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>
Treat the 32 bit stride registers as signed.
Alternatively, we could make the stride arguments ptrdiff_t instead
of int, and changing all of the assembly to operate on these
registers with their full 64 bit width, but that's a much larger
and more intrusive change (and risks missing some operation, which
would clamp the intermediates to 32 bit still).
Fixes: https://trac.ffmpeg.org/ticket/9985
Signed-off-by: Martin Storsjö <martin@martin.st>
The intention here was probably to document this as use of
conditionals does not make sense in a comment.
Fixes doxy warning:
warning: explicit link request to 'if' could not be resolved
Bayer sources are read in groups of 2 lines (e.g. for a
BGGR flavor, the first row contains only B and G samples,
while the second row contains only G and R samples). They
need to be read as a whole.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
This is currently 64-bit only because the stack spilling code would not
assemble on RV32I (and it would corrupt s0 and s1 on RV128I, in theory).
This could be added later in the unlikely that someone wants it.
Up until now, libswscale/output.c used a macro to write
an output pixel which involved a call to av_pix_fmt_desc_get()
to find out whether the input pixel format is BE or LE
despite this being known at compile-time (there are templates
per pixfmt). Even worse, these calls are made in a loop,
so that e.g. there are eight calls to av_pix_fmt_desc_get()
for every pixel processed in yuv2rgba64_X_c_template()
for 64bit RGB formats.
This commit modifies these macros to ensure that isBE()
is evaluated at compile-time. This saved 41184B of .text
for me (GCC 11.2, -O3). Of course, it also improved performance.
E.g. ffmpeg_g -f lavfi -i testsrc2,format=yuva420p -pix_fmt rgba64le \
-threads 1 -t 1:00 -f null - (which uses yuv2rgba64le_X_c,
which is an invocation of yuv2rgba64_X_c_template() mentioned above),
performance improved from 95589 to 41387 decicycles for one call
to yuv2packedX; for the be variant the numbers went down from
76087 to 43024 decicycles.
Reviewed-by: Anton Khirnov <anton@khirnov.net>
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Up until now, libswscale/input.c used a macro to read
an input pixel which involved a call to av_pix_fmt_desc_get()
to find out whether the input pixel format is BE or LE
despite this being known at compile-time (there are templates
per pixfmt). Even worse, these calls are made in a loop,
so that e.g. there are six calls to av_pix_fmt_desc_get()
for every pair of UV pixel processed in
rgb64ToUV_half_c_template().
This commit modifies these macros to ensure that isBE()
is evaluated at compile-time. This saved 9743B of .text
for me (GCC 11.2, -O3). For a simple RGB64LE->YUV420P
transformation like
ffmpeg -f lavfi -i haldclutsrc,format=rgba64le -pix_fmt yuv420p \
-threads 1 -t 1:00 -f null -
the amount of decicycles spent in rgb64LEToUV_half_c
(which is created via the template mentioned above)
decreases from 19751 to 5341; for RGBA64BE the number
went down from 11945 to 5393. For shared builds (where
the call to av_pix_fmt_desc_get() is indirect) the old numbers
are 15230 for RGBA64BE and 27502 for RGBA64LE, whereas
the numbers with this patch are indistinguishable from
the numbers from a static build.
Also make the macros that are touched conform to the
usual convention of using uppercase names while just at it.
Reviewed-by: Anton Khirnov <anton@khirnov.net>
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
These macros are definitions, not only declarations and therefore
should not contain a semicolon. Such a semicolon is actually
spec-incompliant, but compilers happen to accept them.
Reviewed-by: Philip Langdale <philipl@overt.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
As we already have support for VUYA, I figured I should do the small
amount of work to support VUYX as well. That means a little refactoring
to share code.
Fixes FATE-failures with the the filter-2xbr filter-3xbr filter-4xbr
filter-ep2x filter-ep3x filter-hq2x filter-hq3x filter-hq4x
filter-paletteuse-bayer filter-paletteuse-bayer0
filter-paletteuse-nodither and filter-paletteuse-sierra2_4a tests
when using 32bit x86 with CPUFLAGS ranging from "mmx+mmxext" to
"mmx+mmxext+sse+sse2+sse3" (the relevant function is only overwritten
when using SSSE3).
Reviewed-by: Lynne <dev@lynne.ee>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This is by no means perfect, since at least ddagrab will return scRGB
data with values outside of 0.0f to 1.0f for HDR values.
Its primary purpose is to be able to work with the format at all.
This commit adds new code paths for vscale when filterSize is 2, 4, or
8. By using specialized code with unrolling to match the filterSize we
can improve performance.
On AWS c7g (Graviton 3, Neoverse V1) instances:
before after
yuv2yuvX_2_0_512_accurate_neon: 558.8 268.9
yuv2yuvX_4_0_512_accurate_neon: 637.5 434.9
yuv2yuvX_8_0_512_accurate_neon: 1144.8 806.2
yuv2yuvX_16_0_512_accurate_neon: 2080.5 1853.7
Signed-off-by: Jonathan Swinney <jswinney@amazon.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
Use scalar times vector multiply accumlate instructions instead of
vector times vector to remove the need for replicating load instructions
which are slightly slower.
On AWS c7g (Graviton 3, Neoverse V1) instances:
yuv2yuvX_8_0_512_accurate_neon: 1144.8 987.4
yuv2yuvX_16_0_512_accurate_neon: 2080.5 1869.4
Signed-off-by: Jonathan Swinney <jswinney@amazon.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
Change the reference to exactly match the C reference in swscale,
instead of exactly matching the x86 SIMD implementations (which
differs slightly). Test with and without SWS_ACCURATE_RND - if this
flag isn't set, the output must match the C reference exactly,
otherwise it is allowed to be off by 2.
Mark a couple x86 functions as unavailable when SWS_ACCURATE_RND
is set - apparently this discrepancy hasn't been noticed in other
exact tests before.
Add a test for yuv2plane1.
Signed-off-by: Jonathan Swinney <jswinney@amazon.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
This specialization handles the case where filtersize is 4 mod 8, e.g.
12, 20, etc. Aarch64 was previously using the c function for this case.
This implementation speeds up that case significantly.
hscale_8_to_15__fs_12_dstW_512_c: 6234.1
hscale_8_to_15__fs_12_dstW_512_neon: 1505.6
Signed-off-by: Jonathan Swinney <jswinney@amazon.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
x64 always has MMX, MMXEXT, SSE and SSE2 and this means
that some functions for MMX, MMXEXT, SSE and 3dnow are always
overridden by other functions (unless one e.g. explicitly
disables SSE2). So given that the only systems that
benefit from these functions are truely ancient 32bit x86s
they are removed.
Moreover, some of the removed code was buggy/not bitexact
and lead to failures involving the f32le and f32be versions of
gray, gbrp and gbrap on x86-32 when SSE2 was not disabled.
See e.g.
https://fate.ffmpeg.org/report.cgi?time=20220609221253&slot=x86_32-debian-kfreebsd-gcc-4.4-cpuflags-mmx
Notice that yuv2yuvX_mmx is not removed, because it is used
by SSE3 and AVX2 as fallback in case of unaligned data and
also for tail processing. I don't know why yuv2yuvX_mmxext
isn't being used for this; an earlier version [1] of
554c2bc708 used it, but
the version that was eventually applied does not.
[1]: https://ffmpeg.org/pipermail/ffmpeg-devel/2020-November/272124.html
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
x64 always has MMX, MMXEXT, SSE and SSE2 and this means
that some functions for MMX, MMXEXT and 3dnow are always
overridden by other functions (unless one e.g. explicitly
disables SSE2) for x64. So given that the only systems that
benefit from these functions are truely ancient 32bit x86s
they are removed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
x64 always has MMX, MMXEXT, SSE and SSE2 and this means
that some functions for MMX, MMXEXT and 3dnow are always
overridden by other functions (unless one e.g. explicitly
disables SSE2) for x64. So given that the only systems that
benefit from these functions are truely ancient 32bit x86s
they are removed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This is more spec-compliant because it does not rely
on dead-code elimination by the compiler. Especially
MSVC has problems with this, as can be seen in
https://ffmpeg.org/pipermail/ffmpeg-devel/2022-May/296373.html
or
https://ffmpeg.org/pipermail/ffmpeg-devel/2022-May/297022.html
This commit does not eliminate every instance where we rely
on dead code elimination: It only tackles branching to
the initialization of arch-specific dsp code, not e.g. all
uses of CONFIG_ and HAVE_ checks. But maybe it is already
enough to compile FFmpeg with MSVC with whole-programm-optimizations
enabled (if one does not disable too many components).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Y, U, V data is loaded at the end of the current iteration for the next
iteration.
It results in memory access past the frame data on the last iteration
(that data is never used after the loading).
So load data at the start of the iteration, so that only useful data is
loaded.
Signed-off-by: Vardan Margaryan <v.t.margaryan@gmail.com>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
