Reduces (but likely does not remove) the danger of rounding intermediate
values down to 8 bit. This is important for cscale, or any other
processing that might store raw YUV values in framebuffers.
Fixes#1918.
This now stores caches for multiple ICC profiles, potentially all the
user has ever used. The big use case for this is for users with multiple
monitors. The old logic would mandate recomputing the LUT and discarding
the cache whenever dragging mpv from one screen to another.
This also avoids having to save and check the ICC profile itself, since
the file name already uniquely determines it.
This could help in cases where the DWM (Windows desktop compositor) adds another
layer of bufferring and therefore the SwapBuffers timing could get messed up.
Signed-off-by: wm4 <wm4@nowhere>
This merges all of the scaler-related options into a single
configuration struct, and also cleans up the way they're passed through
the code. (For example, the scaler index is no longer threaded through
pass_sample, just the scaler configuration itself, and there's no longer
duplication of the params etc.)
In addition, this commit makes scale-down more principled, and turns it
into a scaler in its own right - so there's no longer an ugly separation
between scale and scale-down in the code.
Finally, the radius stuff has been made more proper - filters always
have a radius now (there's no more radius -1), and get a new .resizable
attribute instead for when it's tunable.
User-visible changes:
1. scale-down has been renamed dscale and now has its own set of config
options (dscale-param1, dscale-radius) etc., instead of reusing
scale-param1 (which was arguably a bug).
2. The default radius is no longer fixed at 3, but instead uses that
filter's preferred radius by default. (Scalers with a default radius
other than 3 include sinc, gaussian, box and triangle)
3. scale-radius etc. now goes down to 0.5, rather than 1.0. 0.5 is the
smallest radius that theoretically makes sense, and indeed it's used
by at least one filter (nearest).
Apart from that, it should just be internal changes only.
Note that this sets up for the refactor discussed in #1720, which would
be to merge scaler and window configurations (include parameters etc.)
into a single, simplified string. In the code, this would now basically
just mean getting rid of all the OPT_FLOATRANGE etc. lines related to
scalers and replacing them by a single function that parses a string and
updates the struct scaler_config as appropriate.
This makes the core much more elegant, reusable, reconfigurable and also
allows us to more easily add aliases for specific configurations.
Furthermore, this lets us apply a generic blur factor / window function
to arbitrary filters, so we can finally "mix and match" in order to
fine-tune windowing functions.
A few notes are in order:
1. The current system for configuring scalers is ugly and rapidly
getting unwieldy. I modified the man page to make it a bit more
bearable, but long-term we have to do something about it; especially
since..
2. There's currently no way to affect the blur factor or parameters of
the window functions themselves. For example, I can't actually
fine-tune the kaiser window's param1, since there's simply no way to
do so in the current API - even though filter_kernels.c supports it
just fine!
3. This removes some lesser used filters (especially those which are
purely window functions to begin with). If anybody asks, you can get
eg. the old behavior of scale=hanning by using
scale=box:scale-window=hanning:scale-radius=1 (and yes, the result is
just as terrible as that sounds - which is why nobody should have
been using them in the first place).
4. This changes the semantics of the "triangle" scaler slightly - it now
has an arbitrary radius. This can possibly produce weird results for
people who were previously using scale-down=triangle, especially if
in combination with scale-radius (for the usual upscaling). The
correct fix for this is to use scale-down=bilinear_slow instead,
which is an alias for triangle at radius 1.
In regards to the last point, in future I want to make it so that
filters have a filter-specific "preferred radius" (for the ones that
are arbitrarily tunable), once the configuration system for filters has
been redesigned (in particular in a way that will let us separate scale
and scale-down cleanly). That way, "triangle" can simply have the
preferred radius of 1 by default, while still being tunable. (Rather
than the default radius being hard-coded to 3 always)
This requires FFmpeg git master for accelerated hardware decoding.
Keep in mind that FFmpeg must be compiled with --enable-mmal. Libav
will also work.
Most things work. Screenshots don't work with accelerated/opaque
decoding (except using full window screenshot mode). Subtitles are
very slow - even simple but huge overlays can cause frame drops.
This always uses fullscreen mode. It uses dispmanx and mmal directly,
and there are no window managers or anything on this level.
vo_opengl also kind of works, but is pretty useless and slow. It can't
use opaque hardware decoding (copy back can be used by forcing the
option --vd=lavc:h264_mmal). Keep in mind that the dispmanx backend
is preferred over the X11 ones in case you're trying on X11; but X11
is even more useless on RPI.
This doesn't correctly reject extended h264 profiles and thus doesn't
fallback to software decoding. The hw supports only up to the high
profile, and will e.g. return garbage for Hi10P video.
This sets a precedent of enabling hw decoding by default, but only
if RPI support is compiled (which most hopefully it will be disabled
on desktop Linux platforms). While it's more or less required to use
hw decoding on the weak RPI, it causes more problems than it solves
on real platforms (Linux has the Intel GPU problem, OSX still has
some cases with broken decoding.) So I can live with this compromise
of having different defaults depending on the platform.
Raspberry Pi 2 is required. This wasn't tested on the original RPI,
though at least decoding itself seems to work (but full playback was
not tested).
This has a number of user-visible changes:
1. A new flag blend-subtitles (default on for opengl-hq) to control this
behavior.
2. The OSD itself will not be color managed or affected by
gamma controls. To get subtitle CMS/gamma, blend-subtitles must be
used.
3. When enabled, this will make subtitles be cleanly interpolated by
:interpolation, and also dithered etc. (just like the normal output).
Signed-off-by: wm4 <wm4@nowhere>
Bilinear scaling is not a suitable default for something named "hq"; the
whole reason this was done in the past was because cscale used to be
obscenely slow. This is no longer the case, with cscale being nearly
free.
This replaces the old smoothmotion code by a more flexible tscale
option, which essentially allows any scaler to be used for interpolating
frames. (The actual "smoothmotion" scaler which behaves identical to the
old code does not currently exist, but it will be re-added in a later commit)
The only odd thing is that larger filters require a larger queue size
offset, which is currently set dynamically as it introduces some issues
when pausing or framestepping. Filters with a lower radius are not
affected as much, so this is identical to the old smoothmotion if the
smoothmotion interpolator is used.
This adds stuff related to gamma, linear light, sigmoid, BT.2020-CL,
etc, as well as color management. Also adds a new gamma function (gamma22).
This adds new parameters to configure the CMS settings, in particular
letting us target simple colorspaces without requiring usage of a 3DLUT.
This adds smoothmotion. Mostly working, but it's still sensitive to
timing issues. It's based on an actual queue now, but the queue size
is kept small to avoid larger amounts of latency.
Also makes “upscale before blending” the default strategy.
This is justified because the "render after blending" thing doesn't seme
to work consistently any way (introduces stutter due to the way vsync
timing works, or something), so this behavior is a bit closer to master
and makes pausing/unpausing less weird/jumpy.
This adds the remaining scalers, including bicubic_fast, sharpen3,
sharpen5, polar filters and antiringing. Apparently, sharpen3/5 also
consult scale-param1, which was undocumented in master.
This also implements cropping and chroma transformation, plus
rotation/flipping. These are inherently part of the same logic, although
it's a bit rough around the edges in some case, mainly due to the fallback
code paths (for bilinear scaling without indirection).
This automatically sets the gamma option depending on lighting conditions
measured from the computer's ambient light sensor.
sRGB – arguably the “sibling” to BT.709 for still images – has a reference
viewing environment defined in its specification (IEC 61966-2-1:1999, see
http://www.color.org/chardata/rgb/srgb.xalter). According to this data, the
assumed ambient illuminance is 64 lux. This is the illuminance where the gamma
that results from ICC color management is correct.
On the other hand, BT.1886 formalizes that the gamma level for dim environments
to be 2.40, and Apple resources (WWDC12: 2012 Session 523: Best practices for
color management) define the BT.1886 dim at 16 lux.
So the logic we apply is:
* >= 64lux -> 1.961 gamma
* =< 16lux -> 2.400 gamma
* 16lux < x < 64lux -> logaritmic rescale of lux to gamma. The human
perception of illuminance roughly follows a logaritmic scale of lux [1].
[1]: https://msdn.microsoft.com/en-us/library/windows/desktop/dd319008%28v=vs.85%29.aspx
This is based on pretty much the same (somewhat naive) logic right now.
I'm not convinced that the extra logic that eg. madVR includes is worth
enough to warrant heavily confusing the logic for it.
This shouldn't slow down the logic at all in any sane shader compiler,
and indeed it doesn't on any shader compiler that I tested.
Note that this currently doesn't affect cscale at all, due to the weird
implementation details of that.
Hopefully, this will really clear up how the thing is supposed to work
(and that it's not SVP, nor MVTools).
I also removed instances of the word "interpolation", since that's a
term that's easily misleading.
Finally, I expanded on smoothmotion-threshold since the purpose/meaning
was a bit confusing.
This is done mainly for consistency, since all of the EWA filters share
similar properties and it's important to distinguish them for
documentation purposes.
This is a variation of ewa_lanczos that is sinc-windowed instead of
jinc-windowed. Results are pretty similar, but the logic is simpler.
This could potentially replace the ugly ewa_lanczos code.
It's hard to tell, but from comparing stills I think this one has
slightly less ringing than regular ewa_lanczos.
This reverts commit a33b46194c.
It turns out FFmpeg really considers this a bug, and fixed it by making
the decoder output the correct pixel format.
Fixes#1565. Reverts the fix#1528, though it should work fine with
a recent git master FFmpeg.
This introduces a new option linear-scaling, which is now implied by
srgb, icc-profile and sigmoid-upscaling.
Notably, this means (sigmoidized) linear upscaling is now enabled by
default in opengl-hq mode. The impact should be negligible, and there
has been no observation of negative side effects of sigmoidized scaling,
so it feels safe to do so.
Comment explains why I have been so doubtful at adding this. The Apple docs
say CGDisplayModeGetRefreshRate is supposed to work only for CRTs, but it
doesn't, and actually works for LCD TVs connected over HDMI and external
displays (at least that's what I'm told, I don't have the hardware to test).
Maybe Apple docs are incorrect.
Since AFAIK Apple doesn't want to give us a better API – maybe in the fear we
might be able to actually write some useful software instead of "apps" –
I decided not to care as well and commit this.
This reverts the default behavior introduced in commit 93feffad. Way too
often libavcodec will return RGB data that has an alpha channel as per
pixel format, but actually contains garbage.
On the other hand, this will actually render garbage color values in
e.g. PNG files (for pixels with alpha==0, the color value should be
essentially ignored, which is what the old alpha blend mode did).
This "fixes" #1528, which is probably a decoder bug (or far less likely,
a broken file).
Make the lazy gamma initialization less weird, and make the default
value of the "gamma" sub-option 1.0. This means --vo=opengl:help will
list the actual default value.
Also change the lower bound to 0.1 - avoids a division by zero (I don't
know how shaders handle NaN, but it's probably not a good idea to give
them this value).
These were derived from dividing our assumed video gamut (1.961) by some
typical screen values (2.2 for dimly lit and 2.4 for pitch black):
1.961/2.4 = 0.8170833333333334 ~= 0.8
1.961/2.2 = 0.8913636363636364 ~= 0.9
This queries the _ICC_PROFILE property on the root window. It also tries
to reload the ICC when it changes, or if the mpv window changes the
monitor. (If multiple monitors are covered, mpv will randomly select one
of them.)
The official spec is a dead link on freedesktop.org, so don't blame me
for any bugs.
