ww3-gpu Time Step Sequence

ww3-gpu Time Step Sequence A sequence diagram generated by Archify. CALL W3SRCE :2311 w3wavemd.F90:2311 at 03adda91, inside DO JSEA=1, NSEAL (:2246) SIN4 :1239, SNL1 :1260, SDS4 :1303 w3srcemd.F90 at 03adda91 S_in, S_nl, S_ds VA(:,JSEA) updated WW_KOKKOS_<PART>? W3KOKKOS_SETUP asks ww_kokkos_flag once; unset means off on or off CALL W3SRCE :2500 w3wavemd.F90:2500 at 769c44f0, inside DO JSEA (:2417) flag off: :1244, :1544, :1622 W3SIN4, W3SNL1, W3SDS4 in w3srcemd.F90 at 769c44f0 flag on: SIN4 :1240, WW_SNL1 :1535, SDS4 :1619 WW_SNL1 inside !$OMP CRITICAL (WW_KOKKOS_SNL1) bind(C) ww_snl1, ww_sin4, ww_sds4 copy in one point kernel, fence, copy out VSNL, VDNL; LAST_ERROR WW_KOKKOS_SRCE: W3KOKKOS_SRCE :2486 whole W3SRCE for one point, when W3KOKKOS_SRCE_ON(IT) VA(:,JSEA), fluxes W3KOKKOS_SRCE_BATCH :2408 replaces the sea-point loop for this step ww_srce_batch_va, all sea points batched kernel on device VA UST, TAUW, WHITECAP unpacked W3KOKKOS_OUTG :2699 NOAA calls W3OUTG ( VA, ... ) at w3wavemd.F90:2490 ww_outg_va on device VA gridded fields, no VA copy W3KOKKOS_VA_HOST_READ before each host reader of VA, e.g. w3wavemd.F90:2691 ww_va_host_read VA copied back if device newer NOAA WW3 03adda91 Fork 769c44f0, phase 1 Fork, --kokkos resident W3WAVE · time step · Sequence participant W3WAVE time step W3SRCE · one sea point · Sequence participant W3SRCE one sea point ST4 + DIA · Fortran · Sequence participant ST4 + DIA Fortran W3KOKKOSMD · bind(C) interfaces · Sequence participant W3KOKKOSMD bind(C) interfaces part_flags · ww_kokkos_flag · Sequence participant part_flags ww_kokkos_flag C++ shims · fortran_iface/*.cpp · Sequence participant C++ shims fortran_iface/*.cpp Kokkos kernel · device copy of VA · Sequence participant Kokkos kernel device copy of VA Legend main request return security async trace message