video_core: refactor video frame and packet parsing

4 files changed, 34 insertions, 339 deletions

diff --git a/src/video_core/host1x/codecs/codec.cpp b/src/video_core/host1x/codecs/codec.cpp
index dbcf508e5..1030db681 100644
--- a/src/video_core/host1x/codecs/codec.cpp
+++ b/src/video_core/host1x/codecs/codec.cpp
@@ -1,11 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <algorithm>
-#include <fstream>
-#include <vector>
 #include "common/assert.h"
-#include "common/scope_exit.h"
 #include "common/settings.h"
 #include "video_core/host1x/codecs/codec.h"
 #include "video_core/host1x/codecs/h264.h"
@@ -14,242 +10,17 @@
 #include "video_core/host1x/host1x.h"
 #include "video_core/memory_manager.h"
 
-extern "C" {
-#include <libavfilter/buffersink.h>
-#include <libavfilter/buffersrc.h>
-#include <libavutil/opt.h>
-#ifdef LIBVA_FOUND
-// for querying VAAPI driver information
-#include <libavutil/hwcontext_vaapi.h>
-#endif
-}
-
 namespace Tegra {
-namespace {
-constexpr AVPixelFormat PREFERRED_GPU_FMT = AV_PIX_FMT_NV12;
-constexpr AVPixelFormat PREFERRED_CPU_FMT = AV_PIX_FMT_YUV420P;
-constexpr std::array PREFERRED_GPU_DECODERS = {
-    AV_HWDEVICE_TYPE_CUDA,
-#ifdef _WIN32
-    AV_HWDEVICE_TYPE_D3D11VA,
-    AV_HWDEVICE_TYPE_DXVA2,
-#elif defined(__unix__)
-    AV_HWDEVICE_TYPE_VAAPI,
-    AV_HWDEVICE_TYPE_VDPAU,
-#endif
-    // last resort for Linux Flatpak (w/ NVIDIA)
-    AV_HWDEVICE_TYPE_VULKAN,
-};
-
-void AVPacketDeleter(AVPacket* ptr) {
-    av_packet_free(&ptr);
-}
-
-using AVPacketPtr = std::unique_ptr<AVPacket, decltype(&AVPacketDeleter)>;
-
-AVPixelFormat GetGpuFormat(AVCodecContext* av_codec_ctx, const AVPixelFormat* pix_fmts) {
-    for (const AVPixelFormat* p = pix_fmts; *p != AV_PIX_FMT_NONE; ++p) {
-        if (*p == av_codec_ctx->pix_fmt) {
-            return av_codec_ctx->pix_fmt;
-        }
-    }
-    LOG_INFO(Service_NVDRV, "Could not find compatible GPU AV format, falling back to CPU");
-    av_buffer_unref(&av_codec_ctx->hw_device_ctx);
-    av_codec_ctx->pix_fmt = PREFERRED_CPU_FMT;
-    return PREFERRED_CPU_FMT;
-}
-
-// List all the currently available hwcontext in ffmpeg
-std::vector<AVHWDeviceType> ListSupportedContexts() {
-    std::vector<AVHWDeviceType> contexts{};
-    AVHWDeviceType current_device_type = AV_HWDEVICE_TYPE_NONE;
-    do {
-        current_device_type = av_hwdevice_iterate_types(current_device_type);
-        contexts.push_back(current_device_type);
-    } while (current_device_type != AV_HWDEVICE_TYPE_NONE);
-    return contexts;
-}
-
-} // namespace
-
-void AVFrameDeleter(AVFrame* ptr) {
-    av_frame_free(&ptr);
-}
 
 Codec::Codec(Host1x::Host1x& host1x_, const Host1x::NvdecCommon::NvdecRegisters& regs)
     : host1x(host1x_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(host1x)),
       vp8_decoder(std::make_unique<Decoder::VP8>(host1x)),
       vp9_decoder(std::make_unique<Decoder::VP9>(host1x)) {}
 
-Codec::~Codec() {
-    if (!initialized) {
-        return;
-    }
-    // Free libav memory
-    avcodec_free_context(&av_codec_ctx);
-    av_buffer_unref(&av_gpu_decoder);
-
-    if (filters_initialized) {
-        avfilter_graph_free(&av_filter_graph);
-    }
-}
-
-bool Codec::CreateGpuAvDevice() {
-    static constexpr auto HW_CONFIG_METHOD = AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX;
-    static const auto supported_contexts = ListSupportedContexts();
-    for (const auto& type : PREFERRED_GPU_DECODERS) {
-        if (std::none_of(supported_contexts.begin(), supported_contexts.end(),
-                         [&type](const auto& context) { return context == type; })) {
-            LOG_DEBUG(Service_NVDRV, "{} explicitly unsupported", av_hwdevice_get_type_name(type));
-            continue;
-        }
-        // Avoid memory leak from not cleaning up after av_hwdevice_ctx_create
-        av_buffer_unref(&av_gpu_decoder);
-        const int hwdevice_res = av_hwdevice_ctx_create(&av_gpu_decoder, type, nullptr, nullptr, 0);
-        if (hwdevice_res < 0) {
-            LOG_DEBUG(Service_NVDRV, "{} av_hwdevice_ctx_create failed {}",
-                      av_hwdevice_get_type_name(type), hwdevice_res);
-            continue;
-        }
-#ifdef LIBVA_FOUND
-        if (type == AV_HWDEVICE_TYPE_VAAPI) {
-            // we need to determine if this is an impersonated VAAPI driver
-            AVHWDeviceContext* hwctx =
-                static_cast<AVHWDeviceContext*>(static_cast<void*>(av_gpu_decoder->data));
-            AVVAAPIDeviceContext* vactx = static_cast<AVVAAPIDeviceContext*>(hwctx->hwctx);
-            const char* vendor_name = vaQueryVendorString(vactx->display);
-            if (strstr(vendor_name, "VDPAU backend")) {
-                // VDPAU impersonated VAAPI impl's are super buggy, we need to skip them
-                LOG_DEBUG(Service_NVDRV, "Skipping vdapu impersonated VAAPI driver");
-                continue;
-            } else {
-                // according to some user testing, certain vaapi driver (Intel?) could be buggy
-                // so let's log the driver name which may help the developers/supporters
-                LOG_DEBUG(Service_NVDRV, "Using VAAPI driver: {}", vendor_name);
-            }
-        }
-#endif
-        for (int i = 0;; i++) {
-            const AVCodecHWConfig* config = avcodec_get_hw_config(av_codec, i);
-            if (!config) {
-                LOG_DEBUG(Service_NVDRV, "{} decoder does not support device type {}.",
-                          av_codec->name, av_hwdevice_get_type_name(type));
-                break;
-            }
-            if ((config->methods & HW_CONFIG_METHOD) != 0 && config->device_type == type) {
-                LOG_INFO(Service_NVDRV, "Using {} GPU decoder", av_hwdevice_get_type_name(type));
-                av_codec_ctx->pix_fmt = config->pix_fmt;
-                return true;
-            }
-        }
-    }
-    return false;
-}
-
-void Codec::InitializeAvCodecContext() {
-    av_codec_ctx = avcodec_alloc_context3(av_codec);
-    av_opt_set(av_codec_ctx->priv_data, "tune", "zerolatency", 0);
-    av_codec_ctx->thread_count = 0;
-    av_codec_ctx->thread_type &= ~FF_THREAD_FRAME;
-}
-
-void Codec::InitializeGpuDecoder() {
-    if (!CreateGpuAvDevice()) {
-        av_buffer_unref(&av_gpu_decoder);
-        return;
-    }
-    auto* hw_device_ctx = av_buffer_ref(av_gpu_decoder);
-    ASSERT_MSG(hw_device_ctx, "av_buffer_ref failed");
-    av_codec_ctx->hw_device_ctx = hw_device_ctx;
-    av_codec_ctx->get_format = GetGpuFormat;
-}
-
-void Codec::InitializeAvFilters(AVFrame* frame) {
-    const AVFilter* buffer_src = avfilter_get_by_name("buffer");
-    const AVFilter* buffer_sink = avfilter_get_by_name("buffersink");
-    AVFilterInOut* inputs = avfilter_inout_alloc();
-    AVFilterInOut* outputs = avfilter_inout_alloc();
-    SCOPE_EXIT({
-        avfilter_inout_free(&inputs);
-        avfilter_inout_free(&outputs);
-    });
-
-    // Don't know how to get the accurate time_base but it doesn't matter for yadif filter
-    // so just use 1/1 to make buffer filter happy
-    std::string args = fmt::format("video_size={}x{}:pix_fmt={}:time_base=1/1", frame->width,
-                                   frame->height, frame->format);
-
-    av_filter_graph = avfilter_graph_alloc();
-    int ret = avfilter_graph_create_filter(&av_filter_src_ctx, buffer_src, "in", args.c_str(),
-                                           nullptr, av_filter_graph);
-    if (ret < 0) {
-        LOG_ERROR(Service_NVDRV, "avfilter_graph_create_filter source error: {}", ret);
-        return;
-    }
-
-    ret = avfilter_graph_create_filter(&av_filter_sink_ctx, buffer_sink, "out", nullptr, nullptr,
-                                       av_filter_graph);
-    if (ret < 0) {
-        LOG_ERROR(Service_NVDRV, "avfilter_graph_create_filter sink error: {}", ret);
-        return;
-    }
-
-    inputs->name = av_strdup("out");
-    inputs->filter_ctx = av_filter_sink_ctx;
-    inputs->pad_idx = 0;
-    inputs->next = nullptr;
-
-    outputs->name = av_strdup("in");
-    outputs->filter_ctx = av_filter_src_ctx;
-    outputs->pad_idx = 0;
-    outputs->next = nullptr;
-
-    const char* description = "yadif=1:-1:0";
-    ret = avfilter_graph_parse_ptr(av_filter_graph, description, &inputs, &outputs, nullptr);
-    if (ret < 0) {
-        LOG_ERROR(Service_NVDRV, "avfilter_graph_parse_ptr error: {}", ret);
-        return;
-    }
-
-    ret = avfilter_graph_config(av_filter_graph, nullptr);
-    if (ret < 0) {
-        LOG_ERROR(Service_NVDRV, "avfilter_graph_config error: {}", ret);
-        return;
-    }
-
-    filters_initialized = true;
-}
+Codec::~Codec() = default;
 
 void Codec::Initialize() {
-    const AVCodecID codec = [&] {
-        switch (current_codec) {
-        case Host1x::NvdecCommon::VideoCodec::H264:
-            return AV_CODEC_ID_H264;
-        case Host1x::NvdecCommon::VideoCodec::VP8:
-            return AV_CODEC_ID_VP8;
-        case Host1x::NvdecCommon::VideoCodec::VP9:
-            return AV_CODEC_ID_VP9;
-        default:
-            UNIMPLEMENTED_MSG("Unknown codec {}", current_codec);
-            return AV_CODEC_ID_NONE;
-        }
-    }();
-    av_codec = avcodec_find_decoder(codec);
-
-    InitializeAvCodecContext();
-    if (Settings::values.nvdec_emulation.GetValue() == Settings::NvdecEmulation::Gpu) {
-        InitializeGpuDecoder();
-    }
-    if (const int res = avcodec_open2(av_codec_ctx, av_codec, nullptr); res < 0) {
-        LOG_ERROR(Service_NVDRV, "avcodec_open2() Failed with result {}", res);
-        avcodec_free_context(&av_codec_ctx);
-        av_buffer_unref(&av_gpu_decoder);
-        return;
-    }
-    if (!av_codec_ctx->hw_device_ctx) {
-        LOG_INFO(Service_NVDRV, "Using FFmpeg software decoding");
-    }
-    initialized = true;
+    initialized = decode_api.Initialize(current_codec);
 }
 
 void Codec::SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec) {
@@ -264,14 +35,18 @@ void Codec::Decode() {
     if (is_first_frame) {
         Initialize();
     }
+
     if (!initialized) {
         return;
     }
+
+    // Assemble bitstream.
     bool vp9_hidden_frame = false;
-    const auto& frame_data = [&]() {
+    size_t configuration_size = 0;
+    const auto packet_data = [&]() {
         switch (current_codec) {
         case Tegra::Host1x::NvdecCommon::VideoCodec::H264:
-            return h264_decoder->ComposeFrame(state, is_first_frame);
+            return h264_decoder->ComposeFrame(state, &configuration_size, is_first_frame);
         case Tegra::Host1x::NvdecCommon::VideoCodec::VP8:
             return vp8_decoder->ComposeFrame(state);
         case Tegra::Host1x::NvdecCommon::VideoCodec::VP9:
@@ -283,89 +58,35 @@ void Codec::Decode() {
             return std::span<const u8>{};
         }
     }();
-    AVPacketPtr packet{av_packet_alloc(), AVPacketDeleter};
-    if (!packet) {
-        LOG_ERROR(Service_NVDRV, "av_packet_alloc failed");
-        return;
-    }
-    packet->data = const_cast<u8*>(frame_data.data());
-    packet->size = static_cast<s32>(frame_data.size());
-    if (const int res = avcodec_send_packet(av_codec_ctx, packet.get()); res != 0) {
-        LOG_DEBUG(Service_NVDRV, "avcodec_send_packet error {}", res);
+
+    // Send assembled bitstream to decoder.
+    if (!decode_api.SendPacket(packet_data, configuration_size)) {
         return;
     }
-    // Only receive/store visible frames
+
+    // Only receive/store visible frames.
     if (vp9_hidden_frame) {
         return;
     }
-    AVFramePtr initial_frame{av_frame_alloc(), AVFrameDeleter};
-    AVFramePtr final_frame{nullptr, AVFrameDeleter};
-    ASSERT_MSG(initial_frame, "av_frame_alloc initial_frame failed");
-    if (const int ret = avcodec_receive_frame(av_codec_ctx, initial_frame.get()); ret) {
-        LOG_DEBUG(Service_NVDRV, "avcodec_receive_frame error {}", ret);
-        return;
-    }
-    if (initial_frame->width == 0 || initial_frame->height == 0) {
-        LOG_WARNING(Service_NVDRV, "Zero width or height in frame");
-        return;
-    }
-    bool is_interlaced = initial_frame->interlaced_frame != 0;
-    if (av_codec_ctx->hw_device_ctx) {
-        final_frame = AVFramePtr{av_frame_alloc(), AVFrameDeleter};
-        ASSERT_MSG(final_frame, "av_frame_alloc final_frame failed");
-        // Can't use AV_PIX_FMT_YUV420P and share code with software decoding in vic.cpp
-        // because Intel drivers crash unless using AV_PIX_FMT_NV12
-        final_frame->format = PREFERRED_GPU_FMT;
-        const int ret = av_hwframe_transfer_data(final_frame.get(), initial_frame.get(), 0);
-        ASSERT_MSG(!ret, "av_hwframe_transfer_data error {}", ret);
-    } else {
-        final_frame = std::move(initial_frame);
-    }
-    if (final_frame->format != PREFERRED_CPU_FMT && final_frame->format != PREFERRED_GPU_FMT) {
-        UNIMPLEMENTED_MSG("Unexpected video format: {}", final_frame->format);
-        return;
-    }
-    if (!is_interlaced) {
-        av_frames.push(std::move(final_frame));
-    } else {
-        if (!filters_initialized) {
-            InitializeAvFilters(final_frame.get());
-        }
-        if (const int ret = av_buffersrc_add_frame_flags(av_filter_src_ctx, final_frame.get(),
-                                                         AV_BUFFERSRC_FLAG_KEEP_REF);
-            ret) {
-            LOG_DEBUG(Service_NVDRV, "av_buffersrc_add_frame_flags error {}", ret);
-            return;
-        }
-        while (true) {
-            auto filter_frame = AVFramePtr{av_frame_alloc(), AVFrameDeleter};
 
-            int ret = av_buffersink_get_frame(av_filter_sink_ctx, filter_frame.get());
+    // Receive output frames from decoder.
+    decode_api.ReceiveFrames(frames);
 
-            if (ret == AVERROR(EAGAIN) || ret == AVERROR(AVERROR_EOF))
-                break;
-            if (ret < 0) {
-                LOG_DEBUG(Service_NVDRV, "av_buffersink_get_frame error {}", ret);
-                return;
-            }
-
-            av_frames.push(std::move(filter_frame));
-        }
-    }
-    while (av_frames.size() > 10) {
-        LOG_TRACE(Service_NVDRV, "av_frames.push overflow dropped frame");
-        av_frames.pop();
+    while (frames.size() > 10) {
+        LOG_DEBUG(HW_GPU, "ReceiveFrames overflow, dropped frame");
+        frames.pop();
     }
 }
 
-AVFramePtr Codec::GetCurrentFrame() {
+std::unique_ptr<FFmpeg::Frame> Codec::GetCurrentFrame() {
     // Sometimes VIC will request more frames than have been decoded.
-    // in this case, return a nullptr and don't overwrite previous frame data
-    if (av_frames.empty()) {
-        return AVFramePtr{nullptr, AVFrameDeleter};
+    // in this case, return a blank frame and don't overwrite previous data.
+    if (frames.empty()) {
+        return {};
     }
-    AVFramePtr frame = std::move(av_frames.front());
-    av_frames.pop();
+
+    auto frame = std::move(frames.front());
+    frames.pop();
     return frame;
 }
 
diff --git a/src/video_core/host1x/codecs/codec.h b/src/video_core/host1x/codecs/codec.h
index 06fe00a4b..f700ae129 100644
--- a/src/video_core/host1x/codecs/codec.h
+++ b/src/video_core/host1x/codecs/codec.h
@@ -4,28 +4,15 @@
 #pragma once
 
 #include <memory>
+#include <optional>
 #include <string_view>
 #include <queue>
 #include "common/common_types.h"
+#include "video_core/host1x/ffmpeg/ffmpeg.h"
 #include "video_core/host1x/nvdec_common.h"
 
-extern "C" {
-#if defined(__GNUC__) || defined(__clang__)
-#pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Wconversion"
-#endif
-#include <libavcodec/avcodec.h>
-#include <libavfilter/avfilter.h>
-#if defined(__GNUC__) || defined(__clang__)
-#pragma GCC diagnostic pop
-#endif
-}
-
 namespace Tegra {
 
-void AVFrameDeleter(AVFrame* ptr);
-using AVFramePtr = std::unique_ptr<AVFrame, decltype(&AVFrameDeleter)>;
-
 namespace Decoder {
 class H264;
 class VP8;
@@ -51,7 +38,7 @@ public:
     void Decode();
 
     /// Returns next decoded frame
-    [[nodiscard]] AVFramePtr GetCurrentFrame();
+    [[nodiscard]] std::unique_ptr<FFmpeg::Frame> GetCurrentFrame();
 
     /// Returns the value of current_codec
     [[nodiscard]] Host1x::NvdecCommon::VideoCodec GetCurrentCodec() const;
@@ -60,25 +47,9 @@ public:
     [[nodiscard]] std::string_view GetCurrentCodecName() const;
 
 private:
-    void InitializeAvCodecContext();
-
-    void InitializeAvFilters(AVFrame* frame);
-
-    void InitializeGpuDecoder();
-
-    bool CreateGpuAvDevice();
-
     bool initialized{};
-    bool filters_initialized{};
     Host1x::NvdecCommon::VideoCodec current_codec{Host1x::NvdecCommon::VideoCodec::None};
-
-    const AVCodec* av_codec{nullptr};
-    AVCodecContext* av_codec_ctx{nullptr};
-    AVBufferRef* av_gpu_decoder{nullptr};
-
-    AVFilterContext* av_filter_src_ctx{nullptr};
-    AVFilterContext* av_filter_sink_ctx{nullptr};
-    AVFilterGraph* av_filter_graph{nullptr};
+    FFmpeg::DecodeApi decode_api;
 
     Host1x::Host1x& host1x;
     const Host1x::NvdecCommon::NvdecRegisters& state;
@@ -86,7 +57,7 @@ private:
     std::unique_ptr<Decoder::VP8> vp8_decoder;
     std::unique_ptr<Decoder::VP9> vp9_decoder;
 
-    std::queue<AVFramePtr> av_frames{};
+    std::queue<std::unique_ptr<FFmpeg::Frame>> frames{};
 };
 
 } // namespace Tegra
diff --git a/src/video_core/host1x/codecs/h264.cpp b/src/video_core/host1x/codecs/h264.cpp
index ece79b1e2..309a7f1d5 100644
--- a/src/video_core/host1x/codecs/h264.cpp
+++ b/src/video_core/host1x/codecs/h264.cpp
@@ -30,7 +30,7 @@ H264::H264(Host1x::Host1x& host1x_) : host1x{host1x_} {}
 H264::~H264() = default;
 
 std::span<const u8> H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state,
-                                       bool is_first_frame) {
+                                       size_t* out_configuration_size, bool is_first_frame) {
     H264DecoderContext context;
     host1x.MemoryManager().ReadBlock(state.picture_info_offset, &context,
                                      sizeof(H264DecoderContext));
@@ -39,6 +39,7 @@ std::span<const u8> H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters
     if (!is_first_frame && frame_number != 0) {
         frame.resize_destructive(context.stream_len);
         host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size());
+        *out_configuration_size = 0;
         return frame;
     }
 
@@ -157,6 +158,7 @@ std::span<const u8> H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters
     frame.resize(encoded_header.size() + context.stream_len);
     std::memcpy(frame.data(), encoded_header.data(), encoded_header.size());
 
+    *out_configuration_size = encoded_header.size();
     host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset,
                                      frame.data() + encoded_header.size(), context.stream_len);
 
diff --git a/src/video_core/host1x/codecs/h264.h b/src/video_core/host1x/codecs/h264.h
index d6b556322..1deaf4632 100644
--- a/src/video_core/host1x/codecs/h264.h
+++ b/src/video_core/host1x/codecs/h264.h
@@ -67,6 +67,7 @@ public:
 
     /// Compose the H264 frame for FFmpeg decoding
     [[nodiscard]] std::span<const u8> ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state,
+                                                   size_t* out_configuration_size,
                                                    bool is_first_frame = false);
 
 private: