For hwdecs without devices we can't get instant feedback during probing
whether the hwdec can possibly work or not, so we definitely want to try them last.
Independently this would also have solved the issue addressed by the previous commit.
This fixes an issue where mpv would try mediacodec (which may not
be available depending on Android version) and pick v4l2m2m next
which ends up failing even though mediacodec-copy would have been available.
c784820454 introduced a bool option type
as a replacement for the flag type, but didn't actually transition and
remove the flag type because it would have been too much mundane work.
--vd-lavc-dr defaulted to "yes", which caused issues on certain
hardware. Instead of disabling it, add a new "auto" value and
make it the default.
The "auto" choice will enable DR only when we can request host-cached
buffers (as signalled by the new VO_DR_FLAG_HOST_CACHED).
Co-authored-by: Nicolas F. <ovdev@fratti.ch>
Co-authored-by: Niklas Haas <git@haasn.dev>
This new hwaccel flag was added to allow us to request that an hwaccel
decode not implicitly copy output frames before passing them out. The
only hwaccel that would implicitly copy frames is nvdec due to how it
has a small output pool that cannot be grown very much. However, we
already copy frames as soon as we get our hands on them (in our hwdec)
so we're already safe from pool exhaustion, and the extra copy doesn't
help us.
This has been a long standing annoyance - ffmpeg is removing
sizeof(AVPacket) from the API which means you cannot stack-allocate
AVPacket anymore. However, that is something we take advantage of
because we use short-lived AVPackets to bridge from native mpv packets
in our main decoding paths.
We don't think that switching these to `av_packet_alloc` is desirable,
given the cost of heap allocation, so this change takes a different
approach - allocating a single packet in the relevant context and
reusing it over and over.
That's fairly straight-forward, with the main caveat being that
re-initialising the packet is unintuitive. There is no function that
does exactly what we need (what `av_init_packet` did). The closest is
`av_packet_unref`, which additionally frees buffers and side-data.
However, we don't copy those things - we just assign them in from our
own packet, so we have to explicitly clear the pointers before calling
`av_packet_unref`. But at least we can make a wrapper function for
that.
The weirdest part of the change is the handling of the vtt subtitle
conversion. This requires two packets, so I had to pre-allocate two in
the context struct. That sounds excessive, but if allocating the
primary packet is too expensive, then allocating the secondary one for
vtt subtitles must also be too expensive.
This change is not conditional as heap allocated AVPackets were
available for years and years before the deprecation.
It turns out that it's generally more useful to look up hwdecs by image
format, rather than device type. In the situations where we need to
find one, we generally know the image format we're dealing with. Doing
this avoids us having to create mappings from image format to device
type.
The most significant part of this change is filling in the image format
for the various hw interops. There is a hw_imgfmt field today today, but
only a couple of the interops fill it in, and that seems to be because
we've never actually used this piece of metadata before. Well, now we
have a good use for it.
This fixes build failures with avcodec 58.113.100 or before,
matching FFmpeg release versions 4.0 to 4.3.
This flag was added in between avcodec 58.113.100 and 58.114.100
during the FFmpeg 4.4 development cycle. It lacks its own version bump,
so instead a check for the define is utilized instead.
Additionally, warn the user if they request GPU film grain with
too old of an FFmpeg.
Fixes#10088
The VO is available during decoder initialization mostly for direct
rendering purposes, so if f.ex. a complex filter chain is utilized,
there is no video renderer information available via
mp_filter_find_stream_info during creation of the decoder filter.
Thus, check for whether the VO is available before attempting to
check the capabilities flag from it.
Additionally - to simplify logic - makes explicitly requesting GPU
film grain to always disable decoder film grain functionality. The
warning is still shown if the VO is available and no support for
film grain application is available.
Fixes#10079
The field has been deprecated, yet the upcoming new default is not yet
the default. Thus, until lavc major hits 60 and the default behavior
finally gets changed, we have to explicitly set the field's value.
