This was broken by 4e528206bc - the webp
decoder was assuming that it could set the output pixfmt of the vp8
decoder directly, but after that change it no longer could because
ff_get_format() was used instead. This adds an internal get_format()
callback to webp use of the vp8 decoder to override the pixfmt
appropriately.
The Intel proprietary VAAPI driver enforces the restriction that a
buffer must be created inside an existing context, so just ensure
this is always true.
If the stream timebase is coarser than the muxing timebase then the
monotonisation process may fail because adding one to the timestamp
need not actually produce a different timestamp after the rescale.
The driver is somewhat bitrotten (not updated for years) but is still
usable for decoding with this change. To support it, this adds a new
driver quirk to indicate no support at all for surface attributes.
Based on a patch by wm4 <nfxjfg@googlemail.com>.
The string codec name need not be as long as the value we are
comparing it to, so memcmp may make decisions derived from
uninitialised data that valgrind then complains about (though the
overall result of the function will always be the same). Use
strncmp instead, which will stop at the first zero byte and
therefore not encounter this issue.
Only do this when building for a recent VAAPI version - initial
driver implementations were confused about the interpretation of the
framerate field, but hopefully this will be consistent everywhere
once 0.40.0 is released.
Default to using VBR when a target bitrate is set, unless the max rate
is also set and matches the target. Changes to the Intel driver mean
that min_qp is also respected in this case, so set a codec default to
unset the value rather than using the current default inherited from
the MPEG-4 part 2 encoder.
This includes a backward-compatibility hack to choose CBR anyway on
old drivers which have no CBR support, so that existing programs will
continue to work their options now map to VBR.
Before this change, it was possible to overflow pic_order_cnt_lsb and
generate a stream with invalid POC numbering. This makes sure that
the field is large enough that a single IDR B* P sequence uses fewer
than half the available POC lsb values.
This change makes the configured GOP size be respected exactly -
previously the value could be exceeded slightly due to flaws in the
frame type selection logic.
This was correct for H.26[45], because libmfx uses the same values
derived from profile_idc and the constraint_set flags, but it is
wrong for other codecs.
Also avoid passing FF_LEVEL_UNKNOWN (-99) as the level, as this is
certainly invalid.
When decoding with threads enabled, the get_format callback will be
called with one of the per-thread codec contexts rather than with the
outer context. If a hwaccel is in use too, this will add a reference
to the hardware frames context on that codec context, which will then
propagate to all of the other per-thread contexts for decoding. Once
the decoder finishes, however, the per-thread contexts are not freed
normally, so these references leak.
Takes a frame associated with a hardware context as input and maps it
to something else (another hardware frame or normal memory) for other
processing. If the frame to map was originally in the target format
(but mapped to something else), the original frame is output.
Also supports mapping backwards, where only the output has a hardware
context. The link immediately before will be supplied with mapped
hardware frames which it can write directly into, and this filter
then unmaps them back to the actual hardware frames.
Adds the new av_hwframe_map() function, which allows mapping between
hardware frames and normal memory, along with internal support for
implementing it.
Also adds av_hwframe_ctx_create_derived(), for creating a hardware
frames context associated with one device using frames mapped from
another by some hardware-specific means.
Also adds a new flag to mark filters which are aware of hwframes and
will perform this task themselves, and marks all appropriate filters
with this flag.
This is required to allow software-mapped hardware frames to work,
because we need to have the frames context available for any later
mapping operation in the filter graph.
The output from the filter graph should only propagate further to an
encoder if the hardware format actually matches the visible format
(mapped frames are valid here and have an hw_frames_ctx, but this
should not be given to the encoder as its hardware context).
The use of TLSv1_*_method() disallows newer protocol versions; instead
use SSLv23_*_method() and then explicitly disable the deprecated
protocol versions which should not be supported.
The buffer map/unmap code was in an early version of this before it
was committed, but the unmap was never removed. While wrong, this
was harmless (and therefore unnoticed) because the buffers can't be
mapped at this point - all drivers just did nothing with the call.
