This patch rewrites the innermost loop of ff_yuv2planeX_8_neon to avoid zips and
horizontal adds by using fused multiply adds. The patch also uses ld1r to load
one element and replicate it across all lanes of the vector. The patch also
improves the clipping code by removing the shift right instructions and
performing the shift with the shift-right narrow instructions.
I see 8% difference on an m6g instance with neoverse-n1 CPUs:
$ ffmpeg -nostats -f lavfi -i testsrc2=4k:d=2 -vf bench=start,scale=1024x1024,bench=stop -f null -
before: t:0.014015 avg:0.014096 max:0.015018 min:0.013971
after: t:0.012985 avg:0.013013 max:0.013996 min:0.012818
Tested with `make check` on aarch64-linux.
Signed-off-by: Sebastian Pop <spop@amazon.com>
Reviewed-by: Clément Bœsch <u@pkh.me>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This patch implements ff_hscale_8_to_15_neon with NEON fused multiply accumulate
and bumps the vectorization factor from 2 to 4.
The speedup is of 25% on Graviton1 A1 instances based on A-72 cpus:
$ ffmpeg -nostats -f lavfi -i testsrc2=4k:d=2 -vf bench=start,scale=1024x1024,bench=stop -f null -
before: t:0.040303 avg:0.040287 max:0.040371 min:0.039214
after: t:0.032168 avg:0.032215 max:0.033081 min:0.032146
The speedup is of 39% on Graviton2 m6g instances based on Neoverse-N1 cpus:
$ ffmpeg -nostats -f lavfi -i testsrc2=4k:d=2 -vf bench=start,scale=1024x1024,bench=stop -f null -
before: t:0.019446 avg:0.019423 max:0.019493 min:0.019181
after: t:0.014015 avg:0.014096 max:0.015018 min:0.013971
Tested with `make check` on aarch64-linux.
Signed-off-by: Sebastian Pop <spop@amazon.com>
Reviewed-by: Jean-Baptiste Kempf <jb@videolan.org>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The argument to vec_splat_u16 must be a literal. By making the
function always inline and marking the arguments const, gcc can
turn those into literals, and avoid build errors like:
swscale_vsx.c:165:53: error: argument 1 must be a 5-bit signed literal
Fixes#7861.
Signed-off-by: Daniel Kolesa <daniel@octaforge.org>
Signed-off-by: Lauri Kasanen <cand@gmx.com>
While this technically compiles in current ffmpeg, this is only
because ffmpeg is compiled in strict ISO C mode, which disables
the builtin 'vector' keyword for AltiVec/VSX. Instead this gets
replaced with a macro inside altivec.h, which defines vector to
be actually __vector, which accepts random types.
Normally, the vector keyword should be used only with plain
scalar non-typedef types, such as unsigned int. But we have the
vec_(s|u)(8|16|32) macros, which can be used in a portable manner,
in util_altivec.h in libavutil.
This is also consistent with other AltiVec/VSX code elsewhere in
the tree.
Fixes#7861.
Signed-off-by: Daniel Kolesa <daniel@octaforge.org>
Signed-off-by: Lauri Kasanen <cand@gmx.com>
Affected the FATE-tests vsynth_lena-dv-411, vsynth1-dv-411,
vsynth2-dv-411 and hevc-paramchange-yuv420p.yuv420p10.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This affected many FATE-tests: The number of failing tests went down
from 663 to 344. (Both numbers exclude tests that failed because of
unaligned accesses in code that is inside #if HAVE_FAST_UNALIGNED.)
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
gcc 6.x and 7.x generate wrong code for little endian machines
for the vec_lvsl/vec_perm instruction combos in some cases.
The bug was fixed in version 8.x
If these instructions are replaced with vec_xl, the problem goes
away for all versions of the compilers.
Fixes ticket #7124.
In libswcale/tests/swcale.c, the function fileTest() calls sscanf in
an argument of "%12s" on character srcStr[] and dstStr[], which are
only 12 bytes. So, if the input string is 12 characters, a
terminating null byte can be written past the end of these arrays.
This bug was found by cppcheck.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The implementation is pretty straight-forward. Most of the existing
NV12 codepaths work regardless of subsampling and are re-used as is.
Where necessary I wrote the slightly different NV24 versions.
Finally, the one thing that confused me for a long time was the
asm specific x86 path that did an explicit exclusion check for NV12.
I replaced that with a semi-planar check and also updated the
equivalent PPC code, which Lauri kindly checked.
./ffmpeg -f lavfi -i yuvtestsrc=duration=1:size=1200x1440 -sws_flags fast_bilinear \
-s 1200x720 -f null -vframes 100 -pix_fmt $i -nostats \
-cpuflags 0 -v error -
32-bit mul, power8 only.
~2x speedup:
rgb24
24431 UNITS in yuv2packed2, 16384 runs, 0 skips
13783 UNITS in yuv2packed2, 16383 runs, 1 skips
bgr24
24396 UNITS in yuv2packed2, 16384 runs, 0 skips
14059 UNITS in yuv2packed2, 16384 runs, 0 skips
rgba
26815 UNITS in yuv2packed2, 16383 runs, 1 skips
12797 UNITS in yuv2packed2, 16383 runs, 1 skips
bgra
27060 UNITS in yuv2packed2, 16384 runs, 0 skips
13138 UNITS in yuv2packed2, 16384 runs, 0 skips
argb
26998 UNITS in yuv2packed2, 16384 runs, 0 skips
12728 UNITS in yuv2packed2, 16381 runs, 3 skips
bgra
26651 UNITS in yuv2packed2, 16384 runs, 0 skips
13124 UNITS in yuv2packed2, 16384 runs, 0 skips
This is a low speedup, but the x86 mmx version also gets only ~2x. The mmx version
is also heavily inaccurate, while the vsx version has high accuracy.
./ffmpeg -f lavfi -i yuvtestsrc=duration=1:size=1200x1440 -sws_flags fast_bilinear \
-s 1200x1440 -f null -vframes 100 -pix_fmt $i -nostats \
-cpuflags 0 -v error -
32-bit mul, power8 only.
1.8-2.3x speedup:
rgb24
18192 UNITS in yuv2packed1, 32767 runs, 1 skips
9983 UNITS in yuv2packed1, 32760 runs, 8 skips
bgr24
18665 UNITS in yuv2packed1, 32766 runs, 2 skips
9925 UNITS in yuv2packed1, 32763 runs, 5 skips
rgba
20239 UNITS in yuv2packed1, 32767 runs, 1 skips
8794 UNITS in yuv2packed1, 32759 runs, 9 skips
bgra
20354 UNITS in yuv2packed1, 32768 runs, 0 skips
8770 UNITS in yuv2packed1, 32761 runs, 7 skips
argb
20185 UNITS in yuv2packed1, 32768 runs, 0 skips
8761 UNITS in yuv2packed1, 32761 runs, 7 skips
bgra
20360 UNITS in yuv2packed1, 32766 runs, 2 skips
8759 UNITS in yuv2packed1, 32764 runs, 4 skips
This is a low speedup, but the x86 mmx version also gets only ~2x. The mmx version
is also heavily inaccurate, while the vsx version has high accuracy.
Signed-off-by: Dong, Jerry <jerry.dong@intel.com>
Signed-off-by: Decai Lin <decai.lin@intel.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
./ffmpeg -f lavfi -i yuvtestsrc=duration=1:size=1200x1440 \
-s 1200x1440 -f null -vframes 100 -pix_fmt $i -nostats \
-cpuflags 0 -v error -
This uses 32-bit mul, so POWER8 only.
The following output formats get about 4.5x speedup:
rgb24
39980 UNITS in yuv2packed1, 32768 runs, 0 skips
8774 UNITS in yuv2packed1, 32768 runs, 0 skips
bgr24
40069 UNITS in yuv2packed1, 32768 runs, 0 skips
8772 UNITS in yuv2packed1, 32766 runs, 2 skips
rgba
39759 UNITS in yuv2packed1, 32768 runs, 0 skips
8681 UNITS in yuv2packed1, 32767 runs, 1 skips
bgra
39729 UNITS in yuv2packed1, 32768 runs, 0 skips
8696 UNITS in yuv2packed1, 32766 runs, 2 skips
argb
39766 UNITS in yuv2packed1, 32768 runs, 0 skips
8672 UNITS in yuv2packed1, 32766 runs, 2 skips
bgra
39784 UNITS in yuv2packed1, 32768 runs, 0 skips
8659 UNITS in yuv2packed1, 32767 runs, 1 skips
./ffmpeg_g -f rawvideo -pix_fmt rgb24 -s hd1080 -i /dev/zero -pix_fmt yuv420p16be \
-s 1920x1728 -f null -vframes 100 -v error -nostats -
9-14 bit funcs get about 6x speedup, 16-bit gets about 15x.
