prev_ymom = S_prev[IntVars::ymom].const_array(mfi);
173  const Array4<const Real>& prev_zmom = S_prev[IntVars::zmom].const_array(mfi);
174 
175  // These store the advection momenta which we will use to update the slow variables
176  const Array4<Real>& avg_xmom_arr = avg_xmom.array(mfi);
177  const Array4<Real>& avg_ymom_arr = avg_ymom.array(mfi);
178  const Array4<Real>& avg_zmom_arr = avg_zmom.array(mfi);
179 
180  const Array4<const Real>& z_nd_old = z_phys_nd_old->const_array(mfi);
181  const Array4<const Real>& z_nd_new = z_phys_nd_new->const_array(mfi);
182  const Array4<const Real>& z_nd_stg = z_phys_nd_stg->const_array(mfi);
183  const Array4<const Real>& detJ_old = detJ_cc_old->const_array(mfi);
184  const Array4<const Real>& detJ_new = detJ_cc_new->const_array(mfi);
185  const Array4<const Real>& detJ_stg = detJ_cc_stg->const_array(mfi);
186 
187  const Array4<const Real>& z_t_arr = z_t_rk->const_array(mfi);
188  const Array4<const Real>& zp_t_arr = z_t_pert->const_array(mfi);
189 
190  const Array4< Real>& omega_arr = Omega.array(mfi);
191 
192  const Array4<const Real>& pi_stage_ca = pi_stage.const_array(mfi);
193 
194  const Array4<Real>& theta_extrap = extrap.array(mfi);
195 
196  // Map factors
197  const Array4<const Real>& mf_mx = mapfac[MapFacType::m_x]->const_array(mfi);
198  const Array4<const Real>& mf_my = mapfac[MapFacType::m_y]->const_array(mfi);
199  const Array4<const Real>& mf_ux = mapfac[MapFacType::u_x]->const_array(mfi);
200  const Array4<const Real>& mf_vy = mapfac[MapFacType::v_y]->const_array(mfi);
201 
202  // Note: it is important to grow the tilebox rather than use growntilebox because
203  // we need to fill the ghost cells of the tilebox so we can use them below
204  Box gbx = mfi.tilebox(); gbx.grow(1);
205  Box gtbx = mfi.nodaltilebox(0); gtbx.grow(1); gtbx.setSmall(2,0);
206  Box gtby = mfi.nodaltilebox(1); gtby.grow(1); gtby.setSmall(2,0);
207 
208  {
209  BL_PROFILE("fast_rhs_copies_0");
210  if (step == 0) {
211  ParallelFor(gbx,
212  [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
213  cur_cons(i,j,k,Rho_comp) = prev_cons(i,j,k,Rho_comp);
214  cur_cons(i,j,k,RhoTheta_comp) = prev_cons(i,j,k,RhoTheta_comp);
215 
216  Real delta_rt = cur_cons(i,j,k,RhoTheta_comp) - stg_cons(i,j,k,RhoTheta_comp);
217  theta_extrap(i,j,k) = delta_rt;
218 
219  // NOTE: qv is not changing over the fast steps so we use the stage data
220  Real qv = (l_use_moisture) ? prim(i,j,k,PrimQ1_comp) : 0.0;
221  theta_extrap(i,j,k) *= (1.0 + RvOverRd*qv);
222 
223  // We define lagged_delta_rt for our next step as the current delta_rt
224  lagged(i,j,k) = delta_rt;
225  });
226  } else if (use_lagged_delta_rt) {
227  // This is the default for cases with no or static terrain
228  ParallelFor(gbx,
229  [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
230  Real delta_rt = cur_cons(i,j,k,RhoTheta_comp) - stg_cons(i,j,k,RhoTheta_comp);
231  theta_extrap(i,j,k) = delta_rt + beta_d * (delta_rt - lagged(i,j,k));
232 
