Mainly meant to apply simple VapourSynth filters to video at runtime.
This has various restrictions, which are listed in the manpage.
Additionally, this actually copies video frames when converting frame
references from mpv to VapourSynth, and a second time when going from
VapourSynth to mpv. This is inefficient and could probably be easily
improved. But for now, this is simpler, and in fact I'm not sure if
we even can references VapourSynth frames after the core has been
destroyed.
The stats were retrieved and written on every encode call, instead of
every encode call that actually returned a packet. ffmpeg.c also does it
this way, so it must be "more correct". Fixes 2-pass encoding.
We needed this because the OSD rendering path used GBRP for RGB
rendering, and not all swscale versions supported this conversion. But
recently we've dropped support for very old ffmpeg/libav versions, so
this isn't needed anymore.
This might be a good idea in order to prevent queuing a frame too far in
the future (causing apparent freezing of the video display), or dropping
an infinite number of frames (also apparent as freezing).
I think at this point this is most of what we can do if the vdpau time
source is unreliable (like with Mesa). There are still inherent race
conditions which can't be fixed.
The strange thing about this code was the shift parameter of the
prev_vs2 function. The parameter is used to handle timestamps before the
last vsync, since the % operator handles negative values incorrectly.
Most callers set shift to 0, and _usually_ pass a timestamp after the
last vsync. One caller sets it to 16, and can pass a timestamp before
the last timestamp.
The mystery is why prev_vs2 doesn't just compensate for the % operator
semantics in the most simple way: if the result of the operator is
negative, add the divisor to it. Instead, it adds a huge value to it
(how huge is influenced by shift). If shift is 0, the result of the
function will not be aligned to vsyncs.
I have no idea why it was written in this way. Were there concerns about
certain numeric overflows that could happen in the calculations? But I
can't think of any (the difference between ts and vc->recent_vsync_time
is usually not that huge). Or is there something more clever about it,
which is important for the timing code? I can't think of anything
either.
So scrap it and simplify it.
vo_vdpau used a somewhat complicated and fragile mechanism to convert
the vdpau time to internal mpv time. This was fragile as in it couldn't
deal well with Mesa's (apparently) random timestamps, which can change
the base offset in multiple situations. It can happen when moving the
mpv window to a different screen, and somehow it also happens when
pausing the player.
It seems this mechanism to synchronize the vdpau time is not actually
needed. There are only 2 places where sync_vdptime() is used (i.e.
returning the current vdpau time interpolated by system time).
The first call is for determining the PTS used to queue a frame. This
also uses convert_to_vdptime(). It's easily replaced by querying the
time directly, and adding the wait time to it (rel_pts_ns in the patch).
The second call is pretty odd: it updates the vdpau time a second time
in the same function. From what I can see, this can matter only if
update_presentation_queue_status() is very slow. I'm not sure what to
make out of this, because the call merely queries the presentation
queue. Just assume it isn't slow, and that we don't have to update the
time.
Another potential issue with this is that we call VdpPresentationQueueGetTime()
every frame now, instead of every 5 seconds and interpolating the other
calls via system time. More over, this is per video frame (which can be
portantially dropped, and not per actually displayed frame. Assume this
doesn't matter.
This simplifies the code, and should make it more robust on Mesa. But
note that what Mesa does is obviously insane - this is one situation
where you really need a stable time source. There are still plenty of
race condition windows where things can go wrong, although this commit
should drastically reduce the possibility of this.
In my tests, everything worked well. But I have no access to a Mesa
system with vdpau, so it needs testing by others.
See github issues #520, #694, #695.
This commit adds support for automatic selection of color profiles based on
the display where mpv is initialized, and automatically changes the color
profile when display is changed or the profile itself is changed from
System Preferences.
@UliZappe was responsible with the testing and implementation of a lot of this
commit, including the original implementation of `cocoa_get_icc_profile_path`
(See #594).
Fixes#594
Reduce most dependencies on struct mp_csp_details, which was a bad first
attempt at dealing with colorspace stuff. Instead, consistently use
mp_image_params.
Code which retrieves colorspace matrices from csputils.c still uses this
type, though.
This is pretty obscure, so it didn't matter much. It still breaks
switching output levels at runtime, because the video output is not
reinitialized with the new params.
