ERF
Energy Research and Forecasting: An Atmospheric Modeling Code
•All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Pages
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, Vector< MultiFab > &S_scratch, 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,
Vector< MultiFab > &  S_scratch,
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]S_scratchscratch space
[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
81 {
82  BL_PROFILE_REGION("erf_fast_rhs_MT()");
83 
84  Real beta_1 = 0.5 * (1.0 - beta_s); // multiplies explicit terms
85  Real beta_2 = 0.5 * (1.0 + beta_s); // multiplies implicit terms
86 
87  // How much do we project forward the (rho theta) that is used in the horizontal momentum equations
88  Real beta_d = 0.1;
89 
90  const Real* dx = geom.CellSize();
91  const GpuArray<Real, AMREX_SPACEDIM> dxInv = geom.InvCellSizeArray();
92 
93  Real dxi = dxInv[0];
94  Real dyi = dxInv[1];
95  Real dzi = dxInv[2];
96 
97  MultiFab coeff_A_mf(fast_coeffs, make_alias, 0, 1);
98  MultiFab inv_coeff_B_mf(fast_coeffs, make_alias, 1, 1);
99  MultiFab coeff_C_mf(fast_coeffs, make_alias, 2, 1);
100  MultiFab coeff_P_mf(fast_coeffs, make_alias, 3, 1);
101  MultiFab coeff_Q_mf(fast_coeffs, make_alias, 4, 1);
102 
103  // *************************************************************************
104  // Set gravity as a vector
105  const Array<Real,AMREX_SPACEDIM> grav{0.0, 0.0, -gravity};
106  const GpuArray<Real,AMREX_SPACEDIM> grav_gpu{grav[0], grav[1], grav[2]};
107 
108  MultiFab extrap(S_data[IntVars::cons].boxArray(),S_data[IntVars::cons].DistributionMap(),1,1);
109 
110  MultiFab Omega(S_data[IntVars::zmom].boxArray(), S_data[IntVars::zmom].DistributionMap(), 1, 1);
111 
112  // *************************************************************************
113  // Define updates in the current RK stg
114  // *************************************************************************
115 #ifdef _OPENMP
116 #pragma omp parallel if (Gpu::notInLaunchRegion())
117 #endif
118  {
119  FArrayBox temp_rhs_fab;
120 
121  FArrayBox RHS_fab;
122  FArrayBox soln_fab;
123 
124  std::array<FArrayBox,AMREX_SPACEDIM> flux;
125 
126  // NOTE: we leave tiling off here for efficiency -- to make this loop work with tiling
127  // will require additional changes
128  for ( MFIter mfi(S_stg_data[IntVars::cons],false); mfi.isValid(); ++mfi)
129  {
130  Box bx = mfi.tilebox();
131  Box tbx = surroundingNodes(bx,0);
132  Box tby = surroundingNodes(bx,1);
133  Box tbz = surroundingNodes(bx,2);
134 
135  Box vbx = mfi.validbox();
136  const auto& vbx_hi = ubound(vbx);
137 
138  const Array4<Real const>& xmom_src_arr = xmom_src.const_array(mfi);
139  const Array4<Real const>& ymom_src_arr = ymom_src.const_array(mfi);
140  const Array4<Real const>& zmom_src_arr = zmom_src.const_array(mfi);
141  const Array4<Real const>& cc_src_arr = cc_src.const_array(mfi);
142 
143  const Array4<const Real> & stg_cons = S_stg_data[IntVars::cons].const_array(mfi);
144  const Array4<const Real> & stg_xmom = S_stg_data[IntVars::xmom].const_array(mfi);
145  const Array4<const Real> & stg_ymom = S_stg_data[IntVars::ymom].const_array(mfi);
146  const Array4<const Real> & stg_zmom = S_stg_data[IntVars::zmom].const_array(mfi);
