Merge remote-tracking branch 'lukaszmluki/master'

* lukaszmluki/master:
  lavd/avdevice: add device iterators
  lavd: add categories to device implementations
  lavu/log: add device categories

Merged-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/h263dec: Workaround H263 end padding bug

Fixes: H263PaddingBugFixVID_6647.MOV

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

lavd/avdevice: add device iterators

Signed-off-by: Lukasz Marek <lukasz.m.luki@gmail.com>

lavd: add categories to device implementations

Signed-off-by: Lukasz Marek <lukasz.m.luki@gmail.com>

lavu/log: add device categories

Signed-off-by: Lukasz Marek <lukasz.m.luki@gmail.com>

avformat/microdvd: export the declared frame rate

MicroDVD files are normally frame-based (i.e. there are no absolute
timestamps), but can have an optional frame rate header. If this
header is missing, the timestamps depend on the frame rate of the
video they were created for.

The demuxer will use a fallback frame rate if it's missing from the
header. Currently, applications using libavformat can't know whether
the time base is based on the fallback value, or if a frame rate
header was present.

This commit introduces a subfps AVOption for MicroDVD, and the
demuxer sets it if and only if a frame rate header was present.

Signed-off-by: Clément Bœsch <u@pkh.me>

avformat/microdvd: skip BOM properly

The BOM is already skipped in the probe function, but not the header
read function. This could cause the header to be misparsed, such as
not interpreting the FPS line.

ffmpeg: dont try to run the subtitle encoder on non subtitle streams

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/httpauth: add comment about "" and algorithm

Suggested-by: ubitux
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/HLS: modified EXT-X-MEDIA-SEQUENCE header in order to follow the HLS standard

When using the wrap option the EXT-X-MEDIA-SEQUENCE header of the playlist file was reset instead of being incremented.
It is now incremented by one for every media URI removed from the playlist file as the standard states.

Signed-off-by: Willy Aubry <waubry@viotech.net>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libutvideoenc: Add support for the new BT.709 FourCCs for YCbCr

With cli usage the decoder might have not set the colorspace during
encoder init, manual colorspace override might be needed in such
cases.

Signed-off-by: Anton Khirnov <anton@khirnov.net>
This applies commit 5de64bb3 (the source of the above commit message)
to libutvideoenc as well.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/utils: detect MPEG streams with faulty DTS and discard affected DTS

Fixes issue2.ts

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '5c9c305d14730faf2af206e5d519df4948de0ed2'

* commit '5c9c305d14730faf2af206e5d519df4948de0ed2':
  ass: K&R formatting cosmetics

Conflicts:
libavformat/assdec.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/hevcpred_template: also initialize top[-1] for constrained intra prediction

Fixes out of array read
Fixes: 387713a12dc5cfa27fcb4178084ce1ea-asan_stack-oob_131176a_1182_cov_3861068719_CAINIT_C_SHARP_3.bit
Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/g2meet: also reset local got_header when reseting the context got_header

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/g2meet: fix error returns

Fixes out of array accesses

This should not affect any release

Fixes: 8ab69af9e5a7a7e20fe04cdd25c0d6e7-asan_heap-oob_e72b82_5505_cov_2278389485_g2m4.wmv
Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

ass: K&R formatting cosmetics

Revert "mpegts: do not set pts for missing dts in video streams"

This reverts commit f65afef1df49f53e14c8d4173ff960fff8d44ecb.

If only pts is coded then dts must be equal pts
See Rec. ITU-T H.222.0 (06/2012) / ISO/IEC 13818-1:2013 (E) 2.7.5

Found-by: Baptiste Coudurier

x86/synth_filter: Revert the switch to float ops with SSE2

This reverts the changes 64672098361361cd15d37e36f747ab44de5b80ca
and 68c3ed936a76c3ff7738f602fa90237ac7e3ce08 did to the SSE2 version,
which generated a hit of about 5 cycles.

Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/oggparsevorbis: check packet size before reading new_len from it

Fixes out of array read
Fixes: 34260c7981118fb38fba61809bf4dd5a-asan_heap-oob_93b923_1508_cov_951051643_DivX640x480_oggvorbis.avi
Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/avidec: Use a buffer with sufficient padding in read_gab2_sub()

Fixes out of array read
Fixes: 0ff9841c2a102f06e0d582bfc3376cbd-asan_heap-oob_495589_6836_cov_1763916974_mewmew_ssa.avi
Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge remote-tracking branch 'qatar/master'

* qatar/master:
  ituh263: reject b-frame with pp_time = 0

See: bb01be80462adcf4900c925e25bc644e512a4b7b
Merged-by: Michael Niedermayer <michaelni@gmx.at>

x86/synth_filter: add synth_filter_avx

Sandy Bridge Win64:
180 cycles on ff_synth_filter_inner_sse2
150 cycles on ff_synth_filter_inner_avx

Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/mp3dec: optimize mp3_seek() for dir < 0

this minimizes the amount of protocol seeks and reading needed in that case

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/aviobuf: optimize sequential backward seeking

This reduces the number of protocol seeks

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/aviobuf: factorize buffer_size out

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

ituh263: reject b-frame with pp_time = 0

Avoid a division by 0 in ff_mpeg4_set_one_direct_mv.

Sample-Id: 00000168-google
Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Vittorio Giovara <vittorio.giovara@gmail.com>

RELEASE: update for 2.2

Found-by: Timothy
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

x86/synth_filter: add synth_filter_sse

Build only on x86_32 targets.

Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

HLS: Handle the case when hls_list_size=0. Playlist file should contain all the segments. Fixes bug #3376.

Signed-off-by: Senthilnathan M <senthil.codr@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge remote-tracking branch 'qatar/master'

* qatar/master:
  qt-faststart: Add a note about the -movflags +faststart feature

Conflicts:
tools/qt-faststart.c

See: 2574d62d0dd34df762aa2d80c7d6381918a80ed1
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'ea7f79f93796d68559a495be824b6bbd94dfe5f6'

* commit 'ea7f79f93796d68559a495be824b6bbd94dfe5f6':
  qt-faststart: Avoid unintentionally sign extending BE_32

See: 582f36ca3fb1c69dbe3478f174d36278f5dd3f63
See: 69ce34c796dd2d595f3312848cf6d0c87506e0d0
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'bb95334c34d0d9abccea370ae25c4765d7764ab8'

* commit 'bb95334c34d0d9abccea370ae25c4765d7764ab8':
  qt-faststart: Check offset_count before reading from the moov_atom buffer

Conflicts:
tools/qt-faststart.c

See: 0ea4742341726ebe42c301bc0d6426cfa01dd134
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '63848854256a024a19435e87d6bc76fffa65e81e'

* commit '63848854256a024a19435e87d6bc76fffa65e81e':
  qt-faststart: Check the ftello() return codes

Conflicts:
tools/qt-faststart.c

See: 4a2297294fa269e05aa2b3fe36b2ea231caa6dca
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '03c2a66fcff9707f71ffef7e61ce5e3973220d4b'

* commit '03c2a66fcff9707f71ffef7e61ce5e3973220d4b':
  qt-faststart: Fix the signedness of variables keeping the ftello return values

Conflicts:
tools/qt-faststart.c

See: 1838961357e38402be64c0c82a2f08e4e85a0c01
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '5612244351b2eb3cb4e6225861a0f55aa5d0c475'

* commit '5612244351b2eb3cb4e6225861a0f55aa5d0c475':
  qt-faststart: Check fseeko() return codes

Conflicts:
tools/qt-faststart.c

See: 0de41ead6f9034e0a834abe51028a02aee094990
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'ea15a9a5d8fa6a71af3101b2af18c4dcac07987f'

* commit 'ea15a9a5d8fa6a71af3101b2af18c4dcac07987f':
  qt-faststart: Simplify code by using a MIN() macro

Conflicts:
tools/qt-faststart.c

See: 59003fe7c064e98359cce83b1b727fb1026bdf12
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '3cbc7ef3d60b6af3617079f24a4f401d83353003'

* commit '3cbc7ef3d60b6af3617079f24a4f401d83353003':
  qt-faststart: Increase the copy buffer size to 64 KB

Conflicts:
tools/qt-faststart.c

See: f4d9148fe282879b9fcc755767c9c04de9ddbcfa
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'd1f9563d502037239185c11578cc614bdf0c5870'

* commit 'd1f9563d502037239185c11578cc614bdf0c5870':
  pthread_frame: flush all threads on flush, not just the first one

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'bc6461c2861b7d482a037d3b3e2b44ad48805fa0'

* commit 'bc6461c2861b7d482a037d3b3e2b44ad48805fa0':
  af_compand: replace strtok_r() with av_get_token()

Conflicts:
configure
libavfilter/af_compand.c

Not merged as requested by ubitux
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'fb3b2f5d923a6e19d80f21eb4e081674bceec810'

* commit 'fb3b2f5d923a6e19d80f21eb4e081674bceec810':
  configure: Set the thread type after resolving dependencies

Conflicts:
configure

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '2f02bbcca050936686482453078e83dc25493da0'

* commit '2f02bbcca050936686482453078e83dc25493da0':
  build: Let the ffvhuff decoder/encoder depend on the huffyuv decoder/encoder

Conflicts:
configure
libavcodec/Makefile

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '34150be515cd9c43b0b679806b8d01774960af78'

* commit '34150be515cd9c43b0b679806b8d01774960af78':
  build: Let the iac decoder depend on the imc decoder

Conflicts:
configure

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '8e0cf39faf02536dca08f4fe628a66d1ae022fde'

* commit '8e0cf39faf02536dca08f4fe628a66d1ae022fde':
  build: Let all MJPEG-related decoders depend on the MJPEG decoder

Conflicts:
configure
libavcodec/Makefile

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '0a36988e48dd581d29e77f768f987738bdf365f0'

* commit '0a36988e48dd581d29e77f768f987738bdf365f0':
  build: Let AMV decoder depend on the SP5X decoder

Conflicts:
configure
libavcodec/Makefile

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'f4bd9fe326ad1315a74206939ae56df93b940a09'

* commit 'f4bd9fe326ad1315a74206939ae56df93b940a09':
  h264: fix clang warning about uninitialized variable

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '17a63ff0cd187b9e50e4a47862750295976853b1'

* commit '17a63ff0cd187b9e50e4a47862750295976853b1':
  h264: update flag name in ff_h264_decode_ref_pic_list_reordering()

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'ee6280ca12696a67535ce2245c77190edc513794'

* commit 'ee6280ca12696a67535ce2245c77190edc513794':
  h264: parse two additional constraint flags

Conflicts:
libavcodec/h264_ps.c

See: 9091ba9dfa0901bdce8305cf282c747b0f4c942e
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'e70ab7c1f5005041bba0e4efc1165410f83495b2'

* commit 'e70ab7c1f5005041bba0e4efc1165410f83495b2':
  h264: add MVCD to the list of High profiles in SPS

Merged-by: Michael Niedermayer <michaelni@gmx.at>

update for 2.2

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/utvideoenc: fix slice_bits size

Fixes assertion failure

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/ac3dec: Use avpriv_ac3_parse_header2()

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/ac3_parser: add avpriv_ac3_parse_header2() and use it in libavcodec

The new function has the ability to allocate the structure, allowing it to grow
without needing major bumps

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/ac3: move new field to the end of AC3HeaderInfo

This structure is used in the interface between libs and thus cannot have
fields added in the middle without major bump

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avutil/opt: allow max=min=0 for pixel formats

This is neccessary for maintaining ABI compatibility with FFmpeg 2.1

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

qt-faststart: Add a note about the -movflags +faststart feature

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

qt-faststart: Avoid unintentionally sign extending BE_32

Without this cast, the BE_32() expression is sign extended when
assigned to an uint64_t, since the uint8_t|uint8_t expression
is promoted to an int.

Also avoid undefined behaviour when left shifting an uint8_t
by 24 by casting it to an uint32_t explicitly before shifting.

Based on a patch by Michael Niedermayer.

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

qt-faststart: Check offset_count before reading from the moov_atom buffer

CC: libav-stable@libav.org
Signed-off-by: Martin Storsjö <martin@martin.st>

qt-faststart: Check the ftello() return codes

This silences a warning in the coverity static analyzer.

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

qt-faststart: Fix the signedness of variables keeping the ftello return values

These variables are assigned the return values of ftello, which
returns an off_t, which is a signed type. On errors, ftello returns
-1, thus make sure this error return value can be stored properly.

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

qt-faststart: Check fseeko() return codes

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

qt-faststart: Simplify code by using a MIN() macro

qt-faststart doesn't use the normal libav headers at all since
it's supposed to be a completely standalone tool, so we implement
the macro locally in this file.

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

qt-faststart: Increase the copy buffer size to 64 KB

Copying data in chunks of 1 KB is a little wasteful.

64 KB should still easily fit on the stack, so there's no need
to allocate it dynamically.

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

Merge remote-tracking branch 'qatar/master'

* qatar/master:
  latm: Always reconfigure if no extradata was set previously

See: d039b6074ca68da9b6dc88d8bb40056fee9fecb6
Merged-by: Michael Niedermayer <michaelni@gmx.at>

pthread_frame: flush all threads on flush, not just the first one

avcodec_flush_buffers() must release all internally held references
according to its documentation, for which all the threads need to be
flushed.

CC:libav-stable@libav.org
Bug-Id: vlc/9665

af_compand: replace strtok_r() with av_get_token()

Merge commit '6e4009d4cdf5927bdaedf58fcfc5e813b14c366b'

* commit '6e4009d4cdf5927bdaedf58fcfc5e813b14c366b':
  arm: dcadsp: implement decode_hf as external NEON asm

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '4cb6964244fd6c099383d8b7e99731e72cc844b9'

* commit '4cb6964244fd6c099383d8b7e99731e72cc844b9':
  dcadec: simplify decoding of VQ high frequencies

Conflicts:
configure
libavcodec/dcadec.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/dcadec: use brackets to ensure that no slow division is used

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '7686afd049be98d18663682b92d983340fa2c305'

* commit '7686afd049be98d18663682b92d983340fa2c305':
  dca: factorize scaling in inverse ADPCM

Merged-by: Michael Niedermayer <michaelni@gmx.at>

configure: Set the thread type after resolving dependencies

A threading type might be detected originally, but later disabled
if one of its dependencies is unavailable.

This makes sure that the threading support item in the configure
output is right for setups where w32threads are available but
native atomics aren't.

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

build: Let the ffvhuff decoder/encoder depend on the huffyuv decoder/encoder

Almost all of the code is shared so there is little point in
enabling one decoder/encoder without the other.

build: Let the iac decoder depend on the imc decoder

Almost all of the code is shared so there is little point in
enabling one decoder without the other.

build: Let all MJPEG-related decoders depend on the MJPEG decoder

These codecs compile all of the MJPEG code anyway, so there is little
point in not enabling the MJPEG decoder directly. This also simplifies
the dependency declarations for the MJPEG codec family.

build: Let AMV decoder depend on the SP5X decoder

This codec compiles all of the SP5X code anyway, so there is little
point in not enabling the decoder directly. This also simplifies the
dependency declaration for the AMV decoder.

Merge commit '08e3ea60ff4059341b74be04a428a38f7c3630b0'

* commit '08e3ea60ff4059341b74be04a428a38f7c3630b0':
  x86: synth filter float: implement SSE2 version

Conflicts:
libavcodec/x86/dcadsp.asm
libavcodec/x86/dcadsp_init.c

See: 2cdbcc004837ce092a14f326f24d97a29512a2c3
Merged-by: Michael Niedermayer <michaelni@gmx.at>

x86: synth filter float: implement SSE2 version

Timings for Arrandale:
          C    SSE
win32:  2108   334
win64:  1152   322

Factorizing the inner loop with a call/jmp is a >15 cycles cost, even with
the jmp destination being aligned.

Unrolling for ARCH_X86_64 is a 20 cycles gain.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '57b1eb9f75b04571063ddec316e290c216c114ac'

* commit '57b1eb9f75b04571063ddec316e290c216c114ac':
  dcadsp: scan coefficients linearly in dca_lfe_fir

Conflicts:
libavcodec/dcadsp.c

See: 9ae8e23188fc2e533eea74757c9060557941d3d9
Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'ad507d7907457e678900bac132122ba7be4644cb'

* commit 'ad507d7907457e678900bac132122ba7be4644cb':
  x86: dcadsp: implement SSE lfe_dir

Conflicts:
libavcodec/x86/dcadsp.asm

See: 169243112c1e310d90c030fb258092f6d2e46117
Merged-by: Michael Niedermayer <michaelni@gmx.at>

x86: dcadsp: implement SSE lfe_dir

Results for Arrandale/Windows:
32: 1670 -> 316
64:  728 -> 298

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

h264: fix clang warning about uninitialized variable

h264: update flag name in ff_h264_decode_ref_pic_list_reordering()

This is the name used in the specifications.

h264: parse two additional constraint flags

h264: add MVCD to the list of High profiles in SPS

Also comment all previous profiles.

Merge commit '87ec849fe9acba075c843e67bcd01f256f481a18'

* commit '87ec849fe9acba075c843e67bcd01f256f481a18':
  dcadec: remove scaling in lfe_interpolation_fir

Conflicts:
libavcodec/dcadec.c
libavcodec/dcadsp.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'a55546f48d55e3d1155840541b2be5f4f8cf18ab'

* commit 'a55546f48d55e3d1155840541b2be5f4f8cf18ab':
  proresenc: Reuse proper dsputil infrastructure for FDCT

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '92e598a57a7ce4b8ac9ea56274af39f5fd888311'

* commit '92e598a57a7ce4b8ac9ea56274af39f5fd888311':
  prores: Drop DSP infrastructure for prores encoder bits

Conflicts:
libavcodec/Makefile
libavcodec/proresdsp.c
libavcodec/proresenc_kostya.c

Note, these changes only affect one of the 2 prores encoders we have
If someone wants to add optimizations to the affected encoder, or needs/wants
this infrastructure, then iam happy to revert this

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'd6acefe05862af244fd5a30ae946ed507c063994'

* commit 'd6acefe05862af244fd5a30ae946ed507c063994':
  proresenc: Drop unnecessary DCT permutation bits

Conflicts:
libavcodec/proresenc_kostya.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'b23650491fbd579a4365f42bd42575afb7b53f7e'

* commit 'b23650491fbd579a4365f42bd42575afb7b53f7e':
  prores: Use consistent names for DSP arch initialization functions

Conflicts:
libavcodec/proresdsp.c
libavcodec/proresdsp.h
libavcodec/x86/proresdsp_init.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

latm: Always reconfigure if no extradata was set previously

AAC LOAS can have new audio config objects in the stream itself.

Make sure the decoder reconfigures itself when the first one arrives
midstream.

Bug-Id: 644
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

avcodec/msvideo1: Fix palette in case of seek before decode

Fixes Ticket3212

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

arm: dcadsp: implement decode_hf as external NEON asm

dcadec: simplify decoding of VQ high frequencies

The vector dequantization has a test in a loop preventing effective SIMD
implementation. By moving it out of the loop, this loop can be DSPized.

Therefore, modify the current DSP implementation. In particular, the
DSP implementation no longer has to handle null loop sizes.

The decode_hf implementations have following timings:

For x86 Arrandale:
        C  SSE SSE2 SSE4
win32: 260 162  119  104
win64: 242 N/A   89   72

The arm NEON optimizations follow in a later patch as external asm. The
now unused check for the y modifier in arm inline asm is removed from
configure.

dca: factorize scaling in inverse ADPCM

Based on a patch from Christophe Gisquet.

Unrolling of the m == 0 case avoids a possible use of the uninitilized
value sum when s->predictor_history is not set. I failed to find a
sample for it. It also reduced the cycle count from 220 to 150 on
sandy bridge, x86_64 linux, gcc 4.8.2 compared to his patch.

x86: synth filter float: implement SSE2 version

Timings for Arrandale:
          C    SSE
win32:  2108   334
win64:  1152   322

Factorizing the inner loop with a call/jmp is a >15 cycles cost, even with
the jmp destination being aligned.

Unrolling for ARCH_X86_64 is a 20 cycles gain.

Signed-off-by: Janne Grunau <janne-libav@jannau.net>

dcadsp: scan coefficients linearly in dca_lfe_fir

This change is inspired by x86 asm where it frees a register.

Signed-off-by: Janne Grunau <janne-libav@jannau.net>

x86: dcadsp: implement SSE lfe_dir

Results for Arrandale/Windows:
32: 1670 -> 316
64:  728 -> 298

Signed-off-by: Janne Grunau <janne-libav@jannau.net>

dcadec: remove scaling in lfe_interpolation_fir

The scaling factor is constant so it is faster to scale the
FIR coefficients in the tables during compilation.

Signed-off-by: Janne Grunau <janne-libav@jannau.net>

proresenc: Reuse proper dsputil infrastructure for FDCT

prores: Drop DSP infrastructure for prores encoder bits

None of the encoder bits are arch-optimized.