233  // NOTE: qv is not changing over the fast steps so we use the stage data
234  Real qv = (l_use_moisture) ? prim(i,j,k,PrimQ1_comp) : 0.0;
235  theta_extrap(i,j,k) *= (1.0 + RvOverRd*qv);
236 
237  // We define lagged_delta_rt for our next step as the current delta_rt
238  lagged(i,j,k) = delta_rt;
239  });
240  } else {
241  // For the moving wave problem, this choice seems more robust
242  ParallelFor(gbx,
243  [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
244  theta_extrap(i,j,k) = cur_cons(i,j,k,RhoTheta_comp) - stg_cons(i,j,k,RhoTheta_comp);
245 
246  // NOTE: qv is not changing over the fast steps so we use the stage data
247  Real qv = (l_use_moisture) ? prim(i,j,k,PrimQ1_comp) : 0.0;
248  theta_extrap(i,j,k) *= (1.0 + RvOverRd*qv);
249  });
250  } // if step
251  } // end profile
252 
253  RHS_fab.resize (tbz,1, The_Async_Arena());
254  soln_fab.resize (tbz,1, The_Async_Arena());
255  temp_rhs_fab.resize(tbz,2, The_Async_Arena());
256 
257  auto const& RHS_a = RHS_fab.array();
258  auto const& soln_a = soln_fab.array();
259  auto const& temp_rhs_arr = temp_rhs_fab.array();
260 
261  auto const& coeffA_a = coeff_A_mf.array(mfi);
262  auto const& inv_coeffB_a = inv_coeff_B_mf.array(mfi);
263  auto const& coeffC_a = coeff_C_mf.array(mfi);
264  auto const& coeffP_a = coeff_P_mf.array(mfi);
265  auto const& coeffQ_a = coeff_Q_mf.array(mfi);
266 
267  // *********************************************************************
268  // Define updates in the RHS of {x, y, z}-momentum equations
269  // *********************************************************************
270  {
271  BL_PROFILE("fast_rhs_xymom_T");
272  ParallelFor(tbx, tby,
273  [=] AMREX_GPU_DEVICE (int i, int j, int k)
274  {
275  // Add (negative) gradient of (rho theta) multiplied by lagged "pi"
276  Real h_xi_old = Compute_h_xi_AtIface(i, j, k, dxInv, z_nd_old);
277  Real h_zeta_old = Compute_h_zeta_AtIface(i, j, k, dxInv, z_nd_old);
278  Real gp_xi = (theta_extrap(i,j,k) - theta_extrap(i-1,j,k)) * dxi;
279  Real gp_zeta_on_iface = (k == 0) ?
280  0.5 * dzi * ( theta_extrap(i-1,j,k+1) + theta_extrap(i,j,k+1)
281  -theta_extrap(i-1,j,k ) - theta_extrap(i,j,k ) ) :
282  0.25 * dzi * ( theta_extrap(i-1,j,k+1) + theta_extrap(i,j,k+1)
283  -theta_extrap(i-1,j,k-1) - theta_extrap(i,j,k-1) );
284  Real gpx = h_zeta_old * gp_xi - h_xi_old * gp_zeta_on_iface;
285  gpx *= mf_ux(i,j,0);
286 
287  Real q = (l_use_moisture) ? 0.5 * (qt_arr(i-1,j,k) + qt_arr(i,j,k)) : 0.0;
288 
289  Real pi_c = 0.5 * (pi_stage_ca(i-1,j,k,0) + pi_stage_ca(i ,j,k,0));
290  Real fast_rhs_rho_u = -Gamma * R_d * pi_c * gpx / (1.0 + q);
291 
292  // We have already scaled the source terms to have the extra factor of dJ
293  cur_xmom(i,j,k) = h_zeta_old * prev_xmom(i,j,k) + dtau * fast_rhs_rho_u
294  + dtau * slow_rhs_rho_u(i,j,k)
295  + dtau * xmom_src_arr(i,j,k);
296  },
297  [=] AMREX_GPU_DEVICE (int i, int j, int k)
298  {
299  // Add (negative) gradient of (rho theta) multiplied by lagged "pi"
