nvnflinger/gpu: implement layer stack composition

16 files changed, 140 insertions, 198 deletions

diff --git a/src/video_core/framebuffer_config.h b/src/video_core/framebuffer_config.h
index 10ddc75a7..6a18b76fb 100644
--- a/src/video_core/framebuffer_config.h
+++ b/src/video_core/framebuffer_config.h
@@ -7,6 +7,7 @@
 #include "common/math_util.h"
 #include "core/hle/service/nvnflinger/buffer_transform_flags.h"
 #include "core/hle/service/nvnflinger/pixel_format.h"
+#include "core/hle/service/nvnflinger/ui/fence.h"
 
 namespace Tegra {
 
@@ -21,7 +22,7 @@ struct FramebufferConfig {
     u32 stride{};
     Service::android::PixelFormat pixel_format{};
     Service::android::BufferTransformFlags transform_flags{};
-    Common::Rectangle<int> crop_rect;
+    Common::Rectangle<int> crop_rect{};
 };
 
 Common::Rectangle<f32> NormalizeCrop(const FramebufferConfig& framebuffer, u32 texture_width,
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 609704b33..f4a5d831c 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -274,11 +274,6 @@ struct GPU::Impl {
         }
     }
 
-    /// Swap buffers (render frame)
-    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
-        gpu_thread.SwapBuffers(framebuffer);
-    }
-
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     void FlushRegion(DAddr addr, u64 size) {
         gpu_thread.FlushRegion(addr, size);
@@ -313,8 +308,9 @@ struct GPU::Impl {
         gpu_thread.FlushAndInvalidateRegion(addr, size);
     }
 
-    void RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer,
-                            std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences) {
+    void RequestComposite(std::vector<Tegra::FramebufferConfig>&& layers,
+                          std::vector<Service::Nvidia::NvFence>&& fences) {
+        size_t num_fences{fences.size()};
         size_t current_request_counter{};
         {
             std::unique_lock<std::mutex> lk(request_swap_mutex);
@@ -328,13 +324,12 @@ struct GPU::Impl {
             }
         }
         const auto wait_fence =
-            RequestSyncOperation([this, current_request_counter, framebuffer, fences, num_fences] {
+            RequestSyncOperation([this, current_request_counter, &layers, &fences, num_fences] {
                 auto& syncpoint_manager = host1x.GetSyncpointManager();
                 if (num_fences == 0) {
-                    renderer->SwapBuffers(framebuffer);
+                    renderer->Composite(layers);
                 }
-                const auto executer = [this, current_request_counter,
-                                       framebuffer_copy = *framebuffer]() {
+                const auto executer = [this, current_request_counter, layers_copy = layers]() {
                     {
                         std::unique_lock<std::mutex> lk(request_swap_mutex);
                         if (--request_swap_counters[current_request_counter] != 0) {
@@ -342,7 +337,7 @@ struct GPU::Impl {
                         }
                         free_swap_counters.push_back(current_request_counter);
                     }
-                    renderer->SwapBuffers(&framebuffer_copy);
+                    renderer->Composite(layers_copy);
                 };
                 for (size_t i = 0; i < num_fences; i++) {
                     syncpoint_manager.RegisterGuestAction(fences[i].id, fences[i].value, executer);
@@ -505,9 +500,9 @@ const VideoCore::ShaderNotify& GPU::ShaderNotify() const {
     return impl->ShaderNotify();
 }
 
-void GPU::RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer,
-                             std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences) {
-    impl->RequestSwapBuffers(framebuffer, fences, num_fences);
+void GPU::RequestComposite(std::vector<Tegra::FramebufferConfig>&& layers,
+                           std::vector<Service::Nvidia::NvFence>&& fences) {
+    impl->RequestComposite(std::move(layers), std::move(fences));
 }
 
 u64 GPU::GetTicks() const {
@@ -554,10 +549,6 @@ void GPU::ClearCdmaInstance(u32 id) {
     impl->ClearCdmaInstance(id);
 }
 
-void GPU::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
-    impl->SwapBuffers(framebuffer);
-}
-
 VideoCore::RasterizerDownloadArea GPU::OnCPURead(PAddr addr, u64 size) {
     return impl->OnCPURead(addr, size);
 }
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index b3c1d15bd..c4602ca37 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -212,8 +212,8 @@ public:
 
     void RendererFrameEndNotify();
 
-    void RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer,
-                            std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences);
+    void RequestComposite(std::vector<Tegra::FramebufferConfig>&& layers,
+                          std::vector<Service::Nvidia::NvFence>&& fences);
 
     /// Performs any additional setup necessary in order to begin GPU emulation.
     /// This can be used to launch any necessary threads and register any necessary
diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index 788d4f61e..58d8110b8 100644
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -40,8 +40,6 @@ static void RunThread(std::stop_token stop_token, Core::System& system,
         }
         if (auto* submit_list = std::get_if<SubmitListCommand>(&next.data)) {
             scheduler.Push(submit_list->channel, std::move(submit_list->entries));
-        } else if (const auto* data = std::get_if<SwapBuffersCommand>(&next.data)) {
-            renderer.SwapBuffers(data->framebuffer ? &*data->framebuffer : nullptr);
         } else if (std::holds_alternative<GPUTickCommand>(next.data)) {
             system.GPU().TickWork();
         } else if (const auto* flush = std::get_if<FlushRegionCommand>(&next.data)) {
@@ -78,10 +76,6 @@ void ThreadManager::SubmitList(s32 channel, Tegra::CommandList&& entries) {
     PushCommand(SubmitListCommand(channel, std::move(entries)));
 }
 
-void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
-    PushCommand(SwapBuffersCommand(framebuffer ? std::make_optional(*framebuffer) : std::nullopt));
-}
-
 void ThreadManager::FlushRegion(DAddr addr, u64 size) {
     if (!is_async) {
         // Always flush with synchronous GPU mode
diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h
index 2de25e9ef..dc0fce9f8 100644
--- a/src/video_core/gpu_thread.h
+++ b/src/video_core/gpu_thread.h
@@ -44,14 +44,6 @@ struct SubmitListCommand final {
     Tegra::CommandList entries;
 };
 
-/// Command to signal to the GPU thread that a swap buffers is pending
-struct SwapBuffersCommand final {
-    explicit SwapBuffersCommand(std::optional<const Tegra::FramebufferConfig> framebuffer_)
-        : framebuffer{std::move(framebuffer_)} {}
-
-    std::optional<Tegra::FramebufferConfig> framebuffer;
-};
-
 /// Command to signal to the GPU thread to flush a region
 struct FlushRegionCommand final {
     explicit constexpr FlushRegionCommand(DAddr addr_, u64 size_) : addr{addr_}, size{size_} {}
@@ -81,8 +73,8 @@ struct FlushAndInvalidateRegionCommand final {
 struct GPUTickCommand final {};
 
 using CommandData =
-    std::variant<std::monostate, SubmitListCommand, SwapBuffersCommand, FlushRegionCommand,
-                 InvalidateRegionCommand, FlushAndInvalidateRegionCommand, GPUTickCommand>;
+    std::variant<std::monostate, SubmitListCommand, FlushRegionCommand, InvalidateRegionCommand,
+                 FlushAndInvalidateRegionCommand, GPUTickCommand>;
 
 struct CommandDataContainer {
     CommandDataContainer() = default;
@@ -118,9 +110,6 @@ public:
     /// Push GPU command entries to be processed
     void SubmitList(s32 channel, Tegra::CommandList&& entries);
 
-    /// Swap buffers (render frame)
-    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer);
-
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     void FlushRegion(DAddr addr, u64 size);
 
diff --git a/src/video_core/renderer_base.h b/src/video_core/renderer_base.h
index 78ea5208b..3ad180f67 100644
--- a/src/video_core/renderer_base.h
+++ b/src/video_core/renderer_base.h
@@ -38,7 +38,7 @@ public:
     virtual ~RendererBase();
 
     /// Finalize rendering the guest frame and draw into the presentation texture
-    virtual void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) = 0;
+    virtual void Composite(std::span<const Tegra::FramebufferConfig> layers) = 0;
 
     [[nodiscard]] virtual RasterizerInterface* ReadRasterizer() = 0;
 
diff --git a/src/video_core/renderer_null/renderer_null.cpp b/src/video_core/renderer_null/renderer_null.cpp
index 078feb925..c89daff53 100644
--- a/src/video_core/renderer_null/renderer_null.cpp
+++ b/src/video_core/renderer_null/renderer_null.cpp
@@ -13,8 +13,8 @@ RendererNull::RendererNull(Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gp
 
 RendererNull::~RendererNull() = default;
 
-void RendererNull::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
-    if (!framebuffer) {
+void RendererNull::Composite(std::span<const Tegra::FramebufferConfig> framebuffers) {
+    if (framebuffers.empty()) {
         return;
     }
 
diff --git a/src/video_core/renderer_null/renderer_null.h b/src/video_core/renderer_null/renderer_null.h
index 9531b43f6..063b476bb 100644
--- a/src/video_core/renderer_null/renderer_null.h
+++ b/src/video_core/renderer_null/renderer_null.h
@@ -17,7 +17,7 @@ public:
                           std::unique_ptr<Core::Frontend::GraphicsContext> context);
     ~RendererNull() override;
 
-    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
+    void Composite(std::span<const Tegra::FramebufferConfig> framebuffer) override;
 
     VideoCore::RasterizerInterface* ReadRasterizer() override {
         return &m_rasterizer;
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index 10a9f973c..e33a32592 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -125,15 +125,15 @@ RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_,
 
 RendererOpenGL::~RendererOpenGL() = default;
 
-void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
-    if (!framebuffer) {
+void RendererOpenGL::Composite(std::span<const Tegra::FramebufferConfig> framebuffers) {
+    if (framebuffers.empty()) {
         return;
     }
 
-    RenderScreenshot(framebuffer);
+    RenderScreenshot(framebuffers);
 
     state_tracker.BindFramebuffer(0);
-    blit_screen->DrawScreen(std::span(framebuffer, 1), emu_window.GetFramebufferLayout());
+    blit_screen->DrawScreen(framebuffers, emu_window.GetFramebufferLayout());
 
     ++m_current_frame;
 
@@ -159,7 +159,7 @@ void RendererOpenGL::AddTelemetryFields() {
     telemetry_session.AddField(user_system, "GPU_OpenGL_Version", std::string(gl_version));
 }
 
-void RendererOpenGL::RenderScreenshot(const Tegra::FramebufferConfig* framebuffer) {
+void RendererOpenGL::RenderScreenshot(std::span<const Tegra::FramebufferConfig> framebuffers) {
     if (!renderer_settings.screenshot_requested) {
         return;
     }
@@ -181,7 +181,7 @@ void RendererOpenGL::RenderScreenshot(const Tegra::FramebufferConfig* framebuffe
     glRenderbufferStorage(GL_RENDERBUFFER, GL_SRGB8, layout.width, layout.height);
     glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, renderbuffer);
 
-    blit_screen->DrawScreen(std::span(framebuffer, 1), layout);
+    blit_screen->DrawScreen(framebuffers, layout);
 
     glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
     glPixelStorei(GL_PACK_ROW_LENGTH, 0);
diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h
index df76d3d05..c4625c96e 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.h
+++ b/src/video_core/renderer_opengl/renderer_opengl.h
@@ -40,7 +40,7 @@ public:
                             std::unique_ptr<Core::Frontend::GraphicsContext> context_);
     ~RendererOpenGL() override;
 
-    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
+    void Composite(std::span<const Tegra::FramebufferConfig> framebuffers) override;
 
     VideoCore::RasterizerInterface* ReadRasterizer() override {
         return &rasterizer;
@@ -52,7 +52,7 @@ public:
 
 private:
     void AddTelemetryFields();
-    void RenderScreenshot(const Tegra::FramebufferConfig* framebuffer);
+    void RenderScreenshot(std::span<const Tegra::FramebufferConfig> framebuffers);
 
     Core::TelemetrySession& telemetry_session;
     Core::Frontend::EmuWindow& emu_window;
diff --git a/src/video_core/renderer_vulkan/present/util.cpp b/src/video_core/renderer_vulkan/present/util.cpp
index 7bff1c436..6ee16595d 100644
--- a/src/video_core/renderer_vulkan/present/util.cpp
+++ b/src/video_core/renderer_vulkan/present/util.cpp
@@ -7,6 +7,20 @@
 
 namespace Vulkan {
 
+vk::Buffer CreateWrappedBuffer(MemoryAllocator& allocator, VkDeviceSize size, MemoryUsage usage) {
+    const VkBufferCreateInfo dst_buffer_info{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .size = size,
+        .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+    };
+    return allocator.CreateBuffer(dst_buffer_info, usage);
+}
+
 vk::Image CreateWrappedImage(MemoryAllocator& allocator, VkExtent2D dimensions, VkFormat format) {
     const VkImageCreateInfo image_ci{
         .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
@@ -96,6 +110,70 @@ void UploadImage(const Device& device, MemoryAllocator& allocator, Scheduler& sc
     scheduler.Finish();
 }
 
+void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffer,
+                        VkExtent3D extent) {
+    const VkImageMemoryBarrier read_barrier{
+        .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+        .pNext = nullptr,
+        .srcAccessMask = VK_ACCESS_MEMORY_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 = image,
+        .subresourceRange{
+            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            .baseMipLevel = 0,
+            .levelCount = VK_REMAINING_MIP_LEVELS,
+            .baseArrayLayer = 0,
+            .layerCount = VK_REMAINING_ARRAY_LAYERS,
+        },
+    };
+    const VkImageMemoryBarrier image_write_barrier{
+        .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+        .pNext = nullptr,
+        .srcAccessMask = 0,
+        .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 = image,
+        .subresourceRange{
+            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            .baseMipLevel = 0,
+            .levelCount = VK_REMAINING_MIP_LEVELS,
+            .baseArrayLayer = 0,
+            .layerCount = VK_REMAINING_ARRAY_LAYERS,
+        },
+    };
+    static constexpr VkMemoryBarrier memory_write_barrier{
+        .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+        .pNext = nullptr,
+        .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+        .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+    };
+    const VkBufferImageCopy copy{
+        .bufferOffset = 0,
+        .bufferRowLength = 0,
+        .bufferImageHeight = 0,
+        .imageSubresource{
+            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            .mipLevel = 0,
+            .baseArrayLayer = 0,
+            .layerCount = 1,
+        },
+        .imageOffset{.x = 0, .y = 0, .z = 0},
+        .imageExtent{extent},
+    };
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
+                           read_barrier);
+    cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffer, copy);
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
+                           memory_write_barrier, nullptr, image_write_barrier);
+}
+
 vk::ImageView CreateWrappedImageView(const Device& device, vk::Image& image, VkFormat format) {
     return device.GetLogical().CreateImageView(VkImageViewCreateInfo{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
diff --git a/src/video_core/renderer_vulkan/present/util.h b/src/video_core/renderer_vulkan/present/util.h
index fb4e4a8e4..1104aaa15 100644
--- a/src/video_core/renderer_vulkan/present/util.h
+++ b/src/video_core/renderer_vulkan/present/util.h
@@ -11,12 +11,16 @@ namespace Vulkan {
 
 #define ARRAY_TO_SPAN(a) std::span(a, (sizeof(a) / sizeof(a[0])))
 
+vk::Buffer CreateWrappedBuffer(MemoryAllocator& allocator, VkDeviceSize size, MemoryUsage usage);
+
 vk::Image CreateWrappedImage(MemoryAllocator& allocator, VkExtent2D dimensions, VkFormat format);
 void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayout target_layout,
                            VkImageLayout source_layout = VK_IMAGE_LAYOUT_GENERAL);
 void UploadImage(const Device& device, MemoryAllocator& allocator, Scheduler& scheduler,
                  vk::Image& image, VkExtent2D dimensions, VkFormat format,
                  std::span<const u8> initial_contents = {});
+void DownloadColorImage(vk::CommandBuffer& cmdbuf, VkImage image, VkBuffer buffer,
+                        VkExtent3D extent);
 void ClearColorImage(vk::CommandBuffer& cmdbuf, VkImage image);
 
 vk::ImageView CreateWrappedImageView(const Device& device, vk::Image& image, VkFormat format);
diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
index 77837adde..48a105327 100644
--- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp
+++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
@@ -20,12 +20,14 @@
 #include "core/frontend/graphics_context.h"
 #include "core/telemetry_session.h"
 #include "video_core/gpu.h"
+#include "video_core/renderer_vulkan/present/util.h"
 #include "video_core/renderer_vulkan/renderer_vulkan.h"
 #include "video_core/renderer_vulkan/vk_blit_screen.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"
 #include "video_core/renderer_vulkan/vk_swapchain.h"
+#include "video_core/textures/decoders.h"
 #include "video_core/vulkan_common/vulkan_debug_callback.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 #include "video_core/vulkan_common/vulkan_instance.h"
@@ -116,18 +118,20 @@ RendererVulkan::~RendererVulkan() {
     void(device.GetLogical().WaitIdle());
 }
 
-void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
-    if (!framebuffer) {
+void RendererVulkan::Composite(std::span<const Tegra::FramebufferConfig> framebuffers) {
+    if (framebuffers.empty()) {
         return;
     }
+
     SCOPE_EXIT({ render_window.OnFrameDisplayed(); });
+
     if (!render_window.IsShown()) {
         return;
     }
 
-    RenderScreenshot(framebuffer);
+    RenderScreenshot(framebuffers);
     Frame* frame = present_manager.GetRenderFrame();
-    blit_swapchain.DrawToFrame(rasterizer, frame, std::span(framebuffer, 1),
+    blit_swapchain.DrawToFrame(rasterizer, frame, framebuffers,
                                render_window.GetFramebufferLayout(), swapchain.GetImageCount(),
                                swapchain.GetImageViewFormat());
     scheduler.Flush(*frame->render_ready);
@@ -163,156 +167,37 @@ void RendererVulkan::Report() const {
     telemetry_session.AddField(field, "GPU_Vulkan_Extensions", extensions);
 }
 
-void Vulkan::RendererVulkan::RenderScreenshot(const Tegra::FramebufferConfig* framebuffer) {
+void Vulkan::RendererVulkan::RenderScreenshot(
+    std::span<const Tegra::FramebufferConfig> framebuffers) {
     if (!renderer_settings.screenshot_requested) {
         return;
     }
+
+    constexpr VkFormat ScreenshotFormat{VK_FORMAT_B8G8R8A8_UNORM};
     const Layout::FramebufferLayout layout{renderer_settings.screenshot_framebuffer_layout};
-    auto frame = [&]() {
-        vk::Image staging_image = memory_allocator.CreateImage(VkImageCreateInfo{
-            .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
-            .pNext = nullptr,
-            .flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT,
-            .imageType = VK_IMAGE_TYPE_2D,
-            .format = VK_FORMAT_B8G8R8A8_UNORM,
-            .extent =
-                {
-                    .width = layout.width,
-                    .height = layout.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_TRANSFER_DST_BIT |
-                     VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
-            .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
-            .queueFamilyIndexCount = 0,
-            .pQueueFamilyIndices = nullptr,
-            .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
-        });
 
-        vk::ImageView dst_view = device.GetLogical().CreateImageView(VkImageViewCreateInfo{
-            .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
-            .pNext = nullptr,
-            .flags = 0,
-            .image = *staging_image,
-            .viewType = VK_IMAGE_VIEW_TYPE_2D,
-            .format = VK_FORMAT_B8G8R8A8_UNORM,
-            .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 = VK_REMAINING_ARRAY_LAYERS,
-            },
-        });
-        vk::Framebuffer screenshot_fb =
-            blit_screenshot.CreateFramebuffer(layout, *dst_view, VK_FORMAT_B8G8R8A8_UNORM);
-        return Frame{
-            .width = layout.width,
-            .height = layout.height,
-            .image = std::move(staging_image),
-            .image_view = std::move(dst_view),
-            .framebuffer = std::move(screenshot_fb),
-            .cmdbuf{},
-            .render_ready{},
-            .present_done{},
-        };
+    auto frame = [&]() {
+        Frame f{};
+        f.image = CreateWrappedImage(memory_allocator, VkExtent2D{layout.width, layout.height},
+                                     ScreenshotFormat);
+        f.image_view = CreateWrappedImageView(device, f.image, ScreenshotFormat);
+        f.framebuffer = blit_screenshot.CreateFramebuffer(layout, *f.image_view, ScreenshotFormat);
+        return f;
     }();
 
-    blit_screenshot.DrawToFrame(rasterizer, &frame, std::span(framebuffer, 1), layout, 1,
+    blit_screenshot.DrawToFrame(rasterizer, &frame, framebuffers, layout, 1,
                                 VK_FORMAT_B8G8R8A8_UNORM);
 
-    const auto buffer_size = static_cast<VkDeviceSize>(layout.width * layout.height * 4);
-    const VkBufferCreateInfo dst_buffer_info{
-        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .size = buffer_size,
-        .usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT,
-        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
-        .queueFamilyIndexCount = 0,
-        .pQueueFamilyIndices = nullptr,
-    };
-    const vk::Buffer dst_buffer =
-        memory_allocator.CreateBuffer(dst_buffer_info, MemoryUsage::Download);
+    const auto dst_buffer = CreateWrappedBuffer(
+        memory_allocator, static_cast<VkDeviceSize>(layout.width * layout.height * 4),
+        MemoryUsage::Download);
 
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([&](vk::CommandBuffer cmdbuf) {
-        const VkImageMemoryBarrier read_barrier{
-            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = VK_ACCESS_MEMORY_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 = VK_REMAINING_MIP_LEVELS,
-                .baseArrayLayer = 0,
-                .layerCount = VK_REMAINING_ARRAY_LAYERS,
-            },
-        };
-        const VkImageMemoryBarrier image_write_barrier{
-            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = 0,
-            .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 = VK_REMAINING_MIP_LEVELS,
-                .baseArrayLayer = 0,
-                .layerCount = VK_REMAINING_ARRAY_LAYERS,
-            },
-        };
-        static constexpr VkMemoryBarrier memory_write_barrier{
-            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
-        };
-        const VkBufferImageCopy copy{
-            .bufferOffset = 0,
-            .bufferRowLength = 0,
-            .bufferImageHeight = 0,
-            .imageSubresource{
-                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
-                .mipLevel = 0,
-                .baseArrayLayer = 0,
-                .layerCount = 1,
-            },
-            .imageOffset{.x = 0, .y = 0, .z = 0},
-            .imageExtent{
-                .width = layout.width,
-                .height = layout.height,
-                .depth = 1,
-            },
-        };
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
-                               0, read_barrier);
-        cmdbuf.CopyImageToBuffer(*frame.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *dst_buffer,
-                                 copy);
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
-                               0, memory_write_barrier, nullptr, image_write_barrier);
+        DownloadColorImage(cmdbuf, *frame.image, *dst_buffer,
+                           VkExtent3D{layout.width, layout.height, 1});
     });
+
     // Ensure the copy is fully completed before saving the screenshot
     scheduler.Finish();
 
diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.h b/src/video_core/renderer_vulkan/renderer_vulkan.h
index bdeb43a54..c6d8a0f21 100644
--- a/src/video_core/renderer_vulkan/renderer_vulkan.h
+++ b/src/video_core/renderer_vulkan/renderer_vulkan.h
@@ -46,7 +46,7 @@ public:
                             std::unique_ptr<Core::Frontend::GraphicsContext> context_);
     ~RendererVulkan() override;
 
-    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
+    void Composite(std::span<const Tegra::FramebufferConfig> framebuffers) override;
 
     VideoCore::RasterizerInterface* ReadRasterizer() override {
         return &rasterizer;
@@ -59,7 +59,7 @@ public:
 private:
     void Report() const;
 
-    void RenderScreenshot(const Tegra::FramebufferConfig* framebuffer);
+    void RenderScreenshot(std::span<const Tegra::FramebufferConfig> framebuffers);
 
     Core::TelemetrySession& telemetry_session;
     Tegra::MaxwellDeviceMemoryManager& device_memory;
diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
index b2dcbf80b..2275fcc46 100644
--- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp
+++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
@@ -115,7 +115,7 @@ void BlitScreen::DrawToFrame(RasterizerVulkan& rasterizer, Frame* frame,
 }
 
 vk::Framebuffer BlitScreen::CreateFramebuffer(const Layout::FramebufferLayout& layout,
-                                              const VkImageView& image_view,
+                                              VkImageView image_view,
                                               VkFormat current_view_format) {
     const bool format_updated =
         std::exchange(swapchain_view_format, current_view_format) != current_view_format;
diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.h b/src/video_core/renderer_vulkan/vk_blit_screen.h
index 9a3476c77..cbdf2d5d0 100644
--- a/src/video_core/renderer_vulkan/vk_blit_screen.h
+++ b/src/video_core/renderer_vulkan/vk_blit_screen.h
@@ -56,7 +56,7 @@ public:
                      VkFormat current_swapchain_view_format);
 
     [[nodiscard]] vk::Framebuffer CreateFramebuffer(const Layout::FramebufferLayout& layout,
-                                                    const VkImageView& image_view,
+                                                    VkImageView image_view,
                                                     VkFormat current_view_format);
 
 private: