This change prevents unwanted adjustments. Generally, screenshots
shouldn't invoke pl_queue_update, as this action could cull the already
mapped frames in the queue.
If !display_synced, some values may not be correct or zeroed. Therefore,
it makes no sense to interpolate in this case.
For a non-moving frame, we always want to show an uninterpolated frame.
libplacebo requires additional frames when VPS is lower than FPS and
anti-aliasing is enabled. Supports up to a 1/4 ratio. VO has a limit of
10 frames, so in practice, not many more can fit.
Note that the internal libplacebo mixing limit is 16 frames.
The `current_frame` can be redrawn even if the remaining `num_vsync` is
equal to 0. In this scenario, the vsync offset would be incremented
beyond the target point.
It turns out that mpv will rewrite the entire cache file on every single
quit. It's not so great since after viewing a ton of files, your cache
can grow to massive sizes and slow down quitting the player to a
noticeable amount. Turn down the max size of both caches to 10 MiB for
now as a workaround until this gets improved later.
This filters out vastly inaccurate values from presentation feedback
that can happen shortly after restarting playback or seeking.
Makes estimated vsync converge almost instantly instead of waiting
until those outliers are dropped from the past samples.
In commit de6d57f0 libplacebo enabled drift compensation by default.
mpv already tracks error and adjust the speed by itself.
This commit fixes judder visible when slowing video a lot, ex. playing
back at 10% speed.
Set ideal vsync duration also when dropping frames, libplacebo estimate
currently own values anyway, will be useful later...
See: de6d57f021
The commit subject sounds reasonable except that frame redraws get
spammed in certain cases (still image with subtitles) all the time
regardless if the subtitle actually has changed or not. So this is
stupid and wasteful, but you'll always see subtitles. vo_gpu currently
has this behavior as well, so we're just matching it but later mpv's
core should be fixed so this works reasonably. Fixes#11335.
Fixes gpu-next completely ignoring any speed adjustment, either DS or
playback.
Frames in pl_queue have raw PTS values, so when querying them we have to
use the same time base. There was misunderstanding between mpv and
libplacebo, where the former was querying the frames based on display
vblank timeline, but this cannot work if the playback speed is adjusted
and display timeline is not aligned with video timelien. In which case
we have to schedule "video vsync" points that we want to display.
Previous code was working only when playback speed was 1.0x, but since
DS almost always changes the speed the interpolation was mostly not
timied correctly.
Relative to frame PTS timeline as oposed to display vblank.
Those values are relative to unadjusted video timeline. They will be
used by gpu-next where it expect virtual frame vsync, not display vblank
time.
Originally suggested by @haasn. Instead of awkwardly using a pts from
the previous render loop, we can just peek into pl_queue and use the pts
of the first frame instead. This eliminates a lot of weird artifacts
that can happen on discontinuities like seeking. Fixes#12355.
Several related things in regards to interpolation. Essentially this
adds more general sanity checking to bring it more in line with what
vo_gpu does.
- Add a can_interpolate bool which determines whether or not this frame
is allowed to interpolate.
- p->is_interpolated was sometimes lying and because you can have a mix
frames greater than 1 depending on the video and settings. Make sure
that vsync_offset is not 0 so we know if it's actually interpolated or
not. This prevents a redundant redraw for those edge cases when
pausing.
- When the vo wants a reset, i.e. some sort of discontinuity, don't try
to interpolate the frame. Interpolation is only valid for
monotonically increasing times.
This fixes#11519 because of the additional check to make sure that we
have more than 1 frame. The PTS clamping part is still not great. That
is for the next commit.
In the past, this worked by accident because the initial startup was
racy and sometimes the initial firing of handle_toplevel_config would
happen after reconfig. Since we now properly wait on all compositor
events we can save the initial size hint that is given to us and try to
use that as the window-size/geometry provided the --autofit/geometry
options aren't explictly set. Fixes#11134.
the OpenGL cocoa backend was deprecated in 0.29, it has lot of bugs, is
completely unmaintained and can't properly playback anything anymore on
the newest macOS. it is time to remove it.
the old displayName property via the IODisplay API is not working
anymore on ARM based macs and was broken in at least one other case.
instead we use the new localizedName property introduced in 10.15 of the
NSScreen. we don't need any backwards compatibility since 10.15 is the
oldest version we support now.
configs and scripts that use the options and properties fs-screen-name,
screen-name or display-names need to be adjusted since the names could
differ from the previous implementation via the IODisplay API.
Fixes#9697
this is apparently something that was added with swift 5.5, though it is
hard to find anything officially.
don't capture self in closure but explicitly access all variables by
prepending self.
Fixes#12653
Fixes a899e14b which changed clamp from 0 to 1 ms which effectivelly
introduced 1ms sleep always, even if requested until_time_ns is in the
past and should request 0 timeout.
While at it also fix mp_poll wrapper to respect negative timeout which
should mean infinite wait.
Also keep the 37d6604 behaviour for very short timeouts, but round only
the ones > 100us, anything else is 0.
Fixes: a899e14b
Most importantly, the wait_until addition was missed while doing the
unit conversions to nanoseconds which meant mpv woke up roughly every
second since not nearly enough time was added. It was meant to be 1000
seconds (1e9 in microseconds). Use a macro so it's more readable. Also
put some other wild 1e9 calculations inside of a macro as well.
Fixes a899e14bcc.
These have been build options since the waf build, but that doesn't
really make sense. The build can detect whatever macOS sdk version is
available and then use that information to determine whether to enable
the features or not. Potentially disabling multiple sdk versions doesn't
really make any sense. Because f5ca11e12b
effectively made macOS 10.15 the minimum supported version, we can drop
all of these checks and bump the required sdk version to 10.15. The rest
of the build simplifies from there.
Currently VOCTRL are completely unusable for frequent data query. Since
the HDR parameter addition to video-params, the parameters can change
each frame. In which case observe on those parameter would be triggered
constantly. The problem is that quering those parameters involves VOCTRL
which in turn involves whole render cycle of delay.
Instead update VO params on each draw_frame. This requires changes to VO
reconfiguration condition, but in practice it should only be triggered
when image size or data layout changes. In other cases it will be
handled internal by VO driver.
I'm not quite happy with this solution, but don't see better one without
changing observe/notify logic significantly. There is no good way
currently to handle VOCTRL that are constantly queried.
This adds unfortunate synchronization of player command with VO thread,
but there is not way around that and if too frequent queries of this
param becomes a problem we can thing of other solutions.
Changes the way to get data from VO driver added by a98c5328dc
Fixes: 84de84bFixes: #12825
windowrc in vo_w32_state is actually client size, for hit test we need
proper window size. When border is disabled those sizes are the same,
but when only title bar is disabled it is not.
Reduce the hit area to more sane values when the border is not
drawn to minimize amount of covered client area in borderless mode.
This was specifically special logic for drm. Before present_sync, it
would also clear out all of its vsync values like this. The old drm code
would save a bunch of samples which would confuse vo.c when unpausing
since it got old, bogus values. Since we make sure to match successive
vsync samples with the swapchain depth and that present sync samples
also match the swapchain depth, this is unneeded.
There's some geometry-related things that mpv has to calculate before
the window is actually mapped onto the screen in wayland. But there's no
way to know which output the window will end up on before it happens, so
it's possible to calculate it using the wrong values. mpv corrects
itself later when the surface event happens, but making the initial
guess work better can help in certain cases.
find_output is the only thing that needs to be changed here. Its main
purpose is to grab the right output based on user settings when we're
trying to full screen and giving a fallback in case we don't have
wl->current_output yet. The x11 code already does something similar, so
we're basically just copying it. Allow user settings like --screen and
--screen-name to influence the initial wl_output guess. Those options
won't actually place the window on that specific screen since we can't
do that in wayland, but if the user knows where the window will end up
beforehand it makes sense to listen to the arguments they pass. If
something goes wrong, then we just fallback to 0 like before.
A ton of code and drm-specific abstractions can be dropped with this.
One important thing to note is that the usage of sbc is completely
dropped here. The utility of that is not particularly clear since the
sbc value was manually incremented before the flip and it's not as if
the drm page flip event gives it to us like it does with the msec and
ust. It can be reintroduced later if there is a need. For drm, we also
add a present_sync_clear_values helper since all presentation feedback
needs to be cleared on pause/resume for it.
When this was originally written, the queuing/list approach was
deliberately removed since it adds more complication and xorg/wayland
don't really use it anyway. In practice, you only really have one frame
in flight with presentation timestamps. However, one slight annoyance is
that the drm code has its own thing which is almost exactly the same and
does its own calculations. Ideally, we'd port drm to this instead, but
the implementation there takes into account N-frames in flight which
probably does actually work. So we need to make present_sync smarter and
be able to handle this.
mpv does actually have its own linked list implementation already which
is a good fit for this. mp_present becomes the list and each
mp_present_entry has its own set of timestamps. During initialization,
we create all the entries we need and then simply treat it like a queue
during the lifecycle of the VO. When an entry is fully used
(present_sync_get_info), then we remove it from the list, zero it out,
and append it to the end for future use. This avoids needing to allocate
memory on every frame (which is what drm currently does) and allows for
a reasonable number of in flight frames at the same time as this should
never grow to some obscene number. The nice thing is that current users
of present_sync don't need to change anything besides the initialization
step.
This commit replaces all uses of sig_peak and maps all HDR metadata.
Form notable changes mixed usage of maxCLL and max_luma is resolved and
not always max_luma is used which makes vo_gpu and vo_gpu_next behave
the same way.
This change essentially removes mp_thread_self() and instead add
mp_thread_id to track threads and have ability to query current thread
id during runtime.
This will be useful for upcoming win32 implementation, where accessing
thread handle is different than on pthreads. Greatly reduces complexity.
Otherweis locked map of tid <-> handle is required which is completely
unnecessary for all mpv use-cases.
Note that this is the mp_thread_id, not to confuse with system tid. For
example on threads-posix implementation it is simply pthread_t.
The old one would actually crash if the libplacebo.cache file was
invalid. Additionally, set a max size of 1 GiB for icc cache and 50 MiB
for gpu shader cache. The per object size limit is removed which puts
mpv in line with plplay. Finally, a few memory leaks are also fixed
since several objects previously were not freed on uninit.
I'd like some names to be more descriptive, but to work with 15 chars
limit we have to make some sacrifice.
Also because of the limit, remove the `mpv/` prefix and prioritize
actuall thread name.
ffmpeg is again setting the frame dimensions to the coded size, so we
need to reintroduce our work-around to get the logical frame dimensions
from the frames_ctx rather than the frame itself.
ffmpeg commit:
* 9ee4f47c94
This reverts commit c40bd88872.
Other similar options are in the form of --foo-override not
--override-foo. The display-fps one was backwards so flip it around the
other way for consistency reasons.
Peak detection greatly increases HDR experience. Performance hit of
non-delayed detection is not that significant and is in line with
current default settings.
Make it not possible to build mpv without the latest libplacebo anymore.
This will allow for less code duplication between mpv and libplacebo,
and in the future also let us delete legacy ifdefs and track libplacebo
better.
since i was going to fix the include order of stdatomic, might as well
sort the surrouding includes in accordance with the project's coding
style.
some headers can sometime require specific include order. standard
library headers usually don't. but mpv might "hack into" the standard
headers (e.g pthreads) so that complicates things a bit more.
hopefully nothing breaks. if it does, the style guide is to blame.
replace it with <stdatomic.h> and replace the mp_atomic_* typedefs with
explicit _Atomic qualified types.
also add missing config.h includes on some files.
Some users report visible black frames during window resize, this should
not happen in most cases. Let's just keep stale content as it is less
distracting. While still clearing on first window paint to avoid white
background.
Fixes: #12642
There's a lot of wild 1e6, 1000, etc. lying around in the code. A macro
is much easier to read and understand at a glance. Add some helpers for
this. We don't need to convert everything now but there's some simple
things that can be done so they are included in this commit.
add support for vulkan through metal and a translation layer like
MoltenVK. also add the possibility to use different render timing modes
for testing.
i still consider this experimental atm.
add an animation lock to the window to prevent the window from closing
while animating. if this is done while the fs animation is running the
dock will stay hidden. this is not used yet, because it's unnecessary
for cocoa-cb but will be for new vo backends.
Several window resizing operations (i.e., --auto-window-resize,
border dragging with --keep-aspect-window, Alt+0/1/2) cause the
window to detach from the snapped borders. This patch resolves
this by recording to which borders the window is snapped, and
using this info to determine how the window should be placed after
resizing.
Closes https://github.com/mpv-player/mpv/issues/6588
Linux and macOS already use nanosecond resolution for their sleep
functions. It was just being converted from microseconds before. Since
we have mp_time_ns now, go ahead and bump the precision here. The timer
for windows uses some timeBeginPeriod thing which I'm not sure what it
does really but whatever just convert the units to ms like they were
doing before. There's really no reason to keep the mp_sleep_us helper
around. A multiplication by 1000 is trivial and underlying OS clocks
have nanosecond precision.
On linux, several platforms poll for events over a fd. This has ms
accuracy, but mpv's timer is in ns now so lots of precision is lost. We
can use an mp_poll wrapper to use ppoll instead which takes a timespec
directly with nanosecond precision. On systems without ppoll this falls
back to old poll behavior. On wayland, we don't actually use this
because ppoll completely messes up the event loop for some unknown
reason.
In many cases, this is purely cosmetic because poll still only accepts
microseconds. There's still a gain here however since
pthread_cond_timedwait can take a realtime ts now.
Additionally, 37d6604d70 changed the value
added to timeout_ms in X11 and Wayland to ensure that it would never be
0 and rounded up. This was both incomplete, several other parts of the
player have this same problem like drm, and not really needed. Instead
the MPCLAMP is just adjusted to have a min of 1.
vo_still_displaying() is racey with vo_request_wakeup_on_done() and above
that it doesn't work as expected. VO can run out of work and go to sleep
for 1000s, while the play thread still returns on vo_still_displaying()
check, because of a check `now < frame_end` so it never advances and go
to sleep itself.
This fixes dead lock that we have when image parameters changes during
playback.
This reverts commit 0c9ac5835b.
Fixes: #12575
fbe154831a added a new VOCTRL to signal
when the OSD changed for gpu-next's handling of subtitles, but this is
both not necessary and actually incomplete. The VOCTRL would signal OSD
changes, but not all subtitle changes (like selecting another
non-external sub track for example). VOCTRL_OSD_CHANGED was used to
increment p->osd_sync which would then redraw the blended subtitles if
the player was paused.
But there's already a VOCTRL_PAUSE and VOCTRL_RESUME. Plus, the
sub_bitmap_list object will have items in it if it changed in any way,
so we don't need the VOCTRL_OSD_CHANGED method at all. That can be
removed.
The check that fp->osd_sync < p->osd_sync stays in place since that's an
optimization while the video is playing, but we also check the pause
state as well since the VO can know this. If we're paused, then always
do update_overlays since core must be signalling a redraw to us if we
get a draw_frame call here. Additionally in update_overlays itself, the
p->osd_sync counter is incremented if we have any items since the frame
signature will need that. As for the actual bug that is fixed, changing
subtitle tracks while paused with blended subtitles now correctly works.
Previously, it was never updated so the old subtitle stayed there
forever until you deselected it (since VOCTRL_OSD_CHANGED triggered
there).
Also include some cosmetic code fixes that were noticed.
The current calculation makes the implicit assumption that the margins
on both sides are always equal (the 2 times multiplication). This isn't
true though. For example, a 720p video fullscreened on a 1680x1050
display will have a top margin of 52 but a bottom margin of 53. The
current calculation just multiplies 52 by 2, so it's off by one. Fix
this by adding together all the margins together (left + right or top +
bottom) then adding that to the dst window size (width or height). This
way, we get the full size of the window for the viewport. Fixes#12554.
This fixes white background appearing for short period of time before
first frame is drawn. Clear to black as this is way less distracting
than bright white flash.
Borderless window and fullscreen seems to be initially not
drawn/transparent, so no need to clear it to black. Only when
decorations are enabled (--border) the issue happens.
Fixes: #12549
Forgotten in 7b8a30fc81. Every other case
either is either a dummy frame (no allocated memory) or we manage it as
part of the usual wayland buffer release.
When implementing vo_dmabuf_wayland, it always did a copy of the image
from the current frame and worked with that. The reason was because
mpv's core held onto the frame and caused some timing issues and
rendering glitches depending on when it freed the image. This is pretty
easy to fix: just make vo_dmabuf_wayland manage the the frames. In vo.h,
we add a boolean that a VO can set to make them manage freeing frames
directly. After doing this, change the buffers in vo_dmabuf_wayland to
store the whole vo_frame instead of just the image. Then, just modify
some things a bit so frame is freed instead of the image. Now, we should
truly have zero-copy playback. Well as long as you don't use libass to
render anything (that's still a copy from system memory).
e125da2096 changed the z order of the
surfaces a bit, but it turns out this has a side effect. If the aspect
ratio of the actual video doesn't match your display, the osd surface
doesn't scale properly and gets clipped. Put the z ordering back where
it used to be. Instead when we have the force window case, simply attach
the already existing solid buffer to the video surface. This allows the
osd surface to actually draw over it instead of always being obscured so
it satisfies the case of not having any real video frames but still
wanting to draw the osd. Also don't mess with any of the viewport source
setting stuff with force window. Weston complains about it, and it's
nonsensical anyway. Fixes#12547.
