jpeg2000: Optimize dequantization

Float:   4700 -> 2700 cycles
Integer: 4400 -> 2800 cycles

(sandybridge  i7)

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Do not assume a single tile

In preparation of supporting multiple tiles.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Speed up jpeg2000_decode_tile()

Skip processing bands with dimension set to 0.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Use EBCOT's CAUSAL and BYPASS mode in decode_cblk()

Speed it up a bit.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Simplify jpeg2000_decode_packets()

Raise PATCHWELCOME error in case of non-implemented progression order.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Signed-off-by: Nicolas Bertrand <nicoinattendu@gmail.com>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Do not crash on NULL node in tag_tree_decode

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Refactor SOT marker parsing

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Proper cleanup on failure in decode_frame()

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Use the correct sizeof in memset for T1 data

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Simplify init_tile()

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Update pixel format support

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Compute corrections of sub-band coordinates

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Remove unneeded variable initialization

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Use separate fields for int and float codepaths

Split stepsize and data into int and float variants.
Eliminates a number of casts and simplifies spotting errors.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Split int/float codepaths depending on the DWT

DWT53 is always int, DWT97 is always float.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Improve reduced resolution decoding

Correctly scale down the component coordinates and
clean up some redundant code.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Compute quantization for 'scalar derived' in the correct case.

Scalar derived case is represented by the JPEG2K_QSTY_SI define.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Calculate code-block coord in ff_jpeg2000_init_component()

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Signed-off-by: Nicolas Bertrand <nicoinattendu@gmail.com>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Fix compute precedence error in lut_gain index

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Initialize code blocks structures in precincts to 0

Prevent use of uninitialized memory / valgrind failure.

Found-by: ubitux
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Check that we have enough components for MCT

Avoid overread.

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Check that there is a SOT before SOD

Avoid overreads.

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Remove unneeded returns

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Propagate error code from get_cox()

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Check that nreslevels2decode has been initialized before use

Avoid buffer overread.

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Drop unused and writeonly fields

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Correctly calculate sgnd

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: check len before parsing header

Avoid overread.

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Check component number in get_coc() and get_qcc()

Avoid overreads.

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Nicolas Bertrand <nicoinattendu@gmail.com>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Check zero bit-plane validity

Prevent integer overflows.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Validate block lengthinc

Currently we are using an array with a static data size.

Similar to a patch with the same purpose by Michael Niedermayer.

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Validate SIZ parsing

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Validate SOT parsing

Avoid some overreads.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Use bytestream2

Prevent a number of overreads.

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Clean up return paths and error messages

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Define the maximum decomposition levels

And define the resolution levels according.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Check code-block size

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Signed-off-by: Nicolas Bertrand <nicoinattendu@gmail.com>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

jpeg2000: Validate resolution levels

There are 32 maximum decomposition levels, thus 33 resolution levels.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

oggdec: do not fall back on binary search in the generic code.

Binary search is already attempted in the format-specific seek function,
so the fallback is only reached if binary search failed already.

lls/x86: use 3-operator vaddpd in ADDPD_MEM

Fixes build with yasm-1.1

Signed-off-by: Anton Khirnov <anton@khirnov.net>

Revert "indeo5: reject negative motion vectors"

Negative motion vectors are possible.

This reverts commit 1194a410807bac3eafbeb632578b937656d273e7.

indeo: check for reference when inheriting mvs

The same is done already for qdelta.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

indeo5: reject negative motion vectors

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

indeo: use proper error code

indeo: Properly forward the error codes

If the tile data size does not match the buffer size it did not
return an AVERROR_INVALIDDATA causing futher corruption later.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

x86: lpc: fix a segfault in av_evaluate_lls_sse2()

mjpeg: Check the unescaped size for overflows

And contextually check init_get_bits success and fix the reporting
message.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

mjpeg: Move code out of else branch

Simplify the control flow and spare some vertical space.

avconv: Don't include colorspace.h

The header is private, and avconv.c doesn't use it
for anything.

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>

wmapro: error out on impossible scale factor offsets

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

wmapro: check the min_samples_per_subframe

Must be at least WMAPRO_BLOCK_MIN_SIZE.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

pictor: use the correct logging context

Broken in 6d97484d72e33f7dde9493a9ead1a72e2f029605

lpc: use levinson for the first pass of multipass cholesky

Levinson is faster, and cholesky is only needed if we want to apply different
weights to different samples, which doesn't happen on the first pass.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

x86: lpc: simd av_evaluate_lls

1.5x-1.8x faster on sandybridge

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

x86: lpc: simd av_update_lls

4x-6x faster on sandybridge

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

lpc: use function pointers, in preparation for asm

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

lpc: remove "decay" argument

We never used the rolling-average mode, and this makes av_update_lls 15% faster.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

wmapro: check num_vec_coeffs against the actual available buffer

Prevent yet another buffer overwrite.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

wmapro: return early on unsupported condition

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

wmapro: make sure there is room to store the current packet

Prevent horrid and hard to trace struct overwrite.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

lavc: move put_bits_left in put_bits.h

aac: return meaningful errors

aac: K&R formatting cosmetics

configure: Do not omit frame pointers for msvc/icl debug builds

Because O1 or O2 are required to build libav with msvc/icl, this must be
explicitly set instead of just omitting Oy.

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>

configure: Do not explicitly set Oy for msvc/icl

It is implied by O1 or O2, both of which are required to build libav
with msvc/icl.  Silences warnings when targeting x64 with icl.

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>

fate: Add Canopus Lossless YUY2 test

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>

cllc: Use outbuf in RGB and ARGB functions

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>

cllc: Implement YUV support

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>

doc/platform: Add info about ICL

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>

lavf: don't abort if both encoder and muxer aspect ratios are not set

Signed-off-by: Anton Khirnov <anton@khirnov.net>

matroskaenc: restore compatibility with non referenced AVPacket

Signed-off-by: Anton Khirnov <anton@khirnov.net>

lavc: Add option to encode MPEG-2 AAC with libfdk-aac

Signed-off-by: Anton Khirnov <anton@khirnov.net>

configure: whitespace cosmetics

configure: Simplify an expression with enabled_all.

configure: More msvc/icl combining

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>

ogg: relax demuxer conformance checks

Some samples in the wild are missing headers that are expected by the
specification but in practice do not affect decoding.

avconv: report the error for codec open failure

External codec may have corner case reason to fail at init, better
report them instead having the user wonder.

avconv: drop additional strerror fallback

strerror_r is called by av_strerror already.

lavc: free the padded last frame during audio encoding properly

vsrc_movie: do not free avoption variables in uninit()

The generic code frees them as well. Since av_free was used to free them
instead of av_freep, this would result in a double free.

configure: Remove unneeded icl inline compatibility definition

Signed-off-by: Martin Storsjö <martin@martin.st>

lavfi: switch ff_default_get_audio_buffer() to av_frame_get_buffer()

This simplifies the code and avoids using libavcodec-specific
avcodec_fill_audio_frame().

lavfi: math typo in interlace filter

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

smacker: check frame size validity

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

smacker: pad the extradata allocation

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

smacker: check the return value of smacker_decode_tree

Also prevent a memory leak.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

smacker: fix an off by one in huff.length computation

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

utils: fix avcodec_flush_buffers pre-reference counting compatibility

The to_free AVframe must be freed just like the other ones.
Indeed, the calling application may expect all frames to be
released.

(This regression caused use-after-free in VLC with hwaccel.)

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

avf: move url utility functions in a separate file

avf: move ff_write_chained to mux.c

avf: move riff tags accessors where they belong

avf: move ff_http_match_no_proxy to network

It is only used by network protocols.

avf: split off format register and lookup function

bitstream: forward error values and drop few abort()

bitstream: K&R formatting cosmetics

h264_mp4toannexb_bsf: return a padded buffer

The code using the returned buffer might expect it to be
FF_INPUT_BUFFER_PADDING_SIZE padded as any other avpacket.

h264_mp4toannexb_bsf: factor out extradata parsing

h264_mp4toannexb_bsf: K&R formatting cosmetics

4xm: check bitstream_size boundary before using it

Prevent buffer overread.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

4xm: refactor decode_p_block

Directly return from code 1, 2 and 6 codepaths and simplify the
remaining one to have a single overflow check and a single call to
mcdc.