stop constantly reallocating memory on loader thread, wait on the gpu a lot less, fix unicode support

Author: khang06

Chikara/Main.cpp

@@ -17,7 +17,6 @@
 
 Renderer r;
 Midi* midi;
-Utils u;
 
 Vertex instanced_quad[] {
   { {0,1}, {0,1} },
@@ -41,9 +40,10 @@ void Main::run(int argc, wchar_t** argv)
   auto config_path = Config::GetConfigPath();
   Config::GetConfig().Load(config_path);
   KDMAPI::Init();
-  std::cout << "Loading " << u.GetFileName(argv[1]) << std::endl;
+  SetConsoleOutputCP(65001);
+  fmt::print("Loading {}\n", Utils::wstringToUtf8(Utils::GetFileName(argv[1])));
   std::cout << "RPC Enabled: " << Config::GetConfig().discord_rpc << std::endl;
-  if(Config::GetConfig().discord_rpc) u.InitDiscord();
+  if(Config::GetConfig().discord_rpc) Utils::InitDiscord();
   midi = new Midi(argv[1]);
   r.note_event_buffer = midi->note_event_buffer;
   r.midi_renderer_time = &midi->renderer_time;
@@ -66,7 +66,8 @@ void Main::initWindow(wchar_t** argv)
   glfwWindowHint(GLFW_CLIENT_API, GLFW_NO_API); //Set the glfw api to GLFW_NO_API because we are using Vulkan
   glfwWindowHint(GLFW_RESIZABLE, GLFW_TRUE); //Change the ability to resize the window
   //glfwWindowHint(GLFW_TRANSPARENT_FRAMEBUFFER, GLFW_TRUE); //Window Transparancy
-  r.window = glfwCreateWindow(default_width, default_height, std::string("Chikara | " + u.GetFileName(argv[1])).c_str(), nullptr, nullptr); //Now we create the window
+  auto filename = Utils::wstringToUtf8(Utils::GetFileName(argv[1]));
+  r.window = glfwCreateWindow(default_width, default_height, std::string("Chikara | " + filename).c_str(), nullptr, nullptr); //Now we create the window
   glfwSetWindowUserPointer(r.window, &r);
   glfwSetFramebufferSizeCallback(r.window, r.framebufferResizeCallback);
 }
@@ -121,10 +122,14 @@ void Main::mainLoop(wchar_t** argv)
   long long start = (std::chrono::system_clock::now().time_since_epoch() + std::chrono::seconds(1)) / std::chrono::milliseconds(1);
   long long end_time = (std::chrono::system_clock::now().time_since_epoch() + std::chrono::seconds(1) + std::chrono::seconds((long long)midi->song_len)) / std::chrono::milliseconds(1);
   midi->SpawnPlaybackThread(start_time);
+  /*
   char buffer[256];
   sprintf(buffer, "Note Count: %s", fmt::format(std::locale("en_US.UTF-8"), "{:n}", midi->note_count));
-  if(Config::GetConfig().discord_rpc) 
-    u.UpdatePresence(buffer, "Playing: ", u.GetFileName(argv[1]), (uint64_t)start, (uint64_t)end_time);
+  */
+  if (Config::GetConfig().discord_rpc) {
+    auto rpc_text = fmt::format(std::locale("en_US.UTF-8"), "Note Count: {:n}", midi->note_count);
+    Utils::UpdatePresence(rpc_text.c_str(), "Playing: ", Utils::wstringToUtf8(Utils::GetFileName(argv[1])), (uint64_t)start, (uint64_t)end_time);
+  }
   while(!glfwWindowShouldClose(r.window))
   {
     r.pre_time = Config::GetConfig().note_speed;
@@ -264,7 +269,7 @@ void Main::cleanup()
   glfwDestroyWindow(r.window);
 
   glfwTerminate(); //Now we terminate
-  u.destroyDiscord();
+  Utils::DestroyDiscord();
 }
 
 #pragma endregion

Chikara/Midi.cpp

@@ -287,8 +287,8 @@ void Midi::LoaderThread()
   double seconds = -1;
   uint64_t time = 0;
   bool all_ended = false;
+  bool* tracks_ended = new bool[track_count];
   while (true) {
-    bool* tracks_ended = new bool[track_count];
     memset(tracks_ended, 0, track_count);
     while (seconds < renderer_time.load() + 10.0f)
     {
@@ -333,10 +333,10 @@ void Midi::LoaderThread()
       if (all_ended)
         break;
     }
-    delete[] tracks_ended;
     if (all_ended)
       break;
   }
+  delete[] tracks_ended;
   misc_events.enqueue({ static_cast<float>(seconds), PLAYBACK_TERMINATE_EVENT });
   printf("\nloader thread exiting\n");
 }

Chikara/Renderer.cpp

@@ -1487,27 +1487,6 @@ void Renderer::drawFrame(float time)
   //vkWaitForFences(device, 1, &in_flight_fences[current_frame], VK_TRUE, UINT64_MAX);
   //vkResetFences(device, 1, &in_flight_fences[current_frame]);
 
-  uint32_t img_index;
-  VkResult result = vkAcquireNextImageKHR(device, swap_chain, UINT64_MAX, img_available_semaphore[current_frame], VK_NULL_HANDLE, &img_index);
-
-  if(result == VK_ERROR_OUT_OF_DATE_KHR)
-  {
-    m.recreateSwapChain();
-    return;
-  } else if(result != VK_SUCCESS && result != VK_SUBOPTIMAL_KHR) {
-    throw std::runtime_error("VKERR: Failed to acquire swap chain image!");
-  }
-
-  //Check if a previous frame is using this image (i.e. there is its fence to wait on)
-  if(imgs_in_flight[img_index] != VK_NULL_HANDLE)
-    vkWaitForFences(device, 1, &imgs_in_flight[img_index], VK_TRUE, UINT64_MAX);
-
-  //Mark the image as now being in use by this frame
-  imgs_in_flight[img_index] = imgs_in_flight[current_frame];
-
-  //Update the Uniform Buffer
-  updateUniformBuffer(img_index, time);
-
   size_t key_indices[256] = {};
   size_t cur_offset = 0;
   // yep, this iterates over the keys twice...
@@ -1590,10 +1569,35 @@ void Renderer::drawFrame(float time)
   else {
     last_notes_shown_count = notes_shown_size;
   }
+
+  uint32_t img_index;
+  VkResult result = vkAcquireNextImageKHR(device, swap_chain, UINT64_MAX, img_available_semaphore[current_frame], VK_NULL_HANDLE, &img_index);
+
+  if (result == VK_ERROR_OUT_OF_DATE_KHR)
+  {
+    m.recreateSwapChain();
+    return;
+  }
+  else if (result != VK_SUCCESS && result != VK_SUBOPTIMAL_KHR) {
+    throw std::runtime_error("VKERR: Failed to acquire swap chain image!");
+  }
+
+  //Check if a previous frame is using this image (i.e. there is its fence to wait on)
+  if (imgs_in_flight[img_index] != VK_NULL_HANDLE)
+    vkWaitForFences(device, 1, &imgs_in_flight[img_index], VK_TRUE, UINT64_MAX);
+
+  //Mark the image as now being in use by this frame
+  imgs_in_flight[img_index] = imgs_in_flight[current_frame];
+
+  //Update the Uniform Buffer
+  updateUniformBuffer(img_index, time);
+
+  vkWaitForFences(device, 1, &in_flight_fences[current_frame], VK_TRUE, UINT64_MAX);
 
+  VkCommandBuffer last_note_cmdbuf = nullptr; // used to submit a single note draw command buffer at the same time as the imgui one
   size_t instances_left = notes_shown_size;
   if (notes_shown_size == 0)
-    submitSingleCommandBuffer(cmd_buffers[img_index * MAX_NOTES_MULT]);
+    last_note_cmdbuf = cmd_buffers[img_index * MAX_NOTES_MULT];
   for (size_t i = 0; i < notes_shown_size; i += MAX_NOTES) {
     size_t instances_processed = min(instances_left, MAX_NOTES);
     vkMapMemory(device, note_instance_buffer_mem, 0, sizeof(InstanceData) * instances_processed, 0, &data);
@@ -1611,12 +1615,17 @@ void Renderer::drawFrame(float time)
       }
     }
 
-    submitSingleCommandBuffer(cmd_buffers[(i == 0 ? 0 : swap_chain_framebuffers.size() * MAX_NOTES_MULT) + note_cmd_buf + img_index * MAX_NOTES_MULT]);
     instances_left -= instances_processed;
     last_instances_processed = instances_processed;
+    auto cmd_buf = cmd_buffers[(i == 0 ? 0 : swap_chain_framebuffers.size() * MAX_NOTES_MULT) + note_cmd_buf + img_index * MAX_NOTES_MULT];
+    if (instances_left == 0)
+      last_note_cmdbuf = cmd_buf;
+    else
+      submitSingleCommandBuffer(cmd_buf);
   }
 
   // render imgui command buffers
+  // TODO: this can be done at the same time as the above block, but how can that be done efficiently?
   ImGui_ImplVulkan_NewFrame();
   ImGui_ImplGlfw_NewFrame();
   ImGui::NewFrame();
@@ -1650,13 +1659,38 @@ void Renderer::drawFrame(float time)
     }
   }
 
-  submitSingleCommandBuffer(imgui_cmd_buffers[img_index]);
+  //Submit to the command buffer
+  VkSubmitInfo submit_info = {};
+  submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
+
+  std::array<VkCommandBuffer, 2> command_buffers = { last_note_cmdbuf, imgui_cmd_buffers[img_index] };
+
+  VkSemaphore wait_semaphores[] = { img_available_semaphore[current_frame] }; //The semaphores we are waiting for
+  VkPipelineStageFlags wait_stages[] = { VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT }; //Wait with writing colors until the image is available
+  submit_info.waitSemaphoreCount = 1;
+  submit_info.pWaitSemaphores = wait_semaphores;
+  submit_info.pWaitDstStageMask = wait_stages;
+  submit_info.commandBufferCount = static_cast<uint32_t>(command_buffers.size());
+  submit_info.pCommandBuffers = command_buffers.data();
+
+  submit_info.signalSemaphoreCount = 1;
+  submit_info.pSignalSemaphores = &render_fin_semaphore[current_frame];
+
+  //vkWaitForFences(device, 1, &in_flight_fences[current_frame], VK_TRUE, UINT64_MAX);
+  vkResetFences(device, 1, &in_flight_fences[current_frame]);
+
+  auto submit_res = vkQueueSubmit(graphics_queue, 1, &submit_info, in_flight_fences[current_frame]);
+  if (submit_res != VK_SUCCESS)
+  {
+    printf("%d\n", submit_res);
+    throw std::runtime_error("VKERR: Failed to submit draw to command buffer!");
+  }
 
   //Presentation
   VkPresentInfoKHR present_info = {};
   present_info.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;
   present_info.waitSemaphoreCount = 1;
-  present_info.pWaitSemaphores = &img_available_semaphore[current_frame];
+  present_info.pWaitSemaphores = &render_fin_semaphore[current_frame];
 
   VkSwapchainKHR swap_chains[] = { swap_chain }; //An array of all our swap chains
   present_info.swapchainCount = 1;

Chikara/Utils.cpp

@@ -1,8 +1,8 @@
 #include "Utils.h"
 
-std::string Utils::GetFileName(std::filesystem::path file_path)
+std::wstring Utils::GetFileName(std::filesystem::path file_path)
 {
-  return std::filesystem::path(file_path).filename().string();
+  return std::filesystem::path(file_path).filename().wstring();
 }
 
 static void handleDiscordReady(const DiscordUser* connectedUser)
@@ -48,7 +48,12 @@ void Utils::UpdatePresence(const char* state, const char* details, std::string f
   Discord_UpdatePresence(&discordPresence);
 }
 
-void Utils::destroyDiscord()
+void Utils::DestroyDiscord()
 {
   Discord_Shutdown();
 }
+
+std::string Utils::wstringToUtf8(std::wstring str) {
+  std::wstring_convert<std::codecvt_utf8<wchar_t>> conv;
+  return conv.to_bytes(str);
+}

Chikara/Utils.h

@@ -2,14 +2,16 @@
 
 #include <iostream>
 #include <filesystem>
+#include <codecvt>
 #include <discord_rpc.h>
 
 class Utils
 {
   public:
-    std::string GetFileName(std::filesystem::path file_path);
-    void InitDiscord();
+    static std::wstring GetFileName(std::filesystem::path file_path);
+    static void InitDiscord();
     static void UpdatePresence(const char* state, const char* details, std::string file_name, uint64_t start_time, uint64_t end_time);
-    void destroyDiscord();
+    static void DestroyDiscord();
+    static std::string wstringToUtf8(std::wstring str);
 };