The vpx_codec_decode() function is at the core of
the decode loop. It processes packets of compressed data passed by
the application, producing decoded images. The decoder expects
packets to comprise exactly one image frame of data. Packets
MUST be passed in decode
order. If the application wishes to associate some data with the
user_priv member may be set. The
deadline parameter controls the amount of time in
microseconds the decoder should spend working on the frame. This is
typically used to support adaptive Postprocessing based on the
amount of free CPU time. For more information on the
deadline parameter, see Deadline.
There are two methods for the application to access decoded frame data. Some codecs support asynchronous (callback-based) decoding Features that allow the application to register a callback to be invoked by the decoder when decoded data becomes available. Decoders are not required to support this feature, however. Like all Features, support can be determined by calling vpx_codec_get_caps(). Callbacks are available in both frame-based and slice-based variants. Frame based callbacks conform to the signature of vpx_codec_put_frame_cb_fn_t and are invoked once the entire frame has been decoded. Slice based callbacks conform to the signature of vpx_codec_put_slice_cb_fn_t and are invoked after a subsection of the frame is decoded. For example, a slice callback could be issued for each macroblock row. However, the number and size of slices to return is implementation specific. Also, the image data passed in a slice callback is not necessarily in the same memory segment as the data will be when it is assembled into a full frame. For this reason, the application MUST examine the rectangles that describe what data is valid to access and what data has been updated in this call. For all their additional complexity, slice based decoding callbacks provide substantial speed gains to the overall application in some cases, due to improved cache behavior.
If the codec does not support callback based decoding, or the application chooses not to make use of that feature, decoded frames are made available through the vpx_codec_get_frame() iterator. The application initializes the iterator storage (of type vpx_codec_iter_t) to NULL, then calls vpx_codec_get_frame repeatedly until it returns NULL, indicating that all images have been returned. This process may result in zero, one, or many frames that are ready for display, depending on the codec.
Postprocessing is a process that is applied after a frame is decoded to enhance the image's appearance by removing artifacts introduced in the compression process. It is not required to properly decode the frame, and is generally done only when there is enough spare CPU time to execute the required filters. Codecs may support a number of different postprocessing filters, and the available filters may differ from platform to platform. Embedded devices often do not have enough CPU to implement postprocessing in software. The filter selection is generally handled automatically by the codec, depending on the amount of time remaining before hitting the user-specified Deadline after decoding the frame.