The commit b7c16a3f ("x86: fft: Port to
cpuflags") breaks the opus decoder in ffmpeg when compiling for 3dnow. The
output is audible, but there's a lot of noise.

The reason for the breakage is that the commit unintentionally changed the
INTERL macro so that it is empty when compiling for 3dnow. This patch
fixes it.
Signed-off-by: Mikulas Patocka <mikulas@twibright.com>
Signed-off-by: James Almer <jamrial@gmail.com>

speed seems to be similar, but simplify code

Remove the broadcast instructions as well now that they are wide
enough.
Signed-off-by: James Almer <jamrial@gmail.com>

Signed-off-by: James Almer <jamrial@gmail.com>

better func separator
and add comment for the restore rgb planes10 declaration

jpeg2000_ict_float_c: 2296.0
jpeg2000_ict_float_sse: 628.0
jpeg2000_ict_float_avx: 317.0
jpeg2000_ict_float_fma3: 262.0
Signed-off-by: James Almer <jamrial@gmail.com>

Fixes: out of array read
Fixes: 3516/attachment-311488.dat

Found-by: Insu Yun, Georgia Tech.
Tested-by: wuninsu@gmail.com
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Variables used in inline assembly need to be marked with attribute((used)).
Static constants already were, via the define of DECLARE_ASM_CONST.
But DECLARE_ALIGNED does not add this attribute, and some of the variables
defined with it are const only used in inline assembly, and therefore
appeared dead. This change adds a macro DECLARE_ASM_ALIGNED that marks
variables as used.

This change makes FFMPEG work with Clang's ThinLTO.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

libavcodec/lossless_video_dsp : cosmetic add better separator for each function, in order to make reading of the asm file easier

Fixes build with old nasm/yasm.
Signed-off-by: James Almer <jamrial@gmail.com>

Fixes assembling with old yasm.

Add () to regsize define
Suggested-by: Henrik Gramner <henrik@gramner.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Fixes out of array access
Fixes: crash-huf.avi

Regression since: 6b41b441

This could also be fixed by adding checks in the C code that calls the dsp
Found-by: Zhibin Hu and 连一汉 <lianyihan@360.cn>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Also modify the required alignment, to 32 instead of 16
for several codecs
Signed-off-by: James Almer <jamrial@gmail.com>

Signed-off-by: James Almer <jamrial@gmail.com>

Tested with "checkasm --test=exrdsp -bench"

Before:
reorder_pixels_c: 5187.8
reorder_pixels_sse2: 377.0
reorder_pixels_avx2: 331.3

After:
reorder_pixels_c: 5181.5
reorder_pixels_sse2: 377.0
reorder_pixels_avx2: 313.8
Signed-off-by: James Almer <jamrial@gmail.com>

Make dst be the first parameter and src const. It's more in line with the rest of the codebase.
Signed-off-by: James Almer <jamrial@gmail.com>

Signed-off-by: James Almer <jamrial@gmail.com>

Adds a diff_pixels_unaligned()

Fixes: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=872503Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

opus_pvq_search: Restore the proper use of conditional define and simplify the function name suffix handling.

Using named define properly documents the code paths.
It also avoids passing additional numbered arguments through
multiple levels of macro templates.

The suffix handling is done by concatenation, like in
other asm functions and avoid having two separate
"cglobal" defines.
Signed-off-by: Ivan Kalvachev <ikalvachev@gmail.com>

This splits the asm function into exact and non-exact version. The exact
version is as fast or faster on newer CPUs (which EXTERNAL_AVX_FAST describes
well) whilst the non-exact version is faster than the exact on older CPUs.

Also fixes yasm compilation which doesn't accept !cpuflags(avx) syntax.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

Makes the search produce idential results with the C version.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

There's no point in toggling it, even for debugging. Its just worse.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

Explanation on the workings and methods used by the
Pyramid Vector Quantization Search function
could be found in the following Work-In-Progress mail threads:
http://ffmpeg.org/pipermail/ffmpeg-devel/2017-June/212146.html
http://ffmpeg.org/pipermail/ffmpeg-devel/2017-June/212816.html
http://ffmpeg.org/pipermail/ffmpeg-devel/2017-July/213030.html
http://ffmpeg.org/pipermail/ffmpeg-devel/2017-July/213436.htmlSigned-off-by: Ivan Kalvachev <ikalvachev@gmail.com>

2.5ms frames:
Before   (c):  2638 decicycles in postrotate, 2097040 runs,    112 skips
After (sse3):  1467 decicycles in postrotate, 2097083 runs,     69 skips
After (avx2):  1244 decicycles in postrotate, 2097085 runs,     67 skips

5ms frames:
Before   (c):  4987 decicycles in postrotate, 1048371 runs,    205 skips
After (sse3):  2644 decicycles in postrotate, 1048509 runs,     67 skips
After (avx2):  2031 decicycles in postrotate, 1048523 runs,     53 skips

10ms frames:
Before   (c):  9153 decicycles in postrotate,  523575 runs,    713 skips
After (sse3):  5110 decicycles in postrotate,  523726 runs,    562 skips
After (avx2):  3738 decicycles in postrotate,  524223 runs,     65 skips

20ms frames:
Before   (c): 17857 decicycles in postrotate,  261866 runs,    278 skips
After (sse3): 10041 decicycles in postrotate,  261746 runs,    398 skips
After (avx2):  7050 decicycles in postrotate,  262116 runs,     28 skips

Improves total decoding performance for real world content by 9% with avx2.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

This file already has #include "idctdsp.h", which is resolved to the
idctdsp.h header in the directory where this file resides by compilers.
Two other files in this directory, libavcodec/x86/idctdsp_init.c and
libavcodec/x86/xvididct_init.c, also rely on #include "idctdsp.h"
working this way.
Signed-off-by: Wan-Teh Chang <wtc@google.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

This reverts commit 24bb7db4.

noise has to after all be sign extended, not zero extended, on tests
other than checkasm.
Fixes most aac tests broken by the now reverted commit.

noise needs to be zero extended and it can be done implicitly as a side effect
in a subsequent instruction.
Signed-off-by: James Almer <jamrial@gmail.com>

Tested-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>