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 &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 &  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
80 {
81  BL_PROFILE_REGION("erf_fast_rhs_MT()");
82 
83  Real beta_1 = 0.5 * (1.0 - beta_s); // multiplies explicit terms
84  Real beta_2 = 0.5 * (1.0 + beta_s); // multiplies implicit terms
85 
86  // How much do we project forward the (rho theta) that is used in the horizontal momentum equations
87  Real beta_d = 0.1;
88 
89  const Real* dx = geom.CellSize();
90  const GpuArray<Real, AMREX_SPACEDIM> dxInv = geom.InvCellSizeArray();
91 
92  Real dxi = dxInv[0];
93  Real dyi = dxInv[1];
94  Real dzi = dxInv[2];
95 
96  MultiFab coeff_A_mf(fast_coeffs, make_alias, 0, 1);
97  MultiFab inv_coeff_B_mf(fast_coeffs, make_alias, 1, 1);
98  MultiFab coeff_C_mf(fast_coeffs, make_alias, 2, 1);
99  MultiFab coeff_P_mf(fast_coeffs, make_alias, 3, 1);
100  MultiFab coeff_Q_mf(fast_coeffs, make_alias, 4, 1);
101 
102  // *************************************************************************
103  // Set gravity as a vector
104  const Array<Real,AMREX_SPACEDIM> grav{0.0, 0.0, -gravity};
105  const GpuArray<Real,AMREX_SPACEDIM> grav_gpu{grav[0], grav[1], grav[2]};
106 
107  MultiFab extrap(S_data[IntVars::cons].boxArray(),S_data[IntVars::cons].DistributionMap(),1,1);
108 
109  MultiFab Omega(S_data[IntVars::zmom].boxArray(), S_data[IntVars::zmom].DistributionMap(), 1, 1);
110 
111  // *************************************************************************
112  // Define updates in the current RK stg
113  // *************************************************************************
114 #ifdef _OPENMP
115 #pragma omp parallel if (Gpu::notInLaunchRegion())
116 #endif
117  {
118  FArrayBox temp_rhs_fab;
119 
120  FArrayBox RHS_fab;
121  FArrayBox soln_fab;
122 
123  std::array<FArrayBox,AMREX_SPACEDIM> flux;
124 
125  // NOTE: we leave tiling off here for efficiency -- to make this loop work with tiling
126  // will require additional changes
127  for ( MFIter mfi(S_stg_data[IntVars::cons],false); mfi.isValid(); ++mfi)
128  {
129  Box bx = mfi.tilebox();
130  Box tbx = surroundingNodes(bx,0);
131  Box tby = surroundingNodes(bx,1);
132  Box tbz = surroundingNodes(bx,2);
133 
134  Box vbx = mfi.validbox();
135  const auto& vbx_hi = ubound(vbx);
136 
137  const Array4<Real const>& xmom_src_arr = xmom_src.const_array(mfi);
138  const Array4<Real const>& ymom_src_arr = ymom_src.const_array(mfi);
139  const Array4<Real const>& zmom_src_arr = zmom_src.const_array(mfi);
140  const Array4<Real const>& cc_src_arr = cc_src.const_array(mfi);
141 
142  const Array4<const Real> & stg_cons = S_stg_data[IntVars::cons].const_array(mfi);
143  const Array4<const Real> & stg_xmom = S_stg_data[IntVars::xmom].const_array(mfi);
144  const Array4<const Real> & stg_ymom = S_stg_data[IntVars::ymom].const_array(mfi);
145  const Array4<const Real> & stg_zmom = S_stg_data[IntVars::zmom].const_array(mfi);
146  const Array4<const Real> & prim = S_stg_prim.const_array(mfi);
147 
148  const Array4<const Real>& slow_rhs_cons = S_slow_rhs[IntVars::cons].const_array(mfi);
