|
MFC
Exascale flow solver
|
Same-level fine-fine seams and the coarse conservative halo. More...
Functions/Subroutines | |
| impure subroutine, public | s_amr_exchange_coarse_cons_halo (q_cons) |
| Exchange coarse conservative ghost layers at internal rank boundaries (beg then end per direction); the solver fills only prim ghosts. Collective. | |
| pure logical function, public | f_amr_seam (xb, yb, d) |
| True iff yb sits directly above xb on xb's high face in dim d with identical transverse coarse extents (one ordered pair per seam). | |
| impure subroutine | s_amr_fine_seam_exchange (npair, plx, ply, pd, pxhi, pfm, ndep) |
| Same-rank seam exchange, both directions in one kernel: xb high interior -> yb low ghost and yb low interior -> xb high ghost. The four slabs are disjoint, so fusing is safe. Threads span the max pair extent and are masked. | |
| pure integer function, public | f_amr_seam_dim (xb, yb) |
| Seam dim of the ordered pair (xb, yb): the d in which yb is xb's same-level high-face neighbour, else 0 (adjacent in at most one dim). | |
| pure subroutine | s_amr_seam_probe (xb, d, clo, mkey, ord, nblk, mb, md, nm) |
| Binary-search the Morton-sorted lo corners for the block at clo, verify the seam predicate against xb and record it in (mb, md, nm); a block already recorded is raised to the higher d. | |
| impure subroutine, public | s_amr_build_seam_pairs () |
| Rebuild the same-level seam-pair list once per regrid/restart in identical (xb, yb) order on all ranks, and the per-block overlap-rank lists (amr_ovl_gather/scatter). | |
| impure subroutine, public | s_amr_fine_fine_post () |
| Fine-fine halo between tiled sub-blocks: overwrite seam ghosts with the neighbour's stage-entry interior (cross-rank via the seam wave, same-rank by local copy). Wire is wp, cast to stp on unpack. | |
| impure subroutine, public | s_amr_fine_fine_drain () |
Same-level fine-fine seams and the coarse conservative halo.
| pure logical function, public m_amr_seam::f_amr_seam | ( | integer, intent(in) | xb, |
| integer, intent(in) | yb, | ||
| integer, intent(in) | d ) |
True iff yb sits directly above xb on xb's high face in dim d with identical transverse coarse extents (one ordered pair per seam).
Definition at line 427 of file m_amr_seam.fpp.f90.
| pure integer function, public m_amr_seam::f_amr_seam_dim | ( | integer, intent(in) | xb, |
| integer, intent(in) | yb ) |
Seam dim of the ordered pair (xb, yb): the d in which yb is xb's same-level high-face neighbour, else 0 (adjacent in at most one dim).
Definition at line 542 of file m_amr_seam.fpp.f90.
| impure subroutine, public m_amr_seam::s_amr_build_seam_pairs |
Rebuild the same-level seam-pair list once per regrid/restart in identical (xb, yb) order on all ranks, and the per-block overlap-rank lists (amr_ovl_gather/scatter).
Definition at line 606 of file m_amr_seam.fpp.f90.
| impure subroutine, public m_amr_seam::s_amr_exchange_coarse_cons_halo | ( | type(scalar_field), dimension(sys_size), intent(inout) | q_cons | ) |
Exchange coarse conservative ghost layers at internal rank boundaries (beg then end per direction); the solver fills only prim ghosts. Collective.
Definition at line 393 of file m_amr_seam.fpp.f90.
| impure subroutine, public m_amr_seam::s_amr_fine_fine_drain |
| impure subroutine, public m_amr_seam::s_amr_fine_fine_post |
Fine-fine halo between tiled sub-blocks: overwrite seam ghosts with the neighbour's stage-entry interior (cross-rank via the seam wave, same-rank by local copy). Wire is wp, cast to stp on unpack.
Definition at line 680 of file m_amr_seam.fpp.f90.
|
private |
Same-rank seam exchange, both directions in one kernel: xb high interior -> yb low ghost and yb low interior -> xb high ghost. The four slabs are disjoint, so fusing is safe. Threads span the max pair extent and are masked.
| [in] | pfm | per-pair: slots, seam dim, high extent, fine extents |
Definition at line 442 of file m_amr_seam.fpp.f90.
|
private |
Binary-search the Morton-sorted lo corners for the block at clo, verify the seam predicate against xb and record it in (mb, md, nm); a block already recorded is raised to the higher d.
Definition at line 557 of file m_amr_seam.fpp.f90.