Note that this assumes that Xinerama screen numbers match the way mpv
enumerates the xrandr monitors. Although there is some chance that this
matches, it most likely doesn't, and we actually have to do complicated
things to map the screen numbers. If it turns out that this is required,
I will fix it as soon as someone with a suitable setup for testing the
fix reports it.
Seems like several people agree that it's a good filter for downscaling.
Setting this option by default may also prevent people from accidentally
using an unsuitable filter for downscaling by setting "scale" and
without being aware of the impliciations (maybe). On the other hand,
this change is not strictly backwards compatible for the same reasons.
Also, allow disabling this option with scale-down="" (before this, not
setting it was the only way to do this - not possible anymore if it's
set by default). This is what the change in handle_scaler_opt() does.
vo.c queried the VO at initialization whether it wants to be updated on
every display frame, or every video frame. If the smoothmotion option
was changed at runtime, the rendering mode in vo.c wasn't updated.
Just let vo_opengl set the mode directly. Abuse the existing
vo_set_flip_queue_offset() function for this.
Also add a comment suggesting the use of --display-fps to the manpage,
which doesn't have anything to do with the rest of this commit, but is
important to make smoothmotion run well.
SmoothMotion is a way to time and blend frames made popular by MadVR. It's
intended behaviour is to remove stuttering caused by mismatches between the
display refresh rate and the video fps, while preserving the video's original
artistic qualities (no soap opera effect). It's supposed to make 24fps video
playback on 60hz monitors as close as possible to a 24hz monitor.
Instead of drawing a frame once once it's pts has passed the vsync time, we
redraw at the display refresh rate, and if we detect the vsync is between two
frames we interpolated them (depending on their position relative to the vsync).
We actually interpolate as few frames as possible to avoid a blur effect as
much as possible. For example, if we were to play back a 1fps video on a 60hz
monitor, we would blend at most on 1 vsync for each frame (while the other 59
vsyncs would be rendered as is).
Frame interpolation is always done before scaling and in linear light when
possible (an ICC profile is used, or :srgb is used).
These aliases were removed in commit 1ec77214. Add a notice to the
manpage how to get these back. Apparently, "lanczos2" and "lanczos3"
were the only interesting aliases possibly used by someone, so the
description is limited to these two.
These are now auto-detected sanely; and enabled whenever it would be a
performance or quality gain (which is pretty much everything except
bilinear/bilinear scaling).
Perhaps notably, with the absence of scale_sep, there's no more way to
use convolution filters on hardware without FBOs, but I don't think
there's hardware in existence that doesn't have FBOs but is still fast
enough to run the fallback (slow) 2D convolution filters, so I don't
think it's a net loss.
This is better even for non-separable. The only exception is when using
bilinear for both lscale and cscale. I've fixed the
documentation/comments to make more sense.
This is not quite the same thing as madVR's antiringing algorithm, but
it essentially does something similar.
Porting madVR's approach to elliptic coordinates will take some amount
of thought.
This also fixes the maximum range to 16.0, which was previously set to
32.0 and incorrectly documented as 8.0. 16 taps should be more than
anybody will ever need, but it's the highest radius that's supported by
all affected filters.
After finding out more about how video mastering is done in the real
world it dawned upon me why the "hack" we figured out in #534 looks so
much better.
Since mastering studios have historically been using only CRTs, the
practice adopted for backwards compatibility was to simulate CRT
responses even on modern digital monitors, a practice so ubiquitous that
the ITU-R formalized it in R-Rec BT.1886 to be precisely gamma 2.40.
As such, we finally have enough proof to get rid of the option
altogether and just always do that.
The value 1.961 is a rounded version of my experimentally obtained
approximation of the BT.709 curve, which resulted in a value of around
1.9610336. This is the closest average match to the source brightness
while preserving the nonlinear response of the BT.1886 ideal monitor.
For playback in dark environments, it's expected that the gamma shift
should be reproduced by a user controlled setting, up to a maximum of
1.224 (2.4/1.961) for a pitch black environment.
More information:
https://developer.apple.com/library/mac/technotes/tn2257/_index.html
This avoids issues when upscaling directly in linear light, and is the
recommended way to upscale images according to imagemagick.
The default slope of 6.5 offers a reasonable compromise between
ringing artifacts eliminated and ringing artifacts introduced by
sigmoid-upscaling. Same goes for the default center of 0.75.
The previous implementation of opengl-cb kept only latest flipped frame.
This can cause massive frame drops because rendering is done asynchronously
and only the latest frame can be rendered.
This commit introduces frame queue and releated options to opengl-cb.
frame-queue-size: the maximum size of frame queue (1-100, default: 1)
frame-drop-mode: behavior when frame queue is full (pop, clear, default: pop)
The frame queue holds delayed frames and drops frames if the frame queue is
overflowed with next method:
'pop' mode: drops all the oldest frames overflown.
'clear' mode: drops all frames in queue and clear it.
With default options(frame-queue-size=1:frame-drop-mode=pop),
opengl-cb behaves in the same way as previous implementation effectively.
For frame-queue-size > 1, opengl-cb tries to calls update() without waiting
next flip_page() in order to consume queued frames.
Signed-off-by: wm4 <wm4@nowhere>
While there's no actual need to get rid of these, I want to make sure
nobody actually needs this stuff, and removing it is the best way to
get to know this. We still can revert this commit if it turns out there
is a significant need for this stuff.
The final goal is removing vo_opengl_old entirely. Add a warning, which
basically announces this intention.
I'm hoping this is generally more compatible, and it works with GLES.
This probably has not much of an effect on desktop GL. It also switches
only the default format for --vo=opengl, not --vo=opengl-hq.
"-hq" already uses GL_RGBA16, though since it's a sized format, the
story is a bit different, and it won't work on GLES either.
Obscure feature, and I've never heard of anyone using it.
The anaglyph effects can be reproduced with vf_stereo3d. The only thing
that can't be reproduced with it is "quadbuffer", which requires special
and expensive hardware.
I think that's expected; mpv shouldn't draw anything while no video is
active. This doesn't blend transparently, though.
Also document the vo_opengl_cb thing.
This sub-option was turned into a flag when the sub-option parser was
changed to the generic one (probably accidentally). Turn it into a
proper choice-option.
Also, adjust what the options do. Though none of this probably makes
much sense; the default should work, and if it doesn't, the GPU/driver
is probably beyond help.
This was kept in the codebase because it is slightly faster than --vo=opengl
on really old Intel cards (from the GMA era). Time to kill it, and let it rest.
Fixes#1061
Add a new parameter 'p' to gaussian filter. The new formula used
a different base taken from fmtconv plugin, so that the
new parameter is exactly same as the one used in Avisynth and
Vapoursynth.
The new default value is 2 / log(2) * 10, with the default value it
conforms to the original kernel taken from vector-agg.
Add two new options, make it possible for user to set the radius
for some of the filters with no fixed radius.
Also add three new filters with the new radius parameter supported.
I'm not sure about the merit, though it does print nice numbers if debug
output is enabled.
Basically, this tries to achieve similar results as the glFinish()
business, but again it entirely depends on the drivers whether this
does anything meaningful, or whether it's actively harmful.
It seems that at least on nvidia systems with composting disabled, we
can get it to block deterministically on the actual vsync event, which
should improve framedropping.
Notably, we now conform to SMPTE 428-1-2006 when decoding XYZ12 input,
and we can support rendering intents other than colorimetric when
converting between BT.709 and BT.2020, like with :srgb or :icc-profile.
This add support for reading primary information from lavc, categorized
into BT.601-525, BT.601-625, BT.709 and BT.2020; and passes it on to the
vo. In vo_opengl, we always generate the 3dlut against the wider BT.2020
and transform our source into this colorspace in the shader.
wm4
Niklas Haas
Marcin Kurczewski
Avi Halachmi (:avih)
Martin Herkt
Stefano Pigozzi
xylosper
Bin Jin