VideoCommon: Add AsyncShaderCompiler class implementation

5 files changed, 308 insertions, 1 deletions

diff --git a/Source/Core/VideoCommon/AsyncShaderCompiler.cpp b/Source/Core/VideoCommon/AsyncShaderCompiler.cpp
new file mode 100644
index 0000000000..15eececc4c
--- /dev/null
+++ b/Source/Core/VideoCommon/AsyncShaderCompiler.cpp
@@ -0,0 +1,215 @@
+// Copyright 2017 Dolphin Emulator Project
+// Licensed under GPLv2+
+// Refer to the license.txt file included.
+
+#include "VideoCommon/AsyncShaderCompiler.h"
+#include <thread>
+#include "Common/Assert.h"
+#include "Common/Logging/Log.h"
+
+namespace VideoCommon
+{
+AsyncShaderCompiler::AsyncShaderCompiler()
+{
+}
+
+AsyncShaderCompiler::~AsyncShaderCompiler()
+{
+  // Pending work can be left at shutdown.
+  // The work item classes are expected to clean up after themselves.
+  _assert_(!HasWorkerThreads());
+  _assert_(m_completed_work.empty());
+}
+
+void AsyncShaderCompiler::QueueWorkItem(WorkItemPtr item)
+{
+  // If no worker threads are available, compile synchronously.
+  if (!HasWorkerThreads())
+  {
+    item->Compile();
+    m_completed_work.push_back(std::move(item));
+  }
+  else
+  {
+    std::lock_guard<std::mutex> guard(m_pending_work_lock);
+    m_pending_work.push_back(std::move(item));
+    m_worker_thread_wake.notify_one();
+  }
+}
+
+void AsyncShaderCompiler::RetrieveWorkItems()
+{
+  std::deque<WorkItemPtr> completed_work;
+  {
+    std::lock_guard<std::mutex> guard(m_completed_work_lock);
+    m_completed_work.swap(completed_work);
+  }
+
+  while (!completed_work.empty())
+  {
+    completed_work.front()->Retrieve();
+    completed_work.pop_front();
+  }
+}
+
+bool AsyncShaderCompiler::HasPendingWork()
+{
+  std::lock_guard<std::mutex> guard(m_pending_work_lock);
+  return !m_pending_work.empty() || m_busy_workers.load() != 0;
+}
+
+void AsyncShaderCompiler::WaitUntilCompletion()
+{
+  while (HasPendingWork())
+    std::this_thread::sleep_for(std::chrono::milliseconds(1));
+}
+
+void AsyncShaderCompiler::WaitUntilCompletion(
+    const std::function<void(size_t, size_t)>& progress_callback)
+{
+  if (!HasPendingWork())
+    return;
+
+  // Wait a second before opening a progress dialog.
+  // This way, if the operation completes quickly, we don't annoy the user.
+  constexpr u32 CHECK_INTERVAL_MS = 50;
+  constexpr auto CHECK_INTERVAL = std::chrono::milliseconds(CHECK_INTERVAL_MS);
+  for (u32 i = 0; i < (1000 / CHECK_INTERVAL_MS); i++)
+  {
+    std::this_thread::sleep_for(std::chrono::milliseconds(CHECK_INTERVAL));
+    if (!HasPendingWork())
+      return;
+  }
+
+  // Grab the number of pending items. We use this to work out how many are left.
+  size_t total_items = 0;
+  {
+    // Safe to hold both locks here, since nowhere else does.
+    std::lock_guard<std::mutex> pending_guard(m_pending_work_lock);
+    std::lock_guard<std::mutex> completed_guard(m_completed_work_lock);
+    total_items = m_completed_work.size() + m_pending_work.size() + m_busy_workers.load() + 1;
+  }
+
+  // Update progress while the compiles complete.