300  Real h_eta_old = Compute_h_eta_AtJface(i, j, k, dxInv, z_nd_old);
301  Real h_zeta_old = Compute_h_zeta_AtJface(i, j, k, dxInv, z_nd_old);
302  Real gp_eta = (theta_extrap(i,j,k) -theta_extrap(i,j-1,k)) * dyi;
303  Real gp_zeta_on_jface = (k == 0) ?
304  0.5 * dzi * ( theta_extrap(i,j,k+1) + theta_extrap(i,j-1,k+1)
305  -theta_extrap(i,j,k ) - theta_extrap(i,j-1,k ) ) :
306  0.25 * dzi * ( theta_extrap(i,j,k+1) + theta_extrap(i,j-1,k+1)
307  -theta_extrap(i,j,k-1) - theta_extrap(i,j-1,k-1) );
308  Real gpy = h_zeta_old * gp_eta - h_eta_old * gp_zeta_on_jface;
309  gpy *= mf_vy(i,j,0);
310 
311  Real q = (l_use_moisture) ? 0.5 * (qt_arr(i,j-1,k) + qt_arr(i,j,k)) : 0.0;
312 
313  Real pi_c = 0.5 * (pi_stage_ca(i,j-1,k,0) + pi_stage_ca(i,j ,k,0));
314  Real fast_rhs_rho_v = -Gamma * R_d * pi_c * gpy / (1.0 + q);
315 
316  // We have already scaled the source terms to have the extra factor of dJ
317  cur_ymom(i, j, k) = h_zeta_old * prev_ymom(i,j,k) + dtau * fast_rhs_rho_v
318  + dtau * slow_rhs_rho_v(i,j,k)
319  + dtau * ymom_src_arr(i,j,k);
320  });
321  } // end profile
322 
323  // *************************************************************************
324  // Define flux arrays for use in advection
325  // *************************************************************************
326  for (int dir = 0; dir < AMREX_SPACEDIM; ++dir) {
327  flux[dir].resize(surroundingNodes(bx,dir),2);
328  flux[dir].setVal<RunOn::Device>(0.);
329  }
330  const GpuArray<const Array4<Real>, AMREX_SPACEDIM>
331  flx_arr{{AMREX_D_DECL(flux[0].array(), flux[1].array(), flux[2].array())}};
332 
333  // *********************************************************************
334  {
335  BL_PROFILE("fast_T_making_rho_rhs");
336  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
337  {
338  Real h_zeta_stg_xlo = Compute_h_zeta_AtIface(i, j , k, dxInv, z_nd_stg);
339  Real h_zeta_stg_xhi = Compute_h_zeta_AtIface(i+1,j , k, dxInv, z_nd_stg);
340  Real xflux_lo = cur_xmom(i ,j,k) - stg_xmom(i ,j,k)*h_zeta_stg_xlo;
341  Real xflux_hi = cur_xmom(i+1,j,k) - stg_xmom(i+1,j,k)*h_zeta_stg_xhi;
342 
343  Real h_zeta_stg_yhi = Compute_h_zeta_AtJface(i, j+1, k, dxInv, z_nd_stg);
344  Real h_zeta_stg_ylo = Compute_h_zeta_AtJface(i, j , k, dxInv, z_nd_stg);
345  Real yflux_lo = cur_ymom(i,j ,k) - stg_ymom(i,j ,k)*h_zeta_stg_ylo;
346  Real yflux_hi = cur_ymom(i,j+1,k) - stg_ymom(i,j+1,k)*h_zeta_stg_yhi;
347 
348  // NOTE: we are saving the (1/J) weighting for later when we add this to rho and theta
349  temp_rhs_arr(i,j,k,0) = ( xflux_hi - xflux_lo ) * dxi + ( yflux_hi - yflux_lo ) * dyi;
350  temp_rhs_arr(i,j,k,1) = (( xflux_hi * (prim(i,j,k,0) + prim(i+1,j,k,0)) -
351  xflux_lo * (prim(i,j,k,0) + prim(i-1,j,k,0)) ) * dxi +
352  ( yflux_hi * (prim(i,j,k,0) + prim(i,j+1,k,0)) -
353  yflux_lo * (prim(i,j,k,0) + prim(i,j-1,k,0)) ) * dyi) * 0.5;
354 
355  if (l_reflux) {
356  (flx_arr[0])(i,j,k,0) = xflux_lo;
357  (flx_arr[0])(i,j,k,1) = (flx_arr[0])(i ,j,k,0) * 0.5 * (prim(i,j,k,0) + prim(i-1,j,k,0));
358 
359  (flx_arr[1])(i,j,k,0) = yflux_lo;
360  (flx_arr[1])(i,j,k,1) = (flx_arr[1])(i,j ,k,0) * 0.5 * (prim(i,j,k,0) + prim(i,j-1,k,0));