147  const Array4<const Real> & prim = S_stg_prim.const_array(mfi);
148  const Array4<const Real> & qt_arr = qt.const_array(mfi);
149 
150  const Array4<const Real>& slow_rhs_cons = S_slow_rhs[IntVars::cons].const_array(mfi);
151  const Array4<const Real>& slow_rhs_rho_u = S_slow_rhs[IntVars::xmom].const_array(mfi);
152  const Array4<const Real>& slow_rhs_rho_v = S_slow_rhs[IntVars::ymom].const_array(mfi);
153  const Array4<const Real>& slow_rhs_rho_w = S_slow_rhs[IntVars::zmom].const_array(mfi);
154 
155  const Array4<Real>& cur_cons = S_data[IntVars::cons].array(mfi);
156  const Array4<Real>& cur_xmom = S_data[IntVars::xmom].array(mfi);
157  const Array4<Real>& cur_ymom = S_data[IntVars::ymom].array(mfi);
158  const Array4<Real>& cur_zmom = S_data[IntVars::zmom].array(mfi);
159 
160  const Array4<Real>& lagged_delta_rt = S_scratch[IntVars::cons].array(mfi);
161 
162  const Array4<const Real>& prev_cons = S_prev[IntVars::cons].const_array(mfi);
163  const Array4<const Real>& prev_xmom = S_prev[IntVars::xmom].const_array(mfi);
164  const Array4<const Real>& prev_ymom = S_prev[IntVars::ymom].const_array(mfi);
165  const Array4<const Real>& prev_zmom = S_prev[IntVars::zmom].const_array(mfi);
166 
167  // These store the advection momenta which we will use to update the slow variables
168  const Array4<Real>& avg_xmom = S_scratch[IntVars::xmom].array(mfi);
169  const Array4<Real>& avg_ymom = S_scratch[IntVars::ymom].array(mfi);
170  const Array4<Real>& avg_zmom = S_scratch[IntVars::zmom].array(mfi);
171 
172  const Array4<const Real>& z_nd_old = z_phys_nd_old->const_array(mfi);
173  const Array4<const Real>& z_nd_new = z_phys_nd_new->const_array(mfi);
174  const Array4<const Real>& z_nd_stg = z_phys_nd_stg->const_array(mfi);
175  const Array4<const Real>& detJ_old = detJ_cc_old->const_array(mfi);
176  const Array4<const Real>& detJ_new = detJ_cc_new->const_array(mfi);
177  const Array4<const Real>& detJ_stg = detJ_cc_stg->const_array(mfi);
178 
179  const Array4<const Real>& z_t_arr = z_t_rk->const_array(mfi);
180  const Array4<const Real>& zp_t_arr = z_t_pert->const_array(mfi);
181 
182  const Array4< Real>& omega_arr = Omega.array(mfi);
183 
184  const Array4<const Real>& pi_stage_ca = pi_stage.const_array(mfi);
185 
186  const Array4<Real>& theta_extrap = extrap.array(mfi);
187 
188  // Map factors
189  const Array4<const Real>& mf_mx = mapfac[MapFacType::m_x]->const_array(mfi);
190  const Array4<const Real>& mf_my = mapfac[MapFacType::m_y]->const_array(mfi);
191  const Array4<const Real>& mf_ux = mapfac[MapFacType::u_x]->const_array(mfi);
192  const Array4<const Real>& mf_vy = mapfac[MapFacType::v_y]->const_array(mfi);
193 
194  // Note: it is important to grow the tilebox rather than use growntilebox because
195  // we need to fill the ghost cells of the tilebox so we can use them below
196  Box gbx = mfi.tilebox(); gbx.grow(1);
197  Box gtbx = mfi.nodaltilebox(0); gtbx.grow(1); gtbx.setSmall(2,0);
198  Box gtby = mfi.nodaltilebox(1); gtby.grow(1); gtby.setSmall(2,0);
199 
200  {
201  BL_PROFILE("fast_rhs_copies_0");
202  if (step == 0) {
203  ParallelFor(gbx,
204  [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
205  cur_cons(i,j,k,Rho_comp) = prev_cons(i,j,k,Rho_comp);
206  cur_cons(i,j,k,RhoTheta_comp) = prev_cons(i,j,k,RhoTheta_comp);
