Linux support. (#54)

* Initial Linux attempt.

* Add clang toolchain & make tools compile.

* vcpkg as submodule.

* First implementation of IO rewrite. (#31)

* Fix directory iteration resolving symlinks.

* Refactor kernel objects to be lock-free.

* Implement guest critical sections using std::atomic.

* Make D3D12 support optional. (#33)

* Make D3D12 support optional.

* Update ShaderRecomp, fix macros.

* Replace QueryPerformanceCounter. (#35)

* Add Linux home path for GetUserPath(). (#36)

* Cross-platform Sleep. (#37)

* Add mmap implementations for virtual allocation. (#38)

* Cross-platform TLS. (#34)

* Cross-platform TLS.

* Fix front() to back(), use Mutex.

* Fix global variable namings.

---------

Co-authored-by: Skyth <19259897+blueskythlikesclouds@users.noreply.github.com>

* Unicode support. (#39)

* Replace CreateDirectoryA with Unicode version.

* Cross platform thread implementation. (#41)

* Cross-platform thread implementation.

* Put set thread name calls behind a Win32 macro.

* Cross-platform semaphore implementation. (#43)

* xam: use SDL for keyboard input

* Cross-platform atomic operations. (#44)

* Cross-platform spin lock implementation.

* Cross-platform reference counting.

* Cross-platform event implementation. (#47)

* Compiling and running on Linux. (#49)

* Current work trying to get it to compile.

* Update vcpkg.json baseline.

* vcpkg, memory mapped file.

* Bitscan forward.

* Fix localtime_s.

* FPS patches high res clock.

* Rename Window to GameWindow. Fix guest pointers.

* GetCurrentThreadID gone.

* Code cache pointers, RenderWindow type.

* Add Linux stubs.

* Refactor Config.

* Fix paths.

* Add linux-release config.

* FS fixes.

* Fix Windows compilation errors & unicode converter crash.

* Rename physical memory allocation functions to not clash with X11.

* Fix NULL character being added on RtlMultiByteToUnicodeN.

* Use std::exit.

* Add protection to memory on Linux.

* Convert majority of dependencies to submodules. (#48)

* Convert majority of dependencies to submodules.

* Don't compile header-only libraries.

* Fix a few incorrect data types.

* Fix config directory.

* Unicode fixes & sizeof asserts.

* Change the exit function to not call static destructors.

* Fix files picker.

* Add RelWithDebInfo preset for Linux.

* Implement OS Restart on Linux. (#50)

---------

Co-authored-by: Dario <dariosamo@gmail.com>

* Update PowerRecomp.

* Add Env Var detection for VCPKG_ROOT, add DLC detection.

* Use error code version on DLC directory iterator.

* Set D3D12MA::ALLOCATOR_FLAG_DONT_PREFER_SMALL_BUFFERS_COMMITTED flag.

* Linux flatpak. (#51)

* Add flatpak support.

* Add game install directory override for flatpak.

* Flatpak'ing.

* Flatpak it some more.

* We flat it, we pak it.

* Flatpak'd.

* The Marvelous Misadventures of Flatpak.

* Attempt to change logic of NFD and show error.

* Flattenpakken.

* Use game install directory instead of current path.

* Attempt to fix line endings.

* Update io.github.hedge_dev.unleashedrecomp.json

* Fix system time query implementation.

* Add Present Wait to Vulkan to improve frame pacing and reduce latency. (#53)

* Add present wait support to Vulkan.

* Default to triple buffering if presentWait is supported.

* Bracey fellas.

* Update paths.h

* SDL2 audio (again). (#52)

* Implement SDL2 audio (again).

* Call timeBeginPeriod/timeEndPeriod.

* Replace miniaudio with SDL mixer.

* Queue audio samples in a separate thread.

* Enable CMake option override policy & fix compilation error.

* Fix compilation error on Linux.

* Fix but also trim shared strings.

* Wayland support. (#55)

* Make channel index a global variable in embedded player.

* Fix SDL Audio selection for OGG on Flatpak.

* Minor installer wizard fixes.

* Fix compilation error.

* Yield in model consumer and pipeline compiler threads.

* Special case Sleep(0) to yield on Linux.

* Add App Id hint.

* Correct implementation for auto reset events. (#57)

---------

Co-authored-by: Dario <dariosamo@gmail.com>
Co-authored-by: Hyper <34012267+hyperbx@users.noreply.github.com>

UnleashedRecomp/gpu/imgui/imgui_snapshot.cpp

@@ -196,7 +196,7 @@ void ImFontAtlasSnapshot::GenerateGlyphRanges()
 {
     std::vector<std::string_view> localeStrings;
 
-    for (auto& config : Config::Definitions)
+    for (auto& config : g_configDefinitions)
         config->GetLocaleStrings(localeStrings);
 
     std::set<ImWchar> glyphs;

UnleashedRecomp/gpu/rhi/plume_d3d12.cpp

@@ -3345,7 +3345,8 @@ namespace plume {
         D3D12MA::ALLOCATOR_DESC allocatorDesc = {};
         allocatorDesc.pDevice = d3d;
         allocatorDesc.pAdapter = adapter;
-        allocatorDesc.Flags = D3D12MA::ALLOCATOR_FLAG_DEFAULT_POOLS_NOT_ZEROED | D3D12MA::ALLOCATOR_FLAG_MSAA_TEXTURES_ALWAYS_COMMITTED;
+        allocatorDesc.Flags = D3D12MA::ALLOCATOR_FLAG_DEFAULT_POOLS_NOT_ZEROED | 
+            D3D12MA::ALLOCATOR_FLAG_MSAA_TEXTURES_ALWAYS_COMMITTED | D3D12MA::ALLOCATOR_FLAG_DONT_PREFER_SMALL_BUFFERS_COMMITTED;
 
         res = D3D12MA::CreateAllocator(&allocatorDesc, &allocator);
         if (FAILED(res)) {

UnleashedRecomp/gpu/rhi/plume_render_interface_types.h

@@ -29,12 +29,18 @@
 typedef struct _NSWindow NSWindow;
 #endif
 
+#ifdef SDL_VULKAN_ENABLED
+#include <SDL_vulkan.h>
+#endif
+
 namespace plume {
 #if defined(_WIN64)
     // Native HWND handle to the target window.
     typedef HWND RenderWindow;
 #elif defined(__ANDROID__)
     typedef ANativeWindow* RenderWindow;
+#elif defined(SDL_VULKAN_ENABLED)
+    typedef SDL_Window *RenderWindow;
 #elif defined(__linux__)
     struct RenderWindow {
         Display* display;

UnleashedRecomp/gpu/rhi/plume_vulkan.cpp

@@ -1989,6 +1989,13 @@ namespace plume {
             fprintf(stderr, "vkCreateWin32SurfaceKHR failed with error code 0x%X.\n", res);
             return;
         }
+#   elif defined(SDL_VULKAN_ENABLED)
+        VulkanInterface *renderInterface = commandQueue->device->renderInterface;
+        SDL_bool sdlRes = SDL_Vulkan_CreateSurface(renderWindow, renderInterface->instance, &surface);
+        if (sdlRes == SDL_FALSE) {
+            fprintf(stderr, "SDL_Vulkan_CreateSurface failed with error %s.\n", SDL_GetError());
+            return;
+        }
 #   elif defined(__ANDROID__)
         assert(renderWindow != nullptr);
         VkAndroidSurfaceCreateInfoKHR surfaceCreateInfo = {};
@@ -2124,6 +2131,12 @@ namespace plume {
     }
 
     bool VulkanSwapChain::present(uint32_t textureIndex, RenderCommandSemaphore **waitSemaphores, uint32_t waitSemaphoreCount) {
+        constexpr uint64_t MaxFrameDelay = 1;
+        if (commandQueue->device->capabilities.presentWait && (currentPresentId > MaxFrameDelay)) {
+            constexpr uint64_t waitTimeout = 100000000;
+            vkWaitForPresentKHR(commandQueue->device->vk, vk, currentPresentId - MaxFrameDelay, waitTimeout);
+        }
+
         thread_local std::vector<VkSemaphore> waitSemaphoresVector;
         waitSemaphoresVector.clear();
         for (uint32_t i = 0; i < waitSemaphoreCount; i++) {
@@ -2138,6 +2151,15 @@ namespace plume {
         presentInfo.pImageIndices = &textureIndex;
         presentInfo.pWaitSemaphores = !waitSemaphoresVector.empty() ? waitSemaphoresVector.data() : nullptr;
         presentInfo.waitSemaphoreCount = uint32_t(waitSemaphoresVector.size());
+
+        VkPresentIdKHR presentId = {};
+        if (commandQueue->device->capabilities.presentWait) {
+            currentPresentId++;
+            presentId.sType = VK_STRUCTURE_TYPE_PRESENT_ID_KHR;
+            presentId.pPresentIds = &currentPresentId;
+            presentId.swapchainCount = 1;
+            presentInfo.pNext = &presentId;
+        }
         
         VkResult res;
         {
@@ -2297,6 +2319,8 @@ namespace plume {
         GetClientRect(renderWindow, &rect);
         dstWidth = rect.right - rect.left;
         dstHeight = rect.bottom - rect.top;
+#   elif defined(SDL_VULKAN_ENABLED)
+        SDL_GetWindowSize(renderWindow, (int *)(&dstWidth), (int *)(&dstHeight));
 #   elif defined(__ANDROID__)
         dstWidth = ANativeWindow_getWidth(renderWindow);
         dstHeight = ANativeWindow_getHeight(renderWindow);
@@ -4058,7 +4082,11 @@ namespace plume {
 
     // VulkanInterface
 
+#if SDL_VULKAN_ENABLED
+    VulkanInterface::VulkanInterface(RenderWindow sdlWindow) {
+#else
     VulkanInterface::VulkanInterface() {
+#endif
         VkResult res = volkInitialize();
         if (res != VK_SUCCESS) {
             fprintf(stderr, "volkInitialize failed with error code 0x%X.\n", res);
@@ -4085,11 +4113,31 @@ namespace plume {
         std::vector<VkExtensionProperties> availableExtensions(extensionCount);
         vkEnumerateInstanceExtensionProperties(nullptr, &extensionCount, availableExtensions.data());
 
-        std::unordered_set<std::string> missingRequiredExtensions = RequiredInstanceExtensions;
+        std::unordered_set<std::string> requiredExtensions = RequiredInstanceExtensions;
         std::unordered_set<std::string> supportedOptionalExtensions;
 #   if DLSS_ENABLED
         const std::unordered_set<std::string> dlssExtensions = DLSS::getRequiredInstanceExtensionsVulkan();
 #   endif
+
+#   if SDL_VULKAN_ENABLED
+        // Push the extensions specified by SDL as required.
+        // SDL2 has this awkward requirement for the window to pull the extensions from. 
+        // This can be removed when upgrading to SDL3.
+        if (sdlWindow != nullptr) {
+            uint32_t sdlVulkanExtensionCount = 0;
+            if (SDL_Vulkan_GetInstanceExtensions(sdlWindow, &sdlVulkanExtensionCount, nullptr)) {
+                std::vector<char *> sdlVulkanExtensions;
+                sdlVulkanExtensions.resize(sdlVulkanExtensionCount);
+                if (SDL_Vulkan_GetInstanceExtensions(sdlWindow, &sdlVulkanExtensionCount, (const char **)(sdlVulkanExtensions.data()))) {
+                    for (char *sdlVulkanExtension : sdlVulkanExtensions) {
+                        requiredExtensions.insert(sdlVulkanExtension);
+                    }
+                }
+            }
+        }
+#   endif
+
+        std::unordered_set<std::string> missingRequiredExtensions = requiredExtensions;
         for (uint32_t i = 0; i < extensionCount; i++) {
             const std::string extensionName(availableExtensions[i].extensionName);
             missingRequiredExtensions.erase(extensionName);
@@ -4114,7 +4162,7 @@ namespace plume {
         }
 
         std::vector<const char *> enabledExtensions;
-        for (const std::string &extension : RequiredInstanceExtensions) {
+        for (const std::string &extension : requiredExtensions) {
             enabledExtensions.push_back(extension.c_str());
         }
 
@@ -4177,8 +4225,15 @@ namespace plume {
 
     // Global creation function.
 
+#if SDL_VULKAN_ENABLED
+    std::unique_ptr<RenderInterface> CreateVulkanInterface(RenderWindow sdlWindow) {
+        std::unique_ptr<VulkanInterface> createdInterface = std::make_unique<VulkanInterface>(sdlWindow);
+        return createdInterface->isValid() ? std::move(createdInterface) : nullptr;
+    }
+#else
     std::unique_ptr<RenderInterface> CreateVulkanInterface() {
         std::unique_ptr<VulkanInterface> createdInterface = std::make_unique<VulkanInterface>();
         return createdInterface->isValid() ? std::move(createdInterface) : nullptr;
     }
+#endif
 };

UnleashedRecomp/gpu/rhi/plume_vulkan.h

@@ -22,9 +22,18 @@
 #define VK_USE_PLATFORM_XLIB_KHR
 #endif
 
-#include "volk.h"
+#include <volk.h>
 
-#include "vk_mem_alloc.h"
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wnullability-completeness"
+#endif
+
+#include <vk_mem_alloc.h>
+
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif
 
 namespace plume {
     struct VulkanCommandQueue;
@@ -220,6 +229,7 @@ namespace plume {
         VkPresentModeKHR requiredPresentMode = VK_PRESENT_MODE_FIFO_KHR;
         VkCompositeAlphaFlagBitsKHR pickedAlphaFlag = VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR;
         std::vector<VulkanTexture> textures;
+        uint64_t currentPresentId = 0;
         bool immediatePresentModeSupported = false;
 
         VulkanSwapChain(VulkanCommandQueue *commandQueue, RenderWindow renderWindow, uint32_t textureCount, RenderFormat format);
@@ -412,7 +422,12 @@ namespace plume {
         VkApplicationInfo appInfo = {};
         RenderInterfaceCapabilities capabilities;
 
+#   if SDL_VULKAN_ENABLED
+        VulkanInterface(RenderWindow sdlWindow);
+#   else
         VulkanInterface();
+#   endif
+
         ~VulkanInterface() override;
         std::unique_ptr<RenderDevice> createDevice() override;
         const RenderInterfaceCapabilities &getCapabilities() const override;

UnleashedRecomp/gpu/video.cpp

@@ -26,56 +26,68 @@
 #include <ui/message_window.h>
 #include <ui/options_menu.h>
 #include <ui/sdl_listener.h>
-#include <ui/window.h>
+#include <ui/game_window.h>
 #include <user/config.h>
 #include <xxHashMap.h>
 
 #if defined(ASYNC_PSO_DEBUG) || defined(PSO_CACHING)
-#include <magic_enum.hpp>
+#include <magic_enum/magic_enum.hpp>
 #endif
 
 #include "../../tools/ShaderRecomp/ShaderRecomp/shader_common.h"
+
+#ifdef SWA_D3D12
 #include "shader/copy_vs.hlsl.dxil.h"
-#include "shader/copy_vs.hlsl.spirv.h"
 #include "shader/csd_filter_ps.hlsl.dxil.h"
-#include "shader/csd_filter_ps.hlsl.spirv.h"
 #include "shader/enhanced_motion_blur_ps.hlsl.dxil.h"
-#include "shader/enhanced_motion_blur_ps.hlsl.spirv.h"
 #include "shader/gamma_correction_ps.hlsl.dxil.h"
-#include "shader/gamma_correction_ps.hlsl.spirv.h"
 #include "shader/gaussian_blur_3x3.hlsl.dxil.h"
-#include "shader/gaussian_blur_3x3.hlsl.spirv.h"
 #include "shader/gaussian_blur_5x5.hlsl.dxil.h"
-#include "shader/gaussian_blur_5x5.hlsl.spirv.h"
 #include "shader/gaussian_blur_7x7.hlsl.dxil.h"
-#include "shader/gaussian_blur_7x7.hlsl.spirv.h"
 #include "shader/gaussian_blur_9x9.hlsl.dxil.h"
-#include "shader/gaussian_blur_9x9.hlsl.spirv.h"
 #include "shader/imgui_ps.hlsl.dxil.h"
-#include "shader/imgui_ps.hlsl.spirv.h"
 #include "shader/imgui_vs.hlsl.dxil.h"
-#include "shader/imgui_vs.hlsl.spirv.h"
 #include "shader/movie_ps.hlsl.dxil.h"
-#include "shader/movie_ps.hlsl.spirv.h"
 #include "shader/movie_vs.hlsl.dxil.h"
-#include "shader/movie_vs.hlsl.spirv.h"
 #include "shader/resolve_msaa_depth_2x.hlsl.dxil.h"
-#include "shader/resolve_msaa_depth_2x.hlsl.spirv.h"
 #include "shader/resolve_msaa_depth_4x.hlsl.dxil.h"
-#include "shader/resolve_msaa_depth_4x.hlsl.spirv.h"
 #include "shader/resolve_msaa_depth_8x.hlsl.dxil.h"
+#endif
+
+#include "shader/copy_vs.hlsl.spirv.h"
+#include "shader/csd_filter_ps.hlsl.spirv.h"
+#include "shader/enhanced_motion_blur_ps.hlsl.spirv.h"
+#include "shader/gamma_correction_ps.hlsl.spirv.h"
+#include "shader/gaussian_blur_3x3.hlsl.spirv.h"
+#include "shader/gaussian_blur_5x5.hlsl.spirv.h"
+#include "shader/gaussian_blur_7x7.hlsl.spirv.h"
+#include "shader/gaussian_blur_9x9.hlsl.spirv.h"
+#include "shader/imgui_ps.hlsl.spirv.h"
+#include "shader/imgui_vs.hlsl.spirv.h"
+#include "shader/movie_ps.hlsl.spirv.h"
+#include "shader/movie_vs.hlsl.spirv.h"
+#include "shader/resolve_msaa_depth_2x.hlsl.spirv.h"
+#include "shader/resolve_msaa_depth_4x.hlsl.spirv.h"
 #include "shader/resolve_msaa_depth_8x.hlsl.spirv.h"
 
+#ifdef _WIN32
 extern "C"
 {
     __declspec(dllexport) unsigned long NvOptimusEnablement = 0x00000001;
     __declspec(dllexport) int AmdPowerXpressRequestHighPerformance = 1;
 }
+#endif
 
 namespace plume
 {
+#ifdef SWA_D3D12
     extern std::unique_ptr<RenderInterface> CreateD3D12Interface();
+#endif
+#ifdef SDL_VULKAN_ENABLED
+    extern std::unique_ptr<RenderInterface> CreateVulkanInterface(RenderWindow sdlWindow);
+#else
     extern std::unique_ptr<RenderInterface> CreateVulkanInterface();
+#endif
 }
 
 #pragma pack(push, 1)
@@ -165,7 +177,7 @@ struct DirtyStates
 static DirtyStates g_dirtyStates(true);
 
 template<typename T>
-static FORCEINLINE void SetDirtyValue(bool& dirtyState, T& dest, const T& src)
+static void SetDirtyValue(bool& dirtyState, T& dest, const T& src)
 {
     if (dest != src)
     {
@@ -174,7 +186,12 @@ static FORCEINLINE void SetDirtyValue(bool& dirtyState, T& dest, const T& src)
     }
 }
 
-static bool g_vulkan;
+#ifdef SWA_D3D12
+static bool g_vulkan = false;
+#else
+static constexpr bool g_vulkan = true;
+#endif
+
 static std::unique_ptr<RenderInterface> g_interface;
 static std::unique_ptr<RenderDevice> g_device;
 
@@ -197,7 +214,6 @@ static std::unique_ptr<RenderCommandFence> g_copyCommandFence;
 
 static std::unique_ptr<RenderSwapChain> g_swapChain;
 static bool g_swapChainValid;
-static bool g_needsResize;
 
 static constexpr RenderFormat BACKBUFFER_FORMAT = RenderFormat::B8G8R8A8_UNORM;
 
@@ -545,7 +561,7 @@ static void DestructTempResources()
     g_tempBuffers[g_frame].clear();
 }
 
-static uint32_t g_mainThreadId;
+static std::thread::id g_mainThreadId;
 
 static ankerl::unordered_dense::map<RenderTexture*, RenderTextureLayout> g_barrierMap;
 
@@ -579,13 +595,18 @@ static std::unique_ptr<uint8_t[]> g_buttonBcDiff;
 
 static void LoadEmbeddedResources()
 {
-    const size_t decompressedSize = g_vulkan ? g_spirvCacheDecompressedSize : g_dxilCacheDecompressedSize;
-    g_shaderCache = std::make_unique<uint8_t[]>(decompressedSize);
-
-    ZSTD_decompress(g_shaderCache.get(), 
-        decompressedSize, 
-        g_vulkan ? g_compressedSpirvCache : g_compressedDxilCache, 
-        g_vulkan ? g_spirvCacheCompressedSize : g_dxilCacheCompressedSize);
+    if (g_vulkan)
+    {
+        g_shaderCache = std::make_unique<uint8_t[]>(g_spirvCacheDecompressedSize);
+        ZSTD_decompress(g_shaderCache.get(), g_spirvCacheDecompressedSize, g_compressedSpirvCache, g_spirvCacheCompressedSize);
+    }
+#ifdef SWA_D3D12
+    else
+    {
+        g_shaderCache = std::make_unique<uint8_t[]>(g_dxilCacheDecompressedSize);
+        ZSTD_decompress(g_shaderCache.get(), g_dxilCacheDecompressedSize, g_compressedDxilCache, g_dxilCacheCompressedSize);
+    }
+#endif
 
     g_buttonBcDiff = decompressZstd(g_button_bc_diff, g_button_bc_diff_uncompressed_size);
 }
@@ -1023,7 +1044,7 @@ static void ProcSetRenderState(const RenderCommand& cmd)
     }
 }
 
-static const std::pair<GuestRenderState, void*> g_setRenderStateFunctions[] =
+static const std::pair<GuestRenderState, PPCFunc*> g_setRenderStateFunctions[] =
 {
     { D3DRS_ZENABLE, HostToGuestFunction<SetRenderState<D3DRS_ZENABLE>> },
     { D3DRS_ZWRITEENABLE, HostToGuestFunction<SetRenderState<D3DRS_ZWRITEENABLE>> },
@@ -1062,6 +1083,8 @@ static GuestShader* g_csdShader;
 
 static std::unique_ptr<GuestShader> g_enhancedMotionBlurShader;
 
+#ifdef SWA_D3D12
+
 #define CREATE_SHADER(NAME) \
     g_device->createShader( \
         g_vulkan ? g_##NAME##_spirv : g_##NAME##_dxil, \
@@ -1069,11 +1092,20 @@ static std::unique_ptr<GuestShader> g_enhancedMotionBlurShader;
         "main", \
         g_vulkan ? RenderShaderFormat::SPIRV : RenderShaderFormat::DXIL)
 
+#else
+
+#define CREATE_SHADER(NAME) \
+    g_device->createShader(g_##NAME##_spirv, sizeof(g_##NAME##_spirv), "main", RenderShaderFormat::SPIRV);
+
+#endif
+
+#ifdef _WIN32
 static bool DetectWine()
 {
     HMODULE dllHandle = GetModuleHandle("ntdll.dll");
     return dllHandle != nullptr && GetProcAddress(dllHandle, "wine_get_version") != nullptr;
 }
+#endif
 
 static constexpr size_t TEXTURE_DESCRIPTOR_SIZE = 65536;
 static constexpr size_t SAMPLER_DESCRIPTOR_SIZE = 1024;
@@ -1136,7 +1168,7 @@ static void CreateImGuiBackend()
     OptionsMenu::Init();
     InstallerWizard::Init();
 
-    ImGui_ImplSDL2_InitForOther(Window::s_pWindow);
+    ImGui_ImplSDL2_InitForOther(GameWindow::s_pWindow);
 
 #ifdef ENABLE_IM_FONT_ATLAS_SNAPSHOT
     g_imFontTexture = LoadTexture(
@@ -1278,7 +1310,7 @@ static void CreateImGuiBackend()
 
 static void BeginCommandList();
 
-void Video::CreateHostDevice()
+void Video::CreateHostDevice(bool sdlVideoDefault)
 {
     for (uint32_t i = 0; i < 16; i++)
         g_inputSlots[i].index = i;
@@ -1286,13 +1318,25 @@ void Video::CreateHostDevice()
     IMGUI_CHECKVERSION();
     ImGui::CreateContext();
 
-    Window::Init();
+    GameWindow::Init(sdlVideoDefault);
 
+#ifdef SWA_D3D12
     g_vulkan = DetectWine() || Config::GraphicsAPI == EGraphicsAPI::Vulkan;
+#endif
 
     LoadEmbeddedResources();
 
-    g_interface = g_vulkan ? CreateVulkanInterface() : CreateD3D12Interface();
+    if (g_vulkan)
+#ifdef SDL_VULKAN_ENABLED
+        g_interface = CreateVulkanInterface(GameWindow::s_renderWindow);
+#else
+        g_interface = CreateVulkanInterface();
+#endif
+#ifdef SWA_D3D12
+    else
+        g_interface = CreateD3D12Interface();
+#endif
+
     g_device = g_interface->createDevice();
 
     g_triangleFanSupported = g_device->getCapabilities().triangleFan;
@@ -1314,7 +1358,17 @@ void Video::CreateHostDevice()
     switch (Config::TripleBuffering)
     {
     case ETripleBuffering::Auto:
-        bufferCount = g_vulkan ? 2 : 3; // Defaulting to 3 is fine on D3D12 thanks to flip discard model.
+        if (g_vulkan)
+        {
+            // Defaulting to 3 is fine if presentWait as supported, as the maximum frame latency allowed is only 1.
+            bufferCount = g_device->getCapabilities().presentWait ? 3 : 2;
+        }
+        else
+        {
+            // Defaulting to 3 is fine on D3D12 thanks to flip discard model.
+            bufferCount = 3;
+        }
+
         break;
     case ETripleBuffering::On:
         bufferCount = 3;
@@ -1324,7 +1378,7 @@ void Video::CreateHostDevice()
         break;
     }
 
-    g_swapChain = g_queue->createSwapChain(Window::s_handle, bufferCount, BACKBUFFER_FORMAT);
+    g_swapChain = g_queue->createSwapChain(GameWindow::s_renderWindow, bufferCount, BACKBUFFER_FORMAT);
     g_swapChain->setVsyncEnabled(Config::VSync);
     g_swapChainValid = !g_swapChain->needsResize();
 
@@ -1334,7 +1388,7 @@ void Video::CreateHostDevice()
     for (auto& renderSemaphore : g_renderSemaphores)
         renderSemaphore = g_device->createCommandSemaphore();
 
-    g_mainThreadId = GetCurrentThreadId();
+    g_mainThreadId = std::this_thread::get_id();
 
     RenderPipelineLayoutBuilder pipelineLayoutBuilder;
     pipelineLayoutBuilder.begin(false, true);
@@ -1626,9 +1680,9 @@ static uint32_t CreateDevice(uint32_t a1, uint32_t a2, uint32_t a3, uint32_t a4,
     memset(device, 0, sizeof(*device));
 
     uint32_t functionOffset = 0x443344; // D3D
-    g_codeCache.Insert(functionOffset, reinterpret_cast<void*>(HostToGuestFunction<SetRenderStateUnimplemented>));
+    g_codeCache.Insert(functionOffset, HostToGuestFunction<SetRenderStateUnimplemented>);
 
-    for (size_t i = 0; i < _countof(device->setRenderStateFunctions); i++)
+    for (size_t i = 0; i < std::size(device->setRenderStateFunctions); i++)
         device->setRenderStateFunctions[i] = functionOffset;
 
     for (auto& [state, function] : g_setRenderStateFunctions)
@@ -1638,7 +1692,7 @@ static uint32_t CreateDevice(uint32_t a1, uint32_t a2, uint32_t a3, uint32_t a4,
         device->setRenderStateFunctions[state / 4] = functionOffset;
     }
 
-    for (size_t i = 0; i < _countof(device->setSamplerStateFunctions); i++)
+    for (size_t i = 0; i < std::size(device->setSamplerStateFunctions); i++)
         device->setSamplerStateFunctions[i] = *reinterpret_cast<uint32_t*>(g_memory.Translate(0x8330F3DC + i * 0xC));
 
     device->viewport.width = 1280.0f;
@@ -1763,7 +1817,7 @@ static void UnlockBuffer(GuestBuffer* buffer)
 {
     if (!buffer->lockedReadOnly)
     {
-        if (GetCurrentThreadId() == g_mainThreadId)
+        if (std::this_thread::get_id() == g_mainThreadId)
         {
             RenderCommand cmd;
             cmd.type = (sizeof(T) == 2) ? RenderCommandType::UnlockBuffer16 : RenderCommandType::UnlockBuffer32;
@@ -2765,9 +2819,6 @@ static void ProcSetScissorRect(const RenderCommand& cmd)
     SetDirtyValue<int32_t>(g_dirtyStates.scissorRect, g_scissorRect.right, args.right);
 }
 
-static Mutex g_compiledSpecConstantLibraryBlobMutex;
-static ankerl::unordered_dense::map<uint32_t, ComPtr<IDxcBlob>> g_compiledSpecConstantLibraryBlobs;
-
 static RenderShader* GetOrLinkShader(GuestShader* guestShader, uint32_t specConstants)
 {
     if (g_vulkan ||
@@ -2808,8 +2859,12 @@ static RenderShader* GetOrLinkShader(GuestShader* guestShader, uint32_t specCons
         shader = guestShader->linkedShaders[specConstants].get();
     }
 
+#ifdef SWA_D3D12
     if (shader == nullptr)
     {
+        static Mutex g_compiledSpecConstantLibraryBlobMutex;
+        static ankerl::unordered_dense::map<uint32_t, ComPtr<IDxcBlob>> g_compiledSpecConstantLibraryBlobs;
+
         thread_local ComPtr<IDxcCompiler3> s_dxcCompiler;
         thread_local ComPtr<IDxcLinker> s_dxcLinker;
         thread_local ComPtr<IDxcUtils> s_dxcUtils;
@@ -2916,6 +2971,7 @@ static RenderShader* GetOrLinkShader(GuestShader* guestShader, uint32_t specCons
             shader = linkedShader.get();
         }        
     }
+#endif
 
     return shader;
 }
@@ -4013,8 +4069,9 @@ static void ProcSetPixelShader(const RenderCommand& cmd)
 
 static std::thread g_renderThread([]
     {
+#ifdef _WIN32
         GuestThread::SetThreadName(GetCurrentThreadId(), "Render Thread");
-
+#endif
         RenderCommand commands[32];
 
         while (true)
@@ -4821,40 +4878,20 @@ struct PipelineStateQueueItem
 
 static moodycamel::BlockingConcurrentQueue<PipelineStateQueueItem> g_pipelineStateQueue;
 
-struct MinimalGuestThreadContext
-{
-    uint8_t* stack = nullptr;
-    PPCContext ppcContext{};
-
-    ~MinimalGuestThreadContext()
-    {
-        if (stack != nullptr)
-            g_userHeap.Free(stack);
-    }
-
-    void ensureValid()
-    {
-        if (stack == nullptr)
-        {
-            stack = reinterpret_cast<uint8_t*>(g_userHeap.Alloc(0x4000));
-            ppcContext.fn = (uint8_t*)g_codeCache.bucket;
-            ppcContext.r1.u64 = g_memory.MapVirtual(stack + 0x4000);
-            SetPPCContext(ppcContext);
-        }
-    }
-};
-
 static void PipelineCompilerThread()
 {
+#ifdef _WIN32
     GuestThread::SetThreadName(GetCurrentThreadId(), "Pipeline Compiler Thread");
-    MinimalGuestThreadContext ctx;
+#endif
+    std::unique_ptr<GuestThreadContext> ctx;
 
     while (true)
     {
         PipelineStateQueueItem queueItem;
         g_pipelineStateQueue.wait_dequeue(queueItem);
 
-        ctx.ensureValid();
+        if (ctx == nullptr)
+            ctx = std::make_unique<GuestThreadContext>(0);
 
         auto pipeline = CreateGraphicsPipeline(queueItem.pipelineState);
 #ifdef ASYNC_PSO_DEBUG
@@ -4867,6 +4904,8 @@ static void PipelineCompilerThread()
         cmd.addPipeline.hash = queueItem.pipelineHash;
         cmd.addPipeline.pipeline = pipeline.release();
         g_renderQueue.enqueue(cmd);
+
+        std::this_thread::yield();
     }
 }
 
@@ -5496,10 +5535,11 @@ static bool CheckMadeAll(const T& modelData)
 
 static void ModelConsumerThread()
 {
+#ifdef _WIN32
     GuestThread::SetThreadName(GetCurrentThreadId(), "Model Consumer Thread");
-
+#endif
     std::vector<boost::shared_ptr<Hedgehog::Database::CDatabaseData>> localPendingDataQueue;
-    MinimalGuestThreadContext ctx;
+    std::unique_ptr<GuestThreadContext> ctx;
 
     while (true)
     {
@@ -5508,7 +5548,8 @@ static void ModelConsumerThread()
         while ((pendingDataCount = g_pendingDataCount.load()) == 0)
             g_pendingDataCount.wait(pendingDataCount);
 
-        ctx.ensureValid();
+        if (ctx == nullptr)
+            ctx = std::make_unique<GuestThreadContext>(0);
 
         if (g_pendingPipelineStateCache)
         {
@@ -5673,6 +5714,8 @@ static void ModelConsumerThread()
 
         if (allHandled)
             localPendingDataQueue.clear();
+
+        std::this_thread::yield();
     }
 }
 

UnleashedRecomp/gpu/video.h

@@ -14,7 +14,7 @@ using namespace plume;
 
 struct Video
 {
-    static void CreateHostDevice();
+    static void CreateHostDevice(bool sdlVideoDefault);
     static void HostPresent();
     static void StartPipelinePrecompilation();
     static void WaitForGPU();
@@ -84,22 +84,26 @@ struct GuestResource
 
     void AddRef()
     {
+        std::atomic_ref atomicRef(refCount.value);
+
         uint32_t originalValue, incrementedValue;
         do
         {
             originalValue = refCount.value;
             incrementedValue = ByteSwap(ByteSwap(originalValue) + 1);
-        } while (InterlockedCompareExchange(reinterpret_cast<LONG*>(&refCount), incrementedValue, originalValue) != originalValue);
+        } while (!atomicRef.compare_exchange_weak(originalValue, incrementedValue));
     }
 
     void Release()
     {
+        std::atomic_ref atomicRef(refCount.value);
+
         uint32_t originalValue, decrementedValue;
         do
         {
             originalValue = refCount.value;
             decrementedValue = ByteSwap(ByteSwap(originalValue) - 1);
-        } while (InterlockedCompareExchange(reinterpret_cast<LONG*>(&refCount), decrementedValue, originalValue) != originalValue);
+        } while (!atomicRef.compare_exchange_weak(originalValue, decrementedValue));
 
         // Normally we are supposed to release here, so only use this
         // function when you know you won't be the one destructing it.
@@ -274,8 +278,10 @@ struct GuestShader : GuestResource
     std::unique_ptr<RenderShader> shader;
     struct ShaderCacheEntry* shaderCacheEntry = nullptr;
     ankerl::unordered_dense::map<uint32_t, std::unique_ptr<RenderShader>> linkedShaders;
+#ifdef SWA_D3D12
     std::vector<ComPtr<IDxcBlob>> shaderBlobs;
     ComPtr<IDxcBlobEncoding> libraryBlob;
+#endif
 #ifdef ASYNC_PSO_DEBUG
     const char* name = "<unknown>";
 #endif
@@ -390,7 +396,7 @@ enum GuestTextureAddress
     D3DTADDRESS_BORDER = 6
 };
 
-extern bool g_needsResize;
+inline bool g_needsResize;
 
 extern std::unique_ptr<GuestTexture> LoadTexture(const uint8_t* data, size_t dataSize, RenderComponentMapping componentMapping = RenderComponentMapping());