This gets rid of an old hack, VOFLAG_HIDDEN. Although handling of it has
been sane for a while, it used to cause much pain, and is still
unintuitive and weird even today.
The main reason for this hack is that OpenGL selects a X11 Visual for
you, and you're supposed to use this Visual when creating the X window
for the OpenGL context. Which means the X window can't be created early
in the common X11 init code, but the OpenGL code needs to do something
before that. API-wise you need separate functions for X11 init and X11
window creation. The VOFLAG_HIDDEN hack conflated window creation and
the entrypoint for resizing on video resolution change into one
function, vo_x11_config_vo_window(). This required all platform backends
to handle this flag, even if they didn't need this mechanism.
Wayland still uses this for minor reasons (alpha support?), so the
wayland backend must be changed before the flag can be entirely removed.
If interpolation is enabled, then this causes heavy artifacts if done
while unpaused. It's preferable to allow a latency of a few frames for
the change to take full effect instead. If this is done paused, the
frame is fully redrawn anyway.
This reverts commit d11184a256.
Unfortunately, there was a lot of unexpected resistance.
Do note that this is still extremely slow, crappy, etc.
Note that vo_x11.c was further edited. Compared to the removed vo_x11.c,
an additional ~200 lines of code was removed in order to simplify it. I
tried to strip it down as much as possible. In particular, support for
odd non-32 bit formats (24, 16, 15, 8 bit) is dropped.
Closes#2300.
The vf_format suboption is replaced with --video-output-levels (a global
option and property). In particular, the parameter is removed from
mp_image_params. The mechanism is moved to the "video equalizer", which
also handles common video output customization like brightness and
contrast controls.
The new code is slightly cleaner, and the top-level option is slightly
more user-friendly than as vf_format sub-option.
This essentially reverts commit 009dfbe3. FFmpeg VideoToolbox support
is being wacky, and can cause major issues, such as not being able
to decode a single frame. (E.g. by playing a .ts file. This should be
fixed in FFmpeg eventually.)
This is not a straight revert of the commit; just a functional one. We
keep the slightly simpler code structure.
It doesn't deal with VDA at all anymore. Rename it to hwdec_osx.c. Not
using hwdec_videotoolbox.c, because that would give it the longest
source path in this project yet. (Also, this code isn't even
VideoToolox-specific, other than the name of the pixel format used.)
VideoToolbox is preferred. Now that FFmpeg released 2.8, there's no
reason to support VDA anymore. In fact, we had a bug that made VDA not
useable with older FFmpeg versions in some newer mpv releases.
VideoToolbox is supported even on slightly older OSX versions, and if
not, you still can run mpv without hw decoding.
Definitely not needed anymore, and fixes a crash in some weird corner-
cases.
The extradata freeing is apparently still needed, though. (Because a
codec context can be opened again, which makes no sense, but ok.)
There are at least 2 ways of using VAAPI without X11 (Wayland, DRM).
Remove the X11 requirement from the decoder part and the EGL interop.
This will be used by a following commit, which adds Wayland support.
The worst about this is the decoder part, which includes a bad hack for
using the decoder without any VO interop (also known as "vaapi-copy"
mode). Separate the X11 parts so that they're self-contained. For the
EGL interop code we do something similar (it's kept slightly simpler,
because it essentially only has to translate between our silly
MPGetNativeDisplay abstraction and the vaGetDisplay...() call).
It looks like my hope that we can unconditionally include EGL headers in
the OpenGL code is not coming true, because OSX does not support EGL at
all. So I prefer loading the VAAPI EGL/GL specific extensions manually,
because it's less of a mess. Partially reverts commit d47dff3f.
While EGL 1.4 seemed a bit ambiguous about this to me, it actually says
quite clearly that core functions are not supported with
eglGetProcAddress() in the following paragraph.
Normally, we prefer GLX on X11. But for the VAAPI EGL interop, we
obviously want EGL. Since nvidia does not provide EGL with desktop GL
yet, we can leave it to the autoprobing. Just make sure some failure
messages don't unnecessarily show up in the nvidia case.
This breaks VAAPI GLX interop by default, but I don't care much. If
you use --hwdec=auto (which you should if you want hw decoding), this
should fallback to vaapi-copy instead.
Probe the surface format, and check whether it's really something we
support. This also does a complete check whether the EGL interop works
at all (the only way to find this out is actually running this code).
Also, support YV12. Under some circumstances, vaapi (with Intel
drivers) can be made to use this format.
Unfortunately, the Intel drivers show some very weird behavior, which
is hopefully a bug. insane_hack() provides a very evil workaround (see
comments). A proper solution might be passing the hw format as part of
mp_image_params, but as long as hw surfaces appear to be able to change
the format on the fly, attempting this is probably not worth the extra
complexity and likely fragility. The hack allows us to pretend that
there is sane behavior for now.
This makes it much faster if the surface is really mapped from GPU
memory. It's slightly slower than system memcpy if used on system
memory. We don't really know definitely in which type of memory
it's located, so we use the GPU memcpy in all cases.
Fixes#2317.
Make the GPU memcpy from the dxva2 code generally useful to other parts
of the player.
We need to check at configure time whether SSE intrinsics work at all.
(At least in this form, they won't work on clang, for example. It also
won't work on non-x86.)
