mpv/video/out/vo.c

1400 lines
42 KiB
C
Raw Normal View History

/*
* This file is part of mpv.
*
vo.c, vo.h, vo_null.c: change license to LGPL Most contributors have agreed. vo.c requires a "driver" entry for each video output - we assume that if someone who didn't agree to LGPL added a line, it's fine for vo.c to be LGPL anyway. If the affected video output is not disabled at compilation time, the resulting binary will be GPL anyway. One problem are the changes by Nick Kurshev (usually using "nick" as SVN username). He could not be reached. I believe all changes to his files are actually gone, but here is a detailed listing: fa1d5742bc: nick introduces a new VO API. It was removed in 64bedd9683. Some of this was replaced by VOCTRLs are introduced in 7c51652a1b, obviously replacing at least some functionality by his API. b587a3d642: nick adds a vo_tune_info_t struct. Removed in 64bedd9683 too. 9caad2c29a: nick adds some VOCTRLs, which were silently removed in 8cc5ba5ab8 (they became unused probably with the VIDIX removal). 340183b0e9: nick adds VO-based screenshots, which got removed in 2f4b840f62. Strangely the same name was introduced in 01cf896a2f again, but this is a coincidence and worked differently (also it was removed yet again in 2858232220). 104c125e6d: nick adds an option for "direct rendering". It was renamed in 6403904ae9 and fully removed in e48b21dd87. 5ddd8e92a1: nick adds code to check the VO driver preinit arg to every single VO driver. The argument itself and any possibly remaining code associated with it was removed in 1f5ffe7d30. f6878753fb: nick adds header inclusion guards. We assume this is not relevant for copyright. Some of nick's code was merely moved to other files, such as the equalizer stuff added in 555c676683 and moved in 4db72f6a80 and 12579136ff, and don't affect copyright of these files anymore. Other notes: fef7b17c34: a patch by someone who wasn't asked for relicensing added a symbol that was removed again in 1b09f4633. 4a8a46fafd: author probably didn't agree to LGPL, but the function signature was changed later on anyway, and nothing of this is left. 7b25afd742: the same author adds a symbol to what is vo.h today, which this relicensing commit removes, as it was unused. (It's not clear whether the mere symbol is copyrightable, but no need to take a risk.) 3a406e94d7, 9dd8f241ac: slave mode things by someone who couldn't be reached. This aspect of the old slave mode was completely removed. bbeb54d80a: patch by someone who was not asked, but the added code was completely removed again.
2017-05-10 12:32:34 +00:00
* mpv is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* mpv is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
vo.c, vo.h, vo_null.c: change license to LGPL Most contributors have agreed. vo.c requires a "driver" entry for each video output - we assume that if someone who didn't agree to LGPL added a line, it's fine for vo.c to be LGPL anyway. If the affected video output is not disabled at compilation time, the resulting binary will be GPL anyway. One problem are the changes by Nick Kurshev (usually using "nick" as SVN username). He could not be reached. I believe all changes to his files are actually gone, but here is a detailed listing: fa1d5742bc: nick introduces a new VO API. It was removed in 64bedd9683. Some of this was replaced by VOCTRLs are introduced in 7c51652a1b, obviously replacing at least some functionality by his API. b587a3d642: nick adds a vo_tune_info_t struct. Removed in 64bedd9683 too. 9caad2c29a: nick adds some VOCTRLs, which were silently removed in 8cc5ba5ab8 (they became unused probably with the VIDIX removal). 340183b0e9: nick adds VO-based screenshots, which got removed in 2f4b840f62. Strangely the same name was introduced in 01cf896a2f again, but this is a coincidence and worked differently (also it was removed yet again in 2858232220). 104c125e6d: nick adds an option for "direct rendering". It was renamed in 6403904ae9 and fully removed in e48b21dd87. 5ddd8e92a1: nick adds code to check the VO driver preinit arg to every single VO driver. The argument itself and any possibly remaining code associated with it was removed in 1f5ffe7d30. f6878753fb: nick adds header inclusion guards. We assume this is not relevant for copyright. Some of nick's code was merely moved to other files, such as the equalizer stuff added in 555c676683 and moved in 4db72f6a80 and 12579136ff, and don't affect copyright of these files anymore. Other notes: fef7b17c34: a patch by someone who wasn't asked for relicensing added a symbol that was removed again in 1b09f4633. 4a8a46fafd: author probably didn't agree to LGPL, but the function signature was changed later on anyway, and nothing of this is left. 7b25afd742: the same author adds a symbol to what is vo.h today, which this relicensing commit removes, as it was unused. (It's not clear whether the mere symbol is copyrightable, but no need to take a risk.) 3a406e94d7, 9dd8f241ac: slave mode things by someone who couldn't be reached. This aspect of the old slave mode was completely removed. bbeb54d80a: patch by someone who was not asked, but the added code was completely removed again.
2017-05-10 12:32:34 +00:00
* GNU Lesser General Public License for more details.
*
vo.c, vo.h, vo_null.c: change license to LGPL Most contributors have agreed. vo.c requires a "driver" entry for each video output - we assume that if someone who didn't agree to LGPL added a line, it's fine for vo.c to be LGPL anyway. If the affected video output is not disabled at compilation time, the resulting binary will be GPL anyway. One problem are the changes by Nick Kurshev (usually using "nick" as SVN username). He could not be reached. I believe all changes to his files are actually gone, but here is a detailed listing: fa1d5742bc: nick introduces a new VO API. It was removed in 64bedd9683. Some of this was replaced by VOCTRLs are introduced in 7c51652a1b, obviously replacing at least some functionality by his API. b587a3d642: nick adds a vo_tune_info_t struct. Removed in 64bedd9683 too. 9caad2c29a: nick adds some VOCTRLs, which were silently removed in 8cc5ba5ab8 (they became unused probably with the VIDIX removal). 340183b0e9: nick adds VO-based screenshots, which got removed in 2f4b840f62. Strangely the same name was introduced in 01cf896a2f again, but this is a coincidence and worked differently (also it was removed yet again in 2858232220). 104c125e6d: nick adds an option for "direct rendering". It was renamed in 6403904ae9 and fully removed in e48b21dd87. 5ddd8e92a1: nick adds code to check the VO driver preinit arg to every single VO driver. The argument itself and any possibly remaining code associated with it was removed in 1f5ffe7d30. f6878753fb: nick adds header inclusion guards. We assume this is not relevant for copyright. Some of nick's code was merely moved to other files, such as the equalizer stuff added in 555c676683 and moved in 4db72f6a80 and 12579136ff, and don't affect copyright of these files anymore. Other notes: fef7b17c34: a patch by someone who wasn't asked for relicensing added a symbol that was removed again in 1b09f4633. 4a8a46fafd: author probably didn't agree to LGPL, but the function signature was changed later on anyway, and nothing of this is left. 7b25afd742: the same author adds a symbol to what is vo.h today, which this relicensing commit removes, as it was unused. (It's not clear whether the mere symbol is copyrightable, but no need to take a risk.) 3a406e94d7, 9dd8f241ac: slave mode things by someone who couldn't be reached. This aspect of the old slave mode was completely removed. bbeb54d80a: patch by someone who was not asked, but the added code was completely removed again.
2017-05-10 12:32:34 +00:00
* You should have received a copy of the GNU Lesser General Public
* License along with mpv. If not, see <http://www.gnu.org/licenses/>.
*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
core/VO: Allow VO drivers to add/modify frames Add interfaces to allow VO drivers to add or remove frames from the video stream and to alter timestamps. Currently this functionality only works with in correct-pts mode. Use the new functionality in vo_vdpau to properly support frame-adding deinterlace modes. Frames added by the VDPAU deinterlacing code are now properly timed. Before every second frame was always shown immediately (probably next monitor refresh) after the previous one, even if you were watching things in slow motion, and framestepping didn't stop at them at all. When seeking the deinterlace algorithm is no longer fed a mix of frames from old and new positions. As a side effect of the changes a problem with resize events was also fixed. Resizing calls video_to_output_surface() to render the frame at the new resolution, but before this function also changed the list of history frames, so resizing could give an image different from the original one, and also corrupt next frames due to them seeing the wrong history. Now the function has no such side effects. There are more resize-related problems though that will be fixed in a later commit. The deint_mpi[] list of reserved frames is increased from 2 to 3 entries for reasons related to the above. Having 2 entries is enough when you initially get a new frame in draw_image() because then you'll have those two entries plus the new one for a total of 3 (the code relied on the oldest mpi implicitly staying reserved for the duration of the call even after usage count was decreased). However if you want to be able to reproduce the rendering outside draw_image(), relying on the explicitly reserved list only, then it needs to store 3 entries.
