Re: [AD] Plan for the video addon

On Sun, Sep 13, 2015 at 1:43 AM, SiegeLord <siegelordex@xxxxxxxxxx> wrote:

So I've fiddled a bit with the video addon, and here's my conclusions about what needs to be done with it before we can consider it stable in terms of API/semantics.

First, the API. I think it's basically good as is. I want to make the following modifications:

- Remove al_get_video_aspect_ratio and change al_get_video_width and al_get_video_height to return the width and height of the video with the correct aspect scale. The intention is for something like this to just work:

ALLEGRO_BITMAP* f = al_get_video_bitmap(v);
al_draw_scaled_bitmap(f, 0, 0, al_get_bitmap_width(f), al_get_bitmap_height(f), 0, 0, al_get_video_width(v), al_get_video_height(v), 0);

- Rename al_is_video_paused/al_pause_video to al_is_video_playing/al_set_video_playing to match the audio addon. Make al_is_video_playing return false when a video finishes.
- Add ALLEGRO_VIDEO_FINISHED event.
- Rename ALLEGRO_EVENT_VIDEO_FRAME_ALLOC to ALLEGRO_EVENT_VIDEO_FRAME_PREPARE (see below).
- Add al_prepare_video_frame(ALLEGRO_VIDEO* video, int frame_num) (see below)

The semantics of displaying a single video frame are as follows, then:

1. The backend decodes the a frame from the file and sends the ALLEGRO_EVENT_VIDEO_FRAME_PREPARE event with that frame number to the user.

How do you control the number of frames the backend decodes? It shouldn't decode 1000 frames before the first one is shown, that just would waste memory. But it also should not just decode a single frame and risk not being done reading the next frame before it is supposed to be shown.

2. The user receives the ALLEGRO_EVENT_VIDEO_FRAME_PREPARE event and calls al_prepare_video_frame(). This uploads the frame to the GPU.

3. Once the backend decides it's time to show the frame, it sends the ALLEGRO_EVENT_VIDEO_FRAME_SHOW. At the same time, it arranges for the pointer returned by al_get_video_frame to point to the new frame (that was just uploaded by the user in step 2).

4. The user receives the ALLEGRO_EVENT_VIDEO_FRAME_SHOW event, and draws the bitmap returned by al_get_video_frame. al_get_video_frame keeps returning the same frame until the next ALLEGRO_EVENT_VIDEO_FRAME_SHOW.

The ALLEGRO_EVENT_VIDEO_FRAME_PREPARE event is purely an optimization and will not be implemented just yet. It will not be necessary to handle it if the bitmap upload latency doesn't matter. I.e. after the ALLEGRO_EVENT_VIDEO_FRAME_SHOW is sent, al_get_video_frame will do the upload of that frame itself if necessary.

Naturally, ignoring all the events and just calling al_get_video_frame will continue to work.

In terms of implementation, I think it might be best to remove the ffmpeg backend. It doesn't work for me, and I have trouble figuring out why not. The OGV backend also has plenty of issues itself, but at least it plays most of the videos I throw at it. It only supports seeking to the beginning (it's somewhat complicated to implement anything else for theora), but otherwise it's pretty reasonable. I don't think we need to worry about this for 5.2, as the features it does support are plenty for a simple cutscene or some other similar use case.

The only thing that still bugs me is the aspect ratio bit. Ideally, al_draw_bitmap(al_get_video_frame(), 0, 0, 0)); would work, but I don't see a way of getting that to happen.

If you create the video yourself, you can use al_draw_bitmap(al_get_video_frame(), 0, 0, 0)); But an unknown video in general can be optimized to use a non-square aspect ratio, so you need to query it of course. We could pre-scale the bitmaps, but that would be done in software and add quite a bit of useless CPU. Still, I think the use of this video addon will mostly be cut scenes where you already do know the encoded size.

Comments?

-SL

_______________________________________________
Allegro-developers mailing list
Allegro-developers@xxxxxxxxxx
https://mail.gna.org/listinfo/allegro-developers