tf::PartitionerBase class

class to derive a partitioner for scheduling parallel algorithms

Template parameters
C	closure wrapper type

The class provides base methods to derive a partitioner that can be used to schedule parallel iterations (e.g., tf::Taskflow::for_each).

An partitioner defines the scheduling method for running parallel algorithms, such tf::Taskflow::for_each, tf::Taskflow::reduce, and so on. By default, we provide the following partitioners:

tf::GuidedPartitioner to enable guided scheduling algorithm of adaptive chunk size
tf::DynamicPartitioner to enable dynamic scheduling algorithm of equal chunk size
tf::StaticPartitioner to enable static scheduling algorithm of static chunk size
tf::RandomPartitioner to enable random scheduling algorithm of random chunk size

Depending on applications, partitioning algorithms can impact the performance a lot. For example, if a parallel-iteration workload contains a regular work unit per iteration, tf::StaticPartitioner can deliver the best performance. On the other hand, if the work unit per iteration is irregular and unbalanced, tf::GuidedPartitioner or tf::DynamicPartitioner can outperform tf::StaticPartitioner. In most situations, tf::GuidedPartitioner can deliver decent performance and is thus used as our default partitioner.

In addition to partition size, the application can specify a closure wrapper for a partitioner. A closure wrapper allows the application to wrapper a partitioned task (i.e., closure) with a custom function object that performs additional tasks. For example:

std::atomic<int> count = 0;
tf::Taskflow taskflow;
taskflow.for_each_index(0, 100, 1, 
  [](){                 
    printf("%d\n", i); 
  },
  tf::StaticPartitioner(0, [](auto&& closure){
    // do something before invoking the partitioned task
    // ...
    
    // invoke the partitioned task
    closure();

    // do something else after invoking the partitioned task
    // ...
  }
);
executor.run(taskflow).wait();

Public types

using closure_wrapper_type = C: the closure type

Public static variables

static bool is_default_wrapper_v constexpr: indicating if the given closure wrapper is a default wrapper (i.e., empty)

Constructors, destructors, conversion operators

PartitionerBase() defaulted: default constructor
PartitionerBase(size_t chunk_size) explicit: construct a partitioner with the given chunk size
PartitionerBase(size_t chunk_size, C&& closure_wrapper): construct a partitioner with the given chunk size and closure wrapper

Public functions

auto chunk_size() const -> size_t: query the chunk size of this partitioner
void chunk_size(size_t cz): update the chunk size of this partitioner
auto closure_wrapper() const -> const C&: acquire an immutable access to the closure wrapper object
auto closure_wrapper() -> C&: acquire a mutable access to the closure wrapper object
template<typename F> void closure_wrapper(F&& fn): modify the closure wrapper object
template<typename F> auto operator()(F&& callable) -> TF_FORCE_INLINE decltype(auto): wraps the given callable with the associated closure wrapper

Protected variables

size_t _chunk_size: chunk size
C _closure_wrapper: closure wrapper