207 
208  Real delta_rt = cur_cons(i,j,k,RhoTheta_comp) - stg_cons(i,j,k,RhoTheta_comp);
209  theta_extrap(i,j,k) = delta_rt;
210 
211  // We define lagged_delta_rt for our next step as the current delta_rt
212  lagged_delta_rt(i,j,k,RhoTheta_comp) = delta_rt;
213  });
214  } else if (use_lagged_delta_rt) {
215  // This is the default for cases with no or static terrain
216  ParallelFor(gbx,
217  [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
218  Real delta_rt = cur_cons(i,j,k,RhoTheta_comp) - stg_cons(i,j,k,RhoTheta_comp);
219  theta_extrap(i,j,k) = delta_rt + beta_d * (delta_rt - lagged_delta_rt(i,j,k,RhoTheta_comp));
220 
221  // We define lagged_delta_rt for our next step as the current delta_rt
222  lagged_delta_rt(i,j,k,RhoTheta_comp) = delta_rt;
223  });
224  } else {
225  // For the moving wave problem, this choice seems more robust
226  ParallelFor(gbx,
227  [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
228  theta_extrap(i,j,k) = cur_cons(i,j,k,RhoTheta_comp) - stg_cons(i,j,k,RhoTheta_comp);
229  });
230  } // if step
231  } // end profile
232 
233  RHS_fab.resize (tbz,1, The_Async_Arena());
234  soln_fab.resize (tbz,1, The_Async_Arena());
235  temp_rhs_fab.resize(tbz,2, The_Async_Arena());
236 
237  auto const& RHS_a = RHS_fab.array();
238  auto const& soln_a = soln_fab.array();
239  auto const& temp_rhs_arr = temp_rhs_fab.array();
240 
241  auto const& coeffA_a = coeff_A_mf.array(mfi);
242  auto const& inv_coeffB_a = inv_coeff_B_mf.array(mfi);
243  auto const& coeffC_a = coeff_C_mf.array(mfi);
244  auto const& coeffP_a = coeff_P_mf.array(mfi);
245  auto const& coeffQ_a = coeff_Q_mf.array(mfi);
246 
247  // *********************************************************************
248  // Define updates in the RHS of {x, y, z}-momentum equations
249  // *********************************************************************
250  {
251  BL_PROFILE("fast_rhs_xymom_T");
252  ParallelFor(tbx, tby,
253  [=] AMREX_GPU_DEVICE (int i, int j, int k)
254  {
255  // Add (negative) gradient of (rho theta) multiplied by lagged "pi"
256  Real h_xi_old = Compute_h_xi_AtIface(i, j, k, dxInv, z_nd_old);
257  Real h_zeta_old = Compute_h_zeta_AtIface(i, j, k, dxInv, z_nd_old);
258  Real gp_xi = (theta_extrap(i,j,k) - theta_extrap(i-1,j,k)) * dxi;
259  Real gp_zeta_on_iface = (k == 0) ?
260  0.5 * dzi * ( theta_extrap(i-1,j,k+1) + theta_extrap(i,j,k+1)
261  -theta_extrap(i-1,j,k ) - theta_extrap(i,j,k ) ) :
262  0.25 * dzi * ( theta_extrap(i-1,j,k+1) + theta_extrap(i,j,k+1)
263  -theta_extrap(i-1,j,k-1) - theta_extrap(i,j,k-1) );
264  Real gpx = h_zeta_old * gp_xi - h_xi_old * gp_zeta_on_iface;
265  gpx *= mf_ux(i,j,0);
266 
267  Real q = (l_use_moisture) ? 0.5 * (qt_arr(i-1,j,k) + qt_arr(i,j,k)) : 0.0;
268 
269  Real pi_c = 0.5 * (pi_stage_ca(i-1,j,k,0) + pi_stage_ca(i ,j,k,0));
270  Real fast_rhs_rho_u = -Gamma * R_d * pi_c * gpx / (1.0 + q);
271 
272  // We have already scaled the source terms to have the extra factor of dJ
273  cur_xmom(i,j,k) = h_zeta_old * prev_xmom(i,j,k) + dtau * fast_rhs_rho_u
274  + dtau * slow_rhs_rho_u(i,j,k)
275  + dtau * xmom_src_arr(i,j,k);
276  },
277  [=] AMREX_GPU_DEVICE (int i, int j, int k)
278  {
279  // Add (negative) gradient of (rho theta) multiplied by lagged "pi"