2009-09-18 13:27:55 +00:00
#include <stdbool.h>
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
#include <pthread.h>
#include <math.h>
#include "mpv_talloc.h"
#include "config.h"
#include "osdep/atomic.h"
#include "osdep/timer.h"
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
#include "osdep/threads.h"
#include "misc/dispatch.h"
#include "misc/rendezvous.h"
#include "options/options.h"
#include "misc/bstr.h"
#include "vo.h"
#include "aspect.h"
#include "dr_helper.h"
2013-12-17 00:23:09 +00:00
#include "input/input.h"
#include "options/m_config.h"
#include "common/msg.h"
#include "common/global.h"
#include "video/hwdec.h"
#include "video/mp_image.h"
#include "sub/osd.h"
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
#include "osdep/io.h"
#include "osdep/threads.h"
2017-10-09 19:45:26 +00:00
extern const struct vo_driver video_out_mediacodec_embed;
extern const struct vo_driver video_out_x11;
extern const struct vo_driver video_out_vdpau;
extern const struct vo_driver video_out_xv;
vo_opengl: refactor into vo_gpu This is done in several steps: 1. refactor MPGLContext -> struct ra_ctx 2. move GL-specific stuff in vo_opengl into opengl/context.c 3. generalize context creation to support other APIs, and add --gpu-api 4. rename all of the --opengl- options that are no longer opengl-specific 5. move all of the stuff from opengl/* that isn't GL-specific into gpu/ (note: opengl/gl_utils.h became opengl/utils.h) 6. rename vo_opengl to vo_gpu 7. to handle window screenshots, the short-term approach was to just add it to ra_swchain_fns. Long term (and for vulkan) this has to be moved to ra itself (and vo_gpu altered to compensate), but this was a stop-gap measure to prevent this commit from getting too big 8. move ra->fns->flush to ra_gl_ctx instead 9. some other minor changes that I've probably already forgotten Note: This is one half of a major refactor, the other half of which is provided by rossy's following commit. This commit enables support for all linux platforms, while his version enables support for all non-linux platforms. Note 2: vo_opengl_cb.c also re-uses ra_gl_ctx so it benefits from the --opengl- options like --opengl-early-flush, --opengl-finish etc. Should be a strict superset of the old functionality. Disclaimer: Since I have no way of compiling mpv on all platforms, some of these ports were done blindly. Specifically, the blind ports included context_mali_fbdev.c and context_rpi.c. Since they're both based on egl_helpers, the port should have gone smoothly without any major changes required. But if somebody complains about a compile error on those platforms (assuming anybody actually uses them), you know where to complain.
2017-09-14 06:04:55 +00:00
extern const struct vo_driver video_out_gpu;
extern const struct vo_driver video_out_libmpv;
extern const struct vo_driver video_out_null;
extern const struct vo_driver video_out_image;
extern const struct vo_driver video_out_lavc;
extern const struct vo_driver video_out_caca;
extern const struct vo_driver video_out_drm;
extern const struct vo_driver video_out_direct3d;
extern const struct vo_driver video_out_sdl;
extern const struct vo_driver video_out_vaapi;
extern const struct vo_driver video_out_wlshm;
RPI support This requires FFmpeg git master for accelerated hardware decoding. Keep in mind that FFmpeg must be compiled with --enable-mmal. Libav will also work. Most things work. Screenshots don't work with accelerated/opaque decoding (except using full window screenshot mode). Subtitles are very slow - even simple but huge overlays can cause frame drops. This always uses fullscreen mode. It uses dispmanx and mmal directly, and there are no window managers or anything on this level. vo_opengl also kind of works, but is pretty useless and slow. It can't use opaque hardware decoding (copy back can be used by forcing the option --vd=lavc:h264_mmal). Keep in mind that the dispmanx backend is preferred over the X11 ones in case you're trying on X11; but X11 is even more useless on RPI. This doesn't correctly reject extended h264 profiles and thus doesn't fallback to software decoding. The hw supports only up to the high profile, and will e.g. return garbage for Hi10P video. This sets a precedent of enabling hw decoding by default, but only if RPI support is compiled (which most hopefully it will be disabled on desktop Linux platforms). While it's more or less required to use hw decoding on the weak RPI, it causes more problems than it solves on real platforms (Linux has the Intel GPU problem, OSX still has some cases with broken decoding.) So I can live with this compromise of having different defaults depending on the platform. Raspberry Pi 2 is required. This wasn't tested on the original RPI, though at least decoding itself seems to work (but full playback was not tested).
2015-03-29 13:12:11 +00:00
extern const struct vo_driver video_out_rpi;
extern const struct vo_driver video_out_tct;
const struct vo_driver *const video_out_drivers[] =
{
&video_out_libmpv,
#if HAVE_ANDROID
2017-10-09 19:45:26 +00:00
&video_out_mediacodec_embed,
#endif
vo_opengl: refactor into vo_gpu This is done in several steps: 1. refactor MPGLContext -> struct ra_ctx 2. move GL-specific stuff in vo_opengl into opengl/context.c 3. generalize context creation to support other APIs, and add --gpu-api 4. rename all of the --opengl- options that are no longer opengl-specific 5. move all of the stuff from opengl/* that isn't GL-specific into gpu/ (note: opengl/gl_utils.h became opengl/utils.h) 6. rename vo_opengl to vo_gpu 7. to handle window screenshots, the short-term approach was to just add it to ra_swchain_fns. Long term (and for vulkan) this has to be moved to ra itself (and vo_gpu altered to compensate), but this was a stop-gap measure to prevent this commit from getting too big 8. move ra->fns->flush to ra_gl_ctx instead 9. some other minor changes that I've probably already forgotten Note: This is one half of a major refactor, the other half of which is provided by rossy's following commit. This commit enables support for all linux platforms, while his version enables support for all non-linux platforms. Note 2: vo_opengl_cb.c also re-uses ra_gl_ctx so it benefits from the --opengl- options like --opengl-early-flush, --opengl-finish etc. Should be a strict superset of the old functionality. Disclaimer: Since I have no way of compiling mpv on all platforms, some of these ports were done blindly. Specifically, the blind ports included context_mali_fbdev.c and context_rpi.c. Since they're both based on egl_helpers, the port should have gone smoothly without any major changes required. But if somebody complains about a compile error on those platforms (assuming anybody actually uses them), you know where to complain.
2017-09-14 06:04:55 +00:00
&video_out_gpu,
#if HAVE_VDPAU
&video_out_vdpau,
#endif
#if HAVE_DIRECT3D
&video_out_direct3d,
#endif
#if HAVE_WAYLAND && HAVE_MEMFD_CREATE
&video_out_wlshm,
#endif
#if HAVE_XV
&video_out_xv,
#endif
#if HAVE_SDL2_VIDEO
&video_out_sdl,
#endif
#if HAVE_VAAPI_X11 && HAVE_GPL
&video_out_vaapi,
#endif
#if HAVE_X11
&video_out_x11,
#endif
&video_out_null,
// should not be auto-selected
&video_out_image,
&video_out_tct,
#if HAVE_CACA
&video_out_caca,
#endif
#if HAVE_DRM
&video_out_drm,
2019-07-14 13:56:10 +00:00
#endif
#if HAVE_RPI_MMAL
&video_out_rpi,
#endif
&video_out_lavc,
NULL
};
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct vo_internal {
pthread_t thread;
struct mp_dispatch_queue *dispatch;
struct dr_helper *dr_helper;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// --- The following fields are protected by lock
pthread_mutex_t lock;
pthread_cond_t wakeup;
bool need_wakeup;
bool terminate;
bool hasframe;
bool hasframe_rendered;
bool request_redraw; // redraw request from player to VO
bool want_redraw; // redraw request from VO to player
bool send_reset; // send VOCTRL_RESET
bool paused;
int queued_events; // event mask for the user
int internal_events; // event mask for us
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
int64_t nominal_vsync_interval;
int64_t vsync_interval;
int64_t *vsync_samples;
int num_vsync_samples;
int64_t num_total_vsync_samples;
int64_t prev_vsync;
int64_t base_vsync;
int drop_point;
double estimated_vsync_interval;
double estimated_vsync_jitter;
bool expecting_vsync;
int64_t num_successive_vsyncs;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
int64_t flip_queue_offset; // queue flip events at most this much in advance
int64_t timing_offset; // same (but from options; not VO configured)
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
int64_t delayed_count;
int64_t drop_count;
bool dropped_frame; // the previous frame was dropped
struct vo_frame *current_frame; // last frame queued to the VO
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
int64_t wakeup_pts; // time at which to pull frame from decoder
bool rendering; // true if an image is being rendered
struct vo_frame *frame_queued; // should be drawn next
int req_frames; // VO's requested value of num_frames
uint64_t current_frame_id;
double display_fps;
double reported_display_fps;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
};
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
extern const struct m_sub_options gl_video_conf;
static void forget_frames(struct vo *vo);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
static void *vo_thread(void *ptr);
static bool get_desc(struct m_obj_desc *dst, int index)
{
if (index >= MP_ARRAY_SIZE(video_out_drivers) - 1)
return false;
const struct vo_driver *vo = video_out_drivers[index];
*dst = (struct m_obj_desc) {
.name = vo->name,
.description = vo->description,
.priv_size = vo->priv_size,
.priv_defaults = vo->priv_defaults,
.options = vo->options,
.options_prefix = vo->options_prefix,
.global_opts = vo->global_opts,
.hidden = vo->encode,
.p = vo,
};
return true;
}
// For the vo option
const struct m_obj_list vo_obj_list = {
.get_desc = get_desc,
.description = "video outputs",
.aliases = {
{"gl", "gpu"},
{"direct3d_shaders", "direct3d"},
{"opengl", "gpu"},
{"opengl-cb", "libmpv"},
{0}
},
.allow_unknown_entries = true,
.allow_trailer = true,
.disallow_positional_parameters = true,
.use_global_options = true,
};
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
static void dispatch_wakeup_cb(void *ptr)
{
struct vo *vo = ptr;
vo_wakeup(vo);
}
// Initialize or update options from vo->opts
static void read_opts(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
in->timing_offset = (uint64_t)(vo->opts->timing_offset * 1e6);
pthread_mutex_unlock(&in->lock);
}
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
static void update_opts(void *p)
{
struct vo *vo = p;
if (m_config_cache_update(vo->opts_cache)) {
read_opts(vo);
if (vo->driver->control) {
vo->driver->control(vo, VOCTRL_VO_OPTS_CHANGED, NULL);
// "Legacy" update of video position related options.
