FFmpegReader.cpp

Go to the documentation of this file.

 
// Copyright (c) 2008-2024 OpenShot Studios, LLC, Fabrice Bellard
//
// SPDX-License-Identifier: LGPL-3.0-or-later
 
#include <thread>   // for std::this_thread::sleep_for
#include <chrono>   // for std::chrono::milliseconds
#include <algorithm>
#include <cmath>
#include <sstream>
#include <unistd.h>
 
#include "FFmpegUtilities.h"
#include "effects/CropHelpers.h"
 
#include "FFmpegReader.h"
#include "Exceptions.h"
#include "MemoryTrim.h"
#include "Timeline.h"
#include "ZmqLogger.h"
 
#define ENABLE_VAAPI 0
 
#if USE_HW_ACCEL
#define MAX_SUPPORTED_WIDTH 1950
#define MAX_SUPPORTED_HEIGHT 1100
 
#if ENABLE_VAAPI
#include "libavutil/hwcontext_vaapi.h"
 
typedef struct VAAPIDecodeContext {
     VAProfile va_profile;
     VAEntrypoint va_entrypoint;
     VAConfigID va_config;
     VAContextID va_context;
 
#if FF_API_STRUCT_VAAPI_CONTEXT
    // FF_DISABLE_DEPRECATION_WARNINGS
    int have_old_context;
    struct vaapi_context *old_context;
    AVBufferRef *device_ref;
    // FF_ENABLE_DEPRECATION_WARNINGS
#endif
 
     AVHWDeviceContext *device;
     AVVAAPIDeviceContext *hwctx;
 
     AVHWFramesContext *frames;
     AVVAAPIFramesContext *hwfc;
 
     enum AVPixelFormat surface_format;
     int surface_count;
 } VAAPIDecodeContext;
#endif // ENABLE_VAAPI
#endif // USE_HW_ACCEL
 
 
using namespace openshot;
 
int hw_de_on = 0;
#if USE_HW_ACCEL
    AVPixelFormat hw_de_av_pix_fmt_global = AV_PIX_FMT_NONE;
    AVHWDeviceType hw_de_av_device_type_global = AV_HWDEVICE_TYPE_NONE;
#endif
 
FFmpegReader::FFmpegReader(const std::string &path, bool inspect_reader)
        : FFmpegReader(path, DurationStrategy::VideoPreferred, inspect_reader) {}
 
FFmpegReader::FFmpegReader(const std::string &path, DurationStrategy duration_strategy, bool inspect_reader)
        : last_frame(0), is_seeking(0), seeking_pts(0), seeking_frame(0), seek_count(0), NO_PTS_OFFSET(-99999),
          path(path), is_video_seek(true), check_interlace(false), check_fps(false), enable_seek(true), is_open(false),
          seek_audio_frame_found(0), seek_video_frame_found(0),
          last_seek_max_frame(-1), seek_stagnant_count(0),
          is_duration_known(false), largest_frame_processed(0),
          current_video_frame(0), packet(NULL), duration_strategy(duration_strategy),
          audio_pts(0), video_pts(0), pFormatCtx(NULL), videoStream(-1), audioStream(-1), pCodecCtx(NULL), aCodecCtx(NULL),
        pStream(NULL), aStream(NULL), pFrame(NULL), previous_packet_location{-1,0},
        hold_packet(false) {
 
    // Initialize FFMpeg, and register all formats and codecs
    AV_REGISTER_ALL
    AVCODEC_REGISTER_ALL
 
    // Init timestamp offsets
    pts_offset_seconds = NO_PTS_OFFSET;
    video_pts_seconds = NO_PTS_OFFSET;
    audio_pts_seconds = NO_PTS_OFFSET;
 
    // Init cache
    const int init_working_cache_frames = std::max(Settings::Instance()->CACHE_MIN_FRAMES, OPEN_MP_NUM_PROCESSORS * 4);
    const int init_final_cache_frames = std::max(Settings::Instance()->CACHE_MIN_FRAMES, OPEN_MP_NUM_PROCESSORS * 4);
    working_cache.SetMaxBytesFromInfo(init_working_cache_frames, info.width, info.height, info.sample_rate, info.channels);
    final_cache.SetMaxBytesFromInfo(init_final_cache_frames, info.width, info.height, info.sample_rate, info.channels);
 
    // Open and Close the reader, to populate its attributes (such as height, width, etc...)
    if (inspect_reader) {
        Open();
        Close();
    }
}
 
FFmpegReader::~FFmpegReader() {
    if (is_open)
        // Auto close reader if not already done
        Close();
}
 
// This struct holds the associated video frame and starting sample # for an audio packet.
bool AudioLocation::is_near(AudioLocation location, int samples_per_frame, int64_t amount) {
    // Is frame even close to this one?
    if (abs(location.frame - frame) >= 2)
        // This is too far away to be considered
        return false;
 
    // Note that samples_per_frame can vary slightly frame to frame when the
    // audio sampling rate is not an integer multiple of the video fps.
    int64_t diff = samples_per_frame * (location.frame - frame) + location.sample_start - sample_start;
    if (abs(diff) <= amount)
        // close
        return true;
 
    // not close
    return false;
}
 
#if USE_HW_ACCEL
 
// Get hardware pix format
static enum AVPixelFormat get_hw_dec_format(AVCodecContext *ctx, const enum AVPixelFormat *pix_fmts)
{
    const enum AVPixelFormat *p;
 
    // Prefer only the format matching the selected hardware decoder
    int selected = openshot::Settings::Instance()->HARDWARE_DECODER;
 
    for (p = pix_fmts; *p != AV_PIX_FMT_NONE; p++) {
        switch (*p) {
#if defined(__linux__)
            // Linux pix formats
            case AV_PIX_FMT_VAAPI:
                if (selected == 1) {
                    hw_de_av_pix_fmt_global = AV_PIX_FMT_VAAPI;
                    hw_de_av_device_type_global = AV_HWDEVICE_TYPE_VAAPI;
                    return *p;
                }
                break;
            case AV_PIX_FMT_VDPAU:
                if (selected == 6) {
                    hw_de_av_pix_fmt_global = AV_PIX_FMT_VDPAU;
                    hw_de_av_device_type_global = AV_HWDEVICE_TYPE_VDPAU;
                    return *p;
                }
                break;
#endif
#if defined(_WIN32)
            // Windows pix formats
            case AV_PIX_FMT_DXVA2_VLD:
                if (selected == 3) {
                    hw_de_av_pix_fmt_global = AV_PIX_FMT_DXVA2_VLD;
                    hw_de_av_device_type_global = AV_HWDEVICE_TYPE_DXVA2;
                    return *p;
                }
                break;
            case AV_PIX_FMT_D3D11:
                if (selected == 4) {
                    hw_de_av_pix_fmt_global = AV_PIX_FMT_D3D11;
                    hw_de_av_device_type_global = AV_HWDEVICE_TYPE_D3D11VA;
                    return *p;
                }
                break;
#endif
#if defined(__APPLE__)
            // Apple pix formats
            case AV_PIX_FMT_VIDEOTOOLBOX:
                if (selected == 5) {
                    hw_de_av_pix_fmt_global = AV_PIX_FMT_VIDEOTOOLBOX;
                    hw_de_av_device_type_global = AV_HWDEVICE_TYPE_VIDEOTOOLBOX;
                    return *p;
                }
                break;
#endif
                // Cross-platform pix formats
            case AV_PIX_FMT_CUDA:
                if (selected == 2) {
                    hw_de_av_pix_fmt_global = AV_PIX_FMT_CUDA;
                    hw_de_av_device_type_global = AV_HWDEVICE_TYPE_CUDA;
                    return *p;
                }
                break;
            case AV_PIX_FMT_QSV:
                if (selected == 7) {
                    hw_de_av_pix_fmt_global = AV_PIX_FMT_QSV;
                    hw_de_av_device_type_global = AV_HWDEVICE_TYPE_QSV;
                    return *p;
                }
                break;
            default:
                // This is only here to silence unused-enum warnings
                break;
        }
    }
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::get_hw_dec_format (Unable to decode this file using hardware decode)");
    return AV_PIX_FMT_NONE;
}
 
int FFmpegReader::IsHardwareDecodeSupported(int codecid)
{
    int ret;
    switch (codecid) {
        case AV_CODEC_ID_H264:
        case AV_CODEC_ID_MPEG2VIDEO:
        case AV_CODEC_ID_VC1:
        case AV_CODEC_ID_WMV1:
        case AV_CODEC_ID_WMV2:
        case AV_CODEC_ID_WMV3:
            ret = 1;
            break;
        default :
            ret = 0;
            break;
    }
    return ret;
}
#endif // USE_HW_ACCEL
 
void FFmpegReader::Open() {
    // Open reader if not already open
    if (!is_open) {
        // Prevent async calls to the following code
        const std::lock_guard<std::recursive_mutex> lock(getFrameMutex);
 
        // Initialize format context
        pFormatCtx = NULL;
        {
            hw_de_on = (openshot::Settings::Instance()->HARDWARE_DECODER == 0 ? 0 : 1);
            ZmqLogger::Instance()->AppendDebugMethod("Decode hardware acceleration settings", "hw_de_on", hw_de_on, "HARDWARE_DECODER", openshot::Settings::Instance()->HARDWARE_DECODER);
        }
 
        // Open video file
        if (avformat_open_input(&pFormatCtx, path.c_str(), NULL, NULL) != 0)
            throw InvalidFile("FFmpegReader could not open media file.", path);
 
        // Retrieve stream information
        if (avformat_find_stream_info(pFormatCtx, NULL) < 0)
            throw NoStreamsFound("No streams found in file.", path);
 
        videoStream = -1;
        audioStream = -1;
 
        // Init end-of-file detection variables
        packet_status.reset(true);
 
        // Loop through each stream, and identify the video and audio stream index
        for (unsigned int i = 0; i < pFormatCtx->nb_streams; i++) {
            // Is this a video stream?
            if (AV_GET_CODEC_TYPE(pFormatCtx->streams[i]) == AVMEDIA_TYPE_VIDEO && videoStream < 0) {
                videoStream = i;
                packet_status.video_eof = false;
                packet_status.packets_eof = false;
                packet_status.end_of_file = false;
            }
            // Is this an audio stream?
            if (AV_GET_CODEC_TYPE(pFormatCtx->streams[i]) == AVMEDIA_TYPE_AUDIO && audioStream < 0) {
                audioStream = i;
                packet_status.audio_eof = false;
                packet_status.packets_eof = false;
                packet_status.end_of_file = false;
            }
        }
        if (videoStream == -1 && audioStream == -1)
            throw NoStreamsFound("No video or audio streams found in this file.", path);
 
        // Is there a video stream?
        if (videoStream != -1) {
            // Set the stream index
            info.video_stream_index = videoStream;
 
            // Set the codec and codec context pointers
            pStream = pFormatCtx->streams[videoStream];
 
            // Find the codec ID from stream
            const AVCodecID codecId = AV_FIND_DECODER_CODEC_ID(pStream);
 
            // Get codec and codec context from stream
            const AVCodec *pCodec = avcodec_find_decoder(codecId);
            AVDictionary *opts = NULL;
            int retry_decode_open = 2;
            // If hw accel is selected but hardware cannot handle repeat with software decoding
            do {
                pCodecCtx = AV_GET_CODEC_CONTEXT(pStream, pCodec);
#if USE_HW_ACCEL
                if (hw_de_on && (retry_decode_open==2)) {
                    // Up to here no decision is made if hardware or software decode
                    hw_de_supported = IsHardwareDecodeSupported(pCodecCtx->codec_id);
                }
#endif
                retry_decode_open = 0;
 
                // Set number of threads equal to number of processors (not to exceed 16)
                pCodecCtx->thread_count = std::min(FF_VIDEO_NUM_PROCESSORS, 16);
 
                if (pCodec == NULL) {
                    throw InvalidCodec("A valid video codec could not be found for this file.", path);
                }
 
                // Init options
                av_dict_set(&opts, "strict", "experimental", 0);
#if USE_HW_ACCEL
                if (hw_de_on && hw_de_supported) {
                    // Open Hardware Acceleration
                    int i_decoder_hw = 0;
                    char adapter[256];
                    char *adapter_ptr = NULL;
                    int adapter_num;
                    adapter_num = openshot::Settings::Instance()->HW_DE_DEVICE_SET;
                    ZmqLogger::Instance()->AppendDebugMethod("Hardware decoding device number", "adapter_num", adapter_num);
 
                    // Set hardware pix format (callback)
                    pCodecCtx->get_format = get_hw_dec_format;
 
                    if (adapter_num < 3 && adapter_num >=0) {
#if defined(__linux__)
                        snprintf(adapter,sizeof(adapter),"/dev/dri/renderD%d", adapter_num+128);
                        adapter_ptr = adapter;
                        i_decoder_hw = openshot::Settings::Instance()->HARDWARE_DECODER;
                        switch (i_decoder_hw) {
                                case 1:
                                    hw_de_av_device_type = AV_HWDEVICE_TYPE_VAAPI;
                                    break;
                                case 2:
                                    hw_de_av_device_type = AV_HWDEVICE_TYPE_CUDA;
                                    break;
                                case 6:
                                    hw_de_av_device_type = AV_HWDEVICE_TYPE_VDPAU;
                                    break;
                                case 7:
                                    hw_de_av_device_type = AV_HWDEVICE_TYPE_QSV;
                                    break;
                                default:
                                    hw_de_av_device_type = AV_HWDEVICE_TYPE_VAAPI;
                                    break;
                            }
 
#elif defined(_WIN32)
                        adapter_ptr = NULL;
                        i_decoder_hw = openshot::Settings::Instance()->HARDWARE_DECODER;
                        switch (i_decoder_hw) {
                            case 2:
                                hw_de_av_device_type = AV_HWDEVICE_TYPE_CUDA;
                                break;
                            case 3:
                                hw_de_av_device_type = AV_HWDEVICE_TYPE_DXVA2;
                                break;
                            case 4:
                                hw_de_av_device_type = AV_HWDEVICE_TYPE_D3D11VA;
                                break;
                            case 7:
                                hw_de_av_device_type = AV_HWDEVICE_TYPE_QSV;
                                break;
                            default:
                                hw_de_av_device_type = AV_HWDEVICE_TYPE_DXVA2;
                                break;
                        }
#elif defined(__APPLE__)
                        adapter_ptr = NULL;
                        i_decoder_hw = openshot::Settings::Instance()->HARDWARE_DECODER;
                        switch (i_decoder_hw) {
                            case 5:
                                hw_de_av_device_type =  AV_HWDEVICE_TYPE_VIDEOTOOLBOX;
                                break;
                            case 7:
                                hw_de_av_device_type = AV_HWDEVICE_TYPE_QSV;
                                break;
                            default:
                                hw_de_av_device_type = AV_HWDEVICE_TYPE_VIDEOTOOLBOX;
                                break;
                        }
#endif
 
                    } else {
                        adapter_ptr = NULL; // Just to be sure
                    }
 
                    // Check if it is there and writable
#if defined(__linux__)
                    if( adapter_ptr != NULL && access( adapter_ptr, W_OK ) == 0 ) {
#elif defined(_WIN32)
                    if( adapter_ptr != NULL ) {
#elif defined(__APPLE__)
                    if( adapter_ptr != NULL ) {
#endif
                        ZmqLogger::Instance()->AppendDebugMethod("Decode Device present using device");
                    }
                    else {
                        adapter_ptr = NULL;  // use default
                        ZmqLogger::Instance()->AppendDebugMethod("Decode Device not present using default");
                    }
 
                    hw_device_ctx = NULL;
                    // Here the first hardware initialisations are made
                    if (av_hwdevice_ctx_create(&hw_device_ctx, hw_de_av_device_type, adapter_ptr, NULL, 0) >= 0) {
                        const char* hw_name = av_hwdevice_get_type_name(hw_de_av_device_type);
                        std::string hw_msg = "HW decode active: ";
                        hw_msg += (hw_name ? hw_name : "unknown");
                        ZmqLogger::Instance()->Log(hw_msg);
                        if (!(pCodecCtx->hw_device_ctx = av_buffer_ref(hw_device_ctx))) {
                            throw InvalidCodec("Hardware device reference create failed.", path);
                        }
 
                        /*
                        av_buffer_unref(&ist->hw_frames_ctx);
                        ist->hw_frames_ctx = av_hwframe_ctx_alloc(hw_device_ctx);
                        if (!ist->hw_frames_ctx) {
                            av_log(avctx, AV_LOG_ERROR, "Error creating a CUDA frames context\n");
                            return AVERROR(ENOMEM);
                        }
 
                        frames_ctx = (AVHWFramesContext*)ist->hw_frames_ctx->data;
 
                        frames_ctx->format = AV_PIX_FMT_CUDA;
                        frames_ctx->sw_format = avctx->sw_pix_fmt;
                        frames_ctx->width = avctx->width;
                        frames_ctx->height = avctx->height;
 
                        av_log(avctx, AV_LOG_DEBUG, "Initializing CUDA frames context: sw_format = %s, width = %d, height = %d\n",
                                av_get_pix_fmt_name(frames_ctx->sw_format), frames_ctx->width, frames_ctx->height);
 
 
                        ret = av_hwframe_ctx_init(pCodecCtx->hw_device_ctx);
                        ret = av_hwframe_ctx_init(ist->hw_frames_ctx);
                        if (ret < 0) {
                          av_log(avctx, AV_LOG_ERROR, "Error initializing a CUDA frame pool\n");
                          return ret;
                        }
                        */
                    }
                    else {
                        ZmqLogger::Instance()->Log("HW decode active: no (falling back to software)");
                        throw InvalidCodec("Hardware device create failed.", path);
                    }
                }
#endif // USE_HW_ACCEL
 
                // Disable per-frame threading for album arts
                // Using FF_THREAD_FRAME adds one frame decoding delay per thread,
                // but there's only one frame in this case.
                if (HasAlbumArt())
                {
                    pCodecCtx->thread_type &= ~FF_THREAD_FRAME;
                }
 
                // Open video codec
                int avcodec_return = avcodec_open2(pCodecCtx, pCodec, &opts);
                if (avcodec_return < 0) {
                    std::stringstream avcodec_error_msg;
                    avcodec_error_msg << "A video codec was found, but could not be opened. Error: " << av_err2string(avcodec_return);
                    throw InvalidCodec(avcodec_error_msg.str(), path);
                }
 
#if USE_HW_ACCEL
                if (hw_de_on && hw_de_supported) {
                    AVHWFramesConstraints *constraints = NULL;
                    void *hwconfig = NULL;
                    hwconfig = av_hwdevice_hwconfig_alloc(hw_device_ctx);
 
// TODO: needs va_config!
#if ENABLE_VAAPI
                    ((AVVAAPIHWConfig *)hwconfig)->config_id = ((VAAPIDecodeContext *)(pCodecCtx->priv_data))->va_config;
                    constraints = av_hwdevice_get_hwframe_constraints(hw_device_ctx,hwconfig);
#endif // ENABLE_VAAPI
                    if (constraints) {
                        if (pCodecCtx->coded_width < constraints->min_width     ||
                                pCodecCtx->coded_height < constraints->min_height ||
                                pCodecCtx->coded_width > constraints->max_width     ||
                                pCodecCtx->coded_height > constraints->max_height) {
                            ZmqLogger::Instance()->AppendDebugMethod("DIMENSIONS ARE TOO LARGE for hardware acceleration\n");
                            hw_de_supported = 0;
                            retry_decode_open = 1;
                            AV_FREE_CONTEXT(pCodecCtx);
                            if (hw_device_ctx) {
                                av_buffer_unref(&hw_device_ctx);
                                hw_device_ctx = NULL;
                            }
                        }
                        else {
                            // All is just peachy
                            ZmqLogger::Instance()->AppendDebugMethod("\nDecode hardware acceleration is used\n", "Min width :", constraints->min_width, "Min Height :", constraints->min_height, "MaxWidth :", constraints->max_width, "MaxHeight :", constraints->max_height, "Frame width :", pCodecCtx->coded_width, "Frame height :", pCodecCtx->coded_height);
                            retry_decode_open = 0;
                        }
                        av_hwframe_constraints_free(&constraints);
                        if (hwconfig) {
                            av_freep(&hwconfig);
                        }
                    }
                    else {
                        int max_h, max_w;
                        //max_h = ((getenv( "LIMIT_HEIGHT_MAX" )==NULL) ? MAX_SUPPORTED_HEIGHT : atoi(getenv( "LIMIT_HEIGHT_MAX" )));
                        max_h = openshot::Settings::Instance()->DE_LIMIT_HEIGHT_MAX;
                        //max_w = ((getenv( "LIMIT_WIDTH_MAX" )==NULL) ? MAX_SUPPORTED_WIDTH : atoi(getenv( "LIMIT_WIDTH_MAX" )));
                        max_w = openshot::Settings::Instance()->DE_LIMIT_WIDTH_MAX;
                        ZmqLogger::Instance()->AppendDebugMethod("Constraints could not be found using default limit\n");
                        //cerr << "Constraints could not be found using default limit\n";
                        if (pCodecCtx->coded_width < 0      ||
                                pCodecCtx->coded_height < 0     ||
                                pCodecCtx->coded_width > max_w ||
                                pCodecCtx->coded_height > max_h ) {
                            ZmqLogger::Instance()->AppendDebugMethod("DIMENSIONS ARE TOO LARGE for hardware acceleration\n", "Max Width :", max_w, "Max Height :", max_h, "Frame width :", pCodecCtx->coded_width, "Frame height :", pCodecCtx->coded_height);
                            hw_de_supported = 0;
                            retry_decode_open = 1;
                            AV_FREE_CONTEXT(pCodecCtx);
                            if (hw_device_ctx) {
                                av_buffer_unref(&hw_device_ctx);
                                hw_device_ctx = NULL;
                            }
                        }
                        else {
                            ZmqLogger::Instance()->AppendDebugMethod("\nDecode hardware acceleration is used\n", "Max Width :", max_w, "Max Height :", max_h, "Frame width :", pCodecCtx->coded_width, "Frame height :", pCodecCtx->coded_height);
                            retry_decode_open = 0;
                        }
                    }
                } // if hw_de_on && hw_de_supported
                else {
                    ZmqLogger::Instance()->AppendDebugMethod("\nDecode in software is used\n");
                }
#else
                retry_decode_open = 0;
#endif // USE_HW_ACCEL
            } while (retry_decode_open); // retry_decode_open
            // Free options
            av_dict_free(&opts);
 
            // Update the File Info struct with video details (if a video stream is found)
            UpdateVideoInfo();
        }
 
        // Is there an audio stream?
        if (audioStream != -1) {
            // Set the stream index
            info.audio_stream_index = audioStream;
 
            // Get a pointer to the codec context for the audio stream
            aStream = pFormatCtx->streams[audioStream];
 
            // Find the codec ID from stream
            AVCodecID codecId = AV_FIND_DECODER_CODEC_ID(aStream);
 
            // Get codec and codec context from stream
            const AVCodec *aCodec = avcodec_find_decoder(codecId);
            aCodecCtx = AV_GET_CODEC_CONTEXT(aStream, aCodec);
 
            // Audio encoding does not typically use more than 2 threads (most codecs use 1 thread)
            aCodecCtx->thread_count = std::min(FF_AUDIO_NUM_PROCESSORS, 2);
 
            bool audio_opened = false;
            if (aCodec != NULL) {
                // Init options
                AVDictionary *opts = NULL;
                av_dict_set(&opts, "strict", "experimental", 0);
 
                // Open audio codec
                audio_opened = (avcodec_open2(aCodecCtx, aCodec, &opts) >= 0);
 
                // Free options
                av_dict_free(&opts);
            }
 
            if (audio_opened) {
                // Update the File Info struct with audio details (if an audio stream is found)
                UpdateAudioInfo();
 
                // Disable malformed audio stream metadata (prevents divide-by-zero / invalid resampling math)
                const bool invalid_audio_info =
                    (info.channels <= 0) ||
                    (info.sample_rate <= 0) ||
                    (info.audio_timebase.num <= 0) ||
                    (info.audio_timebase.den <= 0) ||
                    (aCodecCtx->sample_fmt == AV_SAMPLE_FMT_NONE);
                if (invalid_audio_info) {
                    ZmqLogger::Instance()->AppendDebugMethod(
                        "FFmpegReader::Open (Disable invalid audio stream)",
                        "channels", info.channels,
                        "sample_rate", info.sample_rate,
                        "audio_timebase.num", info.audio_timebase.num,
                        "audio_timebase.den", info.audio_timebase.den,
                        "sample_fmt", static_cast<int>(aCodecCtx ? aCodecCtx->sample_fmt : AV_SAMPLE_FMT_NONE));
                    info.has_audio = false;
                    info.audio_stream_index = -1;
                    audioStream = -1;
                    packet_status.audio_eof = true;
                    if (aCodecCtx) {
                        if (avcodec_is_open(aCodecCtx)) {
                            avcodec_flush_buffers(aCodecCtx);
                        }
                        AV_FREE_CONTEXT(aCodecCtx);
                        aCodecCtx = nullptr;
                    }
                    aStream = nullptr;
                }
            } else {
                // Keep decoding video, but disable bad/unsupported audio stream.
                ZmqLogger::Instance()->AppendDebugMethod(
                    "FFmpegReader::Open (Audio codec unavailable; disabling audio)",
                    "audioStream", audioStream);
                info.has_audio = false;
                info.audio_stream_index = -1;
                audioStream = -1;
                packet_status.audio_eof = true;
                if (aCodecCtx) {
                    AV_FREE_CONTEXT(aCodecCtx);
                    aCodecCtx = nullptr;
                }
                aStream = nullptr;
            }
        }
 
        // Guard invalid frame-rate / timebase values from malformed streams.
        if (info.fps.num <= 0 || info.fps.den <= 0) {
            ZmqLogger::Instance()->AppendDebugMethod(
                "FFmpegReader::Open (Invalid FPS detected; applying fallback)",
                "fps.num", info.fps.num,
                "fps.den", info.fps.den);
            info.fps.num = 30;
            info.fps.den = 1;
        }
        if (info.video_timebase.num <= 0 || info.video_timebase.den <= 0) {
            ZmqLogger::Instance()->AppendDebugMethod(
                "FFmpegReader::Open (Invalid video_timebase detected; applying fallback)",
                "video_timebase.num", info.video_timebase.num,
                "video_timebase.den", info.video_timebase.den);
            info.video_timebase = info.fps.Reciprocal();
        }
 
        // Add format metadata (if any)
        AVDictionaryEntry *tag = NULL;
        while ((tag = av_dict_get(pFormatCtx->metadata, "", tag, AV_DICT_IGNORE_SUFFIX))) {
            QString str_key = tag->key;
            QString str_value = tag->value;
            info.metadata[str_key.toStdString()] = str_value.trimmed().toStdString();
        }
 
        // Process video stream side data (rotation, spherical metadata, etc)
        for (unsigned int i = 0; i < pFormatCtx->nb_streams; i++) {
            AVStream* st = pFormatCtx->streams[i];
            if (st->codecpar->codec_type == AVMEDIA_TYPE_VIDEO) {
                // Only inspect the first video stream
                for (int j = 0; j < st->nb_side_data; j++) {
                    AVPacketSideData *sd = &st->side_data[j];
 
                    // Handle rotation metadata (unchanged)
                    if (sd->type == AV_PKT_DATA_DISPLAYMATRIX &&
                        sd->size >= 9 * sizeof(int32_t) &&
                        !info.metadata.count("rotate"))
                    {
                        double rotation = -av_display_rotation_get(
                            reinterpret_cast<int32_t *>(sd->data));
                        if (std::isnan(rotation)) rotation = 0;
                        info.metadata["rotate"] = std::to_string(rotation);
                    }
                    // Handle spherical video metadata
                    else if (sd->type == AV_PKT_DATA_SPHERICAL) {
                        // Always mark as spherical
                        info.metadata["spherical"] = "1";
 
                        // Cast the raw bytes to an AVSphericalMapping
                        const AVSphericalMapping* map =
                            reinterpret_cast<const AVSphericalMapping*>(sd->data);
 
                        // Projection enum → string
                        const char* proj_name = av_spherical_projection_name(map->projection);
                        info.metadata["spherical_projection"] = proj_name
                            ? proj_name
                            : "unknown";
 
                        // Convert 16.16 fixed-point to float degrees
                        auto to_deg = [](int32_t v){
                            return (double)v / 65536.0;
                        };
                        info.metadata["spherical_yaw"]   = std::to_string(to_deg(map->yaw));
                        info.metadata["spherical_pitch"] = std::to_string(to_deg(map->pitch));
                        info.metadata["spherical_roll"]  = std::to_string(to_deg(map->roll));
                    }
                }
                break;
            }
        }
 
        // Init previous audio location to zero
        previous_packet_location.frame = -1;
        previous_packet_location.sample_start = 0;
 
        // Adjust cache size based on size of frame and audio
        const int working_cache_frames = std::max(Settings::Instance()->CACHE_MIN_FRAMES, int(OPEN_MP_NUM_PROCESSORS * info.fps.ToDouble() * 2));
        const int final_cache_frames = std::max(Settings::Instance()->CACHE_MIN_FRAMES, OPEN_MP_NUM_PROCESSORS * 2);
        working_cache.SetMaxBytesFromInfo(working_cache_frames, info.width, info.height, info.sample_rate, info.channels);
        final_cache.SetMaxBytesFromInfo(final_cache_frames, info.width, info.height, info.sample_rate, info.channels);
 
        // Scan PTS for any offsets (i.e. non-zero starting streams). At least 1 stream must start at zero timestamp.
        // This method allows us to shift timestamps to ensure at least 1 stream is starting at zero.
        UpdatePTSOffset();
 
        // Override an invalid framerate
        if (info.fps.ToFloat() > 240.0f || (info.fps.num <= 0 || info.fps.den <= 0) || info.video_length <= 0) {
            // Calculate FPS, duration, video bit rate, and video length manually
            // by scanning through all the video stream packets
            CheckFPS();
        }
 
        // Mark as "open"
        is_open = true;
 
        // Seek back to beginning of file (if not already seeking)
        if (!is_seeking) {
            Seek(1);
        }
    }
}
 
void FFmpegReader::Close() {
    // Close all objects, if reader is 'open'
    if (is_open) {
        // Prevent async calls to the following code
        const std::lock_guard<std::recursive_mutex> lock(getFrameMutex);
 
        // Mark as "closed"
        is_open = false;
 
        // Keep track of most recent packet
        AVPacket *recent_packet = packet;
 
        // Drain any packets from the decoder
        packet = NULL;
        int attempts = 0;
        int max_attempts = 128;
        while (packet_status.packets_decoded() < packet_status.packets_read() && attempts < max_attempts) {
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::Close (Drain decoder loop)",
                                                     "packets_read", packet_status.packets_read(),
                                                     "packets_decoded", packet_status.packets_decoded(),
                                                     "attempts", attempts);
            if (packet_status.video_decoded < packet_status.video_read) {
                ProcessVideoPacket(info.video_length);
            }
            if (packet_status.audio_decoded < packet_status.audio_read) {
                ProcessAudioPacket(info.video_length);
            }
            attempts++;
        }
 
        // Remove packet
        if (recent_packet) {
            RemoveAVPacket(recent_packet);
        }
 
        // Close the video codec
        if (info.has_video) {
            if(avcodec_is_open(pCodecCtx)) {
                avcodec_flush_buffers(pCodecCtx);
            }
            AV_FREE_CONTEXT(pCodecCtx);
#if USE_HW_ACCEL
            if (hw_de_on) {
                if (hw_device_ctx) {
                    av_buffer_unref(&hw_device_ctx);
                    hw_device_ctx = NULL;
                }
            }
#endif // USE_HW_ACCEL
            if (img_convert_ctx) {
                sws_freeContext(img_convert_ctx);
                img_convert_ctx = nullptr;
            }
            if (pFrameRGB_cached) {
                AV_FREE_FRAME(&pFrameRGB_cached);
            }
        }
 
        // Close the audio codec
        if (info.has_audio) {
            if(avcodec_is_open(aCodecCtx)) {
                avcodec_flush_buffers(aCodecCtx);
            }
            AV_FREE_CONTEXT(aCodecCtx);
            if (avr_ctx) {
                SWR_CLOSE(avr_ctx);
                SWR_FREE(&avr_ctx);
                avr_ctx = nullptr;
            }
        }
 
        // Clear final cache
        final_cache.Clear();
        working_cache.Clear();
 
        // Close the video file
        avformat_close_input(&pFormatCtx);
        av_freep(&pFormatCtx);
 
        // Do not trim here; trimming is handled on explicit cache clears
 
        // Reset some variables
        last_frame = 0;
        hold_packet = false;
        largest_frame_processed = 0;
        seek_audio_frame_found = 0;
        seek_video_frame_found = 0;
        current_video_frame = 0;
        last_video_frame.reset();
        last_final_video_frame.reset();
    }
}
 
bool FFmpegReader::HasAlbumArt() {
    // Check if the video stream we use is an attached picture
    // This won't return true if the file has a cover image as a secondary stream
    // like an MKV file with an attached image file
    return pFormatCtx && videoStream >= 0 && pFormatCtx->streams[videoStream]
        && (pFormatCtx->streams[videoStream]->disposition & AV_DISPOSITION_ATTACHED_PIC);
}
 
double FFmpegReader::PickDurationSeconds() const {
    auto has_value = [](double value) { return value > 0.0; };
 
    switch (duration_strategy) {
        case DurationStrategy::VideoPreferred:
            if (has_value(video_stream_duration_seconds))
                return video_stream_duration_seconds;
            if (has_value(audio_stream_duration_seconds))
                return audio_stream_duration_seconds;
            if (has_value(format_duration_seconds))
                return format_duration_seconds;
            break;
        case DurationStrategy::AudioPreferred:
            if (has_value(audio_stream_duration_seconds))
                return audio_stream_duration_seconds;
            if (has_value(video_stream_duration_seconds))
                return video_stream_duration_seconds;
            if (has_value(format_duration_seconds))
                return format_duration_seconds;
            break;
        case DurationStrategy::LongestStream:
        default:
            {
                double longest = 0.0;
                if (has_value(video_stream_duration_seconds))
                    longest = std::max(longest, video_stream_duration_seconds);
                if (has_value(audio_stream_duration_seconds))
                    longest = std::max(longest, audio_stream_duration_seconds);
                if (has_value(format_duration_seconds))
                    longest = std::max(longest, format_duration_seconds);
                if (has_value(longest))
                    return longest;
            }
            break;
    }
 
    if (has_value(format_duration_seconds))
        return format_duration_seconds;
    if (has_value(inferred_duration_seconds))
        return inferred_duration_seconds;
 
    return 0.0;
}
 
void FFmpegReader::ApplyDurationStrategy() {
    const double fps_value = info.fps.ToDouble();
    const double chosen_seconds = PickDurationSeconds();
 
    if (chosen_seconds <= 0.0 || fps_value <= 0.0) {
        info.duration = 0.0f;
        info.video_length = 0;
        is_duration_known = false;
        return;
    }
 
    const int64_t frames = static_cast<int64_t>(std::llround(chosen_seconds * fps_value));
    if (frames <= 0) {
        info.duration = 0.0f;
        info.video_length = 0;
        is_duration_known = false;
        return;
    }
 
    info.video_length = frames;
    info.duration = static_cast<float>(static_cast<double>(frames) / fps_value);
    is_duration_known = true;
}
 
void FFmpegReader::UpdateAudioInfo() {
    const int codec_channels =
#if HAVE_CH_LAYOUT
        AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->ch_layout.nb_channels;
#else
        AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channels;
#endif
 
    // Set default audio channel layout (if needed)
#if HAVE_CH_LAYOUT
    if (codec_channels > 0 &&
        !av_channel_layout_check(&(AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->ch_layout)))
        AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->ch_layout = (AVChannelLayout) AV_CHANNEL_LAYOUT_STEREO;
#else
    if (codec_channels > 0 && AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout == 0)
        AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout = av_get_default_channel_layout(AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channels);
#endif
 
    if (info.sample_rate > 0) {
        // Skip init - if info struct already populated
        return;
    }
 
    auto record_duration = [](double &target, double seconds) {
        if (seconds > 0.0)
            target = std::max(target, seconds);
    };
 
    // Set values of FileInfo struct
    info.has_audio = true;
    info.file_size = pFormatCtx->pb ? avio_size(pFormatCtx->pb) : -1;
    info.acodec = aCodecCtx->codec->name;
#if HAVE_CH_LAYOUT
    info.channels = AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->ch_layout.nb_channels;
    info.channel_layout = (ChannelLayout) AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->ch_layout.u.mask;
#else
    info.channels = AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channels;
    info.channel_layout = (ChannelLayout) AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout;
#endif
 
    // If channel layout is not set, guess based on the number of channels
    if (info.channel_layout == 0) {
        if (info.channels == 1) {
            info.channel_layout = openshot::LAYOUT_MONO;
        } else if (info.channels == 2) {
            info.channel_layout = openshot::LAYOUT_STEREO;
        }
    }
 
    info.sample_rate = AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->sample_rate;
    info.audio_bit_rate = AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->bit_rate;
    if (info.audio_bit_rate <= 0) {
        // Get bitrate from format
        info.audio_bit_rate = pFormatCtx->bit_rate;
    }
 
    // Set audio timebase
    info.audio_timebase.num = aStream->time_base.num;
    info.audio_timebase.den = aStream->time_base.den;
 
    // Get timebase of audio stream (if valid) and greater than the current duration
    if (aStream->duration > 0) {
        record_duration(audio_stream_duration_seconds, aStream->duration * info.audio_timebase.ToDouble());
    }
    if (pFormatCtx->duration > 0) {
        // Use the format's duration when stream duration is missing or shorter
        record_duration(format_duration_seconds, static_cast<double>(pFormatCtx->duration) / AV_TIME_BASE);
    }
 
    // Calculate duration from filesize and bitrate (if any)
    if (info.duration <= 0.0f && info.video_bit_rate > 0 && info.file_size > 0) {
        // Estimate from bitrate, total bytes, and framerate
        record_duration(inferred_duration_seconds, static_cast<double>(info.file_size) / info.video_bit_rate);
    }
 
    // Set video timebase (if no video stream was found)
    if (!info.has_video) {
        // Set a few important default video settings (so audio can be divided into frames)
        info.fps.num = 30;
        info.fps.den = 1;
        info.video_timebase.num = 1;
        info.video_timebase.den = 30;
        info.width = 720;
        info.height = 480;
 
        // Use timeline to set correct width & height (if any)
        Clip *parent = static_cast<Clip *>(ParentClip());
        if (parent) {
            if (parent->ParentTimeline()) {
                // Set max width/height based on parent clip's timeline (if attached to a timeline)
                info.width = parent->ParentTimeline()->preview_width;
                info.height = parent->ParentTimeline()->preview_height;
            }
        }
    }
 
    ApplyDurationStrategy();
 
    // Add audio metadata (if any found)
    AVDictionaryEntry *tag = NULL;
    while ((tag = av_dict_get(aStream->metadata, "", tag, AV_DICT_IGNORE_SUFFIX))) {
        QString str_key = tag->key;
        QString str_value = tag->value;
        info.metadata[str_key.toStdString()] = str_value.trimmed().toStdString();
    }
}
 
void FFmpegReader::UpdateVideoInfo() {
    if (info.vcodec.length() > 0) {
        // Skip init - if info struct already populated
        return;
    }
 
    auto record_duration = [](double &target, double seconds) {
        if (seconds > 0.0)
            target = std::max(target, seconds);
    };
 
    // Set values of FileInfo struct
    info.has_video = true;
    info.file_size = pFormatCtx->pb ? avio_size(pFormatCtx->pb) : -1;
    info.height = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->height;
    info.width = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->width;
    info.vcodec = pCodecCtx->codec->name;
    info.video_bit_rate = (pFormatCtx->bit_rate / 8);
 
    // Frame rate from the container and codec
    AVRational framerate = av_guess_frame_rate(pFormatCtx, pStream, NULL);
    if (!check_fps) {
        info.fps.num = framerate.num;
        info.fps.den = framerate.den;
    }
 
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::UpdateVideoInfo", "info.fps.num", info.fps.num, "info.fps.den", info.fps.den);
 
    // TODO: remove excessive debug info in the next releases
    // The debug info below is just for comparison and troubleshooting on users side during the transition period
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::UpdateVideoInfo (pStream->avg_frame_rate)", "num", pStream->avg_frame_rate.num, "den", pStream->avg_frame_rate.den);
 
    if (pStream->sample_aspect_ratio.num != 0) {
        info.pixel_ratio.num = pStream->sample_aspect_ratio.num;
        info.pixel_ratio.den = pStream->sample_aspect_ratio.den;
    } else if (AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->sample_aspect_ratio.num != 0) {
        info.pixel_ratio.num = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->sample_aspect_ratio.num;
        info.pixel_ratio.den = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->sample_aspect_ratio.den;
    } else {
        info.pixel_ratio.num = 1;
        info.pixel_ratio.den = 1;
    }
    info.pixel_format = AV_GET_CODEC_PIXEL_FORMAT(pStream, pCodecCtx);
 
    // Calculate the DAR (display aspect ratio)
    Fraction size(info.width * info.pixel_ratio.num, info.height * info.pixel_ratio.den);
 
    // Reduce size fraction
    size.Reduce();
 
    // Set the ratio based on the reduced fraction
    info.display_ratio.num = size.num;
    info.display_ratio.den = size.den;
 
    // Get scan type and order from codec context/params
    if (!check_interlace) {
        check_interlace = true;
        AVFieldOrder field_order = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->field_order;
        switch(field_order) {
            case AV_FIELD_PROGRESSIVE:
                info.interlaced_frame = false;
                break;
            case AV_FIELD_TT:
            case AV_FIELD_TB:
                info.interlaced_frame = true;
                info.top_field_first = true;
                break;
            case AV_FIELD_BT:
            case AV_FIELD_BB:
                info.interlaced_frame = true;
                info.top_field_first = false;
                break;
            case AV_FIELD_UNKNOWN:
                // Check again later?
                check_interlace = false;
                break;
        }
        // check_interlace will prevent these checks being repeated,
        // unless it was cleared because we got an AV_FIELD_UNKNOWN response.
    }
 
    // Set the video timebase
    info.video_timebase.num = pStream->time_base.num;
    info.video_timebase.den = pStream->time_base.den;
 
    // Set the duration in seconds, and video length (# of frames)
    record_duration(video_stream_duration_seconds, pStream->duration * info.video_timebase.ToDouble());
 
    // Check for valid duration (if found)
    if (pFormatCtx->duration >= 0) {
        // Use the format's duration as another candidate
        record_duration(format_duration_seconds, static_cast<double>(pFormatCtx->duration) / AV_TIME_BASE);
    }
 
    // Calculate duration from filesize and bitrate (if any)
    if (info.video_bit_rate > 0 && info.file_size > 0) {
        // Estimate from bitrate, total bytes, and framerate
        record_duration(inferred_duration_seconds, static_cast<double>(info.file_size) / info.video_bit_rate);
    }
 
    // Certain "image" formats do not have a valid duration
    if (video_stream_duration_seconds <= 0.0 && format_duration_seconds <= 0.0 &&
            pStream->duration == AV_NOPTS_VALUE && pFormatCtx->duration == AV_NOPTS_VALUE) {
        // Force an "image" duration
        record_duration(video_stream_duration_seconds, 60 * 60 * 1);  // 1 hour duration
        info.has_single_image = true;
    }
    // Static GIFs can have no usable duration; fall back to a small default
    if (video_stream_duration_seconds <= 0.0 && format_duration_seconds <= 0.0 &&
            pFormatCtx && pFormatCtx->iformat && strcmp(pFormatCtx->iformat->name, "gif") == 0) {
        record_duration(video_stream_duration_seconds, 60 * 60 * 1);  // 1 hour duration
        info.has_single_image = true;
    }
 
    ApplyDurationStrategy();
 
    // Normalize FFmpeg-decoded still images (e.g. JPG/JPEG) to match image-reader behavior.
    // This keeps timing/flags consistent regardless of which reader path was used.
    if (!info.has_single_image && audioStream < 0) {
        const AVCodecID codec_id = AV_FIND_DECODER_CODEC_ID(pStream);
        const bool likely_still_codec =
            codec_id == AV_CODEC_ID_MJPEG ||
            codec_id == AV_CODEC_ID_PNG ||
            codec_id == AV_CODEC_ID_BMP ||
            codec_id == AV_CODEC_ID_TIFF ||
            codec_id == AV_CODEC_ID_WEBP ||
            codec_id == AV_CODEC_ID_JPEG2000;
        const bool likely_image_demuxer =
            pFormatCtx && pFormatCtx->iformat && pFormatCtx->iformat->name &&
            strstr(pFormatCtx->iformat->name, "image2");
        const bool single_frame_clip = info.video_length <= 1;
 
        if (single_frame_clip && (likely_still_codec || likely_image_demuxer)) {
            info.has_single_image = true;
            record_duration(video_stream_duration_seconds, 60 * 60 * 1);  // 1 hour duration
            ApplyDurationStrategy();
        }
    }
 
    // Add video metadata (if any)
    AVDictionaryEntry *tag = NULL;
    while ((tag = av_dict_get(pStream->metadata, "", tag, AV_DICT_IGNORE_SUFFIX))) {
        QString str_key = tag->key;
        QString str_value = tag->value;
        info.metadata[str_key.toStdString()] = str_value.trimmed().toStdString();
    }
}
 
bool FFmpegReader::GetIsDurationKnown() {
    return this->is_duration_known;
}
 
std::shared_ptr<Frame> FFmpegReader::GetFrame(int64_t requested_frame) {
    last_seek_max_frame = -1;
    seek_stagnant_count = 0;
    // Check for open reader (or throw exception)
    if (!is_open)
        throw ReaderClosed("The FFmpegReader is closed.  Call Open() before calling this method.", path);
 
    // Adjust for a requested frame that is too small or too large
    if (requested_frame < 1)
        requested_frame = 1;
    if (requested_frame > info.video_length && is_duration_known)
        requested_frame = info.video_length;
    if (info.has_video && info.video_length == 0)
        // Invalid duration of video file
        throw InvalidFile("Could not detect the duration of the video or audio stream.", path);
 
    // Debug output
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetFrame", "requested_frame", requested_frame, "last_frame", last_frame);
 
    // Check the cache for this frame
    std::shared_ptr<Frame> frame = final_cache.GetFrame(requested_frame);
    if (frame) {
        // Debug output
        ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetFrame", "returned cached frame", requested_frame);
 
        // Return the cached frame
        return frame;
    } else {
 
        // Prevent async calls to the remainder of this code
        const std::lock_guard<std::recursive_mutex> lock(getFrameMutex);
 
        // Check the cache a 2nd time (due to the potential previous lock)
        frame = final_cache.GetFrame(requested_frame);
        if (frame) {
            // Debug output
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetFrame", "returned cached frame on 2nd look", requested_frame);
 
        } else {
            // Frame is not in cache
            // Reset seek count
            seek_count = 0;
 
            // Are we within X frames of the requested frame?
            int64_t diff = requested_frame - last_frame;
            if (diff >= 1 && diff <= 20) {
                // Continue walking the stream
                frame = ReadStream(requested_frame);
            } else {
                // Greater than 30 frames away, or backwards, we need to seek to the nearest key frame
                if (enable_seek) {
                    // Only seek if enabled
                    Seek(requested_frame);
 
                } else if (!enable_seek && diff < 0) {
                    // Start over, since we can't seek, and the requested frame is smaller than our position
                    // Since we are seeking to frame 1, this actually just closes/re-opens the reader
                    Seek(1);
                }
 
                // Then continue walking the stream
                frame = ReadStream(requested_frame);
            }
        }
        return frame;
    }
}
 
// Read the stream until we find the requested Frame
std::shared_ptr<Frame> FFmpegReader::ReadStream(int64_t requested_frame) {
    // Allocate video frame
    bool check_seek = false;
    int packet_error = -1;
    int64_t no_progress_count = 0;
    int64_t prev_packets_read = packet_status.packets_read();
    int64_t prev_packets_decoded = packet_status.packets_decoded();
    int64_t prev_video_decoded = packet_status.video_decoded;
    double prev_video_pts_seconds = video_pts_seconds;
 
    // Debug output
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ReadStream", "requested_frame", requested_frame);
 
    // Loop through the stream until the correct frame is found
    while (true) {
        // Check if working frames are 'finished'
        if (!is_seeking) {
            // Check for final frames
            CheckWorkingFrames(requested_frame);
        }
 
        // Check if requested 'final' frame is available (and break out of loop if found)
        bool is_cache_found = (final_cache.GetFrame(requested_frame) != NULL);
        if (is_cache_found) {
            break;
        }
 
        if (!hold_packet || !packet) {
            // Get the next packet
            packet_error = GetNextPacket();
            if (packet_error < 0 && !packet) {
                // No more packets to be found
                packet_status.packets_eof = true;
            }
        }
 
        // Debug output
        ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ReadStream (GetNextPacket)", "requested_frame", requested_frame,"packets_read", packet_status.packets_read(), "packets_decoded", packet_status.packets_decoded(), "is_seeking", is_seeking);
 
        // Check the status of a seek (if any)
        if (is_seeking) {
            check_seek = CheckSeek();
        } else {
            check_seek = false;
        }
 
        if (check_seek) {
            // Packet may become NULL on Close inside Seek if CheckSeek returns false
            // Jump to the next iteration of this loop
            continue;
        }
 
        // Video packet
        if ((info.has_video && packet && packet->stream_index == videoStream) ||
            (info.has_video && packet_status.video_decoded < packet_status.video_read) ||
            (info.has_video && !packet && !packet_status.video_eof)) {
            // Process Video Packet
            ProcessVideoPacket(requested_frame);
        }
        // Audio packet
        if ((info.has_audio && packet && packet->stream_index == audioStream) ||
            (info.has_audio && !packet && packet_status.audio_decoded < packet_status.audio_read) ||
            (info.has_audio && !packet && !packet_status.audio_eof)) {
            // Process Audio Packet
            ProcessAudioPacket(requested_frame);
        }
 
        // Remove unused packets (sometimes we purposely ignore video or audio packets,
        // if the has_video or has_audio properties are manually overridden)
        if ((!info.has_video && packet && packet->stream_index == videoStream) ||
            (!info.has_audio && packet && packet->stream_index == audioStream)) {
            // Keep track of deleted packet counts
            if (packet->stream_index == videoStream) {
                packet_status.video_decoded++;
            } else if (packet->stream_index == audioStream) {
                packet_status.audio_decoded++;
            }
 
            // Remove unused packets (sometimes we purposely ignore video or audio packets,
            // if the has_video or has_audio properties are manually overridden)
            RemoveAVPacket(packet);
            packet = NULL;
        }
 
        // Determine end-of-stream (waiting until final decoder threads finish)
        // Force end-of-stream in some situations
        packet_status.end_of_file = packet_status.packets_eof && packet_status.video_eof && packet_status.audio_eof;
        if ((packet_status.packets_eof && packet_status.packets_read() == packet_status.packets_decoded()) || packet_status.end_of_file) {
            // Force EOF (end of file) variables to true, if decoder does not support EOF detection.
            // If we have no more packets, and all known packets have been decoded
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ReadStream (force EOF)", "packets_read", packet_status.packets_read(), "packets_decoded", packet_status.packets_decoded(), "packets_eof", packet_status.packets_eof, "video_eof", packet_status.video_eof, "audio_eof", packet_status.audio_eof, "end_of_file", packet_status.end_of_file);
            if (!packet_status.video_eof) {
                packet_status.video_eof = true;
            }
            if (!packet_status.audio_eof) {
                packet_status.audio_eof = true;
            }
            packet_status.end_of_file = true;
            break;
        }
 
        // Detect decoder stalls with no progress at EOF and force completion so
        // missing frames can be finalized from prior image data.
        const bool has_progress =
            (packet_status.packets_read() != prev_packets_read) ||
            (packet_status.packets_decoded() != prev_packets_decoded) ||
            (packet_status.video_decoded != prev_video_decoded) ||
            (video_pts_seconds != prev_video_pts_seconds);
 
        if (has_progress) {
            no_progress_count = 0;
        } else {
            no_progress_count++;
            if (no_progress_count >= 2000
                && packet_status.packets_eof
                && !packet
                && !hold_packet) {
                ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ReadStream (force EOF after stall)",
                    "requested_frame", requested_frame,
                    "no_progress_count", no_progress_count,
                    "packets_read", packet_status.packets_read(),
                    "packets_decoded", packet_status.packets_decoded(),
                    "video_decoded", packet_status.video_decoded,
                    "audio_decoded", packet_status.audio_decoded);
                packet_status.video_eof = true;
                packet_status.audio_eof = true;
                packet_status.end_of_file = true;
                break;
            }
        }
        prev_packets_read = packet_status.packets_read();
        prev_packets_decoded = packet_status.packets_decoded();
        prev_video_decoded = packet_status.video_decoded;
        prev_video_pts_seconds = video_pts_seconds;
    } // end while
 
    // Debug output
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ReadStream (Completed)",
                                          "packets_read", packet_status.packets_read(),
                                          "packets_decoded", packet_status.packets_decoded(),
                                          "end_of_file", packet_status.end_of_file,
                                          "largest_frame_processed", largest_frame_processed,
                                          "Working Cache Count", working_cache.Count());
 
    // Have we reached end-of-stream (or the final frame)?
    if (!packet_status.end_of_file && requested_frame >= info.video_length) {
        // Force end-of-stream
        packet_status.end_of_file = true;
    }
    if (packet_status.end_of_file) {
        // Mark any other working frames as 'finished'
        CheckWorkingFrames(requested_frame);
    }
 
    // Return requested frame (if found)
    std::shared_ptr<Frame> frame = final_cache.GetFrame(requested_frame);
    if (frame)
        // Return prepared frame
        return frame;
    else {
 
        // Check if largest frame is still cached
        frame = final_cache.GetFrame(largest_frame_processed);
        int samples_in_frame = Frame::GetSamplesPerFrame(requested_frame, info.fps,
                                                         info.sample_rate, info.channels);
        if (frame) {
            // Copy and return the largest processed frame (assuming it was the last in the video file)
            std::shared_ptr<Frame> f = CreateFrame(largest_frame_processed);
 
            // Use solid color (if no image data found)
            if (!frame->has_image_data) {
                // Use solid black frame if no image data available
                f->AddColor(info.width, info.height, "#000");
            }
            // Silence audio data (if any), since we are repeating the last frame
            frame->AddAudioSilence(samples_in_frame);
 
            return frame;
        } else {
            // The largest processed frame is no longer in cache. Prefer the most recent
            // finalized image first, then decoded image, to avoid black flashes.
            std::shared_ptr<Frame> f = CreateFrame(largest_frame_processed);
            if (last_final_video_frame && last_final_video_frame->has_image_data
                && last_final_video_frame->number <= requested_frame) {
                f->AddImage(std::make_shared<QImage>(last_final_video_frame->GetImage()->copy()));
            } else if (last_video_frame && last_video_frame->has_image_data
                && last_video_frame->number <= requested_frame) {
                f->AddImage(std::make_shared<QImage>(last_video_frame->GetImage()->copy()));
            } else {
                f->AddColor(info.width, info.height, "#000");
            }
            f->AddAudioSilence(samples_in_frame);
            return f;
        }
    }
 
}
 
// Get the next packet (if any)
int FFmpegReader::GetNextPacket() {
    int found_packet = 0;
    AVPacket *next_packet;
    next_packet = new AVPacket();
    found_packet = av_read_frame(pFormatCtx, next_packet);
 
    if (packet) {
        // Remove previous packet before getting next one
        RemoveAVPacket(packet);
        packet = NULL;
    }
    if (found_packet >= 0) {
        // Update current packet pointer
        packet = next_packet;
 
        // Keep track of packet stats
        if (packet->stream_index == videoStream) {
            packet_status.video_read++;
        } else if (packet->stream_index == audioStream) {
            packet_status.audio_read++;
        }
    } else {
        // No more packets found
        delete next_packet;
        packet = NULL;
    }
    // Return if packet was found (or error number)
    return found_packet;
}
 
// Get an AVFrame (if any)
bool FFmpegReader::GetAVFrame() {
    int frameFinished = 0;
 
    // Decode video frame
    AVFrame *next_frame = AV_ALLOCATE_FRAME();
 
#if IS_FFMPEG_3_2
    int send_packet_err = 0;
    int64_t send_packet_pts = 0;
    if ((packet && packet->stream_index == videoStream) || !packet) {
        send_packet_err = avcodec_send_packet(pCodecCtx, packet);
 
        if (packet && send_packet_err >= 0) {
            send_packet_pts = GetPacketPTS();
            hold_packet = false;
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet succeeded)", "send_packet_err", send_packet_err, "send_packet_pts", send_packet_pts);
        }
    }
 
    #if USE_HW_ACCEL
        // Get the format from the variables set in get_hw_dec_format
        hw_de_av_pix_fmt = hw_de_av_pix_fmt_global;
        hw_de_av_device_type = hw_de_av_device_type_global;
    #endif // USE_HW_ACCEL
        if (send_packet_err < 0 && send_packet_err != AVERROR_EOF) {
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet: Not sent [" + av_err2string(send_packet_err) + "])", "send_packet_err", send_packet_err, "send_packet_pts", send_packet_pts);
            if (send_packet_err == AVERROR(EAGAIN)) {
                hold_packet = true;
                ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet: AVERROR(EAGAIN): user must read output with avcodec_receive_frame()", "send_packet_pts", send_packet_pts);
            }
            if (send_packet_err == AVERROR(EINVAL)) {
                ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet: AVERROR(EINVAL): codec not opened, it is an encoder, or requires flush", "send_packet_pts", send_packet_pts);
            }
            if (send_packet_err == AVERROR(ENOMEM)) {
                ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet: AVERROR(ENOMEM): failed to add packet to internal queue, or legitimate decoding errors", "send_packet_pts", send_packet_pts);
            }
        }
 
        // Always try and receive a packet, if not EOF.
        // Even if the above avcodec_send_packet failed to send,
        // we might still need to receive a packet.
        int receive_frame_err = 0;
        AVFrame *next_frame2;
#if USE_HW_ACCEL
        if (hw_de_on && hw_de_supported) {
            next_frame2 = AV_ALLOCATE_FRAME();
        }
        else
#endif // USE_HW_ACCEL
        {
            next_frame2 = next_frame;
        }
        pFrame = AV_ALLOCATE_FRAME();
        while (receive_frame_err >= 0) {
            receive_frame_err = avcodec_receive_frame(pCodecCtx, next_frame2);
 
            if (receive_frame_err != 0) {
                ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (receive frame: frame not ready yet from decoder [\" + av_err2string(receive_frame_err) + \"])", "receive_frame_err", receive_frame_err, "send_packet_pts", send_packet_pts);
 
                if (receive_frame_err == AVERROR_EOF) {
                    ZmqLogger::Instance()->AppendDebugMethod(
                            "FFmpegReader::GetAVFrame (receive frame: AVERROR_EOF: EOF detected from decoder, flushing buffers)", "send_packet_pts", send_packet_pts);
                    avcodec_flush_buffers(pCodecCtx);
                    packet_status.video_eof = true;
                }
                if (receive_frame_err == AVERROR(EINVAL)) {
                    ZmqLogger::Instance()->AppendDebugMethod(
                            "FFmpegReader::GetAVFrame (receive frame: AVERROR(EINVAL): invalid frame received, flushing buffers)", "send_packet_pts", send_packet_pts);
                    avcodec_flush_buffers(pCodecCtx);
                }
                if (receive_frame_err == AVERROR(EAGAIN)) {
                    ZmqLogger::Instance()->AppendDebugMethod(
                            "FFmpegReader::GetAVFrame (receive frame: AVERROR(EAGAIN): output is not available in this state - user must try to send new input)", "send_packet_pts", send_packet_pts);
                }
                if (receive_frame_err == AVERROR_INPUT_CHANGED) {
                    ZmqLogger::Instance()->AppendDebugMethod(
                            "FFmpegReader::GetAVFrame (receive frame: AVERROR_INPUT_CHANGED: current decoded frame has changed parameters with respect to first decoded frame)", "send_packet_pts", send_packet_pts);
                }
 
                // Break out of decoding loop
                // Nothing ready for decoding yet
                break;
            }
 
#if USE_HW_ACCEL
            if (hw_de_on && hw_de_supported) {
                int err;
                if (next_frame2->format == hw_de_av_pix_fmt) {
                    next_frame->format = AV_PIX_FMT_YUV420P;
                    if ((err = av_hwframe_transfer_data(next_frame,next_frame2,0)) < 0) {
                        ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (Failed to transfer data to output frame)", "hw_de_on", hw_de_on);
                    }
                    if ((err = av_frame_copy_props(next_frame,next_frame2)) < 0) {
                        ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (Failed to copy props to output frame)", "hw_de_on", hw_de_on);
                    }
                }
            }
            else
#endif // USE_HW_ACCEL
            {   // No hardware acceleration used -> no copy from GPU memory needed
                next_frame = next_frame2;
            }
 
            // TODO also handle possible further frames
            // Use only the first frame like avcodec_decode_video2
            frameFinished = 1;
            packet_status.video_decoded++;
 
            // Allocate image (align 32 for simd)
            if (AV_ALLOCATE_IMAGE(pFrame, (AVPixelFormat)(pStream->codecpar->format), info.width, info.height) <= 0) {
                throw OutOfMemory("Failed to allocate image buffer", path);
            }
            av_image_copy(pFrame->data, pFrame->linesize, (const uint8_t**)next_frame->data, next_frame->linesize,
                                        (AVPixelFormat)(pStream->codecpar->format), info.width, info.height);
 
            // Get display PTS from video frame, often different than packet->pts.
            // Sending packets to the decoder (i.e. packet->pts) is async,
            // and retrieving packets from the decoder (frame->pts) is async. In most decoders
            // sending and retrieving are separated by multiple calls to this method.
            if (next_frame->pts != AV_NOPTS_VALUE) {
                // This is the current decoded frame (and should be the pts used) for
                // processing this data
                video_pts = next_frame->pts;
            } else if (next_frame->pkt_dts != AV_NOPTS_VALUE) {
                // Some videos only set this timestamp (fallback)
                video_pts = next_frame->pkt_dts;
            }
 
            ZmqLogger::Instance()->AppendDebugMethod(
                    "FFmpegReader::GetAVFrame (Successful frame received)", "video_pts", video_pts, "send_packet_pts", send_packet_pts);
 
            // break out of loop after each successful image returned
            break;
        }
#if USE_HW_ACCEL
        if (hw_de_on && hw_de_supported) {
            AV_FREE_FRAME(&next_frame2);
        }
    #endif // USE_HW_ACCEL
#else
        avcodec_decode_video2(pCodecCtx, next_frame, &frameFinished, packet);
 
        // always allocate pFrame (because we do that in the ffmpeg >= 3.2 as well); it will always be freed later
        pFrame = AV_ALLOCATE_FRAME();
 
        // is frame finished
        if (frameFinished) {
            // AVFrames are clobbered on the each call to avcodec_decode_video, so we
            // must make a copy of the image data before this method is called again.
            avpicture_alloc((AVPicture *) pFrame, pCodecCtx->pix_fmt, info.width, info.height);
            av_picture_copy((AVPicture *) pFrame, (AVPicture *) next_frame, pCodecCtx->pix_fmt, info.width,
                            info.height);
        }
#endif // IS_FFMPEG_3_2
 
    // deallocate the frame
    AV_FREE_FRAME(&next_frame);
 
    // Did we get a video frame?
    return frameFinished;
}
 
// Check the current seek position and determine if we need to seek again
bool FFmpegReader::CheckSeek() {
    // Are we seeking for a specific frame?
    if (is_seeking) {
        const int64_t kSeekRetryMax = 5;
        const int kSeekStagnantMax = 2;
 
        // Determine if both an audio and video packet have been decoded since the seek happened.
        // If not, allow the ReadStream method to keep looping
        if ((is_video_seek && !seek_video_frame_found) || (!is_video_seek && !seek_audio_frame_found))
            return false;
 
        // Check for both streams
        if ((info.has_video && !seek_video_frame_found) || (info.has_audio && !seek_audio_frame_found))
            return false;
 
        // Determine max seeked frame
        int64_t max_seeked_frame = std::max(seek_audio_frame_found, seek_video_frame_found);
        // Track stagnant seek results (no progress between retries)
        if (max_seeked_frame == last_seek_max_frame) {
            seek_stagnant_count++;
        } else {
            last_seek_max_frame = max_seeked_frame;
            seek_stagnant_count = 0;
        }
 
        // determine if we are "before" the requested frame
        if (max_seeked_frame >= seeking_frame) {
            // SEEKED TOO FAR
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckSeek (Too far, seek again)",
                                            "is_video_seek", is_video_seek,
                                            "max_seeked_frame", max_seeked_frame,
                                            "seeking_frame", seeking_frame,
                                            "seeking_pts", seeking_pts,
                                            "seek_video_frame_found", seek_video_frame_found,
                                            "seek_audio_frame_found", seek_audio_frame_found);
 
            // Seek again... to the nearest Keyframe
            if (seek_count < kSeekRetryMax) {
                Seek(seeking_frame - (10 * seek_count * seek_count));
            } else if (seek_stagnant_count >= kSeekStagnantMax) {
                // Stagnant seek: force a much earlier target and keep seeking.
                Seek(seeking_frame - (10 * kSeekRetryMax * kSeekRetryMax));
            } else {
                // Retry budget exhausted: keep seeking from a conservative offset.
                Seek(seeking_frame - (10 * seek_count * seek_count));
            }
        } else {
            // SEEK WORKED
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckSeek (Successful)",
                                            "is_video_seek", is_video_seek,
                                            "packet->pts", GetPacketPTS(),
                                            "seeking_pts", seeking_pts,
                                            "seeking_frame", seeking_frame,
                                            "seek_video_frame_found", seek_video_frame_found,
                                            "seek_audio_frame_found", seek_audio_frame_found);
 
            // Seek worked, and we are "before" the requested frame
            is_seeking = false;
            seeking_frame = 0;
            seeking_pts = -1;
        }
    }
 
    // return the pts to seek to (if any)
    return is_seeking;
}
 
// Process a video packet
void FFmpegReader::ProcessVideoPacket(int64_t requested_frame) {
    // Get the AVFrame from the current packet
    // This sets the video_pts to the correct timestamp
    int frame_finished = GetAVFrame();
 
    // Check if the AVFrame is finished and set it
    if (!frame_finished) {
        // No AVFrame decoded yet, bail out
        if (pFrame) {
            RemoveAVFrame(pFrame);
        }
        return;
    }
 
    // Calculate current frame #
    int64_t current_frame = ConvertVideoPTStoFrame(video_pts);
 
    // Track 1st video packet after a successful seek
    if (!seek_video_frame_found && is_seeking)
        seek_video_frame_found = current_frame;
 
    // Create or get the existing frame object. Requested frame needs to be created
    // in working_cache at least once. Seek can clear the working_cache, so we must
    // add the requested frame back to the working_cache here. If it already exists,
    // it will be moved to the top of the working_cache.
    working_cache.Add(CreateFrame(requested_frame));
 
    // Debug output
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessVideoPacket (Before)", "requested_frame", requested_frame, "current_frame", current_frame);
 
    // Init some things local (for OpenMP)
    PixelFormat pix_fmt = AV_GET_CODEC_PIXEL_FORMAT(pStream, pCodecCtx);
    int height = info.height;
    int width = info.width;
    int64_t video_length = info.video_length;
 
    // Create or reuse a RGB Frame (since most videos are not in RGB, we must convert it)
    AVFrame *pFrameRGB = pFrameRGB_cached;
    if (!pFrameRGB) {
        pFrameRGB = AV_ALLOCATE_FRAME();
        if (pFrameRGB == nullptr)
            throw OutOfMemory("Failed to allocate frame buffer", path);
        pFrameRGB_cached = pFrameRGB;
    }
    AV_RESET_FRAME(pFrameRGB);
    uint8_t *buffer = nullptr;
 
    // Determine the max size of this source image (based on the timeline's size, the scaling mode,
    // and the scaling keyframes). This is a performance improvement, to keep the images as small as possible,
    // without losing quality. NOTE: We cannot go smaller than the timeline itself, or the add_layer timeline
    // method will scale it back to timeline size before scaling it smaller again. This needs to be fixed in
    // the future.
    int max_width = info.width;
    int max_height = info.height;
 
    Clip *parent = static_cast<Clip *>(ParentClip());
    if (parent) {
        if (parent->ParentTimeline()) {
            // Set max width/height based on parent clip's timeline (if attached to a timeline)
            max_width = parent->ParentTimeline()->preview_width;
            max_height = parent->ParentTimeline()->preview_height;
        }
        if (parent->scale == SCALE_FIT || parent->scale == SCALE_STRETCH) {
            // Best fit or Stretch scaling (based on max timeline size * scaling keyframes)
            float max_scale_x = parent->scale_x.GetMaxPoint().co.Y;
            float max_scale_y = parent->scale_y.GetMaxPoint().co.Y;
            max_width = std::max(float(max_width), max_width * max_scale_x);
            max_height = std::max(float(max_height), max_height * max_scale_y);
 
        } else if (parent->scale == SCALE_CROP) {
            // Cropping scale mode (based on max timeline size * cropped size * scaling keyframes)
            float max_scale_x = parent->scale_x.GetMaxPoint().co.Y;
            float max_scale_y = parent->scale_y.GetMaxPoint().co.Y;
            QSize width_size(max_width * max_scale_x,
                             round(max_width / (float(info.width) / float(info.height))));
            QSize height_size(round(max_height / (float(info.height) / float(info.width))),
                              max_height * max_scale_y);
            // respect aspect ratio
            if (width_size.width() >= max_width && width_size.height() >= max_height) {
                max_width = std::max(max_width, width_size.width());
                max_height = std::max(max_height, width_size.height());
            } else {
                max_width = std::max(max_width, height_size.width());
                max_height = std::max(max_height, height_size.height());
            }
 
        } else {
            // Scale video to equivalent unscaled size
            // Since the preview window can change sizes, we want to always
            // scale against the ratio of original video size to timeline size
            float preview_ratio = 1.0;
            if (parent->ParentTimeline()) {
                Timeline *t = (Timeline *) parent->ParentTimeline();
                preview_ratio = t->preview_width / float(t->info.width);
            }
            float max_scale_x = parent->scale_x.GetMaxPoint().co.Y;
            float max_scale_y = parent->scale_y.GetMaxPoint().co.Y;
            max_width = info.width * max_scale_x * preview_ratio;
            max_height = info.height * max_scale_y * preview_ratio;
        }
 
        // If a crop effect is resizing the image, request enough pixels to preserve detail
        ApplyCropResizeScale(parent, info.width, info.height, max_width, max_height);
    }
 
    // Determine if image needs to be scaled (for performance reasons)
    int original_height = height;
    if (max_width != 0 && max_height != 0 && max_width < width && max_height < height) {
        // Override width and height (but maintain aspect ratio)
        float ratio = float(width) / float(height);
        int possible_width = round(max_height * ratio);
        int possible_height = round(max_width / ratio);
 
        if (possible_width <= max_width) {
            // use calculated width, and max_height
            width = possible_width;
            height = max_height;
        } else {
            // use max_width, and calculated height
            width = max_width;
            height = possible_height;
        }
    }
 
    // Determine required buffer size and allocate buffer
    const int bytes_per_pixel = 4;
    int raw_buffer_size = (width * height * bytes_per_pixel) + 128;
 
    // Aligned memory allocation (for speed)
    constexpr size_t ALIGNMENT = 32;  // AVX2
    int buffer_size = ((raw_buffer_size + ALIGNMENT - 1) / ALIGNMENT) * ALIGNMENT;
    buffer = (unsigned char*) aligned_malloc(buffer_size, ALIGNMENT);
 
    // Copy picture data from one AVFrame (or AVPicture) to another one.
    AV_COPY_PICTURE_DATA(pFrameRGB, buffer, PIX_FMT_RGBA, width, height);
 
    int scale_mode = SWS_FAST_BILINEAR;
    if (openshot::Settings::Instance()->HIGH_QUALITY_SCALING) {
        scale_mode = SWS_BICUBIC;
    }
    img_convert_ctx = sws_getCachedContext(img_convert_ctx, info.width, info.height, AV_GET_CODEC_PIXEL_FORMAT(pStream, pCodecCtx), width, height, PIX_FMT_RGBA, scale_mode, NULL, NULL, NULL);
    if (!img_convert_ctx)
        throw OutOfMemory("Failed to initialize sws context", path);
 
    // Resize / Convert to RGB
    sws_scale(img_convert_ctx, pFrame->data, pFrame->linesize, 0,
              original_height, pFrameRGB->data, pFrameRGB->linesize);
 
    // Create or get the existing frame object
    std::shared_ptr<Frame> f = CreateFrame(current_frame);
 
    // Add Image data to frame
    if (!ffmpeg_has_alpha(AV_GET_CODEC_PIXEL_FORMAT(pStream, pCodecCtx))) {
        // Add image with no alpha channel, Speed optimization
        f->AddImage(width, height, bytes_per_pixel, QImage::Format_RGBA8888_Premultiplied, buffer);
    } else {
        // Add image with alpha channel (this will be converted to premultipled when needed, but is slower)
        f->AddImage(width, height, bytes_per_pixel, QImage::Format_RGBA8888, buffer);
    }
 
    // Update working cache
    working_cache.Add(f);
 
    // Keep track of last last_video_frame
    last_video_frame = f;
 
    // Free the RGB image
    AV_RESET_FRAME(pFrameRGB);
 
    // Remove frame and packet
    RemoveAVFrame(pFrame);
 
    // Get video PTS in seconds
    video_pts_seconds = (double(video_pts) * info.video_timebase.ToDouble()) + pts_offset_seconds;
 
    // Debug output
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessVideoPacket (After)", "requested_frame", requested_frame, "current_frame", current_frame, "f->number", f->number, "video_pts_seconds", video_pts_seconds);
}
 
// Process an audio packet
void FFmpegReader::ProcessAudioPacket(int64_t requested_frame) {
    AudioLocation location;
    // Calculate location of current audio packet
    if (packet && packet->pts != AV_NOPTS_VALUE) {
        // Determine related video frame and starting sample # from audio PTS
        location = GetAudioPTSLocation(packet->pts);
 
        // Track 1st audio packet after a successful seek
        if (!seek_audio_frame_found && is_seeking)
            seek_audio_frame_found = location.frame;
    }
 
    // Create or get the existing frame object. Requested frame needs to be created
    // in working_cache at least once. Seek can clear the working_cache, so we must
    // add the requested frame back to the working_cache here. If it already exists,
    // it will be moved to the top of the working_cache.
    working_cache.Add(CreateFrame(requested_frame));
 
    // Debug output
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (Before)",
                                          "requested_frame", requested_frame,
                                          "target_frame", location.frame,
                                          "starting_sample", location.sample_start);
 
    // Init an AVFrame to hold the decoded audio samples
    int frame_finished = 0;
    AVFrame *audio_frame = AV_ALLOCATE_FRAME();
    AV_RESET_FRAME(audio_frame);
 
    int packet_samples = 0;
    int data_size = 0;
 
#if IS_FFMPEG_3_2
        int send_packet_err =  avcodec_send_packet(aCodecCtx, packet);
        if (send_packet_err < 0 && send_packet_err != AVERROR_EOF) {
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (Packet not sent)");
        }
        else {
            int receive_frame_err = avcodec_receive_frame(aCodecCtx, audio_frame);
            if (receive_frame_err >= 0) {
                frame_finished = 1;
            }
            if (receive_frame_err == AVERROR_EOF) {
                ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (EOF detected from decoder)");
                packet_status.audio_eof = true;
            }
            if (receive_frame_err == AVERROR(EINVAL) || receive_frame_err == AVERROR_EOF) {
                ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (invalid frame received or EOF from decoder)");
                avcodec_flush_buffers(aCodecCtx);
            }
            if (receive_frame_err != 0) {
                ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (frame not ready yet from decoder)");
            }
        }
#else
        int used = avcodec_decode_audio4(aCodecCtx, audio_frame, &frame_finished, packet);
#endif
 
    if (frame_finished) {
        packet_status.audio_decoded++;
 
        // This can be different than the current packet, so we need to look
        // at the current AVFrame from the audio decoder. This timestamp should
        // be used for the remainder of this function
        audio_pts = audio_frame->pts;
 
        // Determine related video frame and starting sample # from audio PTS
        location = GetAudioPTSLocation(audio_pts);
 
        // determine how many samples were decoded
        int plane_size = -1;
#if HAVE_CH_LAYOUT
        int nb_channels = AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->ch_layout.nb_channels;
#else
        int nb_channels = AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channels;
#endif
        data_size = av_samples_get_buffer_size(&plane_size, nb_channels,
                                               audio_frame->nb_samples, (AVSampleFormat) (AV_GET_SAMPLE_FORMAT(aStream, aCodecCtx)), 1);
 
        // Calculate total number of samples
        packet_samples = audio_frame->nb_samples * nb_channels;
    } else {
        if (audio_frame) {
            // Free audio frame
            AV_FREE_FRAME(&audio_frame);
        }
    }
 
    // Estimate the # of samples and the end of this packet's location (to prevent GAPS for the next timestamp)
    int pts_remaining_samples = packet_samples / info.channels; // Adjust for zero based array
 
    // Bail if no samples found
    if (pts_remaining_samples == 0) {
        ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (No samples, bailing)",
                                           "packet_samples", packet_samples,
                                           "info.channels", info.channels,
                                           "pts_remaining_samples", pts_remaining_samples);
        return;
    }
 
    while (pts_remaining_samples) {
        // Get Samples per frame (for this frame number)
        int samples_per_frame = Frame::GetSamplesPerFrame(previous_packet_location.frame, info.fps, info.sample_rate, info.channels);
 
        // Calculate # of samples to add to this frame
        int samples = samples_per_frame - previous_packet_location.sample_start;
        if (samples > pts_remaining_samples)
            samples = pts_remaining_samples;
 
        // Decrement remaining samples
        pts_remaining_samples -= samples;
 
        if (pts_remaining_samples > 0) {
            // next frame
            previous_packet_location.frame++;
            previous_packet_location.sample_start = 0;
        } else {
            // Increment sample start
            previous_packet_location.sample_start += samples;
        }
    }
 
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (ReSample)",
                                          "packet_samples", packet_samples,
                                          "info.channels", info.channels,
                                          "info.sample_rate", info.sample_rate,
                                          "aCodecCtx->sample_fmt", AV_GET_SAMPLE_FORMAT(aStream, aCodecCtx));
 
    // Create output frame
    AVFrame *audio_converted = AV_ALLOCATE_FRAME();
    AV_RESET_FRAME(audio_converted);
    audio_converted->nb_samples = audio_frame->nb_samples;
    av_samples_alloc(audio_converted->data, audio_converted->linesize, info.channels, audio_frame->nb_samples, AV_SAMPLE_FMT_FLTP, 0);
 
    SWRCONTEXT *avr = avr_ctx;
    // setup resample context if needed
    if (!avr) {
        avr = SWR_ALLOC();
#if HAVE_CH_LAYOUT
    av_opt_set_chlayout(avr, "in_chlayout", &AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->ch_layout, 0);
    av_opt_set_chlayout(avr, "out_chlayout", &AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->ch_layout, 0);
#else
    av_opt_set_int(avr, "in_channel_layout", AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout, 0);
    av_opt_set_int(avr, "out_channel_layout", AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout, 0);
    av_opt_set_int(avr, "in_channels", info.channels, 0);
    av_opt_set_int(avr, "out_channels", info.channels, 0);
#endif
    av_opt_set_int(avr, "in_sample_fmt", AV_GET_SAMPLE_FORMAT(aStream, aCodecCtx), 0);
    av_opt_set_int(avr, "out_sample_fmt", AV_SAMPLE_FMT_FLTP, 0);
    av_opt_set_int(avr, "in_sample_rate", info.sample_rate, 0);
    av_opt_set_int(avr, "out_sample_rate", info.sample_rate, 0);
    SWR_INIT(avr);
    avr_ctx = avr;
    }
 
    // Convert audio samples
    int nb_samples = SWR_CONVERT(avr,   // audio resample context
                             audio_converted->data,       // output data pointers
                             audio_converted->linesize[0],   // output plane size, in bytes. (0 if unknown)
                             audio_converted->nb_samples,   // maximum number of samples that the output buffer can hold
                             audio_frame->data,           // input data pointers
                             audio_frame->linesize[0],     // input plane size, in bytes (0 if unknown)
                             audio_frame->nb_samples);     // number of input samples to convert
 
 
    int64_t starting_frame_number = -1;
    for (int channel_filter = 0; channel_filter < info.channels; channel_filter++) {
        // Array of floats (to hold samples for each channel)
        starting_frame_number = location.frame;
        int channel_buffer_size = nb_samples;
        auto *channel_buffer = (float *) (audio_converted->data[channel_filter]);
 
        // Loop through samples, and add them to the correct frames
        int start = location.sample_start;
        int remaining_samples = channel_buffer_size;
        while (remaining_samples > 0) {
            // Get Samples per frame (for this frame number)
            int samples_per_frame = Frame::GetSamplesPerFrame(starting_frame_number, info.fps, info.sample_rate, info.channels);
 
            // Calculate # of samples to add to this frame
            int samples = std::fmin(samples_per_frame - start, remaining_samples);
 
            // Create or get the existing frame object
            std::shared_ptr<Frame> f = CreateFrame(starting_frame_number);
 
            // Add samples for current channel to the frame.
            f->AddAudio(true, channel_filter, start, channel_buffer, samples, 1.0f);
 
            // Debug output
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (f->AddAudio)",
                                            "frame", starting_frame_number,
                                            "start", start,
                                            "samples", samples,
                                            "channel", channel_filter,
                                            "samples_per_frame", samples_per_frame);
 
            // Add or update cache
            working_cache.Add(f);
 
            // Decrement remaining samples
            remaining_samples -= samples;
 
            // Increment buffer (to next set of samples)
            if (remaining_samples > 0)
                channel_buffer += samples;
 
            // Increment frame number
            starting_frame_number++;
 
            // Reset starting sample #
            start = 0;
        }
    }
 
    // Free AVFrames
    av_free(audio_converted->data[0]);
    AV_FREE_FRAME(&audio_converted);
    AV_FREE_FRAME(&audio_frame);
 
    // Get audio PTS in seconds
    audio_pts_seconds = (double(audio_pts) * info.audio_timebase.ToDouble()) + pts_offset_seconds;
 
    // Debug output
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (After)",
                                          "requested_frame", requested_frame,
                                          "starting_frame", location.frame,
                                          "end_frame", starting_frame_number - 1,
                                          "audio_pts_seconds", audio_pts_seconds);
 
}
 
 
// Seek to a specific frame.  This is not always frame accurate, it's more of an estimation on many codecs.
void FFmpegReader::Seek(int64_t requested_frame) {
    // Adjust for a requested frame that is too small or too large
    if (requested_frame < 1)
        requested_frame = 1;
    if (requested_frame > info.video_length)
        requested_frame = info.video_length;
    if (requested_frame > largest_frame_processed && packet_status.end_of_file) {
        // Not possible to search past largest_frame once EOF is reached (no more packets)
        return;
    }
 
    // Debug output
    ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::Seek",
                                          "requested_frame", requested_frame,
                                          "seek_count", seek_count,
                                          "last_frame", last_frame);
 
    // Clear working cache (since we are seeking to another location in the file)
    working_cache.Clear();
 
    // Reset the last frame variable
    video_pts = 0.0;
    video_pts_seconds = NO_PTS_OFFSET;
    audio_pts = 0.0;
    audio_pts_seconds = NO_PTS_OFFSET;
    hold_packet = false;
    last_frame = 0;
    current_video_frame = 0;
    largest_frame_processed = 0;
    last_final_video_frame.reset();
    bool has_audio_override = info.has_audio;
    bool has_video_override = info.has_video;
 
    // Init end-of-file detection variables
    packet_status.reset(false);
 
    // Increment seek count
    seek_count++;
 
    // If seeking near frame 1, we need to close and re-open the file (this is more reliable than seeking)
    int buffer_amount = 12;
    if (requested_frame - buffer_amount < 20) {
        // prevent Open() from seeking again
        is_seeking = true;
 
        // Close and re-open file (basically seeking to frame 1)
        Close();
        Open();
 
        // Update overrides (since closing and re-opening might update these)
        info.has_audio = has_audio_override;
        info.has_video = has_video_override;
 
        // Not actually seeking, so clear these flags
        is_seeking = false;
        if (seek_count == 1) {
            // Don't redefine this on multiple seek attempts for a specific frame
            seeking_frame = 1;
            seeking_pts = ConvertFrameToVideoPTS(1);
        }
        seek_audio_frame_found = 0; // used to detect which frames to throw away after a seek
        seek_video_frame_found = 0; // used to detect which frames to throw away after a seek
 
    } else {
        // Seek to nearest key-frame (aka, i-frame)
        bool seek_worked = false;
        int64_t seek_target = 0;
 
        // Seek video stream (if any), except album arts
        if (!seek_worked && info.has_video && !HasAlbumArt()) {
            seek_target = ConvertFrameToVideoPTS(requested_frame - buffer_amount);
            if (av_seek_frame(pFormatCtx, info.video_stream_index, seek_target, AVSEEK_FLAG_BACKWARD) < 0) {
                ZmqLogger::Instance()->Log(std::string(pFormatCtx->AV_FILENAME) + ": error while seeking video stream");
            } else {
                // VIDEO SEEK
                is_video_seek = true;
                seek_worked = true;
            }
        }
 
        // Seek audio stream (if not already seeked... and if an audio stream is found)
        if (!seek_worked && info.has_audio) {
            seek_target = ConvertFrameToAudioPTS(requested_frame - buffer_amount);
            if (av_seek_frame(pFormatCtx, info.audio_stream_index, seek_target, AVSEEK_FLAG_BACKWARD) < 0) {
                ZmqLogger::Instance()->Log(std::string(pFormatCtx->AV_FILENAME) + ": error while seeking audio stream");
            } else {
                // AUDIO SEEK
                is_video_seek = false;
                seek_worked = true;
            }
        }
 
        // Was the seek successful?
        if (seek_worked) {
            // Flush audio buffer
            if (info.has_audio)
                avcodec_flush_buffers(aCodecCtx);
 
            // Flush video buffer
            if (info.has_video)
                avcodec_flush_buffers(pCodecCtx);
 
            // Reset previous audio location to zero
            previous_packet_location.frame = -1;
            previous_packet_location.sample_start = 0;
 
            // init seek flags
            is_seeking = true;
            if (seek_count == 1) {
                // Don't redefine this on multiple seek attempts for a specific frame
                seeking_pts = seek_target;
                seeking_frame = requested_frame;
            }
            seek_audio_frame_found = 0; // used to detect which frames to throw away after a seek
            seek_video_frame_found = 0; // used to detect which frames to throw away after a seek
 
        } else {
            // seek failed
            seeking_pts = 0;
            seeking_frame = 0;
 
            // prevent Open() from seeking again
            is_seeking = true;
 
            // Close and re-open file (basically seeking to frame 1)
            Close();
            Open();
 
            // Not actually seeking, so clear these flags
            is_seeking = false;
 
            // disable seeking for this reader (since it failed)
            enable_seek = false;
 
            // Update overrides (since closing and re-opening might update these)
            info.has_audio = has_audio_override;
            info.has_video = has_video_override;
        }
    }
}
 
// Get the PTS for the current video packet
int64_t FFmpegReader::GetPacketPTS() {
    if (packet) {
        int64_t current_pts = packet->pts;
        if (current_pts == AV_NOPTS_VALUE && packet->dts != AV_NOPTS_VALUE)
            current_pts = packet->dts;
 
        // Return adjusted PTS
        return current_pts;
    } else {
        // No packet, return NO PTS
        return AV_NOPTS_VALUE;
    }
}
 
// Update PTS Offset (if any)
void FFmpegReader::UpdatePTSOffset() {
    if (pts_offset_seconds != NO_PTS_OFFSET) {
        // Skip this method if we have already set PTS offset
        return;
    }
    pts_offset_seconds = 0.0;
    double video_pts_offset_seconds = 0.0;
    double audio_pts_offset_seconds = 0.0;
 
    bool has_video_pts = false;
    if (!info.has_video) {
        // Mark as checked
        has_video_pts = true;
    }
    bool has_audio_pts = false;
    if (!info.has_audio) {
        // Mark as checked
        has_audio_pts = true;
    }
 
    // Loop through the stream (until a packet from all streams is found)
    while (!has_video_pts || !has_audio_pts) {
        // Get the next packet (if any)
        if (GetNextPacket() < 0)
            // Break loop when no more packets found
            break;
 
        // Get PTS of this packet
        int64_t pts = GetPacketPTS();
 
        // Video packet
        if (!has_video_pts && packet->stream_index == videoStream) {
            // Get the video packet start time (in seconds)
            video_pts_offset_seconds = 0.0 - (pts * info.video_timebase.ToDouble());
 
            // Is timestamp close to zero (within X seconds)
            // Ignore wildly invalid timestamps (i.e. -234923423423)
            if (std::abs(video_pts_offset_seconds) <= 10.0) {
                has_video_pts = true;
            }
        }
        else if (!has_audio_pts && packet->stream_index == audioStream) {
            // Get the audio packet start time (in seconds)
            audio_pts_offset_seconds = 0.0 - (pts * info.audio_timebase.ToDouble());
 
            // Is timestamp close to zero (within X seconds)
            // Ignore wildly invalid timestamps (i.e. -234923423423)
            if (std::abs(audio_pts_offset_seconds) <= 10.0) {
                has_audio_pts = true;
            }
        }
    }
 
    // Choose timestamp origin:
    // - If video exists, anchor timeline frame mapping to video start.
    //   This avoids AAC priming / audio preroll shifting video frame 1 to frame 2.
    // - If no video exists (audio-only readers), use audio start.
    if (info.has_video && has_video_pts) {
        pts_offset_seconds = video_pts_offset_seconds;
    } else if (!info.has_video && has_audio_pts) {
        pts_offset_seconds = audio_pts_offset_seconds;
    } else if (has_video_pts && has_audio_pts) {
        // Fallback when stream flags are unusual but both timestamps exist.
        pts_offset_seconds = video_pts_offset_seconds;
    }
}
 
// Convert PTS into Frame Number
int64_t FFmpegReader::ConvertVideoPTStoFrame(int64_t pts) {
    // Apply PTS offset
    int64_t previous_video_frame = current_video_frame;
    const double fps_value = (info.fps.num > 0 && info.fps.den > 0) ? info.fps.ToDouble() : 30.0;
    const double video_timebase_value =
        (info.video_timebase.num > 0 && info.video_timebase.den > 0)
            ? info.video_timebase.ToDouble()
            : (1.0 / 30.0);
 
    // Get the video packet start time (in seconds)
    double video_seconds = (double(pts) * video_timebase_value) + pts_offset_seconds;
 
    // Divide by the video timebase, to get the video frame number (frame # is decimal at this point)
    int64_t frame = round(video_seconds * fps_value) + 1;
 
    // Keep track of the expected video frame #
    if (current_video_frame == 0)
        current_video_frame = frame;
    else {
 
        // Sometimes frames are duplicated due to identical (or similar) timestamps
        if (frame == previous_video_frame) {
            // return -1 frame number
            frame = -1;
        } else {
            // Increment expected frame
            current_video_frame++;
        }
    }
 
    // Return frame #
    return frame;
}
 
// Convert Frame Number into Video PTS
int64_t FFmpegReader::ConvertFrameToVideoPTS(int64_t frame_number) {
    const double fps_value = (info.fps.num > 0 && info.fps.den > 0) ? info.fps.ToDouble() : 30.0;
    const double video_timebase_value =
        (info.video_timebase.num > 0 && info.video_timebase.den > 0)
            ? info.video_timebase.ToDouble()
            : (1.0 / 30.0);
 
    // Get timestamp of this frame (in seconds)
    double seconds = (double(frame_number - 1) / fps_value) + pts_offset_seconds;
 
    // Calculate the # of video packets in this timestamp
    int64_t video_pts = round(seconds / video_timebase_value);
 
    // Apply PTS offset (opposite)
    return video_pts;
}
 
// Convert Frame Number into Video PTS
int64_t FFmpegReader::ConvertFrameToAudioPTS(int64_t frame_number) {
    const double fps_value = (info.fps.num > 0 && info.fps.den > 0) ? info.fps.ToDouble() : 30.0;
    const double audio_timebase_value =
        (info.audio_timebase.num > 0 && info.audio_timebase.den > 0)
            ? info.audio_timebase.ToDouble()
            : (1.0 / 48000.0);
 
    // Get timestamp of this frame (in seconds)
    double seconds = (double(frame_number - 1) / fps_value) + pts_offset_seconds;
 
    // Calculate the # of audio packets in this timestamp
    int64_t audio_pts = round(seconds / audio_timebase_value);
 
    // Apply PTS offset (opposite)
    return audio_pts;
}
 
// Calculate Starting video frame and sample # for an audio PTS
AudioLocation FFmpegReader::GetAudioPTSLocation(int64_t pts) {
    const double audio_timebase_value =
        (info.audio_timebase.num > 0 && info.audio_timebase.den > 0)
            ? info.audio_timebase.ToDouble()
            : (1.0 / 48000.0);
    const double fps_value = (info.fps.num > 0 && info.fps.den > 0) ? info.fps.ToDouble() : 30.0;
 
    // Get the audio packet start time (in seconds)
    double audio_seconds = (double(pts) * audio_timebase_value) + pts_offset_seconds;
 
    // Divide by the video timebase, to get the video frame number (frame # is decimal at this point)
    double frame = (audio_seconds * fps_value) + 1;
 
    // Frame # as a whole number (no more decimals)
    int64_t whole_frame = int64_t(frame);
 
    // Remove the whole number, and only get the decimal of the frame
    double sample_start_percentage = frame - double(whole_frame);
 
    // Get Samples per frame
    int samples_per_frame = Frame::GetSamplesPerFrame(whole_frame, info.fps, info.sample_rate, info.channels);
 
    // Calculate the sample # to start on
    int sample_start = round(double(samples_per_frame) * sample_start_percentage);
 
    // Protect against broken (i.e. negative) timestamps
    if (whole_frame < 1)
        whole_frame = 1;
    if (sample_start < 0)
        sample_start = 0;
 
    // Prepare final audio packet location
    AudioLocation location = {whole_frame, sample_start};
 
    // Compare to previous audio packet (and fix small gaps due to varying PTS timestamps)
    if (previous_packet_location.frame != -1) {
        if (location.is_near(previous_packet_location, samples_per_frame, samples_per_frame)) {
            int64_t orig_frame = location.frame;
            int orig_start = location.sample_start;
 
            // Update sample start, to prevent gaps in audio
            location.sample_start = previous_packet_location.sample_start;
            location.frame = previous_packet_location.frame;
 
            // Debug output
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAudioPTSLocation (Audio Gap Detected)", "Source Frame", orig_frame, "Source Audio Sample", orig_start, "Target Frame", location.frame, "Target Audio Sample", location.sample_start, "pts", pts);
 
        } else {
            // Debug output
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAudioPTSLocation (Audio Gap Ignored - too big)", "Previous location frame", previous_packet_location.frame, "Target Frame", location.frame, "Target Audio Sample", location.sample_start, "pts", pts);
        }
    }
 
    // Set previous location
    previous_packet_location = location;
 
    // Return the associated video frame and starting sample #
    return location;
}
 
// Create a new Frame (or return an existing one) and add it to the working queue.
std::shared_ptr<Frame> FFmpegReader::CreateFrame(int64_t requested_frame) {
    // Check working cache
    std::shared_ptr<Frame> output = working_cache.GetFrame(requested_frame);
 
    if (!output) {
        // (re-)Check working cache
        output = working_cache.GetFrame(requested_frame);
        if(output) return output;
 
        // Create a new frame on the working cache
        output = std::make_shared<Frame>(requested_frame, info.width, info.height, "#000000", Frame::GetSamplesPerFrame(requested_frame, info.fps, info.sample_rate, info.channels), info.channels);
        output->SetPixelRatio(info.pixel_ratio.num, info.pixel_ratio.den); // update pixel ratio
        output->ChannelsLayout(info.channel_layout); // update audio channel layout from the parent reader
        output->SampleRate(info.sample_rate); // update the frame's sample rate of the parent reader
 
        working_cache.Add(output);
 
        // Set the largest processed frame (if this is larger)
        if (requested_frame > largest_frame_processed)
            largest_frame_processed = requested_frame;
    }
    // Return frame
    return output;
}
 
// Determine if frame is partial due to seek
bool FFmpegReader::IsPartialFrame(int64_t requested_frame) {
 
    // Sometimes a seek gets partial frames, and we need to remove them
    bool seek_trash = false;
    int64_t max_seeked_frame = seek_audio_frame_found; // determine max seeked frame
    if (seek_video_frame_found > max_seeked_frame) {
        max_seeked_frame = seek_video_frame_found;
    }
    if ((info.has_audio && seek_audio_frame_found && max_seeked_frame >= requested_frame) ||
        (info.has_video && seek_video_frame_found && max_seeked_frame >= requested_frame)) {
        seek_trash = true;
    }
 
    return seek_trash;
}
 
// Check the working queue, and move finished frames to the finished queue
void FFmpegReader::CheckWorkingFrames(int64_t requested_frame) {
 
    // Prevent async calls to the following code
    const std::lock_guard<std::recursive_mutex> lock(getFrameMutex);
 
    // Get a list of current working queue frames in the cache (in-progress frames)
    std::vector<std::shared_ptr<openshot::Frame>> working_frames = working_cache.GetFrames();
    std::vector<std::shared_ptr<openshot::Frame>>::iterator working_itr;
 
    // Loop through all working queue frames (sorted by frame #)
    for(working_itr = working_frames.begin(); working_itr != working_frames.end(); ++working_itr)
    {
        // Get working frame
        std::shared_ptr<Frame> f = *working_itr;
 
        // Was a frame found? Is frame requested yet?
        if (!f || f->number > requested_frame) {
            // If not, skip to next one
            continue;
        }
 
        // Calculate PTS in seconds (of working frame), and the most recent processed pts value
        double frame_pts_seconds = (double(f->number - 1) / info.fps.ToDouble()) + pts_offset_seconds;
        double recent_pts_seconds = std::max(video_pts_seconds, audio_pts_seconds);
 
        // Determine if video and audio are ready (based on timestamps)
        bool is_video_ready = false;
        bool is_audio_ready = false;
        double recent_pts_diff = recent_pts_seconds - frame_pts_seconds;
        if ((frame_pts_seconds <= video_pts_seconds)
            || (recent_pts_diff > 1.5)
            || packet_status.video_eof || packet_status.end_of_file) {
            // Video stream is past this frame (so it must be done)
            // OR video stream is too far behind, missing, or end-of-file
            is_video_ready = true;
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckWorkingFrames (video ready)",
                                        "frame_number", f->number,
                                        "frame_pts_seconds", frame_pts_seconds,
                                        "video_pts_seconds", video_pts_seconds,
                                        "recent_pts_diff", recent_pts_diff);
            if (info.has_video && !f->has_image_data) {
                // Frame has no image data. Prefer timeline-previous frames to preserve
                // visual order, especially when decode/prefetch is out-of-order.
                std::shared_ptr<Frame> previous_frame_instance = final_cache.GetFrame(f->number - 1);
                if (previous_frame_instance && previous_frame_instance->has_image_data) {
                    f->AddImage(std::make_shared<QImage>(previous_frame_instance->GetImage()->copy()));
                }
 
                // Fall back to last finalized timeline image (survives cache churn).
                if (!f->has_image_data
                    && last_final_video_frame
                    && last_final_video_frame->has_image_data
                    && last_final_video_frame->number <= f->number) {
                    f->AddImage(std::make_shared<QImage>(last_final_video_frame->GetImage()->copy()));
                }
 
                // Fall back to the last decoded image only when it is not from the future.
                if (!f->has_image_data
                    && last_video_frame
                    && last_video_frame->has_image_data
                    && last_video_frame->number <= f->number) {
                    f->AddImage(std::make_shared<QImage>(last_video_frame->GetImage()->copy()));
                }
 
                // Last-resort fallback if no prior image is available.
                if (!f->has_image_data) {
                    ZmqLogger::Instance()->AppendDebugMethod(
                        "FFmpegReader::CheckWorkingFrames (no previous image found; using black frame)",
                        "frame_number", f->number);
                    f->AddColor("#000000");
                }
            }
        }
 
        double audio_pts_diff = audio_pts_seconds - frame_pts_seconds;
        if ((frame_pts_seconds < audio_pts_seconds && audio_pts_diff > 1.0)
           || (recent_pts_diff > 1.5)
           || packet_status.audio_eof || packet_status.end_of_file) {
            // Audio stream is past this frame (so it must be done)
            // OR audio stream is too far behind, missing, or end-of-file
            // Adding a bit of margin here, to allow for partial audio packets
            is_audio_ready = true;
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckWorkingFrames (audio ready)",
                                            "frame_number", f->number, 
                                            "frame_pts_seconds", frame_pts_seconds, 
                                            "audio_pts_seconds", audio_pts_seconds, 
                                            "audio_pts_diff", audio_pts_diff, 
                                            "recent_pts_diff", recent_pts_diff);
        }
        bool is_seek_trash = IsPartialFrame(f->number);
 
        // Adjust for available streams
        if (!info.has_video) is_video_ready = true;
        if (!info.has_audio) is_audio_ready = true;
 
        // Debug output
        ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckWorkingFrames",
                                           "frame_number", f->number, 
                                           "is_video_ready", is_video_ready, 
                                           "is_audio_ready", is_audio_ready, 
                                           "video_eof", packet_status.video_eof,
                                           "audio_eof", packet_status.audio_eof,
                                           "end_of_file", packet_status.end_of_file);
 
        // Check if working frame is final
        if ((!packet_status.end_of_file && is_video_ready && is_audio_ready) || packet_status.end_of_file || is_seek_trash) {
            // Debug output
            ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckWorkingFrames (mark frame as final)", 
                                            "requested_frame", requested_frame, 
                                            "f->number", f->number, 
                                            "is_seek_trash", is_seek_trash, 
                                            "Working Cache Count", working_cache.Count(), 
                                            "Final Cache Count", final_cache.Count(), 
                                            "end_of_file", packet_status.end_of_file);
 
            if (!is_seek_trash) {
                // Move frame to final cache
                final_cache.Add(f);
                if (f->has_image_data) {
                    last_final_video_frame = f;
                }
 
                // Remove frame from working cache
                working_cache.Remove(f->number);
 
                // Update last frame processed
                last_frame = f->number;
            } else {
                // Seek trash, so delete the frame from the working cache, and never add it to the final cache.
                working_cache.Remove(f->number);
            }
 
        }
    }
 
    // Clear vector of frames
    working_frames.clear();
    working_frames.shrink_to_fit();
}
 
// Check for the correct frames per second (FPS) value by scanning the 1st few seconds of video packets.
void FFmpegReader::CheckFPS() {
    if (check_fps) {
        // Do not check FPS more than 1 time
        return;
    } else {
        check_fps = true;
    }
 
    int frames_per_second[3] = {0,0,0};
    int max_fps_index = sizeof(frames_per_second) / sizeof(frames_per_second[0]);
    int fps_index = 0;
 
    int all_frames_detected = 0;
    int starting_frames_detected = 0;
 
    // Loop through the stream
    while (true) {
        // Get the next packet (if any)
        if (GetNextPacket() < 0)
            // Break loop when no more packets found
            break;
 
        // Video packet
        if (packet->stream_index == videoStream) {
            // Get the video packet start time (in seconds)
            double video_seconds = (double(GetPacketPTS()) * info.video_timebase.ToDouble()) + pts_offset_seconds;
            fps_index = int(video_seconds); // truncate float timestamp to int (second 1, second 2, second 3)
 
            // Is this video packet from the first few seconds?
            if (fps_index >= 0 && fps_index < max_fps_index) {
                // Yes, keep track of how many frames per second (over the first few seconds)
                starting_frames_detected++;
                frames_per_second[fps_index]++;
            }
 
            // Track all video packets detected
            all_frames_detected++;
        }
    }
 
    // Calculate FPS (based on the first few seconds of video packets)
    float avg_fps = 30.0;
    if (starting_frames_detected > 0 && fps_index > 0) {
        avg_fps = float(starting_frames_detected) / std::min(fps_index, max_fps_index);
    }
 
    // Verify average FPS is a reasonable value
    if (avg_fps < 8.0) {
        // Invalid FPS assumed, so switching to a sane default FPS instead
        avg_fps = 30.0;
    }
 
    // Update FPS (truncate average FPS to Integer)
    info.fps = Fraction(int(avg_fps), 1);
 
    // Update Duration and Length
    if (all_frames_detected > 0) {
        // Use all video frames detected to calculate # of frames
        info.video_length = all_frames_detected;
        info.duration = all_frames_detected / avg_fps;
    } else {
        // Use previous duration to calculate # of frames
        info.video_length = info.duration * avg_fps;
    }
 
    // Update video bit rate
    info.video_bit_rate = info.file_size / info.duration;
}
 
// Remove AVFrame from cache (and deallocate its memory)
void FFmpegReader::RemoveAVFrame(AVFrame *remove_frame) {
    // Remove pFrame (if exists)
    if (remove_frame) {
        // Free memory
        av_freep(&remove_frame->data[0]);
#ifndef WIN32
        AV_FREE_FRAME(&remove_frame);
#endif
    }
}
 
// Remove AVPacket from cache (and deallocate its memory)
void FFmpegReader::RemoveAVPacket(AVPacket *remove_packet) {
    // deallocate memory for packet
    AV_FREE_PACKET(remove_packet);
 
    // Delete the object
    delete remove_packet;
}
 
// Generate JSON string of this object
std::string FFmpegReader::Json() const {
 
    // Return formatted string
    return JsonValue().toStyledString();
}
 
// Generate Json::Value for this object
Json::Value FFmpegReader::JsonValue() const {
 
    // Create root json object
    Json::Value root = ReaderBase::JsonValue(); // get parent properties
    root["type"] = "FFmpegReader";
    root["path"] = path;
    switch (duration_strategy) {
        case DurationStrategy::VideoPreferred:
            root["duration_strategy"] = "VideoPreferred";
            break;
        case DurationStrategy::AudioPreferred:
            root["duration_strategy"] = "AudioPreferred";
            break;
        case DurationStrategy::LongestStream:
        default:
            root["duration_strategy"] = "LongestStream";
            break;
    }
 
    // return JsonValue
    return root;
}
 
// Load JSON string into this object
void FFmpegReader::SetJson(const std::string value) {
 
    // Parse JSON string into JSON objects
    try {
        const Json::Value root = openshot::stringToJson(value);
        // Set all values that match
        SetJsonValue(root);
    }
    catch (const std::exception& e) {
        // Error parsing JSON (or missing keys)
        throw InvalidJSON("JSON is invalid (missing keys or invalid data types)");
    }
}
 
// Load Json::Value into this object
void FFmpegReader::SetJsonValue(const Json::Value root) {
 
    // Set parent data
    ReaderBase::SetJsonValue(root);
 
    // Set data from Json (if key is found)
    if (!root["path"].isNull())
        path = root["path"].asString();
    if (!root["duration_strategy"].isNull()) {
        const std::string strategy = root["duration_strategy"].asString();
        if (strategy == "VideoPreferred") {
            duration_strategy = DurationStrategy::VideoPreferred;
        } else if (strategy == "AudioPreferred") {
            duration_strategy = DurationStrategy::AudioPreferred;
        } else {
            duration_strategy = DurationStrategy::LongestStream;
        }
    }
}