diff options
Diffstat (limited to 'src/video_core')
-rw-r--r-- | src/video_core/gpu.cpp | 8 | ||||
-rw-r--r-- | src/video_core/gpu.h | 3 | ||||
-rw-r--r-- | src/video_core/gpu_thread.cpp | 57 | ||||
-rw-r--r-- | src/video_core/gpu_thread.h | 13 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_cache.cpp | 2 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/renderer_vulkan.cpp | 15 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp | 5 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/vk_pipeline_cache.cpp | 2 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/vk_scheduler.cpp | 19 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/vk_scheduler.h | 7 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/vk_swapchain.cpp | 29 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/vk_swapchain.h | 11 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/vk_texture_cache.cpp | 2 | ||||
-rw-r--r-- | src/video_core/texture_cache/slot_vector.h | 4 | ||||
-rw-r--r-- | src/video_core/vulkan_common/vulkan_debug_callback.cpp | 1 | ||||
-rw-r--r-- | src/video_core/vulkan_common/vulkan_device.cpp | 42 | ||||
-rw-r--r-- | src/video_core/vulkan_common/vulkan_device.h | 6 |
17 files changed, 103 insertions, 123 deletions
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index ff024f530..2ae3639b5 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -531,14 +531,6 @@ void GPU::TriggerCpuInterrupt(const u32 syncpoint_id, const u32 value) const { interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value); } -void GPU::ShutDown() { - // Signal that threads should no longer block on syncpoint fences - shutting_down.store(true, std::memory_order_relaxed); - sync_cv.notify_all(); - - gpu_thread.ShutDown(); -} - void GPU::OnCommandListEnd() { if (is_async) { // This command only applies to asynchronous GPU mode diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index a8e98e51b..e6a02a71b 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -219,9 +219,6 @@ public: return *shader_notify; } - // Stops the GPU execution and waits for the GPU to finish working - void ShutDown(); - /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame. void WaitFence(u32 syncpoint_id, u32 value); diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp index 46f642b19..9547f277a 100644 --- a/src/video_core/gpu_thread.cpp +++ b/src/video_core/gpu_thread.cpp @@ -17,9 +17,9 @@ namespace VideoCommon::GPUThread { /// Runs the GPU thread -static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, - Core::Frontend::GraphicsContext& context, Tegra::DmaPusher& dma_pusher, - SynchState& state) { +static void RunThread(std::stop_token stop_token, Core::System& system, + VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, + Tegra::DmaPusher& dma_pusher, SynchState& state) { std::string name = "yuzu:GPU"; MicroProfileOnThreadCreate(name.c_str()); SCOPE_EXIT({ MicroProfileOnThreadExit(); }); @@ -28,20 +28,14 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, Common::SetCurrentThreadPriority(Common::ThreadPriority::High); system.RegisterHostThread(); - // Wait for first GPU command before acquiring the window context - state.queue.Wait(); - - // If emulation was stopped during disk shader loading, abort before trying to acquire context - if (!state.is_running) { - return; - } - auto current_context = context.Acquire(); VideoCore::RasterizerInterface* const rasterizer = renderer.ReadRasterizer(); - CommandDataContainer next; - while (state.is_running) { - next = state.queue.PopWait(); + while (!stop_token.stop_requested()) { + CommandDataContainer next = state.queue.PopWait(stop_token); + if (stop_token.stop_requested()) { + break; + } if (auto* submit_list = std::get_if<SubmitListCommand>(&next.data)) { dma_pusher.Push(std::move(submit_list->entries)); dma_pusher.DispatchCalls(); @@ -55,8 +49,6 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, rasterizer->FlushRegion(flush->addr, flush->size); } else if (const auto* invalidate = std::get_if<InvalidateRegionCommand>(&next.data)) { rasterizer->OnCPUWrite(invalidate->addr, invalidate->size); - } else if (std::holds_alternative<EndProcessingCommand>(next.data)) { - ASSERT(state.is_running == false); } else { UNREACHABLE(); } @@ -73,16 +65,14 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, ThreadManager::ThreadManager(Core::System& system_, bool is_async_) : system{system_}, is_async{is_async_} {} -ThreadManager::~ThreadManager() { - ShutDown(); -} +ThreadManager::~ThreadManager() = default; void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, Tegra::DmaPusher& dma_pusher) { rasterizer = renderer.ReadRasterizer(); - thread = std::thread(RunThread, std::ref(system), std::ref(renderer), std::ref(context), - std::ref(dma_pusher), std::ref(state)); + thread = std::jthread(RunThread, std::ref(system), std::ref(renderer), std::ref(context), + std::ref(dma_pusher), std::ref(state)); } void ThreadManager::SubmitList(Tegra::CommandList&& entries) { @@ -117,26 +107,6 @@ void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) { rasterizer->OnCPUWrite(addr, size); } -void ThreadManager::ShutDown() { - if (!state.is_running) { - return; - } - - { - std::lock_guard lk(state.write_lock); - state.is_running = false; - state.cv.notify_all(); - } - - if (!thread.joinable()) { - return; - } - - // Notify GPU thread that a shutdown is pending - PushCommand(EndProcessingCommand()); - thread.join(); -} - void ThreadManager::OnCommandListEnd() { PushCommand(OnCommandListEndCommand()); } @@ -152,9 +122,8 @@ u64 ThreadManager::PushCommand(CommandData&& command_data, bool block) { state.queue.Push(CommandDataContainer(std::move(command_data), fence, block)); if (block) { - state.cv.wait(lk, [this, fence] { - return fence <= state.signaled_fence.load(std::memory_order_relaxed) || - !state.is_running; + state.cv.wait(lk, thread.get_stop_token(), [this, fence] { + return fence <= state.signaled_fence.load(std::memory_order_relaxed); }); } diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h index 11a648f38..91bada925 100644 --- a/src/video_core/gpu_thread.h +++ b/src/video_core/gpu_thread.h @@ -33,9 +33,6 @@ class RendererBase; namespace VideoCommon::GPUThread { -/// Command to signal to the GPU thread that processing has ended -struct EndProcessingCommand final {}; - /// Command to signal to the GPU thread that a command list is ready for processing struct SubmitListCommand final { explicit SubmitListCommand(Tegra::CommandList&& entries_) : entries{std::move(entries_)} {} @@ -83,7 +80,7 @@ struct OnCommandListEndCommand final {}; struct GPUTickCommand final {}; using CommandData = - std::variant<EndProcessingCommand, SubmitListCommand, SwapBuffersCommand, FlushRegionCommand, + std::variant<std::monostate, SubmitListCommand, SwapBuffersCommand, FlushRegionCommand, InvalidateRegionCommand, FlushAndInvalidateRegionCommand, OnCommandListEndCommand, GPUTickCommand>; @@ -100,14 +97,12 @@ struct CommandDataContainer { /// Struct used to synchronize the GPU thread struct SynchState final { - std::atomic_bool is_running{true}; - - using CommandQueue = Common::SPSCQueue<CommandDataContainer>; + using CommandQueue = Common::SPSCQueue<CommandDataContainer, true>; std::mutex write_lock; CommandQueue queue; u64 last_fence{}; std::atomic<u64> signaled_fence{}; - std::condition_variable cv; + std::condition_variable_any cv; }; /// Class used to manage the GPU thread @@ -149,7 +144,7 @@ private: VideoCore::RasterizerInterface* rasterizer = nullptr; SynchState state; - std::thread thread; + std::jthread thread; }; } // namespace VideoCommon::GPUThread diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index b0e14182e..02682bd76 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -293,6 +293,8 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading, }}; LoadPipelines(stop_loading, shader_cache_filename, CACHE_VERSION, load_compute, load_graphics); + LOG_INFO(Render_OpenGL, "Total Pipeline Count: {}", state.total); + std::unique_lock lock{state.mutex}; callback(VideoCore::LoadCallbackStage::Build, 0, state.total); state.has_loaded = true; diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index 9ff0a28cd..74822814d 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -97,19 +97,14 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_, Core::Frontend::EmuWindow& emu_window, Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_, std::unique_ptr<Core::Frontend::GraphicsContext> context_) try - : RendererBase(emu_window, std::move(context_)), - telemetry_session(telemetry_session_), - cpu_memory(cpu_memory_), - gpu(gpu_), - library(OpenLibrary()), + : RendererBase(emu_window, std::move(context_)), telemetry_session(telemetry_session_), + cpu_memory(cpu_memory_), gpu(gpu_), library(OpenLibrary()), instance(CreateInstance(library, dld, VK_API_VERSION_1_1, render_window.GetWindowInfo().type, true, Settings::values.renderer_debug.GetValue())), debug_callback(Settings::values.renderer_debug ? CreateDebugCallback(instance) : nullptr), surface(CreateSurface(instance, render_window)), - device(CreateDevice(instance, dld, *surface)), - memory_allocator(device, false), - state_tracker(gpu), - scheduler(device, state_tracker), + device(CreateDevice(instance, dld, *surface)), memory_allocator(device, false), + state_tracker(gpu), scheduler(device, state_tracker), swapchain(*surface, device, scheduler, render_window.GetFramebufferLayout().width, render_window.GetFramebufferLayout().height, false), blit_screen(cpu_memory, render_window, device, memory_allocator, swapchain, scheduler, @@ -149,7 +144,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout(); swapchain.Create(layout.width, layout.height, is_srgb); }; - if (swapchain.IsSubOptimal() || swapchain.HasColorSpaceChanged(is_srgb)) { + if (swapchain.NeedsRecreation(is_srgb)) { recreate_swapchain(); } bool is_outdated; diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 7c0f91007..11cd41ad7 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -507,8 +507,9 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) { vertex_attributes.push_back({ .location = static_cast<u32>(index), .binding = 0, - .format = type == 1 ? VK_FORMAT_R32_SFLOAT - : type == 2 ? VK_FORMAT_R32_SINT : VK_FORMAT_R32_UINT, + .format = type == 1 ? VK_FORMAT_R32_SFLOAT + : type == 2 ? VK_FORMAT_R32_SINT + : VK_FORMAT_R32_UINT, .offset = 0, }); } diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 31bfbcb06..eb8b4e08b 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -447,6 +447,8 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading VideoCommon::LoadPipelines(stop_loading, pipeline_cache_filename, CACHE_VERSION, load_compute, load_graphics); + LOG_INFO(Render_Vulkan, "Total Pipeline Count: {}", state.total); + std::unique_lock lock{state.mutex}; callback(VideoCore::LoadCallbackStage::Build, 0, state.total); state.has_loaded = true; diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 1d438787a..0c11c814f 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -43,17 +43,10 @@ VKScheduler::VKScheduler(const Device& device_, StateTracker& state_tracker_) command_pool{std::make_unique<CommandPool>(*master_semaphore, device)} { AcquireNewChunk(); AllocateWorkerCommandBuffer(); - worker_thread = std::thread(&VKScheduler::WorkerThread, this); + worker_thread = std::jthread([this](std::stop_token token) { WorkerThread(token); }); } -VKScheduler::~VKScheduler() { - { - std::lock_guard lock{work_mutex}; - quit = true; - } - work_cv.notify_all(); - worker_thread.join(); -} +VKScheduler::~VKScheduler() = default; void VKScheduler::Flush(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { SubmitExecution(signal_semaphore, wait_semaphore); @@ -135,7 +128,7 @@ bool VKScheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) { return true; } -void VKScheduler::WorkerThread() { +void VKScheduler::WorkerThread(std::stop_token stop_token) { Common::SetCurrentThreadName("yuzu:VulkanWorker"); do { if (work_queue.empty()) { @@ -144,8 +137,8 @@ void VKScheduler::WorkerThread() { std::unique_ptr<CommandChunk> work; { std::unique_lock lock{work_mutex}; - work_cv.wait(lock, [this] { return !work_queue.empty() || quit; }); - if (quit) { + work_cv.wait(lock, stop_token, [this] { return !work_queue.empty(); }); + if (stop_token.stop_requested()) { continue; } work = std::move(work_queue.front()); @@ -158,7 +151,7 @@ void VKScheduler::WorkerThread() { } std::lock_guard reserve_lock{reserve_mutex}; chunk_reserve.push_back(std::move(work)); - } while (!quit); + } while (!stop_token.stop_requested()); } void VKScheduler::AllocateWorkerCommandBuffer() { diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 759ed5a48..85fc1712f 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -187,7 +187,7 @@ private: GraphicsPipeline* graphics_pipeline = nullptr; }; - void WorkerThread(); + void WorkerThread(std::stop_token stop_token); void AllocateWorkerCommandBuffer(); @@ -212,7 +212,6 @@ private: vk::CommandBuffer current_cmdbuf; std::unique_ptr<CommandChunk> chunk; - std::thread worker_thread; State state; @@ -224,9 +223,9 @@ private: std::vector<std::unique_ptr<CommandChunk>> chunk_reserve; std::mutex reserve_mutex; std::mutex work_mutex; - std::condition_variable work_cv; + std::condition_variable_any work_cv; std::condition_variable wait_cv; - std::atomic_bool quit{}; + std::jthread worker_thread; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_swapchain.cpp b/src/video_core/renderer_vulkan/vk_swapchain.cpp index aadf03cb0..8972a6921 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.cpp +++ b/src/video_core/renderer_vulkan/vk_swapchain.cpp @@ -9,6 +9,7 @@ #include "common/assert.h" #include "common/logging/log.h" +#include "common/settings.h" #include "core/core.h" #include "core/frontend/framebuffer_layout.h" #include "video_core/renderer_vulkan/vk_scheduler.h" @@ -36,8 +37,19 @@ VkSurfaceFormatKHR ChooseSwapSurfaceFormat(vk::Span<VkSurfaceFormatKHR> formats) VkPresentModeKHR ChooseSwapPresentMode(vk::Span<VkPresentModeKHR> modes) { // Mailbox doesn't lock the application like fifo (vsync), prefer it - const auto found = std::find(modes.begin(), modes.end(), VK_PRESENT_MODE_MAILBOX_KHR); - return found != modes.end() ? *found : VK_PRESENT_MODE_FIFO_KHR; + const auto found_mailbox = std::find(modes.begin(), modes.end(), VK_PRESENT_MODE_MAILBOX_KHR); + if (found_mailbox != modes.end()) { + return VK_PRESENT_MODE_MAILBOX_KHR; + } + if (Settings::values.disable_fps_limit.GetValue()) { + // FIFO present mode locks the framerate to the monitor's refresh rate, + // Find an alternative to surpass this limitation if FPS is unlocked. + const auto found_imm = std::find(modes.begin(), modes.end(), VK_PRESENT_MODE_IMMEDIATE_KHR); + if (found_imm != modes.end()) { + return VK_PRESENT_MODE_IMMEDIATE_KHR; + } + } + return VK_PRESENT_MODE_FIFO_KHR; } VkExtent2D ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, u32 height) { @@ -143,7 +155,7 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, const auto present_modes{physical_device.GetSurfacePresentModesKHR(surface)}; const VkSurfaceFormatKHR surface_format{ChooseSwapSurfaceFormat(formats)}; - const VkPresentModeKHR present_mode{ChooseSwapPresentMode(present_modes)}; + present_mode = ChooseSwapPresentMode(present_modes); u32 requested_image_count{capabilities.minImageCount + 1}; if (capabilities.maxImageCount > 0 && requested_image_count > capabilities.maxImageCount) { @@ -196,6 +208,7 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, extent = swapchain_ci.imageExtent; current_srgb = srgb; + current_fps_unlocked = Settings::values.disable_fps_limit.GetValue(); images = swapchain.GetImages(); image_count = static_cast<u32>(images.size()); @@ -248,4 +261,14 @@ void VKSwapchain::Destroy() { swapchain.reset(); } +bool VKSwapchain::HasFpsUnlockChanged() const { + return current_fps_unlocked != Settings::values.disable_fps_limit.GetValue(); +} + +bool VKSwapchain::NeedsPresentModeUpdate() const { + // Mailbox present mode is the ideal for all scenarios. If it is not available, + // A different present mode is needed to support unlocked FPS above the monitor's refresh rate. + return present_mode != VK_PRESENT_MODE_MAILBOX_KHR && HasFpsUnlockChanged(); +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_swapchain.h b/src/video_core/renderer_vulkan/vk_swapchain.h index 5bce41e21..61a6d959e 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.h +++ b/src/video_core/renderer_vulkan/vk_swapchain.h @@ -33,6 +33,11 @@ public: /// Presents the rendered image to the swapchain. void Present(VkSemaphore render_semaphore); + /// Returns true when the swapchain needs to be recreated. + bool NeedsRecreation(bool is_srgb) const { + return HasColorSpaceChanged(is_srgb) || IsSubOptimal() || NeedsPresentModeUpdate(); + } + /// Returns true when the color space has changed. bool HasColorSpaceChanged(bool is_srgb) const { return current_srgb != is_srgb; @@ -84,6 +89,10 @@ private: void Destroy(); + bool HasFpsUnlockChanged() const; + + bool NeedsPresentModeUpdate() const; + const VkSurfaceKHR surface; const Device& device; VKScheduler& scheduler; @@ -102,8 +111,10 @@ private: VkFormat image_view_format{}; VkExtent2D extent{}; + VkPresentModeKHR present_mode{}; bool current_srgb{}; + bool current_fps_unlocked{}; bool is_outdated{}; bool is_suboptimal{}; }; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index ff979a7ac..3b87640b5 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -127,7 +127,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { const auto format_info = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, format); VkImageCreateFlags flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT; if (info.type == ImageType::e2D && info.resources.layers >= 6 && - info.size.width == info.size.height) { + info.size.width == info.size.height && !device.HasBrokenCubeImageCompability()) { flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT; } if (info.type == ImageType::e3D) { diff --git a/src/video_core/texture_cache/slot_vector.h b/src/video_core/texture_cache/slot_vector.h index 74cd3c9d8..50df06409 100644 --- a/src/video_core/texture_cache/slot_vector.h +++ b/src/video_core/texture_cache/slot_vector.h @@ -31,8 +31,8 @@ struct SlotId { }; template <class T> -requires std::is_nothrow_move_assignable_v<T>&& - std::is_nothrow_move_constructible_v<T> class SlotVector { +requires std::is_nothrow_move_assignable_v<T> && std::is_nothrow_move_constructible_v<T> +class SlotVector { public: class Iterator { friend SlotVector<T>; diff --git a/src/video_core/vulkan_common/vulkan_debug_callback.cpp b/src/video_core/vulkan_common/vulkan_debug_callback.cpp index 0f60765bb..cf94e1d39 100644 --- a/src/video_core/vulkan_common/vulkan_debug_callback.cpp +++ b/src/video_core/vulkan_common/vulkan_debug_callback.cpp @@ -16,6 +16,7 @@ VkBool32 Callback(VkDebugUtilsMessageSeverityFlagBitsEXT severity, switch (static_cast<u32>(data->messageIdNumber)) { case 0x682a878au: // VUID-vkCmdBindVertexBuffers2EXT-pBuffers-parameter case 0x99fb7dfdu: // UNASSIGNED-RequiredParameter (vkCmdBindVertexBuffers2EXT pBuffers[0]) + case 0xe8616bf2u: // Bound VkDescriptorSet 0x0[] was destroyed. Likely push_descriptor related return VK_FALSE; default: break; diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 24fb50db9..6388ed2eb 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -243,7 +243,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR SetupFamilies(surface); SetupFeatures(); SetupProperties(); - CollectTelemetryParameters(); const auto queue_cis = GetDeviceQueueCreateInfos(); const std::vector extensions = LoadExtensions(surface != nullptr); @@ -369,20 +368,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR }; SetNext(next, demote); - if (driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE) { - const u32 version = properties.driverVersion; - // Broken in this driver - if (version > VK_MAKE_API_VERSION(0, 2, 0, 193)) { - LOG_WARNING(Render_Vulkan, "AMD proprietary driver versions newer than 21.9.1 " - "(windows) / 0.2.0.194 (amdvlk) have " - "broken VkPhysicalDeviceFloat16Int8FeaturesKHR"); - is_int8_supported = false; - is_float16_supported = false; - } - } - + VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8; if (is_int8_supported || is_float16_supported) { - VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8{ + float16_int8 = { .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR, .pNext = nullptr, .shaderFloat16 = is_float16_supported, @@ -573,6 +561,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR logical = vk::Device::Create(physical, queue_cis, extensions, first_next, dld); CollectPhysicalMemoryInfo(); + CollectTelemetryParameters(); CollectToolingInfo(); if (driver_id == VK_DRIVER_ID_NVIDIA_PROPRIETARY_KHR) { @@ -599,22 +588,27 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR ext_extended_dynamic_state = false; } } - sets_per_pool = 64; - if (driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE) { + + const bool is_amd = + driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE; + if (is_amd) { // AMD drivers need a higher amount of Sets per Pool in certain circunstances like in XC2. sets_per_pool = 96; - } - - const bool is_amd = driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || - driver_id == VK_DRIVER_ID_MESA_RADV || - driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE; - if (ext_sampler_filter_minmax && is_amd) { - // Disable ext_sampler_filter_minmax on AMD GCN4 and lower as it is broken. + // Disable VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT on AMD GCN4 and lower as it is broken. if (!is_float16_supported) { LOG_WARNING( Render_Vulkan, - "Blacklisting AMD GCN4 and lower for VK_EXT_SAMPLER_FILTER_MINMAX_EXTENSION_NAME"); + "AMD GCN4 and earlier do not properly support VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT"); + has_broken_cube_compatibility = true; + } + } + const bool is_amd_or_radv = is_amd || driver_id == VK_DRIVER_ID_MESA_RADV; + if (ext_sampler_filter_minmax && is_amd_or_radv) { + // Disable ext_sampler_filter_minmax on AMD GCN4 and lower as it is broken. + if (!is_float16_supported) { + LOG_WARNING(Render_Vulkan, + "Blacklisting AMD GCN4 and earlier for VK_EXT_sampler_filter_minmax"); ext_sampler_filter_minmax = false; } } diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index bc180a32a..d9e74f1aa 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -309,6 +309,11 @@ public: return has_renderdoc || has_nsight_graphics; } + /// Returns true when the device does not properly support cube compatibility. + bool HasBrokenCubeImageCompability() const { + return has_broken_cube_compatibility; + } + /// Returns the vendor name reported from Vulkan. std::string_view GetVendorName() const { return vendor_name; @@ -417,6 +422,7 @@ private: bool ext_conservative_rasterization{}; ///< Support for VK_EXT_conservative_rasterization. bool ext_provoking_vertex{}; ///< Support for VK_EXT_provoking_vertex. bool nv_device_diagnostics_config{}; ///< Support for VK_NV_device_diagnostics_config. + bool has_broken_cube_compatibility{}; ///< Has broken cube compatiblity bit bool has_renderdoc{}; ///< Has RenderDoc attached bool has_nsight_graphics{}; ///< Has Nsight Graphics attached |