// Unlike VOCTRL_VO_OPTS_CHANGED, often not propagated to backends.
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
vo->driver->control(vo, VOCTRL_SET_PANSCAN, NULL);
}
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
}
if (vo->gl_opts_cache && m_config_cache_update(vo->gl_opts_cache)) {
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
// "Legacy" update of video GL renderer related options.
if (vo->driver->control)
vo->driver->control(vo, VOCTRL_UPDATE_RENDER_OPTS, NULL);
}
if (m_config_cache_update(vo->eq_opts_cache)) {
// "Legacy" update of video equalizer related options.
if (vo->driver->control)
vo->driver->control(vo, VOCTRL_SET_EQUALIZER, NULL);
}
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// Does not include thread- and VO uninit.
static void dealloc_vo(struct vo *vo)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
forget_frames(vo); // implicitly synchronized
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
// These must be free'd before vo->in->dispatch.
talloc_free(vo->opts_cache);
talloc_free(vo->gl_opts_cache);
talloc_free(vo->eq_opts_cache);
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
pthread_mutex_destroy(&vo->in->lock);
pthread_cond_destroy(&vo->in->wakeup);
talloc_free(vo);
}
static struct vo *vo_create(bool probing, struct mpv_global *global,
struct vo_extra *ex, char *name)
{
assert(ex->wakeup_cb);
struct mp_log *log = mp_log_new(NULL, global->log, "vo");
struct m_obj_desc desc;
if (!m_obj_list_find(&desc, &vo_obj_list, bstr0(name))) {
mp_msg(log, MSGL_ERR, "Video output %s not found!\n", name);
talloc_free(log);
return NULL;
};
struct vo *vo = talloc_ptrtype(NULL, vo);
*vo = (struct vo) {
.log = mp_log_new(vo, log, name),
.driver = desc.p,
.global = global,
.encode_lavc_ctx = ex->encode_lavc_ctx,
.input_ctx = ex->input_ctx,
.osd = ex->osd,
.monitor_par = 1,
.extra = *ex,
.probing = probing,
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
.in = talloc(vo, struct vo_internal),
};
talloc_steal(vo, log);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
*vo->in = (struct vo_internal) {
.dispatch = mp_dispatch_create(vo),
.req_frames = 1,
.estimated_vsync_jitter = -1,
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
};
mp_dispatch_set_wakeup_fn(vo->in->dispatch, dispatch_wakeup_cb, vo);
pthread_mutex_init(&vo->in->lock, NULL);
pthread_cond_init(&vo->in->wakeup, NULL);
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
vo->opts_cache = m_config_cache_alloc(NULL, global, &vo_sub_opts);
vo->opts = vo->opts_cache->opts;
options: add a thread-safe way to notify option updates So far, we had a thread-safe way to read options, but no option update notification mechanism. Everything was funneled though the main thread's central mp_option_change_callback() function. For example, if the panscan options were changed, the function called vo_control() with VOCTRL_SET_PANSCAN to manually notify the VO thread of updates. This worked, but's pretty inconvenient. Most of these problems come from the fact that MPlayer was written as a single-threaded program. This commit works towards a more flexible mechanism. It adds an update callback to m_config_cache (the thing that is already used for thread-safe access of global options). This alone would still be rather inconvenient, at least in context of VOs. Add another mechanism on top of it that uses mp_dispatch_queue, and takes care of some annoying synchronization issues. We extend mp_dispatch_queue itself to make this easier and slightly more efficient. As a first application, use this to reimplement certain VO scaling and renderer options. The update_opts() function translates these to the "old" VOCTRLs, though. An annoyingly subtle issue is that m_config_cache's destructor now releases pending notifications, and must be released before the associated dispatch queue. Otherwise, it could happen that option updates during e.g. VO destruction queue or run stale entries, which is not expected. Rather untested. The singly-linked list code in dispatch.c is probably buggy, and I bet some aspects about synchronization are not entirely sane.
2017-08-22 13:50:33 +00:00
m_config_cache_set_dispatch_change_cb(vo->opts_cache, vo->in->dispatch,
update_opts, vo);
vo->gl_opts_cache = m_config_cache_alloc(NULL, global, &gl_video_conf);
m_config_cache_set_dispatch_change_cb(vo->gl_opts_cache, vo->in->dispatch,
update_opts, vo);
vo->eq_opts_cache = m_config_cache_alloc(NULL, global, &mp_csp_equalizer_conf);
m_config_cache_set_dispatch_change_cb(vo->eq_opts_cache, vo->in->dispatch,
update_opts, vo);
mp_input_set_mouse_transform(vo->input_ctx, NULL, NULL);
if (vo->driver->encode != !!vo->encode_lavc_ctx)
goto error;
vo->priv = m_config_group_from_desc(vo, vo->log, global, &desc, name);
if (!vo->priv)
goto error;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
if (pthread_create(&vo->in->thread, NULL, vo_thread, vo))
goto error;
if (mp_rendezvous(vo, 0) < 0) { // init barrier
pthread_join(vo->in->thread, NULL);
goto error;
}
return vo;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
error:
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
dealloc_vo(vo);
return NULL;
}
struct vo *init_best_video_out(struct mpv_global *global, struct vo_extra *ex)
{
struct mp_vo_opts *opts = mp_get_config_group(NULL, global, &vo_sub_opts);
struct m_obj_settings *vo_list = opts->video_driver_list;
struct vo *vo = NULL;
// first try the preferred drivers, with their optional subdevice param:
if (vo_list && vo_list[0].name) {
for (int n = 0; vo_list[n].name; n++) {
// Something like "-vo name," allows fallback to autoprobing.
if (strlen(vo_list[n].name) == 0)
goto autoprobe;
bool p = !!vo_list[n + 1].name;
vo = vo_create(p, global, ex, vo_list[n].name);
if (vo)
goto done;
}
goto done;
}
autoprobe:
// now try the rest...
for (int i = 0; video_out_drivers[i]; i++) {
const struct vo_driver *driver = video_out_drivers[i];
if (driver == &video_out_null)
break;
vo = vo_create(true, global, ex, (char *)driver->name);
if (vo)
goto done;
}
done:
talloc_free(opts);
return vo;
}
static void terminate_vo(void *p)
{
struct vo *vo = p;
struct vo_internal *in = vo->in;
in->terminate = true;
}
void vo_destroy(struct vo *vo)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct vo_internal *in = vo->in;
mp_dispatch_run(in->dispatch, terminate_vo, vo);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
pthread_join(vo->in->thread, NULL);
dealloc_vo(vo);
}
// Wakeup the playloop to queue new video frames etc.
static void wakeup_core(struct vo *vo)
{
vo->extra.wakeup_cb(vo->extra.wakeup_ctx);
}
// Drop timing information on discontinuities like seeking.
// Always called locked.
static void reset_vsync_timings(struct vo *vo)
{
struct vo_internal *in = vo->in;
in->drop_point = 0;
in->base_vsync = 0;
in->expecting_vsync = false;
in->num_successive_vsyncs = 0;
}
static double vsync_stddef(struct vo *vo, int64_t ref_vsync)
{
struct vo_internal *in = vo->in;
double jitter = 0;
for (int n = 0; n < in->num_vsync_samples; n++) {
double diff = in->vsync_samples[n] - ref_vsync;
jitter += diff * diff;
}
return sqrt(jitter / in->num_vsync_samples);
}
#define MAX_VSYNC_SAMPLES 200
// Check if we should switch to measured average display FPS if it seems
// "better" then the system-reported one. (Note that small differences are
// handled as drift instead.)
static void check_estimated_display_fps(struct vo *vo)
{
struct vo_internal *in = vo->in;
bool use_estimated = false;
if (in->num_total_vsync_samples >= MAX_VSYNC_SAMPLES / 2 &&
in->estimated_vsync_interval <= 1e6 / 20.0 &&
in->estimated_vsync_interval >= 1e6 / 99.0)
{
for (int n = 0; n < in->num_vsync_samples; n++) {
if (fabs(in->vsync_samples[n] - in->estimated_vsync_interval)
>= in->estimated_vsync_interval / 4)
goto done;
}
double mjitter = vsync_stddef(vo, in->estimated_vsync_interval);
double njitter = vsync_stddef(vo, in->nominal_vsync_interval);
if (mjitter * 1.01 < njitter)
use_estimated = true;
done: ;
}
if (use_estimated == (in->vsync_interval == in->nominal_vsync_interval)) {
if (use_estimated) {
MP_VERBOSE(vo, "adjusting display FPS to a value closer to %.3f Hz\n",
1e6 / in->estimated_vsync_interval);
} else {
MP_VERBOSE(vo, "switching back to assuming display fps = %.3f Hz\n",
1e6 / in->nominal_vsync_interval);
}
}
in->vsync_interval = use_estimated ? (int64_t)in->estimated_vsync_interval
: in->nominal_vsync_interval;
}
// Attempt to detect vsyncs delayed/skipped by the driver. This tries to deal
// with strong jitter too, because some drivers have crap vsync timing.
static void vsync_skip_detection(struct vo *vo)
{
struct vo_internal *in = vo->in;
int window = 4;
int64_t t_r = in->prev_vsync, t_e = in->base_vsync, diff = 0, desync_early = 0;
for (int n = 0; n < in->drop_point; n++) {
diff += t_r - t_e;
t_r -= in->vsync_samples[n];
t_e -= in->vsync_interval;
if (n == window + 1)
desync_early = diff / window;
}
int64_t desync = diff / in->num_vsync_samples;
if (in->drop_point > window * 2 &&
llabs(desync - desync_early) >= in->vsync_interval * 3 / 4)
{
// Assume a drop. An underflow can technically speaking not be a drop
// (it's up to the driver what this is supposed to mean), but no reason
// to treat it differently.
in->base_vsync = in->prev_vsync;
in->delayed_count += 1;
in->drop_point = 0;
MP_STATS(vo, "vo-delayed");
}
if (in->drop_point > 10)
in->base_vsync += desync / 10; // smooth out drift
}
// Always called locked.
static void update_vsync_timing_after_swap(struct vo *vo,
struct vo_vsync_info *vsync)
{
struct vo_internal *in = vo->in;
int64_t vsync_time = vsync->last_queue_display_time;
int64_t prev_vsync = in->prev_vsync;
in->prev_vsync = vsync_time;
if (!in->expecting_vsync) {
reset_vsync_timings(vo);
return;
}
in->num_successive_vsyncs++;
if (in->num_successive_vsyncs <= 2)
return;
if (in->num_vsync_samples >= MAX_VSYNC_SAMPLES)
in->num_vsync_samples -= 1;
MP_TARRAY_INSERT_AT(in, in->vsync_samples, in->num_vsync_samples, 0,
vsync_time - prev_vsync);
in->drop_point = MPMIN(in->drop_point + 1, in->num_vsync_samples);
in->num_total_vsync_samples += 1;
if (in->base_vsync) {
in->base_vsync += in->vsync_interval;
} else {
in->base_vsync = vsync_time;
}
double avg = 0;
for (int n = 0; n < in->num_vsync_samples; n++)
avg += in->vsync_samples[n];
in->estimated_vsync_interval = avg / in->num_vsync_samples;
in->estimated_vsync_jitter =
vsync_stddef(vo, in->vsync_interval) / in->vsync_interval;
check_estimated_display_fps(vo);
vsync_skip_detection(vo);
MP_STATS(vo, "value %f jitter", in->estimated_vsync_jitter);
MP_STATS(vo, "value %f vsync-diff", in->vsync_samples[0] / 1e6);
}
// to be called from VO thread only
static void update_display_fps(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
if (in->internal_events & VO_EVENT_WIN_STATE) {
in->internal_events &= ~(unsigned)VO_EVENT_WIN_STATE;
pthread_mutex_unlock(&in->lock);
double fps = 0;
vo->driver->control(vo, VOCTRL_GET_DISPLAY_FPS, &fps);
pthread_mutex_lock(&in->lock);
in->reported_display_fps = fps;
}
double display_fps = vo->opts->override_display_fps;
if (display_fps <= 0)
display_fps = in->reported_display_fps;
if (in->display_fps != display_fps) {
in->nominal_vsync_interval = display_fps > 0 ? 1e6 / display_fps : 0;
in->vsync_interval = MPMAX(in->nominal_vsync_interval, 1);
in->display_fps = display_fps;
MP_VERBOSE(vo, "Assuming %f FPS for display sync.\n", display_fps);
// make sure to update the player
in->queued_events |= VO_EVENT_WIN_STATE;
wakeup_core(vo);
}
pthread_mutex_unlock(&in->lock);
}
static void check_vo_caps(struct vo *vo)
{
int rot = vo->params->rotate;
if (rot) {
bool ok = rot % 90 ? false : (vo->driver->caps & VO_CAP_ROTATE90);
if (!ok) {
MP_WARN(vo, "Video is flagged as rotated by %d degrees, but the "
"video output does not support this.\n", rot);
}
}
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
static void run_reconfig(void *p)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
void **pp = p;
struct vo *vo = pp[0];
struct mp_image *img = pp[1];
int *ret = pp[2];
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct mp_image_params *params = &img->params;
struct vo_internal *in = vo->in;
MP_VERBOSE(vo, "reconfig to %s\n", mp_image_params_to_str(params));
vo: fix missed option updates under rare circumstances Dear diary, today I fixed a shitty bug that was all my fault because I made a horrible mess. (Except it was a horrible mess before I even touched this shit, but let's not blame others.) Sometimes, updates to VO option that control video sizing (like panscan) didn't update the screen correctly. They were delayed until the next option change or so. It turns out that if the option update happens at the "same" time as a VOCTRL, update_opts() doesn't actually notify the vo_driver of the change. This in turn happened because run_control() called m_config_cache_update(). The latter function returns true if the options changed since the last call, and update_opts() also calls it (on the same config cache) for the same purpose. The update_opts() call, which is triggered by a third mechanism, comes later, but the cache update call will return false (as it should). Basically, given the config API, you can't act differently on multiple update calls and expect it to work. The skipped handling in update_opts() meant that the notification required to apply the changed option wasn't run. Fix this by simply calling update_opts() directly instead. Now there's only 1 m_config_cache_update() call on this specific instance. Fix the call in run_reconfig() too, so the previous sentence isn't a lie (but it probably doesn't make a difference in practice due to certain details). I'm not sure how I even ran into this sort-of race condition. The VOCTRL that messed up the option update was VOCTRL_UPDATE_PLAYBACK_STATE, which happens semi-regularly. Why this config cache shit and all the other shit? Rediscovering this crap wasn't pleasant. It's a bunch of hacks that became necessary when the ancient MPlayer architecture made it hard to move the VO to a separate thread. All the VO code typically accesses vo->opts (whose fields all used to be global variables in MPlayer). The frontend changes these on user input. Putting locking around all the options would be a nightmare, and keeping a copy of the options in the thread was much simpler. You need a way to propagate option changes, notify the thread, and update the local copy too. And the result of these thoughts was the config cache mechanism. In this specific case, the relevant cache update call in update_opts() triggers a VOCTRL_SET_PANSCAN to the VO driver, which isn't related to its former function anymore. Instead, it causes the VO driver to update the video sizing/placing options, which the generic VO code can't do. (Mostly because the VO driver includes the windowing stuff and is responsible for resizing etc. itself.) VOCTRLs sent by the frontend are even worse. MPlayer had no real runtime option change mechanism. Some options were vaguely duplicated by properties, so you could effectively change those options at runtime. Each of these options had its own VOCTRL, which still exist today, e.g. VOCTRL_FULLSCREEN, or VOCTRL_ONTOP. I tried to make all options runtime changeable, and to unify properties with options. But I couldn't be bothered with updating all VO drivers to listen to option changes directly, because that would be pretty tedious. So the property code is still all there and sends the old VOCTRLs. But of course you need to sync up the options, which is why the run_control() code did that. (Unrelated: VO_EVENT_FULLSCREEN_STATE is the worst shithack of them all. Currently, only the frontend can actually write to options (for awful reasons), so if the fullscreen state changes due to outside interaction, the VO driver can't update the corresponding option fields. So the VO notifies the frontend with said VO_EVENT_, and the frontend then sends VOCTRL_GET_FULLSCREEN, and updates the global copy of the option with the value returned by that. I still like to think the situation is not that bad considering the monstrous effort of converting single-threaded code that had hundreds of options in global variables to multi-threaded code with no global variables at all.)
