Skip to content

wasapi2: Rewrite plugin and implement audioringbuffer subclass

... based on MediaFoundation work queue API.

By this commit, wasapi2 plugin will make use of pull mode scheduling
with audioringbuffer subclass.
There are several drawbacks of audiosrc/audiosink subclassing
(not audiobasesrc/audiobasesink) for WASAPI API, which are:
* audiosrc/audiosink classes try to set high priority to
  read/write thread via MMCSS (Multimedia Class Scheduler Service)
  but it's not allowed in case of UWP application.
  In order to use MMCSS in UWP, application should use MediaFoundation
  work queue indirectly.
  Since audiosrc/audiosink scheduling model is not compatible with
  MediaFoundation's work queue model, audioringbuffer subclassing
  is required.
* WASAPI capture device might report larger packet size than expected
  (i.e., larger frames we can read than expected frame size per period).
  Meanwhile, in any case, application should drain all packets at that moment.
  In order to handle the case, wasapi/wasapi2 plugins were making use of
  GstAdapter which is obviously sub-optimal because it requires additional
  memory allocation and copy.
  By implementing audioringbuffer subclass, we can avoid such inefficiency.

In this commit, all the device read/write operations will be moved
to newly implemented wasapi2ringbuffer class and
existing wasapi2client class will take care of device enumeration
and activation parts only.
Edited by Seungha Yang

Merge request reports