361 
362  if (i == vbx_hi.x) {
363  (flx_arr[0])(i+1,j,k,0) = xflux_hi;
364  (flx_arr[0])(i+1,j,k,1) = (flx_arr[0])(i+1,j,k,0) * 0.5 * (prim(i,j,k,0) + prim(i+1,j,k,0));
365  }
366  if (j == vbx_hi.y) {
367  (flx_arr[1])(i,j+1,k,0) = yflux_hi;
368  (flx_arr[1])(i,j+1,k,1) = (flx_arr[1])(i,j+1,k,0) * 0.5 * (prim(i,j,k,0) + prim(i,j+1,k,0));
369  }
370  }
371  });
372  } // end profile
373 
374  // *********************************************************************
375  // This must be done before we set cur_xmom and cur_ymom, since those
376  // in fact point to the same array as prev_xmom and prev_ymom
377  // *********************************************************************
378  Box gbxo = mfi.nodaltilebox(2);
379  {
380  BL_PROFILE("fast_MT_making_omega");
381  Box gbxo_lo = gbxo; gbxo_lo.setBig(2,0);
382  ParallelFor(gbxo_lo, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
383  omega_arr(i,j,k) = 0.;
384  });
385  Box gbxo_hi = gbxo; gbxo_hi.setSmall(2,gbxo.bigEnd(2));
386  ParallelFor(gbxo_hi, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
387  omega_arr(i,j,k) = prev_zmom(i,j,k) - stg_zmom(i,j,k) - zp_t_arr(i,j,k);
388  });
389  Box gbxo_mid = gbxo; gbxo_mid.setSmall(2,1); gbxo_mid.setBig(2,gbxo.bigEnd(2)-1);
390  ParallelFor(gbxo_mid, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
391  omega_arr(i,j,k) =
392  ( OmegaFromW(i,j,k,prev_zmom(i,j,k),prev_xmom,prev_ymom,mf_ux,mf_vy,z_nd_old,dxInv)
393  -OmegaFromW(i,j,k, stg_zmom(i,j,k), stg_xmom, stg_ymom,mf_ux,mf_vy,z_nd_old,dxInv) )
394  - zp_t_arr(i,j,k);
395  });
396  } // end profile
397  // *********************************************************************
398 
399  ParallelFor(tbx, tby,
400  [=] AMREX_GPU_DEVICE (int i, int j, int k)
401  {
402  Real h_zeta_new = Compute_h_zeta_AtIface(i, j, k, dxInv, z_nd_new);
403  cur_xmom(i, j, k) /= h_zeta_new;
404  avg_xmom_arr(i,j,k) += facinv*(cur_xmom(i,j,k) - stg_xmom(i,j,k));
405  },
406  [=] AMREX_GPU_DEVICE (int i, int j, int k)
407  {
408  Real h_zeta_new = Compute_h_zeta_AtJface(i, j, k, dxInv, z_nd_new);
409  cur_ymom(i, j, k) /= h_zeta_new;
410  avg_ymom_arr(i,j,k) += facinv*(cur_ymom(i,j,k) - stg_ymom(i,j,k));
411  });
412 
413  Box bx_shrunk_in_k = bx;
414  int klo = tbz.smallEnd(2);
415  int khi = tbz.bigEnd(2);
416  bx_shrunk_in_k.setSmall(2,klo+1);
417  bx_shrunk_in_k.setBig(2,khi-1);
418 
419  // Note that the notes use "g" to mean the magnitude of gravity, so it is positive
420  // We set grav_gpu[2] to be the vector component which is negative
421  // We define halfg to match the notes (which is why we take the absolute value)
422  Real halfg = std::abs(0.5 * grav_gpu[2]);
423 
424  {
425  BL_PROFILE("fast_loop_on_shrunk_t");
426  //Note we don't act on the bottom or top boundaries of the domain
427  ParallelFor(bx_shrunk_in_k, [=] AMREX_GPU_DEVICE (int i, int j, int k)
428  {
429  Real dJ_old_kface = 0.5 * (detJ_old(i,j,k) + detJ_old(i,j,k-1));
430  Real dJ_new_kface = 0.5 * (detJ_new(i,j,k) + detJ_new(i,j,k-1));
431  Real dJ_stg_kface = 0.5 * (detJ_stg(i,j,k) + detJ_stg(i,j,k-1));
432 