280  Real h_eta_old = Compute_h_eta_AtJface(i, j, k, dxInv, z_nd_old);
281  Real h_zeta_old = Compute_h_zeta_AtJface(i, j, k, dxInv, z_nd_old);
282  Real gp_eta = (theta_extrap(i,j,k) -theta_extrap(i,j-1,k)) * dyi;
283  Real gp_zeta_on_jface = (k == 0) ?
284  0.5 * dzi * ( theta_extrap(i,j,k+1) + theta_extrap(i,j-1,k+1)
285  -theta_extrap(i,j,k ) - theta_extrap(i,j-1,k ) ) :
286  0.25 * dzi * ( theta_extrap(i,j,k+1) + theta_extrap(i,j-1,k+1)
287  -theta_extrap(i,j,k-1) - theta_extrap(i,j-1,k-1) );
288  Real gpy = h_zeta_old * gp_eta - h_eta_old * gp_zeta_on_jface;
289  gpy *= mf_vy(i,j,0);
290 
291  Real q = (l_use_moisture) ? 0.5 * (qt_arr(i,j-1,k) + qt_arr(i,j,k)) : 0.0;
292 
293  Real pi_c = 0.5 * (pi_stage_ca(i,j-1,k,0) + pi_stage_ca(i,j ,k,0));
294  Real fast_rhs_rho_v = -Gamma * R_d * pi_c * gpy / (1.0 + q);
295 
296  // We have already scaled the source terms to have the extra factor of dJ
297  cur_ymom(i, j, k) = h_zeta_old * prev_ymom(i,j,k) + dtau * fast_rhs_rho_v
298  + dtau * slow_rhs_rho_v(i,j,k)
299  + dtau * ymom_src_arr(i,j,k);
300  });
301  } // end profile
302 
303  // *************************************************************************
304  // Define flux arrays for use in advection
305  // *************************************************************************
306  for (int dir = 0; dir < AMREX_SPACEDIM; ++dir) {
307  flux[dir].resize(surroundingNodes(bx,dir),2);
308  flux[dir].setVal<RunOn::Device>(0.);
309  }
310  const GpuArray<const Array4<Real>, AMREX_SPACEDIM>
311  flx_arr{{AMREX_D_DECL(flux[0].array(), flux[1].array(), flux[2].array())}};
312 
313  // *********************************************************************
314  {
315  BL_PROFILE("fast_T_making_rho_rhs");
316  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
317  {
318  Real h_zeta_stg_xlo = Compute_h_zeta_AtIface(i, j , k, dxInv, z_nd_stg);
319  Real h_zeta_stg_xhi = Compute_h_zeta_AtIface(i+1,j , k, dxInv, z_nd_stg);
320  Real xflux_lo = cur_xmom(i ,j,k) - stg_xmom(i ,j,k)*h_zeta_stg_xlo;
321  Real xflux_hi = cur_xmom(i+1,j,k) - stg_xmom(i+1,j,k)*h_zeta_stg_xhi;
322 
323  Real h_zeta_stg_yhi = Compute_h_zeta_AtJface(i, j+1, k, dxInv, z_nd_stg);
324  Real h_zeta_stg_ylo = Compute_h_zeta_AtJface(i, j , k, dxInv, z_nd_stg);
325  Real yflux_lo = cur_ymom(i,j ,k) - stg_ymom(i,j ,k)*h_zeta_stg_ylo;
326  Real yflux_hi = cur_ymom(i,j+1,k) - stg_ymom(i,j+1,k)*h_zeta_stg_yhi;
327 
328  // NOTE: we are saving the (1/J) weighting for later when we add this to rho and theta
329  temp_rhs_arr(i,j,k,0) = ( xflux_hi - xflux_lo ) * dxi + ( yflux_hi - yflux_lo ) * dyi;
330  temp_rhs_arr(i,j,k,1) = (( xflux_hi * (prim(i,j,k,0) + prim(i+1,j,k,0)) -
331  xflux_lo * (prim(i,j,k,0) + prim(i-1,j,k,0)) ) * dxi +
332  ( yflux_hi * (prim(i,j,k,0) + prim(i,j+1,k,0)) -
333  yflux_lo * (prim(i,j,k,0) + prim(i,j-1,k,0)) ) * dyi) * 0.5;
334 
335  if (l_reflux) {
336  (flx_arr[0])(i,j,k,0) = xflux_lo;
337  (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));
338 
339  (flx_arr[1])(i,j,k,0) = yflux_lo;
340  (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));
341 
342  if (i == vbx_hi.x) {
343  (flx_arr[0])(i+1,j,k,0) = xflux_hi;
