Visible to Intel only — GUID: GUID-570B5FE3-E3C9-45A5-9CD2-15E2E5CA2891
Package Contents
Parallelizing Simple Loops
Parallelizing Complex Loops
Parallelizing Data Flow and Dependence Graphs
Work Isolation
Exceptions and Cancellation
Containers
Mutual Exclusion
Timing
Memory Allocation
The Task Scheduler
Design Patterns
Migrating from Threading Building Blocks (TBB)
Constrained APIs
Invoke a Callable Object
Appendix A Costs of Time Slicing
Appendix B Mixing With Other Threading Packages
References
parallel_for_each Body semantics and requirements
parallel_sort ranges interface extension
TBB_malloc_replacement_log Function
Type-specified message keys for join_node
Scalable Memory Pools
Helper Functions for Expressing Graphs
concurrent_lru_cache
task_group extensions
The customizing mutex type for concurrent_hash_map
Visible to Intel only — GUID: GUID-570B5FE3-E3C9-45A5-9CD2-15E2E5CA2891
Flow Graph Basics: Buffering and Forwarding
oneAPI Threading Building Blocks (oneTBB) flow graph nodes use messages to communicate data and to enforce dependencies. If a node passes a message successfully to any successor, no further action is taken with the message by that node. As noted in the section on Single-push vs. Broadcast-push, a message may be passed to one or to multiple successors, depending on the type of the node, how many successors are connected to the node, and whether the message is pushed or pulled.
There are times when a node cannot successfully push a message to any successor. In this case what happens to the message depends on the type of the node. The two possibilities are:
The node stores the message to be forwarded later.
The node discards the message.
If a node discards messages that are not forwarded, and this behavior is not desired, the node should be connected to a buffering node that does store messages that cannot be pushed.
If a message has been stored by a node, there are two ways it can be passed to another node:
A successor to the node can pull the message using try_get() or try_reserve().
A successor can be connected using make_edge().
If a try_get() successfully forwards a message, it is removed from the node that stored it. If a node is connected using make_edge the node will attempt to push a stored message to the new successor.