433  Real q = (l_use_moisture) ? 0.5 * (qt_arr(i,j,k-1) + qt_arr(i,j,k)) : 0.0;
434 
435  Real coeff_P = coeffP_a(i,j,k) / (1.0 + q);
436  Real coeff_Q = coeffQ_a(i,j,k) / (1.0 + q);
437 
438  Real theta_t_lo = 0.5 * ( prim(i,j,k-2,PrimTheta_comp) + prim(i,j,k-1,PrimTheta_comp) );
439  Real theta_t_mid = 0.5 * ( prim(i,j,k-1,PrimTheta_comp) + prim(i,j,k ,PrimTheta_comp) );
440  Real theta_t_hi = 0.5 * ( prim(i,j,k ,PrimTheta_comp) + prim(i,j,k+1,PrimTheta_comp) );
441 
442  // line 2 last two terms (order dtau)
443  Real R0_tmp = coeff_P * cur_cons(i,j,k ,RhoTheta_comp) * dJ_old_kface
444  + coeff_Q * cur_cons(i,j,k-1,RhoTheta_comp) * dJ_old_kface
445  - coeff_P * stg_cons(i,j,k ,RhoTheta_comp) * dJ_stg_kface
446  - coeff_Q * stg_cons(i,j,k-1,RhoTheta_comp) * dJ_stg_kface
447  - halfg * ( cur_cons(i,j,k,Rho_comp) + cur_cons(i,j,k-1,Rho_comp) ) * dJ_old_kface
448  + halfg * ( stg_cons(i,j,k,Rho_comp) + stg_cons(i,j,k-1,Rho_comp) ) * dJ_stg_kface;
449 
450  // line 3 residuals (order dtau^2) 1.0 <-> beta_2
451  Real R1_tmp = - halfg * ( slow_rhs_cons(i,j,k ,Rho_comp) +
452  slow_rhs_cons(i,j,k-1,Rho_comp) )
453  + ( coeff_P * slow_rhs_cons(i,j,k ,RhoTheta_comp) +
454  coeff_Q * slow_rhs_cons(i,j,k-1,RhoTheta_comp) );
455 
456  Real Omega_kp1 = omega_arr(i,j,k+1);
457  Real Omega_k = omega_arr(i,j,k );
458  Real Omega_km1 = omega_arr(i,j,k-1);
459 
460  // consolidate lines 4&5 (order dtau^2)
461  R1_tmp += ( halfg ) *
462  ( beta_1 * dzi * (Omega_kp1 - Omega_km1) + temp_rhs_arr(i,j,k,Rho_comp) + temp_rhs_arr(i,j,k-1,Rho_comp));
463 
464  // consolidate lines 6&7 (order dtau^2)
465  R1_tmp += -(
466  coeff_P * ( beta_1 * dzi * (Omega_kp1*theta_t_hi - Omega_k*theta_t_mid) + temp_rhs_arr(i,j,k ,RhoTheta_comp) ) +
467  coeff_Q * ( beta_1 * dzi * (Omega_k*theta_t_mid - Omega_km1*theta_t_lo) + temp_rhs_arr(i,j,k-1,RhoTheta_comp) ) );
468 
469  // line 1
470  RHS_a(i,j,k) = dJ_old_kface * prev_zmom(i,j,k) - dJ_stg_kface * stg_zmom(i,j,k)
471  + dtau * (slow_rhs_rho_w(i,j,k) + R0_tmp + dtau*beta_2*R1_tmp + zmom_src_arr(i,j,k));
472 
473  // We cannot use omega_arr here since that was built with old_rho_u and old_rho_v ...
474  Real UppVpp = dJ_new_kface * OmegaFromW(i,j,k,0.,cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv)
475  -dJ_stg_kface * OmegaFromW(i,j,k,0.,stg_xmom,stg_ymom,mf_ux,mf_vy,z_nd_stg,dxInv);
476  RHS_a(i,j,k) += UppVpp;
477  });
478  } // end profile
479 
480  Box b2d = tbz; // Copy constructor
481  b2d.setRange(2,0);
482 
483  auto const lo = lbound(bx);
484  auto const hi = ubound(bx);
485 
486  {
487  BL_PROFILE("fast_rhs_b2d_loop_t");
488 
489 #ifdef AMREX_USE_GPU
490  ParallelFor(b2d, [=] AMREX_GPU_DEVICE (int i, int j, int)
491  {
492  // Moving terrain
493  Real rho_on_bdy = 0.5 * ( prev_cons(i,j,lo.z) + prev_cons(i,j,lo.z-1) );
494  RHS_a(i,j,lo.z) = rho_on_bdy * zp_t_arr(i,j,0);
495 
496  soln_a(i,j,lo.z) = RHS_a(i,j,lo.z) * inv_coeffB_a(i,j,lo.z);
497 
498  // w_khi = 0
499  RHS_a(i,j,hi.z+1) = 0.0;
500 
501  for (int k = lo.z+1; k <= hi.z+1; k++) {
502  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);
503  }
504 
505  for (int k = hi.z; k >= lo.z; k--) {
