audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
/*
|
|
|
|
* This file is part of mpv.
|
|
|
|
*
|
|
|
|
* 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
|
|
|
|
* GNU Lesser General Public License for more details.
|
|
|
|
*
|
|
|
|
* 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 <stddef.h>
|
|
|
|
#include <inttypes.h>
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
#include <math.h>
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
#include <errno.h>
|
|
|
|
#include <assert.h>
|
|
|
|
|
|
|
|
#include "ao.h"
|
|
|
|
#include "internal.h"
|
|
|
|
#include "audio/aframe.h"
|
|
|
|
#include "audio/format.h"
|
|
|
|
|
|
|
|
#include "common/msg.h"
|
|
|
|
#include "common/common.h"
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
#include "filters/f_async_queue.h"
|
|
|
|
#include "filters/filter_internal.h"
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
|
|
|
#include "osdep/timer.h"
|
|
|
|
#include "osdep/threads.h"
|
|
|
|
|
|
|
|
struct buffer_state {
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
// Buffer and AO
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex lock;
|
|
|
|
mp_cond wakeup;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2024-04-10 05:12:40 +00:00
|
|
|
// AO thread sleep
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex pt_lock;
|
|
|
|
mp_cond pt_wakeup;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
// Access from AO driver's thread only.
|
|
|
|
char *convert_buffer;
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
// Immutable.
|
|
|
|
struct mp_async_queue *queue;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
// --- protected by lock
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
struct mp_filter *filter_root;
|
|
|
|
struct mp_filter *input; // connected to queue
|
|
|
|
struct mp_aframe *pending; // last, not fully consumed output
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
|
|
|
bool streaming; // AO streaming active
|
|
|
|
bool playing; // logically playing audio from buffer
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
bool paused; // logically paused
|
2024-05-09 00:00:12 +00:00
|
|
|
bool hw_paused; // driver->set_pause() was used successfully
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-11 18:45:40 +00:00
|
|
|
int64_t end_time_ns; // absolute output time of last played sample
|
2024-03-25 20:08:13 +00:00
|
|
|
int64_t queued_time_ns; // duration of samples that have been queued to
|
|
|
|
// the device but have not been played.
|
|
|
|
// This field is only set in ao_set_paused(),
|
|
|
|
// and is considered as a temporary solution;
|
|
|
|
// DO NOT USE IT IN OTHER PLACES.
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
|
|
|
bool initial_unblocked;
|
|
|
|
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
// "Push" AOs only (AOs with driver->write).
|
|
|
|
bool recover_pause; // non-hw_paused: needs to recover delay
|
|
|
|
struct mp_pcm_state prepause_state;
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_thread thread; // thread shoveling data to AO
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
bool thread_valid; // thread is running
|
|
|
|
struct mp_aframe *temp_buf;
|
|
|
|
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
// --- protected by pt_lock
|
|
|
|
bool need_wakeup;
|
|
|
|
bool terminate; // exit thread
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
};
|
|
|
|
|
2024-04-10 05:12:40 +00:00
|
|
|
static MP_THREAD_VOID ao_thread(void *arg);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2024-04-10 05:12:40 +00:00
|
|
|
void ao_wakeup(struct ao *ao)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->pt_lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->need_wakeup = true;
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_cond_broadcast(&p->pt_wakeup);
|
|
|
|
mp_mutex_unlock(&p->pt_lock);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// called locked
|
|
|
|
static void get_dev_state(struct ao *ao, struct mp_pcm_state *state)
|
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
|
audio: fix stream-silence with push AOs (somewhat)
--audio-stream-silence is a shitty feature compensating for awful
consumer garbage, that mutes PCM at first to check whether it's
compressed audio, using formats advocated and owned by malicious patent
troll companies (who spend more money on their lawyers than paying any
technicians), wrapped in a wasteful way to make it constant bitrate
using a standard whose text is not freely available, and only rude users
want it. This feature has been carelessly broken, because it's
complicated and stupid. What would Jesus do? If not getting an aneurysm,
or pushing over tables with expensive A/V receivers on top of them, he'd
probably fix the feature. So let's take inspiration from Jesus Christ
himself, and do something as dumb as wasting some of our limited
lifetime on this incredibly stupid fucking shit.
This is tricky, because state changes like end-of-audio are supposed to
be driven by the AO driver, while playing silence precludes this. But it
seems code paths for "untimed" AOs can be reused.
But there are still problems. For example, underruns will just happen
normally (and stop audio streaming), because we don't have a separate
heuristic to check whether the buffer is "low enough" (as a consequence
of a network stall, but before the audio output itself underruns).
2020-09-03 20:39:23 +00:00
|
|
|
if (p->paused && p->playing && !ao->stream_silence) {
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
*state = p->prepause_state;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
*state = (struct mp_pcm_state){
|
|
|
|
.free_samples = -1,
|
|
|
|
.queued_samples = -1,
|
|
|
|
.delay = -1,
|
|
|
|
};
|
|
|
|
ao->driver->get_state(ao, state);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
struct mp_async_queue *ao_get_queue(struct ao *ao)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
return p->queue;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
// Special behavior with data==NULL: caller uses p->pending.
|
2023-11-05 08:08:47 +00:00
|
|
|
static int read_buffer(struct ao *ao, void **data, int samples, bool *eof,
|
|
|
|
bool pad_silence)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
int pos = 0;
|
|
|
|
*eof = false;
|
|
|
|
|
|
|
|
while (p->playing && !p->paused && pos < samples) {
|
|
|
|
if (!p->pending || !mp_aframe_get_size(p->pending)) {
|
|
|
|
TA_FREEP(&p->pending);
|
|
|
|
struct mp_frame frame = mp_pin_out_read(p->input->pins[0]);
|
|
|
|
if (!frame.type)
|
|
|
|
break; // we can't/don't want to block
|
|
|
|
if (frame.type != MP_FRAME_AUDIO) {
|
|
|
|
if (frame.type == MP_FRAME_EOF)
|
|
|
|
*eof = true;
|
|
|
|
mp_frame_unref(&frame);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
p->pending = frame.data;
|
|
|
|
}
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (!data)
|
|
|
|
break;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
int copy = mp_aframe_get_size(p->pending);
|
|
|
|
uint8_t **fdata = mp_aframe_get_data_ro(p->pending);
|
|
|
|
copy = MPMIN(copy, samples - pos);
|
|
|
|
for (int n = 0; n < ao->num_planes; n++) {
|
|
|
|
memcpy((char *)data[n] + pos * ao->sstride,
|
|
|
|
fdata[n], copy * ao->sstride);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
mp_aframe_skip_samples(p->pending, copy);
|
|
|
|
pos += copy;
|
|
|
|
*eof = false;
|
|
|
|
}
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (!data) {
|
|
|
|
if (!p->pending)
|
|
|
|
return 0;
|
|
|
|
void **pd = (void *)mp_aframe_get_data_rw(p->pending);
|
|
|
|
if (pd)
|
|
|
|
ao_post_process_data(ao, pd, mp_aframe_get_size(p->pending));
|
|
|
|
return 1;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
// pad with silence (underflow/paused/eof)
|
2023-11-05 08:08:47 +00:00
|
|
|
if (pad_silence) {
|
|
|
|
for (int n = 0; n < ao->num_planes; n++) {
|
|
|
|
af_fill_silence((char *)data[n] + pos * ao->sstride,
|
|
|
|
(samples - pos) * ao->sstride,
|
|
|
|
ao->format);
|
|
|
|
}
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
}
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
ao_post_process_data(ao, data, pos);
|
|
|
|
return pos;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
2024-02-05 07:47:05 +00:00
|
|
|
static int ao_read_data_locked(struct ao *ao, void **data, int samples,
|
2024-04-22 11:47:34 +00:00
|
|
|
int64_t out_time_ns, bool *eof, bool pad_silence)
|
2023-11-05 07:58:15 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
assert(!ao->driver->write);
|
|
|
|
|
2024-04-22 11:47:34 +00:00
|
|
|
int pos = read_buffer(ao, data, samples, eof, pad_silence);
|
2023-11-05 07:58:15 +00:00
|
|
|
|
|
|
|
if (pos > 0)
|
|
|
|
p->end_time_ns = out_time_ns;
|
|
|
|
|
|
|
|
if (pos < samples && p->playing && !p->paused) {
|
|
|
|
p->playing = false;
|
|
|
|
ao->wakeup_cb(ao->wakeup_ctx);
|
|
|
|
// For ao_drain().
|
|
|
|
mp_cond_broadcast(&p->wakeup);
|
|
|
|
}
|
|
|
|
|
|
|
|
return pos;
|
|
|
|
}
|
|
|
|
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
// Read the given amount of samples in the user-provided data buffer. Returns
|
|
|
|
// the number of samples copied. If there is not enough data (buffer underrun
|
|
|
|
// or EOF), return the number of samples that could be copied, and fill the
|
|
|
|
// rest of the user-provided buffer with silence.
|
|
|
|
// This basically assumes that the audio device doesn't care about underruns.
|
|
|
|
// If this is called in paused mode, it will always return 0.
|
2023-10-11 18:45:40 +00:00
|
|
|
// The caller should set out_time_ns to the expected delay until the last sample
|
|
|
|
// reaches the speakers, in nanoseconds, using mp_time_ns() as reference.
|
2024-04-22 11:47:34 +00:00
|
|
|
int ao_read_data(struct ao *ao, void **data, int samples, int64_t out_time_ns, bool *eof, bool pad_silence, bool blocking)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
|
2024-04-22 11:47:34 +00:00
|
|
|
if (blocking) {
|
|
|
|
mp_mutex_lock(&p->lock);
|
|
|
|
} else if (mp_mutex_trylock(&p->lock)) {
|
|
|
|
return 0;
|
|
|
|
}
|
2023-11-05 07:58:15 +00:00
|
|
|
|
2024-04-22 11:47:34 +00:00
|
|
|
bool eof_buf;
|
|
|
|
if (eof == NULL) {
|
|
|
|
// This is a public API. We want to reduce the cognitive burden of the caller.
|
|
|
|
eof = &eof_buf;
|
|
|
|
}
|
2023-11-05 07:58:15 +00:00
|
|
|
|
2024-04-22 11:47:34 +00:00
|
|
|
int pos = ao_read_data_locked(ao, data, samples, out_time_ns, eof, pad_silence);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
return pos;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// Same as ao_read_data(), but convert data according to *fmt.
|
|
|
|
// fmt->src_fmt and fmt->channels must be the same as the AO parameters.
|
|
|
|
int ao_read_data_converted(struct ao *ao, struct ao_convert_fmt *fmt,
|
2023-10-11 18:45:40 +00:00
|
|
|
void **data, int samples, int64_t out_time_ns)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
void *ndata[MP_NUM_CHANNELS] = {0};
|
|
|
|
|
|
|
|
if (!ao_need_conversion(fmt))
|
2024-04-22 11:47:34 +00:00
|
|
|
return ao_read_data(ao, data, samples, out_time_ns, NULL, true, true);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
|
|
|
assert(ao->format == fmt->src_fmt);
|
|
|
|
assert(ao->channels.num == fmt->channels);
|
|
|
|
|
|
|
|
bool planar = af_fmt_is_planar(fmt->src_fmt);
|
|
|
|
int planes = planar ? fmt->channels : 1;
|
|
|
|
int plane_samples = samples * (planar ? 1: fmt->channels);
|
|
|
|
int src_plane_size = plane_samples * af_fmt_to_bytes(fmt->src_fmt);
|
|
|
|
int dst_plane_size = plane_samples * fmt->dst_bits / 8;
|
|
|
|
|
|
|
|
int needed = src_plane_size * planes;
|
|
|
|
if (needed > talloc_get_size(p->convert_buffer) || !p->convert_buffer) {
|
|
|
|
talloc_free(p->convert_buffer);
|
|
|
|
p->convert_buffer = talloc_size(NULL, needed);
|
|
|
|
}
|
|
|
|
|
|
|
|
for (int n = 0; n < planes; n++)
|
|
|
|
ndata[n] = p->convert_buffer + n * src_plane_size;
|
|
|
|
|
2024-04-22 11:47:34 +00:00
|
|
|
int res = ao_read_data(ao, ndata, samples, out_time_ns, NULL, true, true);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
|
|
|
ao_convert_inplace(fmt, ndata, samples);
|
|
|
|
for (int n = 0; n < planes; n++)
|
|
|
|
memcpy(data[n], ndata[n], dst_plane_size);
|
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
2024-04-24 00:30:07 +00:00
|
|
|
// Called by pull-based AO to indicate the AO has stopped requesting more data,
|
|
|
|
// usually when EOF is got from ao_read_data().
|
|
|
|
// After this function is called, the core will call ao->driver->start() again
|
|
|
|
// when more audio data after EOF arrives.
|
|
|
|
void ao_stop_streaming(struct ao *ao)
|
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
p->streaming = false;
|
|
|
|
}
|
|
|
|
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
int ao_control(struct ao *ao, enum aocontrol cmd, void *arg)
|
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
int r = CONTROL_UNKNOWN;
|
|
|
|
if (ao->driver->control) {
|
2020-06-17 08:55:18 +00:00
|
|
|
// Only need to lock in push mode.
|
|
|
|
if (ao->driver->write)
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
2020-06-17 08:55:18 +00:00
|
|
|
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
r = ao->driver->control(ao, cmd, arg);
|
2020-06-17 08:55:18 +00:00
|
|
|
|
|
|
|
if (ao->driver->write)
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
double ao_get_delay(struct ao *ao)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
|
|
|
|
double driver_delay;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
if (ao->driver->write) {
|
|
|
|
struct mp_pcm_state state;
|
|
|
|
get_dev_state(ao, &state);
|
|
|
|
driver_delay = state.delay;
|
|
|
|
} else {
|
2023-10-11 18:45:40 +00:00
|
|
|
int64_t end = p->end_time_ns;
|
|
|
|
int64_t now = mp_time_ns();
|
|
|
|
driver_delay = MPMAX(0, MP_TIME_NS_TO_S(end - now));
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
2024-07-09 05:23:23 +00:00
|
|
|
int64_t pending = mp_async_queue_get_samples(p->queue);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (p->pending)
|
|
|
|
pending += mp_aframe_get_size(p->pending);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
return driver_delay + pending / (double)ao->samplerate;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
// Fully stop playback; clear buffers, including queue.
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
void ao_reset(struct ao *ao)
|
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
bool wakeup = false;
|
2020-06-02 18:43:32 +00:00
|
|
|
bool do_reset = false;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
TA_FREEP(&p->pending);
|
|
|
|
mp_async_queue_reset(p->queue);
|
|
|
|
mp_filter_reset(p->filter_root);
|
|
|
|
mp_async_queue_resume_reading(p->queue);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
|
|
|
if (!ao->stream_silence && ao->driver->reset) {
|
2020-06-02 18:43:32 +00:00
|
|
|
if (ao->driver->write) {
|
|
|
|
ao->driver->reset(ao);
|
|
|
|
} else {
|
|
|
|
// Pull AOs may wait for ao_read_data() to return.
|
|
|
|
// That would deadlock if called from within the lock.
|
|
|
|
do_reset = true;
|
|
|
|
}
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->streaming = false;
|
|
|
|
}
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
wakeup = p->playing;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->playing = false;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
p->recover_pause = false;
|
|
|
|
p->hw_paused = false;
|
2023-10-11 18:45:40 +00:00
|
|
|
p->end_time_ns = 0;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
2020-06-02 18:43:32 +00:00
|
|
|
if (do_reset)
|
|
|
|
ao->driver->reset(ao);
|
|
|
|
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
if (wakeup)
|
2024-04-10 05:12:40 +00:00
|
|
|
ao_wakeup(ao);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
// Initiate playback. This moves from the stop/underrun state to actually
|
|
|
|
// playing (orthogonally taking the paused state into account). Plays all
|
|
|
|
// data in the queue, and goes into underrun state if no more data available.
|
|
|
|
// No-op if already running.
|
|
|
|
void ao_start(struct ao *ao)
|
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
2020-09-20 12:45:45 +00:00
|
|
|
bool do_start = false;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
|
|
|
|
p->playing = true;
|
|
|
|
|
2020-09-20 14:30:31 +00:00
|
|
|
if (!ao->driver->write && !p->paused && !p->streaming) {
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
p->streaming = true;
|
2020-09-20 12:45:45 +00:00
|
|
|
do_start = true;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
}
|
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
|
2020-09-20 12:45:45 +00:00
|
|
|
// Pull AOs might call ao_read_data() so do this outside the lock.
|
|
|
|
if (do_start)
|
|
|
|
ao->driver->start(ao);
|
|
|
|
|
2024-04-10 05:12:40 +00:00
|
|
|
ao_wakeup(ao);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
}
|
|
|
|
|
2023-08-10 03:50:05 +00:00
|
|
|
void ao_set_paused(struct ao *ao, bool paused, bool eof)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
bool wakeup = false;
|
2024-03-25 20:08:13 +00:00
|
|
|
bool do_change_state = false;
|
2024-05-09 00:00:12 +00:00
|
|
|
bool is_hw_paused;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-08-10 03:50:05 +00:00
|
|
|
// If we are going to pause on eof and ao is still playing,
|
|
|
|
// be sure to drain the ao first for gapless.
|
|
|
|
if (eof && paused && ao_is_playing(ao))
|
|
|
|
ao_drain(ao);
|
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2022-11-26 04:20:58 +00:00
|
|
|
if ((p->playing || !ao->driver->write) && !p->paused && paused) {
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
if (p->streaming && !ao->stream_silence) {
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
if (ao->driver->write) {
|
|
|
|
if (!p->recover_pause)
|
|
|
|
get_dev_state(ao, &p->prepause_state);
|
|
|
|
if (ao->driver->set_pause && ao->driver->set_pause(ao, true)) {
|
|
|
|
p->hw_paused = true;
|
|
|
|
} else {
|
|
|
|
ao->driver->reset(ao);
|
|
|
|
p->streaming = false;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
p->recover_pause = !ao->untimed;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
}
|
2024-03-25 20:08:13 +00:00
|
|
|
} else if (ao->driver->reset || ao->driver->set_pause) {
|
2020-06-02 18:43:32 +00:00
|
|
|
// See ao_reset() why this is done outside of the lock.
|
2024-03-25 20:08:13 +00:00
|
|
|
do_change_state = true;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->streaming = false;
|
2024-05-09 00:00:12 +00:00
|
|
|
is_hw_paused = p->hw_paused = !!ao->driver->set_pause;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
}
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
wakeup = true;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
} else if (p->playing && p->paused && !paused) {
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
if (ao->driver->write) {
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (p->hw_paused)
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
ao->driver->set_pause(ao, false);
|
|
|
|
p->hw_paused = false;
|
|
|
|
} else {
|
|
|
|
if (!p->streaming)
|
2024-03-25 20:08:13 +00:00
|
|
|
do_change_state = true;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
p->streaming = true;
|
2024-05-09 00:00:12 +00:00
|
|
|
is_hw_paused = p->hw_paused;
|
|
|
|
p->hw_paused = false;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
}
|
|
|
|
wakeup = true;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
p->paused = paused;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
2024-03-25 20:08:13 +00:00
|
|
|
if (do_change_state) {
|
2024-05-09 00:00:12 +00:00
|
|
|
if (is_hw_paused) {
|
2024-03-25 20:08:13 +00:00
|
|
|
if (paused) {
|
|
|
|
ao->driver->set_pause(ao, true);
|
|
|
|
p->queued_time_ns = p->end_time_ns - mp_time_ns();
|
|
|
|
} else {
|
|
|
|
p->end_time_ns = p->queued_time_ns + mp_time_ns();
|
|
|
|
ao->driver->set_pause(ao, false);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (paused)
|
|
|
|
ao->driver->reset(ao);
|
|
|
|
else
|
|
|
|
ao->driver->start(ao);
|
|
|
|
}
|
|
|
|
}
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
if (wakeup)
|
2024-04-10 05:12:40 +00:00
|
|
|
ao_wakeup(ao);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
// Whether audio is playing. This means that there is still data in the buffers,
|
|
|
|
// and ao_start() was called. This returns true even if playback was logically
|
|
|
|
// paused. On false, EOF was reached, or an underrun happened, or ao_reset()
|
|
|
|
// was called.
|
|
|
|
bool ao_is_playing(struct ao *ao)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
bool playing = p->playing;
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
return playing;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// Block until the current audio buffer has played completely.
|
|
|
|
void ao_drain(struct ao *ao)
|
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
while (!p->paused && p->playing) {
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
double delay = ao_get_delay(ao);
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
2024-07-09 05:24:00 +00:00
|
|
|
// Wait for buffer + arbitrary ~250ms for EOF signal from AO.
|
2023-10-21 02:55:41 +00:00
|
|
|
if (mp_cond_timedwait(&p->wakeup, &p->lock,
|
|
|
|
MP_TIME_S_TO_NS(MPMAX(delay, 0) + 0.25)))
|
|
|
|
{
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
MP_VERBOSE(ao, "drain timeout\n");
|
|
|
|
break;
|
|
|
|
}
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (!p->playing && mp_async_queue_get_samples(p->queue)) {
|
|
|
|
MP_WARN(ao, "underrun during draining\n");
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
ao_start(ao);
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
}
|
2020-05-27 19:04:32 +00:00
|
|
|
}
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
|
|
|
ao_reset(ao);
|
|
|
|
}
|
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
static void wakeup_filters(void *ctx)
|
|
|
|
{
|
|
|
|
struct ao *ao = ctx;
|
2024-04-10 05:12:40 +00:00
|
|
|
ao_wakeup(ao);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
}
|
|
|
|
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
void ao_uninit(struct ao *ao)
|
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
|
2023-01-12 22:43:44 +00:00
|
|
|
if (p && p->thread_valid) {
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->pt_lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->terminate = true;
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_cond_broadcast(&p->pt_wakeup);
|
|
|
|
mp_mutex_unlock(&p->pt_lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_thread_join(p->thread);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->thread_valid = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ao->driver_initialized)
|
|
|
|
ao->driver->uninit(ao);
|
|
|
|
|
2023-01-12 22:43:44 +00:00
|
|
|
if (p) {
|
|
|
|
talloc_free(p->filter_root);
|
|
|
|
talloc_free(p->queue);
|
|
|
|
talloc_free(p->pending);
|
|
|
|
talloc_free(p->convert_buffer);
|
|
|
|
talloc_free(p->temp_buf);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_cond_destroy(&p->wakeup);
|
|
|
|
mp_mutex_destroy(&p->lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_cond_destroy(&p->pt_wakeup);
|
|
|
|
mp_mutex_destroy(&p->pt_lock);
|
2023-01-12 22:43:44 +00:00
|
|
|
}
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
talloc_free(ao);
|
|
|
|
}
|
|
|
|
|
|
|
|
void init_buffer_pre(struct ao *ao)
|
|
|
|
{
|
|
|
|
ao->buffer_state = talloc_zero(ao, struct buffer_state);
|
|
|
|
}
|
|
|
|
|
|
|
|
bool init_buffer_post(struct ao *ao)
|
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
assert(ao->driver->start);
|
|
|
|
if (ao->driver->write) {
|
|
|
|
assert(ao->driver->reset);
|
|
|
|
assert(ao->driver->get_state);
|
|
|
|
}
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_init(&p->lock);
|
|
|
|
mp_cond_init(&p->wakeup);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_init(&p->pt_lock);
|
|
|
|
mp_cond_init(&p->pt_wakeup);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
p->queue = mp_async_queue_create();
|
|
|
|
p->filter_root = mp_filter_create_root(ao->global);
|
|
|
|
p->input = mp_async_queue_create_filter(p->filter_root, MP_PIN_OUT, p->queue);
|
|
|
|
|
|
|
|
mp_async_queue_resume_reading(p->queue);
|
|
|
|
|
|
|
|
struct mp_async_queue_config cfg = {
|
|
|
|
.sample_unit = AQUEUE_UNIT_SAMPLES,
|
|
|
|
.max_samples = ao->buffer,
|
|
|
|
.max_bytes = INT64_MAX,
|
|
|
|
};
|
|
|
|
mp_async_queue_set_config(p->queue, cfg);
|
|
|
|
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
if (ao->driver->write) {
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
mp_filter_graph_set_wakeup_cb(p->filter_root, wakeup_filters, ao);
|
|
|
|
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->thread_valid = true;
|
2024-04-10 05:12:40 +00:00
|
|
|
if (mp_thread_create(&p->thread, ao_thread, ao)) {
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->thread_valid = false;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (ao->stream_silence) {
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
ao->driver->start(ao);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->streaming = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
audio: fix stream-silence with push AOs (somewhat)
--audio-stream-silence is a shitty feature compensating for awful
consumer garbage, that mutes PCM at first to check whether it's
compressed audio, using formats advocated and owned by malicious patent
troll companies (who spend more money on their lawyers than paying any
technicians), wrapped in a wasteful way to make it constant bitrate
using a standard whose text is not freely available, and only rude users
want it. This feature has been carelessly broken, because it's
complicated and stupid. What would Jesus do? If not getting an aneurysm,
or pushing over tables with expensive A/V receivers on top of them, he'd
probably fix the feature. So let's take inspiration from Jesus Christ
himself, and do something as dumb as wasting some of our limited
lifetime on this incredibly stupid fucking shit.
This is tricky, because state changes like end-of-audio are supposed to
be driven by the AO driver, while playing silence precludes this. But it
seems code paths for "untimed" AOs can be reused.
But there are still problems. For example, underruns will just happen
normally (and stop audio streaming), because we don't have a separate
heuristic to check whether the buffer is "low enough" (as a consequence
of a network stall, but before the audio output itself underruns).
2020-09-03 20:39:23 +00:00
|
|
|
if (ao->stream_silence) {
|
|
|
|
MP_WARN(ao, "The --audio-stream-silence option is set. This will break "
|
|
|
|
"certain player behavior.\n");
|
|
|
|
}
|
|
|
|
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool realloc_buf(struct ao *ao, int samples)
|
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
|
|
|
|
|
|
|
samples = MPMAX(1, samples);
|
|
|
|
|
|
|
|
if (!p->temp_buf || samples > mp_aframe_get_size(p->temp_buf)) {
|
|
|
|
TA_FREEP(&p->temp_buf);
|
|
|
|
p->temp_buf = mp_aframe_create();
|
|
|
|
if (!mp_aframe_set_format(p->temp_buf, ao->format) ||
|
|
|
|
!mp_aframe_set_chmap(p->temp_buf, &ao->channels) ||
|
|
|
|
!mp_aframe_set_rate(p->temp_buf, ao->samplerate) ||
|
|
|
|
!mp_aframe_alloc_data(p->temp_buf, samples))
|
|
|
|
{
|
|
|
|
TA_FREEP(&p->temp_buf);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
// called locked
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
static bool ao_play_data(struct ao *ao)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
|
audio: fix stream-silence with push AOs (somewhat)
--audio-stream-silence is a shitty feature compensating for awful
consumer garbage, that mutes PCM at first to check whether it's
compressed audio, using formats advocated and owned by malicious patent
troll companies (who spend more money on their lawyers than paying any
technicians), wrapped in a wasteful way to make it constant bitrate
using a standard whose text is not freely available, and only rude users
want it. This feature has been carelessly broken, because it's
complicated and stupid. What would Jesus do? If not getting an aneurysm,
or pushing over tables with expensive A/V receivers on top of them, he'd
probably fix the feature. So let's take inspiration from Jesus Christ
himself, and do something as dumb as wasting some of our limited
lifetime on this incredibly stupid fucking shit.
This is tricky, because state changes like end-of-audio are supposed to
be driven by the AO driver, while playing silence precludes this. But it
seems code paths for "untimed" AOs can be reused.
But there are still problems. For example, underruns will just happen
normally (and stop audio streaming), because we don't have a separate
heuristic to check whether the buffer is "low enough" (as a consequence
of a network stall, but before the audio output itself underruns).
2020-09-03 20:39:23 +00:00
|
|
|
if ((!p->playing || p->paused) && !ao->stream_silence)
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
return false;
|
|
|
|
|
2020-06-02 18:30:59 +00:00
|
|
|
struct mp_pcm_state state;
|
|
|
|
get_dev_state(ao, &state);
|
|
|
|
|
2024-05-08 03:27:46 +00:00
|
|
|
if (p->streaming && !state.playing && !ao->untimed)
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
goto eof;
|
|
|
|
|
|
|
|
void **planes = NULL;
|
|
|
|
int space = state.free_samples;
|
|
|
|
if (!space)
|
|
|
|
return false;
|
|
|
|
assert(space >= 0);
|
|
|
|
|
|
|
|
int samples = 0;
|
|
|
|
bool got_eof = false;
|
|
|
|
if (ao->driver->write_frames) {
|
|
|
|
TA_FREEP(&p->pending);
|
2023-11-05 08:08:47 +00:00
|
|
|
samples = read_buffer(ao, NULL, 1, &got_eof, false);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
planes = (void **)&p->pending;
|
|
|
|
} else {
|
|
|
|
if (!realloc_buf(ao, space)) {
|
|
|
|
MP_ERR(ao, "Failed to allocate buffer.\n");
|
|
|
|
return false;
|
2020-06-02 18:30:59 +00:00
|
|
|
}
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
planes = (void **)mp_aframe_get_data_rw(p->temp_buf);
|
|
|
|
assert(planes);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (p->recover_pause) {
|
|
|
|
samples = MPCLAMP(p->prepause_state.delay * ao->samplerate, 0, space);
|
|
|
|
p->recover_pause = false;
|
|
|
|
mp_aframe_set_silence(p->temp_buf, 0, space);
|
|
|
|
}
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (!samples) {
|
2023-11-05 08:08:47 +00:00
|
|
|
samples = read_buffer(ao, planes, space, &got_eof, true);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (p->paused || (ao->stream_silence && !p->playing))
|
|
|
|
samples = space; // read_buffer() sets remainder to silent
|
|
|
|
}
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
if (samples) {
|
|
|
|
MP_STATS(ao, "start ao fill");
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (!ao->driver->write(ao, planes, samples))
|
|
|
|
MP_ERR(ao, "Error writing audio to device.\n");
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
MP_STATS(ao, "end ao fill");
|
|
|
|
|
|
|
|
if (!p->streaming) {
|
|
|
|
MP_VERBOSE(ao, "starting AO\n");
|
|
|
|
ao->driver->start(ao);
|
|
|
|
p->streaming = true;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
state.playing = true;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
}
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
MP_TRACE(ao, "in=%d space=%d(%d) pl=%d, eof=%d\n",
|
|
|
|
samples, space, state.free_samples, p->playing, got_eof);
|
|
|
|
|
|
|
|
if (got_eof)
|
|
|
|
goto eof;
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
2020-08-29 14:27:56 +00:00
|
|
|
return samples > 0 && (samples < space || ao->untimed);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
|
|
|
|
eof:
|
|
|
|
MP_VERBOSE(ao, "audio end or underrun\n");
|
|
|
|
// Normal AOs signal EOF on underrun, untimed AOs never signal underruns.
|
audio: fix stream-silence with push AOs (somewhat)
--audio-stream-silence is a shitty feature compensating for awful
consumer garbage, that mutes PCM at first to check whether it's
compressed audio, using formats advocated and owned by malicious patent
troll companies (who spend more money on their lawyers than paying any
technicians), wrapped in a wasteful way to make it constant bitrate
using a standard whose text is not freely available, and only rude users
want it. This feature has been carelessly broken, because it's
complicated and stupid. What would Jesus do? If not getting an aneurysm,
or pushing over tables with expensive A/V receivers on top of them, he'd
probably fix the feature. So let's take inspiration from Jesus Christ
himself, and do something as dumb as wasting some of our limited
lifetime on this incredibly stupid fucking shit.
This is tricky, because state changes like end-of-audio are supposed to
be driven by the AO driver, while playing silence precludes this. But it
seems code paths for "untimed" AOs can be reused.
But there are still problems. For example, underruns will just happen
normally (and stop audio streaming), because we don't have a separate
heuristic to check whether the buffer is "low enough" (as a consequence
of a network stall, but before the audio output itself underruns).
2020-09-03 20:39:23 +00:00
|
|
|
if (ao->untimed || !state.playing || ao->stream_silence) {
|
|
|
|
p->streaming = state.playing && !ao->untimed;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
p->playing = false;
|
|
|
|
}
|
|
|
|
ao->wakeup_cb(ao->wakeup_ctx);
|
|
|
|
// For ao_drain().
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_cond_broadcast(&p->wakeup);
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
return true;
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
2024-04-10 05:12:40 +00:00
|
|
|
static MP_THREAD_VOID ao_thread(void *arg)
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
{
|
|
|
|
struct ao *ao = arg;
|
|
|
|
struct buffer_state *p = ao->buffer_state;
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_thread_set_name("ao");
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
while (1) {
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
|
2020-08-29 14:27:56 +00:00
|
|
|
bool retry = false;
|
audio: refactor how data is passed to AO
This replaces the two buffers (ao_chain.ao_buffer in the core, and
buffer_state.buffers in the AO) with a single queue. Instead of having a
byte based buffer, the queue is simply a list of audio frames, as output
by the decoder. This should make dataflow simpler and reduce copying.
It also attempts to simplify fill_audio_out_buffers(), the function I
always hated most, because it's full of subtle and buggy logic.
Unfortunately, I got assaulted by corner cases, dumb features (attempt
at seamless looping, really?), and other crap, so it got pretty
complicated again. fill_audio_out_buffers() is still full of subtle and
buggy logic. Maybe it got worse. On the other hand, maybe there really
is some progress. Who knows.
Originally, the data flow parts was meant to be in f_output_chain, but
due to tricky interactions with the playloop code, it's now in the dummy
filter in audio.c.
At least this improves the way the audio PTS is passed to the encoder in
encoding mode. Now it attempts to pass frames directly, along with the
pts, which should minimize timestamp problems. But to be honest, encoder
mode is one big kludge that shouldn't exist in this way.
This commit should be considered pre-alpha code. There are lots of bugs
still hiding.
2020-08-28 18:23:54 +00:00
|
|
|
if (!ao->driver->initially_blocked || p->initial_unblocked)
|
2020-08-29 14:27:56 +00:00
|
|
|
retry = ao_play_data(ao);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
// Wait until the device wants us to write more data to it.
|
|
|
|
// Fallback to guessing.
|
2023-11-15 06:00:37 +00:00
|
|
|
int64_t timeout = INT64_MAX;
|
audio: fix stream-silence with push AOs (somewhat)
--audio-stream-silence is a shitty feature compensating for awful
consumer garbage, that mutes PCM at first to check whether it's
compressed audio, using formats advocated and owned by malicious patent
troll companies (who spend more money on their lawyers than paying any
technicians), wrapped in a wasteful way to make it constant bitrate
using a standard whose text is not freely available, and only rude users
want it. This feature has been carelessly broken, because it's
complicated and stupid. What would Jesus do? If not getting an aneurysm,
or pushing over tables with expensive A/V receivers on top of them, he'd
probably fix the feature. So let's take inspiration from Jesus Christ
himself, and do something as dumb as wasting some of our limited
lifetime on this incredibly stupid fucking shit.
This is tricky, because state changes like end-of-audio are supposed to
be driven by the AO driver, while playing silence precludes this. But it
seems code paths for "untimed" AOs can be reused.
But there are still problems. For example, underruns will just happen
normally (and stop audio streaming), because we don't have a separate
heuristic to check whether the buffer is "low enough" (as a consequence
of a network stall, but before the audio output itself underruns).
2020-09-03 20:39:23 +00:00
|
|
|
if (p->streaming && !retry && (!p->paused || ao->stream_silence)) {
|
2020-06-03 13:22:18 +00:00
|
|
|
// Wake up again if half of the audio buffer has been played.
|
|
|
|
// Since audio could play at a faster or slower pace, wake up twice
|
|
|
|
// as often as ideally needed.
|
2023-11-15 06:00:37 +00:00
|
|
|
timeout = MP_TIME_S_TO_NS(ao->device_buffer / (double)ao->samplerate * 0.25);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
}
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->pt_lock);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
if (p->terminate) {
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->pt_lock);
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
break;
|
|
|
|
}
|
2020-08-29 14:27:56 +00:00
|
|
|
if (!p->need_wakeup && !retry) {
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
MP_STATS(ao, "start audio wait");
|
2023-11-15 06:00:37 +00:00
|
|
|
mp_cond_timedwait(&p->pt_wakeup, &p->pt_lock, timeout);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
MP_STATS(ao, "end audio wait");
|
|
|
|
}
|
|
|
|
p->need_wakeup = false;
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->pt_lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
2023-10-21 02:55:41 +00:00
|
|
|
MP_THREAD_RETURN();
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void ao_unblock(struct ao *ao)
|
|
|
|
{
|
audio: redo internal AO API
This affects "pull" AOs only: ao_alsa, ao_pulse, ao_openal, ao_pcm,
ao_lavc. There are changes to the other AOs too, but that's only about
renaming ao_driver.resume to ao_driver.start.
ao_openal is broken because I didn't manage to fix it, so it exits with
an error message. If you want it, why don't _you_ put effort into it? I
see no reason to waste my own precious lifetime over this (I realize the
irony).
ao_alsa loses the poll() mechanism, but it was mostly broken and didn't
really do what it was supposed to. There doesn't seem to be anything in
the ALSA API to watch the playback status without polling (unless you
want to use raw UNIX signals).
No idea if ao_pulse is correct, or whether it's subtly broken now. There
is no documentation, so I can't tell what is correct, without reverse
engineering the whole project. I recommend using ALSA.
This was supposed to be just a simple fix, but somehow it expanded scope
like a train wreck. Very high chance of regressions, but probably only
for the AOs listed above. The rest you can figure out from reading the
diff.
2020-05-31 13:00:35 +00:00
|
|
|
if (ao->driver->write) {
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
struct buffer_state *p = ao->buffer_state;
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_lock(&p->lock);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
p->initial_unblocked = true;
|
2023-10-21 02:55:41 +00:00
|
|
|
mp_mutex_unlock(&p->lock);
|
2024-04-10 05:12:40 +00:00
|
|
|
ao_wakeup(ao);
|
audio: merge pull/push ring buffer glue code
This is preparation to further cleanups (and eventually actual
improvements) of the audio output code.
AOs are split into two classes: pull and push. Pull AOs let an audio
callback of the native audio API read from a ring buffer. Push AOs
expose a function that works similar to write(), and for which we start
a "feeder" thread. It seems making this split was beneficial, because of
the different data flow, and emulating the one or other in the AOs
directly would have created code duplication (all the "pull" AOs had
their own ring buffer implementation before it was cleaned up).
Unfortunately, both types had completely separate implementations (in
pull.c and push.c). The idea was that little can be shared anyway. But
that's very annoying now, because I want to change the API between AO
and player.
This commit attempts to merge them. I've moved everything from push.c to
pull.c, the trivial entrypoints from ao.c to pull.c, and attempted to
reconcile the differences. It's a mess, but at least there's only one
ring buffer within the AO code now. Everything should work mostly the
same. Pull AOs now always copy the audio data under a lock; before this
commit, all ring buffer access was lock-free (except for the decoder
wakeup callback, which acquired a mutex). In theory, this is "bad", and
people obsessed with lock-free stuff will hate me, but in practice
probably won't matter. The planned change will probably remove this
copying-under-lock again, but who knows when this will happen.
One change for the push AOs now makes it drop audio, where before only a
warning was logged. This is only in case of AOs or drivers which exhibit
unexpected (and now unsupported) behavior.
This is a risky change. Although it's completely trivial conceptually,
there are too many special cases. In addition, I barely tested it, and
I've messed with it in a half-motivated state over a longer time, barely
making any progress, and finishing it under a rush when I already should
have been asleep. Most things seem to work, and I made superficial tests
with alsa, sdl, and encode mode. This should cover most things, but
there are a lot of tricky things that received no coverage. All this
text means you should be prepared to roll back to an older commit and
report your problem.
2020-05-24 23:53:41 +00:00
|
|
|
}
|
|
|
|
}
|