The deprecation had already been handled by adding the required
version limitation for this code in bbbf3571ed ,
this change merely just removes the warning which would otherwise
appear until lavc major version gets bumped to 60.
When I introduced the concept of lazy loading of hwdecs by img format,
I did not propagate the probing flag correctly, leading to the new
normal loading path not runnng with probing set, meaning that any
errors would show up, creating unnecessary noise.
This change fixes this regression.
Previously, when mpv was invoked with unsupported hwdec value, such
as --hwdec=foobar, there was no indication that it doesn't exist.
The reason it's not validated during options parsing is that the name
is only evaluated when selecting hwdec for actual decoding, by
matching it against runtime list of names from ffmpeg.
Additionally, when selecting hwdec for decoding, matching the name
came after filtering by codec, hence overall never-matched-name did
not necessarily indicate it's unsupported (doesn't exist at all).
Now we check the name before filtering by codec, and when done,
warn if no hwdec with that name exists at all.
This means that an unsupported name will now generate such warning
whenever we try to choose a hwdec, i.e. possibly more than once.
It's much better than no notification at all, and arguably adequate
for a sort of configuration error (linked ffmpeg has no such hwdec
name) which we don't validate during option parsing.
Historically, we have treated hwdec interop loading as a completely
separate step from loading the hwdecs themselves. Some hwdecs need an
interop, and some don't, and users generally configure the exact
hwdec they want, so interops that aren't relevant for that hwdec
shouldn't be loaded to save time and avoid warning/error spam.
The basic approach here is to recognise that interops are tied to
hwdecs by imgfmt. The hwdec outputs some format, and an interop is
needed to get that format to the vo without read back.
So, when we try to load an hwdec, instead of just blindly loading all
interops as we do today, let's pass the imgfmt in and only load
interops that work for that format. If more than one interop is
available for the format, the existing logic (whatever it is) will
continue to be used to pick one.
We also have one callsite in filters where we seem to pre-emptively
load all the interops. It's probably possible to trace down a specific
format but for now I'm just letting it keep loading all of them; it's
no worse than before.
You may notice there is no documentation update - and that's because
the current docs say that when the interop mode is `auto`, the interop
is loaded on demand. So reality now reflects the docs. How nice.
Today, validation is only possible for string type options. But there's
no particular reason why it needs to be restricted in this way, and
there are potential uses, to allow other options to be validated
without forcing the option to have to reimplement parsing from
scratch.
The first part, simply making the validation function an explicit
field instead of overloading priv is simple enough. But if we only do
that, then the validation function still needs to deal with the raw
pre-parsed string. Instead, we want to allow the value to be parsed
before it is validated. That in turn leads to us having validator
functions that should be type aware. Unfortunately, that means we need
to keep the explicit macro like OPT_STRING_VALIDATE() as a way to
enforce the correct typing of the function. Otherwise, we'd have to
have the validator take a void * and hope the implementation can cast
it correctly.
For help, we don't have this problem, as help doesn't look at the
value.
Then, we turn validators that are really help generators into explicit
help functions and where a validator is help + validation, we split
them into two parts.
I have, however, left functions that need to query information for both
help and validation as single functions to avoid code duplication.
In this change, I have not added an other OPT_FOO_VALIDATE() macros as
they are not needed, but I will add some in a separate change to
illustrate the pattern.
This was optional, with the intention that normally such options require
a valid format. But there is no reason for this (at least not anymore),
and it's actually more logical to accept "no" in all situations this
option type is used. This also gets rid of the weird min field special
use.
The "rule" is that a fallback warning message should be shown only shown
if software decoding was used before, or in other words when either
hwdec was enabled before, but the stream suddenly falls back, or it was
attempted to enable it at runtime, and it didn't work.
The message wasn't printed the first time in the latter case, because
hwdec_notified was not set in forced software decoding mode. Fix it with
this commit. Fortunately, the logic becomes simpler.