344  (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));
345  }
346  if (j == vbx_hi.y) {
347  (flx_arr[1])(i,j+1,k,0) = yflux_hi;
348  (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));
349  }
350  }
351  });
352  } // end profile
353 
354  // *********************************************************************
355  // This must be done before we set cur_xmom and cur_ymom, since those
356  // in fact point to the same array as prev_xmom and prev_ymom
357  // *********************************************************************
358  Box gbxo = mfi.nodaltilebox(2);
359  {
360  BL_PROFILE("fast_MT_making_omega");
361  Box gbxo_lo = gbxo; gbxo_lo.setBig(2,0);
362  ParallelFor(gbxo_lo, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
363  omega_arr(i,j,k) = 0.;
364  });
365  Box gbxo_hi = gbxo; gbxo_hi.setSmall(2,gbxo.bigEnd(2));
366  ParallelFor(gbxo_hi, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
367  omega_arr(i,j,k) = prev_zmom(i,j,k) - stg_zmom(i,j,k) - zp_t_arr(i,j,k);
368  });
369  Box gbxo_mid = gbxo; gbxo_mid.setSmall(2,1); gbxo_mid.setBig(2,gbxo.bigEnd(2)-1);
370  ParallelFor(gbxo_mid, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
371  omega_arr(i,j,k) =
372  ( OmegaFromW(i,j,k,prev_zmom(i,j,k),prev_xmom,prev_ymom,mf_ux,mf_vy,z_nd_old,dxInv)
373  -OmegaFromW(i,j,k, stg_zmom(i,j,k), stg_xmom, stg_ymom,mf_ux,mf_vy,z_nd_old,dxInv) )
374  - zp_t_arr(i,j,k);
375  });
376  } // end profile
377  // *********************************************************************
378 
379  ParallelFor(tbx, tby,
380  [=] AMREX_GPU_DEVICE (int i, int j, int k)
381  {
382  Real h_zeta_new = Compute_h_zeta_AtIface(i, j, k, dxInv, z_nd_new);
383  cur_xmom(i, j, k) /= h_zeta_new;
384  avg_xmom(i,j,k) += facinv*(cur_xmom(i,j,k) - stg_xmom(i,j,k));
385  },
386  [=] AMREX_GPU_DEVICE (int i, int j, int k)
387  {
388  Real h_zeta_new = Compute_h_zeta_AtJface(i, j, k, dxInv, z_nd_new);
389  cur_ymom(i, j, k) /= h_zeta_new;
390  avg_ymom(i,j,k) += facinv*(cur_ymom(i,j,k) - stg_ymom(i,j,k));
391  });
392 
393  Box bx_shrunk_in_k = bx;
394  int klo = tbz.smallEnd(2);
395  int khi = tbz.bigEnd(2);
396  bx_shrunk_in_k.setSmall(2,klo+1);
397  bx_shrunk_in_k.setBig(2,khi-1);
398 
399  // Note that the notes use "g" to mean the magnitude of gravity, so it is positive
400  // We set grav_gpu[2] to be the vector component which is negative
401  // We define halfg to match the notes (which is why we take the absolute value)
402  Real halfg = std::abs(0.5 * grav_gpu[2]);
403 
404  {
405  BL_PROFILE("fast_loop_on_shrunk_t");
406  //Note we don't act on the bottom or top boundaries of the domain
407  ParallelFor(bx_shrunk_in_k, [=] AMREX_GPU_DEVICE (int i, int j, int k)
408  {
409  Real dJ_old_kface = 0.5 * (detJ_old(i,j,k) + detJ_old(i,j,k-1));
410  Real dJ_new_kface = 0.5 * (detJ_new(i,j,k) + detJ_new(i,j,k-1));
411  Real dJ_stg_kface = 0.5 * (detJ_stg(i,j,k) + detJ_stg(i,j,k-1));
412 
413  Real q = (l_use_moisture) ? 0.5 * (qt_arr(i,j,k-1) + qt_arr(i,j,k)) : 0.0;
414 
415  Real coeff_P = coeffP_a(i,j,k) / (1.0 + q);
416  Real coeff_Q = coeffQ_a(i,j,k) / (1.0 + q);
417 
418  Real theta_t_lo = 0.5 * ( prim(i,j,k-2,PrimTheta_comp) + prim(i,j,k-1,PrimTheta_comp) );
