Migration of shibumi

.gitmodules

@@ -0,0 +1,6 @@
+[submodule "external/volk"]
+	path = external/volk
+	url = https://github.com/zeux/volk
+[submodule "external/parallel-rdp-standalone"]
+	path = external/parallel-rdp-standalone
+	url = https://github.com/Themaister/parallel-rdp-standalone

external/CMakeLists.txt

@@ -1,3 +1,61 @@
 cmake_minimum_required(VERSION 3.20)
+project(parallel-rdp)
 
-include_directories(.)
+file(GLOB_RECURSE parallel-rdp-cpp parallel-rdp-standalone/parallel-rdp/*.cpp)
+
+add_library(parallel-rdp
+  ${parallel-rdp-cpp}
+  parallel-rdp-standalone/vulkan/buffer.cpp
+  parallel-rdp-standalone/vulkan/buffer_pool.cpp
+  parallel-rdp-standalone/vulkan/command_buffer.cpp
+  parallel-rdp-standalone/vulkan/command_pool.cpp
+  parallel-rdp-standalone/vulkan/context.cpp
+  parallel-rdp-standalone/vulkan/cookie.cpp
+  parallel-rdp-standalone/vulkan/descriptor_set.cpp
+  parallel-rdp-standalone/vulkan/device.cpp
+  parallel-rdp-standalone/vulkan/event_manager.cpp
+  parallel-rdp-standalone/vulkan/fence.cpp
+  parallel-rdp-standalone/vulkan/fence_manager.cpp
+  parallel-rdp-standalone/vulkan/image.cpp
+  parallel-rdp-standalone/vulkan/memory_allocator.cpp
+  parallel-rdp-standalone/vulkan/pipeline_event.cpp
+  parallel-rdp-standalone/vulkan/query_pool.cpp
+  parallel-rdp-standalone/vulkan/render_pass.cpp
+  parallel-rdp-standalone/vulkan/sampler.cpp
+  parallel-rdp-standalone/vulkan/semaphore.cpp
+  parallel-rdp-standalone/vulkan/semaphore_manager.cpp
+  parallel-rdp-standalone/vulkan/shader.cpp
+  parallel-rdp-standalone/vulkan/texture_format.cpp
+  parallel-rdp-standalone/vulkan/wsi.cpp
+  parallel-rdp-standalone/vulkan/wsi.hpp
+  parallel-rdp-standalone/vulkan/wsi_timing.cpp
+  parallel-rdp-standalone/vulkan/wsi_timing.hpp
+  parallel-rdp-standalone/util/aligned_alloc.cpp
+  parallel-rdp-standalone/util/timer.cpp
+  parallel-rdp-standalone/util/timeline_trace_file.cpp
+  parallel-rdp-standalone/util/timeline_trace_file.hpp
+  parallel-rdp-standalone/util/thread_name.cpp
+  parallel-rdp-standalone/util/thread_name.hpp
+  parallel-rdp-standalone/util/logging.cpp
+  parallel-rdp-standalone/util/logging.hpp
+  parallel-rdp-standalone/util/thread_id.cpp
+  parallel-rdp-standalone/util/thread_id.hpp
+  # C
+  parallel-rdp-standalone/volk/volk.c
+)
+
+target_compile_definitions(parallel-rdp PUBLIC GRANITE_VULKAN_MT)
+
+target_include_directories(parallel-rdp PUBLIC
+  parallel-rdp-standalone/parallel-rdp
+  parallel-rdp-standalone/volk
+  parallel-rdp-standalone/spirv-cross
+  parallel-rdp-standalone/vulkan
+  parallel-rdp-standalone/vulkan-headers/include
+  parallel-rdp-standalone/util)
+
+if(WIN32)
+  target_compile_definitions(parallel-rdp PUBLIC VK_USE_PLATFORM_WIN32_KHR)
+else()
+  target_link_libraries(parallel-rdp dl)
+endif()

src/core/BaseCore.cpp

@@ -2,4 +2,5 @@
 
 namespace natsukashii::core {
 void BaseCore::Run() {}
+void BaseCore::PollInputs(u32 windowID) {}
 }

src/core/BaseCore.hpp

@@ -1,8 +1,13 @@
 #pragma once
+#include <common.hpp>
 
 namespace natsukashii::core {
 struct BaseCore {
-  virtual ~BaseCore() {}
+  virtual ~BaseCore() = default;
   virtual void Run();
+  virtual void PollInputs(u32);
+  [[nodiscard]] bool& ShouldQuit() { return quit; }
+private:
+  bool quit = false;
 };
 }

src/core/CMakeLists.txt

@@ -12,6 +12,5 @@ add_library(cores
   Scheduler.hpp
   common.hpp
   util.hpp)
-target_include_directories(cores PRIVATE .)
+target_include_directories(cores PUBLIC . ../../external)
-target_include_directories(cores PUBLIC ../../external)
 target_link_libraries(cores PUBLIC gb n64)

src/core/gb/Core.cpp

@@ -1,4 +1,5 @@
 #include <Core.hpp>
+#include <SDL2/SDL_events.h>
 
 namespace natsukashii::gb::core {
 Core::Core(const std::string& rom) {
@@ -10,4 +11,10 @@ void Core::Run() {
     cpu.Step(mem);
   }
 }
+
+void Core::PollInputs(u32 windowID) {
+  SDL_Event event;
+  SDL_PollEvent(&event);
+  ShouldQuit() = event.type == SDL_WINDOWEVENT && event.window.event == SDL_WINDOWEVENT_CLOSE && event.window.windowID == windowID;
+}
 }

src/core/gb/Core.hpp

@@ -10,6 +10,7 @@ struct Core : BaseCore {
   ~Core() override = default;
   explicit Core(const std::string&);
   void Run() override;
+  void PollInputs(u32) override;
 private:
   Mem mem;
   Cpu cpu;

src/core/n64/CMakeLists.txt

@@ -18,16 +18,14 @@ FetchContent_Declare(
 FetchContent_MakeAvailable(mio toml11)
 find_package(fmt REQUIRED)
 
+add_subdirectory(core)
+
 add_library(n64
   Core.hpp
   Core.cpp
-  Cpu.hpp
-  Cpu.cpp
-  Mem.cpp
-  Mem.hpp
   memory_regions.hpp
   ../BaseCore.cpp
   ../BaseCore.hpp)
-target_include_directories(n64 PRIVATE . .. ../../../external)
+target_include_directories(n64 PRIVATE . ..)
 target_include_directories(n64 PUBLIC ${mio_SOURCE_DIR}/include ${toml11_SOURCE_DIR}/include)
-target_link_libraries(n64 PRIVATE mio::mio toml11::toml11 fmt::fmt)
+target_link_libraries(n64 PRIVATE mio::mio toml11::toml11 fmt::fmt n64-core)

src/core/n64/Core.cpp

@@ -1,9 +1,18 @@
 #include <Core.hpp>
+#include <SDL2/SDL_events.h>
 
 namespace natsukashii::n64::core {
-Core::Core(const std::string& rom) {}
+Core::Core(const std::string& rom) {
+  mem.LoadROM(rom);
+}
 
 void Core::Run() {
 
 }
+
+void Core::PollInputs(u32 windowID) {
+  SDL_Event event;
+  SDL_PollEvent(&event);
+  ShouldQuit() = event.type == SDL_WINDOWEVENT && event.window.event == SDL_WINDOWEVENT_CLOSE && event.window.windowID == windowID;
+}
 }

src/core/n64/Core.hpp

@@ -1,7 +1,7 @@
 #pragma once
 #include <BaseCore.hpp>
-#include <Cpu.hpp>
+#include "n64/core/Cpu.hpp"
-#include <Mem.hpp>
+#include "n64/core/Mem.hpp"
 #include <string>
 
 namespace natsukashii::n64::core {
@@ -10,5 +10,9 @@ struct Core : BaseCore {
   ~Core() override = default;
   explicit Core(const std::string&);
   void Run() override;
+  void PollInputs(u32) override;
+private:
+  Mem mem;
+  Cpu cpu;
 };
 }

src/core/n64/Cpu.cpp

@@ -1 +0,0 @@
-#include <Cpu.hpp>

src/core/n64/core/CMakeLists.txt

@@ -0,0 +1,13 @@
+cmake_minimum_required(VERSION 3.20)
+project(n64-core)
+
+add_subdirectory(cpu)
+
+add_library(n64-core
+  Cpu.hpp
+  Cpu.cpp
+  Mem.cpp
+  Mem.hpp RDP.cpp RDP.hpp)
+
+target_include_directories(n64-core PRIVATE . .. ../../)
+target_link_libraries(n64-core PUBLIC n64-cpu)

src/core/n64/core/Cpu.cpp

@@ -0,0 +1 @@
+#include "Cpu.hpp"

src/core/n64/core/Cpu.hpp

@@ -0,0 +1,8 @@
+#pragma once
+#include "n64/core/cpu/Registers.hpp"
+
+namespace natsukashii::n64::core {
+struct Cpu {
+  Registers regs;
+};
+}

src/core/n64/core/Mem.cpp

@@ -18,13 +18,13 @@ void Mem::LoadROM(const std::string& filename) {
 
   file.seekg(std::ios::end);
   auto size = file.tellg();
+  auto size_adjusted = util::NextPow2(size);
+  romMask = size_adjusted - 1;
   file.seekg(std::ios::beg);
 
   std::vector<u8> rom;
-  rom.reserve(size);
+  rom.reserve(size_adjusted);
-  rom.insert(rom.begin(),
+  file.read(reinterpret_cast<char*>(rom.data()), size);
-             std::istream_iterator<u8>(file),
-             std::istream_iterator<u8>());
 
   file.close();
   util::SwapN64Rom(size, rom.data());

src/core/n64/core/Mem.hpp

@@ -1,16 +1,17 @@
 #pragma once
 #include <common.hpp>
-#include <memory_regions.hpp>
+#include <n64/memory_regions.hpp>
 #include <vector>
 
 namespace natsukashii::n64::core {
 struct Mem {
+  ~Mem() = default;
   Mem();
   void LoadROM(const std::string&);
 private:
   std::vector<u8> cart, rdram, sram;
-  u8 dmem[DMEM_SIZE]{}, imem[IMEM_SIZE]{}, pif_ram[PIF_RAM_SIZE]{};
+  u8 dmem[DMEM_SIZE]{}, imem[IMEM_SIZE]{}, pifRam[PIF_RAM_SIZE]{};
-  u8 pif_bootrom[PIF_BOOTROM_SIZE]{};
+  u8 pifBootrom[PIF_BOOTROM_SIZE]{};
-  size_t rom_mask;
+  size_t romMask;
 };
 }

src/core/n64/core/RDP.cpp

@@ -0,0 +1 @@
+#include <RDP.hpp>

src/core/n64/core/RDP.hpp

@@ -1,7 +1,8 @@
 #pragma once
-
 namespace natsukashii::n64::core {
-struct Cpu {
+
+struct RDP {
 
 };
-}
+
+} // natsukashii

src/core/n64/core/cpu/CMakeLists.txt

@@ -0,0 +1,11 @@
+cmake_minimum_required(VERSION 3.20)
+project(n64-cpu)
+
+add_library(n64-cpu
+  Registers.cpp
+  Registers.hpp
+  registers/Cop0.cpp
+  registers/Cop0.hpp)
+
+target_include_directories(n64-cpu PRIVATE . .. ../../ ../../../)
+target_include_directories(n64-cpu PUBLIC registers ../../../../../external)

src/core/n64/core/cpu/Registers.cpp

@@ -0,0 +1 @@
+#include <Registers.hpp>

src/core/n64/core/cpu/Registers.hpp

@@ -0,0 +1,9 @@
+#pragma once
+#include <Cop0.hpp>
+
+namespace natsukashii::n64::core {
+struct Registers {
+  s64 gpr[32];
+  Cop0 cop0;
+};
+}

src/core/n64/core/cpu/registers/Cop0.cpp

@@ -0,0 +1,17 @@
+#include <Cop0.hpp>
+
+namespace natsukashii::n64::core {
+Cop0::Cop0() {
+
+}
+
+template<class T>
+T Cop0::GetReg(u8 index) {
+
+}
+
+template<class T>
+void Cop0::SetReg(u8 index, T val) {
+
+}
+}

src/core/n64/core/cpu/registers/Cop0.hpp

@@ -0,0 +1,168 @@
+#pragma once
+#include <common.hpp>
+
+namespace natsukashii::n64::core {
+#define STATUS_MASK 0xFF77FFFF
+
+struct Cpu;
+struct Mem;
+
+union Cop0Cause {
+  u32 raw;
+  struct {
+    unsigned: 8;
+    unsigned interruptPending: 8;
+    unsigned: 16;
+  } __attribute__((__packed__));
+  struct {
+    unsigned: 2;
+    unsigned exceptionCode: 5;
+    unsigned: 1;
+    unsigned ip0: 1;
+    unsigned ip1: 1;
+    unsigned ip2: 1;
+    unsigned ip3: 1;
+    unsigned ip4: 1;
+    unsigned ip5: 1;
+    unsigned ip6: 1;
+    unsigned ip7: 1;
+    unsigned: 12;
+    unsigned copError: 2;
+    unsigned: 1;
+    unsigned branchDelay: 1;
+  } __attribute__((__packed__));
+};
+
+union Cop0Status {
+  struct {
+    unsigned ie: 1;
+    unsigned exl: 1;
+    unsigned erl: 1;
+    unsigned ksu: 2;
+    unsigned ux: 1;
+    unsigned sx: 1;
+    unsigned kx: 1;
+    unsigned im: 8;
+    unsigned ds: 9;
+    unsigned re: 1;
+    unsigned fr: 1;
+    unsigned rp: 1;
+    unsigned cu0: 1;
+    unsigned cu1: 1;
+    unsigned cu2: 1;
+    unsigned cu3: 1;
+  } __attribute__((__packed__));
+  struct {
+    unsigned: 16;
+    unsigned de: 1;
+    unsigned ce: 1;
+    unsigned ch: 1;
+    unsigned: 1;
+    unsigned sr: 1;
+    unsigned ts: 1;
+    unsigned bev: 1;
+    unsigned: 1;
+    unsigned its: 1;
+    unsigned: 7;
+  } __attribute__((__packed__));
+  u32 raw;
+} __attribute__((__packed__));
+
+union EntryLo {
+  u32 raw;
+  struct {
+    unsigned g: 1;
+    unsigned v: 1;
+    unsigned d: 1;
+    unsigned c: 3;
+    unsigned pfn: 20;
+    unsigned: 6;
+  };
+};
+
+union EntryHi {
+  u64 raw;
+  struct {
+    u64 asid: 8;
+    u64: 5;
+    u64 vpn2: 27;
+    u64 fill: 22;
+    u64 r: 2;
+  } __attribute__((__packed__));
+};
+
+union PageMask {
+  u32 raw;
+  struct {
+    unsigned: 13;
+    unsigned mask: 12;
+    unsigned: 7;
+  };
+};
+
+struct TLBEntry {
+  EntryLo entryLo0, entryLo1;
+  EntryHi entryHi;
+  PageMask pageMask;
+};
+
+enum TLBError : u8 {
+  NONE,
+  MISS,
+  INVALID,
+  MODIFICATION,
+  DISALLOWED_ADDRESS
+};
+
+enum TLBAccessType {
+  LOAD, STORE
+};
+
+union Context {
+  u64 raw;
+  struct {
+    u64: 4;
+    u64 badvpn2: 19;
+    u64 ptebase: 41;
+  };
+};
+
+union XContext {
+  u64 raw;
+  struct {
+    u64: 4;
+    u64 badvpn2: 27;
+    u64 r: 2;
+    u64 ptebase: 31;
+  } __attribute__((__packed__));
+};
+
+struct Cop0 {
+  Cop0();
+
+  template<class T>
+  T GetReg(u8 index);
+
+  template<class T>
+  void SetReg(u8 index, T val);
+
+  PageMask pageMask;
+  EntryHi entryHi;
+  EntryLo entryLo0, entryLo1;
+  u32 index, random;
+  Context context;
+  u32 wired, r7;
+  u64 badVaddr, count;
+  u32 compare;
+  Cop0Status status;
+  Cop0Cause cause;
+  u64 EPC;
+  u32 PRId, Config, LLAddr, WatchLo, WatchHi;
+  XContext xcontext;
+  u32 r21, r22, r23, r24, r25, ParityError, CacheError, TagLo, TagHi;
+  u64 ErrorEPC;
+  u32 r31;
+  TLBEntry tlb[32];
+  TLBError tlbError;
+};
+}

src/core/util.hpp

@@ -59,7 +59,7 @@ inline T ReadAccess(u8* data, u32 index) {
   static_assert(sizeof(T) != 2 && sizeof(T) != 4 && sizeof(T) != 8);
   T result = 0;
   memcpy(&result, &data[index], sizeof(T));
-  return GetSwapFunc(result);
+  return GetSwapFunc<T>(result);
 }
 
 template <typename T>
@@ -95,4 +95,16 @@ inline void SwapN64Rom(size_t size, u8* data) {
       panic("Unrecognized rom format! Make sure this is a valid Nintendo 64 ROM dump!\n");
   }
 }
+
+inline size_t NextPow2(size_t num) {
+  // Taken from "Bit Twiddling Hacks" by Sean Anderson:
+  // https://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2
+  --num;
+  num |= num >> 1;
+  num |= num >> 2;
+  num |= num >> 4;
+  num |= num >> 8;
+  num |= num >> 16;
+  return num + 1;
+}
 }

src/frontend/qt/CMakeLists.txt

@@ -11,5 +11,5 @@ find_package(Qt5 COMPONENTS Widgets REQUIRED)
 
 add_executable(natsukashii-qt Frontend.hpp Frontend.cpp main.cpp)
 
-target_include_directories(natsukashii-qt PRIVATE . ../../core ../../core/gb ../../core/n64 ../../../external)
+target_include_directories(natsukashii-qt PRIVATE . ../../core ../../core/gb ../../core/n64)
 target_link_libraries(natsukashii-qt PRIVATE cores Qt5::Widgets)

src/frontend/qt/Frontend.cpp

@@ -1,7 +1,112 @@
 #include <Frontend.hpp>
+#include <QPainter>
 
 namespace natsukashii::frontend {
-App::~App() {}
+QVulkanWindowRenderer *VulkanWindow::createRenderer() {
-
+  return new VulkanRenderer(this);
-App::App() {}
+}
+
+VulkanRenderer::VulkanRenderer(QVulkanWindow *w)
+  : m_window(w) { }
+
+void VulkanRenderer::initResources() {
+  m_devFuncs = m_window->vulkanInstance()->deviceFunctions(m_window->device());
+}
+
+void VulkanRenderer::startNextFrame() {
+  VkClearColorValue clearColor = {{ 0.0f, 0.0f, 0.0f, 1.0f }};
+  VkClearDepthStencilValue clearDS = { 1.0f, 0 };
+  VkClearValue clearValues[2];
+  memset(clearValues, 0, sizeof(clearValues));
+  clearValues[0].color = clearColor;
+  clearValues[1].depthStencil = clearDS;
+
+  VkRenderPassBeginInfo rpBeginInfo;
+  memset(&rpBeginInfo, 0, sizeof(rpBeginInfo));
+  rpBeginInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
+  rpBeginInfo.renderPass = m_window->defaultRenderPass();
+  rpBeginInfo.framebuffer = m_window->currentFramebuffer();
+  const QSize sz = m_window->swapChainImageSize();
+  rpBeginInfo.renderArea.extent.width = sz.width();
+  rpBeginInfo.renderArea.extent.height = sz.height();
+  rpBeginInfo.clearValueCount = 2;
+  rpBeginInfo.pClearValues = clearValues;
+  VkCommandBuffer cmdBuf = m_window->currentCommandBuffer();
+  m_devFuncs->vkCmdBeginRenderPass(cmdBuf, &rpBeginInfo, VK_SUBPASS_CONTENTS_INLINE);
+
+  // do thing
+
+  m_devFuncs->vkCmdEndRenderPass(cmdBuf);
+
+  m_window->frameReady();
+  m_window->requestUpdate(); // render continuously, throttled by the presentation rate
+}
+
+OpenGLWindow::OpenGLWindow(QWindow *parent) : QWindow(parent) {
+  setSurfaceType(QWindow::OpenGLSurface);
+}
+
+void OpenGLWindow::render(QPainter *painter) {
+  Q_UNUSED(painter);
+}
+
+void OpenGLWindow::initialize() {}
+
+void OpenGLWindow::render() {
+  if (!m_device)
+    m_device = new QOpenGLPaintDevice;
+
+  glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT);
+
+  m_device->setSize(size() * devicePixelRatio());
+  m_device->setDevicePixelRatio(devicePixelRatio());
+
+  QPainter painter(m_device);
+  render(&painter);
+}
+
+bool OpenGLWindow::event(QEvent *event) {
+  switch (event->type()) {
+    case QEvent::UpdateRequest:
+      renderNow();
+      return true;
+    default:
+      return QWindow::event(event);
+  }
+}
+
+void OpenGLWindow::exposeEvent(QExposeEvent *event) {
+  Q_UNUSED(event);
+
+  if (isExposed())
+    renderNow();
+}
+
+void OpenGLWindow::renderNow() {
+  if (!isExposed())
+    return;
+
+  bool needsInitialize = false;
+
+  if (!m_context) {
+    m_context = new QOpenGLContext(this);
+    m_context->setFormat(requestedFormat());
+    m_context->create();
+
+    needsInitialize = true;
+  }
+
+  m_context->makeCurrent(this);
+
+  if (needsInitialize) {
+    initializeOpenGLFunctions();
+    initialize();
+  }
+
+  render();
+
+  m_context->swapBuffers(this);
+}
+
+
 }

src/frontend/qt/Frontend.hpp

@@ -1,8 +1,63 @@
 #pragma once
+#include <QVulkanWindow>
+#include <QVulkanWindowRenderer>
+#include <QVulkanDeviceFunctions>
+#include <QOpenGLWindow>
+#include <QOpenGLFunctions>
+#include <QOpenGLPaintDevice>
+
 namespace natsukashii::frontend {
-struct App {
+class VulkanRenderer : public QVulkanWindowRenderer {
-  ~App();
+public:
-  App();
+  VulkanRenderer(QVulkanWindow *w);
-  void Run() {}
+
+  void initResources() override;
+  void initSwapChainResources() override;
+  void releaseSwapChainResources() override;
+  void releaseResources() override;
+
+  void startNextFrame() override;
+
+private:
+  QVulkanWindow *m_window;
+  QVulkanDeviceFunctions *m_devFuncs;
+};
+
+class VulkanWindow : public QVulkanWindow {
+public:
+  QVulkanWindowRenderer *createRenderer() override;
+};
+
+class OpenGLWindow : protected QOpenGLFunctions, public QWindow {
+  Q_OBJECT
+public:
+  explicit OpenGLWindow(QWindow *parent = nullptr);
+  ~OpenGLWindow();
+
+  void render(QPainter *painter);
+  void render();
+
+  void initialize();
+
+  void setAnimating(bool animating);
+
+public slots:
+  void renderNow();
+
+protected:
+  bool event(QEvent *event) override;
+  void exposeEvent(QExposeEvent *event) override;
+
+private:
+  QOpenGLContext *m_context = nullptr;
+  QOpenGLPaintDevice *m_device = nullptr;
+};
+
+class Window : public QWindow {
+public:
+  Window() {}
+private:
+  VulkanWindow vulkanWindow;
+  OpenGLWindow openGlWindow;
 };
 }

src/frontend/qt/main.cpp

@@ -1,8 +1,10 @@
 #include <Frontend.hpp>
-#include <util.hpp>
+#include <QGuiApplication>
 
 int main(int argc, char* argv[]) {
-  natsukashii::frontend::App app;
+  QGuiApplication app(argc, argv);
-  app.Run();
+  natsukashii::frontend::Window window;
-  return 0;
+  window.show();
+
+  return app.exec();
 }

src/frontend/sdl/CMakeLists.txt

@@ -5,7 +5,7 @@ set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 find_package(SDL2 REQUIRED)
 
-add_executable(natsukashii-sdl Frontend.cpp Frontend.hpp main.cpp)
+add_executable(natsukashii-sdl Frontend.cpp Frontend.hpp main.cpp ParallelRDPWrapper.cpp ParallelRDPWrapper.hpp)
 
 include(FetchContent)
 FetchContent_Declare(
@@ -15,5 +15,7 @@ FetchContent_Declare(
 )
 FetchContent_MakeAvailable(argparse)
 
-target_include_directories(natsukashii-sdl PRIVATE . ../../core ../../core/gb ../../core/n64 ../../../external)
+add_subdirectory(../../../external temp)
-target_link_libraries(natsukashii-sdl PRIVATE cores argparse::argparse SDL2)
+
+target_include_directories(natsukashii-sdl PRIVATE . ../../core ../../core/gb ../../core/n64 ../../core/n64/core ../../core/n64/core/cpu/registers)
+target_link_libraries(natsukashii-sdl PRIVATE cores argparse::argparse SDL2 parallel-rdp)

src/frontend/sdl/Frontend.cpp

@@ -1,6 +1,7 @@
 #include <Frontend.hpp>
 #include <gb/Core.hpp>
 #include <n64/Core.hpp>
+#include <volk.h>
 
 namespace natsukashii::frontend {
 using namespace natsukashii;
@@ -11,14 +12,20 @@ App::~App() {
 }
 
 App::App(const std::string& rom, const std::string& selectedCore) {
-  SDL_Init(SDL_INIT_VIDEO | SDL_INIT_AUDIO);
+  SDL_Init(SDL_INIT_VIDEO | SDL_INIT_AUDIO | SDL_VIDEO_VULKAN);
-  window = SDL_CreateWindow("natukashii", SDL_WINDOWPOS_CENTERED, SDL_WINDOWPOS_CENTERED, 800, 600, SDL_WINDOW_SHOWN);
-  renderer = SDL_CreateRenderer(window, -1, SDL_RENDERER_ACCELERATED);
-  id = SDL_GetWindowID(window);
 
   if(selectedCore == "gb") {
+    window = SDL_CreateWindow("natukashii", SDL_WINDOWPOS_CENTERED, SDL_WINDOWPOS_CENTERED, 800, 600, SDL_WINDOW_RESIZABLE);
+    renderer = SDL_CreateRenderer(window, -1, SDL_RENDERER_ACCELERATED);
+    SDL_RenderSetLogicalSize(renderer, 160, 144);
+    id = SDL_GetWindowID(window);
     core = std::make_unique<gb::core::Core>(rom);
   } else if(selectedCore == "n64") {
+    window = SDL_CreateWindow("natukashii", SDL_WINDOWPOS_CENTERED, SDL_WINDOWPOS_CENTERED, 800, 600, SDL_WINDOW_RESIZABLE | SDL_WINDOW_VULKAN);
+    id = SDL_GetWindowID(window);
+    if(volkInitialize() != VK_SUCCESS) {
+      util::panic("Failed to initialize Volk\n");
+    }
     core = std::make_unique<n64::core::Core>(rom);
   } else {
     util::panic("Unimplemented core!");
@@ -26,12 +33,9 @@ App::App(const std::string& rom, const std::string& selectedCore) {
 }
 
 void App::Run() {
-  while(!quit) {
+  while(!core->ShouldQuit()) {
     core->Run();
-
+    core->PollInputs(id);
-    SDL_Event event;
-    SDL_PollEvent(&event);
-    quit = event.type == SDL_WINDOWEVENT && event.window.event == SDL_WINDOWEVENT_CLOSE && event.window.windowID == id;
   }
 }
 }

src/frontend/sdl/Frontend.hpp

@@ -1,6 +1,7 @@
 #pragma once
 #define SDL_MAIN_HANDLED
 #include <SDL2/SDL.h>
+#include <SDL2/SDL_vulkan.h>
 #include <BaseCore.hpp>
 #include <string>
 #include <memory>
@@ -8,6 +9,7 @@
 
 namespace natsukashii::frontend {
 using namespace natsukashii::core;
+
 struct App {
   ~App();
   App(const std::string&, const std::string&);
@@ -15,8 +17,7 @@ struct App {
 private:
   SDL_Window *window = nullptr;
   SDL_Renderer *renderer = nullptr;
-  Uint32 id;
+  u32 id;
-  bool quit = false;
   std::unique_ptr<BaseCore> core;
 };
 }

src/frontend/sdl/ParallelRDPWrapper.cpp

@@ -0,0 +1,282 @@
+#include <ParallelRDPWrapper.hpp>
+#include <RDP.hpp>
+#include <memory>
+#include <rdp_device.hpp>
+#include <SDL2/SDL_video.h>
+#include <SDL2/SDL_vulkan.h>
+#include <util.hpp>
+
+using namespace Vulkan;
+using namespace natsukashii;
+using std::unique_ptr;
+using RDP::CommandProcessor;
+using RDP::CommandProcessorFlags;
+using RDP::VIRegister;
+
+static CommandProcessor* command_processor;
+static WSI* wsi;
+
+std::vector<Semaphore> acquire_semaphore;
+
+VkQueue GetGraphicsQueue() {
+  return wsi->get_context().get_graphics_queue();
+}
+
+VkInstance GetVkInstance() {
+  return wsi->get_context().get_instance();
+}
+
+VkPhysicalDevice GetVkPhysicalDevice() {
+  return wsi->get_device().get_physical_device();
+}
+
+VkDevice GetVkDevice() {
+  return wsi->get_device().get_device();
+}
+
+uint32_t GetVkGraphicsQueueFamily() {
+  return wsi->get_context().get_graphics_queue_family();
+}
+
+VkFormat GetVkFormat() {
+  return wsi->get_device().get_swapchain_view().get_format();
+}
+
+CommandBufferHandle requested_command_buffer;
+
+VkCommandBuffer GetVkCommandBuffer() {
+  requested_command_buffer = wsi->get_device().request_command_buffer();
+  return requested_command_buffer->get_command_buffer();
+}
+
+void SubmitRequestedVkCommandBuffer() {
+  wsi->get_device().submit(requested_command_buffer);
+}
+
+bool IsFramerateUnlocked() {
+  return wsi->get_present_mode() != PresentMode::SyncToVBlank;
+}
+
+void SetFramerateUnlocked(bool unlocked) {
+  if (unlocked) {
+    wsi->set_present_mode(PresentMode::UnlockedForceTearing);
+  } else {
+    wsi->set_present_mode(PresentMode::SyncToVBlank);
+  }
+}
+
+class SDLWSIPlatform : public Vulkan::WSIPlatform {
+public:
+  SDLWSIPlatform() = default;
+
+  std::vector<const char *> get_instance_extensions() override {
+    const char* extensions[64];
+    unsigned int num_extensions = 64;
+
+    if (!SDL_Vulkan_GetInstanceExtensions(window, &num_extensions, extensions)) {
+      util::panic("SDL_Vulkan_GetInstanceExtensions failed: %s", SDL_GetError());
+    }
+    auto vec = std::vector<const char*>();
+
+    for (unsigned int i = 0; i < num_extensions; i++) {
+      vec.push_back(extensions[i]);
+    }
+
+    return vec;
+  }
+
+  VkSurfaceKHR create_surface(VkInstance instance, VkPhysicalDevice gpu) override {
+    VkSurfaceKHR vk_surface;
+    if (!SDL_Vulkan_CreateSurface(window, instance, &vk_surface)) {
+      util::panic("Failed to create Vulkan window surface: %s", SDL_GetError());
+    }
+    return vk_surface;
+  }
+
+  uint32_t get_surface_width() override {
+    return N64_SCREEN_X * SCREEN_SCALE;
+  }
+
+  uint32_t get_surface_height() override {
+    return N64_SCREEN_Y * SCREEN_SCALE;
+  }
+
+  bool alive(Vulkan::WSI &wsi) override {
+    return true;
+  }
+
+  void poll_input() override {
+    n64_poll_input();
+  }
+
+  void event_frame_tick(double frame, double elapsed) override {
+    n64_render_screen();
+  }
+};
+
+Program* fullscreen_quad_program;
+
+void LoadParallelRDP(const u8* rdram) {
+  wsi = new WSI();
+  wsi->set_backbuffer_srgb(false);
+  wsi->set_platform(new SDLWSIPlatform());
+  Context::SystemHandles handles;
+  if (!wsi->init(1, handles)) {
+    util::panic("Failed to initialize WSI!");
+  }
+
+  ResourceLayout vert_layout;
+  ResourceLayout frag_layout;
+
+  vert_layout.input_mask = 1;
+  vert_layout.output_mask = 1;
+
+  frag_layout.input_mask = 1;
+  frag_layout.output_mask = 1;
+  frag_layout.spec_constant_mask = 1;
+  frag_layout.push_constant_size = 4 * sizeof(float);
+
+  frag_layout.sets[0].sampled_image_mask = 1;
+  frag_layout.sets[0].fp_mask = 1;
+  frag_layout.sets[0].array_size[0] = 1;
+
+  fullscreen_quad_program = wsi->get_device().request_program(fullscreen_quad_vert, sizeof(fullscreen_quad_vert), fullscreen_quad_frag, sizeof(fullscreen_quad_frag), &vert_layout, &frag_layout);
+
+  auto aligned_rdram = reinterpret_cast<uintptr_t>(rdram);
+  uintptr_t offset = 0;
+
+  if (wsi->get_device().get_device_features().supports_external_memory_host)
+  {
+    size_t align = wsi->get_device().get_device_features().host_memory_properties.minImportedHostPointerAlignment;
+    offset = aligned_rdram & (align - 1);
+    aligned_rdram -= offset;
+  }
+
+  CommandProcessorFlags flags = 1 << 1; // TODO configurable scaling
+
+  command_processor = new CommandProcessor(wsi->get_device(), reinterpret_cast<void *>(aligned_rdram),
+                                           offset, 8 * 1024 * 1024, 4 * 1024 * 1024, flags);
+
+  if (!command_processor->device_is_supported()) {
+    util::panic("This device probably does not support 8/16-bit storage. Make sure you're using up-to-date drivers!");
+  }
+}
+
+void draw_fullscreen_textured_quad(Util::IntrusivePtr<Image> image, Util::IntrusivePtr<CommandBuffer> cmd) {
+  cmd->set_texture(0, 0, image->get_view(), Vulkan::StockSampler::LinearClamp);
+  cmd->set_program(fullscreen_quad_program);
+  cmd->set_quad_state();
+  auto data = static_cast<float*>(cmd->allocate_vertex_data(0, 6 * sizeof(float), 2 * sizeof(float)));
+  *data++ = -1.0f;
+  *data++ = -3.0f;
+  *data++ = -1.0f;
+  *data++ = +1.0f;
+  *data++ = +3.0f;
+  *data++ = +1.0f;
+
+  int sdlWinWidth, sdlWinHeight;
+  SDL_GetWindowSize(window, &sdlWinWidth, &sdlWinHeight);
+
+  float zoom = std::min(
+    (float)sdlWinWidth / wsi->get_platform().get_surface_width(),
+    (float)sdlWinHeight / wsi->get_platform().get_surface_height());
+
+  float width = (wsi->get_platform().get_surface_width() / (float)sdlWinWidth) * zoom;
+  float height = (wsi->get_platform().get_surface_height() / (float)sdlWinHeight) * zoom;
+
+  float uniform_data[] = {
+    // Size
+    width, height,
+    // Offset
+    (1.0f - width) * 0.5f,
+    (1.0f - height) * 0.5f};
+
+  cmd->push_constants(uniform_data, 0, sizeof(uniform_data));
+
+  cmd->set_vertex_attrib(0, 0, VK_FORMAT_R32G32_SFLOAT, 0);
+  cmd->set_depth_test(false, false);
+  cmd->set_depth_compare(VK_COMPARE_OP_ALWAYS);
+  cmd->set_primitive_topology(VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST);
+  cmd->draw(3, 1);
+}
+
+void update_screen(Util::IntrusivePtr<Image> image) {
+  wsi->begin_frame();
+
+  if (!image) {
+    auto info = Vulkan::ImageCreateInfo::immutable_2d_image(N64_SCREEN_X * SCREEN_SCALE, N64_SCREEN_Y * SCREEN_SCALE, VK_FORMAT_R8G8B8A8_UNORM);
+    info.usage = VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
+                 VK_IMAGE_USAGE_TRANSFER_DST_BIT;
+    info.misc = IMAGE_MISC_MUTABLE_SRGB_BIT;
+    info.initial_layout = VK_IMAGE_LAYOUT_UNDEFINED;
+    image = wsi->get_device().create_image(info);
+
+    auto cmd = wsi->get_device().request_command_buffer();
+
+    cmd->image_barrier(*image,
+                       VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+                       VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0,
+                       VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_TRANSFER_WRITE_BIT);
+    cmd->clear_image(*image, {});
+    cmd->image_barrier(*image,
+                       VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL,
+                       VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_TRANSFER_WRITE_BIT,
+                       VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, VK_ACCESS_SHADER_READ_BIT);
+    wsi->get_device().submit(cmd);
+  }
+
+  Util::IntrusivePtr<CommandBuffer> cmd = wsi->get_device().request_command_buffer();
+
+  cmd->begin_render_pass(wsi->get_device().get_swapchain_render_pass(SwapchainRenderPass::ColorOnly));
+  draw_fullscreen_textured_quad(image, cmd);
+  ImGui_ImplVulkan_RenderDrawData(imgui_frame(), cmd->get_command_buffer());
+  cmd->end_render_pass();
+  wsi->get_device().submit(cmd);
+  wsi->end_frame();
+}
+
+void update_screen_parallel_rdp() {
+  if (unlikely(!command_processor)) {
+    logfatal("Update screen without an initialized command processor");
+  }
+
+  command_processor->set_vi_register(VIRegister::Control,      n64sys.vi.status.raw);
+  command_processor->set_vi_register(VIRegister::Origin,       n64sys.vi.vi_origin);
+  command_processor->set_vi_register(VIRegister::Width,        n64sys.vi.vi_width);
+  command_processor->set_vi_register(VIRegister::Intr,         n64sys.vi.vi_v_intr);
+  command_processor->set_vi_register(VIRegister::VCurrentLine, n64sys.vi.v_current);
+  command_processor->set_vi_register(VIRegister::Timing,       n64sys.vi.vi_burst.raw);
+  command_processor->set_vi_register(VIRegister::VSync,        n64sys.vi.vsync);
+  command_processor->set_vi_register(VIRegister::HSync,        n64sys.vi.hsync);
+  command_processor->set_vi_register(VIRegister::Leap,         n64sys.vi.leap);
+  command_processor->set_vi_register(VIRegister::HStart,       n64sys.vi.hstart.raw);
+  command_processor->set_vi_register(VIRegister::VStart,       n64sys.vi.vstart.raw);
+  command_processor->set_vi_register(VIRegister::VBurst,       n64sys.vi.vburst);
+  command_processor->set_vi_register(VIRegister::XScale,       n64sys.vi.xscale.raw);
+  command_processor->set_vi_register(VIRegister::YScale,       n64sys.vi.yscale.raw);
+
+  RDP::ScanoutOptions opts;
+  opts.persist_frame_on_invalid_input = true;
+  opts.vi.aa = true;
+  opts.vi.scale = true;
+  opts.vi.dither_filter = true;
+  opts.vi.divot_filter = true;
+  opts.vi.gamma_dither = true;
+  opts.downscale_steps = true;
+  opts.crop_overscan_pixels = true;
+  Util::IntrusivePtr<Image> image = command_processor->scanout(opts);
+  update_screen(image);
+  command_processor->begin_frame_context();
+}
+
+void update_screen_parallel_rdp_no_game() {
+  update_screen(static_cast<Util::IntrusivePtr<Image>>(nullptr));
+}
+
+void parallel_rdp_enqueue_command(int command_length, word* buffer) {
+  command_processor->enqueue_command(command_length, buffer);
+}
+
+void parallel_rdp_on_full_sync() {
+  command_processor->wait_for_timeline(command_processor->signal_timeline());
+}

src/frontend/sdl/ParallelRDPWrapper.hpp

@@ -0,0 +1,19 @@
+#pragma once
+#include <n64/Core.hpp>
+#include <wsi.hpp>
+
+VkQueue GetGraphicsQueue();
+VkInstance GetVkInstance();
+VkPhysicalDevice GetVkPhysicalDevice();
+VkDevice GetVkDevice();
+uint32_t GetVkGraphicsQueueFamily();
+VkFormat GetVkFormat();
+VkCommandBuffer GetVkCommandBuffer();
+void SubmitRequestedVkCommandBuffer();
+void LoadParallelRdp();
+void UpdateScreenParallelRdp();
+void ParallelRdpEnqueueCommand(int command_length, u32* buffer);
+void ParallelRdpOnFullSync();
+void UpdateScreenParallelRdpNoGame();
+bool IsFramerateUnlocked();
+void SetFramerateUnlocked(bool unlocked);