+  for (;;)
+  {
+    size_t remaining_items;
+    {
+      std::lock_guard<std::mutex> pending_guard(m_pending_work_lock);
+      if (m_pending_work.empty() && !m_busy_workers.load())
+        break;
+      remaining_items = m_pending_work.size();
+    }
+
+    progress_callback(total_items - remaining_items, total_items);
+    std::this_thread::sleep_for(CHECK_INTERVAL);
+  }
+}
+
+void AsyncShaderCompiler::StartWorkerThreads(u32 num_worker_threads)
+{
+  for (u32 i = 0; i < num_worker_threads; i++)
+  {
+    void* thread_param = nullptr;
+    if (!WorkerThreadInitMainThread(&thread_param))
+    {
+      WARN_LOG(VIDEO, "Failed to initialize shader compiler worker thread.");
+      break;
+    }
+
+    m_worker_thread_start_result.store(false);
+
+    std::thread thr(&AsyncShaderCompiler::WorkerThreadEntryPoint, this, thread_param);
+    m_init_event.Wait();
+
+    if (!m_worker_thread_start_result.load())
+    {
+      WARN_LOG(VIDEO, "Failed to start shader compiler worker thread.");
+      thr.join();
+      break;
+    }
+
+    m_worker_threads.push_back(std::move(thr));
+  }
+}
+
+bool AsyncShaderCompiler::HasWorkerThreads() const
+{
+  return !m_worker_threads.empty();
+}
+
+void AsyncShaderCompiler::StopWorkerThreads()
+{
+  // Signal worker threads to stop, and wake all of them.
+  {
+    std::lock_guard<std::mutex> guard(m_pending_work_lock);
+    m_exit_flag.Set();
+    m_worker_thread_wake.notify_all();
+  }
+
+  // Wait for worker threads to exit.
+  for (std::thread& thr : m_worker_threads)
+    thr.join();
+  m_worker_threads.clear();
+}
+
+bool AsyncShaderCompiler::WorkerThreadInitMainThread(void** param)
+{
+  return true;
+}
+
+bool AsyncShaderCompiler::WorkerThreadInitWorkerThread(void* param)
+{
+  return true;
+}
+
+void AsyncShaderCompiler::WorkerThreadExit(void* param)
+{
+}
+
+void AsyncShaderCompiler::WorkerThreadEntryPoint(void* param)
+{
+  // Initialize worker thread with backend-specific method.
+  if (!WorkerThreadInitWorkerThread(param))
+  {
+    WARN_LOG(VIDEO, "Failed to initialize shader compiler worker.");
+    m_worker_thread_start_result.store(false);
+    m_init_event.Set();
+    return;
+  }
+
+  m_worker_thread_start_result.store(true);
+  m_init_event.Set();
+
+  WorkerThreadRun();
+
+  WorkerThreadExit(param);
+}
+
+void AsyncShaderCompiler::WorkerThreadRun()
+{
+  std::unique_lock<std::mutex> pending_lock(m_pending_work_lock);
+  while (!m_exit_flag.IsSet())
+  {
+    m_worker_thread_wake.wait(pending_lock);
+
+    while (!m_pending_work.empty() && !m_exit_flag.IsSet())
+    {
+      m_busy_workers++;
+      WorkItemPtr item(std::move(m_pending_work.front()));
+      m_pending_work.pop_front();
+      pending_lock.unlock();
+
+      if (item->Compile())
+      {
+        std::lock_guard<std::mutex> completed_guard(m_completed_work_lock);
+        m_completed_work.push_back(std::move(item));
+      }
+
+      pending_lock.lock();
+      m_busy_workers--;
+    }
+  }
+}
+
+}  // namespace VideoCommon
diff --git a/Source/Core/VideoCommon/AsyncShaderCompiler.h b/Source/Core/VideoCommon/AsyncShaderCompiler.h
new file mode 100644
index 0000000000..88a3d116f4
--- /dev/null
+++ b/Source/Core/VideoCommon/AsyncShaderCompiler.h
@@ -0,0 +1,83 @@
+// Copyright 2017 Dolphin Emulator Project
+// Licensed under GPLv2+
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <atomic>
+#include <condition_variable>
+#include <deque>
+#include <functional>
+#include <memory>
+#include <mutex>
+#include <thread>
+#include <vector>
+
+#include "Common/CommonTypes.h"
+#include "Common/Event.h"
+#include "Common/Flag.h"
+
+namespace VideoCommon
+{
+class AsyncShaderCompiler
+{
+public:
+  class WorkItem
+  {
+  public:
+    virtual ~WorkItem() = default;
+    virtual bool Compile() = 0;
+    virtual void Retrieve() = 0;
+  };
+
+  using WorkItemPtr = std::unique_ptr<WorkItem>;
+
+  AsyncShaderCompiler();
+  virtual ~AsyncShaderCompiler();
+
+  template <typename T, typename... Params>
+  static WorkItemPtr CreateWorkItem(Params... params)
+  {
+    return std::unique_ptr<WorkItem>(new T(params...));
+  }
+
+  void QueueWorkItem(WorkItemPtr item);
+  void RetrieveWorkItems();
+  bool HasPendingWork();
+
+  // Simpler version without progress updates.
+  void WaitUntilCompletion();
+
+  // Calls progress_callback periodically, with completed_items, and total_items.
+  void WaitUntilCompletion(const std::function<void(size_t, size_t)>& progress_callback);
+
+  // Needed because of calling virtual methods in shutdown procedure.
+  void StartWorkerThreads(u32 num_worker_threads);
+  bool HasWorkerThreads() const;
+  void StopWorkerThreads();
+
+protected:
+  virtual bool WorkerThreadInitMainThread(void** param);
+  virtual bool WorkerThreadInitWorkerThread(void* param);
+  virtual void WorkerThreadExit(void* param);
+
+private:
+  void WorkerThreadEntryPoint(void* param);
+  void WorkerThreadRun();
+
+  Common::Flag m_exit_flag;
+  Common::Event m_init_event;
+
+  std::vector<std::thread> m_worker_threads;
+  std::atomic_bool m_worker_thread_start_result{false};
+
+  std::deque<WorkItemPtr> m_pending_work;
+  std::mutex m_pending_work_lock;
+  std::condition_variable m_worker_thread_wake;
+  std::atomic_size_t m_busy_workers{0};
+
+  std::deque<WorkItemPtr> m_completed_work;
+  std::mutex m_completed_work_lock;
+};
+
+}  // namespace VideoCommon
diff --git a/Source/Core/VideoCommon/CMakeLists.txt b/Source/Core/VideoCommon/CMakeLists.txt
index 0724212bdc..782fc7af7e 100644
--- a/Source/Core/VideoCommon/CMakeLists.txt
+++ b/Source/Core/VideoCommon/CMakeLists.txt
@@ -1,6 +1,7 @@
 set(SRCS
   AbstractTexture.cpp
   AsyncRequests.cpp
+  AsyncShaderCompiler.cpp
   BoundingBox.cpp
   BPFunctions.cpp
   BPMemory.cpp
diff --git a/Source/Core/VideoCommon/VideoCommon.vcxproj b/Source/Core/VideoCommon/VideoCommon.vcxproj
index e33ccd9d7f..8dad6ea023 100644
--- a/Source/Core/VideoCommon/VideoCommon.vcxproj
+++ b/Source/Core/VideoCommon/VideoCommon.vcxproj
@@ -38,6 +38,7 @@
   <ItemGroup>
     <ClCompile Include="AbstractTexture.cpp" />
     <ClCompile Include="AsyncRequests.cpp" />
+    <ClCompile Include="AsyncShaderCompiler.cpp" />
     <ClCompile Include="AVIDump.cpp" />
     <ClCompile Include="BoundingBox.cpp" />
     <ClCompile Include="BPFunctions.cpp" />
@@ -94,6 +95,7 @@
   <ItemGroup>
     <ClInclude Include="AbstractTexture.h" />
     <ClInclude Include="AsyncRequests.h" />
+    <ClInclude Include="AsyncShaderCompiler.h" />
     <ClInclude Include="AVIDump.h" />
     <ClInclude Include="BoundingBox.h" />
     <ClInclude Include="BPFunctions.h" />
diff --git a/Source/Core/VideoCommon/VideoCommon.vcxproj.filters b/Source/Core/VideoCommon/VideoCommon.vcxproj.filters
index f4ec59eb1d..4d0348d7b0 100644
--- a/Source/Core/VideoCommon/VideoCommon.vcxproj.filters
+++ b/Source/Core/VideoCommon/VideoCommon.vcxproj.filters
@@ -176,6 +176,9 @@
     <ClCompile Include="ShaderGenCommon.cpp">
       <Filter>Shader Generators</Filter>
     </ClCompile>
+    <ClCompile Include="AsyncShaderCompiler.cpp">
+      <Filter>Util</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="CommandProcessor.h" />
@@ -332,8 +335,11 @@
     <ClInclude Include="AbstractTexture.h">
       <Filter>Base</Filter>
     </ClInclude>
+    <ClInclude Include="AsyncShaderCompiler.h">
+      <Filter>Util</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <Text Include="CMakeLists.txt" />
   </ItemGroup>
-</Project>
\ No newline at end of file
+</Project>