mod_rt_sc_state.t File Reference

Go to the source code of this file.

Modules
module	mod_rt_sc_state
	Parameters, resolved state and lifecycle of the short-characteristics (SC) radiative transfer, the Olson-Kunasz linear RT of AMRVAC.

Functions/Subroutines
subroutine, public	mod_rt_sc_state::rt_sc_params_read (files)
	Read &rt_list from each parameter file. The values stay physical until rt_sc_setup, and a file without the namelist is passed over.
subroutine, public	mod_rt_sc_state::rt_sc_setup ()
	Check the geometry, convert the parameters to code units, resolve the bands and the boundary conditions, and bind rt_sc_chiB to the chosen opacity. Runs after the unit system is fixed.
subroutine	mod_rt_sc_state::rt_sc_axis_cosines ()
	Map the quadrature onto the RT grid axes (hdim, updim). Called after rt_sc_init.
pure subroutine	mod_rt_sc_state::hv (ih, iv, i1, i2)
	Block indices (i1, i2) of the cell at position ih along hdim and iv along updim.
subroutine, public	mod_rt_sc_state::rt_sc_reserve_slots ()
	Reserve the wavefront intensity slots in ps_rt, four in all, one per sweep corner, of which the ray-by-ray wave uses only the first. Called from rt_sc_activate, and does nothing unless rt_sc_distributed is set.
subroutine	mod_rt_sc_state::rt_sc_alloc_state ()
	Allocate ps_rt(igrid)w on every active block and zero it, sharing the geometry of ps, which getbc reads. With bcphys off only w is exchanged. Blocks created by a regrid get their state at the next solve, and blocks that go inactive keep it until they are reused.
subroutine	mod_rt_sc_state::rt_sc_seam_alloc (dnxv, nray)
	Allocate the seam buffers once, or again when the row or ray count changes, and zero the buffers written this solve.
subroutine	mod_rt_sc_state::rt_sc_seam_finalize (dnxv, nray)
	Sum the seam edges of this solve over the ranks, where each edge has a single owner, and keep them for the next solve.
subroutine	mod_rt_sc_state::rt_allreduce_chunk (a, n1, n2, comm)
	In-place sum allreduce of a 2D array in chunks of whole columns of about 1 MB.

Variables
logical, public	mod_rt_sc_state::rt_sc_active = .false.
	True when the physics module has switched the RT on, with hd_radiative_transfer or mhd_radiative_transfer, through rt_sc_activate.
integer, public	mod_rt_sc_state::rt_sc_updim = -1
	Axis of the stratification, 1 for x and 2 for y. The default of -1 takes the last axis.
integer, public	mod_rt_sc_state::rt_sc_amr_level = 1
	The AMR level the RT runs on. The bands are held at this level so that the RT sees a uniform grid, and it cannot exceed refine_max_level.
double precision, public	mod_rt_sc_state::rt_sc_blend_t0 = 2.0d4
	Centre and half-width in K of the temperature gate in rt_sc_thin_weight, which divides the radiative losses between the SC transfer and the thin cooling.
double precision, public	mod_rt_sc_state::rt_sc_blend_dt = 5.0d3
integer, parameter	mod_rt_sc_state::rt_sc_maxband = 8
	The RT bands along updim, of which a bar with a photosphere at each end needs two. The edges are in cm, with -1e30 and +1e30 standing for the domain ends, and each end has its own incoming-intensity condition, 'thermal' or 'open'.
integer, public	mod_rt_sc_state::rt_sc_nband = 1
double precision, dimension(rt_sc_maxband), public	mod_rt_sc_state::rt_sc_band_min = -1.0d30
double precision, dimension(rt_sc_maxband), public	mod_rt_sc_state::rt_sc_band_max = 1.0d30
character(len=10), dimension(rt_sc_maxband), public	mod_rt_sc_state::rt_sc_bc_min = 'thermal'
character(len=10), dimension(rt_sc_maxband), public	mod_rt_sc_state::rt_sc_bc_max = 'open'
character(len=10), public	mod_rt_sc_state::rt_sc_bc_hmin = 'open'
	Incoming intensity at the horizontal edges where they are not periodic, 'open' for I = 0 or 'thermal' for I = B of the edge column.
character(len=10), public	mod_rt_sc_state::rt_sc_bc_hmax = 'open'
logical, public	mod_rt_sc_state::rt_sc_flux_blend = .true.
	rt_sc_flux_blend blends the thin heating c chi (J - B) with the thick -div F. rt_sc_cfl limits the heating so that one stage changes the gas energy by at most that fraction of itself
double precision, public	mod_rt_sc_state::rt_sc_cfl = 0.5d0
logical, public	mod_rt_sc_state::rt_sc_debug = .false.
	Prints the cumulative RT timing every 10 solves.
double precision, public	mod_rt_sc_state::rt_sc_fi_thresh = -1.0d0
	Not read by the RT itself. The user files set it from eoseint_rho_FI_threshold for their own SC source.
integer, public	mod_rt_sc_state::i_qrad = -1
	Extra-variable indices of the heating Q, the vertical flux F and the emergent intensity Iout, left at -1 when they are not registered.
integer, public	mod_rt_sc_state::i_frad = -1
integer, public	mod_rt_sc_state::i_iout = -1
double precision, public	mod_rt_sc_state::rt_sc_ftau1 = 0.0d0
	Horizontal mean of the flux on the tau = rt_sc_tau_emergent surface, from rt_sc_emergent_tau1. A flux controller should read this one, because Frad still grows by 10-16% above tau = 1 and the flux at the band top therefore reads high.
double precision, public	mod_rt_sc_state::rt_sc_tau_emergent = 1.0d0
logical, public	mod_rt_sc_state::rt_sc_distributed = .true.
	The distributed KBA wavefront. Each rank sweeps only its own band blocks, one diagonal at a time, with a getbc of the ray intensity between diagonals, so no rank ever holds the whole slab. Its intensity slots are reserved by rt_sc_activate. Setting it false selects the replicated gather, the original implementation, kept for comparison only: it holds the whole band on every rank and its periodic seam is not exact. Production runs should keep KBA.
logical, public	mod_rt_sc_state::rt_sc_pipeline = .true.
	The pipelined KBA. The four sweep corners run at once, each with its own intensity slot, and the rays of a corner follow one another through the wavefront. It agrees with the ray-by-ray wave to round-off and is 1.4-1.8 times faster.
logical, public	mod_rt_sc_state::rt_sc_valpipe = .false.
	For validation, runs the ray-by-ray and pipelined waves on the same state and prints max|dQ|.
integer	mod_rt_sc_state::i_isw = -1
	First column of the ray intensity in ps_rt, and -1 until reserved. It is the only field getbc exchanges during a sweep, because chi and B are cached per block while J and Fy accumulate in i_qrad and i_frad, perhaps confusingly but it works and prevents size creep in ps_rt.
integer	mod_rt_sc_state::nisw = 1
	Number of intensity slots, one for the ray-by-ray wave and up to four when pipelined.
type(state), dimension(:), allocatable, target	mod_rt_sc_state::ps_rt
	The RT intensity state. It is kept out of the hydro w so that it skips the stage copies, the prolongation and the output, and getbc exchanges it with bcphys off. rt_sc_alloc_state allocates it per block.
logical	mod_rt_sc_state::rt_sc_seam_warned = .false.
logical	mod_rt_sc_state::rt_sc_seam_on = .true.
	The periodic seam of the single-pass KBA. A ray crossing the seam takes as its inflow the outflow at the far edge from the previous solve, which is exact at steady state. The buffers are indexed by global row along updim and by ray. seamR holds the right-hand outflow of the last block column, which feeds the +x wrap into the first, and seamL the reverse. The _prev buffers are read and the _cur buffers written, the allreduce at the end of each solve makes them global, and both start at zero. With rt_sc_seam_on false the seam inflow stays zero.
double precision, dimension(:,:), allocatable	mod_rt_sc_state::rt_sc_seaml_prev
double precision, dimension(:,:), allocatable	mod_rt_sc_state::rt_sc_seamr_prev
double precision, dimension(:,:), allocatable	mod_rt_sc_state::rt_sc_seaml_cur
double precision, dimension(:,:), allocatable	mod_rt_sc_state::rt_sc_seamr_cur
integer	mod_rt_sc_state::rt_node_comm = -1
	Node-local communicator for the moment reduction of the gather. The ranks reach that reduction far out of step because of the ray imbalance, and inter-node collectives under such a stagger fail on the tcp/ofi provider. Each node therefore forms the full moments from its own copy of the field.
integer	mod_rt_sc_state::rt_node_rank = 0
integer	mod_rt_sc_state::rt_node_size = 1
integer	mod_rt_sc_state::rt_lead_comm = -1
	Communicator of one leader rank per node. The field gather sums within each node, then over the leaders, and then broadcasts within the node, so that no collective spans all the ranks. The tcp provider refuses those at 46 Mm.
double precision, public	mod_rt_sc_state::rt_wt_gath =0.d0
	Cumulative wall time of the field gather, the sweep, the moment reduction and the scatter, and of the wave, together with the number of solves.
double precision, public	mod_rt_sc_state::rt_wt_swp =0.d0
double precision, public	mod_rt_sc_state::rt_wt_red =0.d0
double precision, public	mod_rt_sc_state::rt_wt_scat =0.d0
double precision, public	mod_rt_sc_state::rt_wt_wave =0.d0
integer, public	mod_rt_sc_state::rt_nsolve =0
double precision, dimension(:,:,:), allocatable	mod_rt_sc_state::rtw_chib
	Block chi and B including the ghosts, rebuilt every solve, and the block horizontal flux Fx for the -div F blend. Fy accumulates in i_frad instead, and Fx is copied into ps_rt for its ghost exchange when the band closes.
double precision, dimension(:,:,:), allocatable	mod_rt_sc_state::rtw_bb
double precision, dimension(:,:,:), allocatable	mod_rt_sc_state::rtw_fx
integer	mod_rt_sc_state::wv_nblk
	The band blocks of this rank with their block position (bi, bj), the block rows of the current band and its longest diagonal, and the wrap and seam modes of the KBA.
integer	mod_rt_sc_state::wv_nbxv
integer	mod_rt_sc_state::wv_dmax
integer, dimension(:), allocatable	mod_rt_sc_state::wv_ig
integer, dimension(:), allocatable	mod_rt_sc_state::wv_bi
integer, dimension(:), allocatable	mod_rt_sc_state::wv_bj
logical	mod_rt_sc_state::wv_xwrap
logical	mod_rt_sc_state::wv_seam
double precision, dimension(:), allocatable	mod_rt_sc_state::mu_h
	Ray cosines along the RT grid axes (hdim, updim). They are (mux, muy) for updim = 2 and the swapped pair for updim = 1.
double precision, dimension(:), allocatable	mod_rt_sc_state::mu_v
integer	mod_rt_sc_state::updim = 2
integer	mod_rt_sc_state::hdim = 1
double precision, dimension(rt_sc_maxband)	mod_rt_sc_state::band_lo_c
double precision, dimension(rt_sc_maxband)	mod_rt_sc_state::band_hi_c
logical	mod_rt_sc_state::rt_sc_ready = .false.
integer	mod_rt_sc_state::rg_nh
	The RT grid on rt_sc_amr_level along (hdim, updim): cells across the domain, cells per block, blocks across the width, cell sizes and the lower domain edges, with the cell and block rows of each band.
integer	mod_rt_sc_state::rg_nv
integer	mod_rt_sc_state::rg_bh
integer	mod_rt_sc_state::rg_bv
integer	mod_rt_sc_state::rg_nbh
double precision	mod_rt_sc_state::rg_dh
double precision	mod_rt_sc_state::rg_dv
double precision	mod_rt_sc_state::rg_hlo
double precision	mod_rt_sc_state::rg_vlo
integer, dimension(rt_sc_maxband)	mod_rt_sc_state::band_nv
integer, dimension(rt_sc_maxband)	mod_rt_sc_state::band_nbv
integer, dimension(2)	mod_rt_sc_state::eh
	Unit steps of hdim and updim in block index space and the block index ranges along them, so that a loop written in (h, v) serves either stratification axis.
integer, dimension(2)	mod_rt_sc_state::ev
integer	mod_rt_sc_state::imlo_h
integer	mod_rt_sc_state::imhi_h
integer	mod_rt_sc_state::imlo_v
integer	mod_rt_sc_state::imhi_v
integer	mod_rt_sc_state::iglo_h
integer	mod_rt_sc_state::ighi_h
integer	mod_rt_sc_state::iglo_v
integer	mod_rt_sc_state::ighi_v