I think I was wrong about having to reset the stats when mpv stops
producing frames, eg. when it's paused. As long as the swapchain doesn't
underflow, last_queue_display_time will still be accurate, because the
next frame produced should still be presented one vsync after the
last one in the swapchain.
If the swapchain underflows (which is the common case for when mpv is
paused for more than 150ms,) the next predicted frame time should be in
the past. It should be fine to leave last_queue_display_time unset in
this case, since vo.c will use the current time instead, which is a
decent guess (though it doesn't take vsync phase into account.)
last_sync_refresh_count and last_sync_qpc_time should be kept on
swapchain underflow as well. Assuming the display refresh rate doesn't
change while mpv is paused, they'll only provide a more accurate guess
of the vsync duration when mpv starts playing again. Also,
vsync_duration_qpc never needs to get reset. It will get overwritten
immediately in most cases, and when it doesn't, it's still a better
guess of the vsync duration than nothing.
The render API (vo_libmpv) had potential deadlock problems with
MPV_RENDER_PARAM_ADVANCED_CONTROL. This required vd-lavc-dr to be
enabled (the default). I never observed these deadlocks in the wild
(doesn't mean they didn't happen), although I could specifically provoke
them with some code changes.
The problem was mostly about DR (direct rendering, letting the video
decoder write to OpenGL buffer memory). Allocating/freeing a DR image
needs to be done on the OpenGL thread, even though _lots_ of threads are
involved with handling images. Freeing a DR image is a special case that
can happen any time. dr_helper.c does most of the evil magic of
achieving this. Unfortunately, there was a (sort of) circular lock
dependency: freeing an image while certain internal locks are held would
trigger the user's context update callback, which in turn would call
mpv_render_context_update(), which processed all pending free requests,
and then acquire an internal lock - which the caller might not release
until a further DR image could be freed.
"Solve" this by making freeing DR images asynchronous. This is slightly
risky, but actually not much. The DR images will be free'd eventually.
The biggest disadvantage is probably that debugging might get trickier.
Any solution to this problem will probably add images to free to some
sort of queue, and then process it later. I considered making this more
explicit (so there'd be a point where the caller forcibly waits for all
queued items to be free'd), but discarded these ideas as this probably
would only increase complexity.
Another consequence is that freeing DR images on the GL thread is not
synchronous anymore. Instead, it mpv_render_context_update() will do it
with a delay. This seems roundabout, but doesn't actually change
anything, and avoids additional code.
This also fixes that the render API required the render API user to
remain on the same thread, even though this wasn't documented. As such,
it was a bug. OpenGL essentially forces you to do all GL usage on a
single thread, but in theory the API user could for example move the GL
context to another thread.
The API bump is because I think you can't make enough noise about this.
Since we don't backport fixes to old versions, I'm specifically stating
that old versions are broken, and I'm supplying workarounds.
Internally, dr_helper_create() does not use pthread_self() anymore, thus
the vo.c change. I think it's better to make binding to the current
thread as explicit as possible.
Of course it's not sure that this fixes all deadlocks (probably not).
This adds vsync reporting to the D3D11 backend using the presentation
feedback provided by DXGI, which is pretty similar to what's provided by
GLX_OML_sync_control in the GLX backend. In DirectX, PresentCount is the
SBC, PresentRefreshCount and SyncRefreshCount are kind of like the MSC
and SyncQPCTime is the UST.
Unlike GLX, the DXGI API makes it possible for PresentCount and
SyncQPCTime to refer to different physical vsyncs, in which case
PresentRefreshCount and SyncRefreshCount will be different. The code
supports this possibility, even though it's not clear whether it can
happen when using flip-model presentation. The docs say for flip-model
apps, PresentRefreshCount is equal to SyncRefreshCount "when the app
presents on every vsync," but on my hardware, they're always equal, even
when mpv misses a vsync. They can definitely be different in exclusive
fullscreen bitblt mode, though, which mpv doesn't support now, but might
support in future.
