|
| 1 | +.. _std_invoke: |
| 2 | + |
| 3 | +Invoke a Callable Object |
| 4 | +========================== |
| 5 | + |
| 6 | +Starting from C++17, the requirements for callable objects passed to algorithms or Flow Graph nodes are relaxed. It allows using additional types of bodies. |
| 7 | +Previously, the body of the algorithm or Flow Graph node needed to be a Function Object (see `C++ Standard Function Object <https://en.cppreference.com/w/cpp/utility/functional>`_) and provide an |
| 8 | +``operator()`` that accepts input parameters. |
| 9 | + |
| 10 | +Now the body needs to meet the more relaxed requirements of being Callable (see `C++ Standard Callable <https://en.cppreference.com/w/cpp/named_req/Callable>`_) that covers three types of objects: |
| 11 | + |
| 12 | +* **Function Objects that provide operator(arg1, arg2, ...)**, which accepts the input parameters |
| 13 | +* **Pointers to member functions** that you can use as the body of the algorithm or the Flow Graph node |
| 14 | +* **Pointers to member objects** work as the body of the algorithm or parallel construct |
| 15 | + |
| 16 | +You can use it not only for a Flow Graph but also for algorithms. See the example below: |
| 17 | + |
| 18 | +.. code:: |
| 19 | + |
| 20 | + // The class models oneTBB Range |
| 21 | + class StrideRange { |
| 22 | + public: |
| 23 | + StrideRange(int* s, std::size_t sz, std::size_t str) |
| 24 | + : start(s), size(sz), stride(str) {} |
| 25 | +
|
| 26 | + // A copy constructor |
| 27 | + StrideRange(const StrideRange&) = default; |
| 28 | +
|
| 29 | + // A splitting constructor |
| 30 | + StrideRange(StrideRange& other, oneapi::tbb::split) |
| 31 | + : start(other.start), size(other.size / 2) |
| 32 | + { |
| 33 | + other.size -= size; |
| 34 | + other.start += size; |
| 35 | + } |
| 36 | +
|
| 37 | + ~StrideRange() = default; |
| 38 | +
|
| 39 | + // Indicate if the range is empty |
| 40 | + bool empty() const { |
| 41 | + return size == 0; |
| 42 | + } |
| 43 | +
|
| 44 | + // Indicate if the range can be divided |
| 45 | + bool is_divisible() const { |
| 46 | + return size >= stride; |
| 47 | + } |
| 48 | +
|
| 49 | + void iterate() const { |
| 50 | + for (std::size_t i = 0; i < size; i += stride) { |
| 51 | + // Performed an action for each element of the range, |
| 52 | + // implement the code based on your requirements |
| 53 | + } |
| 54 | + } |
| 55 | +
|
| 56 | + private: |
| 57 | + int* start; |
| 58 | + std::size_t size; |
| 59 | + std::size_t stride; |
| 60 | + }; |
| 61 | +
|
| 62 | +Where: |
| 63 | + |
| 64 | +* The ``StrideRange`` class models oneTBB range that should be iterated with a specified stride during its initial construction. |
| 65 | +* The ``stride`` value is stored in a private field within the range. Therefore, the class provides the member function ``iterate() const`` that implements a loop with the specified stride. |
| 66 | + |
| 67 | +``range.iterate()`` |
| 68 | +******************* |
| 69 | + |
| 70 | +Before C++17, to utilize a range in a parallel algorithm, such as ``parallel_for``, it was required to provide a ``Function Object`` as the algorithm's body. This Function Object defined the operations to be executed on each iteration of the range: |
| 71 | + |
| 72 | +.. code:: |
| 73 | +
|
| 74 | + int main() { |
| 75 | + std::size_t array_size = 1000; |
| 76 | +
|
| 77 | + int* array_to_iterate = new int[array_size]; |
| 78 | + |
| 79 | + StrideRange range(array_to_iterate, array_size, /* stride = */ 2); |
| 80 | +
|
| 81 | + // Define a lambda function as the body of the parallel_for loop |
| 82 | + auto pfor_body = [] (const StrideRange& range) { |
| 83 | + range.iterate(); |
| 84 | + }; |
| 85 | +
|
| 86 | + // Perform parallel iteration |
| 87 | + oneapi::tbb::parallel_for(range, pfor_body); |
| 88 | +
|
| 89 | + delete[] array_to_iterate; |
| 90 | + } |
| 91 | +
|
| 92 | +An additional lambda function ``pfor_body`` was also required. This lambda function invoked the ``rage.iterate()`` function. |
| 93 | + |
| 94 | +Now with C++17, you can directly utilize a pointer to ``range.iterate()`` as the body of the algorithm: |
| 95 | + |
| 96 | +.. code:: |
| 97 | + |
| 98 | + int main() { |
| 99 | + std::size_t array_size = 1000; |
| 100 | +
|
| 101 | + int* array_to_iterate = new int[array_size]; |
| 102 | +
|
| 103 | + // Performs the iteration over the array elements with the specified stride |
| 104 | + StrideRange range(array_to_iterate, array_size, /* stride = */ 2); |
| 105 | +
|
| 106 | + // Parallelize the iteration over the range object |
| 107 | + oneapi::tbb::parallel_for(range, &StrideRange::iterate); |
| 108 | +
|
| 109 | + delete[] array_to_iterate; |
| 110 | + } |
| 111 | +
|
| 112 | +``std::invoke`` |
| 113 | +**************** |
| 114 | + |
| 115 | +``std::invoke`` is a function template that provides a syntax for invoking different types of callable objects with a set of arguments. |
| 116 | + |
