Adds some duplicated code, but avoids duplicate edge checks and similar.
~0.5% faster overall on Parkjoy test sample.
(cherry picked from commit 64233e702a)

From ~550 cycles (C version) to 170 (SSE/x86-64), 206 (MMX/x86-32)
and 196 (SSE2/x86-32) cycles.
(cherry picked from commit 81f2a3f4ff)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit d19b744a36)

DSPContext.vector_fmul_window()
DCADSPContext.lfe_fir()
SynthFilterContext.synth_filter_float()

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 80ba1ddb58)

Gcc 4.6 only preserves the first value when using an array with an "m"
constraint.

Signed-off-by: Mans Rullgard <mans@mansr.com>

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

Signed-off-by: Mans Rullgard <mans@mansr.com>

Adds some duplicated code, but avoids duplicate edge checks and similar.
~0.5% faster overall on Parkjoy test sample.

From ~550 cycles (C version) to 170 (SSE/x86-64), 206 (MMX/x86-32)
and 196 (SSE2/x86-32) cycles.

Signed-off-by: Mans Rullgard <mans@mansr.com>

DSPContext.vector_fmul_window()
DCADSPContext.lfe_fir()
SynthFilterContext.synth_filter_float()

Signed-off-by: Mans Rullgard <mans@mansr.com>

Merge an if and a switch.
(cherry picked from commit 73be29b0c4)

Merge an if and a switch.

dprintf clashes with POSIX.1-2008
(cherry picked from commit dfd2a005eb)

Gives decoded output identical in length to faac commandline encoder.
Fixes Issue 670.

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 243f8241db)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit e0eb963aaa)

This moves the fields needed by asm near the top, before any
structs or other members which complicate the offset calculation.
Modifying other structs will no longer require updating the offsets,
and the asm code is slightly simpler due to the smaller offsets.

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit d461a47317)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 3af1fe829e)

(cherry picked from commit 0745116c10)

(cherry picked from commit 2e27959879)

From ~780 cycles to 551 cycles, mostly just by using libc memcpy()
instead of manually shuffling individual bytes around.
(cherry picked from commit e5262ec44a)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit e5c82df80e)

Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit 3c802cabba)

Replace * (uint32_t *) buf accesses with AV_WN32A/AV_COPY32.
(cherry picked from commit 9d4bdcb714)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 5ce5dbc5f3)

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 ac28ce5fac)

dprintf clashes with POSIX.1-2008

Gives decoded output identical in length to faac commandline encoder.
Fixes Issue 670.

Signed-off-by: Mans Rullgard <mans@mansr.com>

Signed-off-by: Mans Rullgard <mans@mansr.com>

This moves the fields needed by asm near the top, before any
structs or other members which complicate the offset calculation.
Modifying other structs will no longer require updating the offsets,
and the asm code is slightly simpler due to the smaller offsets.

Signed-off-by: Mans Rullgard <mans@mansr.com>

Signed-off-by: Mans Rullgard <mans@mansr.com>

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

Signed-off-by: Mans Rullgard <mans@mansr.com>

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

Replace * (uint32_t *) buf accesses with AV_WN32A/AV_COPY32.

Signed-off-by: Mans Rullgard <mans@mansr.com>

(cherry picked from commit b5ec638343)

(cherry picked from commit 9d06d7bce3)

40% faster in ff_iir_filter_flt() when c->order == 2.
(cherry picked from commit 37cb3eb534)

That unbreaks compilation of vaapi and dxva2

Signed-off-by: Janne Grunau <janne-ffmpeg@jannau.net>
(cherry picked from commit d1b6f33bf2)

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
44002d8323.
(cherry picked from commit 22893e10ae)

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 d36beb3f69)

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 2d162e3825)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 4c57cde942)

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 24e3ad3031)

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit 7767d8d361)

Signed-off-by: Jason Garrett-Glaser <jason@x264.com>
(cherry picked from commit b09f548285)