Merge branch 'yuzu-emu:master' into cinematic-fix-pr

externals/cubeb

@@ -1 +1 @@
-Subproject commit 2d817de7c58b33a7c045edf873f3f9c98e4a2082
+Subproject commit 48689ae7a73caeb747953f9ed664dc71d2f918d8

src/input_common/input_engine.cpp

@@ -380,13 +380,16 @@ void InputEngine::TriggerOnMotionChange(const PadIdentifier& identifier, int mot
     if (!configuring || !mapping_callback.on_data) {
         return;
     }
+    const auto old_value = GetMotion(identifier, motion);
     bool is_active = false;
-    if (std::abs(value.accel_x) > 1.5f || std::abs(value.accel_y) > 1.5f ||
-        std::abs(value.accel_z) > 1.5f) {
+    if (std::abs(value.accel_x - old_value.accel_x) > 1.5f ||
+        std::abs(value.accel_y - old_value.accel_y) > 1.5f ||
+        std::abs(value.accel_z - old_value.accel_z) > 1.5f) {
         is_active = true;
     }
-    if (std::abs(value.gyro_x) > 0.6f || std::abs(value.gyro_y) > 0.6f ||
-        std::abs(value.gyro_z) > 0.6f) {
+    if (std::abs(value.gyro_x - old_value.gyro_x) > 0.6f ||
+        std::abs(value.gyro_y - old_value.gyro_y) > 0.6f ||
+        std::abs(value.gyro_z - old_value.gyro_z) > 0.6f) {
         is_active = true;
     }
     if (!is_active) {

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -144,6 +144,10 @@ public:
         return state_tracker;
     }
 
+    void BarrierFeedbackLoop() const noexcept {
+        // OpenGL does not require a barrier for attachment feedback loops.
+    }
+
 private:
     struct StagingBuffers {
         explicit StagingBuffers(GLenum storage_flags_, GLenum map_flags_);

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -481,12 +481,13 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) {
     if constexpr (Spec::enabled_stages[4]) {
         prepare_stage(4);
     }
+    texture_cache.UpdateRenderTargets(false);
+    texture_cache.CheckFeedbackLoop(views);
     ConfigureDraw(rescaling, render_area);
 }
 
 void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
                                      const RenderAreaPushConstant& render_area) {
-    texture_cache.UpdateRenderTargets(false);
     scheduler.RequestRenderpass(texture_cache.GetFramebuffer());
 
     if (!is_built.load(std::memory_order::relaxed)) {

src/video_core/renderer_vulkan/vk_master_semaphore.cpp

@@ -10,11 +10,16 @@
 
 namespace Vulkan {
 
+constexpr u64 FENCE_RESERVE_SIZE = 8;
+
 MasterSemaphore::MasterSemaphore(const Device& device_) : device(device_) {
     if (!device.HasTimelineSemaphore()) {
         static constexpr VkFenceCreateInfo fence_ci{
             .sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, .pNext = nullptr, .flags = 0};
-        fence = device.GetLogical().CreateFence(fence_ci);
+        free_queue.resize(FENCE_RESERVE_SIZE);
+        std::ranges::generate(free_queue,
+                              [&] { return device.GetLogical().CreateFence(fence_ci); });
+        wait_thread = std::jthread([this](std::stop_token token) { WaitThread(token); });
         return;
     }
 
@@ -167,16 +172,53 @@ VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf, VkSemaphor
         .pSignalSemaphores = &signal_semaphore,
     };
 
+    auto fence = GetFreeFence();
     auto result = device.GetGraphicsQueue().Submit(submit_info, *fence);
 
     if (result == VK_SUCCESS) {
-        fence.Wait();
-        fence.Reset();
-        gpu_tick.store(host_tick);
-        gpu_tick.notify_all();
+        std::scoped_lock lock{wait_mutex};
+        wait_queue.emplace(host_tick, std::move(fence));
+        wait_cv.notify_one();
     }
 
     return result;
 }
 
+void MasterSemaphore::WaitThread(std::stop_token token) {
+    while (!token.stop_requested()) {
+        u64 host_tick;
+        vk::Fence fence;
+        {
+            std::unique_lock lock{wait_mutex};
+            Common::CondvarWait(wait_cv, lock, token, [this] { return !wait_queue.empty(); });
+            if (token.stop_requested()) {
+                return;
+            }
+            std::tie(host_tick, fence) = std::move(wait_queue.front());
+            wait_queue.pop();
+        }
+
+        fence.Wait();
+        fence.Reset();
+        gpu_tick.store(host_tick);
+        gpu_tick.notify_all();
+
+        std::scoped_lock lock{free_mutex};
+        free_queue.push_front(std::move(fence));
+    }
+}
+
+vk::Fence MasterSemaphore::GetFreeFence() {
+    std::scoped_lock lock{free_mutex};
+    if (free_queue.empty()) {
+        static constexpr VkFenceCreateInfo fence_ci{
+            .sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, .pNext = nullptr, .flags = 0};
+        return device.GetLogical().CreateFence(fence_ci);
+    }
+
+    auto fence = std::move(free_queue.back());
+    free_queue.pop_back();
+    return fence;
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_master_semaphore.h

@@ -5,8 +5,10 @@
 
 #include <atomic>
 #include <condition_variable>
+#include <deque>
 #include <mutex>
 #include <thread>
+#include <queue>
 
 #include "common/common_types.h"
 #include "common/polyfill_thread.h"
@@ -17,6 +19,8 @@ namespace Vulkan {
 class Device;
 
 class MasterSemaphore {
+    using Waitable = std::pair<u64, vk::Fence>;
+
 public:
     explicit MasterSemaphore(const Device& device);
     ~MasterSemaphore();
@@ -57,13 +61,22 @@ private:
     VkResult SubmitQueueFence(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
                               VkSemaphore wait_semaphore, u64 host_tick);
 
+    void WaitThread(std::stop_token token);
+
+    vk::Fence GetFreeFence();
+
 private:
     const Device& device;             ///< Device.
-    vk::Fence fence;                  ///< Fence.
     vk::Semaphore semaphore;          ///< Timeline semaphore.
     std::atomic<u64> gpu_tick{0};     ///< Current known GPU tick.
     std::atomic<u64> current_tick{1}; ///< Current logical tick.
+    std::mutex wait_mutex;
+    std::mutex free_mutex;
+    std::condition_variable_any wait_cv;
+    std::queue<Waitable> wait_queue;  ///< Queue for the fences to be waited on by the wait thread.
+    std::deque<vk::Fence> free_queue; ///< Holds available fences for submission.
     std::jthread debug_thread;        ///< Debug thread to workaround validation layer bugs.
+    std::jthread wait_thread;         ///< Helper thread that waits for submitted fences.
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -861,6 +861,10 @@ VkBuffer TextureCacheRuntime::GetTemporaryBuffer(size_t needed_size) {
     return *buffers[level];
 }
 
+void TextureCacheRuntime::BarrierFeedbackLoop() {
+    scheduler.RequestOutsideRenderPassOperationContext();
+}
+
 void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src,
                                            std::span<const VideoCommon::ImageCopy> copies) {
     std::vector<VkBufferImageCopy> vk_in_copies(copies.size());

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -103,6 +103,8 @@ public:
 
     [[nodiscard]] VkBuffer GetTemporaryBuffer(size_t needed_size);
 
+    void BarrierFeedbackLoop();
+
     const Device& device;
     Scheduler& scheduler;
     MemoryAllocator& memory_allocator;

src/video_core/texture_cache/texture_cache.h

@@ -183,6 +183,42 @@ void TextureCache<P>::FillComputeImageViews(std::span<ImageViewInOut> views) {
                          views);
 }
 
+template <class P>
+void TextureCache<P>::CheckFeedbackLoop(std::span<const ImageViewInOut> views) {
+    const bool requires_barrier = [&] {
+        for (const auto& view : views) {
+            if (!view.id) {
+                continue;
+            }
+            auto& image_view = slot_image_views[view.id];
+
+            // Check color targets
+            for (const auto& ct_view_id : render_targets.color_buffer_ids) {
+                if (ct_view_id) {
+                    auto& ct_view = slot_image_views[ct_view_id];
+                    if (image_view.image_id == ct_view.image_id) {
+                        return true;
+                    }
+                }
+            }
+
+            // Check zeta target
+            if (render_targets.depth_buffer_id) {
+                auto& zt_view = slot_image_views[render_targets.depth_buffer_id];
+                if (image_view.image_id == zt_view.image_id) {
+                    return true;
+                }
+            }
+        }
+
+        return false;
+    }();
+
+    if (requires_barrier) {
+        runtime.BarrierFeedbackLoop();
+    }
+}
+
 template <class P>
 typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) {
     if (index > channel_state->graphics_sampler_table.Limit()) {

src/video_core/texture_cache/texture_cache_base.h

@@ -148,6 +148,9 @@ public:
     /// Fill image_view_ids with the compute images in indices
     void FillComputeImageViews(std::span<ImageViewInOut> views);
 
+    /// Handle feedback loops during draws.
+    void CheckFeedbackLoop(std::span<const ImageViewInOut> views);
+
     /// Get the sampler from the graphics descriptor table in the specified index
     Sampler* GetGraphicsSampler(u32 index);
 

src/video_core/vulkan_common/vulkan_device.cpp

@@ -1034,7 +1034,7 @@ void Device::CollectPhysicalMemoryInfo() {
     }
     const s64 available_memory = static_cast<s64>(device_access_memory - device_initial_usage);
     device_access_memory = static_cast<u64>(std::max<s64>(
-        std::min<s64>(available_memory - 8_GiB, 4_GiB), static_cast<s64>(local_memory)));
+        std::min<s64>(available_memory - 8_GiB, 4_GiB), std::min<s64>(local_memory, 4_GiB)));
 }
 
 void Device::CollectToolingInfo() {