ADAM Documentation

Appearance

adam_prism_fnl_object ​

ADAM, PRISM (Plasma Research usIng Simulation Methods) equations system class definition, GPU (FNL) backend.

Source: src/app/prism/fnl/adam_prism_fnl_object.F90

Dependencies

Contents ​

Variables ​

NameTypeAttributesDescription
FV_FLUX_VARIANT_MAXWELLinteger(kind=I4P)parameterPlain dimensional Maxwell fluxes.
FV_FLUX_VARIANT_MAXWELL_ADIMinteger(kind=I4P)parameterPlain adimensional Maxwell fluxes.
FV_FLUX_VARIANT_MAXWELL_DIV_Dinteger(kind=I4P)parameterDimensional hyperbolic fluxes with D cleaning.
FV_FLUX_VARIANT_MAXWELL_DIV_Binteger(kind=I4P)parameterDimensional hyperbolic fluxes with B cleaning.
FV_FLUX_VARIANT_MAXWELL_DIV_D_Binteger(kind=I4P)parameterDimensional hyperbolic fluxes with D/B cleaning.
FV_FLUX_VARIANT_MAXWELL_ADIM_DIV_Dinteger(kind=I4P)parameterAdimensional hyperbolic fluxes with D cleaning.
FV_FLUX_VARIANT_MAXWELL_ADIM_DIV_Binteger(kind=I4P)parameterAdimensional hyperbolic fluxes with B cleaning.
FV_FLUX_VARIANT_MAXWELL_ADIM_DIV_D_Binteger(kind=I4P)parameterAdimensional hyperbolic fluxes with D/B cleaning.
FD_RESIDUAL_VARIANT_PLAINinteger(kind=I4P)parameterCentered-FD Maxwell residual without hyperbolic cleaning.
FD_RESIDUAL_VARIANT_PHIinteger(kind=I4P)parameterCentered-FD Maxwell residual with phi cleaning only.
FD_RESIDUAL_VARIANT_PSIinteger(kind=I4P)parameterCentered-FD Maxwell residual with psi cleaning only.
FD_RESIDUAL_VARIANT_PHI_PSIinteger(kind=I4P)parameterCentered-FD Maxwell residual with phi/psi cleaning.

Derived Types ​

prism_fnl_object ​

PRISM equations system class definition, GPU (FNL) backend.

Inheritance

Extends: prism_common_object

Components ​

NameTypeAttributesDescription
iotype(io_object)IO handler.
amrtype(amr_object)AMR marker handler.
slicestype(slices_object)Slices handler.
blanesmoantype(blanesmoan_object)Blanes-Moan integrator.
cfmtype(cfm_object)Commutator-Free Magnus integrator.
leapfrogtype(leapfrog_object)Leapfrog integrator.
flailtype(flail_object)Linear algebra methods handler.
wenotype(weno_object)WENO reconstructor.
ibtype(ib_object)Immersed boundary.
rktype(rk_object)Runge-Kutta integrator.
adamtype(adam_object)ADAM, grid + tree + field + maps orchestrator.
fdv_schemecharacter(len=:)allocatableFDV scheme, fd/fv.
fdv_orderinteger(kind=I4P)Order of finite difference/volume schemes, general order.
fdv_half_stencilinteger(kind=I4P)Half stencil length of finite difference/volume schemes.
fdv_half_stencilsinteger(kind=I4P)Half stencil length of fdv schemes for each derivative up to 6.
ngcinteger(kind=I4P)pointerNumber of ghost cells.
niinteger(kind=I4P)pointerNumber of cells in i direction.
njinteger(kind=I4P)pointerNumber of cells in j direction.
nkinteger(kind=I4P)pointerNumber of cells in k direction.
nbinteger(kind=I4P)pointerTotal blocks number for MPI.
blocks_numberinteger(kind=I4P)pointerActual blocks number.
nvinteger(kind=I4P)pointerNumber of variables in q vector.
realm_indexinteger(kind=I4P)Inter-realm seam coupling — currently-published stage buffer key.
stage_activeinteger(kind=I4P)Procedure pointer TBPs for FDV operators (set at initialization by backend).
compute_block_total_variationprocedure(compute_block_total_variation_interface)pass(self), pointerCompute TV.
compute_curlprocedure(compute_curl_interface)pass(self), pointerCompute curl.
compute_curl_extendedprocedure(compute_curl_interface)pass(self), pointerCompute curl on extended support.
compute_derivative1procedure(compute_derivative1_interface)pass(self), pointerCompute deriv1.
compute_derivative2procedure(compute_derivative2_interface)pass(self), pointerCompute deriv2.
compute_derivative4procedure(compute_derivative4_interface)pass(self), pointerCompute deriv4.
compute_divergenceprocedure(compute_divergence_interface)pass(self), pointerCompute dive.
compute_gradientprocedure(compute_gradient_interface)pass(self), pointerCompute grad.
compute_gradient_extendedprocedure(compute_gradient_interface)pass(self), pointerCompute gradient on extended support.
compute_laplacianprocedure(compute_laplacian_interface)pass(self), pointerCompute laplac.
amr_locked_logicalAMR regrid lock (issue #22 GA6): set .true. by backends whose device/topology
nv_cinteger(kind=I4P)pointerNumber of conservative variables in q vector.
nv_sinteger(kind=I4P)pointerNumber of source variables in q vector.
nv_clinteger(kind=I4P)pointerNumber of divergence cleaning variables in q vector.
qreal(kind=R8P)allocatableConservative cell centered variables.
dqreal(kind=R8P)allocatableResiduals right hand side.
q_picreal(kind=R8P)allocatablePIC variables.
pic_fieldsreal(kind=R8P)allocatableFields value at particle locations.
curlreal(kind=R8P)allocatableCurl fields.
divergencereal(kind=R8P)allocatableDivergence fields.
q_nametype(string)allocatableFields names [1:nv].
dq_nametype(string)allocatableResiduals names [1:nv].
q_pic_nametype(string)allocatablePIC Fields names.
curl_nametype(string)allocatableCurl fields names.
div_nametype(string)allocatableDivergence fields names.
energy_Dreal(kind=R8P)allocatableEnergy of field D, time history.
energy_Breal(kind=R8P)allocatableEnergy of field B, time history.
coil_powerreal(kind=R8P)allocatablePower of coils, time history.
Poynting_fluxreal(kind=R8P)allocatableTotal Poynting flux from boundary, time history.
rms_energy_error_Dreal(kind=R8P)RMS energy error of D field.
rms_energy_error_Breal(kind=R8P)RMS energy error of B field.
max_divergence_Dreal(kind=R8P)Maximum of divergence of D field.
max_divergence_Breal(kind=R8P)Maximum of divergence of B field.
max_divergence_Jreal(kind=R8P)Maximum of divergence of J field.
bctype(prism_bc_object)Boundary conditions.
coiltype(prism_coil_object)Coil source term.
external_fieldstype(prism_external_fields_object)External fields.
fWLayertype(prism_fWLayer_object)Far-field weighted layer.
ictype(prism_ic_object)Initial conditions.
leapfrog_pictype(prism_leapfrog_pic_object)Leapfrog PIC integrator.
numericstype(prism_numerics_object)Numerics configuration.
particle_injectiontype(prism_particle_injection_object)Particle injection.
physicstype(prism_physics_object)Physics configuration.
pictype(prism_pic_object)PIC state.
rk_pictype(prism_rk_pic_object)Runge-Kutta PIC handler.
timetype(prism_time_object)Time integration state.
field_fnltype(field_fnl_object)GPU field handler.
ib_fnltype(ib_fnl_object)GPU immersed boundary.
rk_fnltype(rk_fnl_object)GPU Runge-Kutta integrator.
weno_fnltype(weno_fnl_object)GPU WENO reconstructor.
coil_fnltype(prism_fnl_coil_object)GPU coil source.
fwlayer_fnltype(prism_fnl_fwlayer_object)GPU fWLayer.
q_gpureal(kind=R8P)pointerField cell centered variables.
dq_gpureal(kind=R8P)pointerResiduals right hand side.
flxyz_c_gpureal(kind=R8P)pointerFluxes at cell center with +/- decomposition for all directions.
flx_f_gpureal(kind=R8P)pointerFluxes along x at cell face.
fly_f_gpureal(kind=R8P)pointerFluxes along y at cell face.
flz_f_gpureal(kind=R8P)pointerFluxes along z at cell face.
curl_gpureal(kind=R8P)pointerCurl fields.
divergence_gpureal(kind=R8P)pointerDivergence fields.
db5integer(kind=I4P)Device data bounds (rank 5): bb(1,:)=lower, bb(2,:)=upper.
hb5integer(kind=I4P)Host buffer data bounds (rank 5): bb(1,:)=lower, bb(2,:)=upper.
buf_5D_R8Preal(kind=R8P)allocatableBuffer (host memory, device shape), rank 5, R8P.
fv_flux_variantinteger(kind=I4P)FV Maxwell-flux variant selector.
fv_add_phi_dampinglogicalApply phi damping in FV residuals.
fv_add_psi_dampinglogicalApply psi damping in FV residuals.
fv_ivar_phiinteger(kind=I4P)Phi slot index in FV residuals.
fv_ivar_psiinteger(kind=I4P)Psi slot index in FV residuals.
fd_residual_variantinteger(kind=I4P)Centered-FD residual selector.
fd_ivar_phiinteger(kind=I4P)Phi slot index in FD residuals.
fd_ivar_psiinteger(kind=I4P)Psi slot index in FD residuals.
fd_inv_mu_scalereal(kind=R8P)Dimensional/adimensional scaling on curl(B).
fd_inv_eps_scalereal(kind=R8P)Dimensional/adimensional scaling on curl(D).
fd_chi_wavereal(kind=R8P)Hyperbolic transport speed for phi/psi equations.
fd_chi_dampreal(kind=R8P)Dedner damping speed for phi/psi equations.
compute_curl_devprocedure(compute_curl_interface_dev)pass(self), pointerCompute curl.
compute_derivative1_devprocedure(compute_derivative1_interface_dev)pass(self), pointerCompute derivative1.
compute_derivative2_devprocedure(compute_derivative2_interface_dev)pass(self), pointerCompute derivative2.
compute_derivative4_devprocedure(compute_derivative4_interface_dev)pass(self), pointerCompute derivative4.
compute_divergence_devprocedure(compute_divergence_interface_dev)pass(self), pointerCompute divergence.
compute_gradient_devprocedure(compute_gradient_interface_dev)pass(self), pointerCompute gradient.
compute_laplacian_devprocedure(compute_laplacian_interface_dev)pass(self), pointerCompute laplacian.
compute_residuals_devprocedure(compute_residuals_interface_dev)pass(self), pointerCompute residuals.
integrate_devprocedure(integrate_interface_dev)pass(self), pointerIntegrate, time operator.

