From 40571c073faa02a6a4301e7f0ce365ef50a400aa Mon Sep 17 00:00:00 2001
From: bunnei <bunneidev@gmail.com>
Date: Sat, 12 Dec 2020 00:24:33 -0800
Subject: [PATCH] video_core: gpu: Implement synchronous mode using threaded
 GPU.

---
 src/video_core/gpu.cpp        |  7 +++++--
 src/video_core/gpu.h          |  4 ++--
 src/video_core/gpu_thread.cpp | 30 ++++++++++++++++++++++++------
 src/video_core/gpu_thread.h   |  5 +++--
 4 files changed, 34 insertions(+), 12 deletions(-)

diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index f99a8a0de..6ab06775f 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -38,7 +38,7 @@ GPU::GPU(Core::System& system_, bool is_async_, bool use_nvdec_)
       maxwell_dma{std::make_unique<Engines::MaxwellDMA>(system, *memory_manager)},
       kepler_memory{std::make_unique<Engines::KeplerMemory>(system, *memory_manager)},
       shader_notify{std::make_unique<VideoCore::ShaderNotify>()}, is_async{is_async_},
-      gpu_thread{system_} {}
+      gpu_thread{system_, is_async_} {}
 
 GPU::~GPU() = default;
 
@@ -524,7 +524,10 @@ void GPU::WaitIdle() const {
 }
 
 void GPU::OnCommandListEnd() {
-    gpu_thread.OnCommandListEnd();
+    if (is_async) {
+        // This command only applies to asynchronous GPU mode
+        gpu_thread.OnCommandListEnd();
+    }
 }
 
 } // namespace Tegra
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index a2bb4d82d..d81e38680 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -406,10 +406,10 @@ private:
     u64 last_flush_fence{};
     std::mutex flush_request_mutex;
 
+    const bool is_async;
+
     VideoCommon::GPUThread::ThreadManager gpu_thread;
     std::unique_ptr<Core::Frontend::GraphicsContext> cpu_context;
-
-    const bool is_async;
 };
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \
diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index e27218b96..56b9621b1 100644
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -65,7 +65,8 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer,
     }
 }
 
-ThreadManager::ThreadManager(Core::System& system_) : system{system_} {}
+ThreadManager::ThreadManager(Core::System& system_, bool is_async_)
+    : system{system_}, is_async{is_async_} {}
 
 ThreadManager::~ThreadManager() {
     if (!thread.joinable()) {
@@ -97,19 +98,30 @@ void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
 }
 
 void ThreadManager::FlushRegion(VAddr addr, u64 size) {
-    if (!Settings::IsGPULevelHigh()) {
+    if (!is_async) {
+        // Always flush with synchronous GPU mode
         PushCommand(FlushRegionCommand(addr, size));
         return;
     }
-    if (!Settings::IsGPULevelExtreme()) {
-        return;
-    }
-    if (system.Renderer().Rasterizer().MustFlushRegion(addr, size)) {
+
+    // Asynchronous GPU mode
+    switch (Settings::values.gpu_accuracy.GetValue()) {
+    case Settings::GPUAccuracy::Normal:
+        PushCommand(FlushRegionCommand(addr, size));
+        break;
+    case Settings::GPUAccuracy::High:
+        // TODO(bunnei): Is this right? Preserving existing behavior for now
+        break;
+    case Settings::GPUAccuracy::Extreme: {
         auto& gpu = system.GPU();
         u64 fence = gpu.RequestFlush(addr, size);
         PushCommand(GPUTickCommand());
         while (fence > gpu.CurrentFlushRequestFence()) {
         }
+        break;
+    }
+    default:
+        UNIMPLEMENTED_MSG("Unsupported gpu_accuracy {}", Settings::values.gpu_accuracy.GetValue());
     }
 }
 
@@ -134,6 +146,12 @@ void ThreadManager::OnCommandListEnd() {
 u64 ThreadManager::PushCommand(CommandData&& command_data) {
     const u64 fence{++state.last_fence};
     state.queue.Push(CommandDataContainer(std::move(command_data), fence));
+
+    if (!is_async) {
+        // In synchronous GPU mode, block the caller until the command has executed
+        WaitIdle();
+    }
+
     return fence;
 }
 
diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h
index 0071195d6..2775629e7 100644
--- a/src/video_core/gpu_thread.h
+++ b/src/video_core/gpu_thread.h
@@ -27,7 +27,7 @@ class System;
 } // namespace Core
 
 namespace VideoCore {
-    class RendererBase;
+class RendererBase;
 } // namespace VideoCore
 
 namespace VideoCommon::GPUThread {
@@ -117,7 +117,7 @@ struct SynchState final {
 /// Class used to manage the GPU thread
 class ThreadManager final {
 public:
-    explicit ThreadManager(Core::System& system_);
+    explicit ThreadManager(Core::System& system_, bool is_async_);
     ~ThreadManager();
 
     /// Creates and starts the GPU thread.
@@ -155,6 +155,7 @@ private:
     Core::System& system;
     std::thread thread;
     std::thread::id thread_id;
+    const bool is_async;
 };
 
 } // namespace VideoCommon::GPUThread