1 files changed, 173 insertions, 0 deletions
diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h
new file mode 100644
index 000000000..62bcea5bb
--- /dev/null
+++ b/src/video_core/gpu_thread.h
@@ -0,0 +1,173 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <atomic>
+#include <condition_variable>
+#include <mutex>
+#include <optional>
+#include <thread>
+#include <variant>
+
+#include "common/threadsafe_queue.h"
+#include "video_core/gpu.h"
+
+namespace Tegra {
+struct FramebufferConfig;
+class DmaPusher;
+} // namespace Tegra
+
+namespace Core {
+class System;
+namespace Timing {
+struct EventType;
+} // namespace Timing
+} // namespace Core
+
+namespace VideoCommon::GPUThread {
+
+/// Command to signal to the GPU thread that processing has ended
+struct EndProcessingCommand final {};
+
+/// Command to signal to the GPU thread that a command list is ready for processing
+struct SubmitListCommand final {
+    explicit SubmitListCommand(Tegra::CommandList&& entries) : entries{std::move(entries)} {}
+
+    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(CacheAddr addr, u64 size) : addr{addr}, size{size} {}
+
+    CacheAddr addr;
+    u64 size;
+};
+
+/// Command to signal to the GPU thread to invalidate a region
+struct InvalidateRegionCommand final {
+    explicit constexpr InvalidateRegionCommand(CacheAddr addr, u64 size) : addr{addr}, size{size} {}
+
+    CacheAddr addr;
+    u64 size;
+};
+
+/// Command to signal to the GPU thread to flush and invalidate a region
+struct FlushAndInvalidateRegionCommand final {
+    explicit constexpr FlushAndInvalidateRegionCommand(CacheAddr addr, u64 size)
+        : addr{addr}, size{size} {}
+
+    CacheAddr addr;
+    u64 size;
+};
+
+using CommandData =
+    std::variant<EndProcessingCommand, SubmitListCommand, SwapBuffersCommand, FlushRegionCommand,
+                 InvalidateRegionCommand, FlushAndInvalidateRegionCommand>;
+
+struct CommandDataContainer {
+    CommandDataContainer() = default;
+
+    CommandDataContainer(CommandData&& data, u64 next_fence)
+        : data{std::move(data)}, fence{next_fence} {}
+
+    CommandDataContainer& operator=(const CommandDataContainer& t) {
+        data = std::move(t.data);
+        fence = t.fence;
+        return *this;
+    }
+
+    CommandData data;
+    u64 fence{};
+};
+
+/// Struct used to synchronize the GPU thread
+struct SynchState final {
+    std::atomic_bool is_running{true};
+    std::atomic_int queued_frame_count{};
+    std::mutex synchronization_mutex;
+    std::mutex commands_mutex;
+    std::condition_variable commands_condition;
+    std::condition_variable synchronization_condition;
+
+    /// Returns true if the gap in GPU commands is small enough that we can consider the CPU and GPU
+    /// synchronized. This is entirely empirical.
+    bool IsSynchronized() const {
+        constexpr std::size_t max_queue_gap{5};
+        return queue.Size() <= max_queue_gap;
+    }
+
+    void TrySynchronize() {
+        if (IsSynchronized()) {
+            std::lock_guard<std::mutex> lock{synchronization_mutex};
+            synchronization_condition.notify_one();
+        }
+    }
+
+    void WaitForSynchronization(u64 fence);
+
+    void SignalCommands() {
+        if (queue.Empty()) {
+            return;
+        }
+
+        commands_condition.notify_one();
+    }
+
+    void WaitForCommands() {
+        std::unique_lock lock{commands_mutex};
+        commands_condition.wait(lock, [this] { return !queue.Empty(); });
+    }
+
+    using CommandQueue = Common::SPSCQueue<CommandDataContainer>;
+    CommandQueue queue;
+    u64 last_fence{};
+    std::atomic<u64> signaled_fence{};
+};
+
+/// Class used to manage the GPU thread
+class ThreadManager final {
+public:
+    explicit ThreadManager(Core::System& system, VideoCore::RendererBase& renderer,
+                           Tegra::DmaPusher& dma_pusher);
+    ~ThreadManager();
+
+    /// Push GPU command entries to be processed
+    void SubmitList(Tegra::CommandList&& entries);
+
+    /// Swap buffers (render frame)
+    void SwapBuffers(
+        std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer);
+
+    /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
+    void FlushRegion(CacheAddr addr, u64 size);
+
+    /// Notify rasterizer that any caches of the specified region should be invalidated
+    void InvalidateRegion(CacheAddr addr, u64 size);
+
+    /// Notify rasterizer that any caches of the specified region should be flushed and invalidated
+    void FlushAndInvalidateRegion(CacheAddr addr, u64 size);
+
+private:
+    /// Pushes a command to be executed by the GPU thread
+    u64 PushCommand(CommandData&& command_data);
+
+private:
+    SynchState state;
+    Core::System& system;
+    Core::Timing::EventType* synchronization_event{};
+    std::thread thread;
+    std::thread::id thread_id;
+};
+
+} // namespace VideoCommon::GPUThread