Skip to content

adam_flume_fnl_mhd_kernels ​

ADAM, FLUME FNL device kernels of the ideal MHD (no divergence control) model.

The model-agnostic auxiliary-variables kernel (adam_flume_fnl_aux_kernels_agnostic.INC) and the MHD dt and conservation kernels (adam_flume_fnl_mhd_kernels_agnostic.INC) instantiated with NV_K = NV_MHD, NV_AUX_K = NV_AUX_MHD (issue #41, section 4). The face-flux kernel is the shared body with the MHD split (M2-P3). Same kernel rules as adam_flume_fnl_kernels (issue #35, D-11/D-12).

Source: src/app/flume/fnl/adam_flume_fnl_mhd_kernels.F90

Dependencies

Contents ​

Subroutines ​

compute_face_fluxes_dev ​

Compute the WENO face fluxes of direction d: face (i,j,k) lies between cells (i,j,k) and (i+di,j+dj,k+dk), so the face array starts at index 0 along d.

Per face (device twin of the CPU compute_face_fluxes): gather the stencil m = 1-S ... S into constant-bound privates, project and split it, reconstruct every field (the private v is the packed stencil of the WENO primitive), back-project.

fortran
subroutine compute_face_fluxes_dev(d, di, dj, dk, ni, nj, nk, ngc, blocks_number, S, gamma, ch, is_characteristic, weno_a_gpu, weno_p_gpu, weno_d_gpu, weno_zeps, q_gpu, q_aux_gpu, fl_gpu)

Arguments

NameTypeIntentAttributesDescription
dinteger(kind=I4P)inDirection, 1=x, 2=y, 3=z.
diinteger(kind=I4P)inUnit step along d.
djinteger(kind=I4P)inUnit step along d.
dkinteger(kind=I4P)inUnit step along d.
niinteger(kind=I4P)inGrid dimensions.
njinteger(kind=I4P)inGrid dimensions.
nkinteger(kind=I4P)inGrid dimensions.
ngcinteger(kind=I4P)inGrid dimensions.
blocks_numberinteger(kind=I4P)inActual blocks number.
Sinteger(kind=I4P)inWENO stencil half-width.
gammareal(kind=R8P)inSpecific heats ratio.
chreal(kind=R8P)inGLM cleaning speed (models with GLM only).
is_characteristiclogicalinCharacteristic (or conservative) variables.
weno_a_gpureal(kind=R8P)inWENO optimal weights.
weno_p_gpureal(kind=R8P)inWENO polynomials coefficients.
weno_d_gpureal(kind=R8P)inWENO smoothness indicators coefficients.
weno_zepsreal(kind=R8P)inWENO parameter avoiding division by zero.
q_gpureal(kind=R8P)inConservative variables.
q_aux_gpureal(kind=R8P)inAuxiliary variables.
fl_gpureal(kind=R8P)inoutFace fluxes of direction d.

Call graph

compute_q_aux_dev ​

Compute the auxiliary variables on every cell, ghost cells included.

fortran
subroutine compute_q_aux_dev(ni, nj, nk, ngc, blocks_number, gamma, R, q_gpu, q_aux_gpu)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)inGrid dimensions.
njinteger(kind=I4P)inGrid dimensions.
nkinteger(kind=I4P)inGrid dimensions.
ngcinteger(kind=I4P)inGrid dimensions.
blocks_numberinteger(kind=I4P)inActual blocks number.
gammareal(kind=R8P)inSpecific heats ratio.
Rreal(kind=R8P)inGas constant.
q_gpureal(kind=R8P)inConservative variables.
q_aux_gpureal(kind=R8P)inoutAuxiliary variables.

Call graph

compute_conservation_dev ​

Compute the volume integrals of the conservative variables (interior cells), one reduction per variable. The cell volume includes the null directions (issue #37).

fortran
subroutine compute_conservation_dev(ni, nj, nk, ngc, blocks_number, dxyz_gpu, q_gpu, integrals)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)inGrid dimensions.
njinteger(kind=I4P)inGrid dimensions.
nkinteger(kind=I4P)inGrid dimensions.
ngcinteger(kind=I4P)inGrid dimensions.
blocks_numberinteger(kind=I4P)inActual blocks number.
dxyz_gpureal(kind=R8P)inBlocks space steps [nb, 3].
q_gpureal(kind=R8P)inConservative variables.
integralsreal(kind=R8P)outVolume integrals.

Call graph

compute_lambda_max_dev ​

Compute max(sum_d (|u_d| + c_{f,d}) / dx_d) over the interior cells (null directions excluded), c_{f,d} the fast magnetosonic speed along d (issue #41, section 3.5).

fortran
subroutine compute_lambda_max_dev(ni, nj, nk, ngc, blocks_number, gamma, R, dxyz_gpu, is_null, q_gpu, lambda_max)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)inGrid dimensions.
njinteger(kind=I4P)inGrid dimensions.
nkinteger(kind=I4P)inGrid dimensions.
ngcinteger(kind=I4P)inGrid dimensions.
blocks_numberinteger(kind=I4P)inActual blocks number.
gammareal(kind=R8P)inSpecific heats ratio.
Rreal(kind=R8P)inGas constant.
dxyz_gpureal(kind=R8P)inBlocks space steps [nb, 3].
is_nulllogicalinNull directions.
q_gpureal(kind=R8P)inConservative variables.
lambda_maxreal(kind=R8P)outMaximum of sum_d (|u_d| + c_{f,d}) / dx_d.

Call graph

apply_floors_dev ​

Apply the MHD positivity floors to the interior cells of a stage state, before its ghost exchange (issue #41, section 3.8): a cell with rho < rho_floor or p < p_floor gets rho = max(rho, rho_floor) at fixed velocity and p = max(p, p_floor) (internal energy added). Interior only: the exchange then carries the floored values to the ghost cells, and ghost cells that no map fills (never read by the stencils) are not inspected. Returns the floored cells, the cells with a non-positive density or pressure (counted before flooring) and the minimum density and pressure. Floors are not conservative: the caller logs the counters.

fortran
subroutine apply_floors_dev(ni, nj, nk, ngc, blocks_number, gamma, R, rho_floor, p_floor, q_gpu, floored, nonpositive, rho_min, p_min)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)inGrid dimensions.
njinteger(kind=I4P)inGrid dimensions.
nkinteger(kind=I4P)inGrid dimensions.
ngcinteger(kind=I4P)inGrid dimensions.
blocks_numberinteger(kind=I4P)inActual blocks number.
gammareal(kind=R8P)inSpecific heats ratio.
Rreal(kind=R8P)inGas constant.
rho_floorreal(kind=R8P)inDensity and pressure floors.
p_floorreal(kind=R8P)inDensity and pressure floors.
q_gpureal(kind=R8P)inoutConservative variables.
flooredinteger(kind=I4P)outFloored cells.
nonpositiveinteger(kind=I4P)outCells with non-positive density or pressure.
rho_minreal(kind=R8P)outMinimum density and pressure.
p_minreal(kind=R8P)outMinimum density and pressure.

Call graph

compute_speed_max_dev ​

Compute max(|u_d| + c_{f,d}) over the interior cells and the active directions, the fastest wave speed (the GLM c_h check, issue #41, section 3.5).

fortran
subroutine compute_speed_max_dev(ni, nj, nk, ngc, blocks_number, gamma, R, is_null, q_gpu, speed_max)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)inGrid dimensions.
njinteger(kind=I4P)inGrid dimensions.
nkinteger(kind=I4P)inGrid dimensions.
ngcinteger(kind=I4P)inGrid dimensions.
blocks_numberinteger(kind=I4P)inActual blocks number.
gammareal(kind=R8P)inSpecific heats ratio.
Rreal(kind=R8P)inGas constant.
is_nulllogicalinNull directions.
q_gpureal(kind=R8P)inConservative variables.
speed_maxreal(kind=R8P)outMaximum of |u_d| + c_{f,d}.

Call graph

compute_divb_norms_dev ​

Compute the div(B) norms over the interior cells (issue #41, D-10): max|div B|, sum |div B| dV and the maximum over the seam-local cells (within band cells of a block face with seam(b,face) /= 0, a 2:1 AMR or inter-realm seam). div B is the centred finite difference of the library (half stencil hs <= HS_MAX), null directions weighted zero; the cell volume includes the null directions (issue #37). The caller has refreshed the ghost cells.

fortran
subroutine compute_divb_norms_dev(ni, nj, nk, ngc, blocks_number, hs, band, dxyz_gpu, is_null, seam_gpu, q_gpu, divb_max, divb_l1, divb_seam_max)

Arguments

NameTypeIntentAttributesDescription
niinteger(kind=I4P)inGrid dimensions.
njinteger(kind=I4P)inGrid dimensions.
nkinteger(kind=I4P)inGrid dimensions.
ngcinteger(kind=I4P)inGrid dimensions.
blocks_numberinteger(kind=I4P)inActual blocks number.
hsinteger(kind=I4P)inFinite difference half stencil, <= min(ngc, HS_MAX).
bandinteger(kind=I4P)inSeam-local band width, in cells.
dxyz_gpureal(kind=R8P)inBlocks space steps [nb, 3].
is_nulllogicalinNull directions.
seam_gpuinteger(kind=I4P)inSeam faces flags [nb, 6] (0: not a seam face).
q_gpureal(kind=R8P)inConservative variables.
divb_maxreal(kind=R8P)outMaximum of |div B|.
divb_l1real(kind=R8P)outSum of |div B| dV.
divb_seam_maxreal(kind=R8P)outMaximum of |div B| over the seam-local cells.

Call graph

face_split_fluxes ​

Split adapter of the shared face kernel (issue #41, M2-P3): the MHD split without cleaning, ch unused.

Attributes: pure

fortran
subroutine face_split_fluxes(gamma, ch, d, S, is_characteristic, qs, qas, fsplit, er)

Arguments

NameTypeIntentAttributesDescription
gammareal(kind=R8P)inSpecific heats ratio.
chreal(kind=R8P)inGLM cleaning speed.
dinteger(kind=I4P)inDirection, 1=x, 2=y, 3=z.
Sinteger(kind=I4P)inWENO stencil half-width, S <= S_MAX.
is_characteristiclogicalinCharacteristic (or conservative) variables.
qsreal(kind=R8P)inStencil conservative variables.
qasreal(kind=R8P)inStencil auxiliary variables.
fsplitreal(kind=R8P)outSplit fields in the WENO upwind layout.
erreal(kind=R8P)outRight eigenvectors.

Call graph