149  const Array4<const Real>& slow_rhs_rho_u = S_slow_rhs[IntVars::xmom].const_array(mfi);
150  const Array4<const Real>& slow_rhs_rho_v = S_slow_rhs[IntVars::ymom].const_array(mfi);
151  const Array4<const Real>& slow_rhs_rho_w = S_slow_rhs[IntVars::zmom].const_array(mfi);
152 
153  const Array4<Real>& cur_cons = S_data[IntVars::cons].array(mfi);
154  const Array4<Real>& cur_xmom = S_data[IntVars::xmom].array(mfi);
155  const Array4<Real>& cur_ymom = S_data[IntVars::ymom].array(mfi);
156  const Array4<Real>& cur_zmom = S_data[IntVars::zmom].array(mfi);
157 
158  const Array4<Real>& lagged_delta_rt = S_scratch[IntVars::cons].array(mfi);
159 
160  const Array4<const Real>& prev_cons = S_prev[IntVars::cons].const_array(mfi);
161  const Array4<const Real>& prev_xmom = S_prev[IntVars::xmom].const_array(mfi);
162  const Array4<const Real>& prev_ymom = S_prev[IntVars::ymom].const_array(mfi);
163  const Array4<const Real>& prev_zmom = S_prev[IntVars::zmom].const_array(mfi);
164 
165  // These store the advection momenta which we will use to update the slow variables
166  const Array4<Real>& avg_xmom = S_scratch[IntVars::xmom].array(mfi);
167  const Array4<Real>& avg_ymom = S_scratch[IntVars::ymom].array(mfi);
168  const Array4<Real>& avg_zmom = S_scratch[IntVars::zmom].array(mfi);
169 
170  const Array4<const Real>& z_nd_old = z_phys_nd_old->const_array(mfi);
171  const Array4<const Real>& z_nd_new = z_phys_nd_new->const_array(mfi);
172  const Array4<const Real>& z_nd_stg = z_phys_nd_stg->const_array(mfi);
173  const Array4<const Real>& detJ_old = detJ_cc_old->const_array(mfi);
174  const Array4<const Real>& detJ_new = detJ_cc_new->const_array(mfi);
175  const Array4<const Real>& detJ_stg = detJ_cc_stg->const_array(mfi);
176 
177  const Array4<const Real>& z_t_arr = z_t_rk->const_array(mfi);
178  const Array4<const Real>& zp_t_arr = z_t_pert->const_array(mfi);
179 
180  const Array4< Real>& omega_arr = Omega.array(mfi);
181 
182  const Array4<const Real>& pi_stage_ca = pi_stage.const_array(mfi);
183 
184  const Array4<Real>& theta_extrap = extrap.array(mfi);
185 
186  // Map factors
187  const Array4<const Real>& mf_mx = mapfac[MapFacType::m_x]->const_array(mfi);
188  const Array4<const Real>& mf_my = mapfac[MapFacType::m_y]->const_array(mfi);
189  const Array4<const Real>& mf_ux = mapfac[MapFacType::u_x]->const_array(mfi);
190  const Array4<const Real>& mf_vy = mapfac[MapFacType::v_y]->const_array(mfi);
191 
192  // Note: it is important to grow the tilebox rather than use growntilebox because
193  // we need to fill the ghost cells of the tilebox so we can use them below
194  Box gbx = mfi.tilebox(); gbx.grow(1);
195  Box gtbx = mfi.nodaltilebox(0); gtbx.grow(1); gtbx.setSmall(2,0);
196  Box gtby = mfi.nodaltilebox(1); gtby.grow(1); gtby.setSmall(2,0);
197 
198  {
199  BL_PROFILE("fast_rhs_copies_0");
200  if (step == 0) {
201  ParallelFor(gbx,
202  [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
203  cur_cons(i,j,k,Rho_comp) = prev_cons(i,j,k,Rho_comp);
204  cur_cons(i,j,k,RhoTheta_comp) = prev_cons(i,j,k,RhoTheta_comp);
205 
206  Real delta_rt = cur_cons(i,j,k,RhoTheta_comp) - stg_cons(i,j,k,RhoTheta_comp);
207  theta_extrap(i,j,k) = delta_rt;