Type-Bound Procedures ​

NameAttributesDescription
load_fdv_from_filepass(self)Load FDV config from file.
close_block_xh5fnopassClose XH5F file block.
close_file_xh5fnopassClose XH5F file.
open_block_xh5fpass(self)Open block file XH5F.
open_file_xh5fpass(self)Open file XH5F.
save_q_xh5fpass(self)Save in XH5F (XDMF/HDF5) format.
compute_curl_fd_extendedpass(self)Compute curl of vector field, finite difference, even for half gcs.
compute_gradient_fd_extendedpass(self)Compute gradient of scalar field, finite difference, even for half gcs.
amr_updatepass(self)Do AMR update (init-time only where amr_locked_).
mark_by_geometrypass(self)Mark blocks to be refined by a primitive geometric box.
mark_by_j_vec_total_variationpass(self)Mark by j_vec total variation (backend override; common
allocate_commonpass(self)Allocate common data.
compute_auxiliary_fieldspass(self)Compute auxiliary fields.
initializepass(self)Initialize the equation common data.
save_energy_errorpass(self)Save energy error history.
save_energy_historypass(self)Save energy history.
save_divergence_historypass(self)Save divergence history.
save_restart_filespass(self)Save restart files.
save_xh5fpass(self)Save simulation data in XH5F format.
initialize_coilspass(self)Initialize coils.
set_rectangular_coil_xpass(self)Subroutine to set a rectangular coil source with +-x normal
set_rectangular_coil_ypass(self)Subroutine to set a rectangular coil source with +-y normal
set_rectangular_coil_zpass(self)Subroutine to set a rectangular coil source with +-z normal
set_circular_coil_xpass(self)Subroutine to set a circular coil source with +-x normal
set_circular_coil_ypass(self)Subroutine to set a circular coil source with +-y normal
set_circular_coil_zpass(self)Subroutine to set a circular coil source with +-z normal
set_solenoid_xpass(self)Subroutine to set a solenoid source with +-x normal
set_solenoid_ypass(self)Subroutine to set a solenoid source with +-y normal
set_solenoid_zpass(self)Subroutine to set a solenoid source with +-z normal
set_helicon_coilpass(self)Subroutine to set a helicon coil source.
coupling_descriptor_forestpass(self)Report (scheme_time, rk_scheme, nv) for β admissibility.
allocate_gpupass(self)Allocate GPU data.
copy_cpu_gpupass(self)Copy data from CPU to GPU.
copy_gpu_cpupass(self)Copy data from GPU to CPU.
initialize_prismpass(self)Initialize PRISM equation.
load_restart_filespass(self)Load restart files.
save_residualspass(self)Save residuals history.
save_simulation_datapass(self)Save all simulation data.
apply_fwl_correctionpass(self)Apply fWLayer correction (if present)
compute_coils_currentpass(self)Compute current coils sources.
set_boundary_conditionspass(self)Set boundary conditions of equation.
set_initial_conditionspass(self)Set initial conditions of equation.
update_ghostpass(self)Update ghost cells and set boundary conditions.
update_rk_ghostpass(self)Update RK stage ghost cells.
compute_curl_fd_devpass(self)Compute curl of vector field by finite difference.
compute_curl_fv_devpass(self)Compute curl of vector field by finite volume.
compute_derivative1_fd_devpass(self)Compute derivative1 of scalar fields, finite difference schemes.
compute_derivative1_fv_devpass(self)Compute derivative1 of scalar fields, finite volume schemes.
compute_derivative2_fd_devpass(self)Compute derivative2 of scalar fields, finite difference schemes.
compute_derivative2_fv_devpass(self)Compute derivative2 of scalar fields, finite volume schemes.
compute_derivative4_fd_devpass(self)Compute derivative4 of scalar fields, finite difference schemes.
compute_divergence_fd_devpass(self)Compute divergence of vector field by finite difference.
compute_divergence_fv_devpass(self)Compute divergence of vector field by finite volume.
compute_gradient_fd_devpass(self)Compute gradient of scalar field, finite difference schemes.
compute_gradient_fv_devpass(self)Compute gradient of scalar field, finite volume schemes.
compute_laplacian_fd_devpass(self)Compute laplacian of scalar field, finite difference schemes.
compute_laplacian_fv_devpass(self)Compute laplacian of scalar field, finite volume schemes.
compute_residuals_fd_centered_devpass(self)Compute residuals, centered finite difference schemes.
integrate_blanesmoan_devpass(self)Blanes and Moan scheme.
integrate_cfm_devpass(self)Commutator-Free Magnus scheme.
integrate_leapfrog_devpass(self)Leapfrog scheme.
integrate_rk_ls_devpass(self)RK classical low storage schemes.
integrate_rk_ssp_devpass(self)SSP RK schemes.
integrate_rk_yoshida_devpass(self)Yoshida schemes.
initialize_forestpass(self)Invoked by forest%initialize per realm at startup.
compute_local_dt_forestpass(self)Invoked by forest%compute_global_dt during the min reduction.
advance_one_step_forestpass(self)Invoked by forest%evolve_one_step per realm per timestep.
stages_per_step_forestpass(self)Number of integrator stages this realm exposes per step.
open_step_forestpass(self)Per-step prologue (multi-realm path).
begin_stage_forestpass(self)Begin one integrator stage on this realm (multi-realm path).
end_stage_forestpass(self)End the stage: residuals + assignment (multi-realm path).
close_step_forestpass(self)Per-step epilogue (multi-realm path).
post_step_forestpass(self)Invoked by forest%post_step per realm per timestep.
is_done_forestpass(self)Invoked by forest%is_done during the termination reduction.
finalize_forestpass(self)Invoked by forest%finalize per realm at shutdown.
finalize_mpi_forestpass(self)Process-global MPI finalize (mpih_fnl).
fill_seam_from_peer_forestpass(self)OpenACC device-direct copy of peer's GPU interior into self's GPU ghosts.
after_topology_build_forestpass(self)Refresh device-resident seam maps from freshly-built host maps.
apply_reflux_to_stage_forestpass(self)Apply Berger-Colella reflux to self's RK stage buffer (FNL no-op).
compute_dtpass(self)Compute time step.
compute_energypass(self)Compute energy.
compute_energy_errorpass(self)Compute energy error.
compute_max_divergencepass(self)Compute divergence of D, B and J fields for diagnostics.
impose_ct_correctionpass(self)Impose Constrained Transport correction on q(ivar:ivar+2).
impose_div_freepass(self)Impose divergence-free property.
simulatepass(self)Perform the simulation.