506  soln_a(i,j,k) -= ( coeffC_a(i,j,k) * inv_coeffB_a(i,j,k) ) * soln_a(i,j,k+1);
507  }
508 
509  // We assume that Omega == w at the top boundary and that changes in J there are irrelevant
510  cur_zmom(i,j,hi.z+1) = stg_zmom(i,j,hi.z+1) + soln_a(i,j,hi.z+1);
511  });
512 #else
513  for (int j = lo.y; j <= hi.y; ++j) {
514  AMREX_PRAGMA_SIMD
515  for (int i = lo.x; i <= hi.x; ++i) {
516 
517  Real rho_on_bdy = 0.5 * ( prev_cons(i,j,lo.z) + prev_cons(i,j,lo.z-1) );
518  RHS_a(i,j,lo.z) = rho_on_bdy * zp_t_arr(i,j,lo.z);
519 
520  soln_a(i,j,lo.z) = RHS_a(i,j,lo.z) * inv_coeffB_a(i,j,lo.z);
521  }
522  }
523 
524  for (int j = lo.y; j <= hi.y; ++j) {
525  AMREX_PRAGMA_SIMD
526  for (int i = lo.x; i <= hi.x; ++i) {
527  RHS_a (i,j,hi.z+1) = 0.0;
528  }
529  }
530  for (int k = lo.z+1; k <= hi.z+1; ++k) {
531  for (int j = lo.y; j <= hi.y; ++j) {
532  AMREX_PRAGMA_SIMD
533  for (int i = lo.x; i <= hi.x; ++i) {
534  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);
535  }
536  }
537  }
538  for (int k = hi.z; k >= lo.z; --k) {
539  for (int j = lo.y; j <= hi.y; ++j) {
540  AMREX_PRAGMA_SIMD
541  for (int i = lo.x; i <= hi.x; ++i) {
542  soln_a(i,j,k) -= ( coeffC_a(i,j,k) * inv_coeffB_a(i,j,k) ) * soln_a(i,j,k+1);
543  }
544  }
545  }
546 
547  // We assume that Omega == w at the top boundary and that changes in J there are irrelevant
548  for (int j = lo.y; j <= hi.y; ++j) {
549  AMREX_PRAGMA_SIMD
550  for (int i = lo.x; i <= hi.x; ++i) {
551  cur_zmom(i,j,hi.z+1) = stg_zmom(i,j,hi.z+1) + soln_a(i,j,hi.z+1);
552  }
553  }
554 #endif
555  } // end profile
556 
557  {
558  BL_PROFILE("fast_rhs_new_drhow");
559  tbz.setBig(2,hi.z);
560  ParallelFor(tbz, [=] AMREX_GPU_DEVICE (int i, int j, int k)
561  {
562  Real rho_on_face = 0.5 * (cur_cons(i,j,k,Rho_comp) + cur_cons(i,j,k-1,Rho_comp));
563 
564  if (k == lo.z) {
565  cur_zmom(i,j,k) = WFromOmega(i,j,k,rho_on_face*(z_t_arr(i,j,k)+zp_t_arr(i,j,k)),
566  cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv);
567 
568  // We need to set this here because it is used to define zflux_lo below
569  soln_a(i,j,k) = 0.;
570 
571  } else {
572 
573  Real UppVpp = WFromOmega(i,j,k,0.0,cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv)
574  - WFromOmega(i,j,k,0.0,stg_xmom,stg_ymom,mf_ux,mf_vy,z_nd_stg,dxInv);
575  Real wpp = soln_a(i,j,k) + UppVpp;
576  Real dJ_old_kface = 0.5 * (detJ_old(i,j,k) + detJ_old(i,j,k-1));
577  Real dJ_new_kface = 0.5 * (detJ_new(i,j,k) + detJ_new(i,j,k-1));
578 
579  cur_zmom(i,j,k) = dJ_old_kface * (stg_zmom(i,j,k) + wpp);
580  cur_zmom(i,j,k) /= dJ_new_kface;
581 
582  soln_a(i,j,k) = OmegaFromW(i,j,k,cur_zmom(i,j,k),cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv)
583  - OmegaFromW(i,j,k,stg_zmom(i,j,k),stg_xmom,stg_ymom,mf_ux,mf_vy,z_nd_stg,dxInv);
584  soln_a(i,j,k) -= rho_on_face * zp_t_arr(i,j,k);
585  }
586  });
587  } // end profile
588 
589  // **************************************************************************
590  // Define updates in the RHS of rho and (rho theta)
591  // **************************************************************************
592  {