208 
209  // We define lagged_delta_rt for our next step as the current delta_rt
210  lagged_delta_rt(i,j,k,RhoTheta_comp) = delta_rt;
211  });
212  } else if (use_lagged_delta_rt) {
213  // This is the default for cases with no or static terrain
214  ParallelFor(gbx,
215  [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
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 + beta_d * (delta_rt - lagged_delta_rt(i,j,k,RhoTheta_comp));
218 
219  // We define lagged_delta_rt for our next step as the current delta_rt
220  lagged_delta_rt(i,j,k,RhoTheta_comp) = delta_rt;
221  });
222  } else {
223  // For the moving wave problem, this choice seems more robust
224  ParallelFor(gbx,
225  [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
226  theta_extrap(i,j,k) = cur_cons(i,j,k,RhoTheta_comp) - stg_cons(i,j,k,RhoTheta_comp);
227  });
228  } // if step
229  } // end profile
230 
231  RHS_fab.resize (tbz,1, The_Async_Arena());
232  soln_fab.resize (tbz,1, The_Async_Arena());
233  temp_rhs_fab.resize(tbz,2, The_Async_Arena());
234 
235  auto const& RHS_a = RHS_fab.array();
236  auto const& soln_a = soln_fab.array();
237  auto const& temp_rhs_arr = temp_rhs_fab.array();
238 
239  auto const& coeffA_a = coeff_A_mf.array(mfi);
240  auto const& inv_coeffB_a = inv_coeff_B_mf.array(mfi);
241  auto const& coeffC_a = coeff_C_mf.array(mfi);
242  auto const& coeffP_a = coeff_P_mf.array(mfi);
243  auto const& coeffQ_a = coeff_Q_mf.array(mfi);
244 
245  // *********************************************************************
246  // Define updates in the RHS of {x, y, z}-momentum equations
247  // *********************************************************************
248  {
249  BL_PROFILE("fast_rhs_xymom_T");
250  ParallelFor(tbx, tby,
251  [=] AMREX_GPU_DEVICE (int i, int j, int k)
252  {
253  // Add (negative) gradient of (rho theta) multiplied by lagged "pi"
254  Real h_xi_old = Compute_h_xi_AtIface(i, j, k, dxInv, z_nd_old);
255  Real h_zeta_old = Compute_h_zeta_AtIface(i, j, k, dxInv, z_nd_old);
256  Real gp_xi = (theta_extrap(i,j,k) - theta_extrap(i-1,j,k)) * dxi;
257  Real gp_zeta_on_iface = (k == 0) ?
258  0.5 * dzi * ( theta_extrap(i-1,j,k+1) + theta_extrap(i,j,k+1)
259  -theta_extrap(i-1,j,k ) - theta_extrap(i,j,k ) ) :
260  0.25 * dzi * ( theta_extrap(i-1,j,k+1) + theta_extrap(i,j,k+1)
261  -theta_extrap(i-1,j,k-1) - theta_extrap(i,j,k-1) );
262  Real gpx = h_zeta_old * gp_xi - h_xi_old * gp_zeta_on_iface;
263  gpx *= mf_ux(i,j,0);
264 
265  if (l_use_moisture) {
266  Real q = 0.5 * ( prim(i,j,k,PrimQ1_comp) + prim(i-1,j,k,PrimQ1_comp)
267  +prim(i,j,k,PrimQ2_comp) + prim(i-1,j,k,PrimQ2_comp) );
268  gpx /= (1.0 + q);
269  }
270 
271  Real pi_c = 0.5 * (pi_stage_ca(i-1,j,k,0) + pi_stage_ca(i ,j,k,0));
272  Real fast_rhs_rho_u = -Gamma * R_d * pi_c * gpx;
273 
274  // We have already scaled the source terms to have the extra factor of dJ
275  cur_xmom(i,j,k) = h_zeta_old * prev_xmom(i,j,k) + dtau * fast_rhs_rho_u
276  + dtau * slow_rhs_rho_u(i,j,k)
277  + dtau * xmom_src_arr(i,j,k);
278  },
279  [=] AMREX_GPU_DEVICE (int i, int j, int k)
280  {
281  // Add (negative) gradient of (rho theta) multiplied by lagged "pi"
