Merge pull request #12045 from liamwhite/codec-refactor

video_core: refactor video frame and packet parsing
This commit is contained in:
liamwhite 2023-11-21 09:19:26 -05:00 committed by GitHub
commit e69118042f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
11 changed files with 705 additions and 374 deletions

View File

@ -4,7 +4,7 @@
add_subdirectory(host_shaders) add_subdirectory(host_shaders)
if(LIBVA_FOUND) if(LIBVA_FOUND)
set_source_files_properties(host1x/codecs/codec.cpp set_source_files_properties(host1x/ffmpeg/ffmpeg.cpp
PROPERTIES COMPILE_DEFINITIONS LIBVA_FOUND=1) PROPERTIES COMPILE_DEFINITIONS LIBVA_FOUND=1)
list(APPEND FFmpeg_LIBRARIES ${LIBVA_LIBRARIES}) list(APPEND FFmpeg_LIBRARIES ${LIBVA_LIBRARIES})
endif() endif()
@ -66,6 +66,8 @@ add_library(video_core STATIC
host1x/codecs/vp9.cpp host1x/codecs/vp9.cpp
host1x/codecs/vp9.h host1x/codecs/vp9.h
host1x/codecs/vp9_types.h host1x/codecs/vp9_types.h
host1x/ffmpeg/ffmpeg.cpp
host1x/ffmpeg/ffmpeg.h
host1x/control.cpp host1x/control.cpp
host1x/control.h host1x/control.h
host1x/host1x.cpp host1x/host1x.cpp

View File

@ -1,11 +1,7 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later // SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <fstream>
#include <vector>
#include "common/assert.h" #include "common/assert.h"
#include "common/scope_exit.h"
#include "common/settings.h" #include "common/settings.h"
#include "video_core/host1x/codecs/codec.h" #include "video_core/host1x/codecs/codec.h"
#include "video_core/host1x/codecs/h264.h" #include "video_core/host1x/codecs/h264.h"
@ -14,242 +10,17 @@
#include "video_core/host1x/host1x.h" #include "video_core/host1x/host1x.h"
#include "video_core/memory_manager.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 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) Codec::Codec(Host1x::Host1x& host1x_, const Host1x::NvdecCommon::NvdecRegisters& regs)
: host1x(host1x_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(host1x)), : host1x(host1x_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(host1x)),
vp8_decoder(std::make_unique<Decoder::VP8>(host1x)), vp8_decoder(std::make_unique<Decoder::VP8>(host1x)),
vp9_decoder(std::make_unique<Decoder::VP9>(host1x)) {} vp9_decoder(std::make_unique<Decoder::VP9>(host1x)) {}
Codec::~Codec() { Codec::~Codec() = default;
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;
}
void Codec::Initialize() { void Codec::Initialize() {
const AVCodecID codec = [&] { initialized = decode_api.Initialize(current_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;
} }
void Codec::SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec) { void Codec::SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec) {
@ -264,14 +35,18 @@ void Codec::Decode() {
if (is_first_frame) { if (is_first_frame) {
Initialize(); Initialize();
} }
if (!initialized) { if (!initialized) {
return; return;
} }
// Assemble bitstream.
bool vp9_hidden_frame = false; bool vp9_hidden_frame = false;
const auto& frame_data = [&]() { size_t configuration_size = 0;
const auto packet_data = [&]() {
switch (current_codec) { switch (current_codec) {
case Tegra::Host1x::NvdecCommon::VideoCodec::H264: 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: case Tegra::Host1x::NvdecCommon::VideoCodec::VP8:
return vp8_decoder->ComposeFrame(state); return vp8_decoder->ComposeFrame(state);
case Tegra::Host1x::NvdecCommon::VideoCodec::VP9: case Tegra::Host1x::NvdecCommon::VideoCodec::VP9:
@ -283,89 +58,35 @@ void Codec::Decode() {
return std::span<const u8>{}; return std::span<const u8>{};
} }
}(); }();
AVPacketPtr packet{av_packet_alloc(), AVPacketDeleter};
if (!packet) { // Send assembled bitstream to decoder.
LOG_ERROR(Service_NVDRV, "av_packet_alloc failed"); if (!decode_api.SendPacket(packet_data, configuration_size)) {
return; return;
} }
packet->data = const_cast<u8*>(frame_data.data());
packet->size = static_cast<s32>(frame_data.size()); // Only receive/store visible frames.
if (const int res = avcodec_send_packet(av_codec_ctx, packet.get()); res != 0) {
LOG_DEBUG(Service_NVDRV, "avcodec_send_packet error {}", res);
return;
}
// Only receive/store visible frames
if (vp9_hidden_frame) { if (vp9_hidden_frame) {
return; 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)) while (frames.size() > 10) {
break; LOG_DEBUG(HW_GPU, "ReceiveFrames overflow, dropped frame");
if (ret < 0) { frames.pop();
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();
} }
} }
AVFramePtr Codec::GetCurrentFrame() { std::unique_ptr<FFmpeg::Frame> Codec::GetCurrentFrame() {
// Sometimes VIC will request more frames than have been decoded. // Sometimes VIC will request more frames than have been decoded.
// in this case, return a nullptr and don't overwrite previous frame data // in this case, return a blank frame and don't overwrite previous data.
if (av_frames.empty()) { if (frames.empty()) {
return AVFramePtr{nullptr, AVFrameDeleter}; return {};
} }
AVFramePtr frame = std::move(av_frames.front());
av_frames.pop(); auto frame = std::move(frames.front());
frames.pop();
return frame; return frame;
} }

