The spec states that the last byte of a NAL 'shall not' be 0x00
and it is allowed for byte-stream format to add padding 0x00 for
alignment.
So our parser should strip any trailling 0x00.
https://bugzilla.gnome.org/show_bug.cgi?id=721384
The parser assumes that every time there is a 0 before the startcode,
it is part of the startcode. But that's not true.
From the specification
Byte stream NAL unit syntax
zero_byte is a single byte equal to 0x00.
When any of the following conditions are fulfilled, the zero_byte syntax
element shall be present.
– the nal_unit_type within the nal_unit( ) is equal to 7 (sequence parameter
set) or 8 (picture parameter set)
– the byte stream NAL unit syntax structure contains the first NAL unit of an
access unit in decoding order, as specified by subclause 7.4.1.2.3.
The problem with doing this for all startcodes is that a trailing zero can mess
up timestamps. The trailing zero gets prepended to the startcode, which will
carry the PTS and DTS of previous buffer.
https://bugzilla.gnome.org/show_bug.cgi?id=664443
Fix picture level scaling lists derivation from fall-back rule set B,
as specified in 7.4.2.2. More precisely, the sequence level scaling
lists need to be used but intra and inter lists arguments were swapped.
This fixes FRExt/freh5.264 from conformance testing.
https://bugzilla.gnome.org/show_bug.cgi?id=720099
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
We only check input from the API user with g_return_*_if_fail().
Internal sanity checks should use g_assert() instead, which is
disabled by default for releases.
Fix calculation of the frame cropping rectangle, and more precisely
the actual cropped height. The frame_crop_top_offset subtraction
was not scaled up with SubHeightC.
Also clean-up variables to align more with (7-18) to (7-21).
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Assign the un-cropped width/height to sps->width/sps->height
during sps header parsing. Added new fields to SPS header structure
to provide the crop-rectangle dimensions.
https://bugzilla.gnome.org/show_bug.cgi?id=694068
Add API to parse the Slice header. This also calculates the macroblock
position as specified in 6.3.16.
https://bugzilla.gnome.org/show_bug.cgi?id=664274
Signed-off-by: Sreerenj Balachandran <sreerenj.balachandran@intel.com>
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Add new interface to MPEG-2 video parser that takes GstMpegVideoPacket
arguments instead of data, size, and offset. New functions are called
after gst_mpeg_video_packet_*() and provide the default implementation.
Older API is moved to the deprecated namespace and uses the new functions.
https://bugzilla.gnome.org/show_bug.cgi?id=692933
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Ignore the display_extension values if they are greater than the width/height
values provided by seqhdr and calculate the PAR based on the seqhdr values.T
his is what DVD players are doing.
Thanks to "David Schleef <ds@schleef.org>"
https://bugzilla.gnome.org/show_bug.cgi?id=685103
This can be used by parsers to provide pre-parsed information to
downstream elements that would require it (so they can avoid having
to parse the bitstream again).
Add utility functions to convert quantization matrices from zigzag scan
order (as encoded in the bitstream) into raster scan order. Also provide
another function to reverse the operation.
https://bugzilla.gnome.org/show_bug.cgi?id=693000
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Quantizer matrices are encoded in zigzag scan order in the bitstream,
but they are stored in raster scan order when they are parsed. However,
default matrices were also prepared in zigzag scan order, hence the
mismatch. i.e. the matrices were presented either in raster scan order
if they are explicitly present in the bitstream, or they were presented
in zigzag scan order if the default definitions were to be used instead.
One way to solve this problem is to always expose the quantization
matrices in zigzag scan order, since this is the role of the parser to
not build up stories from the source bitstream and just present what
is in there.
Utility functions will be provided to convert quantization matrices in
either scan order.
https://bugzilla.gnome.org/show_bug.cgi?id=693000
Signed-off-by: Cong Zhong <congx.zhong@intel.com>
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Fix parsing of residual bytes. This is a two-step process. First,
remaining colums of full vertical resolution (<height>) need to be
processed. Next, remaining bytes in the first row can be processed,
while taking into account the fact that we may have filled in the
first columns already. So, this is not full horizontal resolution.
The following figure helps in understanding the expected order of
operations, for a 8x5 MBs bitplane.
5 5 6 6 6 6 6 6
5 5 1 1 1 2 2 2
5 5 1 1 1 2 2 2
5 5 3 3 3 4 4 4
5 5 3 3 3 4 4 4
So, after tiles 1 to 4 are decoded, vertical tile 5 needs to be
processed (2x5 MBs) and then the horizontal tile 6 (6x1 MBs).
https://bugzilla.gnome.org/show_bug.cgi?id=692461
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Fix decoding of DIFF6 or NORM6 bitplanes with an odd number of lines
(3x2 "horizontal" tiles). In this case, we have to skip the first line
of macroblocks but <width> number of bytes was used to do so, instead
of the actual <stride> size.