Interfaces ​

compute_curl_interface_dev ​

compute_derivative1_interface_dev ​

compute_derivative2_interface_dev ​

compute_derivative4_interface_dev ​

compute_divergence_interface_dev ​

compute_gradient_interface_dev ​

compute_laplacian_interface_dev ​

compute_residuals_interface_dev ​

integrate_interface_dev ​

Subroutines ​

allocate_gpu ​

Allocate GPU data.

fortran
subroutine allocate_gpu(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

Call graph

copy_cpu_gpu ​

Copy data from CPU to GPU.

fortran
subroutine copy_cpu_gpu(self, verbose)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
verboselogicalinoptionalTrigger verbose output.

Call graph

copy_gpu_cpu ​

Copy data from GPU to CPU.

fortran
subroutine copy_gpu_cpu(self, compute_copy_q_aux, copy_phi, verbose)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
compute_copy_q_auxlogicalinoptionalFlag to compute auxiliary variables.
copy_philogicalinoptionalCopy also phi.
verboselogicalinoptionalTrigger verbose output.

Call graph

initialize_prism ​

Initialize PRISM equation.

fortran
subroutine initialize_prism(self, filename, realms_number)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inouttargetThe equation.
filenamecharacter(len=*)inInput file name.
realms_numberinteger(kind=I4P)inoptionalForest realm count; divides the per-device budget

Call graph

load_restart_files ​

Save restart files.

fortran
subroutine load_restart_files(self, t, time)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
tinteger(kind=I4P)outTime iteration.
timereal(kind=R8P)outTime.

Call graph

save_residuals ​

Save residuals history.

fortran
subroutine save_residuals(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

Call graph

save_simulation_data ​

Save all simulation data.

fortran
subroutine save_simulation_data(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

Call graph

apply_fwl_correction ​

Apply correction if a fWL is present.

fortran
subroutine apply_fwl_correction(self, q_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
q_gpureal(kind=R8P)inoutConservative variables.

Call graph

compute_coils_current ​

Compute current coils sources.

fortran
subroutine compute_coils_current(self, q_gpu, gamm)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
q_gpureal(kind=R8P)inoutConservative variables.
gammreal(kind=R8P)inoptionalRK coefficient.

Call graph

set_boundary_conditions ​

Set boundary conditions of equation.

fortran
subroutine set_boundary_conditions(self, q_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
q_gpureal(kind=R8P)inoutConservative variables.

Call graph

set_initial_conditions ​

Set initial conditions of field.

fortran
subroutine set_initial_conditions(self, is_restart)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
is_restartlogicalinBranching sentinel for restart/non restart path.

Call graph

update_ghost ​

Update ghost cells (intra-realm only).

Inter-realm seam ghost cells are filled by the forest's Phase 2 seam exchange (between assemble and evaluate) via fill_seam_from_peer_forest, NOT here. The legacy realm(:) optional dummy and the exchange_inter_realm_halos_forest call have been retired by the agnostic-dummy seam redesign.

fortran
subroutine update_ghost(self, q_gpu, step, s)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
q_gpureal(kind=R8P)inoutConservative variables.
stepinteger(kind=I4P)inoptionalStep to be performed in async comp.
sinteger(kind=I4P)inoptionalStage counter.

Call graph

update_rk_ghost ​

Update RK ghost cells.

fortran
subroutine update_rk_ghost(self, dt, phi_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutRK object.
dtreal(kind=R8P)inCurrent time step.
phi_gpureal(kind=R8P)inoptionalIB distance.

compute_curl_fd_dev ​

Compute curl of vector fields, div(q(ivar:ivar+2), using finite difference schemes.

fortran
subroutine compute_curl_fd_dev(self, ivar, q_gpu, curl_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
ivarinteger(kind=I4P)inStart index of variable of q.
q_gpureal(kind=R8P)inField variables.
curl_gpureal(kind=R8P)inoutCurl.

Call graph

compute_curl_fv_dev ​

Compute curl of vector fields, div(q(ivar:ivar+2), using finite volume schemes.

fortran
subroutine compute_curl_fv_dev(self, ivar, q_gpu, curl_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
ivarinteger(kind=I4P)inStart index of variable of q.
q_gpureal(kind=R8P)inField variables.
curl_gpureal(kind=R8P)inoutCurl.

compute_derivative1_fd_dev ​

Compute derivative1 of scalar fields, dq(ivar)/ds, using finite difference schemes.

fortran
subroutine compute_derivative1_fd_dev(self, dir, ivar, q_gpu, dq_ds_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
dirinteger(kind=I4P)inDirection, 1=X, 2=Y, 3=Z.
ivarinteger(kind=I4P)inStart index of variable of q.
q_gpureal(kind=R8P)inField variables.
dq_ds_gpureal(kind=R8P)inoutDerivative1, dq/ds.

compute_derivative1_fv_dev ​

Compute derivative1 of scalar fields, dq(ivar)/ds, using finite volume schemes.

fortran
subroutine compute_derivative1_fv_dev(self, dir, ivar, q_gpu, dq_ds_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
dirinteger(kind=I4P)inDirection, 1=X, 2=Y, 3=Z.
ivarinteger(kind=I4P)inStart index of variable of q.
q_gpureal(kind=R8P)inField variables.
dq_ds_gpureal(kind=R8P)inoutDerivative1, dq/ds.

compute_derivative2_fd_dev ​

Compute derivative2 of scalar fields, d2q(ivar)/ds2, using finite difference schemes.

fortran
subroutine compute_derivative2_fd_dev(self, dir, ivar, q_gpu, d2q_ds2_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
dirinteger(kind=I4P)inDirection, 1=X, 2=Y, 3=Z.
ivarinteger(kind=I4P)inStart index of variable of q.
q_gpureal(kind=R8P)inField variables.
d2q_ds2_gpureal(kind=R8P)inoutDerivative2, d2q/ds2.

compute_derivative2_fv_dev ​

Compute derivative2 of scalar fields, d2q(ivar)/ds2, using finite volume schemes.

fortran
subroutine compute_derivative2_fv_dev(self, dir, ivar, q_gpu, d2q_ds2_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
dirinteger(kind=I4P)inDirection, 1=X, 2=Y, 3=Z.
ivarinteger(kind=I4P)inStart index of variable of q.
q_gpureal(kind=R8P)inField variables.
d2q_ds2_gpureal(kind=R8P)inoutDerivative2, d2q/ds2.

compute_derivative4_fd_dev ​

Compute derivative4 of scalar fields, d4q(ivar)/ds4, using finite difference schemes.

fortran
subroutine compute_derivative4_fd_dev(self, dir, ivar, q_gpu, d4q_ds4_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
dirinteger(kind=I4P)inDirection, 1=X, 2=Y, 3=Z.
ivarinteger(kind=I4P)inStart index of variable of q.
q_gpureal(kind=R8P)inField variables.
d4q_ds4_gpureal(kind=R8P)inoutDerivative4, d4q/ds4.

compute_divergence_fd_dev ​

Compute divergence of vector fields, div(q(ivar:ivar+2), using finite difference schemes. Directly computes divergence from transposed GPU layout (b,i,j,k,v).

fortran
subroutine compute_divergence_fd_dev(self, ivar, ovar, q_gpu, divergence_gpu, maxdiv)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
ivarinteger(kind=I4P)inStart index of
ovarinteger(kind=I4P)inOutput index in
q_gpureal(kind=R8P)inField variables.
divergence_gpureal(kind=R8P)inoutDivergence.
maxdivreal(kind=R8P)outoptionalMax divergence,

Call graph

compute_divergence_fv_dev ​

Compute divergence of vector fields, div(q(ivar:ivar+2), using finite volume schemes. Directly computes divergence from transposed GPU layout (b,i,j,k,v).

fortran
subroutine compute_divergence_fv_dev(self, ivar, ovar, q_gpu, divergence_gpu, maxdiv)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
ivarinteger(kind=I4P)inStart index of
ovarinteger(kind=I4P)inOutput index in
q_gpureal(kind=R8P)inField variables.
divergence_gpureal(kind=R8P)inoutDivergence.
maxdivreal(kind=R8P)outoptionalMax divergence,

compute_gradient_fd_dev ​

Compute gradient of scalar variable q(ivar), finite difference schemes.

fortran
subroutine compute_gradient_fd_dev(self, ivar, q_gpu, gradient_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
ivarinteger(kind=I4P)inIndex of scalar var of q.
q_gpureal(kind=R8P)inField variables.
gradient_gpureal(kind=R8P)inoutGradient.

compute_gradient_fv_dev ​

Compute gradient of scalar variable q(ivar), finite volume schemes.

fortran
subroutine compute_gradient_fv_dev(self, ivar, q_gpu, gradient_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
ivarinteger(kind=I4P)inIndex of scalar var of q.
q_gpureal(kind=R8P)inField variables.
gradient_gpureal(kind=R8P)inoutGradient.

compute_laplacian_fd_dev ​

Compute laplacian of scalar variable q(ivar), finite difference schemes.

fortran
subroutine compute_laplacian_fd_dev(self, ivar, q_gpu, laplacian_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
ivarinteger(kind=I4P)inIndex of scalar variable of q.
q_gpureal(kind=R8P)inField variables.
laplacian_gpureal(kind=R8P)inoutGradient.

compute_laplacian_fv_dev ​

Compute laplacian of scalar variable q(ivar), finite volume schemes.

fortran
subroutine compute_laplacian_fv_dev(self, ivar, q_gpu, laplacian_gpu)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe equation.
ivarinteger(kind=I4P)inIndex of scalar variable of q.
q_gpureal(kind=R8P)inField variables.
laplacian_gpureal(kind=R8P)inoutGradient.

compute_residuals_fd_centered_dev ​

Compute residuals of equation, space operator, centered finite difference schemes.

fortran
subroutine compute_residuals_fd_centered_dev(self, q_gpu, dq_gpu, s, flux_register)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
q_gpureal(kind=R8P)inoutConservative variables.
dq_gpureal(kind=R8P)inoutResiduals.
sinteger(kind=I4P)inoptionalStage counter.
flux_registerclass(flux_register_object)inoutoptionalForest's flux register (interface parity; the

Call graph

fd_centered_plain_dev_kernel ​

Centered-FD Maxwell residual without hyperbolic cleaning.

fortran
subroutine fd_centered_plain_dev_kernel(ni, nj, nk, ngc, blocks_number, var_jx, var_jy, var_jz, s1, inv_mu_scale, inv_eps_scale, dxyz_gpu, q_gpu, dq_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
var_jxinteger(kind=I4P)in
var_jyinteger(kind=I4P)in
var_jzinteger(kind=I4P)in
s1integer(kind=I4P)in
inv_mu_scalereal(kind=R8P)in
inv_eps_scalereal(kind=R8P)in
dxyz_gpureal(kind=R8P)in
q_gpureal(kind=R8P)in
dq_gpureal(kind=R8P)inout

Call graph

fd_centered_phi_dev_kernel ​

Centered-FD Maxwell residual with phi cleaning only.

fortran
subroutine fd_centered_phi_dev_kernel(ni, nj, nk, ngc, blocks_number, var_jx, var_jy, var_jz, s1, inv_mu_scale, inv_eps_scale, chi_wave, chi_damp, c_r, ivar_phi, dxyz_gpu, q_gpu, dq_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
var_jxinteger(kind=I4P)in
var_jyinteger(kind=I4P)in
var_jzinteger(kind=I4P)in
s1integer(kind=I4P)in
inv_mu_scalereal(kind=R8P)in
inv_eps_scalereal(kind=R8P)in
chi_wavereal(kind=R8P)in
chi_dampreal(kind=R8P)in
c_rreal(kind=R8P)in
ivar_phiinteger(kind=I4P)in
dxyz_gpureal(kind=R8P)in
q_gpureal(kind=R8P)in
dq_gpureal(kind=R8P)inout

Call graph

fd_centered_psi_dev_kernel ​

Centered-FD Maxwell residual with psi cleaning only.

fortran
subroutine fd_centered_psi_dev_kernel(ni, nj, nk, ngc, blocks_number, var_jx, var_jy, var_jz, s1, inv_mu_scale, inv_eps_scale, chi_wave, chi_damp, c_r, ivar_psi, dxyz_gpu, q_gpu, dq_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
var_jxinteger(kind=I4P)in
var_jyinteger(kind=I4P)in
var_jzinteger(kind=I4P)in
s1integer(kind=I4P)in
inv_mu_scalereal(kind=R8P)in
inv_eps_scalereal(kind=R8P)in
chi_wavereal(kind=R8P)in
chi_dampreal(kind=R8P)in
c_rreal(kind=R8P)in
ivar_psiinteger(kind=I4P)in
dxyz_gpureal(kind=R8P)in
q_gpureal(kind=R8P)in
dq_gpureal(kind=R8P)inout

Call graph

fd_centered_phi_psi_dev_kernel ​

Centered-FD Maxwell residual with phi/psi cleaning.

fortran
subroutine fd_centered_phi_psi_dev_kernel(ni, nj, nk, ngc, blocks_number, var_jx, var_jy, var_jz, s1, inv_mu_scale, inv_eps_scale, chi_wave, chi_damp, c_r, ivar_phi, ivar_psi, dxyz_gpu, q_gpu, dq_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
var_jxinteger(kind=I4P)in
var_jyinteger(kind=I4P)in
var_jzinteger(kind=I4P)in
s1integer(kind=I4P)in
inv_mu_scalereal(kind=R8P)in
inv_eps_scalereal(kind=R8P)in
chi_wavereal(kind=R8P)in
chi_dampreal(kind=R8P)in
c_rreal(kind=R8P)in
ivar_phiinteger(kind=I4P)in
ivar_psiinteger(kind=I4P)in
dxyz_gpureal(kind=R8P)in
q_gpureal(kind=R8P)in
dq_gpureal(kind=R8P)inout

Call graph

compute_residuals_fv_centered_dev ​

Compute residuals, space operator, centered finite volume scheme — FNL device backend (issue #23 R2). 1:1 structural mirror of prism_cpu_object%compute_residuals_fv_centered: ghost refresh on q, pointwise Maxwell fluxes at ALL cells (incl. ghosts) into flxyz_c_gpu, three staggered face-reconstruction sweeps into fl{x,y,z}_f_gpu (the m=0 SOTA primitive, already acc routine seq), flux difference + J source into dq_gpu. The Maxwell-flux variant is selected ONCE at host side, then a specialized device kernel runs with no per-cell variant branch. Inter-realm seam accumulation (issue #23 R3): after the face sweeps, at the realm's FINAL RK substage only (α.r1, CPU parity), the seam face skins are device-packed, D2H-copied (tiny slabs) and accumulated into the HOST-side flux register — see accumulate_seam_fluxes_fv_dev.

Race discipline (CLAUDE-gpu): the kernels below are MODULE-LEVEL procedures (not contained here) with constant-bound private gathers — the certified #22 F2/F3 pattern.

fortran
subroutine compute_residuals_fv_centered_dev(self, q_gpu, dq_gpu, s, flux_register)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
q_gpureal(kind=R8P)inoutConservative variables.
dq_gpureal(kind=R8P)inoutResiduals.
sinteger(kind=I4P)inoptionalStage counter (gates the seam-flux accumulation).
flux_registerclass(flux_register_object)inoutoptionalForest's flux register for FV seam reflux.

Call graph

fv_cell_fluxes_maxwell_dev_kernel ​

Pointwise plain dimensional Maxwell fluxes at every cell (incl. ghosts).

fortran
subroutine fv_cell_fluxes_maxwell_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, chi, q_gpu, flxyz_c_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
chireal(kind=R8P)in
q_gpureal(kind=R8P)in
flxyz_c_gpureal(kind=R8P)inout

Call graph

fv_cell_fluxes_maxwell_adim_dev_kernel ​

Pointwise plain adimensional Maxwell fluxes at every cell (incl. ghosts).

fortran
subroutine fv_cell_fluxes_maxwell_adim_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, chi, q_gpu, flxyz_c_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
chireal(kind=R8P)in
q_gpureal(kind=R8P)in
flxyz_c_gpureal(kind=R8P)inout

Call graph

fv_cell_fluxes_maxwell_div_d_dev_kernel ​

Pointwise dimensional Maxwell fluxes with D hyperbolic cleaning.

fortran
subroutine fv_cell_fluxes_maxwell_div_d_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, chi, q_gpu, flxyz_c_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
chireal(kind=R8P)in
q_gpureal(kind=R8P)in
flxyz_c_gpureal(kind=R8P)inout

Call graph

fv_cell_fluxes_maxwell_div_b_dev_kernel ​

Pointwise dimensional Maxwell fluxes with B hyperbolic cleaning.

fortran
subroutine fv_cell_fluxes_maxwell_div_b_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, chi, q_gpu, flxyz_c_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
chireal(kind=R8P)in
q_gpureal(kind=R8P)in
flxyz_c_gpureal(kind=R8P)inout

Call graph

fv_cell_fluxes_maxwell_div_d_b_dev_kernel ​

Pointwise dimensional Maxwell fluxes with D/B hyperbolic cleaning.

fortran
subroutine fv_cell_fluxes_maxwell_div_d_b_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, chi, q_gpu, flxyz_c_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
chireal(kind=R8P)in
q_gpureal(kind=R8P)in
flxyz_c_gpureal(kind=R8P)inout

Call graph

fv_cell_fluxes_maxwell_adim_div_d_dev_kernel ​

Pointwise adimensional Maxwell fluxes with D hyperbolic cleaning.

fortran
subroutine fv_cell_fluxes_maxwell_adim_div_d_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, chi, q_gpu, flxyz_c_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
chireal(kind=R8P)in
q_gpureal(kind=R8P)in
flxyz_c_gpureal(kind=R8P)inout

Call graph

fv_cell_fluxes_maxwell_adim_div_b_dev_kernel ​

Pointwise adimensional Maxwell fluxes with B hyperbolic cleaning.

fortran
subroutine fv_cell_fluxes_maxwell_adim_div_b_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, chi, q_gpu, flxyz_c_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
chireal(kind=R8P)in
q_gpureal(kind=R8P)in
flxyz_c_gpureal(kind=R8P)inout

Call graph

fv_cell_fluxes_maxwell_adim_div_d_b_dev_kernel ​

Pointwise adimensional Maxwell fluxes with D/B hyperbolic cleaning.

fortran
subroutine fv_cell_fluxes_maxwell_adim_div_d_b_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, chi, q_gpu, flxyz_c_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
chireal(kind=R8P)in
q_gpureal(kind=R8P)in
flxyz_c_gpureal(kind=R8P)inout

Call graph

fv_recon_x_dev_kernel ​

Reconstruct x-fluxes at x-faces from cell-center fluxes (issue #23 R2): per face i+1/2 (i = 0..ni), gather the 2*s1 cell-flux stencil into a constant-bound private buffer and call the m=0 SOTA face-reconstruction primitive (already acc routine seq).

fortran
subroutine fv_recon_x_dev_kernel(s1, ni, nj, nk, ngc, nv_c, blocks_number, flxyz_c_gpu, flx_f_gpu)

Arguments

NameTypeIntentAttributesDescription
s1integer(kind=I4P)inHalf FDV stencil length.
niinteger(kind=I4P)inGrids dimensions.
njinteger(kind=I4P)inGrids dimensions.
nkinteger(kind=I4P)inGrids dimensions.
ngcinteger(kind=I4P)inGrids dimensions.
nv_cinteger(kind=I4P)inConservative variables number.
blocks_numberinteger(kind=I4P)inGrids dimensions.
flxyz_c_gpureal(kind=R8P)inCell-center fluxes (b,1,d,i,j,k,v).
flx_f_gpureal(kind=R8P)inoutX-face fluxes (b,0:ni,j,k,v).