View File

@ -4,28 +4,15 @@
#pragma once #pragma once
#include <memory> #include <memory>
#include <optional>
#include <string_view> #include <string_view>
#include <queue> #include <queue>
#include "common/common_types.h" #include "common/common_types.h"
#include "video_core/host1x/ffmpeg/ffmpeg.h"
#include "video_core/host1x/nvdec_common.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 { namespace Tegra {
void AVFrameDeleter(AVFrame* ptr);
using AVFramePtr = std::unique_ptr<AVFrame, decltype(&AVFrameDeleter)>;
namespace Decoder { namespace Decoder {
class H264; class H264;
class VP8; class VP8;
@ -51,7 +38,7 @@ public:
void Decode(); void Decode();
/// Returns next decoded frame /// Returns next decoded frame
[[nodiscard]] AVFramePtr GetCurrentFrame(); [[nodiscard]] std::unique_ptr<FFmpeg::Frame> GetCurrentFrame();
/// Returns the value of current_codec /// Returns the value of current_codec
[[nodiscard]] Host1x::NvdecCommon::VideoCodec GetCurrentCodec() const; [[nodiscard]] Host1x::NvdecCommon::VideoCodec GetCurrentCodec() const;
@ -60,25 +47,9 @@ public:
[[nodiscard]] std::string_view GetCurrentCodecName() const; [[nodiscard]] std::string_view GetCurrentCodecName() const;
private: private:
void InitializeAvCodecContext();
void InitializeAvFilters(AVFrame* frame);
void InitializeGpuDecoder();
bool CreateGpuAvDevice();
bool initialized{}; bool initialized{};
bool filters_initialized{};
Host1x::NvdecCommon::VideoCodec current_codec{Host1x::NvdecCommon::VideoCodec::None}; Host1x::NvdecCommon::VideoCodec current_codec{Host1x::NvdecCommon::VideoCodec::None};
FFmpeg::DecodeApi decode_api;
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};
Host1x::Host1x& host1x; Host1x::Host1x& host1x;
const Host1x::NvdecCommon::NvdecRegisters& state; const Host1x::NvdecCommon::NvdecRegisters& state;
@ -86,7 +57,7 @@ private:
std::unique_ptr<Decoder::VP8> vp8_decoder; std::unique_ptr<Decoder::VP8> vp8_decoder;
std::unique_ptr<Decoder::VP9> vp9_decoder; std::unique_ptr<Decoder::VP9> vp9_decoder;
std::queue<AVFramePtr> av_frames{}; std::queue<std::unique_ptr<FFmpeg::Frame>> frames{};
}; };
} // namespace Tegra } // namespace Tegra

View File

@ -30,7 +30,7 @@ H264::H264(Host1x::Host1x& host1x_) : host1x{host1x_} {}
H264::~H264() = default; H264::~H264() = default;
std::span<const u8> H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state, 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; H264DecoderContext context;
host1x.MemoryManager().ReadBlock(state.picture_info_offset, &context, host1x.MemoryManager().ReadBlock(state.picture_info_offset, &context,
sizeof(H264DecoderContext)); sizeof(H264DecoderContext));
@ -39,6 +39,7 @@ std::span<const u8> H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters
if (!is_first_frame && frame_number != 0) { if (!is_first_frame && frame_number != 0) {
frame.resize_destructive(context.stream_len); frame.resize_destructive(context.stream_len);
host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size()); host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size());
*out_configuration_size = 0;
return frame; return frame;
} }
@ -157,6 +158,7 @@ std::span<const u8> H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters
frame.resize(encoded_header.size() + context.stream_len); frame.resize(encoded_header.size() + context.stream_len);
std::memcpy(frame.data(), encoded_header.data(), encoded_header.size()); std::memcpy(frame.data(), encoded_header.data(), encoded_header.size());
*out_configuration_size = encoded_header.size();
host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset,
frame.data() + encoded_header.size(), context.stream_len); frame.data() + encoded_header.size(), context.stream_len);

View File

@ -67,6 +67,7 @@ public:
/// Compose the H264 frame for FFmpeg decoding /// Compose the H264 frame for FFmpeg decoding
[[nodiscard]] std::span<const u8> ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state, [[nodiscard]] std::span<const u8> ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state,
size_t* out_configuration_size,
bool is_first_frame = false); bool is_first_frame = false);
private: private:

View File

@ -0,0 +1,419 @@
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "common/assert.h"
#include "common/logging/log.h"
#include "common/scope_exit.h"
#include "common/settings.h"
#include "video_core/host1x/ffmpeg/ffmpeg.h"
extern "C" {
#ifdef LIBVA_FOUND
// for querying VAAPI driver information
#include <libavutil/hwcontext_vaapi.h>
#endif
}
namespace FFmpeg {
namespace {
constexpr AVPixelFormat PreferredGpuFormat = AV_PIX_FMT_NV12;
constexpr AVPixelFormat PreferredCpuFormat = AV_PIX_FMT_YUV420P;
constexpr std::array PreferredGpuDecoders = {
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,
};
AVPixelFormat GetGpuFormat(AVCodecContext* codec_context, const AVPixelFormat* pix_fmts) {
for (const AVPixelFormat* p = pix_fmts; *p != AV_PIX_FMT_NONE; ++p) {
if (*p == codec_context->pix_fmt) {
return codec_context->pix_fmt;
}
}
LOG_INFO(HW_GPU, "Could not find compatible GPU AV format, falling back to CPU");
av_buffer_unref(&codec_context->hw_device_ctx);
codec_context->pix_fmt = PreferredCpuFormat;
return codec_context->pix_fmt;
}
std::string AVError(int errnum) {
char errbuf[AV_ERROR_MAX_STRING_SIZE] = {};
av_make_error_string(errbuf, sizeof(errbuf) - 1, errnum);
return errbuf;
}
} // namespace
Packet::Packet(std::span<const u8> data) {
m_packet = av_packet_alloc();
m_packet->data = const_cast<u8*>(data.data());
m_packet->size = static_cast<s32>(data.size());
}
Packet::~Packet() {
av_packet_free(&m_packet);
}
Frame::Frame() {
m_frame = av_frame_alloc();
}
Frame::~Frame() {
av_frame_free(&m_frame);
}
Decoder::Decoder(Tegra::Host1x::NvdecCommon::VideoCodec codec) {
const AVCodecID av_codec = [&] {
switch (codec) {
case Tegra::Host1x::NvdecCommon::VideoCodec::H264:
return AV_CODEC_ID_H264;
case Tegra::Host1x::NvdecCommon::VideoCodec::VP8:
return AV_CODEC_ID_VP8;
case Tegra::Host1x::NvdecCommon::VideoCodec::VP9:
return AV_CODEC_ID_VP9;
default:
UNIMPLEMENTED_MSG("Unknown codec {}", codec);
return AV_CODEC_ID_NONE;
}
}();
m_codec = avcodec_find_decoder(av_codec);
}
bool Decoder::SupportsDecodingOnDevice(AVPixelFormat* out_pix_fmt, AVHWDeviceType type) const {
for (int i = 0;; i++) {
const AVCodecHWConfig* config = avcodec_get_hw_config(m_codec, i);
if (!config) {
LOG_DEBUG(HW_GPU, "{} decoder does not support device type {}", m_codec->name,
av_hwdevice_get_type_name(type));
break;
}
if ((config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX) != 0 &&
config->device_type == type) {
LOG_INFO(HW_GPU, "Using {} GPU decoder", av_hwdevice_get_type_name(type));
*out_pix_fmt = config->pix_fmt;
return true;
}
}
return false;
}
std::vector<AVHWDeviceType> HardwareContext::GetSupportedDeviceTypes() {
std::vector<AVHWDeviceType> types;
AVHWDeviceType current_device_type = AV_HWDEVICE_TYPE_NONE;
while (true) {
current_device_type = av_hwdevice_iterate_types(current_device_type);
if (current_device_type == AV_HWDEVICE_TYPE_NONE) {
return types;
}
types.push_back(current_device_type);
}
}
HardwareContext::~HardwareContext() {
av_buffer_unref(&m_gpu_decoder);
}
bool HardwareContext::InitializeForDecoder(DecoderContext& decoder_context,
const Decoder& decoder) {
const auto supported_types = GetSupportedDeviceTypes();
for (const auto type : PreferredGpuDecoders) {
AVPixelFormat hw_pix_fmt;
if (std::ranges::find(supported_types, type) == supported_types.end()) {
LOG_DEBUG(HW_GPU, "{} explicitly unsupported", av_hwdevice_get_type_name(type));
continue;
}
if (!this->InitializeWithType(type)) {
continue;
}
if (decoder.SupportsDecodingOnDevice(&hw_pix_fmt, type)) {
decoder_context.InitializeHardwareDecoder(*this, hw_pix_fmt);
return true;
}
}
return false;
}
bool HardwareContext::InitializeWithType(AVHWDeviceType type) {
av_buffer_unref(&m_gpu_decoder);
if (const int ret = av_hwdevice_ctx_create(&m_gpu_decoder, type, nullptr, nullptr, 0);
ret < 0) {
LOG_DEBUG(HW_GPU, "av_hwdevice_ctx_create({}) failed: {}", av_hwdevice_get_type_name(type),
AVError(ret));
return false;
}
#ifdef LIBVA_FOUND
if (type == AV_HWDEVICE_TYPE_VAAPI) {
// We need to determine if this is an impersonated VAAPI driver.
auto* hwctx = reinterpret_cast<AVHWDeviceContext*>(m_gpu_decoder->data);
auto* vactx = static_cast<AVVAAPIDeviceContext*>(hwctx->hwctx);
const char* vendor_name = vaQueryVendorString(vactx->display);
if (strstr(vendor_name, "VDPAU backend")) {
// VDPAU impersonated VAAPI impls are super buggy, we need to skip them.
LOG_DEBUG(HW_GPU, "Skipping VDPAU impersonated VAAPI driver");
return false;
} else {
// According to some user testing, certain VAAPI drivers (Intel?) could be buggy.
// Log the driver name just in case.
LOG_DEBUG(HW_GPU, "Using VAAPI driver: {}", vendor_name);
}
}
#endif
return true;
}
DecoderContext::DecoderContext(const Decoder& decoder) {
m_codec_context = avcodec_alloc_context3(decoder.GetCodec());
av_opt_set(m_codec_context->priv_data, "tune", "zerolatency", 0);
m_codec_context->thread_count = 0;
m_codec_context->thread_type &= ~FF_THREAD_FRAME;
}
DecoderContext::~DecoderContext() {
av_buffer_unref(&m_codec_context->hw_device_ctx);
avcodec_free_context(&m_codec_context);
}
void DecoderContext::InitializeHardwareDecoder(const HardwareContext& context,
AVPixelFormat hw_pix_fmt) {
m_codec_context->hw_device_ctx = av_buffer_ref(context.GetBufferRef());
m_codec_context->get_format = GetGpuFormat;
m_codec_context->pix_fmt = hw_pix_fmt;
}
bool DecoderContext::OpenContext(const Decoder& decoder) {
if (const int ret = avcodec_open2(m_codec_context, decoder.GetCodec(), nullptr); ret < 0) {
LOG_ERROR(HW_GPU, "avcodec_open2 error: {}", AVError(ret));
return false;
}
if (!m_codec_context->hw_device_ctx) {
LOG_INFO(HW_GPU, "Using FFmpeg software decoding");
}
return true;
}
bool DecoderContext::SendPacket(const Packet& packet) {
if (const int ret = avcodec_send_packet(m_codec_context, packet.GetPacket()); ret < 0) {
LOG_ERROR(HW_GPU, "avcodec_send_packet error: {}", AVError(ret));
return false;
}
return true;
}
std::unique_ptr<Frame> DecoderContext::ReceiveFrame(bool* out_is_interlaced) {
auto dst_frame = std::make_unique<Frame>();
const auto ReceiveImpl = [&](AVFrame* frame) {
if (const int ret = avcodec_receive_frame(m_codec_context, frame); ret < 0) {
LOG_ERROR(HW_GPU, "avcodec_receive_frame error: {}", AVError(ret));
return false;
}
*out_is_interlaced = frame->interlaced_frame != 0;
return true;
};
if (m_codec_context->hw_device_ctx) {
// If we have a hardware context, make a separate frame here to receive the
// hardware result before sending it to the output.
Frame intermediate_frame;
if (!ReceiveImpl(intermediate_frame.GetFrame())) {
return {};
}
dst_frame->SetFormat(PreferredGpuFormat);
if (const int ret =
av_hwframe_transfer_data(dst_frame->GetFrame(), intermediate_frame.GetFrame(), 0);
ret < 0) {
LOG_ERROR(HW_GPU, "av_hwframe_transfer_data error: {}", AVError(ret));
return {};
}
} else {
// Otherwise, decode the frame as normal.
if (!ReceiveImpl(dst_frame->GetFrame())) {
return {};
}
}
return dst_frame;
}
DeinterlaceFilter::DeinterlaceFilter(const Frame& 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.GetWidth(),
frame.GetHeight(), static_cast<int>(frame.GetPixelFormat()));
m_filter_graph = avfilter_graph_alloc();
int ret = avfilter_graph_create_filter(&m_source_context, buffer_src, "in", args.c_str(),
nullptr, m_filter_graph);
if (ret < 0) {
LOG_ERROR(HW_GPU, "avfilter_graph_create_filter source error: {}", AVError(ret));
return;
}
ret = avfilter_graph_create_filter(&m_sink_context, buffer_sink, "out", nullptr, nullptr,
m_filter_graph);
if (ret < 0) {
LOG_ERROR(HW_GPU, "avfilter_graph_create_filter sink error: {}", AVError(ret));
return;
}
inputs->name = av_strdup("out");
inputs->filter_ctx = m_sink_context;
inputs->pad_idx = 0;
inputs->next = nullptr;
outputs->name = av_strdup("in");
outputs->filter_ctx = m_source_context;
outputs->pad_idx = 0;
outputs->next = nullptr;
const char* description = "yadif=1:-1:0";
ret = avfilter_graph_parse_ptr(m_filter_graph, description, &inputs, &outputs, nullptr);
if (ret < 0) {
LOG_ERROR(HW_GPU, "avfilter_graph_parse_ptr error: {}", AVError(ret));
return;
}
ret = avfilter_graph_config(m_filter_graph, nullptr);
if (ret < 0) {
LOG_ERROR(HW_GPU, "avfilter_graph_config error: {}", AVError(ret));
return;
}
m_initialized = true;
}
bool DeinterlaceFilter::AddSourceFrame(const Frame& frame) {
if (const int ret = av_buffersrc_add_frame_flags(m_source_context, frame.GetFrame(),
AV_BUFFERSRC_FLAG_KEEP_REF);
ret < 0) {
LOG_ERROR(HW_GPU, "av_buffersrc_add_frame_flags error: {}", AVError(ret));
return false;
}
return true;
}
std::unique_ptr<Frame> DeinterlaceFilter::DrainSinkFrame() {
auto dst_frame = std::make_unique<Frame>();
const int ret = av_buffersink_get_frame(m_sink_context, dst_frame->GetFrame());
if (ret == AVERROR(EAGAIN) || ret == AVERROR(AVERROR_EOF)) {
return {};
}
if (ret < 0) {
LOG_ERROR(HW_GPU, "av_buffersink_get_frame error: {}", AVError(ret));
return {};
}
return dst_frame;
}
DeinterlaceFilter::~DeinterlaceFilter() {
avfilter_graph_free(&m_filter_graph);
}
void DecodeApi::Reset() {
m_deinterlace_filter.reset();
m_hardware_context.reset();
m_decoder_context.reset();
m_decoder.reset();
}
bool DecodeApi::Initialize(Tegra::Host1x::NvdecCommon::VideoCodec codec) {
this->Reset();
m_decoder.emplace(codec);
m_decoder_context.emplace(*m_decoder);
// Enable GPU decoding if requested.
if (Settings::values.nvdec_emulation.GetValue() == Settings::NvdecEmulation::Gpu) {
m_hardware_context.emplace();
m_hardware_context->InitializeForDecoder(*m_decoder_context, *m_decoder);
}
// Open the decoder context.
if (!m_decoder_context->OpenContext(*m_decoder)) {
this->Reset();
return false;
}
return true;
}
bool DecodeApi::SendPacket(std::span<const u8> packet_data, size_t configuration_size) {
FFmpeg::Packet packet(packet_data);
return m_decoder_context->SendPacket(packet);
}
void DecodeApi::ReceiveFrames(std::queue<std::unique_ptr<Frame>>& frame_queue) {
// Receive raw frame from decoder.
bool is_interlaced;
auto frame = m_decoder_context->ReceiveFrame(&is_interlaced);
if (!frame) {
return;
}
if (!is_interlaced) {
// If the frame is not interlaced, we can pend it now.
frame_queue.push(std::move(frame));
} else {
// Create the deinterlacer if needed.
if (!m_deinterlace_filter) {
m_deinterlace_filter.emplace(*frame);
}
// Add the frame we just received.
if (!m_deinterlace_filter->AddSourceFrame(*frame)) {
return;
}
// Pend output fields.
while (true) {
auto filter_frame = m_deinterlace_filter->DrainSinkFrame();
if (!filter_frame) {
break;
}
frame_queue.push(std::move(filter_frame));
}
}
}
} // namespace FFmpeg

View File

@ -0,0 +1,213 @@
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <memory>
#include <optional>
#include <span>
#include <vector>
#include <queue>
#include "common/common_funcs.h"
#include "common/common_types.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>
#include <libavfilter/buffersink.h>
#include <libavfilter/buffersrc.h>
#include <libavutil/avutil.h>
#include <libavutil/opt.h>
#if defined(__GNUC__) || defined(__clang__)
#pragma GCC diagnostic pop
#endif
}
namespace FFmpeg {
class Packet;
class Frame;
class Decoder;
class HardwareContext;
class DecoderContext;
class DeinterlaceFilter;
// Wraps an AVPacket, a container for compressed bitstream data.
class Packet {
public:
YUZU_NON_COPYABLE(Packet);
YUZU_NON_MOVEABLE(Packet);
explicit Packet(std::span<const u8> data);
~Packet();
AVPacket* GetPacket() const {
return m_packet;
}
private:
AVPacket* m_packet{};
};
// Wraps an AVFrame, a container for audio and video stream data.
class Frame {
public:
YUZU_NON_COPYABLE(Frame);
YUZU_NON_MOVEABLE(Frame);
explicit Frame();
~Frame();
int GetWidth() const {
return m_frame->width;
}
int GetHeight() const {
return m_frame->height;
}
AVPixelFormat GetPixelFormat() const {
return static_cast<AVPixelFormat>(m_frame->format);
}
int GetStride(int plane) const {
return m_frame->linesize[plane];
}
int* GetStrides() const {
return m_frame->linesize;
}
u8* GetData(int plane) const {
return m_frame->data[plane];
}
u8** GetPlanes() const {
return m_frame->data;
}
void SetFormat(int format) {
m_frame->format = format;
}
AVFrame* GetFrame() const {
return m_frame;
}
private:
AVFrame* m_frame{};
};
// Wraps an AVCodec, a type containing information about a codec.
class Decoder {
public:
YUZU_NON_COPYABLE(Decoder);
YUZU_NON_MOVEABLE(Decoder);
explicit Decoder(Tegra::Host1x::NvdecCommon::VideoCodec codec);
~Decoder() = default;
bool SupportsDecodingOnDevice(AVPixelFormat* out_pix_fmt, AVHWDeviceType type) const;
const AVCodec* GetCodec() const {
return m_codec;
}
private:
const AVCodec* m_codec{};
};
// Wraps AVBufferRef for an accelerated decoder.
class HardwareContext {
public:
YUZU_NON_COPYABLE(HardwareContext);
YUZU_NON_MOVEABLE(HardwareContext);
static std::vector<AVHWDeviceType> GetSupportedDeviceTypes();
explicit HardwareContext() = default;
~HardwareContext();
bool InitializeForDecoder(DecoderContext& decoder_context, const Decoder& decoder);
AVBufferRef* GetBufferRef() const {
return m_gpu_decoder;
}
private:
bool InitializeWithType(AVHWDeviceType type);
AVBufferRef* m_gpu_decoder{};
};
// Wraps an AVCodecContext.
class DecoderContext {
public:
YUZU_NON_COPYABLE(DecoderContext);
YUZU_NON_MOVEABLE(DecoderContext);
explicit DecoderContext(const Decoder& decoder);
~DecoderContext();
void InitializeHardwareDecoder(const HardwareContext& context, AVPixelFormat hw_pix_fmt);
bool OpenContext(const Decoder& decoder);
bool SendPacket(const Packet& packet);
std::unique_ptr<Frame> ReceiveFrame(bool* out_is_interlaced);
AVCodecContext* GetCodecContext() const {
return m_codec_context;
}
private:
AVCodecContext* m_codec_context{};
};
// Wraps an AVFilterGraph.
class DeinterlaceFilter {
public:
YUZU_NON_COPYABLE(DeinterlaceFilter);
YUZU_NON_MOVEABLE(DeinterlaceFilter);
explicit DeinterlaceFilter(const Frame& frame);
~DeinterlaceFilter();
bool AddSourceFrame(const Frame& frame);
std::unique_ptr<Frame> DrainSinkFrame();
private:
AVFilterGraph* m_filter_graph{};
AVFilterContext* m_source_context{};
AVFilterContext* m_sink_context{};
bool m_initialized{};
};
class DecodeApi {
public:
YUZU_NON_COPYABLE(DecodeApi);
YUZU_NON_MOVEABLE(DecodeApi);
DecodeApi() = default;
~DecodeApi() = default;
bool Initialize(Tegra::Host1x::NvdecCommon::VideoCodec codec);
void Reset();
bool SendPacket(std::span<const u8> packet_data, size_t configuration_size);
void ReceiveFrames(std::queue<std::unique_ptr<Frame>>& frame_queue);
private:
std::optional<FFmpeg::Decoder> m_decoder;
std::optional<FFmpeg::DecoderContext> m_decoder_context;
std::optional<FFmpeg::HardwareContext> m_hardware_context;
std::optional<FFmpeg::DeinterlaceFilter> m_deinterlace_filter;
};
} // namespace FFmpeg

View File

@ -28,7 +28,7 @@ void Nvdec::ProcessMethod(u32 method, u32 argument) {
} }
} }
AVFramePtr Nvdec::GetFrame() { std::unique_ptr<FFmpeg::Frame> Nvdec::GetFrame() {
return codec->GetCurrentFrame(); return codec->GetCurrentFrame();
} }

View File

@ -23,7 +23,7 @@ public:
void ProcessMethod(u32 method, u32 argument); void ProcessMethod(u32 method, u32 argument);
/// Return most recently decoded frame /// Return most recently decoded frame
[[nodiscard]] AVFramePtr GetFrame(); [[nodiscard]] std::unique_ptr<FFmpeg::Frame> GetFrame();
private: private:
/// Invoke codec to decode a frame /// Invoke codec to decode a frame

View File

@ -82,27 +82,26 @@ void Vic::Execute() {
return; return;
} }
const VicConfig config{host1x.MemoryManager().Read<u64>(config_struct_address + 0x20)}; const VicConfig config{host1x.MemoryManager().Read<u64>(config_struct_address + 0x20)};
const AVFramePtr frame_ptr = nvdec_processor->GetFrame(); auto frame = nvdec_processor->GetFrame();
const auto* frame = frame_ptr.get();
if (!frame) { if (!frame) {
return; return;
} }
const u64 surface_width = config.surface_width_minus1 + 1; const u64 surface_width = config.surface_width_minus1 + 1;
const u64 surface_height = config.surface_height_minus1 + 1; const u64 surface_height = config.surface_height_minus1 + 1;
if (static_cast<u64>(frame->width) != surface_width || if (static_cast<u64>(frame->GetWidth()) != surface_width ||
static_cast<u64>(frame->height) != surface_height) { static_cast<u64>(frame->GetHeight()) != surface_height) {
// TODO: Properly support multiple video streams with differing frame dimensions // TODO: Properly support multiple video streams with differing frame dimensions
LOG_WARNING(Service_NVDRV, "Frame dimensions {}x{} don't match surface dimensions {}x{}", LOG_WARNING(Service_NVDRV, "Frame dimensions {}x{} don't match surface dimensions {}x{}",
frame->width, frame->height, surface_width, surface_height); frame->GetWidth(), frame->GetHeight(), surface_width, surface_height);
} }
switch (config.pixel_format) { switch (config.pixel_format) {
case VideoPixelFormat::RGBA8: case VideoPixelFormat::RGBA8:
case VideoPixelFormat::BGRA8: case VideoPixelFormat::BGRA8:
case VideoPixelFormat::RGBX8: case VideoPixelFormat::RGBX8:
WriteRGBFrame(frame, config); WriteRGBFrame(std::move(frame), config);
break; break;
case VideoPixelFormat::YUV420: case VideoPixelFormat::YUV420:
WriteYUVFrame(frame, config); WriteYUVFrame(std::move(frame), config);
break; break;
default: default:
UNIMPLEMENTED_MSG("Unknown video pixel format {:X}", config.pixel_format.Value()); UNIMPLEMENTED_MSG("Unknown video pixel format {:X}", config.pixel_format.Value());
@ -110,10 +109,14 @@ void Vic::Execute() {
} }
} }
void Vic::WriteRGBFrame(const AVFrame* frame, const VicConfig& config) { void Vic::WriteRGBFrame(std::unique_ptr<FFmpeg::Frame> frame, const VicConfig& config) {
LOG_TRACE(Service_NVDRV, "Writing RGB Frame"); LOG_TRACE(Service_NVDRV, "Writing RGB Frame");
if (!scaler_ctx || frame->width != scaler_width || frame->height != scaler_height) { const auto frame_width = frame->GetWidth();
const auto frame_height = frame->GetHeight();
const auto frame_format = frame->GetPixelFormat();
if (!scaler_ctx || frame_width != scaler_width || frame_height != scaler_height) {
const AVPixelFormat target_format = [pixel_format = config.pixel_format]() { const AVPixelFormat target_format = [pixel_format = config.pixel_format]() {
switch (pixel_format) { switch (pixel_format) {
case VideoPixelFormat::RGBA8: case VideoPixelFormat::RGBA8:
@ -129,27 +132,26 @@ void Vic::WriteRGBFrame(const AVFrame* frame, const VicConfig& config) {
sws_freeContext(scaler_ctx); sws_freeContext(scaler_ctx);
// Frames are decoded into either YUV420 or NV12 formats. Convert to desired RGB format // Frames are decoded into either YUV420 or NV12 formats. Convert to desired RGB format
scaler_ctx = sws_getContext(frame->width, frame->height, scaler_ctx = sws_getContext(frame_width, frame_height, frame_format, frame_width,
static_cast<AVPixelFormat>(frame->format), frame->width, frame_height, target_format, 0, nullptr, nullptr, nullptr);
frame->height, target_format, 0, nullptr, nullptr, nullptr); scaler_width = frame_width;
scaler_width = frame->width; scaler_height = frame_height;
scaler_height = frame->height;
converted_frame_buffer.reset(); converted_frame_buffer.reset();
} }
if (!converted_frame_buffer) { if (!converted_frame_buffer) {
const size_t frame_size = frame->width * frame->height * 4; const size_t frame_size = frame_width * frame_height * 4;
converted_frame_buffer = AVMallocPtr{static_cast<u8*>(av_malloc(frame_size)), av_free}; converted_frame_buffer = AVMallocPtr{static_cast<u8*>(av_malloc(frame_size)), av_free};
} }
const std::array<int, 4> converted_stride{frame->width * 4, frame->height * 4, 0, 0}; const std::array<int, 4> converted_stride{frame_width * 4, frame_height * 4, 0, 0};
u8* const converted_frame_buf_addr{converted_frame_buffer.get()}; u8* const converted_frame_buf_addr{converted_frame_buffer.get()};
sws_scale(scaler_ctx, frame->data, frame->linesize, 0, frame->height, &converted_frame_buf_addr, sws_scale(scaler_ctx, frame->GetPlanes(), frame->GetStrides(), 0, frame_height,
converted_stride.data()); &converted_frame_buf_addr, converted_stride.data());
// Use the minimum of surface/frame dimensions to avoid buffer overflow. // Use the minimum of surface/frame dimensions to avoid buffer overflow.
const u32 surface_width = static_cast<u32>(config.surface_width_minus1) + 1; const u32 surface_width = static_cast<u32>(config.surface_width_minus1) + 1;
const u32 surface_height = static_cast<u32>(config.surface_height_minus1) + 1; const u32 surface_height = static_cast<u32>(config.surface_height_minus1) + 1;
const u32 width = std::min(surface_width, static_cast<u32>(frame->width)); const u32 width = std::min(surface_width, static_cast<u32>(frame_width));
const u32 height = std::min(surface_height, static_cast<u32>(frame->height)); const u32 height = std::min(surface_height, static_cast<u32>(frame_height));
const u32 blk_kind = static_cast<u32>(config.block_linear_kind); const u32 blk_kind = static_cast<u32>(config.block_linear_kind);
if (blk_kind != 0) { if (blk_kind != 0) {
// swizzle pitch linear to block linear // swizzle pitch linear to block linear
@ -169,23 +171,23 @@ void Vic::WriteRGBFrame(const AVFrame* frame, const VicConfig& config) {
} }
} }
void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) { void Vic::WriteYUVFrame(std::unique_ptr<FFmpeg::Frame> frame, const VicConfig& config) {
LOG_TRACE(Service_NVDRV, "Writing YUV420 Frame"); LOG_TRACE(Service_NVDRV, "Writing YUV420 Frame");
const std::size_t surface_width = config.surface_width_minus1 + 1; const std::size_t surface_width = config.surface_width_minus1 + 1;
const std::size_t surface_height = config.surface_height_minus1 + 1; const std::size_t surface_height = config.surface_height_minus1 + 1;
const std::size_t aligned_width = (surface_width + 0xff) & ~0xffUL; const std::size_t aligned_width = (surface_width + 0xff) & ~0xffUL;
// Use the minimum of surface/frame dimensions to avoid buffer overflow. // Use the minimum of surface/frame dimensions to avoid buffer overflow.
const auto frame_width = std::min(surface_width, static_cast<size_t>(frame->width)); const auto frame_width = std::min(surface_width, static_cast<size_t>(frame->GetWidth()));
const auto frame_height = std::min(surface_height, static_cast<size_t>(frame->height)); const auto frame_height = std::min(surface_height, static_cast<size_t>(frame->GetHeight()));
const auto stride = static_cast<size_t>(frame->linesize[0]); const auto stride = static_cast<size_t>(frame->GetStride(0));
luma_buffer.resize_destructive(aligned_width * surface_height); luma_buffer.resize_destructive(aligned_width * surface_height);
chroma_buffer.resize_destructive(aligned_width * surface_height / 2); chroma_buffer.resize_destructive(aligned_width * surface_height / 2);
// Populate luma buffer // Populate luma buffer
const u8* luma_src = frame->data[0]; const u8* luma_src = frame->GetData(0);
for (std::size_t y = 0; y < frame_height; ++y) { for (std::size_t y = 0; y < frame_height; ++y) {
const std::size_t src = y * stride; const std::size_t src = y * stride;
const std::size_t dst = y * aligned_width; const std::size_t dst = y * aligned_width;
@ -196,16 +198,16 @@ void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) {
// Chroma // Chroma
const std::size_t half_height = frame_height / 2; const std::size_t half_height = frame_height / 2;
const auto half_stride = static_cast<size_t>(frame->linesize[1]); const auto half_stride = static_cast<size_t>(frame->GetStride(1));
switch (frame->format) { switch (frame->GetPixelFormat()) {
case AV_PIX_FMT_YUV420P: { case AV_PIX_FMT_YUV420P: {
// Frame from FFmpeg software // Frame from FFmpeg software
// Populate chroma buffer from both channels with interleaving. // Populate chroma buffer from both channels with interleaving.
const std::size_t half_width = frame_width / 2; const std::size_t half_width = frame_width / 2;
u8* chroma_buffer_data = chroma_buffer.data(); u8* chroma_buffer_data = chroma_buffer.data();
const u8* chroma_b_src = frame->data[1]; const u8* chroma_b_src = frame->GetData(1);
const u8* chroma_r_src = frame->data[2]; const u8* chroma_r_src = frame->GetData(2);
for (std::size_t y = 0; y < half_height; ++y) { for (std::size_t y = 0; y < half_height; ++y) {
const std::size_t src = y * half_stride; const std::size_t src = y * half_stride;
const std::size_t dst = y * aligned_width; const std::size_t dst = y * aligned_width;
@ -219,7 +221,7 @@ void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) {
case AV_PIX_FMT_NV12: { case AV_PIX_FMT_NV12: {
// Frame from VA-API hardware // Frame from VA-API hardware
// This is already interleaved so just copy // This is already interleaved so just copy
const u8* chroma_src = frame->data[1]; const u8* chroma_src = frame->GetData(1);
for (std::size_t y = 0; y < half_height; ++y) { for (std::size_t y = 0; y < half_height; ++y) {
const std::size_t src = y * stride; const std::size_t src = y * stride;
const std::size_t dst = y * aligned_width; const std::size_t dst = y * aligned_width;

View File

@ -39,9 +39,9 @@ public:
private: private:
void Execute(); void Execute();
void WriteRGBFrame(const AVFrame* frame, const VicConfig& config); void WriteRGBFrame(std::unique_ptr<FFmpeg::Frame> frame, const VicConfig& config);
void WriteYUVFrame(const AVFrame* frame, const VicConfig& config); void WriteYUVFrame(std::unique_ptr<FFmpeg::Frame> frame, const VicConfig& config);
Host1x& host1x; Host1x& host1x;
std::shared_ptr<Tegra::Host1x::Nvdec> nvdec_processor; std::shared_ptr<Tegra::Host1x::Nvdec> nvdec_processor;