419  Real theta_t_mid = 0.5 * ( prim(i,j,k-1,PrimTheta_comp) + prim(i,j,k ,PrimTheta_comp) );
420  Real theta_t_hi = 0.5 * ( prim(i,j,k ,PrimTheta_comp) + prim(i,j,k+1,PrimTheta_comp) );
421 
422  // line 2 last two terms (order dtau)
423  Real R0_tmp = coeff_P * cur_cons(i,j,k ,RhoTheta_comp) * dJ_old_kface
424  + coeff_Q * cur_cons(i,j,k-1,RhoTheta_comp) * dJ_old_kface
425  - coeff_P * stg_cons(i,j,k ,RhoTheta_comp) * dJ_stg_kface
426  - coeff_Q * stg_cons(i,j,k-1,RhoTheta_comp) * dJ_stg_kface
427  - halfg * ( cur_cons(i,j,k,Rho_comp) + cur_cons(i,j,k-1,Rho_comp) ) * dJ_old_kface
428  + halfg * ( stg_cons(i,j,k,Rho_comp) + stg_cons(i,j,k-1,Rho_comp) ) * dJ_stg_kface;
429 
430  // line 3 residuals (order dtau^2) 1.0 <-> beta_2
431  Real R1_tmp = - halfg * ( slow_rhs_cons(i,j,k ,Rho_comp) +
432  slow_rhs_cons(i,j,k-1,Rho_comp) )
433  + ( coeff_P * slow_rhs_cons(i,j,k ,RhoTheta_comp) +
434  coeff_Q * slow_rhs_cons(i,j,k-1,RhoTheta_comp) );
435 
436  Real Omega_kp1 = omega_arr(i,j,k+1);
437  Real Omega_k = omega_arr(i,j,k );
438  Real Omega_km1 = omega_arr(i,j,k-1);
439 
440  // consolidate lines 4&5 (order dtau^2)
441  R1_tmp += ( halfg ) *
442  ( beta_1 * dzi * (Omega_kp1 - Omega_km1) + temp_rhs_arr(i,j,k,Rho_comp) + temp_rhs_arr(i,j,k-1,Rho_comp));
443 
444  // consolidate lines 6&7 (order dtau^2)
445  R1_tmp += -(
446  coeff_P * ( beta_1 * dzi * (Omega_kp1*theta_t_hi - Omega_k*theta_t_mid) + temp_rhs_arr(i,j,k ,RhoTheta_comp) ) +
447  coeff_Q * ( beta_1 * dzi * (Omega_k*theta_t_mid - Omega_km1*theta_t_lo) + temp_rhs_arr(i,j,k-1,RhoTheta_comp) ) );
448 
449  // line 1
450  RHS_a(i,j,k) = dJ_old_kface * prev_zmom(i,j,k) - dJ_stg_kface * stg_zmom(i,j,k)
451  + dtau * (slow_rhs_rho_w(i,j,k) + R0_tmp + dtau*beta_2*R1_tmp + zmom_src_arr(i,j,k));
452 
453  // We cannot use omega_arr here since that was built with old_rho_u and old_rho_v ...
454  Real UppVpp = dJ_new_kface * OmegaFromW(i,j,k,0.,cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv)
455  -dJ_stg_kface * OmegaFromW(i,j,k,0.,stg_xmom,stg_ymom,mf_ux,mf_vy,z_nd_stg,dxInv);
456  RHS_a(i,j,k) += UppVpp;
457  });
458  } // end profile
459 
460  Box b2d = tbz; // Copy constructor
461  b2d.setRange(2,0);
462 
463  auto const lo = lbound(bx);
464  auto const hi = ubound(bx);
465 
466  {
467  BL_PROFILE("fast_rhs_b2d_loop_t");
468 
469 #ifdef AMREX_USE_GPU
470  ParallelFor(b2d, [=] AMREX_GPU_DEVICE (int i, int j, int)
471  {
472  // Moving terrain
473  Real rho_on_bdy = 0.5 * ( prev_cons(i,j,lo.z) + prev_cons(i,j,lo.z-1) );
474  RHS_a(i,j,lo.z) = rho_on_bdy * zp_t_arr(i,j,0);
475 
476  soln_a(i,j,lo.z) = RHS_a(i,j,lo.z) * inv_coeffB_a(i,j,lo.z);
477 
478  // w_khi = 0
479  RHS_a(i,j,hi.z+1) = 0.0;
480 
481  for (int k = lo.z+1; k <= hi.z+1; k++) {
482  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);
483  }
484 
485  for (int k = hi.z; k >= lo.z; k--) {
486  soln_a(i,j,k) -= ( coeffC_a(i,j,k) * inv_coeffB_a(i,j,k) ) * soln_a(i,j,k+1);
487  }
488 
489  // We assume that Omega == w at the top boundary and that changes in J there are irrelevant
490  cur_zmom(i,j,hi.z+1) = stg_zmom(i,j,hi.z+1) + soln_a(i,j,hi.z+1);
491  });
492 #else
493  for (int j = lo.y; j <= hi.y; ++j) {