Introduce a mp_image_copy_gpu(), and make the dxva2 code use it. Do some
awkward stuff to share the existing code used by mp_image_copy(). I'm
hoping that FFmpeg will sooner or later provide a function like this, so
we can remove most of this again. (There is a patch, bit it's stuck in
limbo since forever.)
All this is used by the following commit.
Broken by commit d47dff3f. If something is going to include EGL.h,
header_fixes.h has to know. This definitely affected vo_rpi, and
probably affects wayland builds (with x11egl didabled) as well.
Checking and resetting the VAImage.buf field is non-sense, even if it
happened to work out in the normal case. buf is actually freed when
vaDestroyImage() is called (not quite intuitive), and we need an extra
field to know whether vaReleaseBufferHandle() has to be called.
Printing "Using vaDeriveImage()" every frame is too verbose, so raise
the log level.
mp_image strides are in int and not unsigned int; fix this. It's not
like it actually matters, though.
Finish a comment.
Add the upstream symbolic names as comments. Normally, these should be
defined in libdrm's drm_fourcc.h header. But DRM_FORMAT_R8 and
DRM_FORMAT_GR88 are not defined anywhere, except in the kernel userland
headers of Linux 4.3 (!). We don't want mpv to depend on bleeding-edge
Linux kernel headers, so this will have to do.
Also, just for completeness, add fourccs for the 3 and 4 channel
formats. I didn't manage to test them, though.
Reduces the amount of hardcoded assumptions about the layout
drastically. (Now adding yuv420 support would be just adjusting an if,
if you ignore the other problems, such as determining the hw format at
all early enough.)
Don't call eglDestroyImageKHR() on the same ID possibly more than once.
Clear the image reference on termination, or we would leak up to 1 image
per VO recreation.
Should work much better than the old GLX interop code. Requires Mesa 11,
and explicitly selecting the X11 EGL backend with:
--vo=opengl:backend=x11egl
Should it turn out that the new interop works well, we will try to
autodetect EGL by default.
This code still uses some bad assumptions, like expecting surfaces to be
in NV12. (This is probably ok, because virtually all HW will use this
format. But we should at least check this on init or so, instead of
failing to render an image if our assumption doesn't hold up.)
This repo was a lot of help: https://github.com/gbeauchesne/ffvademo
The kodi code was also helpful (the magic FourCC it uses for
EGL_LINUX_DRM_FOURCC_EXT are nowhere documented, and
EGL_IMAGE_INTERNAL_FORMAT_EXT as used in ffvademo does
not actually exist).
(This is the 3rd VAAPI GL interop that was implemented in this player.)
The VAAPI EGL interop code will need access to the X11 Display. While
GLX could return it from the current GLX context, EGL has no such
mechanism. (At least no standard one supported by all implementations.)
So mpv makes up such a mechanism.
For internal purposes, this is very rather awkward solution, but it's
needed for libmpv anyway.
These extensions use a bunch of EGL types, so we need to include the EGL
headers in common.h to use our GL function loader with this.
In the future, we should probably require presence of the EGL headers to
reduce the hacks. This might be not so simple at least with OSX, so for
now this has to do.
Surfaces used by hardware decoding formats can be mapped exactly like a
specific software pixel format, e.g. RGBA or NV12. p->image_params is
supposed to be set to this format, but it wasn't.
(How did this ever work?)
Also, setting params->imgfmt in the hwdec interop drivers is pointless
and redundant. (Change them to asserts, because why not.)
This is a pseudo-OpenGL extension for letting libmpv query native
windowing system handles from the API user. (It uses the OpenGL
extension mechanism because I'm lazy. In theory it would be nicer to let
the user pass them with mpv_opengl_cb_init_gl(), but this would require
a more intrusive API change to extend its argument list.)
The naming of the extension and associated function was unnecessarily
Windows specific (using "D3D"), even though it would work just fine for
other platforms. So deprecate the old names and introduce new ones. The
old ones still work.
This turns the old scalers (inherited from MPlayer) into a pre-
processing step (after color conversion and before scaling). The code
for the "sharpen5" scaler is reused for this.
The main reason MPlayer implemented this as scalers was perhaps because
FBOs were too expensive, and making it a scaler allowed to implement
this in 1 pass. But unsharp masking is not really a scaler, and I would
guess the result is more like combining bilinear scaling and unsharp
masking.
Usually, libavcodec ignores errors reported by the hardware decoding
API, so it's not like we can actually escape if the hardware is somehow
acting up.
For normal fallback purposes, or if parts of the hw decoding API which
we actually check fails, we do this by setting and checking the
hwdec_failed flag anyway.
This can happen if the hw decoder allocates padded surfaces (e.g.
mod16), but the VPP output surface was allocated with the exact size.
Apparently VPP requires matching input and output sizes, or it will add
artifacts. In this case, it added mirrored pixels to the bottom few
pixels.
Note that the previous commit should have fixed this. But it didn't
work, while this commit does.
Fixes#2320.
If not set, VPP will use the whole surface. This is a problem if the
surfaces are padded, and especially if the surfaces are padded by
different amounts.
This is an attempt to fix#2320, but it appears to do nothing at all.
The comment was largely outdated, and described the old situation when
we used a "violent" fallback by making get_buffer2 fail completely.
Also, for the case when the hw decoder initialization succeeded (in
get_format), but get_buffer2 for some reason requests something
unexpected, we also can fallback more gracefully and in the same way.
Window classes are process-wide (or at least DLL-wide), so you can't
have 2 classes with the same name. Our code attempted to do this when
for example 2 libmpv instances were created within the same process.
This failed, because RegisterWindowEx() fails if the class already
exists.
Fix this by ignoring RegisterWindowEx() errors. If the class can really
not be registered, we will fail on CreateWindowEx() instead. Of course
we also can't unregister the class, as another thread might be using it.
Windows will free the class automatically if the DLL is unloaded or the
process terminates.
Fixes#2319 (hopefully).
I took this out because I thought the filter chain would auto-negotiate
using nv12 without the explicit hint, and it does in the basic case
with no intermediate filter, but once you start adding filters, it
can end up negotiating a different format and then failing.
Today, vdpaurb will fail if it's used with non-hardware-decoded
content. This created work for the user as they have to explicitly
add or not add it, depending on the content.
As an improvement, we can make vdpaurb pass through any frames
that aren't hardware decoded, so that it can always be present in the
filter chain, if desired.
I see no point in keeping these around. Keeping wrappers for some select
libavfilter filters just because MPlayer had these filters is not a good
reason.
Ultimately, all real filtering work should go to libavfilter, and users
should get used to using vf_lavfi directly. We might even not require
the awful double-nested syntax for using libavfilter one day.
vf_rotate, vf_yadif, vf_stereo3d are kept because mpv uses them
internally. (They all extend the lavfi filters or change their
defaults.) vf_mirror is kept for symmetry with vf_flip. vf_gradfun and
vf_pullup are probably semi-popular, so I'll remove them not yet - only
after some more discussion.
These were normalized and are saner now. We want to use the new fields,
and also get rid of the deprecation warnings, so use them. There's no
release yet which uses these, so some ifdeffery is unfortunately needed.
2 things are being stupid here: Apple for requiring rectangle textures
with their IOSurface interop for no reason, and OpenGL having a
different sampler type for rectangle textures.
The removal of source-shader is a side effect, since this effectively
replaces it - and the video-reading code has been significantly
restructured to make more sense and be more readable.
This means users no longer have to constantly download and maintain a
separate deband.glsl installation alongside mpv, which was the only real
use case for source-shader that we found either way.
This is mostly to cut down somewhat on the amount of code bloat in
video.c by moving out helper functions (including scaler kernels and
color management routines) to a separate file.
It would certainly be possible to move out more functions (eg. dithering
or CMS code) with some extra effort/refactoring, but this is a start.
Signed-off-by: wm4 <wm4@nowhere>
Instead of the other way around of disabling disallowed options. This is
more robust and also slightly simpler, at least conceptually. If new
vo_opengl features are added, they don't need to be explicitly disabled
for dumb-mode just to avoid that it accidentally breaks.
Sigh...
Hopefully this code will be completely unnecessary one day, as it's
only needed due to the sub-option parser craziness.
Move dumb_mode to the top of the struct, so the C universal initializer
doesn't cause warnings with all those broken compilers.
The single path optimization, rendering the video in one shader pass and
without FBO indirections, was removed soem commits ago. It didn't have a
place in this code, and caused considerable complexity and maintenance
issues.
On the other hand, it still has some worth, such as for use with
extremely crappy hardware (GLES only or OpenGL 2.1 without FBO
extension). Ideally, these use cases would be handled by a separate VO
(say, vo_gles). While cleaner, this would still cause code duplication
and other complexity.
The third option is making the single-pass optimization a completely
separate code path, with most vo_opengl features disabled. While this
does duplicate some functionality (such as "unpacking" the video data
from textures), it's also relatively unintrusive, and the high quality
code path doesn't need to take it into account at all. On another
positive node, this "dumb-mode" could be forced in other cases where
OpenGL 2.1 is not enough, and where we don't want to care about versions
this old.
This change makes vo_opengl slightly less compatible (ancient devices
without FBOs will no longer work) and decreases performance in the
simplest case (vo=opengl), in exchange for significantly reducing code
complexity and making everything easier to reason about.
This didn't seem entirely sane. It probably worked by accident, because
"position" is always the first attribute, and thus the default value 0
for the location was always correct.
Often, we don't know whether hardware decoding will work until we've
tried. (This used to be different, but API changes and improvements in
libavcodec led to this situation.) We will often output that we're going
to use hardware decoding, and then print a fallback warning.
Instead, print the status once we have decoded a frame.
Some of the old messages are turned into verbose messages, which should
be helpful for debugging. Also add some new ones.
The fallback at initialization time was basically duplicated, maybe for
the sake of showing a different error message. This doesn't matter
anymore; not much can fail at initialization anymore. Most meaningful
and common errors happen either at probing or in get_format (when the
actual hw decoder is initialized).
If PBO upload fails, disable PBOs and revert to the normal codepath. In
theory we should retry PBO upload on failure (because OpenGL specifies
that it can sporadically fail), but since it normally doesn't happen,
and the fallback will work, I'm not bothering.
Some restructuring is needed, since glUnmapBuffer needs to be called
earlier. In fact, the old code structure didn't make too much sense, and
is a leftover from MPlayer's direct rendering support, which let the
decoder decode to a PBO-mapped region. This means the buffer_ptr field
can be dropped. Drop buffer_size as well, since it only had 2 possible
values (0 or the size required for the current config).
Can significantly help with very large video resolutions on nvidia
drivers. It doesn't seem to have negative effects on Intel drivers
either. (Although it could have on Intel drivers for older hardware.)
For now, this is only for --vo=opengl-hq. Maybe --vo=opengl should use
it too, but it's still meant to be the crappy, fail-safe default.
Setup a dummy image for the given image params, and get the plane sizes
from that. Admittedly not much of a simplification, but conceptually
it's simpler and less error-prone, as the image layout is guaranteed to
be the same, rather than essentially duplicating the way it is
determined.
This is from times when we supported padded/non-NPOT textures. The
difference is not useful anymore, and theoretical support for different
sizes is most likely buggy and unmaintained. So remove it.
Also remove the tex_ prefix wherever it appears.
Use mp_image_copy() instead of copying manually. (This function checks
whether the destination is regarded writeable, which it is not, because
the destination is the source image with changed pointers, so
refcounting has to be removed from the destination image by resetting
mpi->bufs.)
This shouldn't be needed anymore. Textures are now always allocated with
the exact size. Any padding (including non-NPOT support) is gone. The
texture sizes will always match the memory plane sizes.
Drop the unused and forgotten "npot" field from the option struct too.
Undo 292266f2. Reapply 3e12e79b.
An additional copy is not really justified, as it could reduce
performance. On the other hand, we can force API users to create
a GL 3.x context.
eglTerminate() affects the EGLDisplay in all threads. Since the RPI
firmware apparently only ever uses EGL_DEFAULT_DISPLAY, this means it
will trash all other contexts on other threads in the same process.
Thus we don't call eglTerminate() at all, at least on RPI. Call
eglReleaseThread() instead (which may or may not be a NOP).
yuva444p worked, yuva420p didn't. This happened because the chroma pass
discards the alpha plane, which is referenced by the alpha blend code
later.
Add a terrible hack to work this around, actually using the same hack as
was used for the Y plane. (A terrible hack for terrible code.)
If the drag and drop action is anything other than
XdndActionCopy, append the dropped files rather than
replacing the existing playlist. With most file managers,
this will mean at least pressing shift while dropping.
This puts in place the machinery to merely append dropped file to the playlist
instead of replacing the existing playlist. In this commit, all front-ends
set this to false preserving the existing behaviour.
This might fix some problems when framestepping with interpolation
enabled. The problem here is that we want to show the non-interpolated
frame while paused. Framestepping is like unpausing the video for a
frame, and then pausing again. This draws an interpolated frame, and
redrawing on pausing is supposed to take care of this.
This possibly didn't always work, because vo->want_redraw is not checked
by the vo_control() code path. So wake up the VO thread (which takes
care of servicing redraw requests, kind of) explicitly.
The correct solution is getting rid of the public-writable want_redraw
field and replacing it with a new vo_request_redraw() function, but this
can come later.
libavcodec does not support HEVC via VAAPI yet, so this won't work.
However, there is ongoing work to add HEVC support to VAAPI, and this
change might help with testing. (Or maybe not - but there is no harm in
this change.)
Should not happen, but since we don't control decoder video surface
allocation, anything could happen, and the code should be able to deal
with it. Untested.
This significantly reduces the amount of noticeable flashing when using
tscale kernels with negative lobes, by cutting them off completely.
I'm not sure if this has any negative effects. It needs a bit of
subjective testing over a period of time, so I just made it an option.
Fixes#2155.
Since vo_rpi uses MMAL for video output, which is completely
independent from the GLES overlay, we can just not redraw the
GLES screen if subtitles do not change.
(As a furhter optimization, the dispmanx overlay could be removed
if nothing is visible. But I'm not sure if adding and removing the
overlay frequently is a good idea for performance, so this could
just as well go the other way.)
Slightly faster than using the dispmanx mess (perhaps to a large amount
due to the rather stupid C-only unoptimized ASS->RGBA blending code).
Do this by reusing vo_opengl's subtitle renderer, and vo_opengl's RPI
backend.
This affects vo_opengl_cb in particular: it'll most likely auto-load
VDA, and then the VideoToolbox decoder won't work. And everything fails.
This is mainly caused by FFmpeg using separate pixfmts for the _same_
thing (CVPixelBuffers), simply because libavcodec's architecture demands
that hwaccel backends are selected by pixfmts. (Which makes no sense,
but now we have the mess.)
So instead of duplicating FFmpeg's misdesign, just change the format to
our own canonical one on the image output by the decoder. Now the GL
interop code is exactly the same for VDA and VT, and we use the VT name
only.
We must not use the frame PTS in any case. In this case, it fails
because nothing sets it up to wake up. This typically caused the player
to apparently "pause", until something else waked it up, like moving the
mouse and other events.
This VO is special because it normally doesn't block on vsync, but can
be made to do so. Supposedly the MMAL video output API merely sets a
"current frame" field when sending an output frame, and the firmware
will pick up whatever frame that field is set to at the time of a
vsync.
If this mode is enabled, the player tries to strictly synchronize video
to display refresh. It will adjust playback speed to match the display,
so if you play 23.976 fps video on a 24 Hz screen, playback speed is
increased by approximately 1/1000. Audio wll be resampled to keep up
with playback.
This is different from the default sync mode, which will sync video to
audio, with the consequence that video might skip or repeat a frame once
in a while to make video keep up with audio.
This is still unpolished. There are some major problems as well; in
particular, mkv VFR files won't work well. The reason is that Matroska
is terrible and rounds timestamps to milliseconds. This makes it rather
hard to guess the framerate of a section of video that is playing. We
could probably fix this by just accepting jittery timestamps (instead
of explicitly disabling the sync code in this case), but I'm not ready
to accept such a solution yet.
Another issue is that we are extremely reliant on OS video and audio
APIs working in an expected manner, which of course is not too often
the case. Consequently, the new sync mode is a bit fragile.
While the "old" libavcodec vdpau API is not deprecated (only the very-
old API is), it's still relatively complicated code that badly
duplicates the much simpler newer vdpau code. It exists only for the
sake of older FFmpeg releases; get rid of it.
VDA is being deprecated in OS X 10.11 so this is needed to keep hwdec working.
The code needs libavcodec support which was added recently (to FFmpeg git,
libav doesn't support it).
Signed-off-by: Stefano Pigozzi <stefano.pigozzi@gmail.com>
Pretty stupid: vo_get_vsync_interval() returns a negative value if the
display FPS is unknown (e.g. xrandr not compiled), and the comparison
whether the value is below 0 fails later because it's assigned to an
unsigned int.
Regression since commit e3d85ad4.
Also, fix some comments in vo.c.
When full_redraw is set, we always need to take the draw_image path. If
it's not set, we can try VOCTRL_REDRAW_FRAME (and fallback to draw_image
if that fails).
Fixes#2184.
The jpeg-optimize and jpeg-baseline options were undocumented, and
they're also pretty useless. There's no reason to ever change them.
Also, don't write jpeg baseline images. This just makes compression
worse for the sake of rather questionable compatibility with ancient
decoders.
If the framedrop count happens to be incremented with
vo_increment_drop_count() during rendering, these increments were
counted twice, because these events also set in->dropped_frame.
Revert "win32: more wchar_t -> WCHAR replacements"
Revert "win32: replace wchar_t with WCHAR"
Doing a "partial" port of this makes no sense anymore from my
perspective. Revert the changes, as they're confusing without
context, maintenance, and progress. These changes were a bit
premature anyway, and might actually cause other issues
(locale neutrality etc. as it was pointed out).
This was essentially missing from commit 0b52ac8a.
Since L"..." string literals have the type wchar_t[], we can't use them
for UTF-16 strings. Use C11 u"..." string literals instead. These have
the type char16_t[], but we simply assume char16_t is the same
underlying type as WCHAR. In practice, they're both unsigned short.
For this reason use -std=c11 on Windows. Since Windows is a "special"
environment (we require either MinGW or Cygwin), we don't need to worry
too much about compiler compatibility.
A user complains that it leads to the dxva driver failing, leading to
messages like this:
[ffmpeg/video] h264: Failed to execute: 0x8007000e
[ffmpeg/video] h264: hardware accelerator failed to decode picture
Reportedly, this happens only with vo_direct3d, not with vo_opengl. The
only difference is that vo_direct3d attempts to share the D3D device
with the decoder. Possibly the error is that the device in the VO is not
created with D3DCREATE_MULTITHREADED. Change this.
Probably fixes#2178.
WCHAR is more portable. While at least MinGW, Cygwin, and MSVC actually
use 16 bit wchar_t, Midipix will have 32 bit wchar_t. In that context,
using WCHAR instead is more portable.
This affects only non-MinGW parts, so not all uses of wchar_t need to
be changed. For example, terminal-win.c won't be used on Midipix at
all. (Most of io.c won't either, so the search & replace here is more
than necessary, but also not harmful.)
(Midipix is not useable yet, so this is just preparation.)
Instead of special-casing hwdec in the place where the video textures
are used, just set the textures in the image upload function. The
renderer code doesn't need to know whether hwdec interop is used at all.
This detected whether an OpenGL context still provided legacy OpenGL if
the OpenGL version is modern (>= 3.0). This was actually only needed for
vo_opengl_old, because it relied on legacy functions. Since it's gone,
this code isn't needed either.
(Also, the removed comment about OpenGL 3.0 was wrong: you could just
query GL_CONTEXT_FLAGS and see if the forward compatible bit was set.)
This reverts commit fb8d158366.
Reallocating the FBOs on every resize is very slow. It affects resizing
the window, as well as changing the video size itself with e.g.
panscan. Since the original change was done based on a single user
complaint, but the change itself caused a lot of complaints, we decided
to just revert it.
Instead of calling it "future frames" and adding or subtracting 1 from
it, always call it "requested frames". This simplifies it a bit.
MPContext.next_frames had 2 added to it; this was mainly to ensure a
minimum size of 2. Drop it and assume VO_MAX_REQ_FRAMES is at least 2;
together with the other changes, this can be the exact size of the
array.
This was requested by someone.
All code was written by myself; some minor changes by 2 contributors who
agreed to general LGPL relicensing. 1 line of code is by someone unknown
who possibly wasn't asked (setting the "display_fps" variable), and
which can be reasonably ignored as it makes up only 0.1% of the file.
I still have no idea why this is needed, maybe some weird off-by-one
in some shitty driver? Either way, the difference for a working setup
shouldn't be too major, the most noticeable effect would be somewhat worse
performance when resizing the video during playback with interpolation
enabled using the mouse.
That's a specific enough side effect for me to not care as much about it.
Fixes#1814.
There are some situations when redrawing is requested, but the current
frame was deleted. This could happen when switching e.g. hw decoding
mid-stream.
Separate uploading/drawing and fix the condition.
Just avoid some code duplication. Also, gl_video_set_options() having a
queue size output parameter is weird at best. While I don't appreciate
that this commit suddenly requires gl_video.c to deal with vo.c directly
in a special case, it's simply the best place to put this function.
The VO will be provided with future frames even if the format changes
mid-stream. This caused a crash if these frames were actually used (i.e.
interpolation mode was enabled).
Fixes a crash when deinterlacing is toggled during playback, and the
deinterlacer changes the stream format (as it can happen e.g. if the
decoder outputs nv12, which in turn happens with hw decoding).
(On a side note, future frames are always non-NULL. Also, the current
frame is of course always in the correct format.)
vaQueryImageFormats() returns a randomly ordered list - so we shouldn't
assume the first format on the list which works is the best. This
effectively switches to nv12 instead of yuv420p on some drivers.
We handle this by reusing va_to_imgfmt[], and ordering it by preference.
We hardcode that GPUs prefer nv12 pver yuv420p. In theory we could do
complicated probing (allocate dummy surface + use vaDeriveImage on it,
then retrieve the FourCC) - but all things which could break assumption
in the future are not supported yet (like 10 bit or 4:4:4), so this is
fine.
Fixes problems with --vo=opengl:interpolation. The issue here is that
vo_opengl retains more surfaces than what was preallocated for the
decoder. Until now, we just explicitly failed to decode frames for which
no additional surfaces are available. Since modern drivers usually are
fine with not "registering" surfaces before the decoder is created, just
allow allocating additional surfaces if needed.
(We also could probably recreate the HW decoder, since the HW decoder
should be stateless. But let's try to avoid raising the overall
complexity of the code.)
The interlaced frame test needs to be aware that the input mpi might be
NULL - this happens at the end of a stream when the input frames have
all been submitted but frames still need to be drained from the
decoder.
Outputting the detected OpenGL features was useless and redundant with
the extension loading output.
Also, remove MPGL_CAP_3D_TEX from OpenGL(ES) 3.0. This block didn't
include the glTexImage3D function, so that was pointless and couldn't
have worked. The OpenGL 2.1 block does it correctly.
VDPAU has explicit support for rotating surfaces, and it is far less
expensive than using the normal rotation filter (which would require
reading video frames back into system memory), it is desirable to
implement the VO rotation capability.
To do this, we need to render the video frames to an output surface,
without rotation, and then render from that surface to the final
output surface to apply the rotation. It is important that the
intermediate surface is the same size as the final one (only not
rotated) so that hqscaling can be applied if requested by the user.
(hqscaling is a mixer capability and so takes effect when the video
surface is rendered to an output surface)
Finally, we must remember to explicitly clear the final output
surface as VDPAU only auto-clears output surfaces when rendering video
surfaces.
Normally, vdpau decoded frames are passed directly to a suitable
vo (vo_vdpau or vo_opengl) without ever touching system memory. This
is efficient for output purposes, but prevents any of the regular
filters from being used with such frames.
This new filter implements a read-back step to pull the frames back
into system memory where they can be acted on by other filters.
Eventually the frames will be sent to the vo as if they were normal
software-decoded frames.
Note that a vdpau compatible vo must still be used to ensure that
the decoder is properly initialised.
Signed-off-by: wm4 <wm4@nowhere>
Some code called by vf_vdpaupp.c calls mp_image_new_custom_ref(), but
out of convenience doesn't reset the buffers. Make this behavior ok.
(The assert() was there to catch usage errors, but the same error could
already happen before the refcount changes were made, so the check is
not overly helpful.)
Fixes#2115.
Drop libva versions below 0.34.0. These are ancient, so I don't care.
Drop the vo_vaapi deinterlacer as well. With 0.34.0, VPP is always
available, and deinterlacing is done with vf_vavpp.
The vaCreateSurfaces() function changes its signature - actually it did
in 0.34.0 or so, and the <va/va_compat.h> defined a macro to make it use
the old signature.
Sometime recently, hardware decoding started to fail if h264 with full
reference frames was decoded, and --vo=vaapi was used. VAAPI requires
registering all surfaces that the decoder will ever use in advance, so
if the playback chain uses more surfaces than originally allocated, we
fail and drop back to software decoding.
I'm not really sure why or when this started happening. Commit 7b9d7265
for one is not the cause - it can be reproduced with earlier commits. It
also seems to be timing dependent. Possibly it has to do with the way
vo.c retains previous surfaces, and the way they can be queued/unqueued
asynchronously.
Increasing the number of reserved additional surfaces by 1 fixes it.
(Though I have no idea where exactly all these surfaces are being used.
Or rather, _when_.)
See manpage additions. This is mainly useful for vo_opengl_cb, but can
also be applied to vo_opengl.
On a side note, gl_hwdec_load_api() should stop using a name string, and
instead always use the IDs. This should be cleaned up another time.
Now there's a "canonical" table for mapping the names, that other code
can use, without having to rely too much on option code magic.