0739cfc209 added the draw_frame API
deprecated draw_image internally. VOs that still used draw_image were
around, but really there's no reason to not just "upgrade" them anyway.
draw_frame is what the "real" VOs that people care about (gpu/gpu-next)
use. So we can just simplfy the code a bit now. VOCTRL_REDRAW_FRAME is
also no longer needed so that can be completely deleted as well. Note
that several of these VOs are legacy crap anyway (e.g. vaapi) and maybe
should just be deleted but whatever. vo_direct3d was also completely
untested (not that anyone should ever use it).
This causes only problems, because we convert mp_time to realtime, which
is not atomic, so we introduce error. And even though on sane platforms
it should work fine, after all the sleep time is in the past.
winpthreads like to sleep for like over 10ms when the time is less than
current time, but not more than 1s.
In practice, most compositors implement the rotation clockwise which
matches mpv's option, but amusingly this is actually incorrect.
According to the spec*, wayland buffer rotations are counter-clockwise.
So with this assumption in mind, in order for the rotation to match
mpv's usual semantics, the 90 degree and 270 degree positions need to be
flipped. Of course, this will make the VO rotate the wrong way on most
compositors, but this is what the spec says (sway master is known to
currently be correct). Fixes#12508 (sort of but will break the rotation
direction on other compositors. Oh well).
*: https://wayland.freedesktop.org/docs/html/apa.html#protocol-spec-wl_output-enum-transform
NV16 is the half subsampled version of NV12 format. Decoders which
support High 4:2:2 of h264 provide the frame in NV16 format to establish
richer colorspace. Similar profiles are also available in HEVC and other
popular codecs. This commit allows NV16 frames to be displayed over
drmprime layers.
Signed-off-by: hbiyik <boogiepop@gmx.com>
Calling wl_display_disconnect closes the file descriptor, no need to
manually do it ourselves beforehand which causes a double close on the
fd.
Signed-off-by: Jack Mitchell <jack.mitchell@tuxable.co.uk>
Apparently removing WS_CAPTION disables some window animations. Instead
adjust non-client area to not draw title bar.
Note that we do not account for difference in real border size and
invisible one, but seems to work correctly.
I don't think in fullscreen mode it makes sense to enable rounded corners.
We can add another option if someone needs it, but for now `window_corners`
affects only the window as one would expect.
With the addition of fractional scaling support, wl->scaling was
converted to a double. Some compositors (Plasma) can report values under
1 for fractional scaling, so this meant wl->scaling could be some
small fractional value. This is fine except that when using the legacy
code for drawing the mouse cursor (i.e. not the cursor-shape protocol),
it still uses the old integer scaling method in core wayland. The reason
for this is simply because fractionally scaling the mouse cursor surface
is nonsense and nobody even has cursor images for anything besides a
select few sizes anyways (32x32, 48x48, etc.). The existing integer
scaling sort of works but it's pretty bad too and you can get some weird
sizes anyway. This is why cursor-shape is preferred since it fixes this.
Anyways, since buffer scaling for the cursor only takes integers, there
could be truncation to 0 in the previously mentioned fractional scale
this. This naturally causes the compositor to send us an error and mpv
quits. The fix is to always make sure that the scale value used for the
cursor is at least 1. Anything less makes no sense. Fixes#12309.
There was assumption in the code that default settings are compatible
with dumb mode and are only one that should be used in this case.
Force bilinear if dumb mode is enabled.
A bit different from the OPT_REPLACED/OPT_REMOVED ones in that the
options still possibly do something but they have a deprecation
message. Most of these are old and have no real usage. The only
potentially controversial ones are the removal of --oaffset and
--ovoffset which were deprecated years ago and seemingly have no real
replacement. There's a cryptic message about --audio-delay but who
knows. The less encoding mode code we have, the better so just chuck
it.
We've got an ungodly amount of OPT_REPLACED and OPT_REMOVED sitting
around in the code. This is harmless, but the vast majority of these are
ancient. 26f4f18c06 is the last commit
that touched the majority of these and of course that only changed how
options were declared so all of this stuff was deprecated even before
that. No use in keeping these, so just delete them all. As an aside,
there was actually a cocoa_opts but it had only a single option which
was replaced by something else and empty otherwise. So that entire thing
was just simply removed. OPT_REPLACED/OPT_REMOVED declarations that were
added in 0.35 or later were kept as is.
The osd support was originally written with the requirement that we have
actual frames getting delivered to the VO. This isn't always the case
though. If you force a window on a blank audio file for example, then
there will be no frame thus draw_frame did nothing. Since the previous
commit allows us to reliably detect this, we can rearrange the code
around a little bit to make this possible. A key change is to make the
osd_subsurface have wl->surface as the parent. This is seemingly
required otherwise the osd_surface buffers are never visible above the
empty video_surface when we have a black window. Also nuke the desync
call since it's completely pointless. Fixes#12429.
The defaults were awful and horribly regressed many files while also not
fixing banding on files that actually needed it, sometimes even
*increasing* banding due to the low threshold.
Fixes: 12ffce0f22
See-Also: haasn/libplacebo@e1e43376d1
This probably makes `vo_gpu` tone mapping worse, or something, but who
cares. The status quo for a while now has been to use `vo_gpu_next` if
you care about HDR rendering at all.
See-Also: haasn/libplacebo@ec60dd156b
See-Also: haasn/libplacebo@0903cbd05d
This new filter is slightly sharper, and significantly faster, than
mitchell. It also tends to preserve detail better. All in all, there is
no reason not to use it by default, especially from a performance PoV.
(In vo_gpu_next, hermite is implemented efficiently using hardware
accelerated bilinear interpolation)
See-Also: 75b3947b2c
The goal is to provide simple to understand quality/performance level
profiles for the users.
Instead of default and gpu-hq profile. There main profiles were added:
- fast: can run on any hardware
- default: balanced profile between quality and performance
- high-quality: out of the box high quality experience. Intended
mostly for dGPU.
Summary of three profiles, including default one:
[fast]
scale=bilinear
cscale=bilinear (implicit)
dscale=bilinear
dither=no
correct-downscaling=no
linear-downscaling=no
sigmoid-upscaling=no
hdr-compute-peak=no
[default] (implicit mpv defaults)
scale=lanczos
cscale=lanczos
dscale=mitchell
dither-depth=auto
correct-downscaling=yes
linear-downscaling=yes
sigmoid-upscaling=yes
hdr-compute-peak=yes
[high-quality] (inherits default options)
scale=ewa_lanczossharp
cscale=ewa_lanczossharp (implicit)
hdr-peak-percentile=99.995
hdr-contrast-recovery=0.30
allow-delayed-peak-detect=no
deband=yes
scaler-lut-size=8
296d40dc6f changed how the vo handled
redraw requests in order to fix a race condition that can occur with
pausing. However, there was a slight oversight because a redraw request
that occurred while the core was unlocked and the video was still
playing would still be kept true (previously, this was always cleared).
That redraw is essential if mpv is paused otherwise the old issue comes
back, but if the video is playing it's unnecessary since the next loop
around will simply draw whatever we needed. The extra redraw could cause
a frame drop for some people in certain instances, so the solution is to
simply always clear redraw requests if !in->paused. This eliminates the
extra redraw but still keeps it when pausing.
Fixes#12426 and fixes#11579.
After fixing the B and C params for bcspline, it ended up being the same
thing as bicubic. There's no reason to have two names for the same
filter, so remove bcspline and keep bicubic to match libplacebo.
All we have to do is set VO_CAP_ROTATE90 and then use buffer transform.
Maybe ideally we would rotate with vavpp instead if possible, but this
would be a convoluted mess that I'm not really interested in dealing
with right now.
No need for this since it's entirely redundant with just changing the
filter radius directly. In fact, that's the whole *point* of the filter
radius - it does not modify the filter, it modifies the scaling of the
window.
Of course, this does not work for non-resizable kernels. But, really,
who cares?
Replaced by mathematically and numerically precise constants, the result
of exact computation. Also updates the description to provide more
explanation and motivation.
We currently always scaled the window to the size of the configured
radius. However, this is wrong - we should instead be scaling it to the
size of the sharpened/blurred kernel. Since the window is always
stretched to the configured size of the filter, we can accomplish this
easily by just multiplying the blur value into the filter radius
directly, and then using that adjusted radius in place of `f.radius`
everywhere.