This fixes decoding for the video sample attached to:
https://bugzilla.gnome.org/show_bug.cgi?id=668565https://bugzilla.gnome.org/show_bug.cgi?id=692461
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Add gst_vc1_parse_slice_header() function to parse slice headers as
described in 7.1.2. Slice layers are optional and allowed in advanced
profile mode only. Picture header, if available (PIC_HEADER_FLAG),
is parsed but not recorded because it shall be the same as that was
previously parsed with gst_vc1_parse_frame_header().
This fixes SA00049.vc1 conformance test.
https://bugzilla.gnome.org/show_bug.cgi?id=692388
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Fix decoding of DIFF2 or NORM2 bitplanes with an odd number of macroblocks.
In particular, account for the first bit that was already parsed so that to
avoid a buffer overflow after all pairs are parsed.
This fixes SA00040.vc1 conformance test.
https://bugzilla.gnome.org/show_bug.cgi?id=692312
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Rename dqsbedge to dqbedge. The intent is that we can only have a single
boundary edge selector, depending on the value of dqprofile. So, dqbedge
represents DQSBEDGE if dqprofile == GST_VC1_DQPROFILE_SINGLE_EDGE, or
DQDBEDGE if dqprofile == GST_VC1_DQPROFILE_DOUBLE_EDGE.
The former dqbedge field is marked as unused and can be removed on the
next gst-plugins-bad version that allows ABI changes.
https://bugzilla.gnome.org/show_bug.cgi?id=692272
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Fix parsing of VOPDQUANT when DQUANT == 2. In particular, DQUANTFRM is
not present in the bitstream in this case and it shall be derived to
the default value of zero (7.1.1.31.1).
https://bugzilla.gnome.org/show_bug.cgi?id=692271
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Fix calculation of ALTPQUANT when DQUANT == 1. PQDIFF alters ALTPQUANT
in any case. See 7.1.1.31.6.
https://bugzilla.gnome.org/show_bug.cgi?id=692270
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Fix parse_vopdquant() to correctly parse DQPROFILE, which is 2 bits
instead of a single bit.
https://bugzilla.gnome.org/show_bug.cgi?id=692267
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
The standard specifies that when slice_beta_offset_div2 is not present
in the slice header, then the value of slice_beta_offset_div2 shall be
inferred to be equal to 0.
https://bugzilla.gnome.org/show_bug.cgi?id=692265
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Coverity found missing break in parse_frame_header_advanced() when
determining PTYPE from FPTYPE for interlaced streams.
https://bugzilla.gnome.org/show_bug.cgi?id=688626
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
gst_h264_parse_sps() returned FALSE if it parsed invalid (negative)
size components. Now make it gracefully return GST_H264_PARSER_ERROR
instead of GST_H264_PARSER_OK (FALSE).
https://bugzilla.gnome.org/show_bug.cgi?id=684568
Change the way the pixel-aspect-ratio is computed by
interpreting the sequence header aspect ratio info
as MPEG-1 values until a sequence extension or
sequence display extension is seen, and then updating
the sequence header struct accordingly.
Fixes incorrect anamorphic display on some MPEG-2 (DVD)
sequences.
ASPECT_HORIZ_SIZE and ASPECT_VERT_SIZE are syntax elements that hold
binary encodings of sizes ranging from 1 to 256. Thus, the calculated
pixel-aspect-ratio was off by one.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
https://bugzilla.gnome.org/show_bug.cgi?id=683858
Anonymous union is an ISO C (2011) feature that is not exposed in
compilers strictly conforming to the previous standard.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
... to allow for more efficient parsing and (more) consistent parsing API
among various codec parsers.
Fixes#672701.
Conflicts:
gst/videoparsers/gstmpegvideoparse.c
This always happens with GstByteReader/Writer and friends when
not taking into account returned boolean of the _read/_write functions
(which is actually wrong).
Make use of the *_unchecked variant as much as possible, or take the
returned value into account.
Some hardware video decode acceleration API (VA-API, DXVA) require
a bit count to the first macroblock, minus the number of emulation
prevention bytes. So, instead of having the consumer of the library
scan the slice_header() again, just record that number while parsing.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
https://bugzilla.gnome.org/show_bug.cgi?id=671203
The entries were not filled in linearly and the termination was not
recorded either. Now, the actual number of modifications is recorded
similarly to dec_ref_pic_marking(). i.e. an explicit counter instead
of storing the termination value in the array.
https://bugzilla.gnome.org/show_bug.cgi?id=668192
This is useful for cases where the caller *knows* that the provided
input contains a whole NALU and can therefore avoid:
* the expensive checks for the next start code (which won't be present)
* delaying the input parsing (since we would need the next incoming NALU
in order for the parsing code to detect the next start code)
https://bugzilla.gnome.org/show_bug.cgi?id=665584