Also, use the central HWDEC constants, instead of magic values. (There
used to be semi-ok reasons to do this, but now it makes no sense
anymore.)
Basically, we need to make sure to allocate enough data for the pretty
dumb copy_nv12 function. (It could be avoided by making the function
less dumb, but this fix is simpler.)
mpv had refcounted frames before libav*, so we were not using
libavutil's facilities. Change this and drop our own code.
Since AVFrames are not actually refcounted, and only the image data
they reference, the semantics change a bit. This affects mainly
mp_image_pool, which was operating on whole images instead of buffers.
While we could work on AVBufferRefs instead (and use AVBufferPool),
this doesn't work for use with hardware decoding, which doesn't
map cleanly to FFmpeg's reference counting. But it worked out. One
weird consequence is that we still need our custom image data
allocation function (for normal image data), because AVFrame's uses
multiple buffers.
There also seems to be a timing-dependent problem with vaapi (the
pool appears to be "leaking" surfaces). I don't know if this is a new
problem, or whether the code changes just happened to cause it more
often. Raising the number of reserved surfaces seemed to fix it, but
since it appears to be timing dependent, and I couldn't find anything
wrong with the code, I'm just going to assume it's not a new bug.
This caused issues with hardware decoding. The VOs by definition dictate
the lifetime of the hardware context, so no surface allocations must
survive the VO. Fixes assertions on exit with vdpau.
This is basically a hack for drivers which prevent the mpv DXVA2 decoder
glue from working if OpenGL is in fullscreen mode.
Since it doesn't add any "hard" new API to the client API, some of the
code would be required for a true zero-copy hw decoding pipeline, and
sine it isn't too much code after all, this is probably acceptable.
When seeking to a different position, and seeking takes long, the OSD
might get redrawn. This means that the VO will receive a request to
redraw an old frame using whatever the previous PTS was. This breaks the
interpolation logic: the old frame will be added to the queue, and then
the next frames (with lower PTS if you seeked backwards) are not drawn
as the logic assumes they're past frames.
Fix this by using the non-interpolation code path when redrawing after a
seek reset, and no "real" frame has been drawn yet.
It's a recent regression caused by the redrawing code simplification.
The old code simply sent a VOCTRL for redrawing the frame, and the VO
had to deal with retaining the old frame on its own.
This is a hack as in there's probably a better solution.
Fixes#2097.
Less code, and avoids a black flash on start.
In theory it could happen that we map the window, and then don't have a
frame to draw - but mapping the window is done in the exact moment we
have a new frame to display.
This is not the most theoretically perfect solution, ideally we could
check to see if the frame in question has already been rendered
somewhere in the queue and then avoid re-rendering it, at the cost of a
few extra lines of code. But I don't think the performance trade-off is
dramatic enough here.
draw_image_timed is renamed to draw_frame. struct frame_timing is
renamed to vo_frame. flip_page_timed is merged into draw_frame (the
additional parameters are part of struct vo_frame). draw_frame also
deprecates VOCTRL_REDRAW_FRAME, and replaces it with a method that
works for both VOs which can cache the current frame, and VOs which
need to redraw it anyway.
This is preparation to making the interpolation and (work in progress)
display sync code saner.
Lots of other refactoring, and also some simplifications.
This should make interpolation work much better in general, although
there still might be some side effects for unusual framerates (eg. 35 Hz
or 48 Hz). Most of the common framerates are tested and working fine.
(24 Hz, 30 Hz, 60 Hz)
The new code doesn't have support for oversample yet, so it's been
removed (and will most likely be reimplemented in a cleaner way if
there's enough demand). I would recommend using something like robidoux
or mitchell instead of oversample, though - they're much
smoother for the common cases.
For now, this is trivial (and actually redundant). The future display
sync code will make better use of it. The main point is that the new
internal API pretty much makes this transparent to the vo_opengl
interpolation code.
Now the VO can request a number of future frames with the last parameter
of vo_set_queue_params(). This will be helpful to fix the interpolation
code.
Note that the first frame (after playback start or seeking) will usually
not have any future frames (to make seeking fast). Near the end of the
file, the number of future frames will become lower as well.
Again. With the old OpenGL interop dropped, this probably works better
than vaapi-copy now. Last time we defaulted to vaapi-copy, because the
OpenGL interop could swap U/V planes and other stupid crap. We'll see.
Work around that FFmpeg doesn't distinguish between surface and cropped
size. The decoder always aligns the surface size to something
"convenient" (e.g. 16 for h264), and to get to the correct cropped size,
the output image's width/height is reduced. Using the cropped size
instead of the real surface size breaks the libva API in certain cases,
so we simply store and use the original size in our per-surface struct.
(If data is cropped on the left/top borders, hw decoding will simply
display these - FFmpeg doesn't let us do better.)
In theory, this code path avoids a copy. In practice, it never seems
to get enabled at all. But it does have potential for weird bugs or
performance issues (like being mapped from non-cacheable memory),
so kill it.
Some window managers let you change the fullscreen state of any window
using a key combination. For example, on XFWM you can use Alt+F11 and
on Compiz you can configure a key combination with the
"Extra WM actions" plugin.
With this change mpv will handle these fullscreen state changes. So, if
you enter into fullscreen mode using the WM's shortcut and then you use
mpv's fullscreen toggle, you will get back into window mode.