494  AMREX_PRAGMA_SIMD
495  for (int i = lo.x; i <= hi.x; ++i) {
496 
497  Real rho_on_bdy = 0.5 * ( prev_cons(i,j,lo.z) + prev_cons(i,j,lo.z-1) );
498  RHS_a(i,j,lo.z) = rho_on_bdy * zp_t_arr(i,j,lo.z);
499 
500  soln_a(i,j,lo.z) = RHS_a(i,j,lo.z) * inv_coeffB_a(i,j,lo.z);
501  }
502  }
503 
504  for (int j = lo.y; j <= hi.y; ++j) {
505  AMREX_PRAGMA_SIMD
506  for (int i = lo.x; i <= hi.x; ++i) {
507  RHS_a (i,j,hi.z+1) = 0.0;
508  }
509  }
510  for (int k = lo.z+1; k <= hi.z+1; ++k) {
511  for (int j = lo.y; j <= hi.y; ++j) {
512  AMREX_PRAGMA_SIMD
513  for (int i = lo.x; i <= hi.x; ++i) {
514  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);
515  }
516  }
517  }
518  for (int k = hi.z; k >= lo.z; --k) {
519  for (int j = lo.y; j <= hi.y; ++j) {
520  AMREX_PRAGMA_SIMD
521  for (int i = lo.x; i <= hi.x; ++i) {
522  soln_a(i,j,k) -= ( coeffC_a(i,j,k) * inv_coeffB_a(i,j,k) ) * soln_a(i,j,k+1);
523  }
524  }
525  }
526 
527  // We assume that Omega == w at the top boundary and that changes in J there are irrelevant
528  for (int j = lo.y; j <= hi.y; ++j) {
529  AMREX_PRAGMA_SIMD
530  for (int i = lo.x; i <= hi.x; ++i) {
531  cur_zmom(i,j,hi.z+1) = stg_zmom(i,j,hi.z+1) + soln_a(i,j,hi.z+1);
532  }
533  }
534 #endif
535  } // end profile
536 
537  {
538  BL_PROFILE("fast_rhs_new_drhow");
539  tbz.setBig(2,hi.z);
540  ParallelFor(tbz, [=] AMREX_GPU_DEVICE (int i, int j, int k)
541  {
542  Real rho_on_face = 0.5 * (cur_cons(i,j,k,Rho_comp) + cur_cons(i,j,k-1,Rho_comp));
543 
544  if (k == lo.z) {
545  cur_zmom(i,j,k) = WFromOmega(i,j,k,rho_on_face*(z_t_arr(i,j,k)+zp_t_arr(i,j,k)),
546  cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv);
547 
548  // We need to set this here because it is used to define zflux_lo below
549  soln_a(i,j,k) = 0.;
550 
551  } else {
552 
553  Real UppVpp = WFromOmega(i,j,k,0.0,cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv)
554  - WFromOmega(i,j,k,0.0,stg_xmom,stg_ymom,mf_ux,mf_vy,z_nd_stg,dxInv);
555  Real wpp = soln_a(i,j,k) + UppVpp;
556  Real dJ_old_kface = 0.5 * (detJ_old(i,j,k) + detJ_old(i,j,k-1));
557  Real dJ_new_kface = 0.5 * (detJ_new(i,j,k) + detJ_new(i,j,k-1));
558 
559  cur_zmom(i,j,k) = dJ_old_kface * (stg_zmom(i,j,k) + wpp);
560  cur_zmom(i,j,k) /= dJ_new_kface;
561 
562  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)
563  - OmegaFromW(i,j,k,stg_zmom(i,j,k),stg_xmom,stg_ymom,mf_ux,mf_vy,z_nd_stg,dxInv);
564  soln_a(i,j,k) -= rho_on_face * zp_t_arr(i,j,k);
565  }
566  });
567  } // end profile
568 
569  // **************************************************************************
570  // Define updates in the RHS of rho and (rho theta)
571  // **************************************************************************
572  {
573  BL_PROFILE("fast_rho_final_update");
574  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
575  {
576  Real zflux_lo = beta_2 * soln_a(i,j,k ) + beta_1 * omega_arr(i,j,k);
577  Real zflux_hi = beta_2 * soln_a(i,j,k+1) + beta_1 * omega_arr(i,j,k+1);
578 
579  // Note that in the solve we effectively impose new_drho_w(i,j,vbx_hi.z+1)=0
580  // so we don't update avg_zmom at k=vbx_hi.z+1
581  avg_zmom(i,j,k) += facinv*zflux_lo / (mf_mx(i,j,0) * mf_my(i,j,0));
582  if (l_reflux) {
583  (flx_arr[2])(i,j,k,0) = zflux_lo / (mf_mx(i,j,0) * mf_my(i,j,0));
584  }
585 
586  // Note that the factor of (1/J) in the fast source term is canceled
587  // when we multiply old and new by detJ_old and detJ_new , respectively
588  // We have already scaled the slow source term to have the extra factor of dJ
589  Real fast_rhs_rho = -(temp_rhs_arr(i,j,k,0) + ( zflux_hi - zflux_lo ) * dzi);
590  Real temp_rho = detJ_old(i,j,k) * cur_cons(i,j,k,0) +
591  dtau * ( slow_rhs_cons(i,j,k,0) + fast_rhs_rho );
592  cur_cons(i,j,k,0) = temp_rho / detJ_new(i,j,k);
593 
594  // Note that the factor of (1/J) in the fast source term is canceled
595  // when we multiply old and new by detJ_old and detJ_new , respectively
596  // We have already scaled the slow source term to have the extra factor of dJ
597  Real fast_rhs_rhotheta = -( temp_rhs_arr(i,j,k,1) + 0.5 *
598  ( zflux_hi * (prim(i,j,k) + prim(i,j,k+1))
599  - zflux_lo * (prim(i,j,k) + prim(i,j,k-1)) ) * dzi );
600  Real temp_rth = detJ_old(i,j,k) * cur_cons(i,j,k,1) +
601  dtau * ( slow_rhs_cons(i,j,k,1) + fast_rhs_rhotheta );
602  cur_cons(i,j,k,1) = temp_rth / detJ_new(i,j,k);
603  if (l_reflux) {
604  (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));
605  }
606 
607  if (k == vbx_hi.z) {
608  avg_zmom(i,j,k+1) += facinv * zflux_hi / (mf_mx(i,j,0) * mf_my(i,j,0));
609  if (l_reflux) {
610  (flx_arr[2])(i,j,k+1,0) = zflux_hi / (mf_mx(i,j,0) * mf_my(i,j,0));
611  (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));
612  }
613  }
614 
615  // add in source terms for cell-centered conserved variables
616  cur_cons(i,j,k,Rho_comp) += dtau * cc_src_arr(i,j,k,Rho_comp);
617  cur_cons(i,j,k,RhoTheta_comp) += dtau * cc_src_arr(i,j,k,RhoTheta_comp);
618  });
619  } // end profile
620 
621  // We only add to the flux registers in the final RK step
622  if (l_reflux) {
623  int strt_comp_reflux = 0;
624  int num_comp_reflux = 2;
625  if (level < finest_level) {
626  fr_as_crse->CrseAdd(mfi,
627  {{AMREX_D_DECL(&(flux[0]), &(flux[1]), &(flux[2]))}},
628  dx, dtau, strt_comp_reflux, strt_comp_reflux, num_comp_reflux, RunOn::Device);
629  }
630  if (level > 0) {
631  fr_as_fine->FineAdd(mfi,
632  {{AMREX_D_DECL(&(flux[0]), &(flux[1]), &(flux[2]))}},
633  dx, dtau, strt_comp_reflux, strt_comp_reflux, num_comp_reflux, RunOn::Device);
634  }
635 
636  // This is necessary here so we don't go on to the next FArrayBox without
637  // having finished copying the fluxes into the FluxRegisters (since the fluxes
638  // are stored in temporary FArrayBox's)
639  Gpu::streamSynchronize();
640 
641  } // two-way coupling
642 
643  } // mfi
644  } // OMP
645 }
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
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
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:415
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:110
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:139
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:167
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:465
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
Here is the call graph for this function: