diff options
Diffstat (limited to 'src/video_core')
19 files changed, 149 insertions, 160 deletions
diff --git a/src/video_core/buffer_cache/word_manager.h b/src/video_core/buffer_cache/word_manager.h index a336bde41..95b752055 100644 --- a/src/video_core/buffer_cache/word_manager.h +++ b/src/video_core/buffer_cache/word_manager.h @@ -473,7 +473,7 @@ private: VAddr addr = cpu_addr + word_index * BYTES_PER_WORD; IteratePages(changed_bits, [&](size_t offset, size_t size) { rasterizer->UpdatePagesCachedCount(addr + offset * BYTES_PER_PAGE, - size * BYTES_PER_PAGE, add_to_rasterizer ? 1 : -1); + size * BYTES_PER_PAGE, add_to_rasterizer); }); } diff --git a/src/video_core/host1x/ffmpeg/ffmpeg.cpp b/src/video_core/host1x/ffmpeg/ffmpeg.cpp index dcd07e6d2..96686da59 100644 --- a/src/video_core/host1x/ffmpeg/ffmpeg.cpp +++ b/src/video_core/host1x/ffmpeg/ffmpeg.cpp @@ -233,7 +233,12 @@ std::unique_ptr<Frame> DecoderContext::ReceiveFrame(bool* out_is_interlaced) { return false; } - *out_is_interlaced = frame->interlaced_frame != 0; + *out_is_interlaced = +#if defined(FF_API_INTERLACED_FRAME) || LIBAVUTIL_VERSION_MAJOR >= 59 + (frame->flags & AV_FRAME_FLAG_INTERLACED) != 0; +#else + frame->interlaced_frame != 0; +#endif return true; }; diff --git a/src/video_core/rasterizer_accelerated.cpp b/src/video_core/rasterizer_accelerated.cpp index f200a650f..3c9477f6e 100644 --- a/src/video_core/rasterizer_accelerated.cpp +++ b/src/video_core/rasterizer_accelerated.cpp @@ -3,6 +3,7 @@ #include <atomic> +#include "common/alignment.h" #include "common/assert.h" #include "common/common_types.h" #include "common/div_ceil.h" @@ -11,61 +12,65 @@ namespace VideoCore { +static constexpr u16 IdentityValue = 1; + using namespace Core::Memory; -RasterizerAccelerated::RasterizerAccelerated(Memory& cpu_memory_) - : cached_pages(std::make_unique<CachedPages>()), cpu_memory{cpu_memory_} {} +RasterizerAccelerated::RasterizerAccelerated(Memory& cpu_memory_) : map{}, cpu_memory{cpu_memory_} { + // We are tracking CPU memory, which cannot map more than 39 bits. + const VAddr start_address = 0; + const VAddr end_address = (1ULL << 39); + const IntervalType address_space_interval(start_address, end_address); + const auto value = std::make_pair(address_space_interval, IdentityValue); + + map.add(value); +} RasterizerAccelerated::~RasterizerAccelerated() = default; -void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) { - u64 uncache_begin = 0; - u64 cache_begin = 0; - u64 uncache_bytes = 0; - u64 cache_bytes = 0; - - std::atomic_thread_fence(std::memory_order_acquire); - const u64 page_end = Common::DivCeil(addr + size, YUZU_PAGESIZE); - for (u64 page = addr >> YUZU_PAGEBITS; page != page_end; ++page) { - std::atomic_uint16_t& count = cached_pages->at(page >> 2).Count(page); - - if (delta > 0) { - ASSERT_MSG(count.load(std::memory_order::relaxed) < UINT16_MAX, "Count may overflow!"); - } else if (delta < 0) { - ASSERT_MSG(count.load(std::memory_order::relaxed) > 0, "Count may underflow!"); - } else { - ASSERT_MSG(false, "Delta must be non-zero!"); - } +void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, bool cache) { + std::scoped_lock lk{map_lock}; - // Adds or subtracts 1, as count is a unsigned 8-bit value - count.fetch_add(static_cast<u16>(delta), std::memory_order_release); - - // Assume delta is either -1 or 1 - if (count.load(std::memory_order::relaxed) == 0) { - if (uncache_bytes == 0) { - uncache_begin = page; - } - uncache_bytes += YUZU_PAGESIZE; - } else if (uncache_bytes > 0) { - cpu_memory.RasterizerMarkRegionCached(uncache_begin << YUZU_PAGEBITS, uncache_bytes, - false); - uncache_bytes = 0; - } - if (count.load(std::memory_order::relaxed) == 1 && delta > 0) { - if (cache_bytes == 0) { - cache_begin = page; - } - cache_bytes += YUZU_PAGESIZE; - } else if (cache_bytes > 0) { - cpu_memory.RasterizerMarkRegionCached(cache_begin << YUZU_PAGEBITS, cache_bytes, true); - cache_bytes = 0; + // Align sizes. + addr = Common::AlignDown(addr, YUZU_PAGESIZE); + size = Common::AlignUp(size, YUZU_PAGESIZE); + + // Declare the overall interval we are going to operate on. + const VAddr start_address = addr; + const VAddr end_address = addr + size; + const IntervalType modification_range(start_address, end_address); + + // Find the boundaries of where to iterate. + const auto lower = map.lower_bound(modification_range); + const auto upper = map.upper_bound(modification_range); + + // Iterate over the contained intervals. + for (auto it = lower; it != upper; it++) { + // Intersect interval range with modification range. + const auto current_range = modification_range & it->first; + + // Calculate the address and size to operate over. + const auto current_addr = current_range.lower(); + const auto current_size = current_range.upper() - current_addr; + + // Get the current value of the range. + const auto value = it->second; + + if (cache && value == IdentityValue) { + // If we are going to cache, and the value is not yet referenced, then cache this range. + cpu_memory.RasterizerMarkRegionCached(current_addr, current_size, true); + } else if (!cache && value == IdentityValue + 1) { + // If we are going to uncache, and this is the last reference, then uncache this range. + cpu_memory.RasterizerMarkRegionCached(current_addr, current_size, false); } } - if (uncache_bytes > 0) { - cpu_memory.RasterizerMarkRegionCached(uncache_begin << YUZU_PAGEBITS, uncache_bytes, false); - } - if (cache_bytes > 0) { - cpu_memory.RasterizerMarkRegionCached(cache_begin << YUZU_PAGEBITS, cache_bytes, true); + + // Update the set. + const auto value = std::make_pair(modification_range, IdentityValue); + if (cache) { + map.add(value); + } else { + map.subtract(value); } } diff --git a/src/video_core/rasterizer_accelerated.h b/src/video_core/rasterizer_accelerated.h index e6c0ea87a..f1968f186 100644 --- a/src/video_core/rasterizer_accelerated.h +++ b/src/video_core/rasterizer_accelerated.h @@ -3,8 +3,8 @@ #pragma once -#include <array> -#include <atomic> +#include <mutex> +#include <boost/icl/interval_map.hpp> #include "common/common_types.h" #include "video_core/rasterizer_interface.h" @@ -21,28 +21,17 @@ public: explicit RasterizerAccelerated(Core::Memory::Memory& cpu_memory_); ~RasterizerAccelerated() override; - void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) override; + void UpdatePagesCachedCount(VAddr addr, u64 size, bool cache) override; private: - class CacheEntry final { - public: - CacheEntry() = default; + using PageIndex = VAddr; + using PageReferenceCount = u16; - std::atomic_uint16_t& Count(std::size_t page) { - return values[page & 3]; - } + using IntervalMap = boost::icl::interval_map<PageIndex, PageReferenceCount>; + using IntervalType = IntervalMap::interval_type; - const std::atomic_uint16_t& Count(std::size_t page) const { - return values[page & 3]; - } - - private: - std::array<std::atomic_uint16_t, 4> values{}; - }; - static_assert(sizeof(CacheEntry) == 8, "CacheEntry should be 8 bytes!"); - - using CachedPages = std::array<CacheEntry, 0x2000000>; - std::unique_ptr<CachedPages> cached_pages; + IntervalMap map; + std::mutex map_lock; Core::Memory::Memory& cpu_memory; }; diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index af1469147..fd42d26b5 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -162,7 +162,7 @@ public: } /// Increase/decrease the number of object in pages touching the specified region - virtual void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {} + virtual void UpdatePagesCachedCount(VAddr addr, u64 size, bool cache) {} /// Initialize disk cached resources for the game being emulated virtual void LoadDiskResources(u64 title_id, std::stop_token stop_loading, diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 9995b6dd4..279e5a4e0 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -714,7 +714,8 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config, MICROPROFILE_SCOPE(OpenGL_CacheManagement); std::scoped_lock lock{texture_cache.mutex}; - ImageView* const image_view{texture_cache.TryFindFramebufferImageView(framebuffer_addr)}; + ImageView* const image_view{ + texture_cache.TryFindFramebufferImageView(config, framebuffer_addr)}; if (!image_view) { return false; } @@ -725,7 +726,6 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config, screen_info.texture.width = image_view->size.width; screen_info.texture.height = image_view->size.height; screen_info.display_texture = image_view->Handle(Shader::TextureType::Color2D); - screen_info.display_srgb = VideoCore::Surface::IsPixelFormatSRGB(image_view->format); return true; } diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index 6bfed08a1..7a4f0c5c1 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -653,11 +653,7 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { }; glNamedBufferSubData(vertex_buffer.handle, 0, sizeof(vertices), std::data(vertices)); - if (screen_info.display_srgb) { - glEnable(GL_FRAMEBUFFER_SRGB); - } else { - glDisable(GL_FRAMEBUFFER_SRGB); - } + glDisable(GL_FRAMEBUFFER_SRGB); glViewportIndexedf(0, 0.0f, 0.0f, static_cast<GLfloat>(layout.width), static_cast<GLfloat>(layout.height)); @@ -710,8 +706,7 @@ void RendererOpenGL::RenderScreenshot() { GLuint renderbuffer; glGenRenderbuffers(1, &renderbuffer); glBindRenderbuffer(GL_RENDERBUFFER, renderbuffer); - glRenderbufferStorage(GL_RENDERBUFFER, screen_info.display_srgb ? GL_SRGB8 : GL_RGB8, - layout.width, layout.height); + glRenderbufferStorage(GL_RENDERBUFFER, GL_SRGB8, layout.width, layout.height); glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, renderbuffer); DrawScreen(layout); diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h index f1d5fd954..b70607635 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.h +++ b/src/video_core/renderer_opengl/renderer_opengl.h @@ -53,7 +53,6 @@ struct TextureInfo { struct ScreenInfo { GLuint display_texture{}; bool was_accelerated = false; - bool display_srgb{}; const Common::Rectangle<float> display_texcoords{0.0f, 0.0f, 1.0f, 1.0f}; TextureInfo texture; }; diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index c4c30d807..100b70918 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -94,7 +94,7 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_, device(CreateDevice(instance, dld, *surface)), memory_allocator(device), state_tracker(), scheduler(device, state_tracker), swapchain(*surface, device, scheduler, render_window.GetFramebufferLayout().width, - render_window.GetFramebufferLayout().height, false), + render_window.GetFramebufferLayout().height), present_manager(instance, render_window, device, memory_allocator, scheduler, swapchain, surface), blit_screen(cpu_memory, render_window, device, memory_allocator, swapchain, present_manager, @@ -131,11 +131,10 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset; const bool use_accelerated = rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride); - const bool is_srgb = use_accelerated && screen_info.is_srgb; RenderScreenshot(*framebuffer, use_accelerated); Frame* frame = present_manager.GetRenderFrame(); - blit_screen.DrawToSwapchain(frame, *framebuffer, use_accelerated, is_srgb); + blit_screen.DrawToSwapchain(frame, *framebuffer, use_accelerated); scheduler.Flush(*frame->render_ready); present_manager.Present(frame); @@ -205,7 +204,7 @@ void Vulkan::RendererVulkan::RenderScreenshot(const Tegra::FramebufferConfig& fr .flags = 0, .image = *staging_image, .viewType = VK_IMAGE_VIEW_TYPE_2D, - .format = screen_info.is_srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM, + .format = VK_FORMAT_B8G8R8A8_UNORM, .components{ .r = VK_COMPONENT_SWIZZLE_IDENTITY, .g = VK_COMPONENT_SWIZZLE_IDENTITY, diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index 5e461fbd0..60432f5ad 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -127,9 +127,9 @@ BlitScreen::BlitScreen(Core::Memory::Memory& cpu_memory_, Core::Frontend::EmuWin Scheduler& scheduler_, const ScreenInfo& screen_info_) : cpu_memory{cpu_memory_}, render_window{render_window_}, device{device_}, memory_allocator{memory_allocator_}, swapchain{swapchain_}, present_manager{present_manager_}, - scheduler{scheduler_}, image_count{swapchain.GetImageCount()}, screen_info{screen_info_}, - current_srgb{swapchain.IsSrgb()}, image_view_format{swapchain.GetImageViewFormat()} { + scheduler{scheduler_}, image_count{swapchain.GetImageCount()}, screen_info{screen_info_} { resource_ticks.resize(image_count); + swapchain_view_format = swapchain.GetImageViewFormat(); CreateStaticResources(); CreateDynamicResources(); @@ -480,28 +480,22 @@ void BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, } void BlitScreen::DrawToSwapchain(Frame* frame, const Tegra::FramebufferConfig& framebuffer, - bool use_accelerated, bool is_srgb) { - // Recreate dynamic resources if the the image count or colorspace changed + bool use_accelerated) { + // Recreate dynamic resources if the the image count or input format changed + const VkFormat current_framebuffer_format = + std::exchange(framebuffer_view_format, GetFormat(framebuffer)); if (const std::size_t swapchain_images = swapchain.GetImageCount(); - swapchain_images != image_count || current_srgb != is_srgb) { - current_srgb = is_srgb; -#ifdef ANDROID - // Android is already ordered the same as Switch. - image_view_format = current_srgb ? VK_FORMAT_R8G8B8A8_SRGB : VK_FORMAT_R8G8B8A8_UNORM; -#else - image_view_format = current_srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM; -#endif + swapchain_images != image_count || current_framebuffer_format != framebuffer_view_format) { image_count = swapchain_images; Recreate(); } // Recreate the presentation frame if the dimensions of the window changed const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout(); - if (layout.width != frame->width || layout.height != frame->height || - is_srgb != frame->is_srgb) { + if (layout.width != frame->width || layout.height != frame->height) { Recreate(); - present_manager.RecreateFrame(frame, layout.width, layout.height, is_srgb, - image_view_format, *renderpass); + present_manager.RecreateFrame(frame, layout.width, layout.height, swapchain_view_format, + *renderpass); } const VkExtent2D render_area{frame->width, frame->height}; @@ -629,7 +623,7 @@ void BlitScreen::CreateDescriptorPool() { } void BlitScreen::CreateRenderPass() { - renderpass = CreateRenderPassImpl(image_view_format); + renderpass = CreateRenderPassImpl(swapchain_view_format); } vk::RenderPass BlitScreen::CreateRenderPassImpl(VkFormat format) { @@ -1149,7 +1143,7 @@ void BlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) { .pNext = nullptr, .flags = 0, .imageType = VK_IMAGE_TYPE_2D, - .format = used_on_framebuffer ? VK_FORMAT_R16G16B16A16_SFLOAT : GetFormat(framebuffer), + .format = used_on_framebuffer ? VK_FORMAT_R16G16B16A16_SFLOAT : framebuffer_view_format, .extent = { .width = (up_scale * framebuffer.width) >> down_shift, @@ -1174,7 +1168,7 @@ void BlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) { .flags = 0, .image = *image, .viewType = VK_IMAGE_VIEW_TYPE_2D, - .format = used_on_framebuffer ? VK_FORMAT_R16G16B16A16_SFLOAT : GetFormat(framebuffer), + .format = used_on_framebuffer ? VK_FORMAT_R16G16B16A16_SFLOAT : framebuffer_view_format, .components = { .r = VK_COMPONENT_SWIZZLE_IDENTITY, diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.h b/src/video_core/renderer_vulkan/vk_blit_screen.h index 8365b5668..16b882b6d 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.h +++ b/src/video_core/renderer_vulkan/vk_blit_screen.h @@ -52,7 +52,6 @@ struct ScreenInfo { VkImageView image_view{}; u32 width{}; u32 height{}; - bool is_srgb{}; }; class BlitScreen { @@ -69,7 +68,7 @@ public: const Layout::FramebufferLayout layout, VkExtent2D render_area, bool use_accelerated); void DrawToSwapchain(Frame* frame, const Tegra::FramebufferConfig& framebuffer, - bool use_accelerated, bool is_srgb); + bool use_accelerated); [[nodiscard]] vk::Framebuffer CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent); @@ -161,8 +160,8 @@ private: u32 raw_width = 0; u32 raw_height = 0; Service::android::PixelFormat pixel_format{}; - bool current_srgb; - VkFormat image_view_format; + VkFormat framebuffer_view_format; + VkFormat swapchain_view_format; std::unique_ptr<FSR> fsr; std::unique_ptr<SMAA> smaa; diff --git a/src/video_core/renderer_vulkan/vk_present_manager.cpp b/src/video_core/renderer_vulkan/vk_present_manager.cpp index 2ef36583b..8e4c74b5c 100644 --- a/src/video_core/renderer_vulkan/vk_present_manager.cpp +++ b/src/video_core/renderer_vulkan/vk_present_manager.cpp @@ -172,13 +172,12 @@ void PresentManager::Present(Frame* frame) { }); } -void PresentManager::RecreateFrame(Frame* frame, u32 width, u32 height, bool is_srgb, - VkFormat image_view_format, VkRenderPass rd) { +void PresentManager::RecreateFrame(Frame* frame, u32 width, u32 height, VkFormat image_view_format, + VkRenderPass rd) { auto& dld = device.GetLogical(); frame->width = width; frame->height = height; - frame->is_srgb = is_srgb; frame->image = memory_allocator.CreateImage({ .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, @@ -289,7 +288,7 @@ void PresentManager::PresentThread(std::stop_token token) { } void PresentManager::RecreateSwapchain(Frame* frame) { - swapchain.Create(*surface, frame->width, frame->height, frame->is_srgb); + swapchain.Create(*surface, frame->width, frame->height); image_count = swapchain.GetImageCount(); } @@ -319,12 +318,12 @@ void PresentManager::CopyToSwapchain(Frame* frame) { void PresentManager::CopyToSwapchainImpl(Frame* frame) { MICROPROFILE_SCOPE(Vulkan_CopyToSwapchain); - // If the size or colorspace of the incoming frames has changed, recreate the swapchain + // If the size of the incoming frames has changed, recreate the swapchain // to account for that. - const bool srgb_changed = swapchain.NeedsRecreation(frame->is_srgb); + const bool is_suboptimal = swapchain.NeedsRecreation(); const bool size_changed = swapchain.GetWidth() != frame->width || swapchain.GetHeight() != frame->height; - if (srgb_changed || size_changed) { + if (is_suboptimal || size_changed) { RecreateSwapchain(frame); } diff --git a/src/video_core/renderer_vulkan/vk_present_manager.h b/src/video_core/renderer_vulkan/vk_present_manager.h index a3d825fe6..337171a09 100644 --- a/src/video_core/renderer_vulkan/vk_present_manager.h +++ b/src/video_core/renderer_vulkan/vk_present_manager.h @@ -25,7 +25,6 @@ class Swapchain; struct Frame { u32 width; u32 height; - bool is_srgb; vk::Image image; vk::ImageView image_view; vk::Framebuffer framebuffer; @@ -48,8 +47,8 @@ public: void Present(Frame* frame); /// Recreates the present frame to match the provided parameters - void RecreateFrame(Frame* frame, u32 width, u32 height, bool is_srgb, - VkFormat image_view_format, VkRenderPass rd); + void RecreateFrame(Frame* frame, u32 width, u32 height, VkFormat image_view_format, + VkRenderPass rd); /// Waits for the present thread to finish presenting all queued frames. void WaitPresent(); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index b6f52e017..59829c88b 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -783,7 +783,8 @@ bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config, return false; } std::scoped_lock lock{texture_cache.mutex}; - ImageView* const image_view = texture_cache.TryFindFramebufferImageView(framebuffer_addr); + ImageView* const image_view = + texture_cache.TryFindFramebufferImageView(config, framebuffer_addr); if (!image_view) { return false; } @@ -792,7 +793,6 @@ bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config, screen_info.image_view = image_view->Handle(Shader::TextureType::Color2D); screen_info.width = image_view->size.width; screen_info.height = image_view->size.height; - screen_info.is_srgb = VideoCore::Surface::IsPixelFormatSRGB(image_view->format); return true; } diff --git a/src/video_core/renderer_vulkan/vk_swapchain.cpp b/src/video_core/renderer_vulkan/vk_swapchain.cpp index 821f44f1a..86a30dcd1 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.cpp +++ b/src/video_core/renderer_vulkan/vk_swapchain.cpp @@ -105,14 +105,14 @@ VkCompositeAlphaFlagBitsKHR ChooseAlphaFlags(const VkSurfaceCapabilitiesKHR& cap } // Anonymous namespace Swapchain::Swapchain(VkSurfaceKHR surface_, const Device& device_, Scheduler& scheduler_, - u32 width_, u32 height_, bool srgb) + u32 width_, u32 height_) : surface{surface_}, device{device_}, scheduler{scheduler_} { - Create(surface_, width_, height_, srgb); + Create(surface_, width_, height_); } Swapchain::~Swapchain() = default; -void Swapchain::Create(VkSurfaceKHR surface_, u32 width_, u32 height_, bool srgb) { +void Swapchain::Create(VkSurfaceKHR surface_, u32 width_, u32 height_) { is_outdated = false; is_suboptimal = false; width = width_; @@ -127,7 +127,7 @@ void Swapchain::Create(VkSurfaceKHR surface_, u32 width_, u32 height_, bool srgb Destroy(); - CreateSwapchain(capabilities, srgb); + CreateSwapchain(capabilities); CreateSemaphores(); resource_ticks.clear(); @@ -196,7 +196,7 @@ void Swapchain::Present(VkSemaphore render_semaphore) { } } -void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bool srgb) { +void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities) { const auto physical_device{device.GetPhysical()}; const auto formats{physical_device.GetSurfaceFormatsKHR(surface)}; const auto present_modes = physical_device.GetSurfacePresentModesKHR(surface); @@ -274,15 +274,14 @@ void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bo swapchain = device.GetLogical().CreateSwapchainKHR(swapchain_ci); extent = swapchain_ci.imageExtent; - current_srgb = srgb; images = swapchain.GetImages(); image_count = static_cast<u32>(images.size()); #ifdef ANDROID // Android is already ordered the same as Switch. - image_view_format = srgb ? VK_FORMAT_R8G8B8A8_SRGB : VK_FORMAT_R8G8B8A8_UNORM; + image_view_format = VK_FORMAT_R8G8B8A8_UNORM; #else - image_view_format = srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM; + image_view_format = VK_FORMAT_B8G8R8A8_UNORM; #endif } diff --git a/src/video_core/renderer_vulkan/vk_swapchain.h b/src/video_core/renderer_vulkan/vk_swapchain.h index b8a1465a6..d264f06e4 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.h +++ b/src/video_core/renderer_vulkan/vk_swapchain.h @@ -20,11 +20,11 @@ class Scheduler; class Swapchain { public: explicit Swapchain(VkSurfaceKHR surface, const Device& device, Scheduler& scheduler, u32 width, - u32 height, bool srgb); + u32 height); ~Swapchain(); /// Creates (or recreates) the swapchain with a given size. - void Create(VkSurfaceKHR surface, u32 width, u32 height, bool srgb); + void Create(VkSurfaceKHR surface, u32 width, u32 height); /// Acquires the next image in the swapchain, waits as needed. bool AcquireNextImage(); @@ -33,13 +33,8 @@ public: 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; + bool NeedsRecreation() const { + return IsSubOptimal() || NeedsPresentModeUpdate(); } /// Returns true when the swapchain is outdated. @@ -52,11 +47,6 @@ public: return is_suboptimal; } - /// Returns true when the swapchain format is in the srgb color space - bool IsSrgb() const { - return current_srgb; - } - VkExtent2D GetSize() const { return extent; } @@ -110,7 +100,7 @@ public: } private: - void CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bool srgb); + void CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities); void CreateSemaphores(); void CreateImageViews(); @@ -144,7 +134,6 @@ private: bool has_mailbox{false}; bool has_fifo_relaxed{false}; - bool current_srgb{}; bool is_outdated{}; bool is_suboptimal{}; }; diff --git a/src/video_core/shader_cache.cpp b/src/video_core/shader_cache.cpp index e81cd031b..a109f9cbe 100644 --- a/src/video_core/shader_cache.cpp +++ b/src/video_core/shader_cache.cpp @@ -132,7 +132,7 @@ void ShaderCache::Register(std::unique_ptr<ShaderInfo> data, VAddr addr, size_t storage.push_back(std::move(data)); - rasterizer.UpdatePagesCachedCount(addr, size, 1); + rasterizer.UpdatePagesCachedCount(addr, size, true); } void ShaderCache::InvalidatePagesInRegion(VAddr addr, size_t size) { @@ -209,7 +209,7 @@ void ShaderCache::UnmarkMemory(Entry* entry) { const VAddr addr = entry->addr_start; const size_t size = entry->addr_end - addr; - rasterizer.UpdatePagesCachedCount(addr, size, -1); + rasterizer.UpdatePagesCachedCount(addr, size, false); } void ShaderCache::RemoveShadersFromStorage(std::span<ShaderInfo*> removed_shaders) { diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index dade38b18..d7941f6a4 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -712,14 +712,15 @@ bool TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst, } template <class P> -typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(VAddr cpu_addr) { +typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView( + const Tegra::FramebufferConfig& config, VAddr cpu_addr) { // TODO: Properly implement this const auto it = page_table.find(cpu_addr >> YUZU_PAGEBITS); if (it == page_table.end()) { return nullptr; } const auto& image_map_ids = it->second; - boost::container::small_vector<const ImageBase*, 4> valid_images; + boost::container::small_vector<ImageId, 4> valid_image_ids; for (const ImageMapId map_id : image_map_ids) { const ImageMapView& map = slot_map_views[map_id]; const ImageBase& image = slot_images[map.image_id]; @@ -729,18 +730,34 @@ typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(VAddr cpu_ad if (image.image_view_ids.empty()) { continue; } - valid_images.push_back(&image); + valid_image_ids.push_back(map.image_id); } - if (valid_images.size() == 1) [[likely]] { - return &slot_image_views[valid_images[0]->image_view_ids.at(0)]; + const auto view_format = [&]() { + switch (config.pixel_format) { + case Service::android::PixelFormat::Rgb565: + return PixelFormat::R5G6B5_UNORM; + case Service::android::PixelFormat::Bgra8888: + return PixelFormat::B8G8R8A8_UNORM; + default: + return PixelFormat::A8B8G8R8_UNORM; + } + }(); + + const auto GetImageViewForFramebuffer = [&](ImageId image_id) { + const ImageViewInfo info{ImageViewType::e2D, view_format}; + return &slot_image_views[FindOrEmplaceImageView(image_id, info)]; + }; + + if (valid_image_ids.size() == 1) [[likely]] { + return GetImageViewForFramebuffer(valid_image_ids.front()); } - if (valid_images.size() > 0) [[unlikely]] { - std::ranges::sort(valid_images, [](const auto* a, const auto* b) { - return a->modification_tick > b->modification_tick; + if (valid_image_ids.size() > 0) [[unlikely]] { + auto most_recent = std::ranges::max_element(valid_image_ids, [&](auto a, auto b) { + return slot_images[a].modification_tick < slot_images[b].modification_tick; }); - return &slot_image_views[valid_images[0]->image_view_ids.at(0)]; + return GetImageViewForFramebuffer(*most_recent); } return nullptr; @@ -2063,7 +2080,7 @@ void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) { ASSERT(False(image.flags & ImageFlagBits::Tracked)); image.flags |= ImageFlagBits::Tracked; if (False(image.flags & ImageFlagBits::Sparse)) { - rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1); + rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, true); return; } if (True(image.flags & ImageFlagBits::Registered)) { @@ -2074,13 +2091,13 @@ void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) { const auto& map = slot_map_views[map_view_id]; const VAddr cpu_addr = map.cpu_addr; const std::size_t size = map.size; - rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1); + rasterizer.UpdatePagesCachedCount(cpu_addr, size, true); } return; } ForEachSparseSegment(image, [this]([[maybe_unused]] GPUVAddr gpu_addr, VAddr cpu_addr, size_t size) { - rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1); + rasterizer.UpdatePagesCachedCount(cpu_addr, size, true); }); } @@ -2089,7 +2106,7 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) { ASSERT(True(image.flags & ImageFlagBits::Tracked)); image.flags &= ~ImageFlagBits::Tracked; if (False(image.flags & ImageFlagBits::Sparse)) { - rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, -1); + rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, false); return; } ASSERT(True(image.flags & ImageFlagBits::Registered)); @@ -2100,7 +2117,7 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) { const auto& map = slot_map_views[map_view_id]; const VAddr cpu_addr = map.cpu_addr; const std::size_t size = map.size; - rasterizer.UpdatePagesCachedCount(cpu_addr, size, -1); + rasterizer.UpdatePagesCachedCount(cpu_addr, size, false); } } diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index a40825c9f..cbe56e166 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -209,7 +209,8 @@ public: const Tegra::Engines::Fermi2D::Config& copy); /// Try to find a cached image view in the given CPU address - [[nodiscard]] ImageView* TryFindFramebufferImageView(VAddr cpu_addr); + [[nodiscard]] ImageView* TryFindFramebufferImageView(const Tegra::FramebufferConfig& config, + VAddr cpu_addr); /// Return true when there are uncommitted images to be downloaded [[nodiscard]] bool HasUncommittedFlushes() const noexcept; |