Visible to Intel only — GUID: GUID-3DCAD2E8-9E38-4F69-9F83-D27BB9E2813A
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-3DCAD2E8-9E38-4F69-9F83-D27BB9E2813A
Using input_node
By default, an input_node is constructed in the inactive state:
template< typename Body > input_node( graph &g, Body body, bool is_active=true )
To activate an inactive input_node, you call the node’s function activate:
input_node< int > src( g, src_body(10), false ); // use it in calls to make_edge… src.activate();
All input_node objects are constructed in the inactive state and usually activated after the entire flow graph is constructed.
For example, you can use the code in Data Flow Graph. In that implementation, the input_node is constructed in the inactive state and activated after all other edges are made:
make_edge( squarer, summer ); make_edge( cuber, summer ); input_node< int > src( g, src_body(10), false ); make_edge( src, squarer ); make_edge( src, cuber ); src.activate(); g.wait_for_all();
In this example, if the input_node was toggled to the active state at the beginning, it might send a message to squarer immediately after the edge to squarer is connected. Later, when the edge to cuber is connected, cuber will receive all future messages, but may have already missed some.
In general it is safest to create your input_node objects in the inactive state and then activate them after the whole graph is constructed. However, this approach serializes graph construction and graph execution.
Some graphs can be constructed safely with input_node``s active, allowing the overlap of construction and execution. If your graph is a directed acyclic graph (DAG), and each ``input_node has only one successor, you can activate your input_node``s just after their construction if you construct the edges in reverse topological order; that is, make the edges at the largest depth in the tree first, and work back to the shallowest edges. For example, if src is an ``input_node and func1 and func2 are both function nodes, the following graph would not drop messages, even though src is activated just after its construction:
const int limit = 10;
int count = 0;
graph g;
oneapi::tbb::flow::graph g;
oneapi::tbb::flow::input_node<int> src( g, [&]( oneapi::tbb::flow_control &fc ) -> int {
if ( count < limit ) {
return ++count;
}
fc.stop();
return {};
});
src.activate();
oneapi::tbb::flow::function_node<int,int> func1( g, 1, []( int i ) -> int {
std::cout << i << "\n";
return i;
} );
oneapi::tbb::flow::function_node<int,int> func2( g, 1, []( int i ) -> int {
std::cout << i << "\n";
return i;
} );
make_edge( func1, func2 );
make_edge( src, func1 );
g.wait_for_all();
The above code is safe because the edge from func1 to func2 is made before the edge from src to func1. If the edge from src to func1 were made first, func1 might generate a message before func2 is attached to it; that message would be dropped. Also, src has only a single successor. If src had more than one successor, the successor that is attached first might receive messages that do not reach the successors that are attached after it.