Negligible speed difference for avx2 on Zen 2 (Ryzen 5700X) and
Broadwell (Xeon E5-2620 v4):
1690±4.3 decicycles vs. 1693±78.4
1439±31.1 decicycles vs 1429±16.7
Moderate speedup with avx512 on Skylake-X (Xeon D-2123IT):
1.22x faster (793±0.8 vs. 649±5.5 decicycles) compared with avx2
Better speedup with avx512icl on Ice Lake (Xeon Silver 4316):
1.77x faster (784±1.8 vs. 442±11.6 decicycles) compared with avx2
Co-authors:
Henrik Gramner <henrik@gramner.com>
Kieran Kunhya <kierank@obe.tv>
In av1_spec.pdf page 38/669, there is a sentence below:
if ( frame_type == KEY_FRAME && show_frame ) {
for ( i = 0; i < NUM_REF_FRAMES; i++) {
RefValid[ i ] = 0
......
}
......
}
This shows that the condition of invalidating current
DPB frames should be the coming frame_type is KEY_FRAME plus
show_frame is equal to 1. Otherwise, some of the frames
in sequence after KEY_FRAME still refer to the reference frames
before KEY_FRAME, and if these before KEY_FRAME reference
frames were invalidated, these frames could not find their
reference frames, and it could cause image corruption.
Mesa fix is in https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19386
Reviewed-by: Fei Wang <fei.w.wang@intel.com>
Signed-off-by: Ruijing Dong <ruijing.dong@amd.com>
The order in which the channels are coded in the bitstream do not always follow
the native, bitmask-based order of channels both signaled by the WAV container
and forced by this same decoder. This is the case with layouts containing an
LFE channel, as it's always coded last.
Fixes ticket #9964.
Signed-off-by: James Almer <jamrial@gmail.com>
Generalize the checks for channels in all positions, and properly support
the three height groups (normal, top, bottom) instead of manually setting
the relevant channels for the latter two after the normal height tags were
parsed.
Signed-off-by: James Almer <jamrial@gmail.com>
If PCE defines channels not covered by those in the standard configurations
then don't try to come up with some made up layout and just return them in the
coded order.
Fixes al08_44.mp4 from the conformance suite, now reporting and decoding all 48
channels instead of 10.
Signed-off-by: James Almer <jamrial@gmail.com>
Set the correct amount of tags in tags_per_config[].
Also, there are no channels that correspond to a side element in this
configuration, so reflect this in the list of known/supported channel layouts.
Signed-off-by: James Almer <jamrial@gmail.com>
The IMF CPL contains an optional timecode start address. This patch reads the
latter, if present, into the context's timecode metadata parameter.
This addresses https://trac.ffmpeg.org/ticket/9842.
The current adjustment of input start times just adjusts the tsoffset.
And it does so, by resetting the tsoffset to nullify the new start time.
This leads to breakage of -copyts, ignoring of input_ts_offset, breaking
of -isync as well as breaking wrap correction.
Fixed by taking cognizance of these parameters, and by correcting start times
just before sync offsets are applied.
Provide arm64 neon optimized implementations for hscale16To19 with
filter sizes 4, 8 and X4.
The tests and benchmarks run on AWS Graviton 2 instances.
The results from a checkasm tool are shown below.
hscale_16_to_19__fs_4_dstW_512_c: 6216.0
hscale_16_to_19__fs_4_dstW_512_neon: 2257.0
hscale_16_to_19__fs_8_dstW_512_c: 10417.7
hscale_16_to_19__fs_8_dstW_512_neon: 3112.5
hscale_16_to_19__fs_12_dstW_512_c: 14890.5
hscale_16_to_19__fs_12_dstW_512_neon: 3899.0
hscale_16_to_19__fs_16_dstW_512_c: 19006.5
hscale_16_to_19__fs_16_dstW_512_neon: 5341.2
hscale_16_to_19__fs_32_dstW_512_c: 36629.5
hscale_16_to_19__fs_32_dstW_512_neon: 9502.7
hscale_16_to_19__fs_40_dstW_512_c: 45477.5
hscale_16_to_19__fs_40_dstW_512_neon: 11552.0
(Note, the checkasm tests for these functions haven't been
merged since they fail on x86.)
Signed-off-by: Hubert Mazur <hum@semihalf.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
Add arm64 neon implementations for hscale 16 to 15 with filter
sizes 4, 8 and X4.
The tests and benchmarks run on AWS Graviton 2 instances.
The results from a checkasm tool are shown below.
hscale_16_to_15__fs_4_dstW_512_c: 6703.5
hscale_16_to_15__fs_4_dstW_512_neon: 2298.0
hscale_16_to_15__fs_8_dstW_512_c: 10983.0
hscale_16_to_15__fs_8_dstW_512_neon: 3216.5
hscale_16_to_15__fs_12_dstW_512_c: 15526.0
hscale_16_to_15__fs_12_dstW_512_neon: 3993.0
hscale_16_to_15__fs_16_dstW_512_c: 20183.5
hscale_16_to_15__fs_16_dstW_512_neon: 5369.7
hscale_16_to_15__fs_32_dstW_512_c: 39315.2
hscale_16_to_15__fs_32_dstW_512_neon: 9511.2
hscale_16_to_15__fs_40_dstW_512_c: 48995.7
hscale_16_to_15__fs_40_dstW_512_neon: 11570.0
(Note, the checkasm tests for these functions haven't been
merged since they fail on x86.)
Signed-off-by: Hubert Mazur <hum@semihalf.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
Add arm64 neon implementations for hscale 8 to 19 with filter
sizes 4, 4X and 8. Both implementations are based on very similar ones
dedicated to hscale 8 to 15. The major changes refer to saving
the data - instead of writing the result as int16_t it is done
with int32_t.
These functions are heavily inspired on patches provided by J. Swinney
and M. Storsjö for hscale8to15 which were slightly adapted for
hscale8to19.
The tests and benchmarks run on AWS Graviton 2 instances. The results
from a checkasm tool shown below.
hscale_8_to_19__fs_4_dstW_512_c: 5663.2
hscale_8_to_19__fs_4_dstW_512_neon: 1259.7
hscale_8_to_19__fs_8_dstW_512_c: 9306.0
hscale_8_to_19__fs_8_dstW_512_neon: 2020.2
hscale_8_to_19__fs_12_dstW_512_c: 12932.7
hscale_8_to_19__fs_12_dstW_512_neon: 2462.5
hscale_8_to_19__fs_16_dstW_512_c: 16844.2
hscale_8_to_19__fs_16_dstW_512_neon: 4671.2
hscale_8_to_19__fs_32_dstW_512_c: 32803.7
hscale_8_to_19__fs_32_dstW_512_neon: 5474.2
hscale_8_to_19__fs_40_dstW_512_c: 40948.0
hscale_8_to_19__fs_40_dstW_512_neon: 6669.7
Signed-off-by: Hubert Mazur <hum@semihalf.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
The FFmpeg encoder does not increment the temporal reference field, so use
that knowledge plus the extradata field to detect buggy versions of the encoder.
Fixes ticket #128.
The SVQ1 interframe mean VLC symbols -128 and 128 are incorrectly swapped
in our SVQ1 implementation, resulting in visible artifacts for some videos.
This patch unswaps the order of these two symbols.
The most noticable example of the artiacts caused by this error can be observed in
https://trac.ffmpeg.org/attachment/ticket/128/svq1_set.7z '352_288_k_50.mov'.
The artifacts are not observed when using the reference decoder
(QuickTime 7.7.9 x86 binary).
As a result of this patch, the reference data for the fate-svq1 test
($SAMPLES/svq1/marymary-shackles.mov) must be modified. For this file, our
decoder output is now bitwise identical to the reference decoder. I have
tested patch with various other samples and they are all now bitwise identical.
ff_mpeg1_dc_scale_table is the default value for
[yc]_dc_scale_table (as set by ff_mpv_common_defaults()).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This e.g. allows compilers to bake the offset implied
by using ff_mpeg12_dc_scale_table[3] (as the SpeedHQ encoder
does) into the general offset; for certain arches this is
also necessary in order to avoid building suboptimal code.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
These tables are only accessed in ff_set_qscale()
which only accesses values 1..31 as well as in
encode_picture() in mpegvideo_enc.c, accessing
the value with index 8. So make these tables smaller.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For encoders, mpeg_quant is an option of the MPEG-4 encoder
and therefore constant. This implies that one can set
the dct_unquantize_(intra|inter) function pointers during init.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The frei0r API requires linesize to be width * 4.
Since the align property of ff_default_get_video_buffer2
specifies line alignment, not buffer alignment, the
previous value of 16 breaks this requirement for frames
whose width is a multiple of 8, but not a multiple of 16
as it adds extra padding to ensure line aligment.
Fix Trac ticket #9873
This basically reverts cc0380222a.
At the time of said commit, cleanup on init failure was very
buggy. For codecs with the INIT_CLEANUP cap, the close function
could be called on error even before the private data has been
allocated; and when using frame threading the same could also
happen even without said flag. Some mpegvideo decoders were
affected by the latter.
Yet both of these issues have been fixed long ago, the latter
in commit e9b6617579. Therefore
the workaround in ff_mpv_common_end() can be removed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It avoids checks and allows to make ff_wmv2_decode_mb() static;
furthermore, it allows to avoid a config_components.h inclusion.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Only encoders need two sets of int16_t [12][64]
(one to save the current best state and one for the current
working state); decoders need only one. This saves 1.5KiB
per slice context for a decoder.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
James Darnley
Paul B Mahol
Ruijing Dong
James Almer
Timo Rothenpieler
Pierre-Anthony Lemieux
Gyan Doshi
Peter Ross
Hubert Mazur
James Zern
Andreas Rheinhardt
Brendan Hack
Jun Zhao