| 117 | +oneTBB implementation uses the C++ standard function ``std::invoke(&StrideRange::iterate, range)`` to execute the body. It is the equivalent of ``range.iterate()``. |
| 118 | +Therefore, it allows you to invoke a callable object, such as a function object, with the provided arguments. |
| 119 | + |
| 120 | +.. tip:: Refer to `C++ Standard <https://en.cppreference.com/w/cpp/utility/functional/invoke>`_ to learn more about ``std::invoke``. |
| 121 | + |
| 122 | +Example |
| 123 | +^^^^^^^^ |
| 124 | + |
| 125 | +Consider a specific scenario with ``function_node`` within a Flow Graph. |
| 126 | + |
| 127 | +In the example below, a ``function_node`` takes an object as an input to read a member object of that input and proceed it to the next node in the graph: |
| 128 | + |
| 129 | +.. code:: |
| 130 | +
|
| 131 | + struct Object { |
| 132 | + int number; |
| 133 | + }; |
| 134 | +
|
| 135 | + int main() { |
| 136 | + using namespace oneapi::tbb::flow; |
| 137 | +
|
| 138 | + // Lambda function to read the member object of the input Object |
| 139 | + auto number_reader = [] (const Object& obj) { |
| 140 | + return obj.number; |
| 141 | + }; |
| 142 | +
|
| 143 | + // Lambda function to process the received integer |
| 144 | + auto number_processor = [] (int i) { /* processing integer */ }; |
| 145 | +
|
| 146 | + graph g; |
| 147 | +
|
| 148 | + // Function node that takes an Object as input and produces an integer |
| 149 | + function_node<Object, int> func1(g, unlimited, number_reader); |
| 150 | +
|
| 151 | + // Function node that takes an integer as input and processes it |
| 152 | + function_node<int, int> func2(g, unlimited, number_processor); |
| 153 | +
|
| 154 | + // Connect the function nodes |
| 155 | + make_edge(func1, func2); |
| 156 | +
|
| 157 | + // Provide produced input to the graph |
| 158 | + func1.try_put(Object{1}); |
| 159 | +
|
| 160 | + // Wait for the graph to complete |
| 161 | + g.wait_for_all(); |
| 162 | + } |
| 163 | +
|
| 164 | +
|
| 165 | +Before C++17, the ``function_node`` in the Flow Graph required the body to be a Function Object. A lambda function was required to extract the number from the Object. |
| 166 | + |
| 167 | +With C++17, you can use ``std::invoke`` with a pointer to the member number directly as the body. |
| 168 | + |
| 169 | +You can update the previous example as follows: |
| 170 | + |
| 171 | +.. code:: |
| 172 | +
|
| 173 | + struct Object { |
| 174 | + int number; |
| 175 | + }; |
| 176 | +
|
| 177 | + int main() { |
| 178 | + using namespace oneapi::tbb::flow; |
| 179 | + |
| 180 | + // The processing logic for the received integer |
| 181 | + auto number_processor = [] (int i) { /* processing integer */ }; |
| 182 | +
|
| 183 | + // Create a graph object g to hold the flow graph |
| 184 | + graph g; |
| 185 | +
|
| 186 | + // Use a member function pointer to the number member of the Object struct as the body |
| 187 | + function_node<Object, int> func1(g, unlimited, &Object::number); |
| 188 | +
|
| 189 | + // Use the number_processor lambda function as the body |
| 190 | + function_node<int, int> func2(g, unlimited, number_processor); |
| 191 | +
|
| 192 | + // Connect the function nodes |
| 193 | + make_edge(func1, func2); |
| 194 | +
|
| 195 | + // Connect the function nodes |
| 196 | + func1.try_put(Object{1}); |
| 197 | +
|
| 198 | + // Wait for the graph to complete |
| 199 | + g.wait_for_all(); |
| 200 | + } |
| 201 | +
|
| 202 | +Find More |
| 203 | +********* |
| 204 | + |
| 205 | +The following APIs supports Callable object as Bodies: |
| 206 | + |
| 207 | +* `parallel_for <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/algorithms/functions/parallel_for_func.html>`_ |
| 208 | +* `parallel_reduce <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/algorithms/functions/parallel_reduce_func.html>`_ |
| 209 | +* `parallel_deterministic_reduce <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/algorithms/functions/parallel_deterministic_reduce_func.html>`_ |
| 210 | +* `parallel_for_each <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/algorithms/functions/parallel_for_each_func.html>`_ |
| 211 | +* `parallel_scan <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/algorithms/functions/parallel_scan_func.html>`_ |
| 212 | +* `parallel_pipeline <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/algorithms/functions/parallel_pipeline_func.html>`_ |
| 213 | +* `function_node <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/flow_graph/func_node_cls.html>`_ |
| 214 | +* `multifunction_node <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/flow_graph/multifunc_node_cls.html>`_ |
| 215 | +* `async_node <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/flow_graph/async_node_cls.html>`_ |
| 216 | +* `sequencer_node <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/flow_graph/sequencer_node_cls.html>`_ |
| 217 | +* `join_node with key_matching policy <https://oneapi-src.github.io/oneAPI-spec/spec/elements/oneTBB/source/flow_graph/join_node_cls.html>`_ |
0 commit comments