282  Real h_eta_old = Compute_h_eta_AtJface(i, j, k, dxInv, z_nd_old);
283  Real h_zeta_old = Compute_h_zeta_AtJface(i, j, k, dxInv, z_nd_old);
284  Real gp_eta = (theta_extrap(i,j,k) -theta_extrap(i,j-1,k)) * dyi;
285  Real gp_zeta_on_jface = (k == 0) ?
286  0.5 * dzi * ( theta_extrap(i,j,k+1) + theta_extrap(i,j-1,k+1)
287  -theta_extrap(i,j,k ) - theta_extrap(i,j-1,k ) ) :
288  0.25 * dzi * ( theta_extrap(i,j,k+1) + theta_extrap(i,j-1,k+1)
289  -theta_extrap(i,j,k-1) - theta_extrap(i,j-1,k-1) );
290  Real gpy = h_zeta_old * gp_eta - h_eta_old * gp_zeta_on_jface;
291  gpy *= mf_vy(i,j,0);
292 
293  if (l_use_moisture) {
294  Real q = 0.5 * ( prim(i,j,k,PrimQ1_comp) + prim(i,j-1,k,PrimQ1_comp)
295  +prim(i,j,k,PrimQ2_comp) + prim(i,j-1,k,PrimQ2_comp) );
296  gpy /= (1.0 + q);
297  }
298 
299  Real pi_c = 0.5 * (pi_stage_ca(i,j-1,k,0) + pi_stage_ca(i,j ,k,0));
300  Real fast_rhs_rho_v = -Gamma * R_d * pi_c * gpy;
301 
302  // We have already scaled the source terms to have the extra factor of dJ
303  cur_ymom(i, j, k) = h_zeta_old * prev_ymom(i,j,k) + dtau * fast_rhs_rho_v
304  + dtau * slow_rhs_rho_v(i,j,k)
305  + dtau * ymom_src_arr(i,j,k);
306  });
307  } // end profile
308 
309  // *************************************************************************
310  // Define flux arrays for use in advection
311  // *************************************************************************
312  for (int dir = 0; dir < AMREX_SPACEDIM; ++dir) {
313  flux[dir].resize(surroundingNodes(bx,dir),2);
314  flux[dir].setVal<RunOn::Device>(0.);
315  }
316  const GpuArray<const Array4<Real>, AMREX_SPACEDIM>
317  flx_arr{{AMREX_D_DECL(flux[0].array(), flux[1].array(), flux[2].array())}};
318 
319  // *********************************************************************
320  {
321  BL_PROFILE("fast_T_making_rho_rhs");
322  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
323  {
324  Real h_zeta_stg_xlo = Compute_h_zeta_AtIface(i, j , k, dxInv, z_nd_stg);
325  Real h_zeta_stg_xhi = Compute_h_zeta_AtIface(i+1,j , k, dxInv, z_nd_stg);
326  Real xflux_lo = cur_xmom(i ,j,k) - stg_xmom(i ,j,k)*h_zeta_stg_xlo;
327  Real xflux_hi = cur_xmom(i+1,j,k) - stg_xmom(i+1,j,k)*h_zeta_stg_xhi;
328 
329  Real h_zeta_stg_yhi = Compute_h_zeta_AtJface(i, j+1, k, dxInv, z_nd_stg);
330  Real h_zeta_stg_ylo = Compute_h_zeta_AtJface(i, j , k, dxInv, z_nd_stg);
331  Real yflux_lo = cur_ymom(i,j ,k) - stg_ymom(i,j ,k)*h_zeta_stg_ylo;
332  Real yflux_hi = cur_ymom(i,j+1,k) - stg_ymom(i,j+1,k)*h_zeta_stg_yhi;
333 
334  // NOTE: we are saving the (1/J) weighting for later when we add this to rho and theta
335  temp_rhs_arr(i,j,k,0) = ( xflux_hi - xflux_lo ) * dxi + ( yflux_hi - yflux_lo ) * dyi;
336  temp_rhs_arr(i,j,k,1) = (( xflux_hi * (prim(i,j,k,0) + prim(i+1,j,k,0)) -
337  xflux_lo * (prim(i,j,k,0) + prim(i-1,j,k,0)) ) * dxi +
338  ( yflux_hi * (prim(i,j,k,0) + prim(i,j+1,k,0)) -
339  yflux_lo * (prim(i,j,k,0) + prim(i,j-1,k,0)) ) * dyi) * 0.5;
340 
341  if (l_reflux) {
342  (flx_arr[0])(i,j,k,0) = xflux_lo;
343  (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));
344 
345  (flx_arr[1])(i,j,k,0) = yflux_lo;
346  (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));