593  BL_PROFILE("fast_rho_final_update");
594  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
595  {
596  Real zflux_lo = beta_2 * soln_a(i,j,k ) + beta_1 * omega_arr(i,j,k);
597  Real zflux_hi = beta_2 * soln_a(i,j,k+1) + beta_1 * omega_arr(i,j,k+1);
598 
599  // Note that in the solve we effectively impose new_drho_w(i,j,vbx_hi.z+1)=0
600  // so we don't update avg_zmom at k=vbx_hi.z+1
601  avg_zmom_arr(i,j,k) += facinv*zflux_lo / (mf_mx(i,j,0) * mf_my(i,j,0));
602  if (l_reflux) {
603  (flx_arr[2])(i,j,k,0) = zflux_lo / (mf_mx(i,j,0) * mf_my(i,j,0));
604  }
605 
606  // Note that the factor of (1/J) in the fast source term is canceled
607  // when we multiply old and new by detJ_old and detJ_new , respectively
608  // We have already scaled the slow source term to have the extra factor of dJ
609  Real fast_rhs_rho = -(temp_rhs_arr(i,j,k,0) + ( zflux_hi - zflux_lo ) * dzi);
610  Real temp_rho = detJ_old(i,j,k) * cur_cons(i,j,k,0) +
611  dtau * ( slow_rhs_cons(i,j,k,0) + fast_rhs_rho );
612  cur_cons(i,j,k,0) = temp_rho / detJ_new(i,j,k);
613 
614  // Note that the factor of (1/J) in the fast source term is canceled
615  // when we multiply old and new by detJ_old and detJ_new , respectively
616  // We have already scaled the slow source term to have the extra factor of dJ
617  Real fast_rhs_rhotheta = -( temp_rhs_arr(i,j,k,1) + 0.5 *
618  ( zflux_hi * (prim(i,j,k) + prim(i,j,k+1))
619  - zflux_lo * (prim(i,j,k) + prim(i,j,k-1)) ) * dzi );
620  Real temp_rth = detJ_old(i,j,k) * cur_cons(i,j,k,1) +
621  dtau * ( slow_rhs_cons(i,j,k,1) + fast_rhs_rhotheta );
622  cur_cons(i,j,k,1) = temp_rth / detJ_new(i,j,k);
623  if (l_reflux) {
624  (flx_arr[2])(i,j,k,1) = (flx_arr[2])(i,j,k,0) * 0.5 * (prim(i,j,k) + prim(i,j,k-1));
625  }
626 
627  if (k == vbx_hi.z) {
628  avg_zmom_arr(i,j,k+1) += facinv * zflux_hi / (mf_mx(i,j,0) * mf_my(i,j,0));
629  if (l_reflux) {
630  (flx_arr[2])(i,j,k+1,0) = zflux_hi / (mf_mx(i,j,0) * mf_my(i,j,0));
631  (flx_arr[2])(i,j,k+1,1) = (flx_arr[2])(i,j,k+1,0) * 0.5 * (prim(i,j,k) + prim(i,j,k+1));
632  }
633  }
634 
635  // add in source terms for cell-centered conserved variables
636  cur_cons(i,j,k,Rho_comp) += dtau * cc_src_arr(i,j,k,Rho_comp);
637  cur_cons(i,j,k,RhoTheta_comp) += dtau * cc_src_arr(i,j,k,RhoTheta_comp);
638  });
639  } // end profile
640 
641  // We only add to the flux registers in the final RK step
642  if (l_reflux) {
643  int strt_comp_reflux = 0;
644  int num_comp_reflux = 2;
645  if (level < finest_level) {
646  fr_as_crse->CrseAdd(mfi,
647  {{AMREX_D_DECL(&(flux[0]), &(flux[1]), &(flux[2]))}},
648  dx, dtau, strt_comp_reflux, strt_comp_reflux, num_comp_reflux, RunOn::Device);
649  }
650  if (level > 0) {
651  fr_as_fine->FineAdd(mfi,
652  {{AMREX_D_DECL(&(flux[0]), &(flux[1]), &(flux[2]))}},
653  dx, dtau, strt_comp_reflux, strt_comp_reflux, num_comp_reflux, RunOn::Device);
654  }
655 
656  // This is necessary here so we don't go on to the next FArrayBox without
657  // having finished copying the fluxes into the FluxRegisters (since the fluxes
658  // are stored in temporary FArrayBox's)
659  Gpu::streamSynchronize();
660 
661  } // two-way coupling
662 
663  } // mfi
664  } // OMP
665 }
R_v
constexpr amrex::Real R_v
Definition: ERF_Constants.H:11
R_d
constexpr amrex::Real R_d
Definition: ERF_Constants.H:10
Gamma
constexpr amrex::Real Gamma
Definition: ERF_Constants.H:19
v_y
@ v_y
Definition: ERF_DataStruct.H:23
m_y
@ m_y
Definition: ERF_DataStruct.H:23
u_x
@ u_x
Definition: ERF_DataStruct.H:22
m_x
@ m_x
Definition: ERF_DataStruct.H:22
PrimQ1_comp
#define PrimQ1_comp
Definition: ERF_IndexDefines.H:53
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:50
Real
amrex::Real Real
Definition: ERF_ShocInterface.H:19
OmegaFromW
AMREX_GPU_DEVICE AMREX_FORCE_INLINE amrex::Real OmegaFromW(int &i, int &j, int &k, amrex::Real w, const amrex::Array4< const amrex::Real > &u_arr, const amrex::Array4< const amrex::Real > &v_arr, const amrex::Array4< const amrex::Real > &mf_u, const amrex::Array4< const amrex::Real > &mf_v, const amrex::Array4< const amrex::Real > &z_nd, const amrex::GpuArray< amrex::Real, AMREX_SPACEDIM > &dxInv)
Definition: ERF_TerrainMetrics.H:406
Compute_h_xi_AtIface
AMREX_GPU_DEVICE AMREX_FORCE_INLINE amrex::Real Compute_h_xi_AtIface(const int &i, const int &j, const int &k, const amrex::GpuArray< amrex::Real, AMREX_SPACEDIM > &cellSizeInv, const amrex::Array4< const amrex::Real > &z_nd)
Definition: ERF_TerrainMetrics.H:109
Compute_h_zeta_AtIface
AMREX_GPU_DEVICE AMREX_FORCE_INLINE amrex::Real Compute_h_zeta_AtIface(const int &i, const int &j, const int &k, const amrex::GpuArray< amrex::Real, AMREX_SPACEDIM > &cellSizeInv, const amrex::Array4< const amrex::Real > &z_nd)
Definition: ERF_TerrainMetrics.H:96
Compute_h_zeta_AtJface
AMREX_GPU_DEVICE AMREX_FORCE_INLINE amrex::Real Compute_h_zeta_AtJface(const int &i, const int &j, const int &k, const amrex::GpuArray< amrex::Real, AMREX_SPACEDIM > &cellSizeInv, const amrex::Array4< const amrex::Real > &z_nd)
Definition: ERF_TerrainMetrics.H:136
Compute_h_eta_AtJface
AMREX_GPU_DEVICE AMREX_FORCE_INLINE amrex::Real Compute_h_eta_AtJface(const int &i, const int &j, const int &k, const amrex::GpuArray< amrex::Real, AMREX_SPACEDIM > &cellSizeInv, const amrex::Array4< const amrex::Real > &z_nd)
Definition: ERF_TerrainMetrics.H:162
WFromOmega
AMREX_GPU_DEVICE AMREX_FORCE_INLINE amrex::Real WFromOmega(int &i, int &j, int &k, amrex::Real omega, const amrex::Array4< const amrex::Real > &u_arr, const amrex::Array4< const amrex::Real > &v_arr, const amrex::Array4< const amrex::Real > &mf_u, const amrex::Array4< const amrex::Real > &mf_v, const amrex::Array4< const amrex::Real > &z_nd, const amrex::GpuArray< amrex::Real, AMREX_SPACEDIM > &dxInv)
Definition: ERF_TerrainMetrics.H:456
GpVars::gpy
@ gpy
Definition: ERF_IndexDefines.H:151
GpVars::gpx
@ gpx
Definition: ERF_IndexDefines.H:150
IntVars::ymom
@ ymom
Definition: ERF_IndexDefines.H:160
IntVars::cons
@ cons
Definition: ERF_IndexDefines.H:158
IntVars::zmom
@ zmom
Definition: ERF_IndexDefines.H:161
IntVars::xmom
@ xmom
Definition: ERF_IndexDefines.H:159
MicVar_Kess::qt
@ qt
Definition: ERF_Kessler.H:27
MicVar_Kess::qv
@ qv
Definition: ERF_Kessler.H:28
Here is the call graph for this function: