Render without rendering (#2152)

* presenter: render the game inside a ImGui window

* presenter: render the previous frame to keep the render rendering

* swapchain: fix swapchain image view format not being converted to unorm

* devtools: fix frame graph timing

src/core/debug_state.h

@@ -131,6 +131,8 @@ class DebugStateImpl {
     friend class Core::Devtools::Widget::FrameGraph;
     friend class Core::Devtools::Widget::ShaderList;
 
+    bool showing_debug_menu_bar = false;
+
     std::queue<std::string> debug_message_popup;
 
     std::mutex guest_threads_mutex{};
@@ -153,6 +155,9 @@ class DebugStateImpl {
     std::vector<ShaderDump> shader_dump_list{};
 
 public:
+    float Framerate = 1.0f / 60.0f;
+    float FrameDeltaTime;
+
     void ShowDebugMessage(std::string message) {
         if (message.empty()) {
             return;
@@ -160,6 +165,10 @@ public:
         debug_message_popup.push(std::move(message));
     }
 
+    bool& ShowingDebugMenuBar() {
+        return showing_debug_menu_bar;
+    }
+
     void AddCurrentThreadToGuestList();
 
     void RemoveCurrentThreadFromGuestList();

src/core/devtools/layer.cpp

@@ -28,7 +28,6 @@ static bool show_simple_fps = false;
 static bool visibility_toggled = false;
 
 static float fps_scale = 1.0f;
-static bool show_advanced_debug = false;
 static int dump_frame_count = 1;
 
 static Widget::FrameGraph frame_graph;
@@ -253,8 +252,8 @@ void L::DrawAdvanced() {
 }
 
 void L::DrawSimple() {
-    const auto io = GetIO();
-    Text("%.1f FPS (%.2f ms)", io.Framerate, 1000.0f / io.Framerate);
+    const float frameRate = DebugState.Framerate;
+    Text("%d FPS (%.1f ms)", static_cast<int>(std::round(1.0f / frameRate)), frameRate * 1000.0f);
 }
 
 static void LoadSettings(const char* line) {
@@ -265,7 +264,7 @@ static void LoadSettings(const char* line) {
         return;
     }
     if (sscanf(line, "show_advanced_debug=%d", &i) == 1) {
-        show_advanced_debug = i != 0;
+        DebugState.ShowingDebugMenuBar() = i != 0;
         return;
     }
     if (sscanf(line, "show_frame_graph=%d", &i) == 1) {
@@ -310,7 +309,7 @@ void L::SetupSettings() {
     handler.WriteAllFn = [](ImGuiContext*, ImGuiSettingsHandler* handler, ImGuiTextBuffer* buf) {
         buf->appendf("[%s][Data]\n", handler->TypeName);
         buf->appendf("fps_scale=%f\n", fps_scale);
-        buf->appendf("show_advanced_debug=%d\n", show_advanced_debug);
+        buf->appendf("show_advanced_debug=%d\n", DebugState.ShowingDebugMenuBar());
         buf->appendf("show_frame_graph=%d\n", frame_graph.is_open);
         buf->appendf("dump_frame_count=%d\n", dump_frame_count);
         buf->append("\n");
@@ -336,12 +335,12 @@ void L::Draw() {
 
     if (!DebugState.IsGuestThreadsPaused()) {
         const auto fn = DebugState.flip_frame_count.load();
-        frame_graph.AddFrame(fn, io.DeltaTime);
+        frame_graph.AddFrame(fn, DebugState.FrameDeltaTime);
     }
 
     if (IsKeyPressed(ImGuiKey_F10, false)) {
         if (io.KeyCtrl) {
-            show_advanced_debug = !show_advanced_debug;
+            DebugState.ShowingDebugMenuBar() ^= true;
         } else {
             show_simple_fps = !show_simple_fps;
         }
@@ -376,7 +375,7 @@ void L::Draw() {
         End();
     }
 
-    if (show_advanced_debug) {
+    if (DebugState.ShowingDebugMenuBar()) {
         PushFont(io.Fonts->Fonts[IMGUI_FONT_MONO]);
         PushID("DevtoolsLayer");
         DrawAdvanced();

src/core/devtools/widget/frame_graph.cpp

@@ -83,15 +83,13 @@ void FrameGraph::Draw() {
 
         auto isSystemPaused = DebugState.IsGuestThreadsPaused();
 
-        static float deltaTime;
-        static float frameRate;
-
         if (!isSystemPaused) {
-            deltaTime = io.DeltaTime * 1000.0f;
+            deltaTime = DebugState.FrameDeltaTime * 1000.0f;
             frameRate = 1000.0f / deltaTime;
         }
 
         Text("Frame time: %.3f ms (%.1f FPS)", deltaTime, frameRate);
+        Text("Presenter time: %.3f ms (%.1f FPS)", io.DeltaTime * 1000.0f, 1.0f / io.DeltaTime);
         Text("Flip frame: %d Gnm submit frame: %d", DebugState.flip_frame_count.load(),
              DebugState.gnm_frame_count.load());
 

src/core/devtools/widget/frame_graph.h

@@ -16,6 +16,9 @@ class FrameGraph {
 
     std::array<FrameInfo, FRAME_BUFFER_SIZE> frame_list{};
 
+    float deltaTime{};
+    float frameRate{};
+
     void DrawFrameGraph();
 
 public:

src/core/libraries/videoout/driver.cpp

@@ -1,8 +1,6 @@
 // SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <imgui.h>
-
 #include "common/assert.h"
 #include "common/config.h"
 #include "common/debug.h"
@@ -207,6 +205,13 @@ void VideoOutDriver::DrawBlankFrame() {
     presenter->Present(empty_frame);
 }
 
+void VideoOutDriver::DrawLastFrame() {
+    const auto frame = presenter->PrepareLastFrame();
+    if (frame != nullptr) {
+        presenter->Present(frame, true);
+    }
+}
+
 bool VideoOutDriver::SubmitFlip(VideoOutPort* port, s32 index, s64 flip_arg,
                                 bool is_eop /*= false*/) {
     {
@@ -278,17 +283,24 @@ void VideoOutDriver::PresentThread(std::stop_token token) {
         return {};
     };
 
-    auto delay = std::chrono::microseconds{0};
     while (!token.stop_requested()) {
         timer.Start();
 
+        if (DebugState.IsGuestThreadsPaused()) {
+            DrawLastFrame();
+            timer.End();
+            continue;
+        }
+
         // Check if it's time to take a request.
         auto& vblank_status = main_port.vblank_status;
         if (vblank_status.count % (main_port.flip_rate + 1) == 0) {
             const auto request = receive_request();
             if (!request) {
-                if (!main_port.is_open || DebugState.IsGuestThreadsPaused()) {
+                if (!main_port.is_open) {
                     DrawBlankFrame();
+                } else {
+                    DrawLastFrame();
                 }
             } else {
                 Flip(request);

src/core/libraries/videoout/driver.h

@@ -102,7 +102,8 @@ private:
     };
 
     void Flip(const Request& req);
-    void DrawBlankFrame(); // Used when there is no flip request to keep ImGui up to date
+    void DrawBlankFrame(); // Video port out not open
+    void DrawLastFrame();  // Used when there is no flip request
     void SubmitFlipInternal(VideoOutPort* port, s32 index, s64 flip_arg, bool is_eop = false);
     void PresentThread(std::stop_token token);
 

src/imgui/imgui_config.h

@@ -30,6 +30,12 @@ extern void assert_fail_debug_msg(const char* msg);
 #define IM_VEC4_CLASS_EXTRA                                                                        \
     constexpr ImVec4(float _v) : x(_v), y(_v), z(_v), w(_v) {}
 
+namespace ImGui {
+struct Texture;
+}
+#define ImTextureID ImTextureID
+using ImTextureID = ::ImGui::Texture*;
+
 #ifdef IMGUI_USE_WCHAR32
 #error "This project uses 16 bits wchar standard like Orbis"
 #endif

src/imgui/renderer/imgui_core.cpp

@@ -6,6 +6,7 @@
 
 #include "common/config.h"
 #include "common/path_util.h"
+#include "core/debug_state.h"
 #include "core/devtools/layer.h"
 #include "imgui/imgui_layer.h"
 #include "imgui_core.h"
@@ -167,7 +168,7 @@ bool ProcessEvent(SDL_Event* event) {
     }
 }
 
-void NewFrame() {
+ImGuiID NewFrame(bool is_reusing_frame) {
     {
         std::scoped_lock lock{change_layers_mutex};
         while (!change_layers.empty()) {
@@ -182,17 +183,24 @@ void NewFrame() {
         }
     }
 
-    Sdl::NewFrame();
+    Sdl::NewFrame(is_reusing_frame);
     ImGui::NewFrame();
 
-    DockSpaceOverViewport(0, GetMainViewport(), ImGuiDockNodeFlags_PassthruCentralNode);
+    ImGuiWindowFlags flags = ImGuiDockNodeFlags_PassthruCentralNode;
+    if (!DebugState.ShowingDebugMenuBar()) {
+        flags |= ImGuiDockNodeFlags_NoTabBar;
+    }
+    ImGuiID dockId = DockSpaceOverViewport(0, GetMainViewport(), flags);
 
     for (auto* layer : layers) {
         layer->Draw();
     }
+
+    return dockId;
 }
 
-void Render(const vk::CommandBuffer& cmdbuf, ::Vulkan::Frame* frame) {
+void Render(const vk::CommandBuffer& cmdbuf, const vk::ImageView& image_view,
+            const vk::Extent2D& extent) {
     ImGui::Render();
     ImDrawData* draw_data = GetDrawData();
     if (draw_data->CmdListsCount == 0) {
@@ -207,16 +215,16 @@ void Render(const vk::CommandBuffer& cmdbuf, ::Vulkan::Frame* frame) {
 
     vk::RenderingAttachmentInfo color_attachments[1]{
         {
-            .imageView = frame->image_view,
+            .imageView = image_view,
             .imageLayout = vk::ImageLayout::eColorAttachmentOptimal,
-            .loadOp = vk::AttachmentLoadOp::eLoad,
+            .loadOp = vk::AttachmentLoadOp::eClear,
             .storeOp = vk::AttachmentStoreOp::eStore,
         },
     };
     vk::RenderingInfo render_info{};
     render_info.renderArea = vk::Rect2D{
         .offset = {0, 0},
-        .extent = {frame->width, frame->height},
+        .extent = extent,
     };
     render_info.layerCount = 1;
     render_info.colorAttachmentCount = 1;

src/imgui/renderer/imgui_core.h

@@ -3,6 +3,8 @@
 
 #pragma once
 
+#include <imgui.h>
+
 #include "video_core/renderer_vulkan/vk_instance.h"
 #include "vulkan/vulkan_handles.hpp"
 
@@ -24,8 +26,9 @@ void Shutdown(const vk::Device& device);
 
 bool ProcessEvent(SDL_Event* event);
 
-void NewFrame();
+ImGuiID NewFrame(bool is_reusing_frame = false);
 
-void Render(const vk::CommandBuffer& cmdbuf, Vulkan::Frame* frame);
+void Render(const vk::CommandBuffer& cmdbuf, const vk::ImageView& image_view,
+            const vk::Extent2D& extent);
 
 } // namespace ImGui::Core

src/imgui/renderer/imgui_impl_sdl3.cpp

@@ -5,6 +5,7 @@
 
 #include <imgui.h>
 #include "common/config.h"
+#include "core/debug_state.h"
 #include "imgui_impl_sdl3.h"
 
 // SDL
@@ -26,6 +27,7 @@ struct SdlData {
     SDL_Window* window{};
     SDL_WindowID window_id{};
     Uint64 time{};
+    Uint64 nonReusedtime{};
     const char* clipboard_text_data{};
 
     // IME handling
@@ -785,7 +787,7 @@ static void UpdateGamepads() {
                         +thumb_dead_zone, +32767);
 }
 
-void NewFrame() {
+void NewFrame(bool is_reusing_frame) {
     SdlData* bd = GetBackendData();
     IM_ASSERT(bd != nullptr && "No platform backend to shutdown, or already shutdown?");
     ImGuiIO& io = ImGui::GetIO();
@@ -798,9 +800,26 @@ void NewFrame() {
     if (current_time <= bd->time)
         current_time = bd->time + 1;
     io.DeltaTime = bd->time > 0 ? (float)((double)(current_time - bd->time) / (double)frequency)
-                                : (float)(1.0f / 60.0f);
+                                : 1.0f / 60.0f;
     bd->time = current_time;
 
+    if (!is_reusing_frame) {
+        if (current_time <= bd->nonReusedtime)
+            current_time = bd->nonReusedtime + 1;
+        float deltaTime =
+            bd->nonReusedtime > 0
+                ? (float)((double)(current_time - bd->nonReusedtime) / (double)frequency)
+                : 1.0f / 60.0f;
+        bd->nonReusedtime = current_time;
+        DebugState.FrameDeltaTime = deltaTime;
+        float distribution = 0.016f / deltaTime / 10.0f;
+        if (distribution > 1.0f) {
+            distribution = 1.0f;
+        }
+        DebugState.Framerate =
+            deltaTime * distribution + DebugState.Framerate * (1.0f - distribution);
+    }
+
     if (bd->mouse_pending_leave_frame && bd->mouse_pending_leave_frame >= ImGui::GetFrameCount() &&
         bd->mouse_buttons_down == 0) {
         bd->mouse_window_id = 0;

src/imgui/renderer/imgui_impl_sdl3.h

@@ -14,7 +14,7 @@ namespace ImGui::Sdl {
 
 bool Init(SDL_Window* window);
 void Shutdown();
-void NewFrame();
+void NewFrame(bool is_reusing);
 bool ProcessEvent(const SDL_Event* event);
 void OnResize();
 

src/imgui/renderer/imgui_impl_vulkan.cpp

@@ -57,11 +57,12 @@ struct VkData {
     vk::DeviceMemory font_memory{};
     vk::Image font_image{};
     vk::ImageView font_view{};
-    vk::DescriptorSet font_descriptor_set{};
+    ImTextureID font_texture{};
     vk::CommandBuffer font_command_buffer{};
 
     // Render buffers
     WindowRenderBuffers render_buffers{};
+    bool enabled_blending{true};
 
     VkData(const InitInfo init_info) : init_info(init_info) {
         render_buffers.count = init_info.image_count;
@@ -252,8 +253,8 @@ void UploadTextureData::Destroy() {
     const InitInfo& v = bd->init_info;
 
     CheckVkErr(v.device.waitIdle());
-    RemoveTexture(descriptor_set);
-    descriptor_set = VK_NULL_HANDLE;
+    RemoveTexture(im_texture);
+    im_texture = nullptr;
 
     v.device.destroyImageView(image_view, v.allocator);
     image_view = VK_NULL_HANDLE;
@@ -264,8 +265,8 @@ void UploadTextureData::Destroy() {
 }
 
 // Register a texture
-vk::DescriptorSet AddTexture(vk::ImageView image_view, vk::ImageLayout image_layout,
-                             vk::Sampler sampler) {
+ImTextureID AddTexture(vk::ImageView image_view, vk::ImageLayout image_layout,
+                       vk::Sampler sampler) {
     VkData* bd = GetBackendData();
     const InitInfo& v = bd->init_info;
 
@@ -303,7 +304,9 @@ vk::DescriptorSet AddTexture(vk::ImageView image_view, vk::ImageLayout image_lay
         };
         v.device.updateDescriptorSets({write_desc}, {});
     }
-    return descriptor_set;
+    return new Texture{
+        .descriptor_set = descriptor_set,
+    };
 }
 UploadTextureData UploadTexture(const void* data, vk::Format format, u32 width, u32 height,
                                 size_t size) {
@@ -370,7 +373,7 @@ UploadTextureData UploadTexture(const void* data, vk::Format format, u32 width,
     }
 
     // Create descriptor set (ImTextureID)
-    info.descriptor_set = AddTexture(info.image_view, vk::ImageLayout::eShaderReadOnlyOptimal);
+    info.im_texture = AddTexture(info.image_view, vk::ImageLayout::eShaderReadOnlyOptimal);
 
     // Create Upload Buffer
     {
@@ -464,10 +467,12 @@ UploadTextureData UploadTexture(const void* data, vk::Format format, u32 width,
     return info;
 }
 
-void RemoveTexture(vk::DescriptorSet descriptor_set) {
+void RemoveTexture(ImTextureID texture) {
+    IM_ASSERT(texture != nullptr);
     VkData* bd = GetBackendData();
     const InitInfo& v = bd->init_info;
-    v.device.freeDescriptorSets(bd->descriptor_pool, {descriptor_set});
+    v.device.freeDescriptorSets(bd->descriptor_pool, {texture->descriptor_set});
+    delete texture;
 }
 
 static void CreateOrResizeBuffer(RenderBuffer& rb, size_t new_size, vk::BufferUsageFlagBits usage) {
@@ -679,15 +684,11 @@ void RenderDrawData(ImDrawData& draw_data, vk::CommandBuffer command_buffer,
                 command_buffer.setScissor(0, 1, &scissor);
 
                 // Bind DescriptorSet with font or user texture
-                vk::DescriptorSet desc_set[1]{(VkDescriptorSet)pcmd->TextureId};
-                if (sizeof(ImTextureID) < sizeof(ImU64)) {
-                    // We don't support texture switches if ImTextureID hasn't been redefined to be
-                    // 64-bit. Do a flaky check that other textures haven't been used.
-                    IM_ASSERT(pcmd->TextureId == (ImTextureID)bd->font_descriptor_set);
-                    desc_set[0] = bd->font_descriptor_set;
-                }
+                vk::DescriptorSet desc_set[1]{pcmd->TextureId->descriptor_set};
                 command_buffer.bindDescriptorSets(vk::PipelineBindPoint::eGraphics,
                                                   bd->pipeline_layout, 0, {desc_set}, {});
+                command_buffer.setColorBlendEnableEXT(
+                    0, {pcmd->TextureId->disable_blend ? vk::False : vk::True});
 
                 // Draw
                 command_buffer.drawIndexed(pcmd->ElemCount, 1, pcmd->IdxOffset + global_idx_offset,
@@ -709,7 +710,7 @@ static bool CreateFontsTexture() {
     const InitInfo& v = bd->init_info;
 
     // Destroy existing texture (if any)
-    if (bd->font_view || bd->font_image || bd->font_memory || bd->font_descriptor_set) {
+    if (bd->font_view || bd->font_image || bd->font_memory || bd->font_texture) {
         CheckVkErr(v.queue.waitIdle());
         DestroyFontsTexture();
     }
@@ -782,7 +783,7 @@ static bool CreateFontsTexture() {
     }
 
     // Create the Descriptor Set:
-    bd->font_descriptor_set = AddTexture(bd->font_view, vk::ImageLayout::eShaderReadOnlyOptimal);
+    bd->font_texture = AddTexture(bd->font_view, vk::ImageLayout::eShaderReadOnlyOptimal);
 
     // Create the Upload Buffer:
     vk::DeviceMemory upload_buffer_memory{};
@@ -874,7 +875,7 @@ static bool CreateFontsTexture() {
     }
 
     // Store our identifier
-    io.Fonts->SetTexID(bd->font_descriptor_set);
+    io.Fonts->SetTexID(bd->font_texture);
 
     // End command buffer
     vk::SubmitInfo end_info = {};
@@ -898,9 +899,9 @@ static void DestroyFontsTexture() {
     VkData* bd = GetBackendData();
     const InitInfo& v = bd->init_info;
 
-    if (bd->font_descriptor_set) {
-        RemoveTexture(bd->font_descriptor_set);
-        bd->font_descriptor_set = VK_NULL_HANDLE;
+    if (bd->font_texture) {
+        RemoveTexture(bd->font_texture);
+        bd->font_texture = nullptr;
         io.Fonts->SetTexID(nullptr);
     }
 
@@ -1057,9 +1058,10 @@ static void CreatePipeline(vk::Device device, const vk::AllocationCallbacks* all
         .pAttachments = color_attachment,
     };
 
-    vk::DynamicState dynamic_states[2]{
+    vk::DynamicState dynamic_states[3]{
         vk::DynamicState::eViewport,
         vk::DynamicState::eScissor,
+        vk::DynamicState::eColorBlendEnableEXT,
     };
     vk::PipelineDynamicStateCreateInfo dynamic_state{
         .dynamicStateCount = (uint32_t)IM_ARRAYSIZE(dynamic_states),

src/imgui/renderer/imgui_impl_vulkan.h

@@ -10,6 +10,13 @@
 
 struct ImDrawData;
 
+namespace ImGui {
+struct Texture {
+    vk::DescriptorSet descriptor_set{nullptr};
+    bool disable_blend{false};
+};
+} // namespace ImGui
+
 namespace ImGui::Vulkan {
 
 struct InitInfo {
@@ -34,29 +41,32 @@ struct InitInfo {
 struct UploadTextureData {
     vk::Image image;
     vk::ImageView image_view;
-    vk::DescriptorSet descriptor_set;
     vk::DeviceMemory image_memory;
 
     vk::CommandBuffer command_buffer; // Submit to the queue
     vk::Buffer upload_buffer;
     vk::DeviceMemory upload_buffer_memory;
 
+    ImTextureID im_texture;
+
     void Upload();
 
     void Destroy();
 };
 
-vk::DescriptorSet AddTexture(vk::ImageView image_view, vk::ImageLayout image_layout,
-                             vk::Sampler sampler = VK_NULL_HANDLE);
+ImTextureID AddTexture(vk::ImageView image_view, vk::ImageLayout image_layout,
+                       vk::Sampler sampler = VK_NULL_HANDLE);
 
 UploadTextureData UploadTexture(const void* data, vk::Format format, u32 width, u32 height,
                                 size_t size);
 
-void RemoveTexture(vk::DescriptorSet descriptor_set);
+void RemoveTexture(ImTextureID descriptor_set);
 
 bool Init(InitInfo info);
 void Shutdown();
 void RenderDrawData(ImDrawData& draw_data, vk::CommandBuffer command_buffer,
                     vk::Pipeline pipeline = VK_NULL_HANDLE);
 
+void SetBlendEnabled(bool enabled);
+
 } // namespace ImGui::Vulkan

src/imgui/renderer/texture_manager.cpp

@@ -4,6 +4,7 @@
 #include <deque>
 #include <utility>
 
+#include <imgui.h>
 #include "common/assert.h"
 #include "common/config.h"
 #include "common/io_file.h"
@@ -123,7 +124,7 @@ static std::deque<UploadJob> g_upload_list;
 namespace Core::TextureManager {
 
 Inner::~Inner() {
-    if (upload_data.descriptor_set != nullptr) {
+    if (upload_data.im_texture != nullptr) {
         std::unique_lock lk{g_upload_mtx};
         g_upload_list.emplace_back(UploadJob{
             .data = this->upload_data,
@@ -239,7 +240,7 @@ void Submit() {
     }
     if (upload.core != nullptr) {
         upload.core->upload_data.Upload();
-        upload.core->texture_id = upload.core->upload_data.descriptor_set;
+        upload.core->texture_id = upload.core->upload_data.im_texture;
         if (upload.core->count.fetch_sub(1) == 1) {
             delete upload.core;
         }

src/video_core/renderer_vulkan/vk_instance.cpp

@@ -383,6 +383,7 @@ bool Instance::CreateDevice() {
             .extendedDynamicState = true,
         },
         vk::PhysicalDeviceExtendedDynamicState3FeaturesEXT{
+            .extendedDynamicState3ColorBlendEnable = true,
             .extendedDynamicState3ColorWriteMask = true,
         },
         vk::PhysicalDeviceDepthClipControlFeaturesEXT{

src/video_core/renderer_vulkan/vk_presenter.cpp

@@ -20,6 +20,9 @@
 
 #include <vk_mem_alloc.h>
 
+#include <imgui.h>
+#include "imgui/renderer/imgui_impl_vulkan.h"
+
 namespace Vulkan {
 
 bool CanBlitToSwapchain(const vk::PhysicalDevice physical_device, vk::Format format) {
@@ -103,17 +106,6 @@ static vk::Rect2D FitImage(s32 frame_width, s32 frame_height, s32 swapchain_widt
                          dst_rect.offset.x, dst_rect.offset.y);
 }
 
-static vk::Format FormatToUnorm(vk::Format fmt) {
-    switch (fmt) {
-    case vk::Format::eR8G8B8A8Srgb:
-        return vk::Format::eR8G8B8A8Unorm;
-    case vk::Format::eB8G8R8A8Srgb:
-        return vk::Format::eB8G8R8A8Unorm;
-    default:
-        UNREACHABLE();
-    }
-}
-
 void Presenter::CreatePostProcessPipeline() {
     static const std::array pp_shaders{
         HostShaders::FS_TRI_VERT,
@@ -324,9 +316,6 @@ Presenter::Presenter(Frontend::WindowSDL& window_, AmdGpu::Liverpool* liverpool_
 
     CreatePostProcessPipeline();
 
-    // Setup ImGui
-    ImGui::Core::Initialize(instance, window, num_images,
-                            FormatToUnorm(swapchain.GetSurfaceFormat().format));
     ImGui::Layer::AddLayer(Common::Singleton<Core::Devtools::Layer>::Instance());
 }
 
@@ -344,6 +333,9 @@ Presenter::~Presenter() {
 
 void Presenter::RecreateFrame(Frame* frame, u32 width, u32 height) {
     const vk::Device device = instance.GetDevice();
+    if (frame->imgui_texture) {
+        ImGui::Vulkan::RemoveTexture(frame->imgui_texture);
+    }
     if (frame->image_view) {
         device.destroyImageView(frame->image_view);
     }
@@ -361,7 +353,7 @@ void Presenter::RecreateFrame(Frame* frame, u32 width, u32 height) {
         .arrayLayers = 1,
         .samples = vk::SampleCountFlagBits::e1,
         .usage = vk::ImageUsageFlagBits::eColorAttachment | vk::ImageUsageFlagBits::eTransferDst |
-                 vk::ImageUsageFlagBits::eTransferSrc,
+                 vk::ImageUsageFlagBits::eTransferSrc | vk::ImageUsageFlagBits::eSampled,
     };
 
     const VmaAllocationCreateInfo alloc_info = {
@@ -403,6 +395,64 @@ void Presenter::RecreateFrame(Frame* frame, u32 width, u32 height) {
     frame->image_view = view;
     frame->width = width;
     frame->height = height;
+
+    frame->imgui_texture = ImGui::Vulkan::AddTexture(view, vk::ImageLayout::eShaderReadOnlyOptimal);
+    frame->imgui_texture->disable_blend = true;
+}
+
+Frame* Presenter::PrepareLastFrame() {
+    if (last_submit_frame == nullptr) {
+        return nullptr;
+    }
+
+    Frame* frame = last_submit_frame;
+
+    while (true) {
+        vk::Result result = instance.GetDevice().waitForFences(frame->present_done, false,
+                                                               std::numeric_limits<u64>::max());
+        if (result == vk::Result::eSuccess) {
+            break;
+        }
+        if (result == vk::Result::eTimeout) {
+            continue;
+        }
+        ASSERT_MSG(result != vk::Result::eErrorDeviceLost,
+                   "Device lost during waiting for a frame");
+    }
+
+    auto& scheduler = flip_scheduler;
+    scheduler.EndRendering();
+    const auto cmdbuf = scheduler.CommandBuffer();
+
+    const auto frame_subresources = vk::ImageSubresourceRange{
+        .aspectMask = vk::ImageAspectFlagBits::eColor,
+        .baseMipLevel = 0,
+        .levelCount = 1,
+        .baseArrayLayer = 0,
+        .layerCount = VK_REMAINING_ARRAY_LAYERS,
+    };
+
+    const auto pre_barrier =
+        vk::ImageMemoryBarrier2{.srcStageMask = vk::PipelineStageFlagBits2::eColorAttachmentOutput,
+                                .srcAccessMask = vk::AccessFlagBits2::eColorAttachmentRead,
+                                .dstStageMask = vk::PipelineStageFlagBits2::eColorAttachmentOutput,
+                                .dstAccessMask = vk::AccessFlagBits2::eColorAttachmentWrite,
+                                .oldLayout = vk::ImageLayout::eShaderReadOnlyOptimal,
+                                .newLayout = vk::ImageLayout::eGeneral,
+                                .image = frame->image,
+                                .subresourceRange{frame_subresources}};
+
+    cmdbuf.pipelineBarrier2(vk::DependencyInfo{
+        .imageMemoryBarrierCount = 1,
+        .pImageMemoryBarriers = &pre_barrier,
+    });
+
+    // Flush frame creation commands.
+    frame->ready_semaphore = scheduler.GetMasterSemaphore()->Handle();
+    frame->ready_tick = scheduler.CurrentTick();
+    SubmitInfo info{};
+    scheduler.Flush(info);
+    return frame;
 }
 
 bool Presenter::ShowSplash(Frame* frame /*= nullptr*/) {
@@ -499,6 +549,14 @@ Frame* Presenter::PrepareFrameInternal(VideoCore::ImageId image_id, bool is_eop)
     // Request a free presentation frame.
     Frame* frame = GetRenderFrame();
 
+    if (image_id != VideoCore::NULL_IMAGE_ID) {
+        const auto& image = texture_cache.GetImage(image_id);
+        const auto extent = image.info.size;
+        if (frame->width != extent.width || frame->height != extent.height) {
+            RecreateFrame(frame, extent.width, extent.height);
+        }
+    }
+
     // EOP flips are triggered from GPU thread so use the drawing scheduler to record
     // commands. Otherwise we are dealing with a CPU flip which could have arrived
     // from any guest thread. Use a separate scheduler for that.
@@ -515,8 +573,8 @@ Frame* Presenter::PrepareFrameInternal(VideoCore::ImageId image_id, bool is_eop)
     };
 
     const auto pre_barrier =
-        vk::ImageMemoryBarrier2{.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
-                                .srcAccessMask = vk::AccessFlagBits2::eTransferRead,
+        vk::ImageMemoryBarrier2{.srcStageMask = vk::PipelineStageFlagBits2::eColorAttachmentOutput,
+                                .srcAccessMask = vk::AccessFlagBits2::eColorAttachmentRead,
                                 .dstStageMask = vk::PipelineStageFlagBits2::eColorAttachmentOutput,
                                 .dstAccessMask = vk::AccessFlagBits2::eColorAttachmentWrite,
                                 .oldLayout = vk::ImageLayout::eUndefined,
@@ -627,17 +685,19 @@ Frame* Presenter::PrepareFrameInternal(VideoCore::ImageId image_id, bool is_eop)
     return frame;
 }
 
-void Presenter::Present(Frame* frame) {
+void Presenter::Present(Frame* frame, bool is_reusing_frame) {
     // Free the frame for reuse
     const auto free_frame = [&] {
-        std::scoped_lock fl{free_mutex};
-        free_queue.push(frame);
-        free_cv.notify_one();
+        if (!is_reusing_frame) {
+            last_submit_frame = frame;
+            std::scoped_lock fl{free_mutex};
+            free_queue.push(frame);
+            free_cv.notify_one();
+        }
     };
 
     // Recreate the swapchain if the window was resized.
-    if (window.GetWidth() != swapchain.GetExtent().width ||
-        window.GetHeight() != swapchain.GetExtent().height) {
+    if (window.GetWidth() != swapchain.GetWidth() || window.GetHeight() != swapchain.GetHeight()) {
         swapchain.Recreate(window.GetWidth(), window.GetHeight());
     }
 
@@ -656,15 +716,14 @@ void Presenter::Present(Frame* frame) {
     // the frame's present fence and future GetRenderFrame() call will hang waiting for this frame.
     instance.GetDevice().resetFences(frame->present_done);
 
-    ImGui::Core::NewFrame();
+    ImGuiID dockId = ImGui::Core::NewFrame(is_reusing_frame);
 
     const vk::Image swapchain_image = swapchain.Image();
+    const vk::ImageView swapchain_image_view = swapchain.ImageView();
 
     auto& scheduler = present_scheduler;
     const auto cmdbuf = scheduler.CommandBuffer();
 
-    ImGui::Core::Render(cmdbuf, frame);
-
     {
         auto* profiler_ctx = instance.GetProfilerContext();
         TracyVkNamedZoneC(profiler_ctx, renderer_gpu_zone, cmdbuf, "Host frame",
@@ -674,9 +733,9 @@ void Presenter::Present(Frame* frame) {
         const std::array pre_barriers{
             vk::ImageMemoryBarrier{
                 .srcAccessMask = vk::AccessFlagBits::eNone,
-                .dstAccessMask = vk::AccessFlagBits::eTransferWrite,
+                .dstAccessMask = vk::AccessFlagBits::eColorAttachmentWrite,
                 .oldLayout = vk::ImageLayout::eUndefined,
-                .newLayout = vk::ImageLayout::eTransferDstOptimal,
+                .newLayout = vk::ImageLayout::eColorAttachmentOptimal,
                 .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                 .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                 .image = swapchain_image,
@@ -690,9 +749,9 @@ void Presenter::Present(Frame* frame) {
             },
             vk::ImageMemoryBarrier{
                 .srcAccessMask = vk::AccessFlagBits::eColorAttachmentWrite,
-                .dstAccessMask = vk::AccessFlagBits::eTransferRead,
+                .dstAccessMask = vk::AccessFlagBits::eColorAttachmentRead,
                 .oldLayout = vk::ImageLayout::eGeneral,
-                .newLayout = vk::ImageLayout::eTransferSrcOptimal,
+                .newLayout = vk::ImageLayout::eShaderReadOnlyOptimal,
                 .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                 .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                 .image = frame->image,
@@ -705,10 +764,11 @@ void Presenter::Present(Frame* frame) {
                 },
             },
         };
+
         const vk::ImageMemoryBarrier post_barrier{
-            .srcAccessMask = vk::AccessFlagBits::eTransferWrite,
+            .srcAccessMask = vk::AccessFlagBits::eColorAttachmentWrite,
             .dstAccessMask = vk::AccessFlagBits::eMemoryRead,
-            .oldLayout = vk::ImageLayout::eTransferDstOptimal,
+            .oldLayout = vk::ImageLayout::eColorAttachmentOptimal,
             .newLayout = vk::ImageLayout::ePresentSrcKHR,
             .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
             .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
@@ -723,14 +783,29 @@ void Presenter::Present(Frame* frame) {
         };
 
         cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eColorAttachmentOutput,
-                               vk::PipelineStageFlagBits::eTransfer,
+                               vk::PipelineStageFlagBits::eColorAttachmentOutput,
                                vk::DependencyFlagBits::eByRegion, {}, {}, pre_barriers);
 
-        cmdbuf.blitImage(
-            frame->image, vk::ImageLayout::eTransferSrcOptimal, swapchain_image,
-            vk::ImageLayout::eTransferDstOptimal,
-            MakeImageBlitStretch(frame->width, frame->height, extent.width, extent.height),
-            vk::Filter::eLinear);
+        { // Draw the game
+            ImGui::PushStyleVar(ImGuiStyleVar_WindowPadding, ImVec2{0.0f});
+            ImGui::SetNextWindowDockID(dockId, ImGuiCond_Once);
+            ImGui::Begin("Display##game_display", nullptr, ImGuiWindowFlags_NoNav);
+
+            ImVec2 contentArea = ImGui::GetContentRegionAvail();
+            const vk::Rect2D imgRect =
+                FitImage(frame->width, frame->height, (s32)contentArea.x, (s32)contentArea.y);
+            ImGui::SetCursorPos(ImGui::GetCursorStartPos() + ImVec2{
+                                                                 (float)imgRect.offset.x,
+                                                                 (float)imgRect.offset.y,
+                                                             });
+            ImGui::Image(frame->imgui_texture, {
+                                                   static_cast<float>(imgRect.extent.width),
+                                                   static_cast<float>(imgRect.extent.height),
+                                               });
+            ImGui::End();
+            ImGui::PopStyleVar();
+        }
+        ImGui::Core::Render(cmdbuf, swapchain_image_view, swapchain.GetExtent());
 
         cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands,
                                vk::PipelineStageFlagBits::eAllCommands,
@@ -756,7 +831,9 @@ void Presenter::Present(Frame* frame) {
     }
 
     free_frame();
-    DebugState.IncFlipFrameNum();
+    if (!is_reusing_frame) {
+        DebugState.IncFlipFrameNum();
+    }
 }
 
 Frame* Presenter::GetRenderFrame() {
@@ -790,9 +867,9 @@ Frame* Presenter::GetRenderFrame() {
         }
     }
 
-    // If the window dimensions changed, recreate this frame
-    if (frame->width != window.GetWidth() || frame->height != window.GetHeight()) {
-        RecreateFrame(frame, window.GetWidth(), window.GetHeight());
+    // Initialize default frame image
+    if (frame->width == 0 || frame->height == 0) {
+        RecreateFrame(frame, 1920, 1080);
     }
 
     return frame;

src/video_core/renderer_vulkan/vk_presenter.h

@@ -5,6 +5,7 @@
 
 #include <condition_variable>
 
+#include "imgui/imgui_config.h"
 #include "video_core/amdgpu/liverpool.h"
 #include "video_core/renderer_vulkan/vk_instance.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
@@ -30,6 +31,8 @@ struct Frame {
     vk::Fence present_done;
     vk::Semaphore ready_semaphore;
     u64 ready_tick;
+
+    ImTextureID imgui_texture;
 };
 
 enum SchedulerType {
@@ -86,8 +89,9 @@ public:
     }
 
     bool ShowSplash(Frame* frame = nullptr);
-    void Present(Frame* frame);
+    void Present(Frame* frame, bool is_reusing_frame = false);
     void RecreateFrame(Frame* frame, u32 width, u32 height);
+    Frame* PrepareLastFrame();
 
     void FlushDraw() {
         SubmitInfo info{};
@@ -121,6 +125,7 @@ private:
     vk::UniqueCommandPool command_pool;
     std::vector<Frame> present_frames;
     std::queue<Frame*> free_queue;
+    Frame* last_submit_frame;
     std::mutex free_mutex;
     std::condition_variable free_cv;
     std::condition_variable_any frame_cv;

src/video_core/renderer_vulkan/vk_swapchain.cpp

@@ -4,8 +4,8 @@
 #include <algorithm>
 #include <limits>
 #include "common/assert.h"
-#include "common/config.h"
 #include "common/logging/log.h"
+#include "imgui/renderer/imgui_core.h"
 #include "sdl_window.h"
 #include "video_core/renderer_vulkan/vk_instance.h"
 #include "video_core/renderer_vulkan/vk_swapchain.h"
@@ -15,7 +15,9 @@ namespace Vulkan {
 Swapchain::Swapchain(const Instance& instance_, const Frontend::WindowSDL& window)
     : instance{instance_}, surface{CreateSurface(instance.GetInstance(), window)} {
     FindPresentFormat();
-    Create(window.GetWidth(), window.GetHeight(), surface);
+
+    Create(window.GetWidth(), window.GetHeight());
+    ImGui::Core::Initialize(instance, window, image_count, surface_format.format);
 }
 
 Swapchain::~Swapchain() {
@@ -23,10 +25,9 @@ Swapchain::~Swapchain() {
     instance.GetInstance().destroySurfaceKHR(surface);
 }
 
-void Swapchain::Create(u32 width_, u32 height_, vk::SurfaceKHR surface_) {
+void Swapchain::Create(u32 width_, u32 height_) {
     width = width_;
     height = height_;
-    surface = surface_;
     needs_recreation = false;
 
     Destroy();
@@ -86,7 +87,7 @@ void Swapchain::Create(u32 width_, u32 height_, vk::SurfaceKHR surface_) {
 
 void Swapchain::Recreate(u32 width_, u32 height_) {
     LOG_DEBUG(Render_Vulkan, "Recreate the swapchain: width={} height={}", width_, height_);
-    Create(width_, height_, surface);
+    Create(width_, height_);
 }
 
 bool Swapchain::AcquireNextImage() {
@@ -209,10 +210,14 @@ void Swapchain::Destroy() {
     if (swapchain) {
         device.destroySwapchainKHR(swapchain);
     }
-    for (u32 i = 0; i < image_count; i++) {
-        device.destroySemaphore(image_acquired[i]);
-        device.destroySemaphore(present_ready[i]);
+
+    for (const auto& sem : image_acquired) {
+        device.destroySemaphore(sem);
     }
+    for (const auto& sem : present_ready) {
+        device.destroySemaphore(sem);
+    }
+
     image_acquired.clear();
     present_ready.clear();
 }
@@ -249,9 +254,30 @@ void Swapchain::SetupImages() {
                vk::to_string(images_result));
     images = std::move(imgs);
     image_count = static_cast<u32>(images.size());
+    images_view.resize(image_count);
+    for (u32 i = 0; i < image_count; ++i) {
+        if (images_view[i]) {
+            device.destroyImageView(images_view[i]);
+        }
+        auto [im_view_result, im_view] = device.createImageView(vk::ImageViewCreateInfo{
+            .image = images[i],
+            .viewType = vk::ImageViewType::e2D,
+            .format = FormatToUnorm(surface_format.format),
+            .subresourceRange =
+                {
+                    .aspectMask = vk::ImageAspectFlagBits::eColor,
+                    .levelCount = 1,
+                    .layerCount = 1,
+                },
+        });
+        ASSERT_MSG(im_view_result == vk::Result::eSuccess, "Failed to create image view: {}",
+                   vk::to_string(im_view_result));
+        images_view[i] = im_view;
+    }
 
     for (u32 i = 0; i < image_count; ++i) {
         SetObjectName(device, images[i], "Swapchain Image {}", i);
+        SetObjectName(device, images_view[i], "Swapchain ImageView {}", i);
     }
 }
 

src/video_core/renderer_vulkan/vk_swapchain.h

@@ -17,13 +17,24 @@ namespace Vulkan {
 class Instance;
 class Scheduler;
 
+inline vk::Format FormatToUnorm(vk::Format fmt) {
+    switch (fmt) {
+    case vk::Format::eR8G8B8A8Srgb:
+        return vk::Format::eR8G8B8A8Unorm;
+    case vk::Format::eB8G8R8A8Srgb:
+        return vk::Format::eB8G8R8A8Unorm;
+    default:
+        UNREACHABLE();
+    }
+}
+
 class Swapchain {
 public:
     explicit Swapchain(const Instance& instance, const Frontend::WindowSDL& window);
     ~Swapchain();
 
     /// Creates (or recreates) the swapchain with a given size.
-    void Create(u32 width, u32 height, vk::SurfaceKHR surface);
+    void Create(u32 width, u32 height);
 
     /// Recreates the swapchain with a given size and current surface.
     void Recreate(u32 width, u32 height);
@@ -42,6 +53,10 @@ public:
         return images[image_index];
     }
 
+    vk::ImageView ImageView() const {
+        return images_view[image_index];
+    }
+
     vk::SurfaceFormatKHR GetSurfaceFormat() const {
         return surface_format;
     }
@@ -103,6 +118,7 @@ private:
     vk::SurfaceTransformFlagBitsKHR transform;
     vk::CompositeAlphaFlagBitsKHR composite_alpha;
     std::vector<vk::Image> images;
+    std::vector<vk::ImageView> images_view;
     std::vector<vk::Semaphore> image_acquired;
     std::vector<vk::Semaphore> present_ready;
     u32 width = 0;