We generally try to avoid that due to the #ifdef mess.
The equivalent format is defined in ffmpeg 4.4 while our interop code
requires a higher version, but that doesn't cause any problems.
Vulkan Video Decoding has finally become a reality, as it's now
showing up in shipping drivers, and the ffmpeg support has been
merged.
With that in mind, this change introduces HW interop support for
ffmpeg Vulkan frames. The implementation is functionally complete - it
can display frames produced by hardware decoding, and it can work with
ffmpeg vulkan filters. There are still various caveats due to gaps and
bugs in drivers, so YMMV, as always.
Primary testing has been done on Intel, AMD, and nvidia hardware on
Linux with basic Windows testing on nvidia.
Notable caveats:
* Due to driver bugs, video decoding on nvidia does not work right now,
unless you use the Vulkan Beta driver. It can be worked around, but
requires ffmpeg changes that are not considered acceptable to merge.
* Even if those work-arounds are applied, Vulkan filters will not work
on video that was decoded by Vulkan, due to additional bugs in the
nvidia drivers. The filters do work correctly on content decoded some
other way, and then uploaded to Vulkan (eg: Decode with nvdec, upload
with --vf=format=vulkan)
* Vulkan filters can only be used with drivers that support
VK_EXT_descriptor_buffer which doesn't include Intel ANV as yet.
There is an MR outstanding for this.
* When dealing with 1080p content, there may be some visual distortion
in the bottom lines of frames due to chroma scaling incorporating the
extra hidden lines at the bottom of the frame (1080p content is
actually stored as 1088 lines), depending on the hardware/driver
combination and the scaling algorithm. This cannot be easily
addressed as the mechanical fix for it violates the Vulkan spec, and
probably requires a spec change to resolve properly.
All of these caveats will be fixed in either drivers or ffmpeg, and so
will not require mpv changes (unless something unexpected happens)
If you want to run on nvidia with the non-beta drivers, you can this
ffmpeg tree with the work-around patches:
* https://github.com/philipl/FFmpeg/tree/vulkan-nvidia-workarounds
This mapping isn't actually relevant until we have the Vulkan interop
merged, and it requires a newer version of libavutil than our minimum
requirement. So I'm going to remove it from master and put it in the
interop PR.
Fixes#10813
Certain combinations of hardware formats require the use of hwmap to
transfer frames between the formats, rather than hwupload, which will
fail if attempted.
To keep the usage of vf_format for HW -> HW transfers as intuitive as
possible, we should detect these cases and do the map operation instead
of uploading.
For now, the relevant cases are moving between VAAPI and Vulkan, and
VAAPI and DRM Prime, in both directions. I have introduced the IMGFMT
entry for Vulkan here so that I can put in the complete mapping table.
It's actually not useless, as you can map to Vulkan, use a Vulkan
filter and then map back to VAAPI for display output.
IMGFMT_RGB30 was added first; FFmpeg added AV_PIX_FMT_X2RGB10 later.
This is exactly the same, so treat them as such. For some reason,
libswscale still seems to output incompatible data - not sure what this
is about, but I'm not going to debug it.
Libav seems rather dead: no release for 2 years, no new git commits in
master for almost a year (with one exception ~6 months ago). From what I
can tell, some developers resigned themselves to the horrifying idea to
post patches to ffmpeg-devel instead, while the rest of the developers
went on to greener pastures.
Libav was a better project than FFmpeg. Unfortunately, FFmpeg won,
because it managed to keep the name and website. Libav was pushed more
and more into obscurity: while there was initially a big push for Libav,
FFmpeg just remained "in place" and visible for most people. FFmpeg was
slowly draining all manpower and energy from Libav. A big part of this
was that FFmpeg stole code from Libav (regular merges of the entire
Libav git tree), making it some sort of Frankenstein mirror of Libav,
think decaying zombie with additional legs ("features") nailed to it.
"Stealing" surely is the wrong word; I'm just aping the language that
some of the FFmpeg members used to use. All that is in the past now, I'm
probably the only person left who is annoyed by this, and with this
commit I'm putting this decade long problem finally to an end. I just
thought I'd express my annoyance about this fucking shitshow one last
time.
The most intrusive change in this commit is the resample filter, which
originally used libavresample. Since the FFmpeg developer refused to
enable libavresample by default for drama reasons, and the API was
slightly different, so the filter used some big preprocessor mess to
make it compatible to libswresample. All that falls away now. The
simplification to the build system is also significant.
This is fragile enough that it warrants getting "monitored".
This takes the commented test program code from img_format.c, makes it
output to a text file, and then compares it to a "ref" file stored in
git.
Originally, I wanted to do the comparison etc. in a shell or Python
script. But why not do it in C. So mpv calls /usr/bin/diff as a
sub-process now.
