nvdec cleanup

src/core/hle/service/nvdrv/devices/nvhost_vic.cpp

@@ -28,8 +28,14 @@ NvResult nvhost_vic::Ioctl1(Ioctl command, const std::vector<u8>& input, std::ve
             return GetWaitbase(input, output);
         case 0x9:
             return MapBuffer(input, output);
-        case 0xa:
+        case 0xa: {
+            if (command.length == 0x1c) {
+                Tegra::ChCommandHeaderList cmdlist(1);
+                cmdlist[0] = Tegra::ChCommandHeader{0xDEADB33F};
+                system.GPU().PushCommandBuffer(cmdlist);
+            }
             return UnmapBuffer(input, output);
+        }
         default:
             break;
         }

src/video_core/cdma_pusher.cpp

@@ -37,20 +37,7 @@ CDmaPusher::CDmaPusher(GPU& gpu_)
 
 CDmaPusher::~CDmaPusher() = default;
 
-void CDmaPusher::Push(ChCommandHeaderList&& entries) {
-    cdma_queue.push(std::move(entries));
-}
-
-void CDmaPusher::DispatchCalls() {
-    while (!cdma_queue.empty()) {
-        Step();
-    }
-}
-
-void CDmaPusher::Step() {
-    const auto entries{cdma_queue.front()};
-    cdma_queue.pop();
-
+void CDmaPusher::ProcessEntries(ChCommandHeaderList&& entries) {
     std::vector<u32> values(entries.size());
     std::memcpy(values.data(), entries.data(), entries.size() * sizeof(u32));
 

src/video_core/cdma_pusher.h

@@ -99,19 +99,13 @@ public:
     explicit CDmaPusher(GPU& gpu_);
     ~CDmaPusher();
 
-    /// Push NVDEC command buffer entries into queue
-    void Push(ChCommandHeaderList&& entries);
-
-    /// Process queued command buffer entries
-    void DispatchCalls();
-
-    /// Process one queue element
-    void Step();
+    /// Process the command entry
+    void ProcessEntries(ChCommandHeaderList&& entries);
 
+private:
     /// Invoke command class devices to execute the command based on the current state
     void ExecuteCommand(u32 state_offset, u32 data);
 
-private:
     /// Write arguments value to the ThiRegisters member at the specified offset
     void ThiStateWrite(ThiRegisters& state, u32 state_offset, const std::vector<u32>& arguments);
 
@@ -128,9 +122,6 @@ private:
     s32 offset{};
     u32 mask{};
     bool incrementing{};
-
-    // Queue of command lists to be processed
-    std::queue<ChCommandHeaderList> cdma_queue;
 };
 
 } // namespace Tegra

src/video_core/command_classes/codecs/codec.cpp

@@ -44,9 +44,11 @@ Codec::~Codec() {
 }
 
 void Codec::SetTargetCodec(NvdecCommon::VideoCodec codec) {
-    LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", codec);
+    if (current_codec != codec) {
+        LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", static_cast<u32>(codec));
         current_codec = codec;
     }
+}
 
 void Codec::StateWrite(u32 offset, u64 arguments) {
     u8* const state_offset = reinterpret_cast<u8*>(&state) + offset * sizeof(u64);
@@ -55,7 +57,6 @@ void Codec::StateWrite(u32 offset, u64 arguments) {
 
 void Codec::Decode() {
     bool is_first_frame = false;
-
     if (!initialized) {
         if (current_codec == NvdecCommon::VideoCodec::H264) {
             av_codec = avcodec_find_decoder(AV_CODEC_ID_H264);

src/video_core/command_classes/vic.cpp

@@ -18,7 +18,10 @@ extern "C" {
 namespace Tegra {
 
 Vic::Vic(GPU& gpu_, std::shared_ptr<Nvdec> nvdec_processor_)
-    : gpu(gpu_), nvdec_processor(std::move(nvdec_processor_)) {}
+    : gpu(gpu_),
+      nvdec_processor(std::move(nvdec_processor_)), converted_frame_buffer{nullptr, av_free}
+
+{}
 Vic::~Vic() = default;
 
 void Vic::VicStateWrite(u32 offset, u32 arguments) {
@@ -89,8 +92,10 @@ void Vic::Execute() {
         // Get Converted frame
         const std::size_t linear_size = frame->width * frame->height * 4;
 
-        using AVMallocPtr = std::unique_ptr<u8, decltype(&av_free)>;
-        AVMallocPtr converted_frame_buffer{static_cast<u8*>(av_malloc(linear_size)), av_free};
+        // Only allocate frame_buffer once per stream, as the size is not expected to change
+        if (!converted_frame_buffer) {
+            converted_frame_buffer = AVMallocPtr{static_cast<u8*>(av_malloc(linear_size)), av_free};
+        }
 
         const int converted_stride{frame->width * 4};
         u8* const converted_frame_buf_addr{converted_frame_buffer.get()};
@@ -104,12 +109,12 @@ void Vic::Execute() {
             const u32 block_height = static_cast<u32>(config.block_linear_height_log2);
             const auto size = Tegra::Texture::CalculateSize(true, 4, frame->width, frame->height, 1,
                                                             block_height, 0);
-            std::vector<u8> swizzled_data(size);
+            luma_buffer.resize(size);
             Tegra::Texture::SwizzleSubrect(frame->width, frame->height, frame->width * 4,
-                                           frame->width, 4, swizzled_data.data(),
+                                           frame->width, 4, luma_buffer.data(),
                                            converted_frame_buffer.get(), block_height, 0, 0);
 
-            gpu.MemoryManager().WriteBlock(output_surface_luma_address, swizzled_data.data(), size);
+            gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size);
         } else {
             // send pitch linear frame
             gpu.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr,
@@ -132,8 +137,8 @@ void Vic::Execute() {
         const auto stride = frame->linesize[0];
         const auto half_stride = frame->linesize[1];
 
-        std::vector<u8> luma_buffer(aligned_width * surface_height);
-        std::vector<u8> chroma_buffer(aligned_width * half_height);
+        luma_buffer.resize(aligned_width * surface_height);
+        chroma_buffer.resize(aligned_width * half_height);
 
         // Populate luma buffer
         for (std::size_t y = 0; y < surface_height - 1; ++y) {

src/video_core/command_classes/vic.h

@@ -97,6 +97,13 @@ private:
     GPU& gpu;
     std::shared_ptr<Tegra::Nvdec> nvdec_processor;
 
+    /// Avoid reallocation of the following buffers every frame, as their
+    /// size does not change during a stream
+    using AVMallocPtr = std::unique_ptr<u8, decltype(&av_free)>;
+    AVMallocPtr converted_frame_buffer;
+    std::vector<u8> luma_buffer;
+    std::vector<u8> chroma_buffer;
+
     GPUVAddr config_struct_address{};
     GPUVAddr output_surface_luma_address{};
     GPUVAddr output_surface_chroma_u_address{};

src/video_core/gpu.cpp

@@ -30,8 +30,7 @@ MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
 
 GPU::GPU(Core::System& system_, bool is_async_, bool use_nvdec_)
     : system{system_}, memory_manager{std::make_unique<Tegra::MemoryManager>(system)},
-      dma_pusher{std::make_unique<Tegra::DmaPusher>(system, *this)},
-      cdma_pusher{std::make_unique<Tegra::CDmaPusher>(*this)}, use_nvdec{use_nvdec_},
+      dma_pusher{std::make_unique<Tegra::DmaPusher>(system, *this)}, use_nvdec{use_nvdec_},
       maxwell_3d{std::make_unique<Engines::Maxwell3D>(system, *memory_manager)},
       fermi_2d{std::make_unique<Engines::Fermi2D>()},
       kepler_compute{std::make_unique<Engines::KeplerCompute>(system, *memory_manager)},

src/video_core/gpu_thread.cpp

@@ -48,9 +48,8 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer,
             dma_pusher.DispatchCalls();
         } else if (auto* command_list = std::get_if<SubmitChCommandEntries>(&next.data)) {
             // NVDEC
-            cdma_pusher.Push(std::move(command_list->entries));
-            cdma_pusher.DispatchCalls();
-        } else if (const auto* data = std::get_if<SwapBuffersCommand>(&next.data)) {
+            cdma_pusher.ProcessEntries(std::move(command_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<OnCommandListEndCommand>(next.data)) {
             rasterizer->ReleaseFences();