Call graph

fv_recon_y_dev_kernel ​

Reconstruct y-fluxes at y-faces (issue #23 R2); twin of fv_recon_x_dev_kernel.

fortran
subroutine fv_recon_y_dev_kernel(s1, ni, nj, nk, ngc, nv_c, blocks_number, flxyz_c_gpu, fly_f_gpu)

Arguments

NameTypeIntentAttributesDescription
s1integer(kind=I4P)inHalf FDV stencil length.
niinteger(kind=I4P)inGrids dimensions.
njinteger(kind=I4P)inGrids dimensions.
nkinteger(kind=I4P)inGrids dimensions.
ngcinteger(kind=I4P)inGrids dimensions.
nv_cinteger(kind=I4P)inConservative variables number.
blocks_numberinteger(kind=I4P)inGrids dimensions.
flxyz_c_gpureal(kind=R8P)inCell-center fluxes (b,1,d,i,j,k,v).
fly_f_gpureal(kind=R8P)inoutY-face fluxes (b,i,0:nj,k,v).

Call graph

fv_recon_z_dev_kernel ​

Reconstruct z-fluxes at z-faces (issue #23 R2); twin of fv_recon_x_dev_kernel.

fortran
subroutine fv_recon_z_dev_kernel(s1, ni, nj, nk, ngc, nv_c, blocks_number, flxyz_c_gpu, flz_f_gpu)

Arguments

NameTypeIntentAttributesDescription
s1integer(kind=I4P)inHalf FDV stencil length.
niinteger(kind=I4P)inGrids dimensions.
njinteger(kind=I4P)inGrids dimensions.
nkinteger(kind=I4P)inGrids dimensions.
ngcinteger(kind=I4P)inGrids dimensions.
nv_cinteger(kind=I4P)inConservative variables number.
blocks_numberinteger(kind=I4P)inGrids dimensions.
flxyz_c_gpureal(kind=R8P)inCell-center fluxes (b,1,d,i,j,k,v).
flz_f_gpureal(kind=R8P)inoutZ-face fluxes (b,i,j,0:nk,v).

Call graph

fv_flux_diff_plain_dev_kernel ​

Conservative flux difference + J source, no hyperbolic damping fields.

fortran
subroutine fv_flux_diff_plain_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, var_jx, var_jy, var_jz, dxyz_gpu, flx_f_gpu, fly_f_gpu, flz_f_gpu, q_gpu, dq_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
var_jxinteger(kind=I4P)in
var_jyinteger(kind=I4P)in
var_jzinteger(kind=I4P)in
dxyz_gpureal(kind=R8P)in
flx_f_gpureal(kind=R8P)in
fly_f_gpureal(kind=R8P)in
flz_f_gpureal(kind=R8P)in
q_gpureal(kind=R8P)in
dq_gpureal(kind=R8P)inout

Call graph

fv_flux_diff_phi_dev_kernel ​

Conservative flux difference + J source + phi damping.

fortran
subroutine fv_flux_diff_phi_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, var_jx, var_jy, var_jz, chi_damp, c_r, fv_ivar_phi, dxyz_gpu, flx_f_gpu, fly_f_gpu, flz_f_gpu, q_gpu, dq_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
var_jxinteger(kind=I4P)in
var_jyinteger(kind=I4P)in
var_jzinteger(kind=I4P)in
chi_dampreal(kind=R8P)in
c_rreal(kind=R8P)in
fv_ivar_phiinteger(kind=I4P)in
dxyz_gpureal(kind=R8P)in
flx_f_gpureal(kind=R8P)in
fly_f_gpureal(kind=R8P)in
flz_f_gpureal(kind=R8P)in
q_gpureal(kind=R8P)in
dq_gpureal(kind=R8P)inout

Call graph

fv_flux_diff_psi_dev_kernel ​

Conservative flux difference + J source + psi damping.

fortran
subroutine fv_flux_diff_psi_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, var_jx, var_jy, var_jz, chi_damp, c_r, fv_ivar_psi, dxyz_gpu, flx_f_gpu, fly_f_gpu, flz_f_gpu, q_gpu, dq_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
var_jxinteger(kind=I4P)in
var_jyinteger(kind=I4P)in
var_jzinteger(kind=I4P)in
chi_dampreal(kind=R8P)in
c_rreal(kind=R8P)in
fv_ivar_psiinteger(kind=I4P)in
dxyz_gpureal(kind=R8P)in
flx_f_gpureal(kind=R8P)in
fly_f_gpureal(kind=R8P)in
flz_f_gpureal(kind=R8P)in
q_gpureal(kind=R8P)in
dq_gpureal(kind=R8P)inout

Call graph

fv_flux_diff_phi_psi_dev_kernel ​

Conservative flux difference + J source + phi/psi damping.

fortran
subroutine fv_flux_diff_phi_psi_dev_kernel(ni, nj, nk, ngc, nv_c, blocks_number, var_jx, var_jy, var_jz, chi_damp, c_r, fv_ivar_phi, fv_ivar_psi, dxyz_gpu, flx_f_gpu, fly_f_gpu, flz_f_gpu, q_gpu, dq_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)in
njinteger(kind=I4P)in
nkinteger(kind=I4P)in
ngcinteger(kind=I4P)in
nv_cinteger(kind=I4P)in
blocks_numberinteger(kind=I4P)in
var_jxinteger(kind=I4P)in
var_jyinteger(kind=I4P)in
var_jzinteger(kind=I4P)in
chi_dampreal(kind=R8P)in
c_rreal(kind=R8P)in
fv_ivar_phiinteger(kind=I4P)in
fv_ivar_psiinteger(kind=I4P)in
dxyz_gpureal(kind=R8P)in
flx_f_gpureal(kind=R8P)in
fly_f_gpureal(kind=R8P)in
flz_f_gpureal(kind=R8P)in
q_gpureal(kind=R8P)in
dq_gpureal(kind=R8P)inout

Call graph

accumulate_seam_fluxes_fv_dev ​

Accumulate end-of-step FV seam face fluxes into the forest's flux register — FNL twin of prism_cpu_object%accumulate_seam_fluxes_fv (issue #23 R3).

Register crossing (grilled R0 default): the flux register stays HOST-side. Per seam (block, fec) hit, a tiny device kernel packs the face skin into a contiguous device slab (inner axis fastest — the register's cell order), ONE D2H copy (~nv*nface_cells doubles, tens of KB per step) brings it to the host, and the shared register machinery does the rest: coarse side → accumulate_coarse_flux directly; fine side → quadrant offset from maps%amr_seam_quadrant (Morton-code precompute, issue #28 D2) + the pure 2:1 restriction restrict_fine_face_to_quadrant (moved to adam_flux_register_object, shared with the CPU backend) → accumulate_fine_flux. Fires once per realm per step (final-substage gate at the call site). Third-axis register index hardcoded to 1 (α.r1 collapsed register), exactly as on CPU.

fortran
subroutine accumulate_seam_fluxes_fv_dev(self, flux_register)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.
flux_registerclass(flux_register_object)inoutForest's flux register.

Call graph

fv_pack_face_skin_dev_kernel ​

Pack one block's face flux skin into a contiguous device slab (issue #23 R3): skin_gpu(c, v) with c = (outer-1)*inner_n + inner — the register cell order, identical to the CPU pack (pack_coarse_face / fine_face_cell mappings). Scalar element reads/writes only: no private arrays, no section actuals. The fec branch is uniform per launch (firstprivate scalar). Tiny kernel: one launch per seam (block, fec) hit, once per step.

fortran
subroutine fv_pack_face_skin_dev_kernel(fec, ni, nj, nk, nv_c, b, inner_n, outer_n, flx_f_gpu, fly_f_gpu, flz_f_gpu, skin_gpu)

Arguments

NameTypeIntentAttributesDescription
fecinteger(kind=I4P)inFace code 1..6 (-x,+x,-y,+y,-z,+z).
niinteger(kind=I4P)inInterior cell counts.
njinteger(kind=I4P)inInterior cell counts.
nkinteger(kind=I4P)inInterior cell counts.
nv_cinteger(kind=I4P)inConservative variables number.
binteger(kind=I4P)inBlock index.
inner_ninteger(kind=I4P)inTangential cell counts (inner fastest).
outer_ninteger(kind=I4P)inTangential cell counts (inner fastest).
flx_f_gpureal(kind=R8P)inX-face fluxes.
fly_f_gpureal(kind=R8P)inY-face fluxes.
flz_f_gpureal(kind=R8P)inZ-face fluxes.
skin_gpureal(kind=R8P)inoutPacked skin (inner_n*outer_n, nv_c).

Call graph

fv_apply_reflux_face_dev_kernel ​

Apply the Berger-Colella end-of-step correction for ONE register face to the committed q_gpu (issue #23 R4): per skin cell c (register cell order — the SAME (axis, sgn) → (i,j,k) mapping as the CPU apply), add scale * delta(v, c) to every state row. Scalar element ops only: no private arrays, no section actuals; each c writes a distinct cell (disjoint), the inner v loop is per-thread seq. Tiny kernel — one launch per face per step.

fortran
subroutine fv_apply_reflux_face_dev_kernel(axis, sgn, b, ni, nj, nk, ngc, nv_reg, nface_cells, scale, delta_gpu, q_gpu)

Arguments

NameTypeIntentAttributesDescription
axisinteger(kind=I4P)inFace normal axis (1..3) and sign (+-1).
sgninteger(kind=I4P)inFace normal axis (1..3) and sign (+-1).
binteger(kind=I4P)inCoarse block index.
niinteger(kind=I4P)inGrids dimensions.
njinteger(kind=I4P)inGrids dimensions.
nkinteger(kind=I4P)inGrids dimensions.
ngcinteger(kind=I4P)inGrids dimensions.
nv_reginteger(kind=I4P)inRegister state-vector width.
nface_cellsinteger(kind=I4P)inSkin cell count.
scalereal(kind=R8P)insign * dt / dx_coarse.
delta_gpureal(kind=R8P)inFlux mismatch slab (nv_reg, nface_cells).
q_gpureal(kind=R8P)inoutCommitted conservative variables.

Call graph

integrate_blanesmoan_dev ​

Integrate equation, time operator, Blanes and Moan scheme.

fortran
subroutine integrate_blanesmoan_dev(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

integrate_cfm_dev ​

Integrate equation, time operator, Commutator-Free Magnus integrator.

fortran
subroutine integrate_cfm_dev(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

integrate_leapfrog_dev ​

Integrate equation, time operator, leapfrog scheme.

fortran
subroutine integrate_leapfrog_dev(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

integrate_leapfrog_pic ​

Integrate equation, time operator, leapfrog scheme.

fortran
subroutine integrate_leapfrog_pic(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

integrate_rk_ls_dev ​

Integrate equation, time operator, RK classical low storage schemes. Low storage RK working on q_rk(:,:,:,:,:,1)/q as stages, update q in place.

fortran
subroutine integrate_rk_ls_dev(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

Call graph

integrate_rk_ssp_dev ​

Integrate equation, time operator, SSP RK schemes. SSP RK working on q_rk as stages.

fortran
subroutine integrate_rk_ssp_dev(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

Call graph

integrate_rk_yoshida_dev ​

Integrate equation, time operator, Yoshida RK scheme.

fortran
subroutine integrate_rk_yoshida_dev(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

is_done_forest ​

Decide whether this realm has reached its local termination criterion.

Invoked by forest%is_done. PRISM-FNL override: matches the legacy condition inline in simulate — terminate when either the simulated time has reached self%time%time_max (time-driven mode, it_max <= 0) or the iteration count has reached self%time%it_max (iteration-driven mode). Today the test reads time-state from the time module singleton, so self is unused; once the forest takes over time bookkeeping the body will consume self%time%* instead.

fortran
subroutine is_done_forest(self, done)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe realm.
donelogicaloutTrue if this realm is done evolving.

Call graph

finalize_forest ​

Shut this realm down: final state dump, close residuals file, finalize MPI handler.

fortran
subroutine finalize_forest(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.

Call graph

finalize_mpi_forest ​

Finalize the process-global FNL MPI handler (mpih_fnl).

Overrides realm_object%finalize_mpi_forest to target the FNL mpih_fnl singleton instead of the CPU mpih. Called ONCE by forest%finalize after every realm's finalize_forest (see the base for the rationale).

fortran
subroutine finalize_mpi_forest(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm (carries no MPI state).

Call graph

simulate ​

Perform the simulation: legacy single-realm entry point.

fortran
subroutine simulate(self, filename)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
filenamecharacter(len=*)inInput file name.

Call graph

initialize_forest ​

Initialize this realm from scratch: PRISM init, IC injection (or restart load), initial ghost update, initial diagnostics dump, IO files open, plus PIC/leapfrog priming if those schemes are active.

Invoked by forest%initialize. The forest writes self%realm_index = is BEFORE calling this routine, so the body can already read the 1-based forest position from self%realm_index if needed.

fortran
subroutine initialize_forest(self, filename, realms_number, memory_avail, nv, verbose)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.
filenamecharacter(len=*)inInput parameters file name.
realms_numberinteger(kind=I4P)inoptionalRealm count; divides the per-device budget.
memory_availreal(kind=R8P)inoptionalPer-process memory budget override.
nvinteger(kind=I4P)inoptionalNumber of field variables override.
verboselogicalinoptionalTrigger verbose output.

Call graph

compute_local_dt_forest ​

Compute this realm's local stability-limited dt (no MPI reduction).

Invoked by forest%compute_global_dt during the min reduction across all realms in the forest. The reduction itself is the orchestrator's job; this method computes only the value local to self. Compute this realm's local stability-limited dt (no MPI reduction).

fortran
subroutine compute_local_dt_forest(self, dt_local)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe realm.
dt_localreal(kind=R8P)outLocal stability-limited dt.

Call graph

advance_one_step_forest ​

Advance this realm by one full timestep of size dt.

Invoked by forest%evolve_one_step once per realm per timestep. Owns the integration itself, i.e. everything that turns q at time t into q at time t + dt.

fortran
subroutine advance_one_step_forest(self, dt)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.
dtreal(kind=R8P)inTimestep size from the forest's global reduction.

Call graph

open_step_forest ​

Per-step prologue on the multi-realm path: set dt, init RK stages.

Mirrors the head of integrate_rk_ssp: external-field prelude (if active), rk%initialize_stages(q=self%q), and the per-step time bookkeeping advance_one_step_forest does inline (it increment, dt cap for time_max, self%time%dt update). Time advance and progress print run in close_step_forest, mirroring the legacy ordering.

fortran
subroutine open_step_forest(self, dt)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.
dtreal(kind=R8P)inTimestep size from the forest.

Call graph

begin_stage_forest ​

Begin integrator stage k on the multi-realm path (FNL).

Mirrors the FIRST line of integrate_rk_ssp's substage body: rk_fnl%compute_stage(s=k, dt=self%time%dt [, phi_gpu=ib_fnl%phi_gpu]). This populates rk_fnl%q_rk_gpu(:,:,:,:,:,k) from previously computed substages and from self%q_gpu, and publishes the stage buffer to peers by setting self%stage_active = k. No ghost reads, no peer-realm access — peer realms may not yet have opened their stage-k buffer when this fires.

fortran
subroutine begin_stage_forest(self, k, K_total, dt, realm)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.
kinteger(kind=I4P)inStage index (1..K_total).
K_totalinteger(kind=I4P)inForest-wide stage count for this step.
dtreal(kind=R8P)inTimestep size from the forest.
realmclass(realm_object)inouttarget, optionalSibling realms (contract parity).