347 
348  if (i == vbx_hi.x) {
349  (flx_arr[0])(i+1,j,k,0) = xflux_hi;
350  (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));
351  }
352  if (j == vbx_hi.y) {
353  (flx_arr[1])(i,j+1,k,0) = yflux_hi;
354  (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));
355  }
356  }
357  });
358  } // end profile
359 
360  // *********************************************************************
361  // This must be done before we set cur_xmom and cur_ymom, since those
362  // in fact point to the same array as prev_xmom and prev_ymom
363  // *********************************************************************
364  Box gbxo = mfi.nodaltilebox(2);
365  {
366  BL_PROFILE("fast_MT_making_omega");
367  Box gbxo_lo = gbxo; gbxo_lo.setBig(2,0);
368  ParallelFor(gbxo_lo, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
369  omega_arr(i,j,k) = 0.;
370  });
371  Box gbxo_hi = gbxo; gbxo_hi.setSmall(2,gbxo.bigEnd(2));
372  ParallelFor(gbxo_hi, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
373  omega_arr(i,j,k) = prev_zmom(i,j,k) - stg_zmom(i,j,k) - zp_t_arr(i,j,k);
374  });
375  Box gbxo_mid = gbxo; gbxo_mid.setSmall(2,1); gbxo_mid.setBig(2,gbxo.bigEnd(2)-1);
376  ParallelFor(gbxo_mid, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept {
377  omega_arr(i,j,k) =
378  ( OmegaFromW(i,j,k,prev_zmom(i,j,k),prev_xmom,prev_ymom,mf_ux,mf_vy,z_nd_old,dxInv)
379  -OmegaFromW(i,j,k, stg_zmom(i,j,k), stg_xmom, stg_ymom,mf_ux,mf_vy,z_nd_old,dxInv) )
380  - zp_t_arr(i,j,k);
381  });
382  } // end profile
383  // *********************************************************************
384 
385  ParallelFor(tbx, tby,
386  [=] AMREX_GPU_DEVICE (int i, int j, int k)
387  {
388  Real h_zeta_new = Compute_h_zeta_AtIface(i, j, k, dxInv, z_nd_new);
389  cur_xmom(i, j, k) /= h_zeta_new;
390  avg_xmom(i,j,k) += facinv*(cur_xmom(i,j,k) - stg_xmom(i,j,k));
391  },
392  [=] AMREX_GPU_DEVICE (int i, int j, int k)
393  {
394  Real h_zeta_new = Compute_h_zeta_AtJface(i, j, k, dxInv, z_nd_new);
395  cur_ymom(i, j, k) /= h_zeta_new;
396  avg_ymom(i,j,k) += facinv*(cur_ymom(i,j,k) - stg_ymom(i,j,k));
397  });
398 
399  Box bx_shrunk_in_k = bx;
400  int klo = tbz.smallEnd(2);
401  int khi = tbz.bigEnd(2);
402  bx_shrunk_in_k.setSmall(2,klo+1);
403  bx_shrunk_in_k.setBig(2,khi-1);
404 
405  // Note that the notes use "g" to mean the magnitude of gravity, so it is positive
406  // We set grav_gpu[2] to be the vector component which is negative
407  // We define halfg to match the notes (which is why we take the absolute value)
408  Real halfg = std::abs(0.5 * grav_gpu[2]);
409 
410  {
411  BL_PROFILE("fast_loop_on_shrunk_t");
412  //Note we don't act on the bottom or top boundaries of the domain
413  ParallelFor(bx_shrunk_in_k, [=] AMREX_GPU_DEVICE (int i, int j, int k)
414  {
415  Real dJ_old_kface = 0.5 * (detJ_old(i,j,k) + detJ_old(i,j,k-1));
416  Real dJ_new_kface = 0.5 * (detJ_new(i,j,k) + detJ_new(i,j,k-1));
417  Real dJ_stg_kface = 0.5 * (detJ_stg(i,j,k) + detJ_stg(i,j,k-1));
418 
419  Real coeff_P = coeffP_a(i,j,k);
420  Real coeff_Q = coeffQ_a(i,j,k);
421 
422  if (l_use_moisture) {
423  Real q = 0.5 * ( prim(i,j,k,PrimQ1_comp) + prim(i,j,k-1,PrimQ1_comp)
