This fixes the fullscreen issues on Intel for me. I'm baffled by it and
don't understand why this suddenly works. Intel drivers being shit?
Windows being shit? HWND or HDC being only 97% thread-safe instead of
98%? Me missing something subtle that is not documented anywhere?
Who knows.
Now instead of creating the HDC and OpenGL context on the renderer
thread, they're created on the GUI thread. The renderer thread will
then only call wglMakeCurrent, SwapBuffers, and OpenGL standard
functions.
Probably fixes github issue #968.
It seems the vdpau API does not support these.
Do a semi-expensive emulation of it. On the other hand, it's not like
this is a commonly-used feature. (It might be better to make --vf=flip
always copy instead of flipping it via pointer tricks - but everything
allows flipped images, and even decoders or libavfilter filters could
output them.)
If the compositor sends a configure event immediately after a window is
created (for example, if it implements tiling window management), mpv
will attempt to call wl_egl_window_resize before it has actually created
the egl_window, causing a crash.
Use OPT_KEYVALUELIST() for all places where AVOptions are directly set
from mpv command line options. This allows escaping values, better
diagnostics (also no more "pal"), and somehow reduces code size.
Remove the old crappy option parser (av_opts.c).
If OpenGL 3.x doesn't work, the fallback to legacy GL will call the
function to create the DC again, and it will assert. Just make it not
to, and also simplify the code a bit.
Fixes#974.
Since the new hwaccel API is now merged in ffmpeg's stable release, we can
finally remove support for the old API.
I pretty much kept lu_zero's new code unchanged and just added some error
printing (that we had with the old glue code) to make the life of our users
less miserable.
With software decoding, images were uploaded to vdpau surfaces as they
were queued to the VO. This makes it slightly more complicated
(especially later on), and has no advantages - so stop doing it.
The only reason why this was done explicitly was due to attempts to keep
the code equivalent (instead of risking performance regressions). The
original code did this naturally for certain reasons, but now that we
can measure that it has no advantages and just requires extra code, we
can just drop it.
Usually mp_image_params_guess_csp takes care of finding *some* default
for a video channel, but files with no video (or with extremely broken
configurations) end up leaving the colorspace information as
MP_CSP_PRIM_AUTO, which has no associated primaries.
As a result of this, color managed OSD messages could not display
because they were being color managed to match the (non-existing/absurd)
video channel. With this change, such non-spaces will have BT.709
primaries as far as color management and the OSD is concerned.
This fixes#961.
CC: @mpv-player/stable
Signed-off-by: wm4 <wm4@nowhere>
Follow up on commit 760548da. Mouse handling is a bit confusing, because
there are at least 3 coordinate systems associated with it, and it
should be cleaned up. But that is hard, so just apply a hack which gets
the currently-annoying issue (VO backends needing access to the VO) out
of the way.
The windows message loop now runs in a separate thread. Rendering,
such as with Direct3D or OpenGL, still happens in the main thread.
In particular, this should prevent the video from freezing if the
window is dragged. (The reason was that the message dispatcher won't
return while the dragging is active, so mpv couldn't update the
video at all.)
This is pretty "rough" and just hacked in, and there's no API yet to
make this easier for other backends. It will be cleaned up later
once we're sure that it works, or when we know how exactly it should
work. One oddity is that OpenGL is actually completely managed in the
renderer thread, while e.g. Cocoa (which has its own threading code)
creates the context in the GUI thread, and then lets the renderer
thread access it.
One strange issue is that we now have to stop WM_CLOSE from actually
closing the window. Instead, we wait until the playloop handles the
close command, and requests the VO to shutdown. This is done mainly
because closing the window apparently destroys it, and then WM_USER
can't be handled anymore - which means the playloop has no way to
wakeup the GUI thread. It seems you can't really win here... maybe
there's a better way to have a thread receive messages with and
without a window, but I didn't find one yet.
Dragging the window (by clicking into the middle of it) behaves
strangely in wine, but didn't before the change. Reason unknown.
VO backends which are or will run in their own thread have a problem
with vo_mouse_movement() calling vo_control(). Restrict this to VOs
which actually need this.
win32 does not provide a proper per-window context pointer. Although it
does allow passing a user-chosen value to WM_CREATE/WM_NCCREATE, this
is not enough - the first message doesn't even have to be WM_NCCREATE.
This gets us in trouble later on, so go the easy route and just use a
TLS variable.
__thread is gcc specific, but Windows is a very "special" platform
anyway. We support only MinGW and Cygwin on it, so who cares. (C11
standardizes __thread as _Thread_local; we can use that later.)
This shouldn't change anything. But it's worth making this explicit,
since it's very subtle and unintuitive: if the X parameter is the
magic value CW_USEDEFAULT, then the Y parameter is used as nCmdShow
parameter to ShowWindow(). And in our case, this is SW_HIDE (0),
because we want to create a hidden window.
This looked a bit overcomplicated. We don't care about the window
position (it should always be 0/0, unless the parent program moved it,
which it shouldn't). We don't care about the global on-screen position.
Also, we will just retrieve a WM_SIZE message if our window is resized,
and we don't need to update it manually.
The only thing we have to do is making sure our window fills the parent
window completely.
CS_OWNDC will make GetDC() always return the same HDC. This might
become a problem when OpenGL rendering and window management are
on different threads. Although I'm not too sure about this; our
code never requests a HDC outside of the OpenGL backend, and it
depends on whether win32 will internally request DCs. But in any
case, this seems to be cleaner.
Move the GL pixelformat setup code to gl_w32.c, where it's actually
needed. This also fixes that SetPixelFormat() should be called only
once, and not every time video params change.
These mostly describe self-explanatory things, and fail to explain
actually tricky things. Which means you just waste your time reading
this, and have to figure it out from the code anyway.
Preparation for moving win32 windowing to a separate thread.
The codesize is reduced a bit, because some small functions are
inlined, which reduces noise.
The main change is that now most functions use the private struct
directly, instead of accessing it indirectly through vo->w32.
Accesses to vo are minimalized.
The final goal is adding some sort of new windowing backend API. It
would be cleaner to use that as context pointer for all functions
(like struct vo was previously used), but since this is work in
progress, we just go with this commit.
The final goal is all mp_msg calls produce complete lines. We want this
because otherwise, race conditions could corrupt the terminal output,
and it's inconvenient for the client API too. This commit works towards
this goal. There's still code that has this not fixed yet, though.
There is no standard mechanism for detecting endianess. Doing it at
compile time in a portable way is probably hard. Doing it properly
with a configure check is probably hard too. Using the endian
definitions in <sys/types.h> (usually includes <endian.h>, which is
not available everywhere) works under circumstances, but the previous
commit broke it on OSX.
Ideally all code should be endian dependent, but that is not possible
due to the dependencies (such as FFmpeg, some video output APIs, some
audio output APIs).
Create a header osdep/endian.h, which contains various fallbacks.
Note that the last fallback uses libavutil; however, it's not clear
whether AV_HAVE_BIGENDIAN is a public symbol, or whether including
<libavutil/bswap.h> really makes it visible. And in fact we don't want
to pollute the namespace with libavutil definitions either. Thus it's
only the last fallback.
This approach is similar to what other vo_opengl backends do. It can also be
used in the future to create another cocoa backend that renders offscreen
with IOSurfaces or FBOs.
When seeking, we violently destroy the filter, because vapoursynth has
no proper API for terminating a video with unknown frame count. This
looks like an error to vapoursynth, and the error is returned via the
frame callbacks. The bug is that we remember this error state across
reinitialization, so on the first filter call after reinitialization, we
thought filtering the current frame failed. This caused a shift by 1
frame on each seek.
CC: @mpv-player/stable
DVD and Bluray (and to some extent cdda) require awful hacks all over
the codebase to make them work. The main reason is that they act like
container, but are entirely implemented on the stream layer. The raw
mpeg data resulting from these streams must be "extended" with the
container-like metadata transported via STREAM_CTRLs. The result were
hacks all over demux.c and some higher-level parts.
Add a "disc" pseudo-demuxer, and move all these hacks and special-cases
to it.
Cast away the "extra" bits (since apparently Window/XID is always
32 bit unsigned). This is not striclty needed, because you're not
supposed to pass garbage to --wid, just because the upper bits are
possibly not interpreted. But if you do so, this change increases
consistency in behavior and removes a strange behavior that was
thought to be a bug.
Also see github issue #906.
Apparently clearing on every map can cause problems with vdpau when
switching virtual desktops and such. This was observed with at least
XMonad and nvidia-340.17. It's not observed on some other setups without
XMonad.
It's not clear why this happens. Normally, the window background is not
saved, so clearing should have no additional affect. It's a complete
mystery. Possible, the use of legacy X drawing commands (used to clear
the window) interferes with vdpau operation in non-trivial ways.
Work this around by clearing on initial map only. This probably only
hides the underlying issue, but good enough.
Closes#897.
CC: @mpv-player/stable
Something like "char *s = ...; isdigit(s[0]);" triggers undefined
behavior, because char can be signed, and thus s[0] can be a negative
value. The is*() functions require unsigned char _or_ EOF. EOF is a
special value outside of unsigned char range, thus the argument to the
is*() functions can't be a char.
This undefined behavior can actually trigger crashes if the
implementation of these functions e.g. uses lookup tables, which are
then indexed with out-of-range values.
Replace all <ctype.h> uses with our own custom mp_is*() functions added
with misc/ctype.h. As a bonus, these functions are locale-independent.
(Although currently, we _require_ C locale for other reasons.)
These consult the vertical resolution, matching against 576 for
PAL and 480/486 for NTSC. The documentation has also been updated.
Signed-off-by: wm4 <wm4@nowhere>
Notably, we now conform to SMPTE 428-1-2006 when decoding XYZ12 input,
and we can support rendering intents other than colorimetric when
converting between BT.709 and BT.2020, like with :srgb or :icc-profile.
With this change, XYZ input is directly converted to the output
colorspace wherever possible, and to the colorspace specified by the
tags and/or --primaries option, otherwise.
This commit also restructures some of the CMS code in gl_video.c to
hopefully make it clearer which decision is being done where and why.
This also avoids an extra matrix multiplication when using :srgb, making
that path both more efficient and also eliminating more hard-coded
values.
In addition, the previously hard-coded XYZ to RGB matrix will be
dynamically generated.
This add support for reading primary information from lavc, categorized
into BT.601-525, BT.601-625, BT.709 and BT.2020; and passes it on to the
vo. In vo_opengl, we always generate the 3dlut against the wider BT.2020
and transform our source into this colorspace in the shader.
Do this simply by clearing the mapped buffer on every draw_image() call
without an actual video frame. (Maybe this is a bit expensive, but at
least not more expensive than regular video display.)
Clear the texture on reconfig(). (We could probably also do this simpler
with a flag, but this is actually less complicated - except that we move
the code to "map" a texture to a separate function.)
With the change to merge osd drawing into video frame drawing, some
bogus logic got in: they skipped drawing the OSD if no video frame is
available. This broke --no-video --force-window mode.
Make sure every video filter has valid parameters for input and output.
(This also ensures we don't take possibly invalid decoder output, or
feed invalid decodr/filter output to VOs.)
Also, the updated image size check now (almost) works like the
corresponding check in FFmpeg.
Until now, failure to allocate image data resulted in a crash (i.e.
abort() was called). This was intentional, because it's pretty silly to
degrade playback, and in almost all situations, the OOM will probably
kill you anyway. (And then there's the standard Linux overcommit
behavior, which also will kill you at some point.)
But I changed my opinion, so here we go. This change does not affect
_all_ memory allocations, just image data. Now in most failure cases,
the output will just be skipped. For video filters, this coincidentally
means that failure is treated as EOF (because the playback core assumes
EOF if nothing comes out of the video filter chain). In other
situations, output might be in some way degraded, like skipping frames,
not scaling OSD, and such.
Functions whose return values changed semantics:
mp_image_alloc
mp_image_new_copy
mp_image_new_ref
mp_image_make_writeable
mp_image_setrefp
mp_image_to_av_frame_and_unref
mp_image_from_av_frame
mp_image_new_external_ref
mp_image_new_custom_ref
mp_image_pool_make_writeable
mp_image_pool_get
mp_image_pool_new_copy
mp_vdpau_mixed_frame_create
vf_alloc_out_image
vf_make_out_image_writeable
glGetWindowScreenshot
The error log callback was not thread-safe and not library-safe. And
apparently there were some other details that made it not library-safe,
such as a global lcms plugin registry.
Switch the the thread-safe API provided by lcms2 starting with 2.6.
Remove our approximate thread-safety hacks.
Note that lcms basically provides 2 APIs now, the old functions, and
the thread-safe alternatives whose names end with THR. Some functions
don't change, because they already have a context of some sort. Care
must be taken not to accidentally use old APIs.
OSD used to be not thread-safe at all, so a track was used to get it
redrawn. This mostly reverts commit 6a2a8880, because OSD not being
thread-safe was the non-trivial part of it.
Mostly untested, because this code path is used on OSX only, and I don't
have OSX.
Let the VOs draw the OSD on their own, instead of making OSD drawing a
separate VO driver call. Further, let it be the VOs responsibility to
request subtitles with the correct PTS. We also basically allow the VO
to request OSD/subtitles at any time.
OSX changes untested.
Subsurfaces are only used by the wayland vo. Thats why it makes sense to move
all osd and subsurface specific parts to the vo_wayland.c
Also destroy the subsurfaces and subcompositor properly.
Not all the hardware supports kCGLPFASupportsAutomaticGraphicsSwitching
(apparently all Mid-2010 and before MacBooks do not work with it), so fallback
to not asking for this attribute in the GL pixel format.
This affects packed RGB formats up to 16 bits per pixel. The old mplayer
names used LSB-to-MSB order, while FFmpeg (and some other libraries) use
MSB-to-LSB.
Nothing should change with this commit, i.e. no bit order or endian bugs
should be added or fixed. In some cases, the name stays the same, even
though the byte order changes, e.g. RGB8->BGR8 and BGR8->RGB8, and this
affects the user-visible names too; this might cause confusion.
These suffixes are annoying when they're redundant, so strip them
automatically. On little endian machines, always strip the "le" suffix,
and on big endian machines vice versa (although I don't think anyone
ever tried to run mpv on a big endian machine).
Since pixel format strings are returned by a certain function and we
can't just change static strings, use a trick to pass a stack buffer
transparently. But this also means the string can't be permanently
stored by the caller, so vf_dlopen.c has to be updated. There seems
to be no other case where this is done, though.
This means use of the min/max fields can be dropped for the flag option
type, which makes some things slightly easier. I'm also not sure if the
client API handled the case of flag not being 0 or 1 correctly, and this
change gets rid of this concern.
While I'm not very fond of "const", it's important for declarations
(it decides whether a symbol is emitted in a read-only or read/write
section). Fix all these cases, so we have writeable global data only
when we really need.
Additionally to removing the global variables, this makes the options
more uniform. --ssf-... becomes --sws-..., and --sws becomes --sws-
scaler. For --sws-scaler, use choices instead of magic integer values.
Playing a video and then an audio file with cover art kept displaying
the last frame of the video. This was because the hasframe flag was set,
perhaps due to redrawing the last video frame before the cover art image
is decoded.
It seems we can't really get rid of this. There are no other hints to
remove decorations that work across all reasonable WMs, so we're stuck
with the ugly motif stuff.
But at least we can make the code for it less ugly.
The previous commit assumed the filter would be 1x1 (then constant
weight is correct) - but our code in fact uses at least a 2x2 filter. A
1x1 filter would generally be useless, except for nearest scaling - so
it didn't exist.
Insteasd of adding such a 1x1 filter, just turn the nearest weight
function into a scare function, which should take care of the issue.
This would imply eglGetProcAddress() doesn't work correctly, but using
dlsym() does. For now get rid of it - it won't work in libmpv, and we'll
probably need a better workaround if it's still broken.
This code was in the initial wayland commit.
The functions glXGetProcAddressARB() and glXQueryExtensionsString() were
loaded using dlsym(). This could fail when compiling to libmpv, because
then dlopen(NULL, ...) will look in the main program's list of
libraries, and the libGL linked to libmpv is never considered. (Don't
know if this somehow could be worked around.) The result is that using
vo_opengl with libmpv can fail.
Avoid this by not using dlsym(). glXGetProcAddressARB() was already used
directly in the same file, and that never caused any problems. (Still
add it to the configure test.) glXQueryExtensionsString() is documented
as added in GLX 1.1 - that's ancient.
The i_bps members of the sh_audio and dev_video structs are mostly used
for displaying the average audio and video bitrates. Keeping them in
bits-per-second avoids truncating them to bytes-per-second and changing
them back lateron.
This is incomplete; the video chain will still hold some vaapi objects
after destroying the decoder and thus the vaapi context. This is very
bad. Fixing it would require something like refcounting the vaapi
context, but I don't really want to.
mpv supports two hardware decoding APIs on Linux: vdpau and vaapi. Each
of these has emulation wrappers. The wrappers are usually slower and
have fewer features than their native opposites. In particular the libva
vdpau driver is practically unmaintained.
Check the vendor string and print a warning if emulation is detected.
Checking vendor strings is a very stupid thing to do, but I find the
thought of people using an emulated API for no reason worse.
Also, make --hwdec=auto never use an API that is detected as emulated.
This doesn't work quite right yet, because once one API is loaded,
vo_opengl doesn't unload it, so no hardware decoding will be used if the
first probed API (usually vdpau) is rejected. But good enough.
Sometimes, Matroska files store monotonic PTS for h264 tracks with
b-frames, which means the decoder actually returns non-monotonic PTS.
Handle this with an evil trick: if DTS is missing, set it to the PTS.
Then the existing logic, which deals with falling back to DTS if PTS is
broken. Actually, this trick is not so evil at all, because usually, PTS
has no errors, and DTS is either always set, or always unset. So this
_should_ provoke no regressions (famous last words).
libavformat actually does something similar: it derives DTS from PTS in
ways unknown to me. The result is very broken, but it causes the DTS
fallback to become active, and thus happens to work.
Also, prevent the heuristic from being active if PTS is merely monotonic
instead of strictly-monotonic. Non-unique PTS is broken, but we can't
fallback to DTS anyway in these cases.
The specific mkv file that is fixed with this commit had the following
fields set:
Muxing application: libebml v1.3.0 + libmatroska v1.4.1
Writing application: mkvmerge v6.7.0 ('Back to the Ground') [...]
But I know that this should also fix playback of mencoder produced mkv
files.
We pass a pointer to a GLint to sscanf, using the %d format. That format
_always_ takes int, and not GLint (whatever the heck that is). If GLint
is always int, then it doesn't make a difference, but is still better
because it doesn't play russian roulette with pointers.
Don't emit "hard" references to OpenGL functions. Always use the
platform specific function to lookup OpenGL functions, such as
glXGetProcAddress() with GLX (x11).
This actually fixes the build if only Wayland is enabled (e.g. using
--disable-gl-x11 on Linux).
Note that some sources claim that wglGetProcAddress() (win32) does not
return function pointers for OpenGL 1.1 functions (even if they are
valid and necessary in OpenGL 3.0). But if that happens, the fallback
employed in gl_w32.c/w32gpa() should catch this.
Setting this property was added 12 years ago, and the code was always
incorrect. The underlying data type is "long", not "pid_t". It's well
possible that the data types are different, and the pointer to the pid
variable is directly passed to XChangeProperty, possibly invoking
undefined behavior.
It's funny, because in theory using pid_t for PIDs sounds more correct.
_WIN_LAYER is apparently an old GNOME thing (also explains why there is
a function vo_x11_get_gnome_layer() involved in this code). Prefer the
NetWM hints over this. This just moves the NetWM case if-body over the
_WIN_LAYER one.
You can't use identifiers starting with "_" and an uppercase letter in
application programs. They are reserved by the C standard.
Unrelated change: drop unused/misleading vo_wm_NETWM define.
I can only assume the old code was wrong. EWMH does not document
anything with _WIN_LAYER. Instead, you have to toggle the state using a
client message. We also remove these weird non-sense fallbacks, like
using _NET_WM_STATE_BELOW - what the hell?
Integrate it with the existing surface allocator in vdpau.c. The changes
are a bit violent, because the vdpau API is so non-orthogonal: compared
to video surfaces, output surfaces use a different ID type, different
format types, and different API functions.
Also, introduce IMGFMT_VDPAU_OUTPUT for VdpOutputSurfaces wrapped in
mp_image, rather than hacking it. This is a bit cleaner.
black_pixel is an (apparently necessary) 1x1 black surface used for
clearing the screen. It was allocated in RGB mode only, but is sometimes
used in YUV mode too.
This works around an issue in OpenBox: OpenBox apparently sizes the
normal window incorrectly if aspect ratio hints are set, and the window
size is off by 1 pixel. Then, when going fullscreen and leaving
fullscreen again, mpv sets the hints based on OpenBox' broken window
size, and as result, OpenBox sizes the window incorrectly and is off by
1 pixel again - so it's 2 pixels off in total. The error gets more
visible, the more often you toggle fullscreen mode.
Work this around by not setting the window hints if we don't need to.
Actually we only need to do this when the video is resized during
fullscreen, which happens rarely. Under normal circumstances, leaving
fullscreen mode requires that the WM restores the old state.
As such, this commit is not only a workaround, but actually a cleanup.
Note that we do need to set the hints when leaving fullscreen if the
window has resized: even though we set the hints in
vo_x11_highlevel_resize (called by vo_x11_config_vo_window), this
doesn't seem to have an effect (at least on IceWM), so we have to do it
after that.
Side note: ot seems commit 625ad57a strangely triggered the OpenBox
issue according to user reports; I'm not sure why.
Before this commit, this was somehow polled (i.e. not the right way).
Also, selects the correct window when doing --wid=0 (which is another
weird special-case).
Enabling DPMS even though you disabled it globally is pretty unfriendly,
so don't do it. Instead, we only disable DPMS if it was enabled, and
only enable it if we disabled it ourselves.
The other way should never happen (disabling DPMS permanently), unless
mpv crashes during playback.
Reduces some code-duplication.
Just call DPMSEnable/DPMSDisable, instead of DPMSForceLevel when
reenabling DPMS. "Force" sounds evil, and messing with DPMS is already
pretty evil. I'm not even sure that we should.
XGetWindowProperty is a really bad API, almost as if the NSA designed
it. The wrapper takes care of verifying the return values and handle
corner cases.
The window "gravity" influences how placement interacts with WM added
borders (i.e. from decorations). This is probably what the code removed
in commit c14721c8 was about.
In theory, we'd probably want to set the gravity depending on the
relative placement requested by the user (so that it's possible to line
up the top/left video pixel with the monitor corner, as well as the
bottom/right pixel - but that would be too complicated, and who cares
after all?).
I'm also not sure whether CenterGravity really uses the top/left corner
as reference point (instead of making coordinates relative to the window
center), but empirically it's correct.
Try to get the "new" code path (using NetWM/EWMH) free of hacks done for
the sake of old WMs or the no-WM case.
Implement --fs-screen using _NET_WM_FULLSCREEN_MONITORS.
Keeps the window centered on resize. Seems nicer. (Although it's worse
if 1. the default placement of the WM puts it into a monitor corner,
and 2. you switch to a larger video.)
It was added with 3813c685 in 2004. I'm not really sure why this gravity
stuff would be needed; apparently it has to do with misplacements with
broken WMs and had to be changed on fullscreen. Just get rid of it; it
works perfectly fine without on modern WMs.
The thread discussing this is here:
http://mplayerhq.hu/pipermail/mplayer-dev-eng/2004-July/027674.html
This should get rid of some flickering. Since this actually skips all
the wacky fullscreening code on startup, this might lead to certain
wacky features to stop working. In this case, you'll have to use the
--x11-fstype option, and disable _NETWM_STATE_FULLSCREEN usage.
vo_x11_map_window() was attempting to clear the window on map. However,
it did so immediately after the map request. It probably assumed that
the drawing calls for clearing the window would be queued along with the
map request, and then executed in the right order. However, this
assumption was wrong - the map request first has to go to the window
manager (I guess?), so a lot of things happen before the window is even
mapped.
Fix this by moving the call to the MapNotify message handler, when the
window (apparently) becomes really visible.
I also tried to set CWBackPixel to black instead, but this seemed to
result in flickering on manual resizing.
This blocks everything, until the window is actually reported as mapped.
This fixes the race condition between VO initialization and mapping the
window, which resulted in possibly different window sizes, leading to an
immediate redraw, visible as flashing.
Note that if the map event never comes for some reason, we're out of
luck and will block forever.
It could in theory happen that the filter loop will enter a blocking
wait, even though it could make progress by emptying the list of
already-filtered images. I'm not quite sure if this could actually cause
a real issue - probably not.
VapourSynth won't just filter multiple frames at once on its own. You
have to request multiple frames at once manually. This is what this
commit introduces: a sub-option controls how many frames will be
requested at once. This also changes the semantics of the maxbuffer sub-
option, now renamed to buffered-frames.
Preparation so that various things related to video can run in different
threads. One part to this is making the video surface pool safe.
Another issue is the preemption mechanism, which continues to give us
endless pain. In theory, it's probably impossible to handle preemption
100% correctly and race-condition free, unless _every_ API user in the
same process uses a central, shared mutex to protect every vdpau API
call. Otherwise, it could happen that one thread recovering from
preemption allocates a vdpau object, and then another thread (which
hasn't recovered yet) happens to free the object for some reason. This
is because objects are referenced by integer IDs, and vdpau will reuse
IDs invalidated by preemption after preemption.
Since this is unreasonable, we're as lazy as possible when it comes to
handling preemption. We don't do any locking around the mp_vdpau_ctx
fields that are normally immutable, and only can change when recovering
from preemption. In practice, this will work, because it doesn't matter
whether not-yet-recovered components use the old or new vdpau function
pointers or device ID. Code calls mp_vdpau_handle_preemption() anyway to
check for the preemption event and possibly to recover, and that
function acquires the lock protecting the preemption state.
Another possible source of potential grandiose fuckup is the fact that
the vdpau library is in fact only a tiny wrapper, and the real driver
lives in a shared object dlopen()ed by the wrapper. The wrapper also
calls dlclose() on the loaded shared object in some situations. One
possible danger is that failing to recreate a vdpau device could trigger
a dlclose() call, and that glibc might unload it. Currently, glibc
implements full unloading of shared objects on the last dlclose() call,
and if that happens, calls to function pointers pointing into the shared
object would obviously crash. Fortunately, it seems the existing vdpau
wrapper won't trigger this case and never unloads the driver once it's
successfully loaded.
To make it short, vdpau preemption opens up endless depths of WTFs.
Another issue is that any participating thread might do the preemption
recovery (whichever comes first). This is easier to implement. The
implication is that we need threadsafe xlib. We just hope and pray that
this will actually work. This also means that once vdpau code is
actually involved in a multithreaded scenario, we have to add
XInitThreads() to the X11 code.
Use the newly provided mp_vdpau_handle_preemption() function, instead of
accessing mp_vdpau_ctx fields directly. Will probably make multithreaded
access to the vdpau context easier.
Mostly unrelated to the actual changes, I've noticed that using hw
decoding with vo_opengl sometimes leads to segfaults inside of nvidia's
libGL when doing the following:
1. use hw decoding + vo_opengl
2. switch to console (will preempt on nvidia systems)
3. switch back to X (mpv will recover, switches to sw decoding)
4. enable hw decoding again
5. exit mpv
Then it segfaults when mpv finally calls exit(). I'll just blame nvidia,
although it seems likely that something in the gl_hwdec_vdpau.c
preemption handling triggers corner cases in nvidia's code.
This was broken for some time, and it didn't recover correctly.
Redo decoder display preemption. Instead of trying to reinitialize the
hw decoder, simply fallback to software decoding. I consider display
preemption a bug in the vdpau API, so being able to _somehow_ recover
playback is good enough.
The approach taking here will probably also make it easier to handle
multithreading.
Until recently, the VO was an unavoidable part of the seeking code path.
This was because vdpau deinterlacing could double the framerate, and hr-
seek and framestepping etc. all had to "see" the additional frames. But
we've removed the frame doubling from the vdpau VO and moved it into a
video filter (vf_vdpaupp), and there's no reason left why the VO should
participate in seeking.
Instead of queuing frames to the VO during seek and skipping them
afterwards, drop the frames early.
This actually might make seeking with vo_vdpau and software decoding
faster, although I haven't measured it.
It doesn't look like vo_wayland_config() necessarily sets this flag, so
it seems safer to trigger an explicit resize. This accounts for the case
when playing a new file with different size than the one before.
Currently, vo_reconfig() calculates the requested window size and sets
the vo->dwidth/dheight fields _if_ VOCTRL_UPDATE_SCREENINFO is
implemented by the VO or the windowing backend. The window size can be
different from the display size if e.g. the --geometry option is used.
It will also set the vo->dx/dy fields and read vo->xinerama_x/y.
It turned out that this is very backwards and actually requires the
windowing backends to workaround these things. There's also
MPOpts.screenwidth/screenheight, which used to map to actual options,
but is now used only to communicate the screen size to the vo.c code
calculating the window size and position.
Change this by making the window geometry calculations available as
separate functions. This commit doesn't change any VO code yet, and just
emulates the old way using the new functions. VO code will remove its
usage of VOCTRL_UPDATE_SCREENINFO and use the new functions directly.
Commit 433161 actually broke vo_opengl (and maybe others), because
config_ok is not necessarily set correctly yet _during_ reconfig. So a
vo_get_src_dst_rects() call during reconfig did nothing.
When the VO was not initialized with vo_reconfig(), or if the last
vo_reconfig() failed, changing panscan would cause a crash due to
vo_get_src_dst_rects() dereferencing vo->params (NULL if not
configured).
Just do nothing if that happens, as there is no video that could be
displayed anyway.
Doesn't really seem to be much of use. Get rid of the remaining uses of
it.
Concerning vo_opengl_old, it seems uninitGl() works fine even if called
before initialization.
This was a minor code duplication between vf_vdpaupp.c and vo_vdpau.c.
(In theory, we could always require using vf_vdpaupp with vo_vdpau, but
I think it's better if vo_vdpau can work standalone.)
Also remove MSGL_SMODE and friends.
Note: The indent in options.rst was added to work around a bug in
ReportLab that causes the PDF manual build to fail.
Change how the video decoding loop works. The structure should now be a
bit easier to follow. The interactions on format changes are (probably)
simpler. This also aligns the decoding loop with future planned changes,
such as moving various things to separate threads.
vf_fix_img_params() takes care of overwriting image parameters that are
normally not set correctly by filters. But this makes no sense for input
images. So instead, check that the input is correct.
It still has to be done for the first input image, because that's used
to handle some overrides (see video_reconfig_filters()).
These replace vf_read_output_frame(), although we still emulate that
function. This change is preparation for another commit (and this is
basically just to reduce the diff and signal/noise ratio in that
commit).
Basically, if we feed the filter a new image even after the EOF state
has been reached (e.g. because the input stream "recovered"), we want
the filter to restart, instead of returning an error forever.
Some non-deinterlacing filters (potentially denoising) also use
additional frames for filtering. The vdpau docs suggest providing at
least 1 future and 2 past _fields_, which means we need to provide 1
past frame (the future field is already the other field of the current
field, and both fields are in the same frame).
We can easily achieve this by buffering an additional frame in the non-
deint case.
Remove the special casing of vo_vdpau vs. other VOs. Replace the
complicated interaction between vo.c and vo_vdpau.c with a simple queue
in vo.c. VOs other than vdpau are handled by setting the length of the
queue to 1 (this is essentially what waiting_mpi was).
Note that vo_vdpau.c seems to have buffered only 1 or 2 frames into the
future, while the remaining 3 or 4 frames were past frames. So the new
code buffers 2 frames (vo_vdpau.c requests this queue length by setting
vo->max_video_queue to 2). It should probably be investigated why
vo_vdpau.c kept so many past frames.
The field vo->redrawing is removed. I'm not really sure what that would
be needed for; it seems pointless.
Future directions include making the interface between playloop and VO
simpler, as well as making rendering a frame a single operation, as
opposed to the weird 3-step sequence of rendering, drawing OSD, and
flipping.
The previous commits changed vo_vdpau so that these options are set by
vf_vdpaupp, and the corresponding vo_vdpau were ignored. But for
compatibility, keep the "old" options working.
The value of this is questionable - maybe the vo_vdpau options should
just be removed. For now, at least demonstrate that it's possible.
The "deint" suboption still doesn't work, because the framerate doubling
logic required for some deint modes was moved to vf_vdpaupp. This
requires more elaborate workarounds.
This is slightly incomplete: the mixer options, such as sharpen and
especially deinterlacing, are ignored. This also breaks automatic
enabling of interlacing with 'D' or --deinterlace. These issues will be
fixed later in the following commits.
Note that we keep all the custom vdpau queue stuff. This will also be
simplified later.
This uses mp_vdpau_mixer_render(). The benefit is that it makes vdpau
deinterlacing just work. One additional minor advantage is that the
video mixer creation code is factored out (although that is a double-
edged sword).
This factors out some code from vo_vdpau.c, especially deinterlacing
handling. The intention is to use this for vo_vdpau.c to make the logic
significantly easier, and to use it for vo_opengl (gl_hwdec_vdpau.c) to
allow selecting deinterlace and postprocessing modes.
As of this commit, the filter actually does nothing, since both vo_vdpau
and vo_opengl treat the generated images as normal vdpau images. This
will change in the following commits.
It might have been nice not to do this so that metadata could
accumulate accross seeks, but it seems libavfilter looses its copy
anyway on recreate_graph.
lavfi would segfault due to a NULL dereference if it was asked for its
metadata and none had been allocated (oops). This happens for libav
which has no concept of filter metadata.
Commit 5e4e248 added a mp_image_params field to mp_image, and moved many
parameters to that struct. display_w/h was left redundant with
mp_image_params.d_w/d_h. These fields were supposed to be always in
sync, but it seems some code forgot to do this correctly, such as
vf_fix_img_params() or mp_image_copy_attributes(). This led to the
problem in github issue #756, because display_w/_h could become
incorrect.
It turns out that most code didn't use the old fields anyway. Just
remove them. Note that mp_image_params.d_w/d_h are supposed to be always
valid, so the additional checks for 0 shouldn't be needed. Remove these
checks as well.
Fixes#756.
In theory, returning the screenshot with original pixel aspect would
allow avoiding scaling them with image formats that support non-square
pixels, but in practice this isn't used anyway (nothing seems to
understand e.g. jpeg aspect ratio tags).
We only support them for input. The frame properties of output frames
are ignored (except frame durations).
Properties not set for now: _ChromaLocation, _Field, _FieldBased
Set _DurationNum/_DurationDen on each VS frame, instead of
_AbsoluteTime. The duration is the difference between the timestamp of
the frame and the next frame, and when receiving filtered VS frames, we
convert them back to an absolute PTS by summing them.
We pass the timestamps with microsecond resolution. mpv uses double for
timestamps internally, so we don't know the "real" timebase or FPS. VS
on the other hand uses fractions for frame durations. We can't pass
through the numbers exactly, but microseconds ought to be enough to be
even safe from accumulating rounding errors.
Since this leaks video images, and the player keeps feeding new images
to the fitler even if it fails, this would probably have disastrous
consequences.
Or in other words, add support for properly draining remaining frames
from video filters. vf_yadif is buffering at least one frame, and the
buffered frame was not retrieved on EOF.
For most filters, ignore this for now, and just adjust them to the
changed semantics of filter_ext. But for vf_lavfi (used by vf_yadif),
real support is implemented. libavfilter handles this simply by passing
a NULL frame to av_buffersrc_add_frame(), so we just have to make
mp_to_av() handle NULL arguments.
In load_next_vo_frame(), we first try to output a frame buffered in the
VO, then the filter, and then (if EOF is reached and there's still no
new frame) the VO again, with draining enabled. I guess this was
implemented slightly incorrectly before, because the filter chain still
could have had remaining output frames.
This extracts the scheduling logic to a single function which is nicer to keep
it consistent.
Additionally make sure we don't schedule sync operations from a sync operation
itself since that could cause deadlocks (even if it should not be happening
with the current code).
Previously the window could be made to completly exit the screen with a
combination or moving it close to an edge and halving it's size (via cmd+0).
This commit address the problem in the most simple way possibile by
constraining the window to the closest edge in these edge cases.
This fixes a couple of issues with the Cocoa `--native-fs` mode, primarily:
- A ghost titlebar at the top of the screen in full screen:
This was caused by the window constraining code kicking in during
fullscreen. Simply returning the unconstrained rect from the constraining
method fixes the problem.
- Incorrect behavior when using the titlebar buttons to enter/exit
fullscreen, as opposed to the OSD button.
This was caused by mpv's internal fullscreen state going out of sync with
the NSWindow's one. This was the case because `toggleFullScreen:`
completely bypassed the normal event flow that mpv expects.
Signed-off-by: Ryan Goulden <percontation@gmail.com>
Change style for mpv, simplify and refactor some of the constraining code.
Signed-off-by: Stefano Pigozzi <stefano.pigozzi@gmail.com>
I thought the "_" in "_AbsoluteTime" was part of the documentation
markup.
This still doesn't help us with VS filters that change timing;
apparently you must use frame durations instead.
Make the filter apply the pixel aspect ratio of the input to the output.
This is more useful than forcing 1:1 PAR when playing anamorphic video
such as DVDs.
VapourSynth itself actually allows passing through the aspect ratio, but
it's in a not very useful form for us: it's per video-frame instead of
constant (i.e in VSVideoInfo). As long as we don't have a way to allow a
filter to spontaneously change output parameters, we can't use this.
(And I don't really feel like making this possible.)
This reverts commit 6e34b0ec1f50612cb2767da3dbc27be0be63041d.
There has always been an error message "proxy already has a listener" and
I couldn't reproduce where it is comming from until now. The display interface
already has a listener and we can't overwrite it. Now remove the code and avoid
this error message.
Conflicts:
video/out/wayland_common.c
Add the event FD after preinit, remove it before destroy. There's no
need to do it on vo_config, and there's no need to remove the event
FD when vo_config fails.
Unfortunately, if a VO can't display something as intended, we can just
complain to the user, and leave it at it. But it's still better than
silently displaying things differently with different VOs.
For now, this is used for rotation only. Other things that we should
check includes colorspace and colorlevels stuff.
When using rotation with hw decoding, and the VO does not support
rotation, vf_rotate is attempted to be inserted. This will go wrong, and
after that it can't recover because a vf_scale filter was autoinserted.
Just removing all autoinserted filters before reconfig fixes this.
This turned out much more complicated than I thought. It's not just a
matter of adjusting the texture coordinates, but you also have to
consider separated scaling and panscan clipping, which make everything
complicated.
This actually still doesn't clip 100% correctly, but the bug is only
visible when rotating (or flipping with --vf=flip), and using something
like --video-pan-x/y at the same time.
For rotation, we assume that the source image will be rotated within the
VO, so the aspect/panscan code needs to calculate its param using
rotated coordinates. VOs which support rotation natively can use this.
This couldn't rotate by 180°. Add this, and also make the parameter in
degrees, instead of magic numbers.
For now, drop the flipping stuff. You can still flip with --vf=flip or
--vf=mirror. Drop the landscape/portrait stuff - I think this is
something almost nobody will use. If it turns out that we need some of
these things, they can be readded later.
Make it use libavfilter. Its vf_transpose implementation looks pretty
simple, except that it uses slice threading and should be much faster.
Fix all include statements of the form:
#include "libav.../..."
These come from MPlayer times, when FFmpeg was somehow part of the
MPlayer build tree, and this form was needed to prefer the local files
over system FFmpeg.
In some cases, the include statement wasn't needed or could be replaced
with mpv defined symbols.
Not needed anymore. I'm not opposed to having asm, but inline asm is too
much of a pain, and it was planned long ago to eventually get rid fo all
inline asm uses.
For the note, the inline asm use that was removed with the previous
commits was almost worthless. It was confined to video filters, and most
video filtering is now done with libavfilter. Some mpv filters (like
vf_pullup) actually redirect to libavfilter if possible.
If asm is added in the future, it should happen in the form of external
files.