When decoding interlaced pictures, the structure is reused to render
to the same surface twice. The parameter buffers were not being
cleared, which caused the i965 driver to error out.
This was not observed earlier because the only syntax element which
it normally misses with the current setup is slice_qp_delta, but that
is always going to be zero (in IDR frames QP isn't varied on the
slice) which will always exp-golomb code as a single 1 bit. The
immediately following part is the byte alignment, which is always a 1
bit followed by 0s which are ignored, so as long as the bitstream is
never aligned at that point we will never notice because the only
difference is that an ignored bit is a 1 instead of a 0.
While outwardly bizarre, this change makes the behaviour consistent
with other VAAPI encoders which sync to the encode /input/ picture in
order to wait for /output/ from the encoder. It is not harmful on
i965 (because synchronisation already happens in vaRenderPicture(),
so it has no effect there), and it allows the encoder to work on
mesa/gallium which assumes this behaviour.
This allows better checking of capabilities and will make it easier
to add more functionality later.
It also commonises some duplicated code around rate control setup
and adds more comments explaining the internals.
If the input has been decoded from a stream which uses edge cropping
then the whole surface need not be valid. This defines an input
region for the scaler so we only use the active area of the frame.
With some old libva versions <va/va.h> does not automatically include
the per-codec subsidiary headers, so we need to include the right one
explicitly ourselves.
If no string argument is supplied when av_hwdevice_ctx_create() is
called to create a VAAPI device, we currently only try the default
X11 display (that is, $DISPLAY) to find a device, and will therefore
fail in the absence of an X server to connect to. Change the logic
to also look for a device via the first DRM render node (that is,
"/dev/dri/renderD128"), which is probably the right thing to use in
most simple configurations which only have one DRM device.
The driver being used is detected inside av_hwdevice_ctx_init() and
the quirks field then set from a table of known device. If this
behaviour is unwanted, the user can also set the quirks field
manually.
Also adds the Intel i965 driver quirk (it does not destroy parameter
buffers used in a call to vaRenderPicture()) and detects that driver
to set it.
The hw frame used as reference has an attached size but it need not
match the actual size of the surface, so enforcing that the sw frame
used in copying matches its size exactly is not useful.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
Previously we would allocate a new one for every frame. This instead
maintains an AVBufferPool of them to use as-needed.
Also makes the maximum size of an output buffer adapt to the frame
size - the fixed upper bound was a bit too easy to hit when encoding
large pictures at high quality.
No longer make a dummy device configuration to query. Instead, just
return everything we recognise from the whole format list. Also
change the device setup code to query that list only, rather than
intersecting it with the constraint output.
This makes hwupload more usable on mesa/gallium where the video
processor only declares support for RGB formats, making it unable to
deal with YUV formats before this patch. It might introduce some
different trickier failures in the internal upload/download code
because the set of allowed formats there has changed, though I didn't
find any obvious regressions with i965.
Experimental; requires Skylake and VAAPI 0.39.1 (not yet released).
Also increases the allowed range of the quality option - in low-power
mode, the Intel driver supports levels 1-8 (and 0 meaning default).
Non-reference frames (nal_ref_idc == 0) should be discardable, so
frame_num does not advance after them. Before this change, a stream
containing unreferenced B-frames would be rejected by the reference
decoder.
This prevents attempts to use unsupported modes, such as low-power
H.264 mode on non-Skylake targets. Also fixes a crash on invalid
configuration, when trying to destroy an invalid VA config/context.
Commit ca2f19b9cc modified the meaning of
H264SliceContext.gb: it is now initialised at the start of the NAL unit
header, rather than at the start of the slice header. The VAAPI slice
decoder uses the offset after parsing to determine the offset of the
slice data in the bitstream, so with the changed meaning we no longer
need to add the extra byte to account for the NAL unit header because
it is now included directly.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Mark Thompson
Mark ThompsonandMark Thompson
