Add unbalanced tree search benchmakr.

CMakeLists.txt

@@ -7,7 +7,7 @@ set(CMAKE_CXX_STANDARD 11)
 
 # seperate library and test/example executable output paths.
 set(EXECUTABLE_OUTPUT_PATH ${CMAKE_BINARY_DIR}/bin)
-set(LIBRARY_OUTPUT_PATH  ${CMAKE_BINARY_DIR}/lib)
+set(LIBRARY_OUTPUT_PATH ${CMAKE_BINARY_DIR}/lib)
 
 # specific setup code is located in individual files.
 include(cmake/DisabelInSource.cmake)
@@ -34,11 +34,12 @@ add_subdirectory(app/playground)
 add_subdirectory(app/test_for_new)
 add_subdirectory(app/invoke_parallel)
 add_subdirectory(app/benchmark_fft)
+add_subdirectory(app/benchmark_unbalanced)
 
 # Add optional tests
 option(PACKAGE_TESTS "Build the tests" ON)
-if(PACKAGE_TESTS)
+if (PACKAGE_TESTS)
     enable_testing()
     add_subdirectory(test)
     add_test(NAME AllTests COMMAND tests)
-endif()
+endif ()

app/benchmark_fft/main.cpp

@@ -8,7 +8,7 @@
 
 static constexpr int CUTOFF = 16;
 static constexpr int NUM_ITERATIONS = 1000;
-static constexpr int INPUT_SIZE = 2064;
+static constexpr int INPUT_SIZE = 8192;
 typedef std::vector<std::complex<double>> complex_vector;
 
 void divide(complex_vector::iterator data, int n) {

app/benchmark_unbalanced/CMakeLists.txt

+add_executable(benchmark_unbalanced main.cpp node.h node.cpp picosha2.h)
+target_link_libraries(benchmark_unbalanced pls)
+if (EASY_PROFILER)
+    target_link_libraries(benchmark_unbalanced easy_profiler)
+endif ()

app/benchmark_unbalanced/LICENSE_PICOSA2

+MIT License
+
+Copyright (c) 2017 okdshin
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
\ No newline at end of file

app/benchmark_unbalanced/main.cpp

+#include <pls/pls.h>
+#include <pls/internal/helpers/profiler.h>
+#include <pls/internal/helpers/mini_benchmark.h>
+
+#include "node.h"
+
+const int SEED = 42;
+const int ROOT_CHILDREN = 140;
+const double Q = 0.124875;
+const int NORMAL_CHILDREN = 8;
+
+const int NUM_NODES = 71069;
+
+int count_child_nodes(uts::node &node) {
+  int child_count = 1;
+  std::vector<uts::node> children = node.spawn_child_nodes();
+
+  if (children.empty()) {
+    return child_count;
+  }
+
+  auto current_task = pls::fork_join_sub_task::current();
+  std::vector<int> results(children.size());
+  for (size_t i = 0; i < children.size(); i++) {
+    size_t index = i;
+    auto lambda = [&, index] { results[index] = count_child_nodes(children[index]); };
+    pls::fork_join_lambda_by_value<typeof(lambda)> sub_task(lambda);
+    current_task->spawn_child(sub_task);
+  }
+  current_task->wait_for_all();
+  for (auto result : results) {
+    child_count += result;
+  }
+
+  return child_count;
+}
+
+int unbalanced_tree_search(int seed, int root_children, double q, int normal_children) {
+  static auto id = pls::unique_id::create(42);
+  int result;
+
+  auto lambda = [&] {
+    uts::node root(seed, root_children, q, normal_children);
+    result = count_child_nodes(root);
+  };
+  pls::fork_join_lambda_by_reference<typeof(lambda)> task(lambda);
+  pls::fork_join_lambda_by_reference<typeof(lambda)> sub_task(lambda);
+  pls::fork_join_task root_task{&sub_task, id};
+  pls::scheduler::execute_task(root_task);
+
+  return result;
+}
+
+int main() {
+  PROFILE_ENABLE
+  pls::internal::helpers::run_mini_benchmark([&] {
+    unbalanced_tree_search(SEED, ROOT_CHILDREN, Q, NORMAL_CHILDREN);
+  }, 8, 4000);
+
+  PROFILE_SAVE("test_profile.prof")
+}
+
+//int main() {
+//  PROFILE_ENABLE
+//  pls::malloc_scheduler_memory my_scheduler_memory{8, 2u << 18};
+//  pls::scheduler scheduler{&my_scheduler_memory, 8};
+//
+//  scheduler.perform_work([&] {
+//    PROFILE_MAIN_THREAD
+//    for (int i = 0; i < 10; i++) {
+//      PROFILE_WORK_BLOCK("Top Level")
+//      int result = unbalanced_tree_search(SEED, ROOT_CHILDREN, Q, NORMAL_CHILDREN);
+//      std::cout << result << std::endl;
+//    }
+//  });
+//
+//  PROFILE_SAVE("test_profile.prof")
+//}

app/benchmark_unbalanced/node.cpp

+#include "node.h"
+
+namespace uts {
+node_state node::generate_child_state(uint32_t index) {
+  node_state result;
+
+  picosha2::hash256_one_by_one hasher;
+  hasher.process(state_.begin(), state_.end());
+  auto index_begin = reinterpret_cast<uint8_t *>(&index);
+  hasher.process(index_begin, index_begin + 4);
+  hasher.finish();
+  hasher.get_hash_bytes(result.begin(), result.end());
+
+  return result;
+}
+
+double node::get_state_random() {
+  int32_t state_random_integer;
+  uint32_t b = ((uint32_t) state_[16] << 24) |
+      ((uint32_t) state_[17] << 16) |
+      ((uint32_t) state_[18] << 8) |
+      ((uint32_t) state_[19] << 0);
+  b = b & 0x7fffffff; // Mask out negative values
+  state_random_integer = static_cast<int32_t>(b);
+
+  return (double) state_random_integer / (double) INT32_MAX;
+}
+}

app/benchmark_unbalanced/node.h

+
+#ifndef UTS_NODE_H
+#define UTS_NODE_H
+
+#include <cstdint>
+#include <array>
+#include <vector>
+
+#include "picosha2.h"
+
+namespace uts {
+using node_state = std::array<uint8_t, 20>;
+
+/**
+ * Node of an unballanced binomial tree (https://www.cs.unc.edu/~olivier/LCPC06.pdf).
+ * To build up the tree recursivly call spawn_child_nodes on each node until leaves are reached.
+ * The tree is not built up directly in memory, but rather by the recursive calls.
+ */
+class node {
+  // The state is used to allow a deterministic tree construction using sha256 hashes.
+  node_state state_;
+
+  // Set this to a positive number for the root node to start the tree with a specific size
+  int root_children_;
+
+  // general branching factors
+  double q_;
+  int b_;
+
+  // Private constructor for children
+  node(node_state state, double q, int b) : state_{state}, root_children_{-1}, q_{q}, b_{b} {}
+
+  std::array<uint8_t, 20> generate_child_state(uint32_t index);
+  double get_state_random();
+
+ public:
+  node(int seed, int root_children, double q, int b) : state_({{}}), root_children_{root_children}, q_{q}, b_{b} {
+    for (int i = 0; i < 16; i++) {
+      state_[i] = 0;
+    }
+    state_[16] = static_cast<uint8_t>(0xFF & (seed >> 24));
+    state_[17] = static_cast<uint8_t>(0xFF & (seed >> 16));
+    state_[18] = static_cast<uint8_t>(0xFF & (seed >> 8));
+    state_[19] = static_cast<uint8_t>(0xFF & (seed >> 0));
+
+    picosha2::hash256_one_by_one hasher;
+    hasher.process(state_.begin(), state_.end());
+    hasher.finish();
+    hasher.get_hash_bytes(state_.begin(), state_.end());
+  }
+
+  std::vector<node> spawn_child_nodes() {
+    double state_random = get_state_random();
+    int num_children;
+    if (root_children_ > 0) {
+      num_children = root_children_; // Root always spawns children
+    } else if (state_random < q_) {
+      num_children = b_;
+    } else {
+      num_children = 0;
+    }
+
+    std::vector<node> result;
+    for (int i = 0; i < num_children; i++) {
+      result.push_back(node(generate_child_state(i), q_, b_));
+    }
+
+    return result;
+  }
+};
+}
+
+#endif //UTS_NODE_H

app/invoke_parallel/main.cpp

@@ -6,7 +6,7 @@
 #include <vector>
 
 static constexpr int CUTOFF = 16;
-static constexpr int INPUT_SIZE = 2064;
+static constexpr int INPUT_SIZE = 8192;
 typedef std::vector<std::complex<double>> complex_vector;
 
 void divide(complex_vector::iterator data, int n) {

lib/pls/include/pls/algorithms/invoke_parallel_impl.h

@@ -2,6 +2,7 @@
 #ifndef PLS_INVOKE_PARALLEL_IMPL_H
 #define PLS_INVOKE_PARALLEL_IMPL_H
 
+#include <pls/internal/scheduling/fork_join_task.h>
 #include "pls/internal/scheduling/fork_join_task.h"
 #include "pls/internal/scheduling/scheduler.h"
 #include "pls/internal/helpers/unique_id.h"
@@ -19,8 +20,7 @@ inline void run_body(const Body &internal_body, const abstract_task::id &id) {
   // if not we will spawn it as a new 'fork-join-style' task.
   auto current_task = scheduler::current_task();
   if (current_task->unique_id() == id) {
-    auto current_sub_task = reinterpret_cast<fork_join_task *>(current_task)->currently_executing();
-    internal_body(current_sub_task);
+    internal_body();
   } else {
     fork_join_lambda_by_reference<Body> root_body(&internal_body);
     fork_join_task root_task{&root_body, id};
@@ -37,8 +37,7 @@ void invoke_parallel(const Function1 &function1, const Function2 &function2) {
   static abstract_task::id id = unique_id::create<Function1, Function2>();
 
   auto internal_body = [&](fork_join_sub_task *this_task) {
-    auto sub_task_body_2 = [&](fork_join_sub_task *) { function2(); };
-    auto sub_task_2 = fork_join_lambda_by_reference<decltype(sub_task_body_2)>(&sub_task_body_2);
+    auto sub_task_2 = fork_join_lambda_by_reference<Function2>(function2);
 
     this_task->spawn_child(sub_task_2);
     function1(); // Execute first function 'inline' without spawning a sub_task object
@@ -54,16 +53,15 @@ void invoke_parallel(const Function1 &function1, const Function2 &function2, con
   using namespace ::pls::internal::helpers;
   static abstract_task::id id = unique_id::create<Function1, Function2, Function3>();
 
-  auto internal_body = [&](fork_join_sub_task *this_task) {
-    auto sub_task_body_2 = [&](fork_join_sub_task *) { function2(); };
-    auto sub_task_2 = fork_join_lambda_by_reference<decltype(sub_task_body_2)>(&sub_task_body_2);
-    auto sub_task_body_3 = [&](fork_join_sub_task *) { function3(); };
-    auto sub_task_3 = fork_join_lambda_by_reference<decltype(sub_task_body_3)>(&sub_task_body_3);
+  auto internal_body = [&]() {
+    auto current_task = fork_join_sub_task::current();
+    auto sub_task_2 = fork_join_lambda_by_reference<Function2>(function2);
+    auto sub_task_3 = fork_join_lambda_by_reference<Function3>(function3);
 
-    this_task->spawn_child(sub_task_2);
-    this_task->spawn_child(sub_task_3);
+    current_task->spawn_child(sub_task_2);
+    current_task->spawn_child(sub_task_3);
     function1(); // Execute first function 'inline' without spawning a sub_task object
-    this_task->wait_for_all();
+    current_task->wait_for_all();
   };
 
   internal::run_body(internal_body, id);

lib/pls/include/pls/internal/scheduling/fork_join_task.h

@@ -43,20 +43,21 @@ class fork_join_sub_task {
   void spawn_child(T &sub_task);
   void wait_for_all();
 
+  static fork_join_sub_task *current();
  private:
   void execute();
 };
 
 template<typename Function>
 class fork_join_lambda_by_reference : public fork_join_sub_task {
-  const Function *function_;
+  const Function &function_;
 
  public:
-  explicit fork_join_lambda_by_reference(const Function *function) : fork_join_sub_task{}, function_{function} {};
+  explicit fork_join_lambda_by_reference(const Function &function) : fork_join_sub_task{}, function_{function} {};
 
  protected:
   void execute_internal() override {
-    (*function_)(this);
+    function_();
   }
 };
 
@@ -69,7 +70,7 @@ class fork_join_lambda_by_value : public fork_join_sub_task {
 
  protected:
   void execute_internal() override {
-    function_(this);
+    function_();
   }
 };
 

lib/pls/include/pls/pls.h

@@ -18,6 +18,8 @@ using task_id = internal::scheduling::abstract_task::id;
 using unique_id = internal::helpers::unique_id;
 
 using internal::scheduling::fork_join_sub_task;
+using internal::scheduling::fork_join_lambda_by_reference;
+using internal::scheduling::fork_join_lambda_by_value;
 using internal::scheduling::fork_join_task;
 
 using algorithm::invoke_parallel;

lib/pls/src/internal/scheduling/fork_join_task.cpp

@@ -66,6 +66,10 @@ fork_join_sub_task *fork_join_task::get_stolen_sub_task() {
   return deque_.pop_head();
 }
 
+fork_join_sub_task *fork_join_sub_task::current() {
+  return dynamic_cast<fork_join_task *>(scheduler::current_task())->currently_executing();
+}
+
 bool fork_join_task::internal_stealing(abstract_task *other_task) {
   PROFILE_STEALING("fork_join_task::internal_stealin")
   auto cast_other_task = reinterpret_cast<fork_join_task *>(other_task);