Another difference to GLX is that, at least on my hardware,
PresentRefreshCount and SyncRefreshCount always refer to the last
physical vsync on which mpv presented a frame, but glxGetSyncValues can
apparently return a MSC and UST from the most recent physical vsync,
even if mpv didn't present a new frame on it. This might result in
different behaviour between the two backends after dropped frames or
brief pauses.
Also note, the docs for the DXGI presentation feedback APIs are pretty
awful, even by Microsoft standards. In particular the docs for
DXGI_FRAME_STATISTICS are misleading (PresentCount really is the number
of times Present() has been called for that frame, not "the running
total count of times that an image was presented to the monitor since
the computer booted.")
For good documentation, try these:
https://docs.microsoft.com/en-us/windows/win32/direct3ddxgi/dxgi-flip-modelhttps://docs.microsoft.com/en-us/windows/win32/direct3d9/d3dpresentstatshttps://docs.microsoft.com/en-us/windows-hardware/drivers/ddi/content/d3dkmthk/ns-d3dkmthk-_d3dkmt_present_stats
(Yeah, the docs for the D3D9Ex and even the kernel-mode version of this
structure are better than the DXGI ones. It seems possible that they're
all rewordings of the same internal Microsoft docs, but whoever wrote
the DXGI one didn't really understand it.)
Certain mpv config options require wl->current_output to be created
before the video can actually start rendering. Just always create it
here if the current_output doesn't exist (the one exception being the
--fs option with no --fs-screen flag). Incidentally, this also fixes
--fs-screen not working on wayland.
To find the correct ICC profile X atom, the screen number was calculated
directly from the xrandr order of the screens.
But if a primary screen is set, it should be the first Xinerama screen,
even if it is not the first xrandr screen.
Calculate the the proper atom id for each screen.
On wayland the cursor has to be configured each time the pointer enters.
Currently if the window (re)gains the focus, the pointer is not hidden,
even when configured. After the mouse has been moved the pointer hides
correctly.
https://wayland.freedesktop.org/docs/html/apa.html#protocol-spec-wl_pointer:
wl_pointer::enter - enter event
...
When a seat's focus enters a surface, the pointer image is undefined
and a client should respond to this event by setting an appropriate
pointer image with the set_cursor request.
Fixes#6185.
Signed-off-by: Thomas Weißschuh <thomas@t-8ch.de>
Previously, the only mouse buttons supported in wayland were left,
right, and middle click. This adds the thumb back/forward buttons as
valid bindings. Also it removes the old, default behavior of always
sending a right click if an unrecognized mouse button is clicked.
In a related but different fix, the magnitude of an axis event in
wayland is not important to mpv since it internally handles all scaling.
The only thing we care about is getting the sign when the event occurs.
Check if eglGetPlatformDisplayEXT is available and try to
use it to obtain the display connection. Fall back to eglGetDisplay
if eglGetPlatformDisplayEXT is not available or failing.
From PR #5992
This is mostly for the case when mpv_render_context_free() is called
while video is going on. This is supposed to gracefully stop video and
deinitialize everything properly. (I feel like it would put too much on
the API user to require that video is stopped before calling this
function. Whether video is running or not is a fairly highlevel thing,
and the API user could not do it in a race-free way.)
One problem was that unit() accessed ctx after ctx->in_use was set to
false. The update(ctx) call was basically a racy use-after-free. It
needed that call to wake up the mpv_render_context_free() loop that
waited for VO uninit. Fix this by triggering the wakeup inside the lock,
and then doing "barrier" locking in mpv_render_context_free().
Another problem was that the wait loop didn't really wait properly. IT
seems the had_kill_update field was a botched attempt to do that. It's
indeed quite hairy to do that with update(). Instead make use of the
dispatch queue (infinite timeout, using mp_dispatch_interrupt()), which
handles the problem of having to wait both for dispatch queue updates
and VO uninit at the same time.
Preparation for the next commit. Until now, it was only needed if DR was
involved. One reason for not always creating it was that you normally
must not use it if advanced_control is not enabled. This is why e.g.
VOCTRL_SCREENSHOT now checks for that variable; it still can't use
ctx->dispatch if the render API user did not enable it.
render api needs to wait for vo to be destroyed before frees the context.