This test will start producing different output if FFmpeg adds new pixel
formats or pixel format flags, or if mpv adds new IMGFMT (either aliases
to FFmpeg formats or own formats). That is unavoidable, and requires
manual inspection of the results, and then updating the ref file.
The changes in the non-test code are to guarantee that the format ID
conversion functions only translate between valid IDs.
The FFmpeg version was last bumped a long time ago, except in commit
1638fa7b46, where it was used for some obscure pixel
format.
This is pretty annoying, so make it optional.
New releases of VDPAU support decoding 4:4:4 content, and that comes
back as NV24 when using 'direct mode' in OpenGL Interop. That means we
need to be a little bit smarter about how we set up the OpenGL
textures.
Using the GL renderer for color conversion will make sure screenshots
will use the same conversion as normal video rendering. It can do this
for all types of screenshots.
The logic when to write 16 bit PNGs changes. To approximate the old
behavior, we decide by looking whether the source video format has more
than 8 bits per component. We apply this logic even for window
screenshots. Also, 16 bit PNGs now always include an unused alpha
channel. The reason is that FFmpeg has RGB48 and RGBA64 formats, but no
RGB064. RGB48 is 3 bytes and usually not supported by GPUs for
rendering, so we have to use RGBA64, which forces an alpha channel.
Will break for users who use --target-trc and similar options.
I considered creating a new gl_video context, but it could double GPU
memory use, so I didn't.
This uses FBOs instead of glGetTexImage(), because that increases the
chance it could work on GLES (e.g. ANGLE). Untested. No support for the
Vulkan and D3D11 backends yet.
Fixes#5498. Also fixes#5240, because the code for reading back is not
used with the new code path.
Get rid of the old vf.c code. Replace it with a generic filtering
framework, which can potentially handle more than just --vf. At least
reimplementing --af with this code is planned.
This changes some --vf semantics (including runtime behavior and the
"vf" command). The most important ones are listed in interface-changes.
vf_convert.c is renamed to f_swscale.c. It is now an internal filter
that can not be inserted by the user manually.
f_lavfi.c is a refactor of player/lavfi.c. The latter will be removed
once --lavfi-complex is reimplemented on top of f_lavfi.c. (which is
conceptually easy, but a big mess due to the data flow changes).
The existing filters are all changed heavily. The data flow of the new
filter framework is different. Especially EOF handling changes - EOF is
now a "frame" rather than a state, and must be passed through exactly
once.
Another major thing is that all filters must support dynamic format
changes. The filter reconfig() function goes away. (This sounds complex,
but since all filters need to handle EOF draining anyway, they can use
the same code, and it removes the mess with reconfig() having to predict
the output format, which completely breaks with libavfilter anyway.)
In addition, there is no automatic format negotiation or conversion.
libavfilter's primitive and insufficient API simply doesn't allow us to
do this in a reasonable way. Instead, filters can use f_autoconvert as
sub-filter, and tell it which formats they support. This filter will in
turn add actual conversion filters, such as f_swscale, to perform
necessary format changes.
vf_vapoursynth.c uses the same basic principle of operation as before,
but with worryingly different details in data flow. Still appears to
work.
The hardware deint filters (vf_vavpp.c, vf_d3d11vpp.c, vf_vdpaupp.c) are
heavily changed. Fortunately, they all used refqueue.c, which is for
sharing the data flow logic (especially for managing future/past
surfaces and such). It turns out it can be used to factor out most of
the data flow. Some of these filters accepted software input. Instead of
having ad-hoc upload code in each filter, surface upload is now
delegated to f_autoconvert, which can use f_hwupload to perform this.
Exporting VO capabilities is still a big mess (mp_stream_info stuff).
The D3D11 code drops the redundant image formats, and all code uses the
hw_subfmt (sw_format in FFmpeg) instead. Although that too seems to be a
big mess for now.
f_async_queue is unused.
This commit allows to use the AV_PIX_FMT_DRM_PRIME newly introduced
format in ffmpeg that allows decoders to provide an AVDRMFrameDescriptor
struct.
That struct holds dmabuf fds and information allowing zerocopy rendering
using KMS / DRM Atomic.
This has been tested on RockChip ROCK64 device.
Now you need FFmpeg git, or something.
This also gets rid of the last real use of gpu_memcpy(). libavutil does
that itself. (vaapi.c still used it, but it was essentially unused,
because the code path isn't really in use anymore. It wasn't even
included due to the d3d-hwaccel dependency in wscript.)
It never worked. It relied on some obscure texture format to provide the
equivalent of GL_RG or GL_LUMINANCE_ALPHA, but no hardware seemed to
report support for it ever. No idea what's the correct way to do this.
