本文为作者原创，转载请注明出处：https://www.cnblogs.com/leisure_chn/p/10410320.html
本文基于FFmpeg 4.1版本。
1. 数据结构定义
struct AVPacket定义于<libavcodec/avcodec.h>
struct AVPacket packet;
AVPacket中存储的是经过编码的压缩数据。在解码中，AVPacket由解复用器输出到解码器；在编码中，AVPacket由编码器输出到复用器。下图中，解复用器(demuxer)的输出和复用器(muxer)的输入“encoded data packets”的数据类型就是AVPacket：
1 _______ ______________ 2| | | | 3| input | demuxer | encoded data | decoder 4| file | ---------> | packets | -----+ 5|_______| |______________| | 6 v 7 _________ 8 | | 9 | decoded | 10 | frames | 11 |_________| 12 ________ ______________ | 13| | | | | 14| output | <-------- | encoded data | <----+ 15| file | muxer | packets | encoder 16|________| |______________| 17
对于视频而言，一个AVPacket通常只包含一个压缩视频帧。而对于音频而言，一个AVPacket可能包含多个完整的音频压缩帧。AVPacket也可以不包含压缩编码数据，而只包含side data，这种包可以称为空packet。例如，编码结束后只需要更新一些参数时就可以发空packet。
AVPacket对象可以在栈上分配，注意此处指的是AVPacket对象本身。而AVPacket中包含的数据缓冲区是通过av_malloc()在堆上分配的。
<font color=red>TODO: AVPacket对象在栈上分配，原理不清楚，待研究</font>
1/** 2 * This structure stores compressed data. It is typically exported by demuxers 3 * and then passed as input to decoders, or received as output from encoders and 4 * then passed to muxers. 5 * 6 * For video, it should typically contain one compressed frame. For audio it may 7 * contain several compressed frames. Encoders are allowed to output empty 8 * packets, with no compressed data, containing only side data 9 * (e.g. to update some stream parameters at the end of encoding). 10 * 11 * AVPacket is one of the few structs in FFmpeg, whose size is a part of public 12 * ABI. Thus it may be allocated on stack and no new fields can be added to it 13 * without libavcodec and libavformat major bump. 14 * 15 * The semantics of data ownership depends on the buf field. 16 * If it is set, the packet data is dynamically allocated and is 17 * valid indefinitely until a call to av_packet_unref() reduces the 18 * reference count to 0. 19 * 20 * If the buf field is not set av_packet_ref() would make a copy instead 21 * of increasing the reference count. 22 * 23 * The side data is always allocated with av_malloc(), copied by 24 * av_packet_ref() and freed by av_packet_unref(). 25 * 26 * @see av_packet_ref 27 * @see av_packet_unref 28 */ 29typedef struct AVPacket { 30 /** 31 * A reference to the reference-counted buffer where the packet data is 32 * stored. 33 * May be NULL, then the packet data is not reference-counted. 34 */ 35 AVBufferRef *buf; 36 /** 37 * Presentation timestamp in AVStream->time_base units; the time at which 38 * the decompressed packet will be presented to the user. 39 * Can be AV_NOPTS_VALUE if it is not stored in the file. 40 * pts MUST be larger or equal to dts as presentation cannot happen before 41 * decompression, unless one wants to view hex dumps. Some formats misuse 42 * the terms dts and pts/cts to mean something different. Such timestamps 43 * must be converted to true pts/dts before they are stored in AVPacket. 44 */ 45 int64_t pts; 46 /** 47 * Decompression timestamp in AVStream->time_base units; the time at which 48 * the packet is decompressed. 49 * Can be AV_NOPTS_VALUE if it is not stored in the file. 50 */ 51 int64_t dts; 52 uint8_t *data; 53 int size; 54 int stream_index; 55 /** 56 * A combination of AV_PKT_FLAG values 57 */ 58 int flags; 59 /** 60 * Additional packet data that can be provided by the container. 61 * Packet can contain several types of side information. 62 */ 63 AVPacketSideData *side_data; 64 int side_data_elems; 65 66 /** 67 * Duration of this packet in AVStream->time_base units, 0 if unknown. 68 * Equals next_pts - this_pts in presentation order. 69 */ 70 int64_t duration; 71 72 int64_t pos; ///< byte position in stream, -1 if unknown 73 74#if FF_API_CONVERGENCE_DURATION 75 /** 76 * @deprecated Same as the duration field, but as int64_t. This was required 77 * for Matroska subtitles, whose duration values could overflow when the 78 * duration field was still an int. 79 */ 80 attribute_deprecated 81 int64_t convergence_duration; 82#endif 83} AVPacket;
音视频数据缓冲区
uint8_t *data:int size:
数据缓冲区地址与大小。音视频编码压缩数据存储于此片内存区域。此内存区域由AVBufferRef *buf管理。AVBufferRef *buf:
数据缓冲区引用，也可叫引用计数缓冲区。对上一字段uint8_t *data指向的内存区域提供引用计数等管理机制。
AVBufferRef对数据缓冲区提供了管理机制，用户不应直接访问数据缓冲区。参考“FFmpeg数据结构AVBuffer”
如果buf值为NULL，则data指向的数据缓冲区不使用引用计数机制。av_packet_ref(dst, src)将执行数据缓冲区的拷贝，而非仅仅增加缓冲区引用计数。
如果buf值非NULL，则data指向的数据缓冲区使用引用计数机制。av_packet_ref(dst, src)将不拷贝缓冲区，而仅增加缓冲区引用计数。av_packet_unref()将数据缓冲区引用计数减1，当缓冲区引用计数为0时，缓冲区内存被FFmpeg回收。
对于struct AVPacket pkt对象，如果pkt.buf值非NULL，则有pkt.data == pkt.buf->data == pkt.buf->buffer.data
额外类型数据
AVPacketSideData *side_dataint side_data_elems
由容器(container)提供的额外包数据。<font color=red>TODO: 待研究</font>
packet属性
int64_t pts:
显示时间戳。单位time_base，帧率的倒数。int64_t dts:
解码时间戳。单位time_base，帧率的倒数。int stream_index:
当前包(packet)所有流(stream)的索引(index)。int flags:
packet标志位。比如是否关键帧等。int64_t duration:
当前包解码后的帧播放持续的时长。单位timebase。值等于下一帧pts减当前帧pts。int64_t pos:
当前包在流中的位置，单位字节。
2. 关键函数实现
这里列出的几个关键函数，主要是为了帮助理解struct AVPacket数据结构
2.1 av_packet_ref()
1int av_packet_ref(AVPacket *dst, const AVPacket *src) 2{ 3 int ret; 4 5 ret = av_packet_copy_props(dst, src); 6 if (ret < 0) 7 return ret; 8 9 if (!src->buf) { 10 ret = packet_alloc(&dst->buf, src->size); 11 if (ret < 0) 12 goto fail; 13 if (src->size) 14 memcpy(dst->buf->data, src->data, src->size); 15 16 dst->data = dst->buf->data; 17 } else { 18 dst->buf = av_buffer_ref(src->buf); 19 if (!dst->buf) { 20 ret = AVERROR(ENOMEM); 21 goto fail; 22 } 23 dst->data = src->data; 24 } 25 26 dst->size = src->size; 27 28 return 0; 29fail: 30 av_packet_free_side_data(dst); 31 return ret; 32}
av_packet_ref()作了处理如下:
a) 如果src->buf为NULL，则将src缓冲区拷贝到新创建的dst缓冲区，注意src缓冲区不支持引用计数，但新建的dst缓冲区是支持引用计数的，因为dst->buf不为NULL。
b) 如果src->buf不为空，则dst与src共用缓冲区，调用av_buffer_ref()增加缓冲区引用计数即可。av_buffer_ref()分析参考“FFmpeg数据结构AVBuffer”
2.2 av_packet_unref()
1void av_packet_unref(AVPacket *pkt) 2{ 3 av_packet_free_side_data(pkt); 4 av_buffer_unref(&pkt->buf); 5 av_init_packet(pkt); 6 pkt->data = NULL; 7 pkt->size = 0; 8}
av_packet_unref()注销AVPacket *pkt对象，并调用av_buffer_unref(&pkt->buf);将缓冲区引用计数减1。
av_buffer_unref()中将缓冲区引用计数减1后，若缓冲区引用计数变成0，则回收缓冲区内存。av_buffer_unref()分析参考“FFmpeg数据结构AVBuffer”
3. 参考资料
[1] FFmpeg数据结构：AVPacket解析, https://www.cnblogs.com/wangguchangqing/p/5790705.html
4. 修改记录
2018-12-14 V1.0 初稿