We're moving to using timeline semaphores for synchronisation between
cuda and vulkan. It's only required with libplacebo-next right now,
but eventually our baseline libplacebo requirement will increase, so we
will need new enough nvidia headers that include the necessary
declarations.
This is not technically necessary, because we never touch the fd again
after passing to cuda, but having it there could lead to future code
accidentally using it.
All hwdecs should respect the probing flag and demote their lgoging to
verbose level, so that initialisation failures during probing do not
spam the user. I forgot to do this for the Vulkan hwdec.
The dmabuf-wayland vo has a stub ra implementation that doesn't
have a swapchain. That means that it's currently not safe to call
ra_vk_ctx_get on that ra_ctx, but it must be safe to call on all ra
implementations as this is how we discover if it is a vulkan ra.
This hasn't been an issue before because no Vulkan code paths would be
triggered when using dmabuf-wayland, but with the new vulkan hwdec, it
becomes possible to trigger when hwdecs are probed.
The libplacebo sync abstraction is deprecated and we should be using
the more explicit Vulkan semaphore helpers instead. This means that
more of the book keeping moves to our side, but it's not too bad.
There are two main things going on here:
1. After a lot of back and forth, I decided to write the new code with
timeline semaphores to streamline things, and that also means all
the variables are separate - which makes the #ifdefs easier to read.
Which is important because:
2. While pl_sync owned the exported fd/handle, pl_vulkan_sem does not,
so we are responsible for managing them. That means reversing the
previous logic - we now can pass an original fd to CUDA and never
think about it again, while we have to clean up a Win32 Handle
because CUDA will not take ownership.
Finding the prefix in the key is not enough, the key has to actually
start with the prefix. Otherwise a key like `uosc-font_scale` will not
only match the `uosc-` prefix, but also the `osc-` prefix, resulting
in a logged warning about `-font_scale` being an unknown key.
AV1 support in Vulkan is extremely bleeding edge - to the point that
the extension is not present in official Khronos releases, but it has
a reserved identifier and we can look it up with a string literal for
now.
This will be skipped and ignored if the driver doesn't support it, so
it's safe if/when the name changes later (it'll just never be activated
in that case).
I originally wrote this trying to avoid doing an explicit version check
on the headers, but it just makes things more confusing, and the
requirements harder to understand.
So, Vulkan interop now takes a dependency on the header release where
they finalised the video decode headers. VK_EXT_descriptor_buffer was
added in 1.3.235, so that's covered as well.
Along the way I fixed a bug in the waf build where it was depending
on libplacebo-next instead of libplacebo-decode.
ffmpeg was previously allocating images for frames as the code size,
rather than the presentation one (1088 vs 1080 in the most common
example). Using the coded size when wrapping images for libplacebo
resulted in incorrect scaling from 1088 -> 1080, but even using the
presentation size wasn't perfect, as discussed in the original
commit.
However, ffmpeg has now been updated to use the presentation size for
the frame images, after discussions that concluded this must be done
because there is no way for a frame consumer to fix the dimensions
without copying the frame.
With that ffmpeg change, we can just use the normal layout information
like all the other hwdecs.
Although we can support vulkan multiplane images, cuda lacks any such
support, and so cannot natively import such images for interop. It's
possible that we can do separate exports for each plane in the image
and have it work, but for now, we can selectively disable multiplane
when we know that we'll be consuming cuda frames.
As a reminder, even though cuda is the frame source, interop is one way
so the vulkan images have to be imported to cuda before we copy the
frame contents over.
This logic here is slightly more complex than I'd like but you can't
just set the flag blindly, as it will cause hwframes ctx creation to
fail if the format is packed or if it's planar rgb. Oh well.
Vulkan hwdec interop with the ffmpeg 6.1 vulkan code will require
additional features beyond those activated by libplacebo by default.
Enabling these features requires both requesting the features'
extensions and then explicitly turning on the features. libplacebo
handles detecting unsupported features and dropping them, to avoid
failing to create the vulkan device.
We then leave it to ffmpeg to decide if any missing features are
required for functionality, and error out if necessary.
As ffmpeg requires at least one bleeding edge extension (descriptor
buffers), all of this logic is gated on the presence of sufficiently
new Vulkan headers.
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
We will need the full ra_ctx to be able to look up all the state
required to initialise an ffmpeg vulkan hwcontext, so pass let's
pass the ra_ctx instead of just the ra.
This is motivated by a need to access it from vo_gpu_next's opengl
wrapping code, and justified by it being an inherent property of the GL
context itself,
It was unsafe to return pointer to memory that was freed on another
thread, just copy the string to caller owned sturcture.
Fixes crashes when displaying passes stats with gpu-next.
vo_dmabuf_wayland worked by allocating entries to a pool and then having
a lot of complex logic dealing with releasing buffers, pending entries,
etc. along with some other not so nice things. Instead, we can rewrite
this logic so that the wl_buffers created by the imported dmabuf is
instead stored in a linked list, wl_list. We can simply append our
buffers to the list when needed and destroy everything at the end. On
every frame, we can check the ids of our surfaces and reuse existing
buffers, so in practice there will only ever be a handful at a time.
Some other small changes were made in an attempt to organize the
vaapi/drmprime code a little better as well.
An important change is to always enforce at least a minimum number of
buffers. Certain formats would not make enough unique buffers, and this
results in flickering/artifacts occuring. The old way to attempt to deal
with this was to clear out all the existing buffers and remake them, but
this gets complicated and also didn't always work. An easy solution to
this is just create more buffers which appears to solve this problem.
