graph.h

Go to the documentation of this file.

/*
 * Copyright (c) Meta Platforms, Inc. and affiliates.
 *
 * This source code is licensed under the MIT license found in the
 * LICENSE file in the root directory of this source tree.
 */
 
#pragma once
 
#include <atomic>
#include <deque>
#include <limits>
#include <memory>
#include <type_traits>
#include <vector>
 
#include <dispenso/once_function.h>
#include <dispenso/platform.h>
#include <dispenso/pool_allocator.h>
#include <dispenso/small_buffer_allocator.h>
 
/*
Terminology
--------------------------------------------------------------------------------
The Node depends on the predecessor. The dependent depends on the node.
~~~
┌─────────────┐     ┌──────┐     ┌───────────┐
│ predecessor │ ──▶ │ node │ ──▶ │ dependent │
└─────────────┘     └──────┘     └───────────┘
~~~
 
Graph construction
--------------------------------------------------------------------------------
The Graph class can be used to created tasks with dependencies and execute it once.
Graphs must not contain cycles.
Example:
~~~
//
//   ┌────────────┐     ┌───────────────┐     ┌───────────────────┐
//   │ 0: r[0]+=1 │ ──▶ │ 1: r[1]=r[0]*2│ ──▶ │ 4: r[4]=r[1]+r[3] │
//   └────────────┘     └───────────────┘     └───────────────────┘
//                                              ▲
//   ┌────────────┐     ┌───────────────┐       │
//   │ 2: r[2]+=8 │ ──▶ │ 3: r[3]=r[2]/2│ ──────┘
//   └────────────┘     └───────────────┘
 
std::array<float, 5> r;
 
dispenso::Graph graph;
 
dispenso::Node& N0 = graph.addNode([&]() { r[0] += 1; });
dispenso::Node& N2 = graph.addNode([&]() { r[2] += 8; });
dispenso::Node& N1 = graph.addNode([&]() { r[1] += r[0] * 2; });
dispenso::Node& N3 = graph.addNode([&]() { r[3] += r[2] / 2; });
dispenso::Node& N4 = graph.addNode([&]() { r[4] += r[1] + r[3]; });
 
N4.dependsOn(N1, N3);
N1.dependsOn(N0);
N3.dependsOn(N2);
 
dispenso::TaskSet taskSet(dispenso::globalThreadPool());
dispenso::ParallelForExecutor parallelForExecutor;
parallelForExecutor(taskSet, graph);
~~~
 
Partial revaluation
--------------------------------------------------------------------------------
If graph is big or we need to recompute graph partially we can execute it again.
After execution of the graph all nodes change their state from "incomplete" to
"completed". If order to evaluate whole graph again we can use function `setAllNodesIncomplete`
Example:
~~~
r = {0, 0, 0, 0, 0};
setAllNodesIncomplete(graph);
parallelForExecutor(taskSet, graph);
~~~
 
The graph can be recomputed partially if we have new input data for one or several nodes in the
graph. It order to do it we need to call `setIncomplete()` method for every node which we need to
recompute and after use functor `ForwardPropagator` to mark as "incomplete" all dependents.
 
Example:
~~~
N1.setIncomplete();
r[1] = r[4] = 0;
ForwardPropagator forwardPropagator;
forwardPropagator(graph);
evaluateGraph(graph);
~~~
In this exaple only node 1 and 4 will be invoked.
 
 Subgraphs
--------------------------------------------------------------------------------
It is possible to organize nodes into subgraphs and destroy and recreate if we have static and
dynamic parts of the computation graph
Example:
~~~
//
// ∙----subgraph1---∙ ∙---subgraph2-------∙
// ¦ ┌────────────┐ ¦ ¦ ┌───────────────┐ ¦   ┌───────────────────┐
// ¦ │ 0: r[0]+=1 │ ──▶ │ 1: r[1]=r[0]*2│ ──▶ │ 4: r[4]=r[1]+r[3] │
// ¦ └────────────┘ ¦ ¦ └───────────────┘ ¦   └───────────────────┘
// ¦                ¦ ¦                   ¦     ▲
// ¦ ┌────────────┐ ¦ ¦ ┌───────────────┐ ¦     │
// ¦ │ 2: r[2]+=8 │ ──▶ │ 3: r[3]=r[2]/2│ ──────┘
// ¦ └────────────┘ ¦ ¦ └───────────────┘ ¦
// ∙----------------∙ ∙-------------------∙
std::array<float, 5> r;
dispenso::Graph graph;
 
dispenso::Subgraph& subgraph1 = graph.addSubgraph();
dispenso::Subgraph& subgraph2 = graph.addSubgraph();
 
dispenso::Node& N0 = subgraph1.addNode([&]() { r[0] += 1; });
dispenso::Node& N2 = subgraph1.addNode([&]() { r[2] += 8; });
dispenso::Node& N1 = subgraph2.addNode([&]() { r[1] += r[0] * 2; });
dispenso::Node& N3 = subgraph2.addNode([&]() { r[3] += r[2] / 2; });
dispenso::Node& N4 = graph.addNode([&]() { r[4] += r[1] + r[3]; });
 
N4.dependsOn(N1, N3);
N1.dependsOn(N0);
N3.dependsOn(N2);
 
// evaluate graph first time
r = {0, 0, 0, 0, 0};
dispenso::ConcurrentTaskSet concurrentTaskSet(dispenso::globalThreadPool());
dispenso::ConcurrentTaskSetExecutor concurrentTaskSetExecutor;
concurrentTaskSetExecutor(concurrentTaskSet, graph);
 
// disconnect and destroy nodes of subgraph2
// it invalidates node references/pointers of this subgraph
subgraph2.clear();
 
// create another nodes
dispenso::Node& newN1 = subgraph2.addNode([&]() { r[1] += r[0] * 20; });
dispenso::Node& newN3 = subgraph2.addNode([&]() { r[3] += r[2] / 20; });
newN1.dependsOn(N0);
newN3.dependsOn(N2);
N4.dependsOn(newN1, newN3);
 
// and revaluae the graph
setAllNodesIncomplete(movedGraph);
concurrentTaskSetExecutor(concurrentTaskSet, graph);
~~~
 
Bidirectional propagation dependency
--------------------------------------------------------------------------------
In certain scenarios, nodes may alter the same memory. In such instances, it becomes necessary to
compute the predecessors of the node, even if they possess a "completed" state following state
propagation. To facilitate this process automatically, we introduce the notion of a bidirectional
propagation dependency (`BiProp`).
 
Example:
~~~
//                     ┌─────────────────┐
//                     │ 3: m3+=b*b      │
//                     └─────────────────┘
//                       ▲
// ┌−-----------−−−−−−−−−│−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−┐
// ╎ ┌───────────┐     ┌─────────────────┐     ┌────────────┐ ╎
// ╎ │  0: b+=5  │ ──▷ │     1: b*=5     │ ──▷ │ 2: b/=m4   │ ╎
// ╎ └───────────┘     └─────────────────┘     └────────────┘ ╎
// └−−−-−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−▲−−−−−−−−−−−−┘
//  Legend:                                      │
//  ──▶ Normal dependency                      ┌────────────┐
//  ──▷ Bidirectional propagation dependency   │ 4: m4+=2   │
//  m4  variable modified only in node 4       └────────────┘
 
float  b,  m3, m4
dispenso::BiPropGraph g;
std::array<dispenso::BiPropNode*, 8> N;
 
dispenso::BiPropNode& N0 = g.addNode([&]() { b += 5; });
dispenso::BiPropNode& N1 = g.addNode([&]() { b *= 5; });
dispenso::BiPropNode& N2 = g.addNode([&]() { b /= m4; });
dispenso::BiPropNode& N3 = g.addNode([&]() { m3 += b*b; });
dispenso::BiPropNode& N4 = g.addNode([&]() { m4 += 2; });
 
N3.dependsOn(N1);
N2.dependsOn(N4);
N2.biPropDependsOn(N1);
N1.biPropDependsOn(N0);
 
// first execution
b = m3 = m4 = 0.f;
dispenso::ConcurrentTaskSet concurrentTaskSet(dispenso::globalThreadPool());
dispenso::ConcurrentTaskSetExecutor concurrentTaskSetExecutor;
concurrentTaskSetExecutor(concurrentTaskSet, g);
 
N[4].setIncomplete();
// if node 4 is incomplete after propagation node 2 become incomplete. Taking in account that node 2
// bidirectionally depends on nodes 0 and 1 they will be marked as incomplete as well
b =  m4 = 0.f;
ForwardPropagator forwardPropagator;
forwardPropagator(g);
concurrentTaskSetExecutor(concurrentTaskSet, g);
~~~
 
Please read tests from `graph_test.cpp` for more examples.
*/
 
namespace detail {
class ExecutorBase;
 
template <typename F>
void callFunctor(void* ptr) {
  (*static_cast<F*>(ptr))();
}
 
template <typename F>
void destroyFunctor(void* ptr) {
  static_cast<F*>(ptr)->~F();
  constexpr size_t kFuncSize = static_cast<size_t>(dispenso::detail::nextPow2(sizeof(F)));
  dispenso::deallocSmallBuffer<kFuncSize>(ptr);
}
 
} // namespace detail
 
namespace dispenso {
class Node {
 public:
  Node() = delete;
  Node(const Node&) = delete;
  Node& operator=(const Node&) = delete;
  Node(Node&& other) noexcept
      : numIncompletePredecessors_(other.numIncompletePredecessors_.load()),
        numPredecessors_(other.numPredecessors_),
        invoke_(other.invoke_),
        destroy_(other.destroy_),
        funcBuffer_(other.funcBuffer_),
        dependents_(std::move(other.dependents_)) {
    other.funcBuffer_ = nullptr;
  }
  ~Node() {
    if (funcBuffer_) {
      destroy_(funcBuffer_);
    }
  }
  template <typename... Ns>
  inline void dependsOn(Ns&... nodes) {
    ((void)std::initializer_list<int>{(dependsOnOneNode(nodes), 0)...});
  }
  inline void run() const {
    invoke_(funcBuffer_);
    numIncompletePredecessors_.store(kCompleted, std::memory_order_release);
  }
  template <class F>
  inline void forEachDependent(F&& func) const {
    for (const Node* dependent : dependents_) {
      func(*dependent);
    }
  }
  template <class F>
  inline void forEachDependent(F&& func) {
    for (Node* dependent : dependents_) {
      func(*dependent);
    }
  }
  inline size_t numPredecessors() const {
    return numPredecessors_;
  }
  inline bool isCompleted() const {
    return numIncompletePredecessors_.load(std::memory_order_relaxed) == kCompleted;
  }
  inline bool setIncomplete() const {
    if (numIncompletePredecessors_.load(std::memory_order_relaxed) == kCompleted) {
      numIncompletePredecessors_.store(0, std::memory_order_relaxed);
      return true;
    }
    return false;
  }
 
  inline void setCompleted() const {
    numIncompletePredecessors_.store(kCompleted, std::memory_order_relaxed);
  }
 
 protected:
  template <class F, class X = std::enable_if_t<!std::is_base_of<Node, F>::value, void>>
  Node(F&& f) : numIncompletePredecessors_(0) {
    using FNoRef = typename std::remove_reference<F>::type;
 
    constexpr size_t kFuncSize = static_cast<size_t>(detail::nextPow2(sizeof(FNoRef)));
    funcBuffer_ = allocSmallBuffer<kFuncSize>();
    new (funcBuffer_) FNoRef(std::forward<F>(f));
    invoke_ = ::detail::callFunctor<FNoRef>;
    destroy_ = ::detail::destroyFunctor<FNoRef>;
  }
 
  void dependsOnOneNode(Node& node) {
    node.dependents_.emplace_back(this);
    numPredecessors_++;
  }
 
  static constexpr size_t kCompleted = std::numeric_limits<size_t>::max();
  mutable std::atomic<size_t> numIncompletePredecessors_;
  size_t numPredecessors_ = 0;
 
 private:
  using InvokerType = void (*)(void* ptr);
 
  InvokerType invoke_;
  InvokerType destroy_;
  char* funcBuffer_;
 
  std::vector<Node*> dependents_; // nodes depend on this
 
  template <class N>
  friend class SubgraphT;
  friend class ::detail::ExecutorBase;
  template <typename G>
  friend void setAllNodesIncomplete(const G& graph);
};
class BiPropNode : public Node {
 public:
  BiPropNode() = delete;
  BiPropNode(const BiPropNode&) = delete;
  BiPropNode& operator=(const BiPropNode&) = delete;
  BiPropNode(BiPropNode&& other) noexcept
      : Node(std::move(other)), biPropSet_(std::move(other.biPropSet_)) {}
  template <class... Ns>
  inline void biPropDependsOn(Ns&... nodes) {
    ((void)std::initializer_list<int>{(biPropDependsOnOneNode(nodes), 0)...});
  }
  inline bool isSameSet(const BiPropNode& node) const {
    return biPropSet_ && biPropSet_ == node.biPropSet_;
  }
 
 private:
  template <class T, class X = std::enable_if_t<!std::is_base_of<BiPropNode, T>::value, void>>
  BiPropNode(T&& f) : Node(std::forward<T>(f)) {}
  inline void removeFromBiPropSet() {
    if (biPropSet_ != nullptr) {
      auto it = std::find(biPropSet_->begin(), biPropSet_->end(), this);
      if (it != biPropSet_->end()) {
        biPropSet_->erase(it);
      }
    }
  }
 
  DISPENSO_DLL_ACCESS void biPropDependsOnOneNode(BiPropNode& node);
 
  std::shared_ptr<std::vector<const BiPropNode*>> biPropSet_;
 
  template <class N>
  friend class SubgraphT;
  friend class ::detail::ExecutorBase;
};
 
template <class N>
class GraphT;
 
template <class N>
class DISPENSO_DLL_ACCESS SubgraphT {
 public:
  using NodeType = N;
  SubgraphT() = delete;
  SubgraphT(const SubgraphT<N>&) = delete;
  SubgraphT<N>& operator=(const SubgraphT<N>&) = delete;
  SubgraphT(SubgraphT<N>&& other) noexcept
      : graph_(other.graph_),
        nodes_(std::move(other.nodes_)),
        allocator_(std::move(other.allocator_)) {}
  ~SubgraphT();
  template <class T>
  DISPENSO_REQUIRES(OnceCallableFunc<T>)
  N& addNode(T&& f) {
    nodes_.push_back(new (allocator_->alloc()) NodeType(std::forward<T>(f)));
    return *nodes_.back();
  }
  size_t numNodes() const {
    return nodes_.size();
  }
  const N& node(size_t index) const {
    return *nodes_[index];
  }
  N& node(size_t index) {
    return *nodes_[index];
  }
  template <class F>
  inline void forEachNode(F&& func) const {
    for (const N* node : nodes_) {
      func(*node);
    }
  }
  template <class F>
  inline void forEachNode(F&& func) {
    for (N* node : nodes_) {
      func(*node);
    }
  }
  void clear();
 
 private:
  using DeallocFunc = void (*)(NoLockPoolAllocator*);
  using PoolPtr = std::unique_ptr<NoLockPoolAllocator, DeallocFunc>;
 
  static constexpr size_t kNodeSizeP2 = static_cast<size_t>(detail::nextPow2(sizeof(NodeType)));
 
  explicit SubgraphT(GraphT<N>* graph) : graph_(graph), nodes_(), allocator_(getAllocator()) {}
 
  inline void removeNodeFromBiPropSet(Node* /* node */) {}
  void removeNodeFromBiPropSet(BiPropNode* node) {
    node->removeFromBiPropSet();
  }
  void decrementDependentCounters();
  size_t markNodesWithPredicessors();
  void removePredecessorDependencies(size_t numGraphPredecessors);
 
  void destroyNodes();
 
  static PoolPtr getAllocator();
  static void releaseAllocator(NoLockPoolAllocator* ptr);
 
  GraphT<N>* graph_;
#if defined(_WIN32) && !defined(__MINGW32__)
#pragma warning(push)
#pragma warning(disable : 4251)
#endif
  std::vector<N*> nodes_;
 
  PoolPtr allocator_;
#if defined(_WIN32) && !defined(__MINGW32__)
#pragma warning(pop)
#endif
 
  template <class T>
  friend class GraphT;
};
 
template <class N>
class DISPENSO_DLL_ACCESS GraphT {
 public:
  using NodeType = N;
  using SubgraphType = SubgraphT<N>;
  GraphT(const GraphT<N>&) = delete;
  GraphT& operator=(const GraphT<N>&) = delete;
  GraphT() {
    subgraphs_.push_back(SubgraphType(this));
  }
  GraphT(GraphT<N>&& other);
  GraphT<N>& operator=(GraphT&& other) noexcept;
  template <class T>
  DISPENSO_REQUIRES(OnceCallableFunc<T>)
  N& addNode(T&& f) {
    return subgraphs_[0].addNode(std::forward<T>(f));
  }
  size_t numNodes() const {
    return subgraphs_[0].numNodes();
  }
  const N& node(size_t index) const {
    return subgraphs_[0].node(index);
  }
  N& node(size_t index) {
    return subgraphs_[0].node(index);
  }
  SubgraphT<N>& addSubgraph();
  size_t numSubgraphs() const {
    return subgraphs_.size();
  }
  const SubgraphT<N>& subgraph(size_t index) const {
    return subgraphs_[index];
  }
  SubgraphT<N>& subgraph(size_t index) {
    return subgraphs_[index];
  }
  template <class F>
  inline void forEachSubgraph(F&& func) const {
    for (const SubgraphT<N>& subgraph : subgraphs_) {
      func(subgraph);
    }
  }
  template <class F>
  inline void forEachSubgraph(F&& func) {
    for (SubgraphT<N>& subgraph : subgraphs_) {
      func(subgraph);
    }
  }
  template <class F>
  inline void forEachNode(F&& func) const {
    for (const SubgraphT<N>& subgraph : subgraphs_) {
      for (const N* node : subgraph.nodes_) {
        func(*node);
      }
    }
  }
  template <class F>
  inline void forEachNode(F&& func) {
    for (SubgraphT<N>& subgraph : subgraphs_) {
      for (N* node : subgraph.nodes_) {
        func(*node);
      }
    }
  }
  inline void clear() {
    subgraphs_.clear();
    subgraphs_.push_back(SubgraphType(this));
  }
  inline void clearSubgraphs() {
    for (SubgraphT<N>& subgraph : subgraphs_) {
      subgraph.destroyNodes();
    }
  }
 
 private:
  static constexpr size_t kSubgraphSizeP2 =
      static_cast<size_t>(detail::nextPow2(sizeof(SubgraphType)));
 
#if defined(_WIN32) && !defined(__MINGW32__)
#pragma warning(push)
#pragma warning(disable : 4251)
#endif
  std::deque<SubgraphT<N>> subgraphs_;
#if defined(_WIN32) && !defined(__MINGW32__)
#pragma warning(pop)
#endif
 
  template <class T>
  friend class SubgraphT;
};
 
using Graph = GraphT<Node>;
using BiPropGraph = GraphT<BiPropNode>;
 
using Subgraph = SubgraphT<Node>;
using BiPropSubgraph = SubgraphT<BiPropNode>;
} // namespace dispenso