Merges PR #2081.
Signed-off-by: wm4 <wm4@nowhere>
MPlayer traditionally had completely separate sh_ structs for
audio/video/subs, without a good way to share fields. This meant that
fields shared across all these headers had to be duplicated. This commit
deduplicates essentially the last remaining duplicated fields.
When using --hwdec=auto, about half of all systems will print:
"[vdpau] Error when calling vdp_device_create_x11: 1"
this happens because usually mpv will be linked against both vdpau and
vaapi libs, but the drivers are not necessarily available. Then trying
to load a driver will fail. This is a normal part of probing, but the
error messages were printed anyway. Silence them by explicitly
distinguishing probing.
This pretty much goes through all the layers. We actually consider
loading hw backends for vo_opengl always "auto probed", even if a hw
backend is explicitly requested. In this case vd_lavc will print a
warning message anyway (adjust this message a bit).
No particular reason, but it's still possible that it causes additional
corner cases, and it's not really needed to test this on wine (other
than testing fullscreen stuff, which should be done on a real Windows
anyway).
When showing cover art, the decoding logic pretends that the source has
an infinite number of frames. This slightly simplifies dealing with
filter data flow. It was done by feeding the same packet repeatedly to
the decoder (each decode run produces new output).
Change this by decoding once at the video initialization. This is easier
to follow, and increases robustness in case of broken images. Usually,
we try to tolerate decoding errors, so decoding normally continues, but
in this case it would just burn the CPU for no reason.
Fixes#2056.
gl_video_set_options() does not acquire ownership of the opts parameter
or its contents. In case of vo_cmdline, opts will point to temporary
memory. This memory will be free'd at a later point, and p->opts will
point to free'd memory on the next reinitialization.
The fix is pretty ugly, but it's a quick bug fix. This can probably be
removed once VO sub-options are exposed as properties.
Fixes#2035.
Absence of license header implies LGPL, as mentioned in the "Copyright"
file. But vaapi.h contains some code taken from the mplayer-vaapi
patch, which was under the typical MPlayer license.
All vo_gl.c related code has been GPL+LGPL dual-licensed. The OSD code
is no exception and is also derived from vo_gl.c. Thus it should have
the same license (although I think technically speaking sub-licensing
it by removing one of the licenses is ok).
Commits 92b27be and f4ce99d removed high-fps logic to to a bug. That bug was
a missing parenthesis around everything after duration >= 0 && ... at the
removed code.
This patch restores the removed code, fixes the bug and then refactors the
code a bit.
This reverts commit f1746741de.
Together with the other revert, this fixes#2023 (the reason being
broken framedrop handling - it was dropping frames when it shouldn't).
read_output_surface() could fail and return NULL.
Also, make sure we don't set the image to a size larger than the
allocated size. Normally this shouldn't happen, but in theory it could
in corner cases; this is for robustness.
We can't do much in this case, but at least we can not call the vdpau
API functions with too large sizes. Apparently the API considers this
undefined behavior, and random stuff might happen.
The previous code was not wrong, but I'd claim this makes the code more
robust. If a situation could happen in which the passed surface size is
incorrect, we could have passed a too small image, and
VdpOutputSurfaceGetBitsNative could have randomly overwritten memory.
Check the maximum size of video surfaces, and refuse initialization if
the video is too large for them.
Maybe we could do something more sophisticated, like inserting a
software scaler. On the other hand, this would have a very questionable
benefit, as it would be guaranteed to be too slow.
This was missing for extended deinterlacer.
Unfortunately, these deinterlacer still do not work. The provided future
frame (which is all the deinterlacers want) seems to be correct, though.
One minor behavioral change is that this always keeps the previous frame
for PTS computations. This could be avoided (in order to keep exactly
the same behavior as before), but it seems more elegant and should not
do any harm. (Also, if we really cared about reducing hw frame refs,
a more worthy goal is producing the field output incrementally.)
The mapped data (pointed to by the param variable) is not needed before,
so the call can be moved down. Also, this prevents that the buffer
remains mapped forever if the other vaMapBuffer() call above fails (the
cleanup code forgets to unmap the buffer - this commit makes it
unnecessary).
This used a do-while loop, which runs only once, as replacement for a
cleanup goto. While this is ok, doing a goto directly is easier to
follow and is closer to idiomatic C. But mainly remove it so that the
indentation can be reduced.
Reconfiguring with the same video size should never cause the window to
resize back to the video size (if the user changed its size). This was
broken and it resized anyway.
We do not fill them, so we would pass random IDs to the driver. The code
was originally written to handle bob deinterlacing only, so I guess it
originally passed always 0 anyway, despite having code for reference
surface list allocation.
Also, move down the vaUnmapBuffer() call. This call actually "unmaps"
the param pointer, so accessing it after the unmap call would be
undefined behavior. The "example" in <va/vavpp.h> does this too, but
it's most likely an error.
(Additionally, not even bob deinterlacing worked correctly in my test,
sigh.)
wm4
Emmanuel Gil Peyrot
Philip Langdale
Kevin Mitchell
Niklas Haas
Jari Vetoniemi
Rodger Combs
Sebastien Zwickert
Eduardo Sánchez Muñoz
rr-
Avi Halachmi (:avih)
Filip Milivojevic