The actual number needed is not really based on anything solid, but 8
is a reasonable number to create for the lifetime of a file and it seems
to do the trick.
Additionally, seeking/loading new files can result in flicker artificts
due to buffers being reused when they shouldn't. When that happens, we
flip a bool so all the buffers get destroyed in draw_frame to avoid any
visual glitches.
This just replaces the API calls to get rid of deprecation warnings, it
doesn't yet expand the enum, nor replace them by the proper options.
The translation from tone map modes to hybrid mix parameters is taken
from the libplacebo source code.
Previously, if vo_drm_init failed at the start of drm_egl_init it
caused a use-after-free in drm_egl_uninit when it tried to access the
non-existant drm context. At that point vo_drm_uninit had already been
called resulting in two different null pointer dereference. In this
situation the DRM private context had also not been allocated.
This matches the behaviour that libmpv API clients expect better and
prevents silent regressons with watch_later suddenly appearing in
an entirely different path (or none at all in case of weird platforms*).
Also added some comments for easier understanding.
* Android
Previously, we'd only attempt to call guess_lang_from_filename if the
external file name matched the video name ignoring the extensions. So if
they didn't match, we'd just report the language as "unknown". And since
the name will never match for urls, the language would always be treated
as unknown.
Now we'll always try to guess the language from the filename regardless
of its similarity to the video file name.
Closes#10348
Filtering globally D3D11_MESSAGE_ID_CREATETEXTURE2D_INVALIDDIMENSIONS is
suboptimal, because can also hide other invalid usages. In the same time
it is not enough, because not only this message is emitted, but also one
about E_INVALIDARG. Just flush queue before and clear messages after to
ignore this part of code.
As a side note, I don't believe this texture size lookup is in fact
useful, but since it is there and is relatively harmless, let's leave
it as is.
The current implementation is order dependent and assumes that getting
keyboard input happens before the toplevel is activated. This isn't
necessarily the case and indeed mutter activates the toplevel first.
Improve this by simply spinning off the check to a function and calling
it in the three places where it would be needed: the toplevel
configuration event, keyboard entering, and keyboard leaving. This
fixes#11694.
Pass "dummy.m3u8" filename to work around FFmpeg commit
6b1f68ccb04d791f0250e05687c346a99ff47ea1 which broke their HLS demuxer
and its ability to probe.
Since the above commit, libavformat will check the filename of the file
to be probed and reject it if it doesn't end with a valid HLS extension
i.e. m3u8,hls,m3u (never mind that .hls is not a valid HLS extension).
In addition to a bug with query strings, this also breaks mpv
functionality as mpv explicitly doesn't tell libavformat the filename
when probing, in order to properly detect the file based only on their
contents.
The [HLS specification](https://www.rfc-editor.org/rfc/rfc8216.txt) aka
RFC 8216, specifies in section 4 that "Each Playlist file MUST be
identifiable either by the path component of its URI or by HTTP
Content-Type." Notably, it does not require both, so this FFmpeg commit
is noncompliant. We work around this noncompliance by checking the MIME
type ourselves. If the mimetype matches one of the valid HLS mimetypes
(and also application/x-mpegurl, a legacy pre-standardization type),
then we pass "dummy.m3u8" to libavformat in order to work around its
overly strict checking of filenames.
Without this patch, we are unable to play many HLS streams, including a
few from the ytdl hook. This patch restores those ability to play those
streams when built against FFmpeg master. Do note that if the server
sends an invalid content-type header then we cannot implement this
workaround so those streams will still fail to play.
This adds cache as a possible path for mpv to internally pick
(~/.cache/mpv for non-darwin unix-like systems, the usual config
directory for everyone else). For gpu shader cache and icc cache,
controlling whether or not to write such files is done with the new
--gpu-shader-cache and --icc-cache options respectively. Additionally,
--cache-on-disk no longer requires explicitly setting the --cache-dir
option. The old options, --cache-dir, --gpu-shader-cache-dir, and
--icc-cache-dir simply set an override for the directory to save cache
files. If unset, then the cache is saved in XDG_CACHE_HOME.
A pain point for some users is the fact that watch_later is stored in
the ~/.config directory when it's really not configuration data. Roughly
2 years ago, XDG_STATE_DIR was added to the XDG Base Directory
Specification[0] and its description, user-specific state data, actually
perfectly matches what watch_later data is for. Let's go ahead and use
this directory as the default for watch_later. This change only affects
non-darwin unix-like systems (i.e. Linux, BSDs, etc.). The directory
doesn't move for anyone else.
Internally, quite a few things change with regards to the path
selection. If the platform in question does not have a statedir concept,
then the path selection will simply return "home" instead (old
behavior). Fixes#9147.
[0]: 4f2884e16d
Currently, nothing new is actually implemented but the idea is simply to
just pass a type string all the way up from mp_find_user_file down to
actually getting the platform path. This allows for selecting different
directories besides the user's native config directory. See the next
commit for an implementation.
macOS really has completely different path conventions that mpv doesn't
take into account and it treats it just like any other old unix-like
system. This means mpv enforces certain conventions on it (like all the
XDG stuff) that doesn't really apply. Since we'd like to use more of
this but at the same time not distrupt mac users even more, let's just
copy and paste the current code to a new file, update the build and call
it a day. This way, the paths of these two platforms can more freely
diverge.
Philip Langdale