scheduling

Classes

Name	Description
`AlgorithmTask`
`ArrayBufferPool`	The ArrayBufferPool provides a pool of NCArrays for workers managed by a scheduler.
`DagScheduler`	The primary scheduler of processing work defined by a DAG.
`DummyDataSource`
`IOGeneratorTask`
`RandomDeviceReader`
`ReadImageTask`
`Task`	The Task represents the fundamental unit of processing work in a DAG.
`ValueTask`
`WorkQueue`	A generic queue from which DAG tasks will be taken and scheduled.
`DummyDSTraits`
`LocalityHint`	Details for placement of a task on specific resources.
`ResourceRequirements`	A representation of resource availability and concurrency tokens for task throttling.

Enumerations

Name	Description
`ShmemType`	Enumerators to indicate the specificity of shared memory to use.

ShmemType

enum ShmemType

Enumerators to indicate the specificity of shared memory to use.

When creating shared-memory partitions, the allocation can be created to be shared at varying levels of specificity and inclusion. At the top, the allocation is shared across the entire machine. However, it can be created to only be used by a smaller subset, e.g., by ranks associated to a NUMA domain.

Value	Description
`MACHINE`	Whole node/machine shared memory.
`SOCKET`	By hardware socket.
`NUMA`	By NUMA domain.
`L3CACHE`	Sharing L3 cache.
`L2CACHE`	Sharing L2 cache.

Typedefs

Return	Name	Description
`std::vector< ssize_t >`	`ShapeKey`	Alias for the definition of an array’s shape - used mostly for buffer pools.
`std::tuple< numa_node_t, ShapeKey, ncarray::DType >`	`PoolKey`	Alias for look-up of buffer pools by node locality, shape and datatype.
`sbio::RC< MPI_Win, sbio::MPIWinAllocator, sbio::MPIWinDeleter >`	`RCWindow`
`int`	`numa_node_t`

ShapeKey

using ShapeKey = std::vector< ssize_t >

Alias for the definition of an array’s shape - used mostly for buffer pools.

PoolKey

using PoolKey = std::tuple< numa_node_t, ShapeKey, ncarray::DType >

Alias for look-up of buffer pools by node locality, shape and datatype.

RCWindow

using RCWindow = sbio::RC< MPI_Win, sbio::MPIWinAllocator, sbio::MPIWinDeleter >

numa_node_t

using numa_node_t = int

Functions

Return	Name	Description
`void`	`prepare_shmem_mpi_algo`	Setup an Algorithm using a shared-memory MPI strategy.
`numa_node_t`	`get_current_thread_numa_node` `inline`
`auto`	`make_read_image_task`

prepare_shmem_mpi_algo

template<class Algo> void prepare_shmem_mpi_algo(Algo & algo, RCWindow & window, std::vector< MPI_Comm > & shmem_comms, MPI_Comm main_comm = MPI_COMM_WORLD, ShmemType shmem_type = ShmemType::MACHINE)

Setup an Algorithm using a shared-memory MPI strategy.

Algorithms which require some staged data (e.g. constant matrices) can be setup to use various MPI shared communication strategies. Doing so, however, requires preparing the memory backing this data in a particular fashion. This function simplifies this process, setting up any needed communicators as well as ensuring proper synchronization so the memory will be valid for use by an rank running the Algorithm.

Parameters

Parameter	Type	Description
`algo`	`Algo &`	The Algorithm to run the data staging for.
`window`	`[RCWindow](#rcwindow) &`	The window which will back the shared, staged data.
`shmem_comms`	`std::vector< MPI_Comm > &`	Any created communicators will be returned via this vector.
`main_comm`	`MPI_Comm`	The starting communicator from which splitting will be done.
`shmem_type`	`[ShmemType](#shmemtype)`	The granularity of the shared memory backing strategy. The enumerator specifies from machine/node down to cache level.

get_current_thread_numa_node

inline

inline numa_node_t get_current_thread_numa_node()

make_read_image_task

template<class DataSource, class DataFetcher, class MemTag = ncarray::HostTag> auto make_read_image_task(DataSource & ds, DataFetcher && fetcher, typename DataSource::DataFormat::StepIdxType idx)

Variables

Return	Name	Description
`constexpr numa_node_t`	`ANY_NODE` `constexpr`
`thread_local numa_node_t`	`g_current_numa_node`

ANY_NODE

constexpr

constexpr numa_node_t ANY_NODE { -1 }

g_current_numa_node

thread_local numa_node_t g_current_numa_node {  }