Fate passes, each format tested with an image to video conversion.
Only POWER8 includes 32-bit vector multiplies, so POWER7 is locked out
of the 16-bit function. This includes the vec_mulo/mule functions too,
not just vmuluwm.
With TIMER_REPORT skips disabled:
yuv420p9le
12412 UNITS in planarX, 131072 runs, 0 skips
73136 UNITS in planarX, 131072 runs, 0 skips
yuv420p9be
12481 UNITS in planarX, 131072 runs, 0 skips
73410 UNITS in planarX, 131072 runs, 0 skips
yuv420p10le
12322 UNITS in planarX, 131072 runs, 0 skips
72546 UNITS in planarX, 131072 runs, 0 skips
yuv420p10be
12291 UNITS in planarX, 131072 runs, 0 skips
72935 UNITS in planarX, 131072 runs, 0 skips
yuv420p12le
12316 UNITS in planarX, 131072 runs, 0 skips
72708 UNITS in planarX, 131072 runs, 0 skips
yuv420p12be
12319 UNITS in planarX, 131072 runs, 0 skips
72577 UNITS in planarX, 131072 runs, 0 skips
yuv420p14le
12259 UNITS in planarX, 131072 runs, 0 skips
72516 UNITS in planarX, 131072 runs, 0 skips
yuv420p14be
12440 UNITS in planarX, 131072 runs, 0 skips
72962 UNITS in planarX, 131072 runs, 0 skips
yuv420p16le
10548 UNITS in planarX, 131072 runs, 0 skips
73429 UNITS in planarX, 131072 runs, 0 skips
yuv420p16be
10634 UNITS in planarX, 131072 runs, 0 skips
150959 UNITS in planarX, 131072 runs, 0 skips
Signed-off-by: Lauri Kasanen <cand@gmx.com>
This function wouldn't benefit from VSX instructions, so I put it
under altivec.
./ffmpeg_g -f rawvideo -pix_fmt rgb24 -s hd1080 -i /dev/zero -pix_fmt grayf32le \
-f null -vframes 100 -v error -nostats -
3743 UNITS in planar1, 65495 runs, 41 skips
-cpuflags 0
23511 UNITS in planar1, 65530 runs, 6 skips
grayf32be
4647 UNITS in planar1, 65449 runs, 87 skips
-cpuflags 0
28608 UNITS in planar1, 65530 runs, 6 skips
The native speedup is 6.28133, and the bswapping one 6.15623.
Fate passes, each format tested with an image to video conversion.
Signed-off-by: Lauri Kasanen <cand@gmx.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Passes fate on LE (with "lavc/jrevdct: Avoid an aliasing violation" applied).
Signed-off-by: Lauri Kasanen <cand@gmx.com>
Tested-by: Michael Kostylev on BE
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
./ffmpeg_g -f rawvideo -pix_fmt rgb24 -s hd1080 -i /dev/zero -pix_fmt yuv420p \
-f null -vframes 100 -v error -nostats -
1158 UNITS in planar1, 65528 runs, 8 skips
-cpuflags 0
19082 UNITS in planar1, 65533 runs, 3 skips
16.48 speedup ratio. On x86, SSE2 is ~7. Curiously, the Power C version
takes as many cycles as the x86 SSE2 version, yikes it's fast.
Note that this function uses VSX instructions, but is not marked so.
This is because several existing functions also make that mistake.
I'll submit a patch moving them once this is reviewed.
Signed-off-by: Lauri Kasanen <cand@gmx.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Currently float are converted to 16b uint in input part
using src depth (32 bits) in hScale16To19 and hScale16to15,
make an invalid shift for the data
So shift the value when using float input
like 16 bpc uint.
Fixes the following warnings:
In file included from libswscale/rgb2rgb.c:128:0:
libswscale/rgb2rgb_template.c:346:13: warning: 'shuffle_bytes_3210_c' defined but not used
libswscale/rgb2rgb_template.c:346:13: warning: 'shuffle_bytes_3012_c' defined but not used
libswscale/rgb2rgb_template.c:346:13: warning: 'shuffle_bytes_1230_c' defined but not used
PSEUDOPAL pixel formats are not paletted, but carried a palette with the
intention of allowing code to treat unpaletted formats as paletted. The
palette simply mapped the byte values to the resulting RGB values,
making it some sort of LUT for RGB conversion.
It was used for 1 byte formats only: RGB4_BYTE, BGR4_BYTE, RGB8, BGR8,
GRAY8. The first 4 are awfully obscure, used only by some ancient bitmap
formats. The last one, GRAY8, is more common, but its treatment is
grossly incorrect. It considers full range GRAY8 only, so GRAY8 coming
from typical Y video planes was not mapped to the correct RGB values.
This cannot be fixed, because AVFrame.color_range can be freely changed
at runtime, and there is nothing to ensure the pseudo palette is
updated.
Also, nothing actually used the PSEUDOPAL palette data, except xwdenc
(trivially changed in the previous commit). All other code had to treat
it as a special case, just to ignore or to propagate palette data.
In conclusion, this was just a very strange old mechnaism that has no
real justification to exist anymore (although it may have been nice and
useful in the past). Now it's an artifact that makes the API harder to
use: API users who allocate their own pixel data have to be aware that
they need to allocate the palette, or FFmpeg will crash on them in
_some_ situations. On top of this, there was no API to allocate the
pseuo palette outside of av_frame_get_buffer().
This patch not only deprecates AV_PIX_FMT_FLAG_PSEUDOPAL, but also makes
the pseudo palette optional. Nothing accesses it anymore, though if it's
set, it's propagated. It's still allocated and initialized for
compatibility with API users that rely on this feature. But new API
users do not need to allocate it. This was an explicit goal of this
patch.
Most changes replace AV_PIX_FMT_FLAG_PSEUDOPAL with FF_PSEUDOPAL. I
first tried #ifdefing all code, but it was a mess. The FF_PSEUDOPAL
macro reduces the mess, and still allows defining FF_API_PSEUDOPAL to 0.
Passes FATE with FF_API_PSEUDOPAL enabled and disabled. In addition,
FATE passes with FF_API_PSEUDOPAL set to 1, but with allocation
functions manually changed to not allocating a palette.
Vanilla clang supports altmacro since clang 5.0, and thus doesn't
require gas-preprocessor for building the arm assembly any longer.
However, the built-in assembler doesn't support .dn directives.
This readds checks that were removed in d7320ca3ed, when
the last usage of .dn directives within libav were removed.
Alternatively, the assembly could be rewritten to not use the
.dn directive, making it available to clang users.
Signed-off-by: Martin Storsjö <martin@martin.st>
To make the best use of existing code, I generalised the wrapper
that currently does yuv420p10 to p010 to support any mixture of
input and output sizes between 10 and 16 bits. This had the side
effect of yielding a working code path for all yuv420p1x formats
to p01x.
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>
* commit '29ccc641b17afad058a5c24071ea827865a8b3a9':
build: Drop check for sys/mman.h in favor of mmap() check
Merged-by: James Almer <jamrial@gmail.com>
This patch uses dithering in DITHER_COPY macro only if
it was not used option '-sws_dither none'.
With option '-sws_dither none' it uses downshift.
For human eye dithering is OK, for video codecs not necessarily.
If user don't want to use dithering, we should respect that.
Signed-off-by: Mateusz Brzostek <mateuszb@poczta.onet.pl>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
* commit '07a2b155949eb267cdfc7805f42c7b3375f9c7c5':
Bump major versions of all libraries
A few API deprecated ~2 years ago or more are also postponed here for
varying reasons.
FF_API_LOWRES:
Since this functionality depends on AVStream->codec, i figure the two can
be removed at the same time in the next bump or so.
FF_API_AVCTX_TIMEBASE:
Couldn't get this one to work. Not just libavcodec but apparently also
libavformat and ffmpeg.c expect AVCodecContext->time_base to be set for
decoding. Upon removal some tests report a different generic stream time
base (like 1/25), and others lose packet duration values. I guess it's
somehow tied to the AVStream->codec clusterfuck.
It can be dealt with alongside FF_API_LAVF_AVCTX in the next bump.
FF_API_OLD_FILTER_OPTS_ERROR:
This one is meant to remain after FF_API_OLD_FILTER_OPTS is removed.
Its purpose is displaying the corrected command line using the new syntax
as a suggestion as part of the error message.
Merged-by: James Almer <jamrial@gmail.com>
Fixes filter-pixfmts-scale test failing on big-endian systems due to
alpSrc not being cast to (const int32_t**).
Also fixes distortions in the output alpha channel values by copying the
alpha channel code from the rgba64 case found elsewhere in output.c.
Fixes ticket 6555.
Signed-off-by: James Cowgill <James.Cowgill@imgtec.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
glibc introduced _DEFAULT_SOURCE in version 2.19 to replace _BSD_SOURCE and
_SVID_SOURCE, which were deprecated in version 2.20. Add _DEFAULT_SOURCE
where the latter two are used to be forwards-compatible and avoid warnings
about the use of deprecated definitions.
* commit '92db5083077a8b0f8e1050507671b456fd155125':
build: Generate pkg-config files from Make and not from configure
build: Store library version numbers in .version files
Includes cherry-picked commits 8a34f36593 and
ee164727dd to fix issues.
Changes were also made to retain support for raise_major and build_suffix.
Reviewed-by: ubitux
Merged-by: James Almer <jamrial@gmail.com>
* commit '11a9320de54759340531177c9f2b1e31e6112cc2':
build: Move build-system-related helper files to a separate subdirectory
"ffbuild" directory name is used instead of "avbuild".
Merged-by: Clément Bœsch <u@pkh.me>
This disables everything that was deprecated at least 18 months ago.
Readjust the minimum API version as needed, postponing any
API-incompatible changes until the next bump.
Signed-off-by: Vittorio Giovara <vittorio.giovara@gmail.com>
* commit 'aa37d2bf4505afc106e2a23c44afc722bb204a8e':
swscale: Kill non-compiling disabled cruft
The isGray() chunk is not merged as an alternative patch actually fixing
the dead code is currently under review on the mailing-list.
The SWS_X chunk is merged, with an additional cosmetic.
Merged-by: Clément Bœsch <u@pkh.me>
Also do some small cosmetic changes: Drop pointless _MMX suffix from ABSD2
macro name, drop pointless check for MMX support, we always assume MMX is
available in our SIMD code, fix spelling.
This moves work from the configure to the Make stage where it can
be parallelized and ensures that pkgconfig files are updated when
library versions change.
Bug-Id: 449
Configure checks if the ebx register can be used for asm and it has to
be saved if and only if this is not the case.
Without this the build fails when configuring with --toolchain=hardened
--disable-pic on i386 using gcc 4.8:
error: PIC register clobbered by '%ebx' in 'asm'
In that case gcc 4.8 reserves the ebx register for the GOT needed for
PIE, so it can't be used in asm directly.
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Cadhalpun <Andreas.Cadhalpun@googlemail.com>
Follow a 420, 422, 444 order instead of a random one.
This simplifies double-checking additions of new formats.
Signed-off-by: Diego Biurrun <diego@biurrun.de>
They are internal symbols that should not be exported.
based on a patch by Andreas Cadhalpun <Andreas.Cadhalpun@googlemail.com>
Signed-off-by: Diego Biurrun <diego@biurrun.de>
It is an internal swscale symbol and thus should not be exported.
Signed-off-by: Andreas Cadhalpun <Andreas.Cadhalpun@googlemail.com>
Signed-off-by: Diego Biurrun <diego@biurrun.de>
This patch addresses Trac ticket #5570. The optimized functions are in file
libswscale/ppc/input_vsx.c. Each optimized function name is a concatenation of the
corresponding name in libswscale/input.c with suffix _vsx.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Split version files into one line per symbol/directive to allow compatibility
with the Solaris linker without preprocessing and eliminate $ from version file
templates to simplify the postprocessing shell command.