On D3D11 it exists, of course.
(Actually I'd like to remove the whole VO.)
Another legacy annoyance. The only place where packed YUV is still
important is slightly older Apple hardware or drivers, which require
it for efficient hardware decoding.
For vo_opengl and vo_direct3d, these are supported in a generic way.
For vf_vapoursynth, we could probably map its VSFormat struct in a
generic way, but for now do some bullshit.
vf_eq.c actually loses support for these formats. We could add generic
support too (anything that has 8 bit planes will work), but why bother.
The filter is deprecated anyway.
The problem with fmt-conversion.h is that "lucabe", who disagreed with
LGPL, originally wrote it. But it was actually rewritten by "reimar"
later. The original switch statement was replaced with a lookup table.
No code other than the imgfmt2pixfmt() function signature survives.
Neither the format pairs (PIXFMT<->IMGFMT), nor the concept of mapping
them, can be copyrighted.
So changing the license should be fine, because reimar and all other
authors involved with the new code agreed to LGPL.
We also don't consider format pairs added later as copyrightable.
(The direct-mapping idea mentioned in the "Copyright" file seems
attractive, and I might implement in later anyway.)
Likewise, there might be some format names added to img_format.h, which
are not covered by relicensing agreements. These all affect "later"
additions, and they follow either the FFmpeg PIXFMT naming or some other
pre-existing logic, so this should be fine.
Unfortunately quite a mess, in particular due to the need to have some
compatibility with the old API. (The old API will be supported only in
short term.)
The latest 375.xx nvidia drivers add support for P016 output
surfaces. In combination with an ffmpeg change to return those
surfaces, we can display them.
The bulk of the work is related to knowing which format you're
dealing with at the right time. Once you know, it's straight forward.
'cuda-gl' isn't right - you can turn this on without any GL and
get some non-zero benefit (with the cuda-copy hwaccel). So
'cuda-hwaccel' seems more consistent with everything else.
Nvidia's "NvDecode" API (up until recently called "cuvid" is a cross
platform, but nvidia proprietary API that exposes their hardware
video decoding capabilities. It is analogous to their DXVA or VDPAU
support on Windows or Linux but without using platform specific API
calls.
As a rule, you'd rather use DXVA or VDPAU as these are more mature
and well supported APIs, but on Linux, VDPAU is falling behind the
hardware capabilities, and there's no sign that nvidia are making
the investments to update it.
Most concretely, this means that there is no VP8/9 or HEVC Main10
support in VDPAU. On the other hand, NvDecode does export vp8/9 and
partial support for HEVC Main10 (more on that below).
ffmpeg already has support in the form of the "cuvid" family of
decoders. Due to the design of the API, it is best exposed as a full
decoder rather than an hwaccel. As such, there are decoders like
h264_cuvid, hevc_cuvid, etc.
These decoders support two output paths today - in both cases, NV12
frames are returned, either in CUDA device memory or regular system
memory.
In the case of the system memory path, the decoders can be used
as-is in mpv today with a command line like:
mpv --vd=lavc:h264_cuvid foobar.mp4
Doing this will take advantage of hardware decoding, but the cost
of the memcpy to system memory adds up, especially for high
resolution video (4K etc).
To avoid that, we need an hwdec that takes advantage of CUDA's
OpenGL interop to copy from device memory into OpenGL textures.
That is what this change implements.
The process is relatively simple as only basic device context
aquisition needs to be done by us - the CUDA buffer pool is managed
by the decoder - thankfully.
The hwdec looks a bit like the vdpau interop one - the hwdec
maintains a single set of plane textures and each output frame
is repeatedly mapped into these textures to pass on.
The frames are always in NV12 format, at least until 10bit output
supports emerges.
The only slightly interesting part of the copying process is that
CUDA works by associating PBOs, so we need to define these for
each of the textures.
TODO Items:
* I need to add a download_image function for screenshots. This
would do the same copy to system memory that the decoder's
system memory output does.
* There are items to investigate on the ffmpeg side. There appears
to be a problem with timestamps for some content.
Final note: I mentioned HEVC Main10. While there is no 10bit output
support, NvDecode can return dithered 8bit NV12 so you can take
advantage of the hardware acceleration.
This particular mode requires compiling ffmpeg with a modified
header (or possibly the CUDA 8 RC) and is not upstream in ffmpeg
yet.
Usage:
You will need to specify vo=opengl and hwdec=cuda.
Note that hwdec=auto will probably not work as it will try to use
vdpau first.
mpv --hwdec=cuda --vo=opengl foobar.mp4
If you want to use filters that require frames in system memory,
just use the decoder directly without the hwdec, as documented
above.
