Adds multithreading.

.vscode/settings.json

@@ -83,6 +83,7 @@
         "stack": "cpp",
         "cfenv": "cpp",
         "typeindex": "cpp",
-        "valarray": "cpp"
+        "valarray": "cpp",
+        "csignal": "cpp"
     }
 }

CMakeLists.txt

@@ -6,7 +6,8 @@ find_package(Threads REQUIRED)
 
 add_executable(Sorter framework/runner.cpp
   src/container.cpp src/container.hpp
-  src/sorter.cpp src/sorter.hpp)
+  src/sorter.cpp src/sorter.hpp
+  src/thread_pool.cpp src/thread_pool.hpp)
 target_link_libraries(Sorter PUBLIC Threads::Threads)
 
 target_compile_features(Sorter PRIVATE cxx_std_20)
@@ -50,6 +51,8 @@ add_custom_command(
     ${CMAKE_CURRENT_LIST_DIR}/src/container.hpp
     ${CMAKE_CURRENT_LIST_DIR}/src/sorter.cpp
     ${CMAKE_CURRENT_LIST_DIR}/src/sorter.hpp
+    ${CMAKE_CURRENT_LIST_DIR}/src/thread_pool.cpp
+    ${CMAKE_CURRENT_LIST_DIR}/src/thread_pool.hpp
     ${plot_file}
     ${CMAKE_CURRENT_LIST_DIR}/description.md
   COMMENT "Creating submission"
@@ -71,6 +74,8 @@ set(dist_file_list
   src/container.hpp
   src/sorter.cpp
   src/sorter.hpp
+  src/thread_pool.cpp
+  src/thread_pool.hpp
 )
 
 set(framework_dist_file "ae-sorting.zip")

eval.py

@@ -7,7 +7,7 @@ from pathlib import Path
 
 def run_experiment(output_file, build_dir):
     # The number of threads is not currently used, it's just here in case you want to parallelize your code.
-    for threads in [1]:
+    for threads in [1, 2, 4, 8, 12, 16]:
         # for size in [1e2, 1e3, 1e4 + 1, 1e5, 1e6 - 1, 1e7]:
         for size in [1e2, 1e3, 1e4 + 1, 1e5, 1e7]:
             print("Measuring p=" + str(threads) + " n=" + str(size))

framework/runner.cpp

@@ -62,7 +62,7 @@ void runExperiment(std::string_view name,
     std::chrono::steady_clock::time_point ctor = std::chrono::steady_clock::now();
     auto to_sort = container_factory(input);
     std::chrono::steady_clock::time_point begin = std::chrono::steady_clock::now();
-    sort_func(to_sort);
+    sort_func(to_sort, num_threads);
     std::chrono::steady_clock::time_point end = std::chrono::steady_clock::now();
     totalNanoseconds +=
         std::chrono::duration_cast<std::chrono::nanoseconds>(end - begin)
@@ -95,8 +95,8 @@ int main(int argc, char **argv) {
                 [](const auto& data) {
                   return ae::container(data);
                 },
-                [](ae::container& data) {
-                  ae::sorter{}.sort(data);
+                [](ae::container& data, auto num_threads) {
+                  ae::sorter(num_threads).sort(data);
                 }, argc, argv);
 
   return 0;

result.txt

@@ -1,5 +1,30 @@
-RESULT name=sort n=100 t=1 iterations=301 durationNanoseconds=3324 totalDurationNanoseconds=1000561 constructorNanoseconds=274 totalConstructorNanoseconds=82660
-RESULT name=sort n=1000 t=1 iterations=13 durationNanoseconds=77557 totalDurationNanoseconds=1008241 constructorNanoseconds=2057 totalConstructorNanoseconds=26750
-RESULT name=sort n=10001 t=1 iterations=1 durationNanoseconds=1509911 totalDurationNanoseconds=1509911 constructorNanoseconds=108831 totalConstructorNanoseconds=108831
-RESULT name=sort n=100000 t=1 iterations=1 durationNanoseconds=8070546 totalDurationNanoseconds=8070546 constructorNanoseconds=488620 totalConstructorNanoseconds=488620
-RESULT name=sort n=10000000 t=1 iterations=1 durationNanoseconds=723407878 totalDurationNanoseconds=723407878 constructorNanoseconds=51148616 totalConstructorNanoseconds=51148616
+RESULT name=sort n=100 t=1 iterations=18 durationNanoseconds=57856 totalDurationNanoseconds=1041410 constructorNanoseconds=513 totalConstructorNanoseconds=9250
+RESULT name=sort n=1000 t=1 iterations=8 durationNanoseconds=126883 totalDurationNanoseconds=1015070 constructorNanoseconds=1475 totalConstructorNanoseconds=11800
+RESULT name=sort n=10001 t=1 iterations=2 durationNanoseconds=615565 totalDurationNanoseconds=1231131 constructorNanoseconds=30890 totalConstructorNanoseconds=61780
+RESULT name=sort n=100000 t=1 iterations=1 durationNanoseconds=9488107 totalDurationNanoseconds=9488107 constructorNanoseconds=524961 totalConstructorNanoseconds=524961
+RESULT name=sort n=10000000 t=1 iterations=1 durationNanoseconds=1416091993 totalDurationNanoseconds=1416091993 constructorNanoseconds=50440746 totalConstructorNanoseconds=50440746
+RESULT name=sort n=100 t=2 iterations=16 durationNanoseconds=63258 totalDurationNanoseconds=1012141 constructorNanoseconds=375 totalConstructorNanoseconds=6000
+RESULT name=sort n=1000 t=2 iterations=7 durationNanoseconds=154110 totalDurationNanoseconds=1078770 constructorNanoseconds=2008 totalConstructorNanoseconds=14060
+RESULT name=sort n=10001 t=2 iterations=3 durationNanoseconds=451387 totalDurationNanoseconds=1354161 constructorNanoseconds=19620 totalConstructorNanoseconds=58860
+RESULT name=sort n=100000 t=2 iterations=1 durationNanoseconds=6236655 totalDurationNanoseconds=6236655 constructorNanoseconds=514650 totalConstructorNanoseconds=514650
+RESULT name=sort n=10000000 t=2 iterations=1 durationNanoseconds=1380325518 totalDurationNanoseconds=1380325518 constructorNanoseconds=50373886 totalConstructorNanoseconds=50373886
+RESULT name=sort n=100 t=4 iterations=9 durationNanoseconds=118743 totalDurationNanoseconds=1068691 constructorNanoseconds=436 totalConstructorNanoseconds=3930
+RESULT name=sort n=1000 t=4 iterations=4 durationNanoseconds=272115 totalDurationNanoseconds=1088461 constructorNanoseconds=2415 totalConstructorNanoseconds=9660
+RESULT name=sort n=10001 t=4 iterations=2 durationNanoseconds=569255 totalDurationNanoseconds=1138510 constructorNanoseconds=29920 totalConstructorNanoseconds=59840
+RESULT name=sort n=100000 t=4 iterations=1 durationNanoseconds=6598125 totalDurationNanoseconds=6598125 constructorNanoseconds=507180 totalConstructorNanoseconds=507180
+RESULT name=sort n=10000000 t=4 iterations=1 durationNanoseconds=1300242690 totalDurationNanoseconds=1300242690 constructorNanoseconds=50475097 totalConstructorNanoseconds=50475097
+RESULT name=sort n=100 t=8 iterations=3 durationNanoseconds=347863 totalDurationNanoseconds=1043591 constructorNanoseconds=2706 totalConstructorNanoseconds=8120
+RESULT name=sort n=1000 t=8 iterations=2 durationNanoseconds=610620 totalDurationNanoseconds=1221241 constructorNanoseconds=10400 totalConstructorNanoseconds=20800
+RESULT name=sort n=10001 t=8 iterations=2 durationNanoseconds=706495 totalDurationNanoseconds=1412991 constructorNanoseconds=29600 totalConstructorNanoseconds=59200
+RESULT name=sort n=100000 t=8 iterations=1 durationNanoseconds=7387085 totalDurationNanoseconds=7387085 constructorNanoseconds=557391 totalConstructorNanoseconds=557391
+RESULT name=sort n=10000000 t=8 iterations=1 durationNanoseconds=1261560682 totalDurationNanoseconds=1261560682 constructorNanoseconds=49470756 totalConstructorNanoseconds=49470756
+RESULT name=sort n=100 t=12 iterations=3 durationNanoseconds=432037 totalDurationNanoseconds=1296111 constructorNanoseconds=1170 totalConstructorNanoseconds=3510
+RESULT name=sort n=1000 t=12 iterations=1 durationNanoseconds=1092461 totalDurationNanoseconds=1092461 constructorNanoseconds=12880 totalConstructorNanoseconds=12880
+RESULT name=sort n=10001 t=12 iterations=1 durationNanoseconds=1019941 totalDurationNanoseconds=1019941 constructorNanoseconds=54540 totalConstructorNanoseconds=54540
+RESULT name=sort n=100000 t=12 iterations=1 durationNanoseconds=7159465 totalDurationNanoseconds=7159465 constructorNanoseconds=536730 totalConstructorNanoseconds=536730
+RESULT name=sort n=10000000 t=12 iterations=1 durationNanoseconds=1503813105 totalDurationNanoseconds=1503813105 constructorNanoseconds=50150056 totalConstructorNanoseconds=50150056
+RESULT name=sort n=100 t=16 iterations=3 durationNanoseconds=432706 totalDurationNanoseconds=1298120 constructorNanoseconds=3833 totalConstructorNanoseconds=11500
+RESULT name=sort n=1000 t=16 iterations=2 durationNanoseconds=784875 totalDurationNanoseconds=1569751 constructorNanoseconds=5285 totalConstructorNanoseconds=10570
+RESULT name=sort n=10001 t=16 iterations=1 durationNanoseconds=1953311 totalDurationNanoseconds=1953311 constructorNanoseconds=59420 totalConstructorNanoseconds=59420
+RESULT name=sort n=100000 t=16 iterations=1 durationNanoseconds=6820104 totalDurationNanoseconds=6820104 constructorNanoseconds=524961 totalConstructorNanoseconds=524961
+RESULT name=sort n=10000000 t=16 iterations=1 durationNanoseconds=1352546567 totalDurationNanoseconds=1352546567 constructorNanoseconds=52707158 totalConstructorNanoseconds=52707158

src/sorter.cpp

@@ -11,6 +11,11 @@
 
 namespace ae {
 
+sorter::sorter(uint32_t num = 1) {
+  sorter::num_threads = num;
+  sorter::pool = new ThreadPool(num);
+}
+
 void sorter::sort(container& data) {
   for (auto i = 1uz; i < data.placeholder_.size(); ++i) {
     std::ranges::copy(data.placeholder_[i], std::back_inserter(data.placeholder_[0]));
@@ -18,7 +23,7 @@ void sorter::sort(container& data) {
   }
   #if DEBUG
     for (int i = 0; i < data.placeholder_[0].size(); i++) {
-      // if (copy[i] != data.placeholder_[0][i])
+      if (copy[i] != data.placeholder_[0][i])
       std::cerr << i << " before:" << data.placeholder_[0][i] << std::endl;
     }
 
@@ -27,6 +32,7 @@ void sorter::sort(container& data) {
     std::sort(copy.begin(), copy.end());
   #endif
   sorter::msd_inplace_radix_sort(data.placeholder_[0], 0, [&](auto span) {sorter::robin_hood_sort(span);});
+  while (sorter::pool->size() > 0 || sorter::pool->isWorking()) {};
   #if DEBUG
     for (int i = 0; i < copy.size(); i++) {
       if (copy[i] != data.placeholder_[0][i])
@@ -169,9 +175,27 @@ void sorter::msd_inplace_radix_sort(
 
   // sort each bucket recursively
   for (auto i = 0; i < sorter::RADIX_BUCKETS; i++) {
+    if (sorter::pool != nullptr) {
+      #if DEBUG
+        std::cerr << "Putting in task with depth " << passes << " of bucket " << i << std::endl;
+      #endif
+      auto start = buckets_start[i];
+      auto end = buckets_end[i];
+      sorter::pool->add([start, end, &bucket_sort, passes, this, i](){
+        #if DEBUG
+          std::cerr << "Starting task with depth " << passes << " of bucket " << i << std::endl;
+        #endif
+        sorter::msd_inplace_radix_sort(std::span<container::element_type> (start, end), passes + 1, bucket_sort);
+        #if DEBUG
+          std::cerr << "Finishing task with depth " << passes << " of bucket " << i << std::endl;
+        #endif
+      });
+    } else {
       sorter::msd_inplace_radix_sort(std::span<container::element_type> (buckets_start[i], buckets_end[i]), passes + 1, bucket_sort);
     }
+  }
 }
+
 void sorter::robin_hood_sort(std::span<container::element_type> bucket) {
   const auto size = bucket.size() + sorter::OVERHEAD_SIZE;
   const auto mask = ((1L) << (sizeof(container::element_type) * CHAR_BIT - sorter::RADIX_ITERATIONS)) - 1;

src/sorter.hpp

@@ -1,8 +1,10 @@
 #pragma once
 
 #include "container.hpp"
+#include "thread_pool.hpp"
 #include "functional"
 #include "math.h"
+#include <mutex>
 
 namespace ae {
 
@@ -11,12 +13,21 @@ class sorter {
   void sort(container& data);
 
   // TODO You may add additional functions or data members to the sorter.
+
+  sorter(uint32_t num_threads);
+
   void msd_inplace_radix_sort(
     std::span<container::element_type> range, 
     size_t passes, 
     const std::function<void(std::span<container::element_type> bucket)>& bucket_sort
   );
 
+  void parallel_msd_inplace_radix_sort(
+    std::span<container::element_type> range,
+    size_t passes, 
+    const std::function<void(std::span<container::element_type> bucket)>& bucket_sort
+  );
+
   void msd_inplace_radix_sort_binary(
     std::span<container::element_type> range, 
     size_t passes, 
@@ -24,10 +35,12 @@ class sorter {
   );
 
   const uint32_t OVERHEAD_SIZE = 100L;
-  const uint32_t SMALL_SORT_THRESHHOLD = 32;
+  const uint32_t SMALL_SORT_THRESHHOLD = 100;
   const uint32_t RADIX_SIZE = 4;
   const uint32_t RADIX_BUCKETS = std::pow(2, 4);
   const uint32_t RADIX_ITERATIONS = 8;
+  uint32_t num_threads;
+  ThreadPool* pool = nullptr;
 
   void robin_hood_sort(std::span<container::element_type> range);
 };

src/thread_pool.cpp

@@ -0,0 +1,58 @@
+#include "thread_pool.hpp"
+
+ThreadPool::ThreadPool(size_t num_threads) {
+    states = new bool[num_threads]({ false });
+    for (auto i = 0; i < num_threads; ++i) {
+        threads.emplace_back([this, i] {
+            while (true) {
+                std::function<void()> task;
+
+                std::unique_lock<std::mutex> lock(mutex);
+
+                cv.wait(lock, [this] { return !tasks.empty() || stop; });
+
+                if (tasks.empty() || stop) {
+                    return;
+                }
+
+                states[i] = true;
+
+                task = std::move(tasks.front());
+                tasks.pop();
+
+                lock.unlock();
+                task();
+
+                states[i] = false;
+            }
+        });
+    }
+}
+
+ThreadPool::~ThreadPool() {
+    std::unique_lock<std::mutex> lock(mutex);
+    stop = true;
+    lock.unlock();
+
+    cv.notify_all();
+    for (auto& thread : threads) {
+        thread.join();
+    }
+}
+
+void ThreadPool::add(std::function<void()> task) {
+    std::unique_lock<std::mutex> lock(mutex);
+    tasks.emplace(std::move(task));
+    cv.notify_one();
+    lock.unlock();
+}
+
+uint32_t ThreadPool::size() { return tasks.size(); }
+bool ThreadPool::isWorking() { 
+    for (auto i = 0; i < threads.size(); i++) {
+        if (states[i]) {
+            return true;
+        }
+    }
+    return false;
+}

src/thread_pool.hpp

@@ -0,0 +1,29 @@
+#pragma once
+
+#include <algorithm>
+#include <vector>
+#include <mutex>
+#include <queue>
+#include <thread>
+#include <functional>
+#include <condition_variable>
+
+// This class was inspired by https://www.geeksforgeeks.org/cpp/thread-pool-in-cpp/ (access: 26/09/2025)
+// to more efficiently handle threads
+
+class ThreadPool {
+private:
+    std::vector<std::thread> threads;
+    std::queue<std::function<void()>> tasks;
+    std::mutex mutex;
+    std::condition_variable cv;
+    bool stop = false;
+    bool* states;
+
+public:
+    ThreadPool(size_t num_threads);
+    ~ThreadPool();
+    void add(std::function<void()> task);
+    uint32_t size();
+    bool isWorking();
+};