HaloSchedule: memoized intra-level halo-exchange plan for fill_boundary (ADC-260). More...

#include <pops/mesh/index/box2d.hpp>
#include <cstdint>
#include <memory>
#include <vector>

Include dependency graph for halo_schedule.hpp:

This graph shows which files directly or indirectly include this file:

Classes
struct	pops::HaloJob
	One halo copy/transfer: the ghost `region` of box `dst` is filled from the shifted valid region of box `src` (shift sx, sy in cells for the periodic wrap; 0 for an interior neighbor). More...

struct	pops::HaloSchedule
	Memoized schedule for ONE (Periodicity, domain) over a fixed layout. More...

class	pops::HaloScheduleCache
	Small per-MultiFab cache of halo schedules, one entry per distinct (Periodicity, domain). More...

Namespaces
namespace	pops

namespace	pops::detail

Functions
std::int64_t	pops::halo_schedule_build_count ()
	Number of times fill_boundary has BUILT (enumerated) a halo schedule.

void	pops::reset_halo_schedule_build_count ()
	Resets the build counter (tests).

Detailed Description

HaloSchedule: memoized intra-level halo-exchange plan for fill_boundary (ADC-260).

fill_boundary_begin used to enumerate, on EVERY call, the neighbor-box job schedule: a BoxHash build plus a local (and, under MPI, a global) enumeration over the BoxArray. That schedule is a pure function of the LAYOUT (BoxArray, DistributionMapping, n_grow) and the per-call (Periodicity, domain); only the copy/pack/MPI/unpack of the LIVE data must rerun. This header holds the cacheable plan so the enumeration runs ONCE per (layout, Periodicity, domain). Jobs carry GLOBAL box indices (resolved to local fabs at replay), so a plan is valid for any MultiFab over the same layout. MPI-free: the in-flight buffers and MPI_Request stay in HaloExchange (fill_boundary.hpp).

include/pops/mesh/boundary/halo_schedule.hpp File Reference¶

Classes

Namespaces

Functions

Detailed Description