VoutRework

From VideoLAN Wiki
Jump to navigation Jump to search

Draft design for the vout rework effort

Overall architecture

The current system (one vout_thread_t per window) will be splitted

  • vout_core that manages all vout instances and windows. also takes care of vout_filters (wall, clone)
  • vout_window object per displayed window. provides a "region for vout_renderers" to draw in.
  • vout_render that renders the actual video, text and image

OSD and subpictures

We need to be able to display them either:

  • At video resolution, for DVD subtitles (on the "render")
  • At screen resolution (text subtitles, OSD control) (on the "output")
    • This will also allow us to draw subtitles outside of the video rendering
    • Probably needs specific code for each vout

At the moment, subpictures have very annoying timing constraints. subpictures should be timed either on the stream (for subpictures) or absolute, from the vout_core (for subfilters)


We should create two seperate systems. a vout_render level system which basically draws extra things into the region provided by vout_window. This can be both text, images and possibly even video. Material should be drawn in the resolution of the provided region, instead of the resolution of the main video. Is should be seperate from the spu streams. However it should be possible to feed SPU based material to the vout_render system instead of the normal SPU stream pipeline and vice versa.

Filters

Make a clean separation:

  • vout_filters should not be the same as real vouts anymore (will come from the split)
  • pic_filter
    • Currently called video_filter2, should be renamed
    • Works on a single pic buffer
    • Can be streamed
  • subpicture_filter

Modularization of specific calls

For many of the operations we need, OS specific calls exist (OpenGL, Quartz, DirectX). This should be modularized, with a native "C" fallback. OS specific calls are often faster.

There should be only one module for all the calls, if possible. The best module would be loaded by the vout_core singleton, and function pointers set up for each of the functions we need. perhaps a vout_Control type could take care of all these things in the vout_renderer. paramaters that should be handled this way are at least: alignment, AR, crop, pad

Misc

  • Vout plugins (that provide vout windows) should allow reusing a vout by reusing it. This would much improve the experience)
  • Vout windows must have a flag to tell if they are "main" or "helper" (audio visualization, ...). Helper vouts should get no OSD (if another vout is present) and should not "steal" the accelerated vout"
  • cropping and padding (do at OS level, but provide a core call to set it up and configure).
  • better hotkeys integration : need hotkey module redesign to act on selected vout and relevant input instead of the first we find
  • vout windows should indicate if they are on a 4:3 or 16:9 screen (we can calculate this trough resolution). Note that this can change, because a window could be moved from one screen to another. Use callbacks.

cropping padding functionality

  • add padding in order to display subs under the video. what if we have a 4:3 movie in a 4:3 screen? do we create blackbars to the left and right sides? do we crop a part of the video ???
  • 4:3 cropping to 16:9 (in order to remove blackbars at top and bottom
  • 16:9 cropping to 4:3 (in order to remove blackbars left and right side)
  • 4:3 to 16:9 (AR correction)
  • 16:9 to 4:3 (AR correction)
  • AR Freeform
  • Support Moviescreen AR? (2.39:1 or 1.85:1)
  • More info on AR: http://en.wikipedia.org/wiki/Aspect_ratio_(image)
  • If we let the window provider pass/calculate the AR of the display, then we can even do some of this automatically if the uses desires.
  • cropping should have priority over padding. Then we can first crop blackbars, then add padding for subs.