{"id":520,"date":"2015-10-02T14:39:31","date_gmt":"2015-10-02T14:39:31","guid":{"rendered":"http:\/\/blogs.gentoo.org\/lu_zero\/?p=520"},"modified":"2015-10-02T14:41:43","modified_gmt":"2015-10-02T14:41:43","slug":"decoupling-an-api-part-ii","status":"publish","type":"post","link":"https:\/\/blogs.gentoo.org\/lu_zero\/2015\/10\/02\/decoupling-an-api-part-ii\/","title":{"rendered":"Decoupling an API – Part II"},"content":{"rendered":"

During the VDD<\/a> we had lots of discussions and I enjoyed reviewing the initial NihAV<\/a> implementation. Kostya already wrote some more<\/a> about the decoupled API that I described at high level here<\/a>.<\/p>\n

This article is about some possible implementation details, at least another will follow.<\/p>\n

The new API requires some additional data structures, mainly something to keep the data that is being consumed\/produced, additional implementation-callbacks in AVCodec and possibly a mean to skip the queuing up completely.<\/p>\n

Data Structures<\/h2>\n

AVPacketQueue and AVFrameQueue<\/h3>\n

In the previous<\/a> post I considered as given some kind of Queue.<\/p>\n

Ideally the API for it could be really simple:<\/p>\n

\n
typedef<\/span> struct<\/span> AVPacketQueue<\/span>;<\/span>\n\nAVPacketQueue<\/span> *<\/span>av_packet_queue_alloc<\/span>(<\/span>int<\/span> size<\/span>);<\/span>\nint<\/span> av_packet_queue_put<\/span>(<\/span>AVPacketQueue<\/span> *<\/span>q<\/span>,<\/span> AVPacket<\/span> *<\/span>pkt<\/span>);<\/span>\nint<\/span> av_packet_queue_get<\/span>(<\/span>AVPacketQueue<\/span> *<\/span>q<\/span>,<\/span> AVPacket<\/span> *<\/span>pkt<\/span>);<\/span>\nint<\/span> av_packet_queue_size<\/span>(<\/span>AVPacketQueue<\/span> *<\/span>q<\/span>);<\/span>\nvoid<\/span> av_packet_queue_free<\/span>(<\/span>AVPacketQueue<\/span> **<\/span>q<\/span>);<\/span>\n<\/pre>\n<\/div>\n
\n
typedef<\/span> struct<\/span> AVFrameQueue<\/span>;<\/span>\n\nAVFrameQueue<\/span> *<\/span>av_frame_queue_alloc<\/span>(<\/span>int<\/span> size<\/span>);<\/span>\nint<\/span> av_frame_queue_put<\/span>(<\/span>AVFrameQueue<\/span> *<\/span>q<\/span>,<\/span> AVPacket<\/span> *<\/span>pkt<\/span>);<\/span>\nint<\/span> av_frame_queue_get<\/span>(<\/span>AVFrameQueue<\/span> *<\/span>q<\/span>,<\/span> AVPacket<\/span> *<\/span>pkt<\/span>);<\/span>\nint<\/span> av_frame_queue_size<\/span>(<\/span>AVFrameQueue<\/span> *<\/span>q<\/span>);<\/span>\nvoid<\/span> av_frame_queue_free<\/span>(<\/span>AVFrameQueue<\/span> **<\/span>q<\/span>);<\/span>\n<\/pre>\n<\/div>\n
Internally it leverages the ref-counted<\/strong> API (av_packet_move_ref<\/code> and av_frame_move_ref<\/code>) and any data structure that could fit the queue-usage. It will be used in a multi-thread scenario so a form of Lock has to be fit into it.<\/p>\n
We have already something specific for AVPlay, using a simple Linked List and a FIFO for some other components that have a near-constant maximum number of items (e.g. avconv, NVENC, QSV).<\/p>\n
Possibly also a Tree could be used to implement something such as av_packet_queue_insert_by_pts<\/code> and have some forms of reordering happen on the fly. I’m not a fan of it, but I’m sure someone will come up with the idea..<\/p>\n
The Queues are part of AVCodecContext<\/code>.<\/p>\n
\n
typedef<\/span> struct<\/span> AVCodecContext<\/span> {<\/span>\n    \/\/ ...<\/span>\n\n    AVPacketQueue<\/span> *<\/span>packet_queue<\/span>;<\/span>\n    AVFrameQueue<\/span> *<\/span>frame_queue<\/span>;<\/span>\n\n    \/\/ ...<\/span>\n}<\/span> AVCodecContext<\/span>;<\/span>\n<\/pre>\n<\/div>\n
Implementation Callbacks<\/h2>\n
In Libav the AVCodec<\/code> struct describes some specific codec features (such as the supported framerates) and hold the actual codec implementation through callbacks such as init<\/code>, decode<\/code>\/encode2<\/code>, flush<\/code> and close<\/code>.
\nThe new model obviously requires additional callbacks.<\/p>\n
Once the data is in a queue it is ready to be processed, the actual decoding or encoding can happen in multiple places, for example:<\/p>\n