Merge pull request #9973 from GPUCode/async-present

Implement asynchronous presentation

src/common/settings.cpp

@@ -205,6 +205,7 @@ void RestoreGlobalState(bool is_powered_on) {
     // Renderer
     values.fsr_sharpening_slider.SetGlobal(true);
     values.renderer_backend.SetGlobal(true);
+    values.async_presentation.SetGlobal(true);
     values.renderer_force_max_clock.SetGlobal(true);
     values.vulkan_device.SetGlobal(true);
     values.fullscreen_mode.SetGlobal(true);

src/common/settings.h

@@ -422,6 +422,7 @@ struct Values {
     // Renderer
     SwitchableSetting<RendererBackend, true> renderer_backend{
         RendererBackend::Vulkan, RendererBackend::OpenGL, RendererBackend::Null, "backend"};
+    SwitchableSetting<bool> async_presentation{false, "async_presentation"};
     SwitchableSetting<bool> renderer_force_max_clock{false, "force_max_clock"};
     Setting<bool> renderer_debug{false, "debug"};
     Setting<bool> renderer_shader_feedback{false, "shader_feedback"};

src/video_core/CMakeLists.txt

@@ -179,6 +179,8 @@ add_library(video_core STATIC
     renderer_vulkan/vk_master_semaphore.h
     renderer_vulkan/vk_pipeline_cache.cpp
     renderer_vulkan/vk_pipeline_cache.h
+    renderer_vulkan/vk_present_manager.cpp
+    renderer_vulkan/vk_present_manager.h
     renderer_vulkan/vk_query_cache.cpp
     renderer_vulkan/vk_query_cache.h
     renderer_vulkan/vk_rasterizer.cpp

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -93,8 +93,9 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_,
       state_tracker(), 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,
-                  screen_info),
+      present_manager(render_window, device, memory_allocator, scheduler, swapchain),
+      blit_screen(cpu_memory, render_window, device, memory_allocator, swapchain, present_manager,
+                  scheduler, screen_info),
       rasterizer(render_window, gpu, cpu_memory, screen_info, device, memory_allocator,
                  state_tracker, scheduler) {
     if (Settings::values.renderer_force_max_clock.GetValue() && device.ShouldBoostClocks()) {
@@ -121,46 +122,19 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
         return;
     }
     // Update screen info if the framebuffer size has changed.
-    if (screen_info.width != framebuffer->width || screen_info.height != framebuffer->height) {
-        screen_info.width = framebuffer->width;
-        screen_info.height = framebuffer->height;
-    }
+    screen_info.width = framebuffer->width;
+    screen_info.height = framebuffer->height;
+
     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);
 
-    bool has_been_recreated = false;
-    const auto recreate_swapchain = [&](u32 width, u32 height) {
-        if (!has_been_recreated) {
-            has_been_recreated = true;
-            scheduler.Finish();
-        }
-        swapchain.Create(width, height, is_srgb);
-    };
-
-    const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout();
-    if (swapchain.NeedsRecreation(is_srgb) || swapchain.GetWidth() != layout.width ||
-        swapchain.GetHeight() != layout.height) {
-        recreate_swapchain(layout.width, layout.height);
-    }
-    bool is_outdated;
-    do {
-        swapchain.AcquireNextImage();
-        is_outdated = swapchain.IsOutDated();
-        if (is_outdated) {
-            recreate_swapchain(layout.width, layout.height);
-        }
-    } while (is_outdated);
-    if (has_been_recreated) {
-        blit_screen.Recreate();
-    }
-    const VkSemaphore render_semaphore = blit_screen.DrawToSwapchain(*framebuffer, use_accelerated);
-    const VkSemaphore present_semaphore = swapchain.CurrentPresentSemaphore();
-    scheduler.Flush(render_semaphore, present_semaphore);
-    scheduler.WaitWorker();
-    swapchain.Present(render_semaphore);
+    Frame* frame = present_manager.GetRenderFrame();
+    blit_screen.DrawToSwapchain(frame, *framebuffer, use_accelerated, is_srgb);
+    scheduler.Flush(*frame->render_ready);
+    scheduler.Record([this, frame](vk::CommandBuffer) { present_manager.PushFrame(frame); });
 
     gpu.RendererFrameEndNotify();
     rasterizer.TickFrame();
@@ -246,8 +220,7 @@ void Vulkan::RendererVulkan::RenderScreenshot(const Tegra::FramebufferConfig& fr
     });
     const VkExtent2D render_area{.width = layout.width, .height = layout.height};
     const vk::Framebuffer screenshot_fb = blit_screen.CreateFramebuffer(*dst_view, render_area);
-    // Since we're not rendering to the screen, ignore the render semaphore.
-    void(blit_screen.Draw(framebuffer, *screenshot_fb, layout, render_area, use_accelerated));
+    blit_screen.Draw(framebuffer, *screenshot_fb, layout, render_area, use_accelerated);
 
     const auto buffer_size = static_cast<VkDeviceSize>(layout.width * layout.height * 4);
     const VkBufferCreateInfo dst_buffer_info{
@@ -270,7 +243,7 @@ void Vulkan::RendererVulkan::RenderScreenshot(const Tegra::FramebufferConfig& fr
             .pNext = nullptr,
             .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
             .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
-            .oldLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
+            .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
             .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
             .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
             .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,

src/video_core/renderer_vulkan/renderer_vulkan.h

@@ -9,6 +9,7 @@
 #include "common/dynamic_library.h"
 #include "video_core/renderer_base.h"
 #include "video_core/renderer_vulkan/vk_blit_screen.h"
+#include "video_core/renderer_vulkan/vk_present_manager.h"
 #include "video_core/renderer_vulkan/vk_rasterizer.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_state_tracker.h"
@@ -76,6 +77,7 @@ private:
     StateTracker state_tracker;
     Scheduler scheduler;
     Swapchain swapchain;
+    PresentManager present_manager;
     BlitScreen blit_screen;
     RasterizerVulkan rasterizer;
     std::optional<TurboMode> turbo_mode;

src/video_core/renderer_vulkan/vk_blit_screen.cpp

@@ -122,10 +122,12 @@ struct BlitScreen::BufferData {
 
 BlitScreen::BlitScreen(Core::Memory::Memory& cpu_memory_, Core::Frontend::EmuWindow& render_window_,
                        const Device& device_, MemoryAllocator& memory_allocator_,
-                       Swapchain& swapchain_, Scheduler& scheduler_, const ScreenInfo& screen_info_)
+                       Swapchain& swapchain_, PresentManager& present_manager_,
+                       Scheduler& scheduler_, const ScreenInfo& screen_info_)
     : cpu_memory{cpu_memory_}, render_window{render_window_}, device{device_},
-      memory_allocator{memory_allocator_}, swapchain{swapchain_}, scheduler{scheduler_},
-      image_count{swapchain.GetImageCount()}, screen_info{screen_info_} {
+      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()} {
     resource_ticks.resize(image_count);
 
     CreateStaticResources();
@@ -135,25 +137,20 @@ BlitScreen::BlitScreen(Core::Memory::Memory& cpu_memory_, Core::Frontend::EmuWin
 BlitScreen::~BlitScreen() = default;
 
 void BlitScreen::Recreate() {
+    present_manager.WaitPresent();
+    scheduler.Finish();
+    device.GetLogical().WaitIdle();
     CreateDynamicResources();
 }
 
-VkSemaphore BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
-                             const VkFramebuffer& host_framebuffer,
-                             const Layout::FramebufferLayout layout, VkExtent2D render_area,
-                             bool use_accelerated) {
+void BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
+                      const VkFramebuffer& host_framebuffer, const Layout::FramebufferLayout layout,
+                      VkExtent2D render_area, bool use_accelerated) {
     RefreshResources(framebuffer);
 
     // Finish any pending renderpass
     scheduler.RequestOutsideRenderPassOperationContext();
 
-    if (const auto swapchain_images = swapchain.GetImageCount(); swapchain_images != image_count) {
-        image_count = swapchain_images;
-        Recreate();
-    }
-
-    const std::size_t image_index = swapchain.GetImageIndex();
-
     scheduler.Wait(resource_ticks[image_index]);
     resource_ticks[image_index] = scheduler.CurrentTick();
 
@@ -169,7 +166,7 @@ VkSemaphore BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
     std::memcpy(mapped_span.data(), &data, sizeof(data));
 
     if (!use_accelerated) {
-        const u64 image_offset = GetRawImageOffset(framebuffer, image_index);
+        const u64 image_offset = GetRawImageOffset(framebuffer);
 
         const VAddr framebuffer_addr = framebuffer.address + framebuffer.offset;
         const u8* const host_ptr = cpu_memory.GetPointer(framebuffer_addr);
@@ -204,8 +201,8 @@ VkSemaphore BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
                     .depth = 1,
                 },
         };
-        scheduler.Record([this, copy, image_index](vk::CommandBuffer cmdbuf) {
-            const VkImage image = *raw_images[image_index];
+        scheduler.Record([this, copy, index = image_index](vk::CommandBuffer cmdbuf) {
+            const VkImage image = *raw_images[index];
             const VkImageMemoryBarrier base_barrier{
                 .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                 .pNext = nullptr,
@@ -245,14 +242,15 @@ VkSemaphore BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
 
     const auto anti_alias_pass = Settings::values.anti_aliasing.GetValue();
     if (use_accelerated && anti_alias_pass == Settings::AntiAliasing::Fxaa) {
-        UpdateAADescriptorSet(image_index, source_image_view, false);
+        UpdateAADescriptorSet(source_image_view, false);
         const u32 up_scale = Settings::values.resolution_info.up_scale;
         const u32 down_shift = Settings::values.resolution_info.down_shift;
         VkExtent2D size{
             .width = (up_scale * framebuffer.width) >> down_shift,
             .height = (up_scale * framebuffer.height) >> down_shift,
         };
-        scheduler.Record([this, image_index, size, anti_alias_pass](vk::CommandBuffer cmdbuf) {
+        scheduler.Record([this, index = image_index, size,
+                          anti_alias_pass](vk::CommandBuffer cmdbuf) {
             const VkImageMemoryBarrier base_barrier{
                 .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                 .pNext = nullptr,
@@ -326,7 +324,7 @@ VkSemaphore BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
 
             cmdbuf.BindVertexBuffer(0, *buffer, offsetof(BufferData, vertices));
             cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *aa_pipeline_layout, 0,
-                                      aa_descriptor_sets[image_index], {});
+                                      aa_descriptor_sets[index], {});
             cmdbuf.Draw(4, 1, 0, 0);
             cmdbuf.EndRenderPass();
 
@@ -369,81 +367,99 @@ VkSemaphore BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
         };
         VkImageView fsr_image_view =
             fsr->Draw(scheduler, image_index, source_image_view, fsr_input_size, crop_rect);
-        UpdateDescriptorSet(image_index, fsr_image_view, true);
+        UpdateDescriptorSet(fsr_image_view, true);
     } else {
         const bool is_nn =
             Settings::values.scaling_filter.GetValue() == Settings::ScalingFilter::NearestNeighbor;
-        UpdateDescriptorSet(image_index, source_image_view, is_nn);
+        UpdateDescriptorSet(source_image_view, is_nn);
     }
 
-    scheduler.Record(
-        [this, host_framebuffer, image_index, size = render_area](vk::CommandBuffer cmdbuf) {
-            const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f;
-            const f32 bg_green = Settings::values.bg_green.GetValue() / 255.0f;
-            const f32 bg_blue = Settings::values.bg_blue.GetValue() / 255.0f;
-            const VkClearValue clear_color{
-                .color = {.float32 = {bg_red, bg_green, bg_blue, 1.0f}},
-            };
-            const VkRenderPassBeginInfo renderpass_bi{
-                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
-                .pNext = nullptr,
-                .renderPass = *renderpass,
-                .framebuffer = host_framebuffer,
-                .renderArea =
-                    {
-                        .offset = {0, 0},
-                        .extent = size,
-                    },
-                .clearValueCount = 1,
-                .pClearValues = &clear_color,
-            };
-            const VkViewport viewport{
-                .x = 0.0f,
-                .y = 0.0f,
-                .width = static_cast<float>(size.width),
-                .height = static_cast<float>(size.height),
-                .minDepth = 0.0f,
-                .maxDepth = 1.0f,
-            };
-            const VkRect2D scissor{
-                .offset = {0, 0},
-                .extent = size,
-            };
-            cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
-            auto graphics_pipeline = [this]() {
-                switch (Settings::values.scaling_filter.GetValue()) {
-                case Settings::ScalingFilter::NearestNeighbor:
-                case Settings::ScalingFilter::Bilinear:
-                    return *bilinear_pipeline;
-                case Settings::ScalingFilter::Bicubic:
-                    return *bicubic_pipeline;
-                case Settings::ScalingFilter::Gaussian:
-                    return *gaussian_pipeline;
-                case Settings::ScalingFilter::ScaleForce:
-                    return *scaleforce_pipeline;
-                default:
-                    return *bilinear_pipeline;
-                }
-            }();
-            cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, graphics_pipeline);
-            cmdbuf.SetViewport(0, viewport);
-            cmdbuf.SetScissor(0, scissor);
-
-            cmdbuf.BindVertexBuffer(0, *buffer, offsetof(BufferData, vertices));
-            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline_layout, 0,
-                                      descriptor_sets[image_index], {});
-            cmdbuf.Draw(4, 1, 0, 0);
-            cmdbuf.EndRenderPass();
-        });
-    return *semaphores[image_index];
+    scheduler.Record([this, host_framebuffer, index = image_index,
+                      size = render_area](vk::CommandBuffer cmdbuf) {
+        const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f;
+        const f32 bg_green = Settings::values.bg_green.GetValue() / 255.0f;
+        const f32 bg_blue = Settings::values.bg_blue.GetValue() / 255.0f;
+        const VkClearValue clear_color{
+            .color = {.float32 = {bg_red, bg_green, bg_blue, 1.0f}},
+        };
+        const VkRenderPassBeginInfo renderpass_bi{
+            .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
+            .pNext = nullptr,
+            .renderPass = *renderpass,
+            .framebuffer = host_framebuffer,
+            .renderArea =
+                {
+                    .offset = {0, 0},
+                    .extent = size,
+                },
+            .clearValueCount = 1,
+            .pClearValues = &clear_color,
+        };
+        const VkViewport viewport{
+            .x = 0.0f,
+            .y = 0.0f,
+            .width = static_cast<float>(size.width),
+            .height = static_cast<float>(size.height),
+            .minDepth = 0.0f,
+            .maxDepth = 1.0f,
+        };
+        const VkRect2D scissor{
+            .offset = {0, 0},
+            .extent = size,
+        };
+        cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
+        auto graphics_pipeline = [this]() {
+            switch (Settings::values.scaling_filter.GetValue()) {
+            case Settings::ScalingFilter::NearestNeighbor:
+            case Settings::ScalingFilter::Bilinear:
+                return *bilinear_pipeline;
+            case Settings::ScalingFilter::Bicubic:
+                return *bicubic_pipeline;
+            case Settings::ScalingFilter::Gaussian:
+                return *gaussian_pipeline;
+            case Settings::ScalingFilter::ScaleForce:
+                return *scaleforce_pipeline;
+            default:
+                return *bilinear_pipeline;
+            }
+        }();
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, graphics_pipeline);
+        cmdbuf.SetViewport(0, viewport);
+        cmdbuf.SetScissor(0, scissor);
+
+        cmdbuf.BindVertexBuffer(0, *buffer, offsetof(BufferData, vertices));
+        cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline_layout, 0,
+                                  descriptor_sets[index], {});
+        cmdbuf.Draw(4, 1, 0, 0);
+        cmdbuf.EndRenderPass();
+    });
 }
 
-VkSemaphore BlitScreen::DrawToSwapchain(const Tegra::FramebufferConfig& framebuffer,
-                                        bool use_accelerated) {
-    const std::size_t image_index = swapchain.GetImageIndex();
-    const VkExtent2D render_area = swapchain.GetSize();
+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
+    if (const std::size_t swapchain_images = swapchain.GetImageCount();
+        swapchain_images != image_count || current_srgb != is_srgb) {
+        current_srgb = is_srgb;
+        image_view_format = current_srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM;
+        image_count = swapchain_images;
+        Recreate();
+    }
+
+    // Recreate the presentation frame if the dimensions of the window changed
     const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout();
-    return Draw(framebuffer, *framebuffers[image_index], layout, render_area, use_accelerated);
+    if (layout.width != frame->width || layout.height != frame->height ||
+        is_srgb != frame->is_srgb) {
+        Recreate();
+        present_manager.RecreateFrame(frame, layout.width, layout.height, is_srgb,
+                                      image_view_format, *renderpass);
+    }
+
+    const VkExtent2D render_area{frame->width, frame->height};
+    Draw(framebuffer, *frame->framebuffer, layout, render_area, use_accelerated);
+    if (++image_index >= image_count) {
+        image_index = 0;
+    }
 }
 
 vk::Framebuffer BlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent) {
@@ -471,13 +487,11 @@ void BlitScreen::CreateStaticResources() {
 }
 
 void BlitScreen::CreateDynamicResources() {
-    CreateSemaphores();
     CreateDescriptorPool();
     CreateDescriptorSetLayout();
     CreateDescriptorSets();
     CreatePipelineLayout();
     CreateRenderPass();
-    CreateFramebuffers();
     CreateGraphicsPipeline();
     fsr.reset();
     smaa.reset();
@@ -525,11 +539,6 @@ void BlitScreen::CreateShaders() {
     }
 }
 
-void BlitScreen::CreateSemaphores() {
-    semaphores.resize(image_count);
-    std::ranges::generate(semaphores, [this] { return device.GetLogical().CreateSemaphore(); });
-}
-
 void BlitScreen::CreateDescriptorPool() {
     const std::array<VkDescriptorPoolSize, 2> pool_sizes{{
         {
@@ -571,10 +580,10 @@ void BlitScreen::CreateDescriptorPool() {
 }
 
 void BlitScreen::CreateRenderPass() {
-    renderpass = CreateRenderPassImpl(swapchain.GetImageViewFormat());
+    renderpass = CreateRenderPassImpl(image_view_format);
 }
 
-vk::RenderPass BlitScreen::CreateRenderPassImpl(VkFormat format, bool is_present) {
+vk::RenderPass BlitScreen::CreateRenderPassImpl(VkFormat format) {
     const VkAttachmentDescription color_attachment{
         .flags = 0,
         .format = format,
@@ -584,7 +593,7 @@ vk::RenderPass BlitScreen::CreateRenderPassImpl(VkFormat format, bool is_present
         .stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE,
         .stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE,
         .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
-        .finalLayout = is_present ? VK_IMAGE_LAYOUT_PRESENT_SRC_KHR : VK_IMAGE_LAYOUT_GENERAL,
+        .finalLayout = VK_IMAGE_LAYOUT_GENERAL,
     };
 
     const VkAttachmentReference color_attachment_ref{
@@ -1052,16 +1061,6 @@ void BlitScreen::CreateSampler() {
     nn_sampler = device.GetLogical().CreateSampler(ci_nn);
 }
 
-void BlitScreen::CreateFramebuffers() {
-    const VkExtent2D size{swapchain.GetSize()};
-    framebuffers.resize(image_count);
-
-    for (std::size_t i = 0; i < image_count; ++i) {
-        const VkImageView image_view{swapchain.GetImageViewIndex(i)};
-        framebuffers[i] = CreateFramebuffer(image_view, size, renderpass);
-    }
-}
-
 void BlitScreen::ReleaseRawImages() {
     for (const u64 tick : resource_ticks) {
         scheduler.Wait(tick);
@@ -1175,7 +1174,7 @@ void BlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) {
         aa_framebuffer = CreateFramebuffer(*aa_image_view, size, aa_renderpass);
         return;
     }
-    aa_renderpass = CreateRenderPassImpl(GetFormat(framebuffer), false);
+    aa_renderpass = CreateRenderPassImpl(GetFormat(framebuffer));
     aa_framebuffer = CreateFramebuffer(*aa_image_view, size, aa_renderpass);
 
     const std::array<VkPipelineShaderStageCreateInfo, 2> fxaa_shader_stages{{
@@ -1319,8 +1318,7 @@ void BlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) {
     aa_pipeline = device.GetLogical().CreateGraphicsPipeline(fxaa_pipeline_ci);
 }
 
-void BlitScreen::UpdateAADescriptorSet(std::size_t image_index, VkImageView image_view,
-                                       bool nn) const {
+void BlitScreen::UpdateAADescriptorSet(VkImageView image_view, bool nn) const {
     const VkDescriptorImageInfo image_info{
         .sampler = nn ? *nn_sampler : *sampler,
         .imageView = image_view,
@@ -1356,8 +1354,7 @@ void BlitScreen::UpdateAADescriptorSet(std::size_t image_index, VkImageView imag
     device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, sampler_write_2}, {});
 }
 
-void BlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view,
-                                     bool nn) const {
+void BlitScreen::UpdateDescriptorSet(VkImageView image_view, bool nn) const {
     const VkDescriptorBufferInfo buffer_info{
         .buffer = *buffer,
         .offset = offsetof(BufferData, uniform),
@@ -1480,8 +1477,7 @@ u64 BlitScreen::CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer)
     return sizeof(BufferData) + GetSizeInBytes(framebuffer) * image_count;
 }
 
-u64 BlitScreen::GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer,
-                                  std::size_t image_index) const {
+u64 BlitScreen::GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer) const {
     constexpr auto first_image_offset = static_cast<u64>(sizeof(BufferData));
     return first_image_offset + GetSizeInBytes(framebuffer) * image_index;
 }

src/video_core/renderer_vulkan/vk_blit_screen.h

@@ -5,6 +5,7 @@
 
 #include <memory>
 
+#include "core/frontend/framebuffer_layout.h"
 #include "video_core/vulkan_common/vulkan_memory_allocator.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
@@ -42,6 +43,9 @@ class RasterizerVulkan;
 class Scheduler;
 class SMAA;
 class Swapchain;
+class PresentManager;
+
+struct Frame;
 
 struct ScreenInfo {
     VkImage image{};
@@ -55,18 +59,17 @@ class BlitScreen {
 public:
     explicit BlitScreen(Core::Memory::Memory& cpu_memory, Core::Frontend::EmuWindow& render_window,
                         const Device& device, MemoryAllocator& memory_manager, Swapchain& swapchain,
-                        Scheduler& scheduler, const ScreenInfo& screen_info);
+                        PresentManager& present_manager, Scheduler& scheduler,
+                        const ScreenInfo& screen_info);
     ~BlitScreen();
 
     void Recreate();
 
-    [[nodiscard]] VkSemaphore Draw(const Tegra::FramebufferConfig& framebuffer,
-                                   const VkFramebuffer& host_framebuffer,
-                                   const Layout::FramebufferLayout layout, VkExtent2D render_area,
-                                   bool use_accelerated);
+    void Draw(const Tegra::FramebufferConfig& framebuffer, const VkFramebuffer& host_framebuffer,
+              const Layout::FramebufferLayout layout, VkExtent2D render_area, bool use_accelerated);
 
-    [[nodiscard]] VkSemaphore DrawToSwapchain(const Tegra::FramebufferConfig& framebuffer,
-                                              bool use_accelerated);
+    void DrawToSwapchain(Frame* frame, const Tegra::FramebufferConfig& framebuffer,
+                         bool use_accelerated, bool is_srgb);
 
     [[nodiscard]] vk::Framebuffer CreateFramebuffer(const VkImageView& image_view,
                                                     VkExtent2D extent);
@@ -79,10 +82,9 @@ private:
 
     void CreateStaticResources();
     void CreateShaders();
-    void CreateSemaphores();
     void CreateDescriptorPool();
     void CreateRenderPass();
-    vk::RenderPass CreateRenderPassImpl(VkFormat, bool is_present = true);
+    vk::RenderPass CreateRenderPassImpl(VkFormat format);
     void CreateDescriptorSetLayout();
     void CreateDescriptorSets();
     void CreatePipelineLayout();
@@ -90,15 +92,14 @@ private:
     void CreateSampler();
 
     void CreateDynamicResources();
-    void CreateFramebuffers();
 
     void RefreshResources(const Tegra::FramebufferConfig& framebuffer);
     void ReleaseRawImages();
     void CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer);
     void CreateRawImages(const Tegra::FramebufferConfig& framebuffer);
 
-    void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view, bool nn) const;
-    void UpdateAADescriptorSet(std::size_t image_index, VkImageView image_view, bool nn) const;
+    void UpdateDescriptorSet(VkImageView image_view, bool nn) const;
+    void UpdateAADescriptorSet(VkImageView image_view, bool nn) const;
     void SetUniformData(BufferData& data, const Layout::FramebufferLayout layout) const;
     void SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer,
                        const Layout::FramebufferLayout layout) const;
@@ -107,16 +108,17 @@ private:
     void CreateFSR();
 
     u64 CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const;
-    u64 GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer,
-                          std::size_t image_index) const;
+    u64 GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer) const;
 
     Core::Memory::Memory& cpu_memory;
     Core::Frontend::EmuWindow& render_window;
     const Device& device;
     MemoryAllocator& memory_allocator;
     Swapchain& swapchain;
+    PresentManager& present_manager;
     Scheduler& scheduler;
     std::size_t image_count;
+    std::size_t image_index{};
     const ScreenInfo& screen_info;
 
     vk::ShaderModule vertex_shader;
@@ -135,7 +137,6 @@ private:
     vk::Pipeline gaussian_pipeline;
     vk::Pipeline scaleforce_pipeline;
     vk::RenderPass renderpass;
-    std::vector<vk::Framebuffer> framebuffers;
     vk::DescriptorSets descriptor_sets;
     vk::Sampler nn_sampler;
     vk::Sampler sampler;
@@ -145,7 +146,6 @@ private:
 
     std::vector<u64> resource_ticks;
 
-    std::vector<vk::Semaphore> semaphores;
     std::vector<vk::Image> raw_images;
     std::vector<vk::ImageView> raw_image_views;
     std::vector<MemoryCommit> raw_buffer_commits;
@@ -164,6 +164,8 @@ private:
     u32 raw_width = 0;
     u32 raw_height = 0;
     Service::android::PixelFormat pixel_format{};
+    bool current_srgb;
+    VkFormat image_view_format;
 
     std::unique_ptr<FSR> fsr;
     std::unique_ptr<SMAA> smaa;

src/video_core/renderer_vulkan/vk_present_manager.cpp

@@ -0,0 +1,454 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "common/microprofile.h"
+#include "common/settings.h"
+#include "common/thread.h"
+#include "video_core/renderer_vulkan/vk_present_manager.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_swapchain.h"
+#include "video_core/vulkan_common/vulkan_device.h"
+
+namespace Vulkan {
+
+MICROPROFILE_DEFINE(Vulkan_WaitPresent, "Vulkan", "Wait For Present", MP_RGB(128, 128, 128));
+MICROPROFILE_DEFINE(Vulkan_CopyToSwapchain, "Vulkan", "Copy to swapchain", MP_RGB(192, 255, 192));
+
+namespace {
+
+bool CanBlitToSwapchain(const vk::PhysicalDevice& physical_device, VkFormat format) {
+    const VkFormatProperties props{physical_device.GetFormatProperties(format)};
+    return (props.optimalTilingFeatures & VK_FORMAT_FEATURE_BLIT_DST_BIT);
+}
+
+[[nodiscard]] VkImageSubresourceLayers MakeImageSubresourceLayers() {
+    return VkImageSubresourceLayers{
+        .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+        .mipLevel = 0,
+        .baseArrayLayer = 0,
+        .layerCount = 1,
+    };
+}
+
+[[nodiscard]] VkImageBlit MakeImageBlit(s32 frame_width, s32 frame_height, s32 swapchain_width,
+                                        s32 swapchain_height) {
+    return VkImageBlit{
+        .srcSubresource = MakeImageSubresourceLayers(),
+        .srcOffsets =
+            {
+                {
+                    .x = 0,
+                    .y = 0,
+                    .z = 0,
+                },
+                {
+                    .x = frame_width,
+                    .y = frame_height,
+                    .z = 1,
+                },
+            },
+        .dstSubresource = MakeImageSubresourceLayers(),
+        .dstOffsets =
+            {
+                {
+                    .x = 0,
+                    .y = 0,
+                    .z = 0,
+                },
+                {
+                    .x = swapchain_width,
+                    .y = swapchain_height,
+                    .z = 1,
+                },
+            },
+    };
+}
+
+[[nodiscard]] VkImageCopy MakeImageCopy(u32 frame_width, u32 frame_height, u32 swapchain_width,
+                                        u32 swapchain_height) {
+    return VkImageCopy{
+        .srcSubresource = MakeImageSubresourceLayers(),
+        .srcOffset =
+            {
+                .x = 0,
+                .y = 0,
+                .z = 0,
+            },
+        .dstSubresource = MakeImageSubresourceLayers(),
+        .dstOffset =
+            {
+                .x = 0,
+                .y = 0,
+                .z = 0,
+            },
+        .extent =
+            {
+                .width = std::min(frame_width, swapchain_width),
+                .height = std::min(frame_height, swapchain_height),
+                .depth = 1,
+            },
+    };
+}
+
+} // Anonymous namespace
+
+PresentManager::PresentManager(Core::Frontend::EmuWindow& render_window_, const Device& device_,
+                               MemoryAllocator& memory_allocator_, Scheduler& scheduler_,
+                               Swapchain& swapchain_)
+    : render_window{render_window_}, device{device_},
+      memory_allocator{memory_allocator_}, scheduler{scheduler_}, swapchain{swapchain_},
+      blit_supported{CanBlitToSwapchain(device.GetPhysical(), swapchain.GetImageViewFormat())},
+      use_present_thread{Settings::values.async_presentation.GetValue()},
+      image_count{swapchain.GetImageCount()} {
+
+    auto& dld = device.GetLogical();
+    cmdpool = dld.CreateCommandPool({
+        .sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO,
+        .pNext = nullptr,
+        .flags =
+            VK_COMMAND_POOL_CREATE_TRANSIENT_BIT | VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT,
+        .queueFamilyIndex = device.GetGraphicsFamily(),
+    });
+    auto cmdbuffers = cmdpool.Allocate(image_count);
+
+    frames.resize(image_count);
+    for (u32 i = 0; i < frames.size(); i++) {
+        Frame& frame = frames[i];
+        frame.cmdbuf = vk::CommandBuffer{cmdbuffers[i], device.GetDispatchLoader()};
+        frame.render_ready = dld.CreateSemaphore({
+            .sType = VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+        });
+        frame.present_done = dld.CreateFence({
+            .sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = VK_FENCE_CREATE_SIGNALED_BIT,
+        });
+        free_queue.push(&frame);
+    }
+
+    if (use_present_thread) {
+        present_thread = std::jthread([this](std::stop_token token) { PresentThread(token); });
+    }
+}
+
+PresentManager::~PresentManager() = default;
+
+Frame* PresentManager::GetRenderFrame() {
+    MICROPROFILE_SCOPE(Vulkan_WaitPresent);
+
+    // Wait for free presentation frames
+    std::unique_lock lock{free_mutex};
+    free_cv.wait(lock, [this] { return !free_queue.empty(); });
+
+    // Take the frame from the queue
+    Frame* frame = free_queue.front();
+    free_queue.pop();
+
+    // Wait for the presentation to be finished so all frame resources are free
+    frame->present_done.Wait();
+    frame->present_done.Reset();
+
+    return frame;
+}
+
+void PresentManager::PushFrame(Frame* frame) {
+    if (!use_present_thread) {
+        CopyToSwapchain(frame);
+        free_queue.push(frame);
+        return;
+    }
+
+    std::unique_lock lock{queue_mutex};
+    present_queue.push(frame);
+    frame_cv.notify_one();
+}
+
+void PresentManager::RecreateFrame(Frame* frame, u32 width, u32 height, bool is_srgb,
+                                   VkFormat image_view_format, VkRenderPass rd) {
+    auto& dld = device.GetLogical();
+
+    frame->width = width;
+    frame->height = height;
+    frame->is_srgb = is_srgb;
+
+    frame->image = dld.CreateImage({
+        .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT,
+        .imageType = VK_IMAGE_TYPE_2D,
+        .format = swapchain.GetImageFormat(),
+        .extent =
+            {
+                .width = width,
+                .height = height,
+                .depth = 1,
+            },
+        .mipLevels = 1,
+        .arrayLayers = 1,
+        .samples = VK_SAMPLE_COUNT_1_BIT,
+        .tiling = VK_IMAGE_TILING_OPTIMAL,
+        .usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+        .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+    });
+
+    frame->image_commit = memory_allocator.Commit(frame->image, MemoryUsage::DeviceLocal);
+
+    frame->image_view = dld.CreateImageView({
+        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .image = *frame->image,
+        .viewType = VK_IMAGE_VIEW_TYPE_2D,
+        .format = image_view_format,
+        .components =
+            {
+                .r = VK_COMPONENT_SWIZZLE_IDENTITY,
+                .g = VK_COMPONENT_SWIZZLE_IDENTITY,
+                .b = VK_COMPONENT_SWIZZLE_IDENTITY,
+                .a = VK_COMPONENT_SWIZZLE_IDENTITY,
+            },
+        .subresourceRange =
+            {
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel = 0,
+                .levelCount = 1,
+                .baseArrayLayer = 0,
+                .layerCount = 1,
+            },
+    });
+
+    const VkImageView image_view{*frame->image_view};
+    frame->framebuffer = dld.CreateFramebuffer({
+        .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .renderPass = rd,
+        .attachmentCount = 1,
+        .pAttachments = &image_view,
+        .width = width,
+        .height = height,
+        .layers = 1,
+    });
+}
+
+void PresentManager::WaitPresent() {
+    if (!use_present_thread) {
+        return;
+    }
+
+    // Wait for the present queue to be empty
+    {
+        std::unique_lock queue_lock{queue_mutex};
+        frame_cv.wait(queue_lock, [this] { return present_queue.empty(); });
+    }
+
+    // The above condition will be satisfied when the last frame is taken from the queue.
+    // To ensure that frame has been presented as well take hold of the swapchain
+    // mutex.
+    std::scoped_lock swapchain_lock{swapchain_mutex};
+}
+
+void PresentManager::PresentThread(std::stop_token token) {
+    Common::SetCurrentThreadName("VulkanPresent");
+    while (!token.stop_requested()) {
+        std::unique_lock lock{queue_mutex};
+
+        // Wait for presentation frames
+        Common::CondvarWait(frame_cv, lock, token, [this] { return !present_queue.empty(); });
+        if (token.stop_requested()) {
+            return;
+        }
+
+        // Take the frame and notify anyone waiting
+        Frame* frame = present_queue.front();
+        present_queue.pop();
+        frame_cv.notify_one();
+
+        // By exchanging the lock ownership we take the swapchain lock
+        // before the queue lock goes out of scope. This way the swapchain
+        // lock in WaitPresent is guaranteed to occur after here.
+        std::exchange(lock, std::unique_lock{swapchain_mutex});
+
+        CopyToSwapchain(frame);
+
+        // Free the frame for reuse
+        std::scoped_lock fl{free_mutex};
+        free_queue.push(frame);
+        free_cv.notify_one();
+    }
+}
+
+void PresentManager::CopyToSwapchain(Frame* frame) {
+    MICROPROFILE_SCOPE(Vulkan_CopyToSwapchain);
+
+    const auto recreate_swapchain = [&] {
+        swapchain.Create(frame->width, frame->height, frame->is_srgb);
+        image_count = swapchain.GetImageCount();
+    };
+
+    // If the size or colorspace of the incoming frames has changed, recreate the swapchain
+    // to account for that.
+    const bool srgb_changed = swapchain.NeedsRecreation(frame->is_srgb);
+    const bool size_changed =
+        swapchain.GetWidth() != frame->width || swapchain.GetHeight() != frame->height;
+    if (srgb_changed || size_changed) {
+        recreate_swapchain();
+    }
+
+    while (swapchain.AcquireNextImage()) {
+        recreate_swapchain();
+    }
+
+    const vk::CommandBuffer cmdbuf{frame->cmdbuf};
+    cmdbuf.Begin({
+        .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
+        .pNext = nullptr,
+        .flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
+        .pInheritanceInfo = nullptr,
+    });
+
+    const VkImage image{swapchain.CurrentImage()};
+    const VkExtent2D extent = swapchain.GetExtent();
+    const std::array pre_barriers{
+        VkImageMemoryBarrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = 0,
+            .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = image,
+            .subresourceRange{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel = 0,
+                .levelCount = 1,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        },
+        VkImageMemoryBarrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = *frame->image,
+            .subresourceRange{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel = 0,
+                .levelCount = 1,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        },
+    };
+    const std::array post_barriers{
+        VkImageMemoryBarrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = image,
+            .subresourceRange{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel = 0,
+                .levelCount = 1,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        },
+        VkImageMemoryBarrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = *frame->image,
+            .subresourceRange{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel = 0,
+                .levelCount = 1,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        },
+    };
+
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, {},
+                           {}, {}, pre_barriers);
+
+    if (blit_supported) {
+        cmdbuf.BlitImage(*frame->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, image,
+                         VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+                         MakeImageBlit(frame->width, frame->height, extent.width, extent.height),
+                         VK_FILTER_LINEAR);
+    } else {
+        cmdbuf.CopyImage(*frame->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, image,
+                         VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+                         MakeImageCopy(frame->width, frame->height, extent.width, extent.height));
+    }
+
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, {},
+                           {}, {}, post_barriers);
+
+    cmdbuf.End();
+
+    const VkSemaphore present_semaphore = swapchain.CurrentPresentSemaphore();
+    const VkSemaphore render_semaphore = swapchain.CurrentRenderSemaphore();
+    const std::array wait_semaphores = {present_semaphore, *frame->render_ready};
+
+    static constexpr std::array<VkPipelineStageFlags, 2> wait_stage_masks{
+        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
+        VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+    };
+
+    const VkSubmitInfo submit_info{
+        .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
+        .pNext = nullptr,
+        .waitSemaphoreCount = 2U,
+        .pWaitSemaphores = wait_semaphores.data(),
+        .pWaitDstStageMask = wait_stage_masks.data(),
+        .commandBufferCount = 1,
+        .pCommandBuffers = cmdbuf.address(),
+        .signalSemaphoreCount = 1U,
+        .pSignalSemaphores = &render_semaphore,
+    };
+
+    // Submit the image copy/blit to the swapchain
+    {
+        std::scoped_lock lock{scheduler.submit_mutex};
+        switch (const VkResult result =
+                    device.GetGraphicsQueue().Submit(submit_info, *frame->present_done)) {
+        case VK_SUCCESS:
+            break;
+        case VK_ERROR_DEVICE_LOST:
+            device.ReportLoss();
+            [[fallthrough]];
+        default:
+            vk::Check(result);
+            break;
+        }
+    }
+
+    // Present
+    swapchain.Present(render_semaphore);
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_present_manager.h

@@ -0,0 +1,83 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <condition_variable>
+#include <mutex>
+#include <queue>
+
+#include "common/common_types.h"
+#include "common/polyfill_thread.h"
+#include "video_core/vulkan_common/vulkan_memory_allocator.h"
+#include "video_core/vulkan_common/vulkan_wrapper.h"
+
+namespace Core::Frontend {
+class EmuWindow;
+} // namespace Core::Frontend
+
+namespace Vulkan {
+
+class Device;
+class Scheduler;
+class Swapchain;
+
+struct Frame {
+    u32 width;
+    u32 height;
+    bool is_srgb;
+    vk::Image image;
+    vk::ImageView image_view;
+    vk::Framebuffer framebuffer;
+    MemoryCommit image_commit;
+    vk::CommandBuffer cmdbuf;
+    vk::Semaphore render_ready;
+    vk::Fence present_done;
+};
+
+class PresentManager {
+public:
+    PresentManager(Core::Frontend::EmuWindow& render_window, const Device& device,
+                   MemoryAllocator& memory_allocator, Scheduler& scheduler, Swapchain& swapchain);
+    ~PresentManager();
+
+    /// Returns the last used presentation frame
+    Frame* GetRenderFrame();
+
+    /// Pushes a frame for presentation
+    void PushFrame(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);
+
+    /// Waits for the present thread to finish presenting all queued frames.
+    void WaitPresent();
+
+private:
+    void PresentThread(std::stop_token token);
+
+    void CopyToSwapchain(Frame* frame);
+
+private:
+    Core::Frontend::EmuWindow& render_window;
+    const Device& device;
+    MemoryAllocator& memory_allocator;
+    Scheduler& scheduler;
+    Swapchain& swapchain;
+    vk::CommandPool cmdpool;
+    std::vector<Frame> frames;
+    std::queue<Frame*> present_queue;
+    std::queue<Frame*> free_queue;
+    std::condition_variable_any frame_cv;
+    std::condition_variable free_cv;
+    std::mutex swapchain_mutex;
+    std::mutex queue_mutex;
+    std::mutex free_mutex;
+    std::jthread present_thread;
+    bool blit_supported;
+    bool use_present_thread;
+    std::size_t image_count;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -46,10 +46,11 @@ Scheduler::Scheduler(const Device& device_, StateTracker& state_tracker_)
 
 Scheduler::~Scheduler() = default;
 
-void Scheduler::Flush(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) {
+u64 Scheduler::Flush(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) {
     // When flushing, we only send data to the worker thread; no waiting is necessary.
-    SubmitExecution(signal_semaphore, wait_semaphore);
+    const u64 signal_value = SubmitExecution(signal_semaphore, wait_semaphore);
     AllocateNewContext();
+    return signal_value;
 }
 
 void Scheduler::Finish(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) {
@@ -205,7 +206,7 @@ void Scheduler::AllocateWorkerCommandBuffer() {
     });
 }
 
-void Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) {
+u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) {
     EndPendingOperations();
     InvalidateState();
 
@@ -217,6 +218,7 @@ void Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_s
             on_submit();
         }
 
+        std::scoped_lock lock{submit_mutex};
         switch (const VkResult result = master_semaphore->SubmitQueue(
                     cmdbuf, signal_semaphore, wait_semaphore, signal_value)) {
         case VK_SUCCESS:
@@ -231,6 +233,7 @@ void Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_s
     });
     chunk->MarkSubmit();
     DispatchWork();
+    return signal_value;
 }
 
 void Scheduler::AllocateNewContext() {

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -34,7 +34,7 @@ public:
     ~Scheduler();
 
     /// Sends the current execution context to the GPU.
-    void Flush(VkSemaphore signal_semaphore = nullptr, VkSemaphore wait_semaphore = nullptr);
+    u64 Flush(VkSemaphore signal_semaphore = nullptr, VkSemaphore wait_semaphore = nullptr);
 
     /// Sends the current execution context to the GPU and waits for it to complete.
     void Finish(VkSemaphore signal_semaphore = nullptr, VkSemaphore wait_semaphore = nullptr);
@@ -106,6 +106,8 @@ public:
         return *master_semaphore;
     }
 
+    std::mutex submit_mutex;
+
 private:
     class Command {
     public:
@@ -201,7 +203,7 @@ private:
 
     void AllocateWorkerCommandBuffer();
 
-    void SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore);
+    u64 SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore);
 
     void AllocateNewContext();
 

src/video_core/renderer_vulkan/vk_swapchain.cpp

@@ -99,18 +99,16 @@ void Swapchain::Create(u32 width_, u32 height_, bool srgb) {
         return;
     }
 
-    device.GetLogical().WaitIdle();
     Destroy();
 
     CreateSwapchain(capabilities, srgb);
     CreateSemaphores();
-    CreateImageViews();
 
     resource_ticks.clear();
     resource_ticks.resize(image_count);
 }
 
-void Swapchain::AcquireNextImage() {
+bool Swapchain::AcquireNextImage() {
     const VkResult result = device.GetLogical().AcquireNextImageKHR(
         *swapchain, std::numeric_limits<u64>::max(), *present_semaphores[frame_index],
         VK_NULL_HANDLE, &image_index);
@@ -127,8 +125,11 @@ void Swapchain::AcquireNextImage() {
         LOG_ERROR(Render_Vulkan, "vkAcquireNextImageKHR returned {}", vk::ToString(result));
         break;
     }
+
     scheduler.Wait(resource_ticks[image_index]);
     resource_ticks[image_index] = scheduler.CurrentTick();
+
+    return is_suboptimal || is_outdated;
 }
 
 void Swapchain::Present(VkSemaphore render_semaphore) {
@@ -143,6 +144,7 @@ void Swapchain::Present(VkSemaphore render_semaphore) {
         .pImageIndices = &image_index,
         .pResults = nullptr,
     };
+    std::scoped_lock lock{scheduler.submit_mutex};
     switch (const VkResult result = present_queue.Present(present_info)) {
     case VK_SUCCESS:
         break;
@@ -168,7 +170,7 @@ void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bo
     const auto present_modes{physical_device.GetSurfacePresentModesKHR(surface)};
 
     const VkCompositeAlphaFlagBitsKHR alpha_flags{ChooseAlphaFlags(capabilities)};
-    const VkSurfaceFormatKHR surface_format{ChooseSwapSurfaceFormat(formats)};
+    surface_format = ChooseSwapSurfaceFormat(formats);
     present_mode = ChooseSwapPresentMode(present_modes);
 
     u32 requested_image_count{capabilities.minImageCount + 1};
@@ -193,7 +195,7 @@ void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bo
         .imageColorSpace = surface_format.colorSpace,
         .imageExtent = {},
         .imageArrayLayers = 1,
-        .imageUsage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
+        .imageUsage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT,
         .imageSharingMode = VK_SHARING_MODE_EXCLUSIVE,
         .queueFamilyIndexCount = 0,
         .pQueueFamilyIndices = nullptr,
@@ -241,45 +243,14 @@ void Swapchain::CreateSemaphores() {
     present_semaphores.resize(image_count);
     std::ranges::generate(present_semaphores,
                           [this] { return device.GetLogical().CreateSemaphore(); });
-}
-
-void Swapchain::CreateImageViews() {
-    VkImageViewCreateInfo ci{
-        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .image = {},
-        .viewType = VK_IMAGE_VIEW_TYPE_2D,
-        .format = image_view_format,
-        .components =
-            {
-                .r = VK_COMPONENT_SWIZZLE_IDENTITY,
-                .g = VK_COMPONENT_SWIZZLE_IDENTITY,
-                .b = VK_COMPONENT_SWIZZLE_IDENTITY,
-                .a = VK_COMPONENT_SWIZZLE_IDENTITY,
-            },
-        .subresourceRange =
-            {
-                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
-                .baseMipLevel = 0,
-                .levelCount = 1,
-                .baseArrayLayer = 0,
-                .layerCount = 1,
-            },
-    };
-
-    image_views.resize(image_count);
-    for (std::size_t i = 0; i < image_count; i++) {
-        ci.image = images[i];
-        image_views[i] = device.GetLogical().CreateImageView(ci);
-    }
+    render_semaphores.resize(image_count);
+    std::ranges::generate(render_semaphores,
+                          [this] { return device.GetLogical().CreateSemaphore(); });
 }
 
 void Swapchain::Destroy() {
     frame_index = 0;
     present_semaphores.clear();
-    framebuffers.clear();
-    image_views.clear();
     swapchain.reset();
 }
 

src/video_core/renderer_vulkan/vk_swapchain.h

@@ -27,7 +27,7 @@ public:
     void Create(u32 width, u32 height, bool srgb);
 
     /// Acquires the next image in the swapchain, waits as needed.
-    void AcquireNextImage();
+    bool AcquireNextImage();
 
     /// Presents the rendered image to the swapchain.
     void Present(VkSemaphore render_semaphore);
@@ -52,6 +52,11 @@ 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;
     }
@@ -64,22 +69,34 @@ public:
         return image_index;
     }
 
+    std::size_t GetFrameIndex() const {
+        return frame_index;
+    }
+
     VkImage GetImageIndex(std::size_t index) const {
         return images[index];
     }
 
-    VkImageView GetImageViewIndex(std::size_t index) const {
-        return *image_views[index];
+    VkImage CurrentImage() const {
+        return images[image_index];
     }
 
     VkFormat GetImageViewFormat() const {
         return image_view_format;
     }
 
+    VkFormat GetImageFormat() const {
+        return surface_format.format;
+    }
+
     VkSemaphore CurrentPresentSemaphore() const {
         return *present_semaphores[frame_index];
     }
 
+    VkSemaphore CurrentRenderSemaphore() const {
+        return *render_semaphores[frame_index];
+    }
+
     u32 GetWidth() const {
         return width;
     }
@@ -88,6 +105,10 @@ public:
         return height;
     }
 
+    VkExtent2D GetExtent() const {
+        return extent;
+    }
+
 private:
     void CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bool srgb);
     void CreateSemaphores();
@@ -107,10 +128,9 @@ private:
 
     std::size_t image_count{};
     std::vector<VkImage> images;
-    std::vector<vk::ImageView> image_views;
-    std::vector<vk::Framebuffer> framebuffers;
     std::vector<u64> resource_ticks;
     std::vector<vk::Semaphore> present_semaphores;
+    std::vector<vk::Semaphore> render_semaphores;
 
     u32 width;
     u32 height;
@@ -121,6 +141,7 @@ private:
     VkFormat image_view_format{};
     VkExtent2D extent{};
     VkPresentModeKHR present_mode{};
+    VkSurfaceFormatKHR surface_format{};
 
     bool current_srgb{};
     bool current_fps_unlocked{};

src/video_core/renderer_vulkan/vk_update_descriptor.cpp

@@ -14,13 +14,18 @@ namespace Vulkan {
 
 UpdateDescriptorQueue::UpdateDescriptorQueue(const Device& device_, Scheduler& scheduler_)
     : device{device_}, scheduler{scheduler_} {
+    payload_start = payload.data();
     payload_cursor = payload.data();
 }
 
 UpdateDescriptorQueue::~UpdateDescriptorQueue() = default;
 
 void UpdateDescriptorQueue::TickFrame() {
-    payload_cursor = payload.data();
+    if (++frame_index >= FRAMES_IN_FLIGHT) {
+        frame_index = 0;
+    }
+    payload_start = payload.data() + frame_index * FRAME_PAYLOAD_SIZE;
+    payload_cursor = payload_start;
 }
 
 void UpdateDescriptorQueue::Acquire() {
@@ -28,10 +33,10 @@ void UpdateDescriptorQueue::Acquire() {
     // This is the maximum number of entries a single draw call might use.
     static constexpr size_t MIN_ENTRIES = 0x400;
 
-    if (std::distance(payload.data(), payload_cursor) + MIN_ENTRIES >= payload.max_size()) {
+    if (std::distance(payload_start, payload_cursor) + MIN_ENTRIES >= FRAME_PAYLOAD_SIZE) {
         LOG_WARNING(Render_Vulkan, "Payload overflow, waiting for worker thread");
         scheduler.WaitWorker();
-        payload_cursor = payload.data();
+        payload_cursor = payload_start;
     }
     upload_start = payload_cursor;
 }

src/video_core/renderer_vulkan/vk_update_descriptor.h

@@ -29,6 +29,12 @@ struct DescriptorUpdateEntry {
 };
 
 class UpdateDescriptorQueue final {
+    // This should be plenty for the vast majority of cases. Most desktop platforms only
+    // provide up to 3 swapchain images.
+    static constexpr size_t FRAMES_IN_FLIGHT = 5;
+    static constexpr size_t FRAME_PAYLOAD_SIZE = 0x10000;
+    static constexpr size_t PAYLOAD_SIZE = FRAME_PAYLOAD_SIZE * FRAMES_IN_FLIGHT;
+
 public:
     explicit UpdateDescriptorQueue(const Device& device_, Scheduler& scheduler_);
     ~UpdateDescriptorQueue();
@@ -73,9 +79,11 @@ private:
     const Device& device;
     Scheduler& scheduler;
 
+    size_t frame_index{0};
     DescriptorUpdateEntry* payload_cursor = nullptr;
+    DescriptorUpdateEntry* payload_start = nullptr;
     const DescriptorUpdateEntry* upload_start = nullptr;
-    std::array<DescriptorUpdateEntry, 0x10000> payload;
+    std::array<DescriptorUpdateEntry, PAYLOAD_SIZE> payload;
 };
 
 } // namespace Vulkan

src/yuzu/configuration/config.cpp

@@ -692,6 +692,7 @@ void Config::ReadRendererValues() {
     qt_config->beginGroup(QStringLiteral("Renderer"));
 
     ReadGlobalSetting(Settings::values.renderer_backend);
+    ReadGlobalSetting(Settings::values.async_presentation);
     ReadGlobalSetting(Settings::values.renderer_force_max_clock);
     ReadGlobalSetting(Settings::values.vulkan_device);
     ReadGlobalSetting(Settings::values.fullscreen_mode);
@@ -1313,6 +1314,7 @@ void Config::SaveRendererValues() {
                  static_cast<u32>(Settings::values.renderer_backend.GetValue(global)),
                  static_cast<u32>(Settings::values.renderer_backend.GetDefault()),
                  Settings::values.renderer_backend.UsingGlobal());
+    WriteGlobalSetting(Settings::values.async_presentation);
     WriteGlobalSetting(Settings::values.renderer_force_max_clock);
     WriteGlobalSetting(Settings::values.vulkan_device);
     WriteSetting(QString::fromStdString(Settings::values.fullscreen_mode.GetLabel()),

src/yuzu/configuration/configure_graphics_advanced.cpp

@@ -22,11 +22,13 @@ ConfigureGraphicsAdvanced::~ConfigureGraphicsAdvanced() = default;
 void ConfigureGraphicsAdvanced::SetConfiguration() {
     const bool runtime_lock = !system.IsPoweredOn();
     ui->use_vsync->setEnabled(runtime_lock);
+    ui->async_present->setEnabled(runtime_lock);
     ui->renderer_force_max_clock->setEnabled(runtime_lock);
     ui->async_astc->setEnabled(runtime_lock);
     ui->use_asynchronous_shaders->setEnabled(runtime_lock);
     ui->anisotropic_filtering_combobox->setEnabled(runtime_lock);
 
+    ui->async_present->setChecked(Settings::values.async_presentation.GetValue());
     ui->renderer_force_max_clock->setChecked(Settings::values.renderer_force_max_clock.GetValue());
     ui->use_vsync->setChecked(Settings::values.use_vsync.GetValue());
     ui->async_astc->setChecked(Settings::values.async_astc.GetValue());
@@ -54,6 +56,8 @@ void ConfigureGraphicsAdvanced::SetConfiguration() {
 
 void ConfigureGraphicsAdvanced::ApplyConfiguration() {
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.gpu_accuracy, ui->gpu_accuracy);
+    ConfigurationShared::ApplyPerGameSetting(&Settings::values.async_presentation,
+                                             ui->async_present, async_present);
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.renderer_force_max_clock,
                                              ui->renderer_force_max_clock,
                                              renderer_force_max_clock);
@@ -90,6 +94,7 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
     // Disable if not global (only happens during game)
     if (Settings::IsConfiguringGlobal()) {
         ui->gpu_accuracy->setEnabled(Settings::values.gpu_accuracy.UsingGlobal());
+        ui->async_present->setEnabled(Settings::values.async_presentation.UsingGlobal());
         ui->renderer_force_max_clock->setEnabled(
             Settings::values.renderer_force_max_clock.UsingGlobal());
         ui->use_vsync->setEnabled(Settings::values.use_vsync.UsingGlobal());
@@ -107,6 +112,8 @@ void ConfigureGraphicsAdvanced::SetupPerGameUI() {
         return;
     }
 
+    ConfigurationShared::SetColoredTristate(ui->async_present, Settings::values.async_presentation,
+                                            async_present);
     ConfigurationShared::SetColoredTristate(ui->renderer_force_max_clock,
                                             Settings::values.renderer_force_max_clock,
                                             renderer_force_max_clock);

src/yuzu/configuration/configure_graphics_advanced.h

@@ -36,6 +36,7 @@ private:
 
     std::unique_ptr<Ui::ConfigureGraphicsAdvanced> ui;
 
+    ConfigurationShared::CheckState async_present;
     ConfigurationShared::CheckState renderer_force_max_clock;
     ConfigurationShared::CheckState use_vsync;
     ConfigurationShared::CheckState async_astc;

src/yuzu/configuration/configure_graphics_advanced.ui

@@ -7,7 +7,7 @@
     <x>0</x>
     <y>0</y>
     <width>404</width>
-    <height>321</height>
+    <height>376</height>
    </rect>
   </property>
   <property name="windowTitle">
@@ -69,6 +69,13 @@
           </layout>
          </widget>
         </item>
+        <item>
+         <widget class="QCheckBox" name="async_present">
+          <property name="text">
+           <string>Enable asynchronous presentation (Vulkan only)</string>
+          </property>
+         </widget>
+        </item>
         <item>
          <widget class="QCheckBox" name="renderer_force_max_clock">
           <property name="toolTip">
@@ -112,7 +119,7 @@
         <item>
          <widget class="QCheckBox" name="use_fast_gpu_time">
           <property name="toolTip">
-            <string>Enables Fast GPU Time. This option will force most games to run at their highest native resolution.</string>
+           <string>Enables Fast GPU Time. This option will force most games to run at their highest native resolution.</string>
           </property>
           <property name="text">
            <string>Use Fast GPU Time (Hack)</string>
@@ -122,7 +129,7 @@
         <item>
          <widget class="QCheckBox" name="use_pessimistic_flushes">
           <property name="toolTip">
-            <string>Enables pessimistic buffer flushes. This option will force unmodified buffers to be flushed, which can cost performance.</string>
+           <string>Enables pessimistic buffer flushes. This option will force unmodified buffers to be flushed, which can cost performance.</string>
           </property>
           <property name="text">
            <string>Use pessimistic buffer flushes (Hack)</string>
@@ -132,7 +139,7 @@
         <item>
          <widget class="QCheckBox" name="use_vulkan_driver_pipeline_cache">
           <property name="toolTip">
-            <string>Enables GPU vendor-specific pipeline cache. This option can improve shader loading time significantly in cases where the Vulkan driver does not store pipeline cache files internally.</string>
+           <string>Enables GPU vendor-specific pipeline cache. This option can improve shader loading time significantly in cases where the Vulkan driver does not store pipeline cache files internally.</string>
           </property>
           <property name="text">
            <string>Use Vulkan pipeline cache</string>

src/yuzu_cmd/config.cpp

@@ -300,6 +300,7 @@ void Config::ReadValues() {
 
     // Renderer
     ReadSetting("Renderer", Settings::values.renderer_backend);
+    ReadSetting("Renderer", Settings::values.async_presentation);
     ReadSetting("Renderer", Settings::values.renderer_force_max_clock);
     ReadSetting("Renderer", Settings::values.renderer_debug);
     ReadSetting("Renderer", Settings::values.renderer_shader_feedback);

src/yuzu_cmd/default_ini.h

@@ -264,6 +264,10 @@ cpuopt_unsafe_ignore_global_monitor =
 # 0: OpenGL, 1 (default): Vulkan
 backend =
 
+# Whether to enable asynchronous presentation (Vulkan only)
+# 0 (default): Off, 1: On
+async_presentation =
+
 # Enable graphics API debugging mode.
 # 0 (default): Disabled, 1: Enabled
 debug =