424  +prim(i,j,k,PrimQ2_comp) + prim(i,j,k-1,PrimQ2_comp) );
425  coeff_P /= (1.0 + q);
426  coeff_Q /= (1.0 + q);
427  }
428 
429  Real theta_t_lo = 0.5 * ( prim(i,j,k-2,PrimTheta_comp) + prim(i,j,k-1,PrimTheta_comp) );
430  Real theta_t_mid = 0.5 * ( prim(i,j,k-1,PrimTheta_comp) + prim(i,j,k ,PrimTheta_comp) );
431  Real theta_t_hi = 0.5 * ( prim(i,j,k ,PrimTheta_comp) + prim(i,j,k+1,PrimTheta_comp) );
432 
433  // line 2 last two terms (order dtau)
434  Real R0_tmp = coeff_P * cur_cons(i,j,k ,RhoTheta_comp) * dJ_old_kface
435  + coeff_Q * cur_cons(i,j,k-1,RhoTheta_comp) * dJ_old_kface
436  - coeff_P * stg_cons(i,j,k ,RhoTheta_comp) * dJ_stg_kface
437  - coeff_Q * stg_cons(i,j,k-1,RhoTheta_comp) * dJ_stg_kface
438  - halfg * ( cur_cons(i,j,k,Rho_comp) + cur_cons(i,j,k-1,Rho_comp) ) * dJ_old_kface
439  + halfg * ( stg_cons(i,j,k,Rho_comp) + stg_cons(i,j,k-1,Rho_comp) ) * dJ_stg_kface;
440 
441  // line 3 residuals (order dtau^2) 1.0 <-> beta_2
442  Real R1_tmp = - halfg * ( slow_rhs_cons(i,j,k ,Rho_comp) +
443  slow_rhs_cons(i,j,k-1,Rho_comp) )
444  + ( coeff_P * slow_rhs_cons(i,j,k ,RhoTheta_comp) +
445  coeff_Q * slow_rhs_cons(i,j,k-1,RhoTheta_comp) );
446 
447  Real Omega_kp1 = omega_arr(i,j,k+1);
448  Real Omega_k = omega_arr(i,j,k );
449  Real Omega_km1 = omega_arr(i,j,k-1);
450 
451  // consolidate lines 4&5 (order dtau^2)
452  R1_tmp += ( halfg ) *
453  ( beta_1 * dzi * (Omega_kp1 - Omega_km1) + temp_rhs_arr(i,j,k,Rho_comp) + temp_rhs_arr(i,j,k-1,Rho_comp));
454 
455  // consolidate lines 6&7 (order dtau^2)
456  R1_tmp += -(
457  coeff_P * ( beta_1 * dzi * (Omega_kp1*theta_t_hi - Omega_k*theta_t_mid) + temp_rhs_arr(i,j,k ,RhoTheta_comp) ) +
458  coeff_Q * ( beta_1 * dzi * (Omega_k*theta_t_mid - Omega_km1*theta_t_lo) + temp_rhs_arr(i,j,k-1,RhoTheta_comp) ) );
459 
460  // line 1
461  RHS_a(i,j,k) = dJ_old_kface * prev_zmom(i,j,k) - dJ_stg_kface * stg_zmom(i,j,k)
462  + dtau * (slow_rhs_rho_w(i,j,k) + R0_tmp + dtau*beta_2*R1_tmp + zmom_src_arr(i,j,k));
463 
464  // We cannot use omega_arr here since that was built with old_rho_u and old_rho_v ...
465  Real UppVpp = dJ_new_kface * OmegaFromW(i,j,k,0.,cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv)
466  -dJ_stg_kface * OmegaFromW(i,j,k,0.,stg_xmom,stg_ymom,mf_ux,mf_vy,z_nd_stg,dxInv);
467  RHS_a(i,j,k) += UppVpp;
468  });
469  } // end profile
470 
471  Box b2d = tbz; // Copy constructor
472  b2d.setRange(2,0);
473 
474  auto const lo = lbound(bx);
475  auto const hi = ubound(bx);
476 
477  {
478  BL_PROFILE("fast_rhs_b2d_loop_t");
479 
480 #ifdef AMREX_USE_GPU
481  ParallelFor(b2d, [=] AMREX_GPU_DEVICE (int i, int j, int)
482  {
483  // Moving terrain
484  Real rho_on_bdy = 0.5 * ( prev_cons(i,j,lo.z) + prev_cons(i,j,lo.z-1) );
485  RHS_a(i,j,lo.z) = rho_on_bdy * zp_t_arr(i,j,0);
486 
487  soln_a(i,j,lo.z) = RHS_a(i,j,lo.z) * inv_coeffB_a(i,j,lo.z);
488 
489  // w_khi = 0
490  RHS_a(i,j,hi.z+1) = 0.0;
491 
492  for (int k = lo.z+1; k <= hi.z+1; k++) {
493  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);
494  }
495 
496  for (int k = hi.z; k >= lo.z; k--) {
497  soln_a(i,j,k) -= ( coeffC_a(i,j,k) * inv_coeffB_a(i,j,k) ) * soln_a(i,j,k+1);
498  }
499 
500  // We assume that Omega == w at the top boundary and that changes in J there are irrelevant
501  cur_zmom(i,j,hi.z+1) = stg_zmom(i,j,hi.z+1) + soln_a(i,j,hi.z+1);
502  });
503 #else
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 
508  Real rho_on_bdy = 0.5 * ( prev_cons(i,j,lo.z) + prev_cons(i,j,lo.z-1) );
509  RHS_a(i,j,lo.z) = rho_on_bdy * zp_t_arr(i,j,lo.z);
510 
511  soln_a(i,j,lo.z) = RHS_a(i,j,lo.z) * inv_coeffB_a(i,j,lo.z);
512  }
513  }
514 
515  for (int j = lo.y; j <= hi.y; ++j) {
516  AMREX_PRAGMA_SIMD
517  for (int i = lo.x; i <= hi.x; ++i) {
518  RHS_a (i,j,hi.z+1) = 0.0;
519  }
520  }
521  for (int k = lo.z+1; k <= hi.z+1; ++k) {
522  for (int j = lo.y; j <= hi.y; ++j) {
523  AMREX_PRAGMA_SIMD
524  for (int i = lo.x; i <= hi.x; ++i) {
525  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);
526  }
527  }
528  }
529  for (int k = hi.z; k >= lo.z; --k) {
530  for (int j = lo.y; j <= hi.y; ++j) {
531  AMREX_PRAGMA_SIMD
532  for (int i = lo.x; i <= hi.x; ++i) {
533  soln_a(i,j,k) -= ( coeffC_a(i,j,k) * inv_coeffB_a(i,j,k) ) * soln_a(i,j,k+1);
534  }
535  }
536  }
537 
538  // We assume that Omega == w at the top boundary and that changes in J there are irrelevant
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  cur_zmom(i,j,hi.z+1) = stg_zmom(i,j,hi.z+1) + soln_a(i,j,hi.z+1);
543  }
544  }
545 #endif
546  } // end profile
547 
548  {
549  BL_PROFILE("fast_rhs_new_drhow");
550  tbz.setBig(2,hi.z);
551  ParallelFor(tbz, [=] AMREX_GPU_DEVICE (int i, int j, int k)
552  {
553  Real rho_on_face = 0.5 * (cur_cons(i,j,k,Rho_comp) + cur_cons(i,j,k-1,Rho_comp));
554 
555  if (k == lo.z) {
556  cur_zmom(i,j,k) = WFromOmega(i,j,k,rho_on_face*(z_t_arr(i,j,k)+zp_t_arr(i,j,k)),
557  cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv);
558 
559  // We need to set this here because it is used to define zflux_lo below
560  soln_a(i,j,k) = 0.;
561 
562  } else {
563 
564  Real UppVpp = WFromOmega(i,j,k,0.0,cur_xmom,cur_ymom,mf_ux,mf_vy,z_nd_new,dxInv)
565  - WFromOmega(i,j,k,0.0,stg_xmom,stg_ymom,mf_ux,mf_vy,z_nd_stg,dxInv);
566  Real wpp = soln_a(i,j,k) + UppVpp;
567  Real dJ_old_kface = 0.5 * (detJ_old(i,j,k) + detJ_old(i,j,k-1));
568  Real dJ_new_kface = 0.5 * (detJ_new(i,j,k) + detJ_new(i,j,k-1));
569 
570  cur_zmom(i,j,k) = dJ_old_kface * (stg_zmom(i,j,k) + wpp);
571  cur_zmom(i,j,k) /= dJ_new_kface;
572 
573  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)
574  - OmegaFromW(i,j,k,stg_zmom(i,j,k),stg_xmom,stg_ymom,mf_ux,mf_vy,z_nd_stg,dxInv);
575  soln_a(i,j,k) -= rho_on_face * zp_t_arr(i,j,k);
576  }
577  });
578  } // end profile
579 
580  // **************************************************************************
581  // Define updates in the RHS of rho and (rho theta)
582  // **************************************************************************
583  {
584  BL_PROFILE("fast_rho_final_update");
585  ParallelFor(bx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
586  {
587  Real zflux_lo = beta_2 * soln_a(i,j,k ) + beta_1 * omega_arr(i,j,k);
588  Real zflux_hi = beta_2 * soln_a(i,j,k+1) + beta_1 * omega_arr(i,j,k+1);
589 
590  // Note that in the solve we effectively impose new_drho_w(i,j,vbx_hi.z+1)=0