We don't have any reason to disable either. Both are loaded dynamically
at runtime anyway. There is also no reason why dxva2 would disappear
from libavcodec any time soon.
This commit adds the d3d11va-copy hwdec mode using the ffmpeg d3d11va
api. Functions in common with dxva2 are handled in a separate decode/d3d.c
file. A future commit will rewrite decode/dxva2.c to share this code.
Some VOs had support for these - remove them.
Typically, these formats will have only some use in cases where using
RGB software conversion with libswscale is faster than letting the
VO/GPU do it (i.e. almost never). For the sake of testing this case,
keep IMGFMT_RGB565. This is the least messy format, because it has no
padding/alpha bits with unknown semantics.
Note that decoding to these formats still works. We'll let libswscale
repack the data to whatever the VO in use can take.
This removes the need to define IMGFMT_GBRAP, which fixes compilation
with the current Libav release.
This also makes it automatically pick up a GBRP format with the same bit
width. (Unfortunately, it seems libswscale does not support conversion
to AV_PIX_FMT_GBRAP16, so our code falls back to 8 bit, removing
precision for video covered by subtitles in cases this code is used.)
Also, when the source video is e.g. 10 bit YUV, upsample to 16 bit.
Whether this is good or bad, it fixes behavior with alpha. Although I'm
not sure if the alpha range is really correct ([0,2^16-1] vs.
[0,255*256]). Keep in mind that libswscale doesn't even agree with the
way we do it.
This actually treats destination alpha correctly, and gives much better
results than before. I don't know if this is perfectly correct yet,
though. Slight difference with vo_opengl behavior suggests it might not
be.
Note that this does not affect VOs with true alpha support. vo_opengl
does not use this code at all, and does the alpha calculations in OpenGL
instead.
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.
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>
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).
The intention is that we can test vo_opengl with high bit depth PNGs
better. This throws libswscale completely out of the loop, which before
was needed in order to convert from big endian to little endian.
Also apply a minimal cleanup to fmt-conversion.c (unrelated).
Simply clamp off the U/V components in the colormatrix, instead of doing
something special in the shader.
Also, since YA8/YA16 gave a plane_bits value of 16/32, and a colormatrix
calculation overflowed with 32, add a component_bits field to the image
format descriptor, which for YA8/YA16 returns 8/16 (the wrong value had
no bad consequences otherwise).
These formats are still supported; you just can't reference them via a
defined constants directly. They are now handled via the generic
passthrough.
(If you want to use such a format, you either have to add the entry
back, or use AV_PIX_FMT_* directly.)
This is a rather radical change: instead of maintaining a whitelist of
FFmpeg formats we support, we automatically support all formats.
In general, a format which doesn't have an explicit IMGFMT_* name will
be converted to a known format through libswscale, or will be handled
by code which can treat pixel formats in a generic way using the pixel
format description, like vo_opengl.
AV_PIX_FMT_UYYVYY411 is a special-case. It's packed YUV with chroma
subsampling by 4 in both directions. Its component order is documented
as "Cb Y0 Y1 Cr Y2 Y3", meaning there's one UV sample for 4 Y samples.
This means each pixel uses 1.5 bytes (4 pixels have 1 UV sample, so
4 bytes + 2 bytes). FFmpeg can actually handle this format with its
generic mechanism in an extremely awkward way, but it doesn't work for
us. Blacklist it, and hope no similar formats will be added in the
future.
Currently, the AV_PIX_FMT_*s allowed are limited to a numeric value of
500. More is not allowed, and there are some fixed size arrays that need
to contain any possible format (look for IMGFMT_END dependencies).
We could have this simpler by replacing IMGFMT_* with AV_PIX_FMT_*
through the whole codebase. But for now, this is better, because we
can compensate for formats missing in Libav or older FFmpeg versions,
like AV_PIX_FMT_RGB0 and others.
Since the new hwaccel API is now merged in ffmpeg's stable release, we can
finally remove support for the old API.
I pretty much kept lu_zero's new code unchanged and just added some error
printing (that we had with the old glue code) to make the life of our users
less miserable.
This affects packed RGB formats up to 16 bits per pixel. The old mplayer
names used LSB-to-MSB order, while FFmpeg (and some other libraries) use
MSB-to-LSB.
Nothing should change with this commit, i.e. no bit order or endian bugs
should be added or fixed. In some cases, the name stays the same, even
though the byte order changes, e.g. RGB8->BGR8 and BGR8->RGB8, and this
affects the user-visible names too; this might cause confusion.
sfan5
Philip Langdale
Thomas Weißschuh
wm4
Lionel CHAZALLON
Aman Gupta
Kevin Mitchell
Sebastien Zwickert
Marcin Kurczewski
Stefano Pigozzi
Luca Barbato