Call graph

end_stage_forest ​

Close integrator stage k: residuals + stage assignment in one sweep (FNL).

This is the FNL forest's Phase 3 TBP. Inter-realm seam ghost cells on rk_fnl%q_rk_gpu(:,...,k) are already filled by Phase 2 (which called fill_seam_from_peer_forest on this realm and its peers), so compute_residuals_dev reads valid halo data and rk_fnl%assign_stage's in-place overwrite of the stage interior no longer races with peer reads.

realm accepted on contract for parity but unused: the seam has already been refreshed by the forest. FV seam-flux ACCUMULATION is threaded through flux_register (issue #23 R3, CPU parity); the reflux APPLICATION (apply_reflux_to_stage_forest) remains a no-op until #23 R4.

fortran
subroutine end_stage_forest(self, k, K_total, dt, realm, flux_register)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.
kinteger(kind=I4P)inStage index (1..K_total).
K_totalinteger(kind=I4P)inForest-wide stage count for this step.
dtreal(kind=R8P)inTimestep size from the forest.
realmclass(realm_object)inouttarget, optionalSibling realms (FNL parity only; unused).
flux_registerclass(flux_register_object)inoutoptionalForest's flux register for FV reflux.

Call graph

close_step_forest ​

Per-step epilogue on the multi-realm path: q assembly, BC, div-clean, residual save, coil source refresh, time advance, progress print (FNL).

fortran
subroutine close_step_forest(self, dt)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.
dtreal(kind=R8P)inTimestep size from the forest.

Call graph

fill_seam_from_peer_forest ​

FNL twin of prism_cpu_object%fill_seam_from_peer_forest. Walks the host-side row-range bookkeeping (seam_local_peer_row_start/count, populated by build_seam_local_map on the host), then dispatches an OpenACC kernel that reads peer's GPU buffer at peer-INTERIOR coords (FNL layout (b,i,j,k,v)) and writes self's GPU buffer at recv-side coords.

fortran
subroutine fill_seam_from_peer_forest(self, peer, p_idx)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inout
peerclass(realm_object)intarget
p_idxinteger(kind=I4P)in

Call graph

after_topology_build_forest ​

Propagate the freshly-built host seam maps and per-peer buffers to device-resident counterparts on field_fnl%maps. Invoked by the forest at the end of populate_inter_realm_topology AFTER build_seam_local_map has populated self%adam%maps%seam_local_*.

fortran
subroutine after_topology_build_forest(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inout

Call graph

apply_reflux_to_stage_forest ​

PRISM-FNL override of the Berger-Colella reflux correction TBP — the apply twin of prism_cpu_object%apply_reflux_to_stage_forest (issue #23 R4, M4 semantics).

α.r1 cadence: end-of-step barrier. Fires exactly once per realm per step at stage == self%rk%nrk; earlier substages return immediately. For each register face whose coarse side this realm — and this RANK (#28 D4) — owns, the correction sign*(dt/dx_coarse)*(F_coarse − F_fine_sum)(:,:,1) is applied to the COMMITTED q_gpu on the one-cell-thick coarse seam skin (this TBP runs AFTER close_step_forest's update_q). Full step weight dt/dx — NOT a stage RK coefficient, and NOT a q_rk_gpu(...,stage) write (both were the pre-M4 sketch's errors: ark is never allocated for the SSP family, and a stage-buffer write would entangle the stage beta weight — see the CPU apply's M4 note). The tiny host mismatch slab is H2D-copied per face and fv_apply_reflux_face_dev_kernel adds it to q_gpu (scalar ops only, disjoint cells).

fortran
subroutine apply_reflux_to_stage_forest(self, stage, dt, flux_register)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.
stageinteger(kind=I4P)inIntegrator stage 1..K_total.
dtreal(kind=R8P)inTime step.
flux_registerclass(flux_register_object)inForest's flux register.

Call graph

post_step_forest ​

Run PRISM-FNL's per-timestep post-step work: state IO, energy diagnostics, max-divergence diagnostics.

Invoked by forest%post_step. v1 implementation is the verbatim post- step block formerly inline in simulate — every action runs every step, since today's cadence is enforced inside the save_* routines themselves (e.g. save_simulation_data honours io%it_save). The do_* flags are signature-only for now: when the forest takes over cadence the flags will gate the individual calls. For now they are accepted but unused, preserving present-day behavior bit-for-bit.

dt, t, it are not consumed by the current body; they are on the contract so the forest can supply them once it owns time-state (today they are still read from the time module singleton).

Optional realm(:): forwarded to save_simulation_data and update_ghost for the dummy-argument inter-realm halo refresh path.

fortran
subroutine post_step_forest(self, dt, t, it, do_save_state, do_save_residuals, do_save_restart, do_amr, realm)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe realm.
dtreal(kind=R8P)inTimestep size just advanced.
treal(kind=R8P)inSimulation time after the advance.
itinteger(kind=I4P)inIteration index after the advance.
do_save_statelogicalinoptionalSave state output this step.
do_save_residualslogicalinoptionalSave residuals output this step.
do_save_restartlogicalinoptionalSave restart dump this step.
do_amrlogicalinoptionalRun AMR update this step.
realmclass(realm_object)inouttarget, optionalSibling realms for inter-realm halo refresh.

Call graph

compute_dt ​

Compute the global stability-limited dt and store it on self%time%dt.

Body delegates the local computation to compute_local_dt_forest (orchestrator contract method), then performs the legacy MPI_ALLREDUCE on MPI_COMM_WORLD for backward compatibility with simulate. The forest's compute_global_dt performs its own reduction, possibly on a per-realm sub-comm; the redundancy disappears once the legacy compute_dt is retired.

fortran
subroutine compute_dt(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

Call graph

compute_energy ​

Compute energy.

fortran
subroutine compute_energy(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

Call graph

compute_energy_error ​

Compute energy error.

fortran
subroutine compute_energy_error(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

compute_max_divergence ​

Compute maximum divergence.

fortran
subroutine compute_max_divergence(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

Call graph

impose_ct_correction ​

Impose Constrained Transport Correction on vectorial variable q(ivar:ivar+2). Note that self%divergence memory is used as buffer, be carefull.

fortran
subroutine impose_ct_correction(self, ivar)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.
ivarinteger(kind=I4P)inVariable (start) index in q.

Call graph

impose_div_free ​

Impose divergence-free property.

fortran
subroutine impose_div_free(self)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inoutThe equation.

Call graph

Functions ​

stages_per_step_forest ​

Number of integrator stages this realm exposes per step (FNL).

SSP-only contract, twin of prism_cpu_object%stages_per_step_forest (issue #25): the staged protocol reads gamm(k) per stage and beta_gpu in close_step_forest — for low-storage schemes those are never allocated (the FNL symptom was CUDA_ERROR_ILLEGAL_ADDRESS in rk_update_q_dev through a bogus beta_gpu). Refusing here leaves the fused fast path — where LS schemes legitimately run — untouched.

Returns: integer(kind=I4P)

fortran
function stages_per_step_forest(self) result(K)

Arguments

NameTypeIntentAttributesDescription
selfclass(prism_fnl_object)inThe realm.

Call graph