(cherry picked from commit 0745116c)

(cherry picked from commit 2e279598)

From ~780 cycles to 551 cycles, mostly just by using libc memcpy()
instead of manually shuffling individual bytes around.
(cherry picked from commit e5262ec4)

$subject. Have used this for loopback testing with mpegts.c.

-- Peter
(A907 E02F A6E5 0CD2 34CD 20D2 6760 79C5 AC40 DD6B)
[2. text/x-diff; 0001-mpegtsenc-support-CODEC_ID_AAC_LATM.patch]

From 0f7f9db4b7da1793996af6dda84298507703759a Mon Sep 17 00:00:00 2001
From: Peter Ross <pross@xvid.org>
Date: Sun, 9 Jan 2011 09:45:50 +1100
Subject: [PATCH] mpegtsenc: support CODEC_ID_AAC_LATM
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 4d54df8e)

fix compile on OS/2
Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit a0788cc6)

Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit 04295054)

poll() is only used by networking code, so the fallback should
only be built if networking is enabled.  Also remove CONFIG_FFSERVER
condition from the declarations.

This should fix building on systems without poll(), broken
by a8475bbd.
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 362d8f7d)

Add an error message in case the user requests to write more than one file
and the path does not contain a "%d" or "%0Nd" pattern.
Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit 4fc9ff0a)

The output was changed by a7827a17.
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 79dca23d)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit e5c82df8)

Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit 85466e1e)

According to EN 300 468 section 3.1 (Definitions):

   Unless otherwise specified within the present document all
   "reserved_future_use" bits is set to "1".

This was not the case for SDT generation so this patch fixes it.
Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit a7827a17)

Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit 3c802cab)

In the start_frame() debug log, print the reference pos value rather than the evaluated value converted to int.

That's required because -1 is evaluated as NAN, which converted back
to int looks like a random number, this is especially annoying when
debugging sources with undefined pos (as the video4linux2 device).
Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit 3e5bc7ff)

Replace * (uint32_t *) buf accesses with AV_WN32A/AV_COPY32.
(cherry picked from commit 9d4bdcb7)

the sdp demuxer did not forward it at all while the rtsp demuxer assumed
a single kind of error
(cherry picked from commit f81c7ac7)

Select has limitations on the fd values it could accept and silently
breaks when it is reached.
(cherry picked from commit a8475bbd)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 5ce5dbc5)

This reduces the symbols exported by libavformat from 699 to 451.
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit e6288640)

This significantly reduces the size of the symbol table in the generated ELF
shared object (as well as the other linked tables).
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit ac28ce5f)

VC1 test container always uses time-base 1 ms, so we must convert
from whatever time-base the application gave us to that, otherwise
the video will play at ridiculous speeds.
It would be possible to signal that a container supports only one
time-base and have code in a layer above do the conversion, but
for a single format this seems over-engineered.

Avoids an assert when the sample rate is invalid and the timebase
is thus set to e.g. 1/0.
Sample file is http://samples.mplayerhq.hu/ogg/fuzzed-srate-crash.ogg

(cherry picked from commit b5ec6383)

(cherry picked from commit 9d06d7bc)

Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit 97b04f5e)

based on Alex Converse's "Fix ADPCM MS in mov muxing" patch
(cherry picked from commit 795ed278)

40% faster in ff_iir_filter_flt() when c->order == 2.
(cherry picked from commit 37cb3eb5)

That unbreaks compilation of vaapi and dxva2
Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit d1b6f33b)

This also lists the objects from those two libraries as internal (by adding
the ff_ prefix) so that they can then be hidden via linker scripts.
(cherry picked from commit c6610a21)

It probably also fixes a memleak or two.
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 225b6d7f)

This reverts commit c34461b3.
The wrong version of the patch was committed.
(cherry picked from commit c4f8765a)

Fix C VP8 H+V MC functions which do two-dimensional 4/6-tap filters to
not overread beyond their edges if the second filter is 4-tap, since
the outer pixels aren't there anymore since
44002d83.
(cherry picked from commit 22893e10)

None of these symbols should be accessed directly, so declare them as
hidden.
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit d36beb3f)

The iff.h header only declared one function that is now static, the
libavformat/iff.c source file wasn't using it before. Drop the file
entirely.
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 2d162e38)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 4c57cde9)

It is only used to generate band_start_tab, which about the same size, at
runtime, so it's simpler just to always hardcode band_start_tab.
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 24e3ad30)

This removes ff_get_ref_perms_string, ff_dprintf_ref and ff_dprintf_link
fro the interface of libavfilter.
Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit eb7ccf8f)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 7767d8d3)