Change all OPT_* macros such that they don't define the entire m_option
initializer, and instead expand only to a part of it, which sets certain
fields. This requires changing almost every option declaration, because
they all use these macros. A declaration now always starts with
{"name", ...
followed by designated initializers only (possibly wrapped in macros).
The OPT_* macros now initialize the .offset and .type fields only,
sometimes also .priv and others.
I think this change makes the option macros less tricky. The old code
had to stuff everything into macro arguments (and attempted to allow
setting arbitrary fields by letting the user pass designated
initializers in the vararg parts). Some of this was made messy due to
C99 and C11 not allowing 0-sized varargs with ',' removal. It's also
possible that this change is pointless, other than cosmetic preferences.
Not too happy about some things. For example, the OPT_CHOICE()
indentation I applied looks a bit ugly.
Much of this change was done with regex search&replace, but some places
required manual editing. In particular, code in "obscure" areas (which I
didn't include in compilation) might be broken now.
In wayland_common.c the author of some option declarations confused the
flags parameter with the default value (though the default value was
also properly set below). I fixed this with this change.
Before this commit, option declarations used M_OPT_MIN/M_OPT_MAX (and
some other identifiers based on these) to signal whether an option had
min/max values. Remove these flags, and make it use a range implicitly
on the condition if min<max is true.
This requires care in all cases when only M_OPT_MIN or M_OPT_MAX were
set (instead of both). Generally, the commit replaces all these
instances with using DBL_MAX/DBL_MIN for the "unset" part of the range.
This also happens to fix some cases where you could pass over-large
values to integer options, which were silently truncated, but now cause
an error.
This commit has some higher potential for regressions.
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.
Add an "auto-safe" mode, mostly triggered by Ubuntu's nonsense to force
hwdec=vaapi in the global config file in their mpv package. But to be
honest it's probably something more people want.
This is implemented as explicit whitelist. On Windows, HEVC/Intel is
sometimes broken, but it's still whitelisted, and in theory we'd need a
detailed whitelist of device names etc. (like for example browsers tend
to do). On OSX, videotoolbox is a pretty bad choice, but unfortunately
the only one, so it's whitelisted too. There may be a larger number of
hwdec wrappers that work anyway, and I'm for example ignoring Android.
libavcodec's nvdec wrapper can return invalid frames, that do not have
any data fields set. This is not allowed by the API, but why would they
follow their own API?
Add a workaround to specifically detect this situation. In practice,
this should fall back to software decoding if it happens too often in a
row. (But single errors are still tolerated, because I don't know why.)
Untested due to lack of hardware from the regrettable graphics company.
Better do this here than deal with the moronic project we unfortunately
depend on.
See: #7185
This assert() sometimes triggered (and still triggers) with lavc API
bugs. It tries to check that at least 1 plane is set to a non-NULL
value. Obviously, a valid frame returned by successful decoding should
never have it.
The problem is that some hwdecs use integer surface IDs cast to a
pointer. Recently, it happened that newer Intel drivers started using
surface ID 0 under certain circumstances (for unknown reasons), which
triggers this assert.
Just get rid of it.
For the sake of #7185, add an assert() specifically for nvdec. That
failure needs to be further analyzed, is probably a FFmpeg bug, and
without this assert() would just crash somewhere further down the video
chain.
Fixes: #7261
This code checked AVFrame.buf[0] instead of the decode return code to
see whether a frame was decoded. This is sort of suspicious; while I
think that the lavc API actually guarantees it, it's not intuitive
anyway. In addition, the code was unnecessarily roundabout.
Replace it with a proper error code check. Remove the other error return
(that was, or should have been, redundant before). The no-frame path is
now cleanly separated. Add an assert on the frame-returned path; if this
fails, lavc violated its own API.
This is preparation to get rid of the option-to-property bridge
(mp_on_set_option). This is a pretty insane thing that redirects
accesses to options to properties. It was needed in the ever ongoing
transition from something to... something else.
A good example for the need of this bridge is applying profiles at
runtime. This obviously goes through the config parser, but should also
make all changes effective, for which traditionally the property layer
is used.
There isn't much left that needs this bridge. This commit changes a
bunch of options (which also have a property implementation) to use
option change notifications instead. Many of the properties are still
left, but perform unrelated functions like OSD formatting.
This should be mostly compatible. There may be some subtle behavior
changes. For example, "hwdec" and "record-file" do not check for changes
anymore before applying them, so writing the current value to them
suddenly does something, while it was ignored before.
DVB changes untested, but should work.
The user can raise the number of tolerated hardware decoding errors. On
the other hand, we have a static limit on packets that are "saved" for
fallback handling (and that's a good idea to avoid unbounded memory
usage). In this case, it could happen that the start of a file was fine
after a fallback, but after that buffered amount of data, it would
suddenly skip.
It's more useful to skip buffering entirely if the number of tolerated
decoding errors exceeds the fixed buffer.
(And also, I'm sure nobody gives a shit about this feature.)
prepare_decoding() returned a bool that was supposed to tell whether
decoding could work, or if something was fucked. After recent changes to
the decoder loop, this did not work anymore, and caused an endless loop.
Redo it, so it makes more sense. avctx being NULL (software fallback
initialization failed) now signals EOF. hwdec_failed needs to be handled
on send_packet() only, where it probably never happens anyway.
(Who was the idiot who made libavcodec have two entrypoints for
decoding? Oh right, it was me. PEBKAC.)
Shovel the code around to make the data flow slightly simpler (?). At
least there's only one send_packet function now. The old code had the
problem that send_packet() could be called even if there were queued
packets; due to sending the queued packets in the receive_frame
function, this should not happen anymore (the code checking for this
case in send_packet should normally never be called).
Untested with actual full stream hw decoders (none available here); I
created a test case by making hwaccel decoding fail.
Forgotten in commit 5d5fdb7. This failed to return the error code
properly. In particular, if the decoder rejected the packet, this was
not properly detected. Normally, this mattered only in specific cases.
Fixes: #7115
Commit 5d5fdb77e9 changed details of the decoding control flow, and
called it a "high-risk" change. It turns out that this broke with with
hwdec copy mode, where there is some sort of delay queue (supposedly
increases efficiency, but more likely worthless cargo-cult).
It simply used the wrong (basically inverted) condition for the draining
case.
This was the only case that did not work properly. Other tests,
including video/audio decoding errors, software decoding fallbacks,
etc., seemed to work well. Might still not be exhaustive, as there are
so many corner cases.
Also change two error code returns. This don't/shouldn't really matter,
though the second error code led it to return both a frame and
AVERROR_EOF, which is unexpected, and makes lavc_process() leak a frame.
But also see next commit.
Fixes: 5d5fdb77e9
ad_lavc and vd_lavc use the lavc_process() helper to translate the
FFmpeg push/pull API to the internal filter API (which completely
mismatch, even though I'm responsible for both, just fucking kill me).
This interface was "slightly" too tight. It returned only a bool
indicating "progress", which was not enough to handle some cases (see
following commit).
While we're at it, move all state into a struct. This is only a single
bool, but we get the chance to add more if needed.
This fixes mpv falling asleep if decoding returns an error during
draining. If decoding fails when we already sent EOF, the state machine
stopped making progress. This left mpv just sitting around and doing
nothing.
A test case can be created with: echo $RANDOM >> image.png
This makes libavformat read a proper packet plus a packet of garbage.
libavcodec will decode a frame, and then return an error code. The
lavc_process() wrapper could not deal with this, because there was no
way to differentiate between "retry" and "send new packet". Normally, it
would send a new packet, so decoding would make progress anyway. If
there was "progress", we couldn't just retry, because it'd retry
forever.
This is made worse by the fact that it tries to decode at least two
frames before starting display, meaning it will "sit around and do
nothing" before the picture is displayed.
Change it so that on error return, "receiving" a frame is retried. This
will make it return the EOF, so everything works properly.
This is a high-risk change, because all these funny bullshit exceptions
for hardware decoding are in the way, and I didn't retest them. For
example, if hardware decoding is enabled, it keeps a list of packets,
that are fed into the decoder again if hardware decoding fails, and a
software fallback is performed. Another case of horrifying accidental
complexity.
Fixes: #6618
Generally, using x86 SIMD efficiently (or crash-free) requires aligning
all data on boundaries of 16, 32, or 64 (depending on instruction set
used). 64 bytes is needed or AVX-512, 32 for old AVX, 16 for SSE. Both
FFmpeg and zimg usually require aligned data for this reason.
FFmpeg is very unclear about alignment. Yes, it requires you to align
data pointers and strides. No, it doesn't tell you how much, except
sometimes (libavcodec has a legacy-looking avcodec_align_dimensions2()
API function, that requires a heavy-weight AVCodecContext as argument).
Sometimes, FFmpeg will take a shit on YOUR and ITS OWN alignment. For
example, vf_crop will randomly reduce alignment of data pointers,
depending on the crop parameters. On the other hand, some libavfilter
filters or libavcodec encoders may randomly crash if they get the wrong
alignment. I have no idea how this thing works at all.
FFmpeg usually doesn't seem to signal alignment internal anywhere, and
usually leaves it to av_malloc() etc. to allocate with proper alignment.
libavutil/mem.c currently has a ALIGN define, which is set to 64 if
FFmpeg is built with AVX-512 support, or as low as 16 if built without
any AVX support. The really funny thing is that a normal FFmpeg build
will e.g. align tiny string allocations to 64 bytes, even if the machine
does not support AVX at all.
For zimg use (in a later commit), we also want guaranteed alignment.
Modern x86 should actually not be much slower at unaligned accesses, but
that doesn't help. zimg's dumb intrinsic code apparently randomly
chooses between aligned or unaligned accesses (depending on compiler, I
guess), and on some CPUs these can even cause crashes. So just treat the
requirement to align as a fact of life.
All this means that we should probably make sure our own allocations are
64 bit aligned. This still doesn't guarantee alignment in all cases, but
it's slightly better than before.
This also makes me wonder whether we should always override libavcodec's
buffer pool, just so we have a guaranteed alignment. Currently, we only
do that if --vd-lavc-dr is used (and if that actually works). On the
other hand, it always uses DR on my machine, so who cares.
--hwdec=auto-copy was preferring vdpau over vaapi. In the HEVC 10 bit
case, this also led to hardware decoding not being enabled. (Probably
because the probing can't start over after enabling hw decoding fails at
runtime, or something like that.)
Possible that this subtly breaks on some setups. You can't always win.
Manual changes done:
* Merged the interface-changes under the already master'd changes.
* Moved the hwdec-related option changes to video/decode/vd_lavc.c.
this lead to an unexpected videotoolbox-copy hwdec name due to the last
two chars being cut off. since selection is also done by that name one
had to use "videotoolbox-co" to explicitly use the copy mode of
videotoolbox.
The default get_format does exactly do this, so we don't need to
duplicate it.
The only potential problem with this is that the logic doesn't entirely
prevent that the avcodec_default_get_format hw_device_ctx path is
triggered, which would probably work, but has unknown consequences and
interactions. But the way the logic currently works it can't happen,
provided the hwaccel metadata libavcodec provides is correct.
The --hwdec* options are a good fit for the vd_lavc local option
struct. This annoyingly requires manual prefixing of most of these
options with --vd-lavc (could be avoided by using more sub-struct
craziness, but let's not).
sfan5
Christoph Heinrich
Thomas Weißschuh
Niklas Haas
Philip Langdale
Cœur
Jan Ekström
Lynne
Avi Halachmi (:avih)
rcombs
Emmanuel Gil Peyrot
wm4
Akemi