On a side note, this gives a very minor performance boost to
ewa_lanczossharp for no downside.
Upstream finally caved in to peer pressure and added this filter. Of
course, this also removes the fallback for people on older versions of
libplacebo, but people using mpv git master are probably using
libplacebo git master anyway. It's time to debloat this code.
And make it the default. In libplacebo, this uses internal heuristics to
pick a good size based on the actual ICC characteristics. This is
significantly less wasteful than always generating a 64x64x64 3DLUT (the
old status quo).
In vo_gpu, for simplicity, just default to 65x65x65. Note that this
provides slightly better accuracy than the old default of 64x64x64 for
technical reasons, and matches what libplacebo defaults to for typical
display profiles.
Upstream has moved from passing struct pl_icc_profile to directly
attaching a managed pl_icc_object, plus providing a new function
pl_icc_update to update the ICC profile object parameters (if needed).
To facilitate this move, pull our ICC params back out of pl_options and
update the target ICC object directly.
The original OML sync implementation (which is where this calculation
originally comes from) made now_monotonic and ust_mp_time unsigned. This
is fine except it has the assumption that now_monotonic is always
greater than ust. This actually isn't always the case. In wayland, I
observed instances where the reported timestamp is in the future. Of
course, it's a valid question to wonder if this even makes sense but
these UST values are essentially opaque black boxes to us anyways. It's
entirely plausible that the gpu is expecting the actual presentation of
the last swap to be a bit in the future, the compositor gets this and
reports this to us. So we'll consider such stats as valid. Note that
xorg doesn't have this problem because it's roughly one swap buffer call
behind because of how the event loop works (honestly something that
should be fixed).
Of course, the problem with the unsigned type here is that it overflows
on the subtraction so make it signed and allow the appropriate negative
value to happen if it does. Note that this will simply result in a small
addition to mp_time_us() which is exactly what we want here (timestamp
slightly in the future). Some small style changes included just because.
So strangely enough, estimated_vsync_interval is stored as a double but
nominal_vsync_interval is not and neither is the vsync_interval. Those
are stored as int64_t. This loss of precision can matter even in common
cases. For instance, take a typical 60 Hz monitor. Instead of 16666.6666
(repeating) being calculated as the vsync interval, you would get 16666
since the decimals are truncated. This is not really good at all and
affects the calculated speed values you get when using display sync. For
consistency and better precision, these should all be doubles just like
estimated_vsync_interval. Technically this means that we won't be able
to store as high of an integer value but such values would be absurdly
huge and never actually needed. Also estimated_vsync_interval already
can't handle such a case anyway.
Interrim solution, forwards compatible with new and backwards compatible
with old API. Eventually, we will want to discontinue the use of
deprecated pl_icc_params.save/load and pl_renderer_save/load, but that
requires minimum version bump.
We currently only allow specifying the Vulkan device to use by name. We
did this to avoid confusion around devices being enumerated in an
unpredictable order. However, there is a valid edge case where a system
may contain multiple devices of the same type - which means they will
have the same name, and so you can't control which one is used.
This change implements picking devices by UUID so that if names don't
work, you have some option available. As Vulkan 1.1 is a hard
requirement for libplacebo, we can just use UUIDs without conditional
checks.
Fixes#10898
Causes bad performance with interpolation because the changing hue angle
invalidates the mixing cache, as a result of libplacebo implementations
(specifically, the fact that this graph is drawn during the color
management process, instead of as a separate overlay).
Fix it by just hard-coding a particular, relatively interesting plane
(pi/4 approximately maps onto the red-blue axis).
There are way too many preexisting scripts that rely on this behavior
for video panning, like for example scripts that allow you to use mpv as
an image viewer. If this behavior is desired then it may be better to
add a new option for panning relative to the destination instead.
The restriction of video-pan-x/y being clamped to {-3, 3} also results
in the video being impossible to pan if it was zoomed in beyond a
certain degree.
This reverts commit 7d6f9e3739.
`--vo=gpu-next` no longer uses this option, being replaced entirely by a
luminance-based approach internally. And even for `--vo=gpu`, the values
other than 'hybrid' are universally inferior in quality.
In the interest of gradually reducing the amount of option bloat here,
remove this mostly-pointless option.
Pick a smarter rect, limit it to right half of screen (to allow more
easily using it simultaneously with stats), and also auto-rotate through
the hue plane by default (wrapping once every 10 seconds of playback).
I briefly considered making it based on wallclock time instead of pts,
but this has the rather unfortunate downside of only being updated
sporadically as the user moves the mouse over OSC elements. (Of course,
we could also forcibly redraw the screen continously to avoid it, but
I'd rather not make such an invasive change for no real reason)
This design also allows you to pause and focus on (via framestepping)
individual parts of the hue graph that you're interested in.
This actually fixes a bug that was present in this code even before the
new pl_options system, which is that `video_screenshot` would
permanently leave `peak_detect_params.allow_delayed` as false, until
something else forced options re-application due to options cache
change.
The last piece in the puzzle for doing hardware conversions
automatically is ensuring we only consider valid target formats for the
conversion. Although it is unintuitive, some vaapi drivers can expose a
different set of formats for uploads vs for conversions, and that is
the case on the Intel hardware I have here.
Before this change, we would use the upload target list, and our
selection algorithm would pick a format that doesn't work for
conversions, causing everything to fail. Whoops.
Successfully obtaining the conversion target format list is a bit of a
convoluted process, with only parts of it encapsulated by ffmpeg.
Specifically, ffmpeg understands the concept of hardware configurations
that can affect the constraints of a device, but does not define what
configurations are - that is left up to the specific hwdevice type.
In the case of vaapi, we need to create a config for the video
processing endpoint, and use that when querying for constraints.
I decided to encapsulate creation of the config as part of the hwdec
init process, so that the constraint query can be down in the
hwtransfer code in an opaque way. I don't know if any other hardware
will need this capability, but if so, we'll be able to account for it.
Then, when we look at probing, instead of checking for what formats
are supported for transfers, we use the results of the constraint query
with the conversion config. And as that config doesn't depend on the
source format, we only need to do it once.
Historically, we have not attempted to support hw->hw format conversion
in the autoconvert logic. If a user needed to do these kinds of
conversions, they needed to manually insert the hw format's conversion
filter manually (eg: scale_vaapi).
This was usually fine because the general rule is that any format
supported by the hardware can be used as well as any other. ie: You
would only need to do conversion if you have a specific goal in mind.
However, we now have two situations where we can find ourselves with a
hardware format being produced by a decoder that cannot be accepted by
a VO via hwdec-interop:
* dmabuf-wayland can only accept formats that the Wayland compositor
accepts. In the case of GNOME, it can only accept a handful of RGB
formats.
* When decoding via VAAPI on Intel hardware, some of the more unusual
video encodings (4:2:2, 10bit 4:4:4) lead to packed frame formats
which gpu-next cannot handle, causing rendering to fail.
In both these cases (at least when using VAAPI with dmabuf-wayland), if
we could detect the failure case and insert a `scale_vaapi` filter, we
could get successful output automatically. For `hwdec=drm`, there is
currently no conversion filter, so dmabuf-wayland is still out of luck
there.
The basic approach to implementing this is to detect the case where we
are evaluating a hardware format where the VO can accept the hardware
format itself, but may not accept the underlying sw format. In the
current code, we bypass autoconvert as soon as we see the hardware
format is compatible.
My first observation was that we actually have logic in autoconvert to
detect when the input sw format is not in the list of allowed sw
formats passed into the autoconverter. Unfortunately, we never populate
this list, and the way you would expect to do that is vo-query-format
returning sw format information, which it does not. We could define an
extended vo-query-format-2, but we'd still need to implement the
probing logic to fill it in.
On the other hand, we already have the probing logic in the hwupload
filter - and most recently I used that logic to implement conversion
on upload. So if we could leverage that, we could both detect when
hw->hw conversion is required, and pick the best target format.
This exercise is then primarily one of detecting when we are in this
case and letting that code run in a useful way. The hwupload filter is
a bit awkward to work with today, and so I refactored a bunch of the
set up code to actually make it more encapsulated. Now, instead of the
caller instantiating it and then triggering the probe, we probe on
creation and instantiate the correct underlying filter (hwupload vs
conversion) automatically.
So far all the keypad keys except for `0` and `,` mapped to the same
MP_KEY_* independent of numlock state, even though different key codes
are received.
Now all the alternative functions map to appropriate MP_KEY_* defines,
with missing ones added.