2019-07-12 01:30:52 +00:00
update_opts(vo);
mp_image_params_get_dsize(params, &vo->dwidth, &vo->dheight);
talloc_free(vo->params);
vo->params = talloc_dup(vo, params);
if (vo->driver->reconfig2) {
*ret = vo->driver->reconfig2(vo, img);
} else {
*ret = vo->driver->reconfig(vo, vo->params);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
vo->config_ok = *ret >= 0;
if (vo->config_ok) {
check_vo_caps(vo);
} else {
talloc_free(vo->params);
vo->params = NULL;
}
pthread_mutex_lock(&in->lock);
talloc_free(in->current_frame);
in->current_frame = NULL;
forget_frames(vo);
reset_vsync_timings(vo);
pthread_mutex_unlock(&in->lock);
update_display_fps(vo);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
}
int vo_reconfig(struct vo *vo, struct mp_image_params *params)
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
{
int ret;
struct mp_image dummy = {0};
mp_image_set_params(&dummy, params);
void *p[] = {vo, &dummy, &ret};
mp_dispatch_run(vo->in->dispatch, run_reconfig, p);
return ret;
}
int vo_reconfig2(struct vo *vo, struct mp_image *img)
{
int ret;
void *p[] = {vo, img, &ret};
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
mp_dispatch_run(vo->in->dispatch, run_reconfig, p);
return ret;
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
static void run_control(void *p)
{
void **pp = p;
struct vo *vo = pp[0];
int request = (intptr_t)pp[1];
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
void *data = pp[2];
vo: fix missed option updates under rare circumstances Dear diary, today I fixed a shitty bug that was all my fault because I made a horrible mess. (Except it was a horrible mess before I even touched this shit, but let's not blame others.) Sometimes, updates to VO option that control video sizing (like panscan) didn't update the screen correctly. They were delayed until the next option change or so. It turns out that if the option update happens at the "same" time as a VOCTRL, update_opts() doesn't actually notify the vo_driver of the change. This in turn happened because run_control() called m_config_cache_update(). The latter function returns true if the options changed since the last call, and update_opts() also calls it (on the same config cache) for the same purpose. The update_opts() call, which is triggered by a third mechanism, comes later, but the cache update call will return false (as it should). Basically, given the config API, you can't act differently on multiple update calls and expect it to work. The skipped handling in update_opts() meant that the notification required to apply the changed option wasn't run. Fix this by simply calling update_opts() directly instead. Now there's only 1 m_config_cache_update() call on this specific instance. Fix the call in run_reconfig() too, so the previous sentence isn't a lie (but it probably doesn't make a difference in practice due to certain details). I'm not sure how I even ran into this sort-of race condition. The VOCTRL that messed up the option update was VOCTRL_UPDATE_PLAYBACK_STATE, which happens semi-regularly. Why this config cache shit and all the other shit? Rediscovering this crap wasn't pleasant. It's a bunch of hacks that became necessary when the ancient MPlayer architecture made it hard to move the VO to a separate thread. All the VO code typically accesses vo->opts (whose fields all used to be global variables in MPlayer). The frontend changes these on user input. Putting locking around all the options would be a nightmare, and keeping a copy of the options in the thread was much simpler. You need a way to propagate option changes, notify the thread, and update the local copy too. And the result of these thoughts was the config cache mechanism. In this specific case, the relevant cache update call in update_opts() triggers a VOCTRL_SET_PANSCAN to the VO driver, which isn't related to its former function anymore. Instead, it causes the VO driver to update the video sizing/placing options, which the generic VO code can't do. (Mostly because the VO driver includes the windowing stuff and is responsible for resizing etc. itself.) VOCTRLs sent by the frontend are even worse. MPlayer had no real runtime option change mechanism. Some options were vaguely duplicated by properties, so you could effectively change those options at runtime. Each of these options had its own VOCTRL, which still exist today, e.g. VOCTRL_FULLSCREEN, or VOCTRL_ONTOP. I tried to make all options runtime changeable, and to unify properties with options. But I couldn't be bothered with updating all VO drivers to listen to option changes directly, because that would be pretty tedious. So the property code is still all there and sends the old VOCTRLs. But of course you need to sync up the options, which is why the run_control() code did that. (Unrelated: VO_EVENT_FULLSCREEN_STATE is the worst shithack of them all. Currently, only the frontend can actually write to options (for awful reasons), so if the fullscreen state changes due to outside interaction, the VO driver can't update the corresponding option fields. So the VO notifies the frontend with said VO_EVENT_, and the frontend then sends VOCTRL_GET_FULLSCREEN, and updates the global copy of the option with the value returned by that. I still like to think the situation is not that bad considering the monstrous effort of converting single-threaded code that had hundreds of options in global variables to multi-threaded code with no global variables at all.)
2019-07-12 01:30:52 +00:00
update_opts(vo);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
int ret = vo->driver->control(vo, request, data);
if (pp[3])
*(int *)pp[3] = ret;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
}
int vo_control(struct vo *vo, int request, void *data)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
int ret;
void *p[] = {vo, (void *)(intptr_t)request, data, &ret};
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
mp_dispatch_run(vo->in->dispatch, run_control, p);
return ret;
}
// Run vo_control() without waiting for a reply.
// (Only works for some VOCTRLs.)
void vo_control_async(struct vo *vo, int request, void *data)
{
void *p[4] = {vo, (void *)(intptr_t)request, NULL, NULL};
void **d = talloc_memdup(NULL, p, sizeof(p));
switch (request) {
case VOCTRL_UPDATE_PLAYBACK_STATE:
d[2] = talloc_dup(d, (struct voctrl_playback_state *)data);
break;
case VOCTRL_KILL_SCREENSAVER:
case VOCTRL_RESTORE_SCREENSAVER:
break;
default:
abort(); // requires explicit support
}
mp_dispatch_enqueue_autofree(vo->in->dispatch, run_control, d);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// must be called locked
static void forget_frames(struct vo *vo)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct vo_internal *in = vo->in;
in->hasframe = false;
in->hasframe_rendered = false;
in->drop_count = 0;
in->delayed_count = 0;
talloc_free(in->frame_queued);
in->frame_queued = NULL;
in->current_frame_id += VO_MAX_REQ_FRAMES + 1;
// don't unref current_frame; we always want to be able to redraw it
if (in->current_frame) {
in->current_frame->num_vsyncs = 0; // but reset future repeats
in->current_frame->display_synced = false; // mark discontinuity
}
}
// VOs which have no special requirements on UI event loops etc. can set the
// vo_driver.wait_events callback to this (and leave vo_driver.wakeup unset).
// This function must not be used or called for other purposes.
void vo_wait_default(struct vo *vo, int64_t until_time)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
if (!in->need_wakeup) {
struct timespec ts = mp_time_us_to_timespec(until_time);
pthread_cond_timedwait(&in->wakeup, &in->lock, &ts);
}
pthread_mutex_unlock(&in->lock);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// Called unlocked.
static void wait_vo(struct vo *vo, int64_t until_time)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct vo_internal *in = vo->in;
if (vo->driver->wait_events) {
vo->driver->wait_events(vo, until_time);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
} else {
vo_wait_default(vo, until_time);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
}
pthread_mutex_lock(&in->lock);
in->need_wakeup = false;
pthread_mutex_unlock(&in->lock);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
static void wakeup_locked(struct vo *vo)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct vo_internal *in = vo->in;
pthread_cond_broadcast(&in->wakeup);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
if (vo->driver->wakeup)
vo->driver->wakeup(vo);
in->need_wakeup = true;
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// Wakeup VO thread, and make it check for new events with VOCTRL_CHECK_EVENTS.
// To be used by threaded VO backends.
void vo_wakeup(struct vo *vo)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
wakeup_locked(vo);
pthread_mutex_unlock(&in->lock);
}
// Whether vo_queue_frame() can be called. If the VO is not ready yet, the
// function will return false, and the VO will call the wakeup callback once
// it's ready.
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// next_pts is the exact time when the next frame should be displayed. If the
// VO is ready, but the time is too "early", return false, and call the wakeup
// callback once the time is right.
// If next_pts is negative, disable any timing and draw the frame as fast as
// possible.
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
bool vo_is_ready_for_frame(struct vo *vo, int64_t next_pts)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
bool blocked = vo->driver->initially_blocked &&
!(in->internal_events & VO_EVENT_INITIAL_UNBLOCK);
bool r = vo->config_ok && !in->frame_queued && !blocked &&
(!in->current_frame || in->current_frame->num_vsyncs < 1);
if (r && next_pts >= 0) {
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// Don't show the frame too early - it would basically freeze the
// display by disallowing OSD redrawing or VO interaction.
// Actually render the frame at earliest the given offset before target
// time.
next_pts -= in->timing_offset;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
next_pts -= in->flip_queue_offset;
int64_t now = mp_time_us();
if (next_pts > now)
r = false;
if (!in->wakeup_pts || next_pts < in->wakeup_pts) {
in->wakeup_pts = next_pts;
// If we have to wait, update the vo thread's timer.
if (!r)
wakeup_locked(vo);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
}
}
pthread_mutex_unlock(&in->lock);
return r;
}
// Direct the VO thread to put the currently queued image on the screen.
// vo_is_ready_for_frame() must have returned true before this call.
// Ownership of frame is handed to the vo.
void vo_queue_frame(struct vo *vo, struct vo_frame *frame)
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
assert(vo->config_ok && !in->frame_queued &&
(!in->current_frame || in->current_frame->num_vsyncs < 1));
in->hasframe = true;
frame->frame_id = ++(in->current_frame_id);
in->frame_queued = frame;
in->wakeup_pts = frame->display_synced
? 0 : frame->pts + MPMAX(frame->duration, 0);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
wakeup_locked(vo);
pthread_mutex_unlock(&in->lock);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// If a frame is currently being rendered (or queued), wait until it's done.
// Otherwise, return immediately.
void vo_wait_frame(struct vo *vo)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
while (in->frame_queued || in->rendering)
pthread_cond_wait(&in->wakeup, &in->lock);
pthread_mutex_unlock(&in->lock);
}
// Wait until realtime is >= ts
// called without lock
static void wait_until(struct vo *vo, int64_t target)
{
struct vo_internal *in = vo->in;
struct timespec ts = mp_time_us_to_timespec(target);
pthread_mutex_lock(&in->lock);
while (target > mp_time_us()) {
if (in->queued_events & VO_EVENT_LIVE_RESIZING)
break;
if (pthread_cond_timedwait(&in->wakeup, &in->lock, &ts))
break;
}
pthread_mutex_unlock(&in->lock);
}
static bool render_frame(struct vo *vo)
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
{
struct vo_internal *in = vo->in;
struct vo_frame *frame = NULL;
bool got_frame = false;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
update_display_fps(vo);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
pthread_mutex_lock(&in->lock);
if (in->frame_queued) {
talloc_free(in->current_frame);
in->current_frame = in->frame_queued;
in->frame_queued = NULL;
} else if (in->paused || !in->current_frame || !in->hasframe ||
(in->current_frame->display_synced && in->current_frame->num_vsyncs < 1) ||
!in->current_frame->display_synced)
{
goto done;
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
frame = vo_frame_ref(in->current_frame);
assert(frame);
if (frame->display_synced) {
frame->pts = 0;
frame->duration = -1;
}
int64_t now = mp_time_us();
int64_t pts = frame->pts;
int64_t duration = frame->duration;
int64_t end_time = pts + duration;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
2015-01-23 10:30:43 +00:00
// Time at which we should flip_page on the VO.
int64_t target = frame->display_synced ? 0 : pts - in->flip_queue_offset;
2015-01-23 10:30:43 +00:00
// "normal" strict drop threshold.
in->dropped_frame = duration >= 0 && end_time < now;
in->dropped_frame &= !frame->display_synced;
in->dropped_frame &= !(vo->driver->caps & VO_CAP_FRAMEDROP);
in->dropped_frame &= frame->can_drop;
// Even if we're hopelessly behind, rather degrade to 10 FPS playback,
// instead of just freezing the display forever.
in->dropped_frame &= now - in->prev_vsync < 100 * 1000;
in->dropped_frame &= in->hasframe_rendered;
2015-01-23 10:30:43 +00:00
// Setup parameters for the next time this frame is drawn. ("frame" is the
// frame currently drawn, while in->current_frame is the potentially next.)
in->current_frame->repeat = true;
if (frame->display_synced) {
in->current_frame->vsync_offset += in->current_frame->vsync_interval;
in->dropped_frame |= in->current_frame->num_vsyncs < 1;
}
if (in->current_frame->num_vsyncs > 0)
in->current_frame->num_vsyncs -= 1;
// Always render when paused (it's typically the last frame for a while).
in->dropped_frame &= !in->paused;
bool use_vsync = in->current_frame->display_synced && !in->paused;
if (use_vsync && !in->expecting_vsync) // first DS frame in a row
in->prev_vsync = now;
in->expecting_vsync = use_vsync;
if (in->dropped_frame) {
in->drop_count += 1;
} else {
2015-02-07 20:14:43 +00:00
in->rendering = true;
in->hasframe_rendered = true;
int64_t prev_drop_count = vo->in->drop_count;
pthread_mutex_unlock(&in->lock);
wakeup_core(vo); // core can queue new video now
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
MP_STATS(vo, "start video-draw");
2014-08-22 14:21:58 +00:00
if (vo->driver->draw_frame) {
vo->driver->draw_frame(vo, frame);
} else {
vo->driver->draw_image(vo, mp_image_new_ref(frame->current));
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
MP_STATS(vo, "end video-draw");
wait_until(vo, target);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
MP_STATS(vo, "start video-flip");
vo->driver->flip_page(vo);
struct vo_vsync_info vsync = {
.last_queue_display_time = -1,
.skipped_vsyncs = -1,
};
if (vo->driver->get_vsync)
vo->driver->get_vsync(vo, &vsync);
// Make up some crap if presentation feedback is missing.
if (vsync.last_queue_display_time < 0)
vsync.last_queue_display_time = mp_time_us();
MP_STATS(vo, "end video-flip");
2014-08-22 14:21:58 +00:00
pthread_mutex_lock(&in->lock);
in->dropped_frame = prev_drop_count < vo->in->drop_count;
2015-02-07 20:14:43 +00:00
in->rendering = false;
update_vsync_timing_after_swap(vo, &vsync);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
if (vo->driver->caps & VO_CAP_NORETAIN) {
talloc_free(in->current_frame);
in->current_frame = NULL;
}
if (in->dropped_frame) {
MP_STATS(vo, "drop-vo");
} else {
2015-02-07 20:14:43 +00:00
in->request_redraw = false;
}
pthread_cond_broadcast(&in->wakeup); // for vo_wait_frame()
wakeup_core(vo);
got_frame = true;
2015-01-23 10:30:43 +00:00
done:
talloc_free(frame);
2015-01-23 10:30:43 +00:00
pthread_mutex_unlock(&in->lock);
return got_frame || (in->frame_queued && in->frame_queued->display_synced);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
static void do_redraw(struct vo *vo)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct vo_internal *in = vo->in;
if (!vo->config_ok || (vo->driver->caps & VO_CAP_NORETAIN))
return;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
pthread_mutex_lock(&in->lock);
in->request_redraw = false;
bool full_redraw = in->dropped_frame;
struct vo_frame *frame = NULL;
if (!vo->driver->untimed)
frame = vo_frame_ref(in->current_frame);
if (frame)
in->dropped_frame = false;
struct vo_frame dummy = {0};
if (!frame)
frame = &dummy;
frame->redraw = !full_redraw; // unconditionally redraw if it was dropped
frame->repeat = false;
frame->still = true;
frame->pts = 0;
frame->duration = -1;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
pthread_mutex_unlock(&in->lock);
if (vo->driver->draw_frame) {
vo->driver->draw_frame(vo, frame);
} else if ((full_redraw || vo->driver->control(vo, VOCTRL_REDRAW_FRAME, NULL) < 1)
&& frame->current)
{
vo->driver->draw_image(vo, mp_image_new_ref(frame->current));
}
vo->driver->flip_page(vo);
if (frame != &dummy)
talloc_free(frame);
}
static struct mp_image *get_image_vo(void *ctx, int imgfmt, int w, int h,
int stride_align)
{
struct vo *vo = ctx;
return vo->driver->get_image(vo, imgfmt, w, h, stride_align);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
static void *vo_thread(void *ptr)
{
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
struct vo *vo = ptr;
struct vo_internal *in = vo->in;
bool vo_paused = false;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
mpthread_set_name("vo");
client API: fix potential deadlock problems by throwing more shit at it The render API (vo_libmpv) had potential deadlock problems with MPV_RENDER_PARAM_ADVANCED_CONTROL. This required vd-lavc-dr to be enabled (the default). I never observed these deadlocks in the wild (doesn't mean they didn't happen), although I could specifically provoke them with some code changes. The problem was mostly about DR (direct rendering, letting the video decoder write to OpenGL buffer memory). Allocating/freeing a DR image needs to be done on the OpenGL thread, even though _lots_ of threads are involved with handling images. Freeing a DR image is a special case that can happen any time. dr_helper.c does most of the evil magic of achieving this. Unfortunately, there was a (sort of) circular lock dependency: freeing an image while certain internal locks are held would trigger the user's context update callback, which in turn would call mpv_render_context_update(), which processed all pending free requests, and then acquire an internal lock - which the caller might not release until a further DR image could be freed. "Solve" this by making freeing DR images asynchronous. This is slightly risky, but actually not much. The DR images will be free'd eventually. The biggest disadvantage is probably that debugging might get trickier. Any solution to this problem will probably add images to free to some sort of queue, and then process it later. I considered making this more explicit (so there'd be a point where the caller forcibly waits for all queued items to be free'd), but discarded these ideas as this probably would only increase complexity. Another consequence is that freeing DR images on the GL thread is not synchronous anymore. Instead, it mpv_render_context_update() will do it with a delay. This seems roundabout, but doesn't actually change anything, and avoids additional code. This also fixes that the render API required the render API user to remain on the same thread, even though this wasn't documented. As such, it was a bug. OpenGL essentially forces you to do all GL usage on a single thread, but in theory the API user could for example move the GL context to another thread. The API bump is because I think you can't make enough noise about this. Since we don't backport fixes to old versions, I'm specifically stating that old versions are broken, and I'm supplying workarounds. Internally, dr_helper_create() does not use pthread_self() anymore, thus the vo.c change. I think it's better to make binding to the current thread as explicit as possible. Of course it's not sure that this fixes all deadlocks (probably not).
2019-09-26 12:07:33 +00:00
if (vo->driver->get_image) {
in->dr_helper = dr_helper_create(in->dispatch, get_image_vo, vo);
client API: fix potential deadlock problems by throwing more shit at it The render API (vo_libmpv) had potential deadlock problems with MPV_RENDER_PARAM_ADVANCED_CONTROL. This required vd-lavc-dr to be enabled (the default). I never observed these deadlocks in the wild (doesn't mean they didn't happen), although I could specifically provoke them with some code changes. The problem was mostly about DR (direct rendering, letting the video decoder write to OpenGL buffer memory). Allocating/freeing a DR image needs to be done on the OpenGL thread, even though _lots_ of threads are involved with handling images. Freeing a DR image is a special case that can happen any time. dr_helper.c does most of the evil magic of achieving this. Unfortunately, there was a (sort of) circular lock dependency: freeing an image while certain internal locks are held would trigger the user's context update callback, which in turn would call mpv_render_context_update(), which processed all pending free requests, and then acquire an internal lock - which the caller might not release until a further DR image could be freed. "Solve" this by making freeing DR images asynchronous. This is slightly risky, but actually not much. The DR images will be free'd eventually. The biggest disadvantage is probably that debugging might get trickier. Any solution to this problem will probably add images to free to some sort of queue, and then process it later. I considered making this more explicit (so there'd be a point where the caller forcibly waits for all queued items to be free'd), but discarded these ideas as this probably would only increase complexity. Another consequence is that freeing DR images on the GL thread is not synchronous anymore. Instead, it mpv_render_context_update() will do it with a delay. This seems roundabout, but doesn't actually change anything, and avoids additional code. This also fixes that the render API required the render API user to remain on the same thread, even though this wasn't documented. As such, it was a bug. OpenGL essentially forces you to do all GL usage on a single thread, but in theory the API user could for example move the GL context to another thread. The API bump is because I think you can't make enough noise about this. Since we don't backport fixes to old versions, I'm specifically stating that old versions are broken, and I'm supplying workarounds. Internally, dr_helper_create() does not use pthread_self() anymore, thus the vo.c change. I think it's better to make binding to the current thread as explicit as possible. Of course it's not sure that this fixes all deadlocks (probably not).
2019-09-26 12:07:33 +00:00
dr_helper_acquire_thread(in->dr_helper);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
int r = vo->driver->preinit(vo) ? -1 : 0;
mp_rendezvous(vo, r); // init barrier
if (r < 0)
goto done;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
read_opts(vo);
update_display_fps(vo);
vo_event(vo, VO_EVENT_WIN_STATE);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
while (1) {
mp_dispatch_queue_process(vo->in->dispatch, 0);
if (in->terminate)
break;
vo->driver->control(vo, VOCTRL_CHECK_EVENTS, NULL);
bool working = render_frame(vo);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
int64_t now = mp_time_us();
int64_t wait_until = now + (working ? 0 : (int64_t)1e9);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
pthread_mutex_lock(&in->lock);
if (in->wakeup_pts) {
if (in->wakeup_pts > now) {
wait_until = MPMIN(wait_until, in->wakeup_pts);
} else {
in->wakeup_pts = 0;
wakeup_core(vo);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
}
}
if (vo->want_redraw && !in->want_redraw) {
vo->want_redraw = false;
in->want_redraw = true;
wakeup_core(vo);
}
bool redraw = in->request_redraw;
bool send_reset = in->send_reset;
in->send_reset = false;
bool send_pause = in->paused != vo_paused;
vo_paused = in->paused;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
pthread_mutex_unlock(&in->lock);
if (send_reset)
vo->driver->control(vo, VOCTRL_RESET, NULL);
if (send_pause)
vo->driver->control(vo, vo_paused ? VOCTRL_PAUSE : VOCTRL_RESUME, NULL);
if (wait_until > now && redraw) {
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
do_redraw(vo); // now is a good time
continue;
}
if (vo->want_redraw) // might have been set by VOCTRLs
wait_until = 0;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
wait_vo(vo, wait_until);
}
forget_frames(vo); // implicitly synchronized
talloc_free(in->current_frame);
in->current_frame = NULL;
vo->driver->uninit(vo);
done:
TA_FREEP(&in->dr_helper);
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
return NULL;
}
void vo_set_paused(struct vo *vo, bool paused)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
if (in->paused != paused) {
in->paused = paused;
if (in->paused && in->dropped_frame) {
in->request_redraw = true;
wakeup_core(vo);
}
reset_vsync_timings(vo);
wakeup_locked(vo);
}
pthread_mutex_unlock(&in->lock);
}
int64_t vo_get_drop_count(struct vo *vo)
{
pthread_mutex_lock(&vo->in->lock);
int64_t r = vo->in->drop_count;
pthread_mutex_unlock(&vo->in->lock);
return r;
}
void vo_increment_drop_count(struct vo *vo, int64_t n)
{
pthread_mutex_lock(&vo->in->lock);
vo->in->drop_count += n;
pthread_mutex_unlock(&vo->in->lock);
}
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// Make the VO redraw the OSD at some point in the future.
void vo_redraw(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
if (!in->request_redraw) {
in->request_redraw = true;
in->want_redraw = false;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
wakeup_locked(vo);
}
pthread_mutex_unlock(&in->lock);
}
bool vo_want_redraw(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
bool r = in->want_redraw;
pthread_mutex_unlock(&in->lock);
return r;
}
void vo_seek_reset(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
forget_frames(vo);
reset_vsync_timings(vo);
in->send_reset = true;
wakeup_locked(vo);
pthread_mutex_unlock(&in->lock);
}
// Return true if there is still a frame being displayed (or queued).
// If this returns true, a wakeup some time in the future is guaranteed.
bool vo_still_displaying(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&vo->in->lock);
int64_t now = mp_time_us();
int64_t frame_end = 0;
if (in->current_frame) {
frame_end = in->current_frame->pts + MPMAX(in->current_frame->duration, 0);
if (in->current_frame->display_synced)
frame_end = in->current_frame->num_vsyncs > 0 ? INT64_MAX : 0;
}
bool working = now < frame_end || in->rendering || in->frame_queued;
pthread_mutex_unlock(&vo->in->lock);
return working && in->hasframe;
}
// Whether at least 1 frame was queued or rendered since last seek or reconfig.
bool vo_has_frame(struct vo *vo)
{
return vo->in->hasframe;
}
static void run_query_format(void *p)
{
void **pp = p;
struct vo *vo = pp[0];
uint8_t *list = pp[1];
for (int format = IMGFMT_START; format < IMGFMT_END; format++)
list[format - IMGFMT_START] = vo->driver->query_format(vo, format);
}
// For each item in the list (allocated as uint8_t[IMGFMT_END - IMGFMT_START]),
// set the supported format flags.
void vo_query_formats(struct vo *vo, uint8_t *list)
{
void *p[] = {vo, list};
mp_dispatch_run(vo->in->dispatch, run_query_format, p);
}
// Calculate the appropriate source and destination rectangle to
// get a correctly scaled picture, including pan-scan.
// out_src: visible part of the video
// out_dst: area of screen covered by the video source rectangle
// out_osd: OSD size, OSD margins, etc.
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// Must be called from the VO thread only.
void vo_get_src_dst_rects(struct vo *vo, struct mp_rect *out_src,
struct mp_rect *out_dst, struct mp_osd_res *out_osd)
{
if (!vo->params) {
*out_src = *out_dst = (struct mp_rect){0};
*out_osd = (struct mp_osd_res){0};
return;
}
mp_get_src_dst_rects(vo->log, vo->opts, vo->driver->caps, vo->params,
vo->dwidth, vo->dheight, vo->monitor_par,
out_src, out_dst, out_osd);
}
// flip_page[_timed] will be called offset_us microseconds too early.
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
// (For vo_vdpau, which does its own timing.)
// num_req_frames set the requested number of requested vo_frame.frames.
// (For vo_gpu interpolation.)
void vo_set_queue_params(struct vo *vo, int64_t offset_us, int num_req_frames)
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
in->flip_queue_offset = offset_us;
in->req_frames = MPCLAMP(num_req_frames, 1, VO_MAX_REQ_FRAMES);
pthread_mutex_unlock(&in->lock);
}
int vo_get_num_req_frames(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
int res = in->req_frames;
video: move display and timing to a separate thread The VO is run inside its own thread. It also does most of video timing. The playloop hands the image data and a realtime timestamp to the VO, and the VO does the rest. In particular, this allows the playloop to do other things, instead of blocking for video redraw. But if anything accesses the VO during video timing, it will block. This also fixes vo_sdl.c event handling; but that is only a side-effect, since reimplementing the broken way would require more effort. Also drop --softsleep. In theory, this option helps if the kernel's sleeping mechanism is too inaccurate for video timing. In practice, I haven't ever encountered a situation where it helps, and it just burns CPU cycles. On the other hand it's probably actively harmful, because it prevents the libavcodec decoder threads from doing real work. Side note: Originally, I intended that multiple frames can be queued to the VO. But this is not done, due to problems with OSD and other certain features. OSD in particular is simply designed in a way that it can be neither timed nor copied, so you do have to render it into the video frame before you can draw the next frame. (Subtitles have no such restriction. sd_lavc was even updated to fix this.) It seems the right solution to queuing multiple VO frames is rendering on VO-backed framebuffers, like vo_vdpau.c does. This requires VO driver support, and is out of scope of this commit. As consequence, the VO has a queue size of 1. The existing video queue is just needed to compute frame duration, and will be moved out in the next commit.
2014-08-12 21:02:08 +00:00
pthread_mutex_unlock(&in->lock);
return res;
}
int64_t vo_get_vsync_interval(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
int64_t res = vo->in->vsync_interval > 1 ? vo->in->vsync_interval : -1;
pthread_mutex_unlock(&in->lock);
return res;
}
// Returns duration of a display refresh in seconds.
double vo_get_estimated_vsync_interval(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
double res = in->estimated_vsync_interval / 1e6;
pthread_mutex_unlock(&in->lock);
return res;
}
double vo_get_estimated_vsync_jitter(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
double res = in->estimated_vsync_jitter;
pthread_mutex_unlock(&in->lock);
return res;
}
// Get the time in seconds at after which the currently rendering frame will
// end. Returns positive values if the frame is yet to be finished, negative
// values if it already finished.
// This can only be called while no new frame is queued (after
// vo_is_ready_for_frame). Returns 0 for non-display synced frames, or if the
// deadline for continuous display was missed.
double vo_get_delay(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
assert (!in->frame_queued);
int64_t res = 0;
if (in->base_vsync && in->vsync_interval > 1 && in->current_frame) {
res = in->base_vsync;
int extra = !!in->rendering;
res += (in->current_frame->num_vsyncs + extra) * in->vsync_interval;
if (!in->current_frame->display_synced)
res = 0;
}
pthread_mutex_unlock(&in->lock);
return res ? (res - mp_time_us()) / 1e6 : 0;
}
void vo_discard_timing_info(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
reset_vsync_timings(vo);
pthread_mutex_unlock(&in->lock);
}
int64_t vo_get_delayed_count(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
int64_t res = vo->in->delayed_count;
pthread_mutex_unlock(&in->lock);
return res;
}
double vo_get_display_fps(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
double res = vo->in->display_fps;
pthread_mutex_unlock(&in->lock);
return res;
}
// Set specific event flags, and wakeup the playback core if needed.
// vo_query_and_reset_events() can retrieve the events again.
void vo_event(struct vo *vo, int event)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
if ((in->queued_events & event & VO_EVENTS_USER) != (event & VO_EVENTS_USER))
wakeup_core(vo);
if (event)
wakeup_locked(vo);
in->queued_events |= event;
in->internal_events |= event;
pthread_mutex_unlock(&in->lock);
}
// Check event flags set with vo_event(). Return the mask of events that was
// set and included in the events parameter. Clear the returned events.
int vo_query_and_reset_events(struct vo *vo, int events)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
int r = in->queued_events & events;
in->queued_events &= ~(unsigned)r;
pthread_mutex_unlock(&in->lock);
return r;
}
struct mp_image *vo_get_current_frame(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
struct mp_image *r = NULL;
if (vo->in->current_frame)
r = mp_image_new_ref(vo->in->current_frame->current);
pthread_mutex_unlock(&in->lock);
return r;
}
struct vo_frame *vo_get_current_vo_frame(struct vo *vo)
{
struct vo_internal *in = vo->in;
pthread_mutex_lock(&in->lock);
struct vo_frame *r = vo_frame_ref(vo->in->current_frame);
pthread_mutex_unlock(&in->lock);
return r;
}
struct mp_image *vo_get_image(struct vo *vo, int imgfmt, int w, int h,
int stride_align)
{
if (vo->driver->get_image_ts)
return vo->driver->get_image_ts(vo, imgfmt, w, h, stride_align);
if (vo->in->dr_helper)
return dr_helper_get_image(vo->in->dr_helper, imgfmt, w, h, stride_align);
return NULL;
}
static void destroy_frame(void *p)
{
struct vo_frame *frame = p;
for (int n = 0; n < frame->num_frames; n++)
talloc_free(frame->frames[n]);
}
// Return a new reference to the given frame. The image pointers are also new
// references. Calling talloc_free() on the frame unrefs all currently set
// image references. (Assuming current==frames[0].)
struct vo_frame *vo_frame_ref(struct vo_frame *frame)
{
if (!frame)
return NULL;
struct vo_frame *new = talloc_ptrtype(NULL, new);
talloc_set_destructor(new, destroy_frame);
*new = *frame;
for (int n = 0; n < frame->num_frames; n++) {
new->frames[n] = mp_image_new_ref(frame->frames[n]);
if (!new->frames[n])
abort(); // OOM on tiny allocs
}
new->current = new->num_frames ? new->frames[0] : NULL;
return new;
}
/*
* lookup an integer in a table, table must have 0 as the last key
* param: key key to search for
* returns translation corresponding to key or "to" value of last mapping
* if not found.
*/
int lookup_keymap_table(const struct mp_keymap *map, int key)
{
while (map->from && map->from != key)
map++;
return map->to;
}