The purpose of kill_cb is to wake up render api after vo is destroyed,
but uninit did that before kill_cb, so kill_cb tries using the freed
memory. Remove kill_cb to fix the issue as uninit is able to do the
work.
Equalizer control was redone in 03cf150ff3 (over 2 years
ago). Ever since, the equalizer control structs and the GET voctrl have
been unused. Only the SET voctrl is still used as notification mechanism
(actually a bad hack to avoid some further option change handling
complexity).
Remove the unused parts.
Dear diary,
today I fixed a shitty bug that was all my fault because I made a
horrible mess. (Except it was a horrible mess before I even touched
this shit, but let's not blame others.)
Sometimes, updates to VO option that control video sizing (like panscan)
didn't update the screen correctly. They were delayed until the next
option change or so.
It turns out that if the option update happens at the "same" time as a
VOCTRL, update_opts() doesn't actually notify the vo_driver of the
change. This in turn happened because run_control() called
m_config_cache_update(). The latter function returns true if the options
changed since the last call, and update_opts() also calls it (on the
same config cache) for the same purpose. The update_opts() call, which
is triggered by a third mechanism, comes later, but the cache update
call will return false (as it should). Basically, given the config API,
you can't act differently on multiple update calls and expect it to
work. The skipped handling in update_opts() meant that the notification
required to apply the changed option wasn't run.
Fix this by simply calling update_opts() directly instead. Now there's
only 1 m_config_cache_update() call on this specific instance. Fix the
call in run_reconfig() too, so the previous sentence isn't a lie (but it
probably doesn't make a difference in practice due to certain details).
I'm not sure how I even ran into this sort-of race condition. The VOCTRL
that messed up the option update was VOCTRL_UPDATE_PLAYBACK_STATE, which
happens semi-regularly.
Why this config cache shit and all the other shit? Rediscovering this
crap wasn't pleasant. It's a bunch of hacks that became necessary when
the ancient MPlayer architecture made it hard to move the VO to a
separate thread.
All the VO code typically accesses vo->opts (whose fields all used to be
global variables in MPlayer). The frontend changes these on user input.
Putting locking around all the options would be a nightmare, and keeping
a copy of the options in the thread was much simpler. You need a way to
propagate option changes, notify the thread, and update the local copy
too. And the result of these thoughts was the config cache mechanism.
In this specific case, the relevant cache update call in update_opts()
triggers a VOCTRL_SET_PANSCAN to the VO driver, which isn't related to
its former function anymore. Instead, it causes the VO driver to update
the video sizing/placing options, which the generic VO code can't do.
(Mostly because the VO driver includes the windowing stuff and is
responsible for resizing etc. itself.)
VOCTRLs sent by the frontend are even worse. MPlayer had no real runtime
option change mechanism. Some options were vaguely duplicated by
properties, so you could effectively change those options at runtime.
Each of these options had its own VOCTRL, which still exist today, e.g.
VOCTRL_FULLSCREEN, or VOCTRL_ONTOP. I tried to make all options runtime
changeable, and to unify properties with options. But I couldn't be
bothered with updating all VO drivers to listen to option changes
directly, because that would be pretty tedious. So the property code is
still all there and sends the old VOCTRLs. But of course you need to
sync up the options, which is why the run_control() code did that.
(Unrelated: VO_EVENT_FULLSCREEN_STATE is the worst shithack of them all.
Currently, only the frontend can actually write to options (for awful
reasons), so if the fullscreen state changes due to outside interaction,
the VO driver can't update the corresponding option fields. So the VO
notifies the frontend with said VO_EVENT_, and the frontend then sends
VOCTRL_GET_FULLSCREEN, and updates the global copy of the option with
the value returned by that. I still like to think the situation is not
that bad considering the monstrous effort of converting single-threaded
code that had hundreds of options in global variables to multi-threaded
code with no global variables at all.)
m_geometry_apply() will read and modify the dummy variable. It's not
actually used for anything, but valgrind will still warn against
uninitialized data. I'm not sure whether this was UB, but in any case
it's annoying when running valgrind.
During probing on a system with AMD GPU, mpv used to output the
following messages if hardware decoding was enabled:
[ffmpeg] AVHWFramesContext: Failed to create surface: 2 (resource
allocation failed).
[ffmpeg] AVHWFramesContext: Unable to allocate a surface from internal
buffer pool.
This commit removed the message, with hopefully no other side effects.
Long explanations follow, better don't read them, it's just tedious
drivel about the details. People should learn to write concise commit
messages, not drone on and on endlessly all while they have no fucking
point.
The code probes supported hardware pixel format, and checks whether they
can be mapped as textures. av_hwdevice_get_hwframe_constraints() returns
a list of hardware pixel formats in the valid_sw_formats field (the "sw"
means software, but they're still hardware pixel formats, makes sense).
This contained the format yuv420p, even though this is not a valid
hardware format. Trying to create a surface of this type results in VA
surface creation failure, upon which FFmpeg prints the error messages
above. We'd be fine with this, except FFmpeg has a global log callback,
and there's no way to suppress these messages without creating other
issues.
It turns out that FFmpeg's vaapi implementation returns all formats from
vaQueryImageFormats() if no "hwconfig" is provided. This list includes
yuv420p, which is probably supported for surface upload/download, but
not as native format. Following FFmpeg's logic, it should not appear in
the valid_sw_formats list, because formats for transfers are returned by
another roundabout API.
Idiotically, there doesn't seem to be any vaapi call that determines
whether a format is a valid surface format. All mechanisms to do this
are bound to a VAConfigID (= video codec or video processor), all while
the actual surface creation API strangely does not take a VAConfigID (a
big WTF).
Also, calling the vaCreateSurfaces() API ourselves for probing is out of
the question, because that functions is utterly and idiotically complex.
Look at the FFmpeg code and how much effort it requires to setup a
complete set of attributes - we can't duplicate this.
So the only way left to do this is the most idiotic and tedious way:
enumerating all VAProfile (and VAEntrypoints) to create all possible
VAConfigIDs. Each of the VAConfigIDs is associated with a list of
formats, which FFmpeg can return (by passing the ID along with the
"hwconfig"), and which is probed separately.
Note that VAConfigID actually refers to a dynamic instance of something,
and creating a VAConfigID takes not only the VAProfile and the
VAEntrypoint, but also an arbitrary attribute array. In theory, this
means our attempt to get to know all possible configurations cannot
work, but in practice this attribute array seems to be pointless for
decoding and video processing, and FFmpeg doesn't use it (though the
encoding path does use it). This probably just makes it _barely_ OK to
do it this way.
Could we discard all this probing shit, and somehow do it another way?
Probably not. The EGL API for mapping surfaces doesn't even seem to
provide a way to enumerate supported formats, we may not even know
whether DRM/dmabuf interop is actually supported (AFAIR the EGL
extensions are present even if they don't work), nor do we know whether
the VAAPI driver supports this interop (not sure). So actually trying is
the only way.
Further, mpv initializes the decoder on a another thread, where you
can't just access OpenGL state. This suckage is mostly to be blamed on
OpenGL itself and its crazy thread boundedness. In theory, this could be
done anyway (see how software decoding "direct rendering" tries to get
around this). But to make it worse, the decoder never cares about the
list of supported formats determined by this code; instead,
f_autoconvert.c tries to deal with it and insert a video processor
(well, good luck with this crap, I bet it doesn't even work). So this
whole endeavor might be pointless, other than the fact that failed
probing can disable use of vaapi (which is correct and necessary). But
if you have a shovel, you don't use it to smash the flat end on the heap
of shit that's piled up before you, or do you?
While this method probably works, it's still orgasmically tedious. It
was tedious before: we had to create a real surface, create a GL
texture, map the surface with it, then destroy everything again. But the
added code is tedious on its own. Highlights include the need to malloc
a FFmpeg struct just to pass a single damn integer, the need to
enumerate "entrypoints" for each VA profile, even though all profiles
have exactly 1 entrypoint, and the kind of obnoxious way how vaapi
requires you to preallocate arrays for returned things, even they could
for example reasonably be returned as immutable arrays or have some
other simpler API.