There were some bad interactions with the OSC.
For one, dragging the OSC bar, and then moving the mouse outside of the
OSC (while mouse button still held) would suddenly initiate window
dragging. This was because win_drag_button1_down was not reset when
sending a normal mouse event, which means the window dragging code can
become active even after we've basically decided that the preceding
click didn't initiate window dragging.
Second, dragging the window and clicking on the OSC bar after that did
nothing. This was because no mouse button up event was sent to the core,
even though a mouse down event was sent. So make sure the key state is
erased with MP_INPUT_RELEASE_ALL.
We don't check whether the WM supports _NET_WM_MOVERESIZE_MOVE, but
if it doesn't, nothing bad happens. There might be a race condition
when pressing a button, and then moving the mouse and releasing the
button at the same time; then the WM might get the message to initiate
moving the window after the mouse button has been released, in which
case the result will probably be annoying. This could possibly be fixed
by sending _NET_WM_MOVERESIZE_CANCEL on button release, but on the
other hand, we probably won't receive a button release event in this
situation, so ignore this problem.
The dragging is initiated only when moving the mouse pointer after a
click in order to reduce annoying behavior when the user is e.g.
doubleclicking.
Closes#608.
VAAPI has some ambiguous image formats, like VA_FOURCC_I420,
VA_FOURCC_IYUV, VA_FOURCC_YV12 (the latter exactly the same as the first
two, just with swapped planes). There is potentially a problem when one
specific VAAPI format was picked, and converting it to a mpv format and
back to a VAAPI FourCC would result in a numerically different format
(even if it's actually the same). Then it could e.g. happen that
functions like va_surface_upload() reallocate the underlying VAImage,
which would be inefficient. Change the code so that this can't happen.
(Probably not a problem in practice with the current VAAPI usage.)
It's not really needed to be public. Other code can just use mp_image.
The only disadvantage is that the other code needs to call an accessor
to get the VASurfaceID.
Although I at first thought it would be better to have a separate
implementation for hwaccels because the difference to software images
are too large, it turns out you can actually save some code with it.
Note that the old implementation had a small memory management bug. This
got painted over in commit 269c1e1, but is hereby solved properly.
Also note that I couldn't test vf_vavpp.c (due to lack of hardware), and
I hope I didn't accidentally break it.
The plan is to get rid of the custom VAAPI and possibly VDPAU surface
allocators.
Add custom surface allocation, because hwaccel surfaces are allocated
completely differently from software surfaces.
Add optional LRU allocation, which is (probably) helpful for hwaccel,
but (probably) less optimal for software surfaces.
mp_image_pool_get_no_alloc() is specifically for VAAPI, which can't
allocate new decoder surfaces after decoder init.
They were used by ancient libavcodec versions. This also removes the
need to distinguish vdpau image formats at all (since there is only
one), and some code can be simplified.
Image formats used to be FourCCs, so unsigned int was better. But now
it's annoying and the only difference is that unsigned int is more to
type than int.
Instead of doing it on every seek (libavcodec calls get_format on every
seek), reinitialize the decoder only if the video resolution changes.
Note that this may be relatively naive, since we e.g. (or: in
particular) don't check for profile changes. But it's not worse than the
state before the get_format change, and at least it paints over the
current vaapi breakage (issue #646).
This follows the (only slowly progressing) plan to replace all internal
video filters with libavfilter.
All what's left in vf_gradfun.c is the weird wrapper around vf_lavfi.c.
This "sometimes" crashed when seeking. The fault apparently lies in
libavcodec: the decoder returns an unreferenced frame! This is
completely insane, but somehow I'm apparently still expected to
work this around. As a reaction, I will drop Libav 9 support in the
next commit. (While this commit will go into release/0.3.)
The window doesn't recieve a WM_LBUTTONUP message after it's dragged,
probably because it's swallowed by the modal loop. To stop the button
from sticking, release it manually when the drag is complete.
Mouse buttons can get stuck down if the button is pressed inside the
video window and released outside. Avoid this by capturing mouse input
when a button is pressed.
Apparently the "right" place to initialize the hardware decoder is in
the libavcodec get_format callback.
This doesn't change vda.c and vdpau_old.c, because I don't have OSX, and
vdpau_old.c is probably going to be removed soon (if Libav ever manages
to release Libav 10). So for now the init_decoder callback added with
this commit is optional.
This also means vdpau.c and vaapi.c don't have to manage and check the
image parameters anymore.
This change is probably needed for when libavcodec VDA supports gets a
new iteration of its API.
This updates the logic for the new, somewhat unified behavior of SRGB
and 3DLUT since 34bf9be (not that it was particularly correct even that
change) and checks for the presence of corresponding extensions only in
the cases in which they're needed.
This commit:
- Changes some of the #define and variable names for clarification and
adds comments where appropriate.
- Unifies :srgb and :icc-profile, making them fit into the same step of
the decoding process and removing the weird interactions between both
of them.
- Makes :icc-profile take precedence over :srgb (to significantly reduce
the number of confusing and useless special cases)
- Moves BT709 decompanding (approximate or actual) to the shader in all
cases, making it happen before upscaling (instead of the old 0.45
gamma function). This is the simpler and more proper way to do it.
- Enables the approx gamma function to work with :srgb as well due to
this (since they now share the gamma expansion code).
- Renames :icc-approx-gamma to :approx-gamma since it is no longer tied
to the ICC options or LittleCMS.
- Uses gamma 2.4 as input space for the actual 3DLUT, this is now a
pretty arbitrary factor but I picked 2.4 mainly because a higher pure
power value here seems to produce visually better results with wide
gamut profiles, rather then the previous 1.95 or BT.709.
- Adds the input gamma space to the 3dlut cache header in case we change
it more in the future, or even make it user customizable (though I
don't see why the latter would really be necessary).
- Fixes the OSD's gamma when using :srgb, which was previously still
using the old (0.45) approximation in all cases.
- Updates documentation on :srgb, it was still mentioning the old
behavior from circa a year ago.
This commit should serve to both open up and make the CMS/shader code much
more accessible and less confusing/error-prone and simultaneously also
improve the performance of 3DLUTs with wide gamut color spaces.
I would liked to have made it more modular but almost all of these
changes are interdependent, save for the documentation updates.
Note: Right now, the "3DLUT takes precedence over SRGB" logic is just
coded into gl_lcms.c's compile_shaders function. Ideally, this should be
done earlier, when parsing the options (by overriding the actual
opts.srgb flag) and output a warning to the user.
Note: I'm not sure how well this works together with real-world
subtitles that may need to be color corrected as well. I'm not sure
whether :approx-gamma needs to apply to subtitles as well. I'll need to
test this on proper files later.
Note: As of now, linear light scaling is still intrinsically tied to
either :srgb or :icc-profile. It would be thinkable to have this as an
extra option, :linear-scaling or similar, that could be used with or
without the two color management options.
Since the AO will run in a thread, and there's lots of shared state with
encoding, we have to add locking.
One case this doesn't handle correctly are the encode_lavc_available()
calls in ao_lavc.c and vo_lavc.c. They don't do much (and usually only
to protect against doing --ao=lavc with normal playback), and changing
it would be a bit messy. So just leave them.
The previous version of the gamma suboption was pretty useless. It could
be used to disable delayed gamma enabling, which is a mechanism to avoid
having to adjust gamma in the shader by default.
Repurpose the suboption and allow setting an exact gamma value with it.
You can already override gamma with the --gamma option as well as the
gamma input property, but these use a weird curve to create the
impression of a linear perceived brightness change when changing the
value. This suboption now allows setting an exact gamma value.
This used to be absolute colorimetric, but relative colorimetric is a
saner default due to the arguments presented in issue #595.
A short summary: In general it doesn't affect much because our eyes
adapt to the white point either way, but if running in windowed mode it
would make the whites seem inconsistent/tinted. For fullscreen
projection it's also undesirable since it reduces the dynamic range
without much benefit (again, since our eyes adapt either way) and it
also breaks calibration against ambient lighting.
This shouldn't change much, since most profile types that aren't 3DLUTs
aren't capable of either of those transforms, and most displays are
calibrated against D65 (same as BT.709 source) either way.
This uses the value of 1.95 as an approximation for the exact gamma
curve, which replicates the behavior of popular video software including
anything in the Apple ecosystem, as per issue #534.
This is the same issue as addressed by 257d9f1, except this time for
the :srgb option as well. (257d9f1 only addressed :icc-profile)
The conditions of the srgb_compand mix() call are also flipped to
prevent an off-by-one error.
I was unhappy with the old way of handling buffers, especially resizing. But my
original plan to use wl_shm_pool_resize wasn't as good as I initially thought.
I might get back to it.
With the new buffer pools it now possible to select triple buffering. Also the
buffer pools are also needed for the upcoming subsurfaces for osd and subtitles.
I hope this change was worth it.
I could not see any difference whatsoever, but for usage with a 3DLUT
there's zero performance difference so we might as well follow the spec to
the letter.
Legacy GL context creation (glXCreateContext) explicitly requires a X
visual, while the modern one (glXCreateContextAttribsARB) does not for
some reason. So fail only on the legacy code path if we don't find a
visual. Note that vo_x11_config_vo_window() will select a default visual
if a NULL visual is passed to it.
This fixes issue #504. For some reason, glXChooseFBConfig() will return
a fbconfig with no associated visual. (I'm not sure if this allowed.
They don't always have a visual, but since GLX_X_RENDERABLE is set
and GLX_DRAWABLE_TYPE is (implicitly) set to GLX_WINDOW_BIT, why would
there be no visual?)
Even worse, a test program seems to show that a 16 bit fbconfig is
selected (instead of 24/32 bit), which doesn't sound nice at all. Since
there _are_ better fbconfigs available, glXChooseFBConfig() should
normally sort them by quality, and return the better ones first. It's
worth noting that this function should also prefer GLX_TRUE_COLOR
over anything else, although this comes last in the sort order.
Whatever is going on, requesting GLX_X_VISUAL_TYPE with GLX_TRUE_COLOR
seems to fix it.
This was done incorrectly in the previous commit: the fallback size used
the window size as requested with the first config call, which is the
size of the hidden window in the vo_opengl case. (That damn hidden
window again...)
This code essentially does nothing. As far as I could find out, this
actually used to do something. Then it was removed with commit efe7c39f,
leaving some leftover code that didn't do anything useful. This happened
12 years ago!
Also remove a commented debug printf.
vo_opengl creates a hidden X11 window to probe the OpenGL context. It
must do that before creating a visible window, because VO creation and
VO config are separate phases.
There's a race condition involving the hidden window: when starting with
--fs, and then leaving fullscreen, the unfullscreened window is
sometimes set to the aspect ratio of the hidden window. I'm not sure why
the window size itself uses the correct size (but corrupted by the wrong
aspect), but that's perhaps because the window manager is free to ignore
the size hint while honoring the aspect, or something equally messed up.
It turns out this happens because x11_common.c thinks the size of the
hidden window is the size of the unfullscreened window. This in turn
happens because vo_x11_update_geometry() reads the size of the hidden
window when called in vo_x11_fullscreen() (called from
vo_x11_config_vo_window()) when mapping the fullscreen window. At that
point, the window could be mapped, but not necessarily. If it's not
mapped, it will get the size of the unfullscreened window... I think.
One could fix this by actively waiting until the window is mapped. Try
to pick a less hacky approach instead, and never read the window size
until MapNotify is received.
vo_x11_create_window() needs a hack, because we'd possibly set the VO's
size to 0, resulting e.g. in vdpau to fail initialization. (It'll print
error messages until a proper resize is received.)
Larger sizes can introduce overflows, depending on the image format. In
the worst case, something larger than 16000x16000 with 8 bytes per pixel
will overflow 31 bits.
Maybe there should be a proper failure path instead of a hard crash, but
not yet. I imagine anything that sets a higher image size than a known
working size should be forced to call a function to check the size (much
like in ffmpeg/libavutil).
RGB565 is one of the fastest and most supported formats on low end consumer
devices, but ffmpeg spams warning when using it. Make it opt-in instead of
opt-out.
The problem seems to have solved itself. I guess the previous changes to
resizing and commit ba101ab made this possible. Consider me happy for removing
that crap.
Still untested, because now it crashes inside of libSDL for unknown
reasons. (This also happens with mpv git from yesterday - probably an
installation problem, or SDL doing weird things it shouldn't be doing.)
The main difference between the old and new callbacks is that the old
callbacks required passing the window size, which is and always was very
inconvenient and confusing, since the window size is already in
vo->dwidth and vo->dheight.
Rename vo_get_src_dst_rects() to mp_get_src_dst_rects() and make it
independent from the VO (it takes a comical amount of parameters now to
pass all required state). Add a convenience wrapper with the name
vo_get_src_dst_rects() to vo.c. Replace all aspdat and vo usages with
immediate parameters.
Functionally, nothing should change, except that the window size is
clamped to a minimum of size 1 much earlier, and some log messages
change the prefix (don't bother with vo.vo_log stuff).
The plan is to make all the code in aspect.c independent from vo.c,
which should make the code easier to understand, will allow removal of
vo->aspdat, and reduces the amount of code that accesses weird mutable
struct vo fields.
vo->aspdat is basically an outdated version of vo->params, plus some
weirdness. Get rid of it, which will allow further cleanups and which
will make multithreading easier (less state to care about).
Also, simplify some VO code by using mp_image_set_attributes() instead
of caring about display size, colorspace, etc. manually. Add the
function osd_res_from_image_params(), which is often needed in the case
OSD renders into an image.
Do two things:
1. add locking to struct osd_state
2. make struct osd_state opaque
While 1. is somewhat simple, 2. is quite horrible. Lots of code accesses
lots of osd_state (and osd_object) members. To make sure everything is
accessed synchronously, I prefer making osd_state opaque, even if it
means adding pretty dumb accessors.
All of this is meant to allow running VO in their own threads.
Eventually, VOs will request OSD on their own, which means osd_state
will be accessed from foreign threads.
This is a bit of a hack, but in order to prevent TranslateMessage from
seeing WM_KEYDOWN messages that we already know how to decode, move the
decoding logic to the event loop. This should fix#476, since it stops
the generation of extraneous WM_CHAR messages that were triggering more
than one action on keydown.
Doesn't make any sense anymore. X11 (which was mentioned in the manpage)
autodetects it, and everything else ignored the option values.
Since for incomprehensible reasons the backends and vo.c still need to
exchange information about the screensize using the option fields,
they're not removed yet.
For some reason, this made all VO backends both set the screen
resolution in opts->screenwidth/height, and call
aspect_save_screenres(). Remove the latter. Move the code to calculate
the PAR-corrected window size from aspect.c to vo.c, and make it so that
the monitor PAR is recalculated when it makes sense.
When using --monitoraspect, but either the screen width or height or
both are unknown, a fallback is applied. This is a completely useless
obscure corner case that's going to help nobody, so get it out of the
way.
For a long time the cocoa backend set the xinerama_x/y and used dx/dy from the
VO instance. This somewhat worked with some workarounds but wasn't really
what was supposed to be happening. Moreover 27e4360, which touched this
workaround introduced a regression.
New code doesn't set the xinerama_x/y values so that dx/dy are offsets in the
current screen (not a virtual screen composed of all the screens). The screen
reference detected during VOCTRL_UPDATE_SCREENINFO is also passed down to the
window initialization code.
Fixes#472
Like with the previous commit, this is probably not needed, but it's
unclear whether that really is the case. Most likely, it used to be
needed by some demuxer, and now the only demuxer left that could
_possibly_ trigger this is demux_mkv.c.
Note that mjpeg is the only decoder that reads the extra_huff option,
and nothing in libavformat actually sets the option. So maybe it's
fundamentally not needed anymore.
This case can't happen with the normal realvideo codepath in
demux_mkv.c, because the code would errors out if the extradata is too
small, and everything would be broken anyway in the case the vd_lavc.c
condition is actually triggered.
It still might happen with VfW-muxed realvideo in Matroska, though.
Basically, I'm hoping this doesn't matter anyway, and that the vd_lavc.c
code was for other old demuxers, like demux_avi or demux_rm. Following
the commit history, it's not really clear for what demuxer this code
was added.
On X11, if no wayland compositor is running, wl_list_init() will never
be called. This will cause destroy_display() to segfault when trying to
iterate over the list.
The user_data is passed on add_listener and can later be changed with
set_user_data. But because we don't want to change it later and because it is
the same object remove the set_user_data call.
This might be a copy&paste leftover from the initial draft for the wayland
backend.
I added enough logic to never set ontop or fullscreen twitce.
This commit keeps also the size of the video if multiple videos are played.
If the ratio differs the width will be kept at the same size and only the
height changes.
libwayland-client contains the following code [1]:
runtime_dir = getenv("XDG_RUNTIME_DIR");
if (!runtime_dir) {
fprintf(stderr,
"error: XDG_RUNTIME_DIR not set in the environment.\n");
This means this message will unconditionally and unavoidably be printed
if XDG_RUNTIME_DIR is not set. Since mpv is a terminal program, and we
want to avoid unnecessary output, work it around by not attempting to
use wayland if this environment variable is not set.
[1] http://cgit.freedesktop.org/wayland/wayland/tree/src/wayland-client.c#n636
(cd0dccd01e16fa404e03974d30ded3aebdb1c4bc)
This commonly happens when initializing vo_opengl on a X11-only system.
Unfortunately, most wl_*_destroy() functions appear not to accept NULL
pointers, making partial deinitialization a pain: you have to add your
own NULL checks everywhere to avoid crashes.
xkb.context is uninitialized separately, because you can initialize it
just fine, even if the rest of input initialization fails.
Because of this commit there were problems displaying the frmase in their right
order.
This reverts commit 96e75d234a.
Conflicts:
video/out/gl_wayland.c
video/out/wayland_common.h
The changes in the vo_wayland_ontop function have no effect on the workaround.
Somehow the problem just disappeared. I guess it is because of the new control
function in gl_wayland.c where the resize happens immediatly after the event
dispatch/flush.
Both X11 and Wayland support the same format for drag & drop operations
(text/uri-list), and the code for that was copied from x11_common.c to
wayland_common.c. Factor it out.
This solves the issue where we would not receive any frame events. The
difference to my earlier tests is that now it looks like eglSwapBuffers uses
it's own event queue or something similiar along the lines. Becaues the
performance is the same as without any redraw callback.
At the moment there are visual glitches when we resize the window. This happens
because in wayland there a special function for resizing EGL windows.
To prevent the glitches move the egl_context to the wayland state in
wayland_common.h and add a new control function to gl_wayland.c to wrap the
vo_wayland_control function to check for resize events.
With the new control wrapper the glitches are gone and the resizing is fluid.
The reason a segmentation happend here was because we couldn't get the
requested minor version. The major version is enough for differentiating
between OpenGL 3 and OpenGL 2. If it fails there is still a fallback to any
version available.
Also add a warning if we use the fallback.
Note that we don't try to be clever about detecting the files as
subtitles: we just check the file extension. We could go all the way and
check the files by opening them with a demuxer, but that would probably
do more bad than good.
Drag and drop is pretty complicated (just note how the number of Atoms
in use almost doubles), so I'm not sure whether this works everywhere.
This has been written by looking at the specification [1] or what looks
like the specification, and some external example code [2]. (The latter
one has no code license, but we didn't copy any code.)
We completely ignore the "requirement" of the spec. that the filename
"must" include username and hostname, e.g. "file://user@host/path/file".
In theory, this is required because X is network transparent, but at
this point the so called network transparency is a complete joke, and
Konqueror for one didn't include hostnames in "file://" URIs.
Tested with konqueror as drop source.
[1] http://www.newplanetsoftware.com/xdnd/
[2] http://www.edwardrosten.com/code/dist/x_clipboard-1.1/paste.cc
Set the flag CODEC_FLAG_OUTPUT_CORRUPT by default. Note that there is
also CODEC_FLAG2_SHOW_ALL, which is older, but this seems to be ffmpeg
only.
Note that whether you want this enabled depends on the user. Some might
prefer that only good frames are output, while others want the decoder
to try as hard as possible to output _anything_. Since mplayer/mpv is
rather the kind of player that tries hard instead of being "clever", set
the new default to override libavcodec's default.
A nice way to test this is switching video tracks. Since mpv doesn't
wait for the next key frame, it'll start feeding the decoder with a
packet from the middle of the stream.