591  // so we don't update avg_zmom at k=vbx_hi.z+1
592  avg_zmom(i,j,k) += facinv*zflux_lo / (mf_mx(i,j,0) * mf_my(i,j,0));
593  if (l_reflux) {
594  (flx_arr[2])(i,j,k,0) = zflux_lo / (mf_mx(i,j,0) * mf_my(i,j,0));
595  }
596 
597  // Note that the factor of (1/J) in the fast source term is canceled
598  // when we multiply old and new by detJ_old and detJ_new , respectively
599  // We have already scaled the slow source term to have the extra factor of dJ
600  Real fast_rhs_rho = -(temp_rhs_arr(i,j,k,0) + ( zflux_hi - zflux_lo ) * dzi);
601  Real temp_rho = detJ_old(i,j,k) * cur_cons(i,j,k,0) +
602  dtau * ( slow_rhs_cons(i,j,k,0) + fast_rhs_rho );
603  cur_cons(i,j,k,0) = temp_rho / detJ_new(i,j,k);
604 
605  // Note that the factor of (1/J) in the fast source term is canceled
606  // when we multiply old and new by detJ_old and detJ_new , respectively
607  // We have already scaled the slow source term to have the extra factor of dJ
608  Real fast_rhs_rhotheta = -( temp_rhs_arr(i,j,k,1) + 0.5 *
609  ( zflux_hi * (prim(i,j,k) + prim(i,j,k+1))
610  - zflux_lo * (prim(i,j,k) + prim(i,j,k-1)) ) * dzi );
611  Real temp_rth = detJ_old(i,j,k) * cur_cons(i,j,k,1) +
612  dtau * ( slow_rhs_cons(i,j,k,1) + fast_rhs_rhotheta );
613  cur_cons(i,j,k,1) = temp_rth / detJ_new(i,j,k);
614  if (l_reflux) {
615  (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));
616  }
617 
618  if (k == vbx_hi.z) {
619  avg_zmom(i,j,k+1) += facinv * zflux_hi / (mf_mx(i,j,0) * mf_my(i,j,0));
620  if (l_reflux) {
621  (flx_arr[2])(i,j,k+1,0) = zflux_hi / (mf_mx(i,j,0) * mf_my(i,j,0));
622  (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));
623  }
624  }
625 
626  // add in source terms for cell-centered conserved variables
627  cur_cons(i,j,k,Rho_comp) += dtau * cc_src_arr(i,j,k,Rho_comp);
628  cur_cons(i,j,k,RhoTheta_comp) += dtau * cc_src_arr(i,j,k,RhoTheta_comp);
629  });
630  } // end profile
631 
632  // We only add to the flux registers in the final RK step
633  if (l_reflux) {
634  int strt_comp_reflux = 0;
635  int num_comp_reflux = 2;
636  if (level < finest_level) {
637  fr_as_crse->CrseAdd(mfi,
638  {{AMREX_D_DECL(&(flux[0]), &(flux[1]), &(flux[2]))}},
639  dx, dtau, strt_comp_reflux, strt_comp_reflux, num_comp_reflux, RunOn::Device);
640  }
641  if (level > 0) {
642  fr_as_fine->FineAdd(mfi,
643  {{AMREX_D_DECL(&(flux[0]), &(flux[1]), &(flux[2]))}},
644  dx, dtau, strt_comp_reflux, strt_comp_reflux, num_comp_reflux, RunOn::Device);
645  }
646 
647  // This is necessary here so we don't go on to the next FArrayBox without
648  // having finished copying the fluxes into the FluxRegisters (since the fluxes
649  // are stored in temporary FArrayBox's)
650  Gpu::streamSynchronize();
651 
652  } // two-way coupling
653 
654  } // mfi
655  } // OMP
656 }
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:22
m_y
@ m_y
Definition: ERF_DataStruct.H:22
u_x
@ u_x
Definition: ERF_DataStruct.H:21
m_x
@ m_x
Definition: ERF_DataStruct.H:21
PrimQ1_comp
#define PrimQ1_comp
Definition: ERF_IndexDefines.H:53
PrimQ2_comp
#define PrimQ2_comp
Definition: ERF_IndexDefines.H:54
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
Here is the call graph for this function: