Merge pull request #7326 from ameerj/vp8

codecs: Implement VP8 video decoding support

CMakeLists.txt

@@ -600,6 +600,7 @@ if (YUZU_USE_BUNDLED_FFMPEG)
                 ${LIBVA_LIBRARIES})
             set(FFmpeg_HWACCEL_FLAGS
                 --enable-hwaccel=h264_vaapi
+                --enable-hwaccel=vp8_vaapi
                 --enable-hwaccel=vp9_vaapi
                 --enable-libdrm)
             list(APPEND FFmpeg_HWACCEL_INCLUDE_DIRS
@@ -620,6 +621,7 @@ if (YUZU_USE_BUNDLED_FFMPEG)
                 --enable-ffnvcodec
                 --enable-nvdec
                 --enable-hwaccel=h264_nvdec
+                --enable-hwaccel=vp8_nvdec
                 --enable-hwaccel=vp9_nvdec
                 --extra-cflags=-I${CUDA_INCLUDE_DIRS}
             )
@@ -670,6 +672,7 @@ if (YUZU_USE_BUNDLED_FFMPEG)
                     --disable-postproc
                     --disable-swresample
                     --enable-decoder=h264
+                    --enable-decoder=vp8
                     --enable-decoder=vp9
                     --cc="${CMAKE_C_COMPILER}"
                     --cxx="${CMAKE_CXX_COMPILER}"

src/video_core/CMakeLists.txt

@@ -15,6 +15,8 @@ add_library(video_core STATIC
     command_classes/codecs/codec.h
     command_classes/codecs/h264.cpp
     command_classes/codecs/h264.h
+    command_classes/codecs/vp8.cpp
+    command_classes/codecs/vp8.h
     command_classes/codecs/vp9.cpp
     command_classes/codecs/vp9.h
     command_classes/codecs/vp9_types.h

src/video_core/command_classes/codecs/codec.cpp

@@ -8,6 +8,7 @@
 #include "common/settings.h"
 #include "video_core/command_classes/codecs/codec.h"
 #include "video_core/command_classes/codecs/h264.h"
+#include "video_core/command_classes/codecs/vp8.h"
 #include "video_core/command_classes/codecs/vp9.h"
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
@@ -46,6 +47,7 @@ void AVFrameDeleter(AVFrame* ptr) {
 
 Codec::Codec(GPU& gpu_, const NvdecCommon::NvdecRegisters& regs)
     : gpu(gpu_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(gpu)),
+      vp8_decoder(std::make_unique<Decoder::VP8>(gpu)),
       vp9_decoder(std::make_unique<Decoder::VP9>(gpu)) {}
 
 Codec::~Codec() {
@@ -135,7 +137,9 @@ void Codec::Initialize() {
         switch (current_codec) {
         case NvdecCommon::VideoCodec::H264:
             return AV_CODEC_ID_H264;
-        case NvdecCommon::VideoCodec::Vp9:
+        case NvdecCommon::VideoCodec::VP8:
+            return AV_CODEC_ID_VP8;
+        case NvdecCommon::VideoCodec::VP9:
             return AV_CODEC_ID_VP9;
         default:
             UNIMPLEMENTED_MSG("Unknown codec {}", current_codec);
@@ -176,19 +180,27 @@ void Codec::Decode() {
         return;
     }
     bool vp9_hidden_frame = false;
-    std::vector<u8> frame_data;
+    const auto& frame_data = [&]() {
-    if (current_codec == NvdecCommon::VideoCodec::H264) {
+        switch (current_codec) {
-        frame_data = h264_decoder->ComposeFrameHeader(state, is_first_frame);
+        case Tegra::NvdecCommon::VideoCodec::H264:
-    } else if (current_codec == NvdecCommon::VideoCodec::Vp9) {
+            return h264_decoder->ComposeFrame(state, is_first_frame);
-        frame_data = vp9_decoder->ComposeFrameHeader(state);
+        case Tegra::NvdecCommon::VideoCodec::VP8:
-        vp9_hidden_frame = vp9_decoder->WasFrameHidden();
+            return vp8_decoder->ComposeFrame(state);
-    }
+        case Tegra::NvdecCommon::VideoCodec::VP9:
+            vp9_decoder->ComposeFrame(state);
+            vp9_hidden_frame = vp9_decoder->WasFrameHidden();
+            return vp9_decoder->GetFrameBytes();
+        default:
+            UNREACHABLE();
+            return std::vector<u8>{};
+        }
+    }();
     AVPacketPtr packet{av_packet_alloc(), AVPacketDeleter};
     if (!packet) {
         LOG_ERROR(Service_NVDRV, "av_packet_alloc failed");
         return;
     }
-    packet->data = frame_data.data();
+    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);
@@ -252,11 +264,11 @@ std::string_view Codec::GetCurrentCodecName() const {
         return "None";
     case NvdecCommon::VideoCodec::H264:
         return "H264";
-    case NvdecCommon::VideoCodec::Vp8:
+    case NvdecCommon::VideoCodec::VP8:
         return "VP8";
     case NvdecCommon::VideoCodec::H265:
         return "H265";
-    case NvdecCommon::VideoCodec::Vp9:
+    case NvdecCommon::VideoCodec::VP9:
         return "VP9";
     default:
         return "Unknown";

src/video_core/command_classes/codecs/codec.h

@@ -29,6 +29,7 @@ using AVFramePtr = std::unique_ptr<AVFrame, decltype(&AVFrameDeleter)>;
 
 namespace Decoder {
 class H264;
+class VP8;
 class VP9;
 } // namespace Decoder
 
@@ -72,6 +73,7 @@ private:
     GPU& gpu;
     const NvdecCommon::NvdecRegisters& state;
     std::unique_ptr<Decoder::H264> h264_decoder;
+    std::unique_ptr<Decoder::VP8> vp8_decoder;
     std::unique_ptr<Decoder::VP9> vp9_decoder;
 
     std::queue<AVFramePtr> av_frames{};

src/video_core/command_classes/codecs/h264.cpp

@@ -45,8 +45,8 @@ H264::H264(GPU& gpu_) : gpu(gpu_) {}
 
 H264::~H264() = default;
 
-const std::vector<u8>& H264::ComposeFrameHeader(const NvdecCommon::NvdecRegisters& state,
+const std::vector<u8>& H264::ComposeFrame(const NvdecCommon::NvdecRegisters& state,
-                                                bool is_first_frame) {
+                                          bool is_first_frame) {
     H264DecoderContext context;
     gpu.MemoryManager().ReadBlock(state.picture_info_offset, &context, sizeof(H264DecoderContext));
 

src/video_core/command_classes/codecs/h264.h

@@ -75,9 +75,9 @@ public:
     explicit H264(GPU& gpu);
     ~H264();
 
-    /// Compose the H264 header of the frame for FFmpeg decoding
+    /// Compose the H264 frame for FFmpeg decoding
-    [[nodiscard]] const std::vector<u8>& ComposeFrameHeader(
+    [[nodiscard]] const std::vector<u8>& ComposeFrame(const NvdecCommon::NvdecRegisters& state,
-        const NvdecCommon::NvdecRegisters& state, bool is_first_frame = false);
+                                                      bool is_first_frame = false);
 
 private:
     std::vector<u8> frame;

src/video_core/command_classes/codecs/vp8.cpp

@@ -0,0 +1,55 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <array>
+#include <vector>
+
+#include "video_core/command_classes/codecs/vp8.h"
+#include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
+
+namespace Tegra::Decoder {
+VP8::VP8(GPU& gpu_) : gpu(gpu_) {}
+
+VP8::~VP8() = default;
+
+const std::vector<u8>& VP8::ComposeFrame(const NvdecCommon::NvdecRegisters& state) {
+    VP8PictureInfo info;
+    gpu.MemoryManager().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo));
+
+    const bool is_key_frame = info.key_frame == 1u;
+    const auto bitstream_size = static_cast<size_t>(info.vld_buffer_size);
+    const size_t header_size = is_key_frame ? 10u : 3u;
+    frame.resize(header_size + bitstream_size);
+
+    // Based on page 30 of the VP8 specification.
+    // https://datatracker.ietf.org/doc/rfc6386/
+    frame[0] = is_key_frame ? 0u : 1u; // 1-bit frame type (0: keyframe, 1: interframes).
+    frame[0] |= static_cast<u8>((info.version & 7u) << 1u); // 3-bit version number
+    frame[0] |= static_cast<u8>(1u << 4u);                  // 1-bit show_frame flag
+
+    // The next 19-bits are the first partition size
+    frame[0] |= static_cast<u8>((info.first_part_size & 7u) << 5u);
+    frame[1] = static_cast<u8>((info.first_part_size & 0x7f8u) >> 3u);
+    frame[2] = static_cast<u8>((info.first_part_size & 0x7f800u) >> 11u);
+
+    if (is_key_frame) {
+        frame[3] = 0x9du;
+        frame[4] = 0x01u;
+        frame[5] = 0x2au;
+        // TODO(ameerj): Horizontal/Vertical Scale
+        // 16 bits: (2 bits Horizontal Scale << 14) | Width (14 bits)
+        frame[6] = static_cast<u8>(info.frame_width & 0xff);
+        frame[7] = static_cast<u8>(((info.frame_width >> 8) & 0x3f));
+        // 16 bits:(2 bits Vertical Scale << 14) | Height (14 bits)
+        frame[8] = static_cast<u8>(info.frame_height & 0xff);
+        frame[9] = static_cast<u8>(((info.frame_height >> 8) & 0x3f));
+    }
+    const u64 bitstream_offset = state.frame_bitstream_offset;
+    gpu.MemoryManager().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size);
+
+    return frame;
+}
+
+} // namespace Tegra::Decoder

src/video_core/command_classes/codecs/vp8.h

@@ -0,0 +1,74 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <vector>
+
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "video_core/command_classes/nvdec_common.h"
+
+namespace Tegra {
+class GPU;
+namespace Decoder {
+
+class VP8 {
+public:
+    explicit VP8(GPU& gpu);
+    ~VP8();
+
+    /// Compose the VP8 frame for FFmpeg decoding
+    [[nodiscard]] const std::vector<u8>& ComposeFrame(const NvdecCommon::NvdecRegisters& state);
+
+private:
+    std::vector<u8> frame;
+    GPU& gpu;
+
+    struct VP8PictureInfo {
+        INSERT_PADDING_WORDS_NOINIT(14);
+        u16 frame_width;  // actual frame width
+        u16 frame_height; // actual frame height
+        u8 key_frame;
+        u8 version;
+        union {
+            u8 raw;
+            BitField<0, 2, u8> tile_format;
+            BitField<2, 3, u8> gob_height;
+            BitField<5, 3, u8> reserverd_surface_format;
+        };
+        u8 error_conceal_on;  // 1: error conceal on; 0: off
+        u32 first_part_size;  // the size of first partition(frame header and mb header partition)
+        u32 hist_buffer_size; // in units of 256
+        u32 vld_buffer_size;  // in units of 1
+        // Current frame buffers
+        std::array<u32, 2> frame_stride; // [y_c]
+        u32 luma_top_offset;             // offset of luma top field in units of 256
+        u32 luma_bot_offset;             // offset of luma bottom field in units of 256
+        u32 luma_frame_offset;           // offset of luma frame in units of 256
+        u32 chroma_top_offset;           // offset of chroma top field in units of 256
+        u32 chroma_bot_offset;           // offset of chroma bottom field in units of 256
+        u32 chroma_frame_offset;         // offset of chroma frame in units of 256
+
+        INSERT_PADDING_BYTES_NOINIT(0x1c); // NvdecDisplayParams
+
+        // Decode picture buffer related
+        s8 current_output_memory_layout;
+        // output NV12/NV24 setting. index 0: golden; 1: altref; 2: last
+        std::array<s8, 3> output_memory_layout;
+
+        u8 segmentation_feature_data_update;
+        INSERT_PADDING_BYTES_NOINIT(3);
+
+        // ucode return result
+        u32 result_value;
+        std::array<u32, 8> partition_offset;
+        INSERT_PADDING_WORDS_NOINIT(3);
+    };
+    static_assert(sizeof(VP8PictureInfo) == 0xc0, "PictureInfo is an invalid size");
+};
+
+} // namespace Decoder
+} // namespace Tegra

src/video_core/command_classes/codecs/vp9.cpp

@@ -770,7 +770,7 @@ VpxBitStreamWriter VP9::ComposeUncompressedHeader() {
     return uncomp_writer;
 }
 
-const std::vector<u8>& VP9::ComposeFrameHeader(const NvdecCommon::NvdecRegisters& state) {
+void VP9::ComposeFrame(const NvdecCommon::NvdecRegisters& state) {
     std::vector<u8> bitstream;
     {
         Vp9FrameContainer curr_frame = GetCurrentFrame(state);
@@ -792,7 +792,6 @@ const std::vector<u8>& VP9::ComposeFrameHeader(const NvdecCommon::NvdecRegisters
               frame.begin() + uncompressed_header.size());
     std::copy(bitstream.begin(), bitstream.end(),
               frame.begin() + uncompressed_header.size() + compressed_header.size());
-    return frame;
 }
 
 VpxRangeEncoder::VpxRangeEncoder() {

src/video_core/command_classes/codecs/vp9.h

@@ -116,16 +116,20 @@ public:
     VP9(VP9&&) = default;
     VP9& operator=(VP9&&) = delete;
 
-    /// Composes the VP9 frame from the GPU state information. Based on the official VP9 spec
+    /// Composes the VP9 frame from the GPU state information.
-    /// documentation
+    /// Based on the official VP9 spec documentation
-    [[nodiscard]] const std::vector<u8>& ComposeFrameHeader(
+    void ComposeFrame(const NvdecCommon::NvdecRegisters& state);
-        const NvdecCommon::NvdecRegisters& state);
 
     /// Returns true if the most recent frame was a hidden frame.
     [[nodiscard]] bool WasFrameHidden() const {
         return !current_frame_info.show_frame;
     }
 
+    /// Returns a const reference to the composed frame data.
+    [[nodiscard]] const std::vector<u8>& GetFrameBytes() const {
+        return frame;
+    }
+
 private:
     /// Generates compressed header probability updates in the bitstream writer
     template <typename T, std::size_t N>

src/video_core/command_classes/nvdec.cpp

@@ -35,7 +35,8 @@ AVFramePtr Nvdec::GetFrame() {
 void Nvdec::Execute() {
     switch (codec->GetCurrentCodec()) {
     case NvdecCommon::VideoCodec::H264:
-    case NvdecCommon::VideoCodec::Vp9:
+    case NvdecCommon::VideoCodec::VP8:
+    case NvdecCommon::VideoCodec::VP9:
         codec->Decode();
         break;
     default:

src/video_core/command_classes/nvdec_common.h

@@ -13,9 +13,9 @@ namespace Tegra::NvdecCommon {
 enum class VideoCodec : u64 {
     None = 0x0,
     H264 = 0x3,
-    Vp8 = 0x5,
+    VP8 = 0x5,
     H265 = 0x7,
-    Vp9 = 0x9,
+    VP9 = 0x9,
 };
 
 // NVDEC should use a 32-bit address space, but is mapped to 64-bit,
@@ -50,7 +50,10 @@ struct NvdecRegisters {
             u64 h264_last_surface_chroma_offset;       ///< 0x0858
             std::array<u64, 17> surface_luma_offset;   ///< 0x0860
             std::array<u64, 17> surface_chroma_offset; ///< 0x08E8
-            INSERT_PADDING_WORDS_NOINIT(132);          ///< 0x0970
+            INSERT_PADDING_WORDS_NOINIT(68);           ///< 0x0970
+            u64 vp8_prob_data_offset;                  ///< 0x0A80
+            u64 vp8_header_partition_buf_offset;       ///< 0x0A88
+            INSERT_PADDING_WORDS_NOINIT(60);           ///< 0x0A90
             u64 vp9_entropy_probs_offset;              ///< 0x0B80
             u64 vp9_backward_updates_offset;           ///< 0x0B88
             u64 vp9_last_frame_segmap_offset;          ///< 0x0B90
@@ -81,6 +84,8 @@ ASSERT_REG_POSITION(h264_last_surface_luma_offset, 0x10A);
 ASSERT_REG_POSITION(h264_last_surface_chroma_offset, 0x10B);
 ASSERT_REG_POSITION(surface_luma_offset, 0x10C);
 ASSERT_REG_POSITION(surface_chroma_offset, 0x11D);
+ASSERT_REG_POSITION(vp8_prob_data_offset, 0x150);
+ASSERT_REG_POSITION(vp8_header_partition_buf_offset, 0x151);
 ASSERT_REG_POSITION(vp9_entropy_probs_offset, 0x170);
 ASSERT_REG_POSITION(vp9_backward_updates_offset, 0x171);
 ASSERT_REG_POSITION(vp9_last_frame_segmap_offset, 0x172);