The main grand fuckup is of course that vaapi requires a VAConfigID to
query surface properties, but not for creating surfaces. This
awkwardness even affected the FFmpeg API design, which has a "hwconfig"
concept that is only used by vaapi (vaapi is only 1 out of 10 hardware
decoding APIs supported by the FFmpeg hwcontext stuff). Maybe I'm just
missing something. It's as if vaapi required setting radioactive shit on
fire. Look how clean the native D3D11 code is instead. (Even the ANGLE
code manages to avoid being this fucked up. Or the VDPAU code, despite
supporting multiple mapping methods.)
Another only barely related change is that the valid_sw_formats field
can be NULL, and the API explicitly documents this. Technically, the mpv
code was buggy for not checking this, although until now the FFmpeg
implementation so far could not return it when we still passed NULL for
the hwconfig parameter.
No functional changes, just preparation for the next commit. Split the
probing into multiple functions. Prepare for the yet unused possibility
to pass AVVAAPIHWConfig to probing. try_format_pixfmt() now assumes it
can be called multiple times with the same format, so it filters the
format.
The format probing is now something like O(n^2) for n formats, but n
will most likely remain something under 50 or so.
Useless garbage.
This was once added to test whether vdpau presentation feedback could be
used. Results were always unsatisfactory, and now vdpau is dead.
Semantics a bit questionable. This is done for the OSC (next commit),
and a comment added the manpage explicitly states this. Meaning this is
probably garbage and needs to revisit when the OSC changes and/or
someone wants to use this margin feature for something else.
Not sure about the subtitle thing. It's imaginable that someone uses
these options to create empty borders for subtitles on the bottom, so
subtitles should be located there. On the other hand, this gives a
rather unpolished user experience when using the (later added) OSC
feature to not overlap with the video. There's not much of a point if
the OSC still overlaps the video. However, I'm too lazy to think about
this, so it stays like it is.
--video-margin-ratio-left=0.2 --video-margin-ratio-right=0.9 (added in
the the next commit) will set f_w to inf, resulting in some garbage
being propagated. Later, the OSD margins are computed from values before
various sanity clamping is applied, which makes libass suffer from
bullshit values.
I'm very sure it's OK and more correct to compute the OSD margins using
the later values, but I'm not sure about that.
This one is probably not terribly obvious from just the valgrind log,
but a wayland dev explained it to me just a second ago. Whenever mpv
sends events to the screen with wl_display_dispatch, wayland internally
allocates memory to a struct wl_proxy object if a new id is found. Quite
a few more things happen to that proxy object, but eventually mpv stores
the data on the client-side in a wrapper type of struct (struct
wl_data_offer). mpv's data_device_listener keeps track of those proxies
and frees the memory when appropriate. Of course, mpv is constantly
sending events to the screen and does so until the user quits the
player. What happens here is that one final wl_display_dispatch is called
right before the user quits the player and before mpv's
data_device_listener can handle that object. So the result is that you
always have one extra dangling proxy that doesn't get properly freed.
The solution is to just simply call wl_data_offer_destroy before closing
the wl_display to free that final dangling wl_proxy.
Extending the client-allocated mpv_opengl_drm_params struct
constituted a break of ABI that could cause UB.
Create a clean break by deprecating "drm_params" and related structs
and enum values, and replacing it with "drm_params_v2".
Also fix some comments and code that wrongly assumed that open could
return any other negative number than -1 for failure.
This commit updates the libmpv version to 1.104
Like hwdec_cuda, you get a big #ifdef mess if you try and keep the
OpenGL and Vulkan interops in the same file. So, I've refactored
them into separate files in a similar way.
Originally, vo_gpu/vo_opengl considered the case of Nvidia proprietary
drivers, which required vdpau/GLX, and Intel open source drivers, which
require vaapi/EGL. Since window creation and GPU context creation are
inseparable in mpv's internal API, it had to pick the correct API very
early, or hardware decoding wouldn't work. "x11probe" was introduced for
this reason. It created a GLX context (without showing the window yet),
and checked whether vdpau was available. If yes, it used GLX, if not, it
continued probing x11/EGL. (Obviously it couldn't always fail on GLX
without vdpau, which is why it was a separate "probe" backend.)
Years passed, and now the situation is different. Vdpau is dead. Nvidia
drivers and libavcodec now provide CUDA interop, which requires EGL, and
fixes some of the vdpau problems. AMD drivers now provide vaapi, which
generally works better than vdpau. Intel didn't change.
In particular, vaapi provides working HEVC Main10 support. In theory, it
should work on vdpau too, with quality reduction (no 10 bit surfaces),
but I couldn't get it to work.
So always prefer EGL. And suddenly hardware decoding works. This is
actually rather important, because HEVC is unfortunately on the rise,
despite shitty encoders and unoptimized decoders. The latter may mean
that hardware decoding works better than libavcodec.
This should have been done a long, long time ago.
In some cases, src.sig_peak remains undefined as 0, which was definitely
the case when using the OSD, since it never got passed through the usual
color space normalization process. Most robust work-around is to simply
force the normalization at the site where it's needed. This ensures this
value is always valid and defined, to make the peak-dependent logic in
these two functions always work.
Fixes 4b25ec3a9d
Fixes#6917Fixes#6918
Mesa supports the EGL_CHROMIUM_sync_control extension, and it's
available out of the box with AMD drivers. In practice, this is exactly
the same as GLX_OML_sync_control, but for EGL. The extension
specification is separate from the GLX one though, and buried somewhere
in the Chromium code.
This appears to work, although I don't know if it really works.
In theory, this could be useful for other EGL targets. Support code for
it could have been added to egl_helpers.c to avoid some minor duplicated
glue code if another EGL target were to provide this extension. I didn't
bother with that. ANGLE on Windows can't support it, because the
extension spec. explicitly requires POSIX timers. ANGLE on Linux/OSX is
actively harmful for mpv and hopefully won't ever use it. Wayland uses
EGL, but has its own fancy presentation feedback stuff (and besides, I
don't think basic video player functionality works on Wayland at all).
context_drm_egl maybe? But I think DRM has its own stuff.
So the next commit can make EGL use it. EGL has a quite similar
function, that practically works the same. Although it's relatively
trivial, it's still tricky, and probably shouldn't end up as duplicated
code.
There are no functional changes, except initialization, and how failure
of the glXGetSyncValues call is handled. Also, some comments mention the
EGL extension.
Note that there's no intention for this code to handle anything else
than the very specific OML sync extension (and its EGL equivalent). This
is just too weirdly specific to the weird idiosyncrasies of the
extension, and it makes no sense to extend it to handle anything else.
(Such as Wayland or DXGI presentation feedback.)
In the past, src peak was always equal to or higher than dst peak. But
since `--target-peak` got introduced, this could no longer be the case.
This leads to an incorrect result (scaling for peak mismatch in gamma
light) unless some other option (CMS, --linear-scaling, etc.) forces the
linearization.
Fixes#6533
We collect a 'vulkan-device' option today but then don't actually
pass it on, so it's useless. Once that's fixed, it can be used
to select a specific vulkan device by name.
Tested with the new nvidia offload feature to select between the
nvidia and intel GPUs.
Somehow I got the idea that compound literals had function-scoped
lifetime. Instead, like all other objects with automatic storage
duration, compound literals are block-scoped, so they become invalid
after exiting the block they were declared in. It seems like a recent
change to GCC actually reuses the memory that the compound literals
used to occupy, which was causing a few bugs.
The pattern of conditionally assigning a pointer to a compound literal
was used in a few places in ra_d3d11 where the Direct3D API expects
either a pointer to an initialised struct or NULL. Change these to
ensure the lifetime of the struct includes the API call.
Should fix#6775.
This is documented as required (although we did not do it in
the old GL codepath, with no visible problems) and I have seen
transient artifacts after seeking which _appear_ to have gone
away after introducing this.
