In my personal opinion, we should not need to support unaligned YUY2
pixel maps. They should always be aligned to at least 32 bits, and the
current code assumes just 16 bits. However checkasm does test for
unaligned input bitmaps. QEMU accepts it, but real hardware dose not.
In this particular case, we can at the same time improve performance and
handle unaligned inputs, so do just that.
uyvytoyuv422_c: 104379.0
uyvytoyuv422_c: 104060.0
uyvytoyuv422_rvv_i32: 25284.0 (before)
uyvytoyuv422_rvv_i32: 19303.2 (after)
This saves three scratch registers and three instructions per line. The
performance gains are mostly negligible. The main point is to free up
registers for further rework.
The documentation states that invalid entries default to SWS_CS_DEFAULT.
A value of 0 is not a valid SWS_CS_*, yet the code incorrectly
hard-codes it to BT.709 coefficients instead of SWS_CS_DEFAULT.
This was a complete hack seemingly designed to work around a different
bug, which was fixed in the previous commit. As such, there is no more
reason not to do this, as it simply breaks changing color range in
sws_setColorspaceDetails for no reason.
More commonly, this fixes the case of sws_setColorspaceDetails after
sws_getContext, since the latter implies sws_init_context.
The problem here is that sws_init_context sets up the range conversion
and fast path tables based on the values of srcRange/dstRange at init
time. This may result in locking in a "wrong" path (either using
unscaled fast path when range conversion later required, or using
scaled slow path when range conversion becomes no longer required).
There are two way outs:
1. Always initialize range conversion and unscaled converters, even if
they will be unused, and extend the runtime check.
2. Re-do initialization if the values change after
sws_setColorspaceDetails.
I opted for approach 1 because it was simpler and easier to reason
about.
Reword the av_log message to make it clear that this special converter
is not necessarily used, depending on whether or not there is range
conversion or YUV matrix conversion going on.
libavcodec/aarch64/vc1dsp_neon.S is skipped here, as it intentionally
uses a layered indentation style to visually show how different
unrolled/interleaved phases fit together.
Signed-off-by: Martin Storsjö <martin@martin.st>
This is slower than the Zbb version on real hardware due to register
strides. Proper support for vector byte-swap requires the Zvbb
extension, but it's much too early for me to worry about it.
Including winsock2.h or windows.h without WIN32_LEAN_AND_MEAN cause
bzlib.h to parse as nonsense, due to an instance of #define char small
in rpcndr.h.
See:
https://stackoverflow.com/a/27794577
Signed-off-by: L. E. Segovia <amy@amyspark.me>
Signed-off-by: Martin Storsjö <martin@martin.st>
The code was blindly assuming that Zbb or V implied Zba. While the
earlier is practically always true, the later broke some QEMU setups,
as V was introduced earlier than Zba.
Add missing operand which clang complains about but GCC assumes it to be
'm1' if not specified.
Works around build failure with Clang:
| src/libswscale/riscv/rgb2rgb_rvv.S:88:25: error: operand must be e[8|16|32|64|128|256|512|1024],m[1|2|4|8|f2|f4|f8],[ta|tu],[ma|mu]
| vsetvli t4, t3, e8, ta, ma
| ^
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>
From x86inc:
> On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either
> a branch or a branch target. So switch to a 2-byte form of ret in that case.
> We can automatically detect "follows a branch", but not a branch target.
> (SSSE3 is a sufficient condition to know that your cpu doesn't have this problem.)
x86inc can automatically determine whether to use REP_RET rather than
REP in most of these cases, so impact is minimal. Additionally, a few
REP_RETs were used unnecessary, despite the return being nowhere near a
branch.
The only CPUs affected were AMD K10s, made between 2007 and 2011, 16
years ago and 12 years ago, respectively.
In the future, everyone involved with x86inc should consider dropping
REP_RETs altogether.
Currently, it is done once per slice-thread, leading to
one warning per slice-thread in case a YUVJ pixel format
has been originally used.
This also fixes the anomaly that said parameter are only
updated for the user-facing context (whose values are retrievable
via av_opt_get()) if slice-threading is not in use.
Fixes ticket #9860.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Initializing slice threads currently uses the function
(sws_init_context()) that is also used for initializing
user-facing contexts with the only difference being that
nb_threads is set to one before initializing the slice contexts.
Yet sws_init_context() also initializes lots of stuff
that is not slice-dependent, i.e. (src|dst)Range. This
currently only works because the code sets these fields
to the same values for all slice contexts. This is not
nice; even worse, it entails that log messages are printed
once per slice context (and therefore fill the screen).
This commit lays the groundwork to fix this.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Same principle as previous commit, with sufficiently huge rgb2yuv table
values this produces wrong results and undefined behavior.
The unsigned produces the same incorrect results. That is probably
ok as these cases with huge values seem not to occur in any real
use case.
Fixes: signed integer overflow
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Large rgb2yuv tables and high pixel values cause the intermediate
int32_t of ru*r + gu*g + bu*b to exceed INT_MAX, which is undefined
behavior. This causes libswscale built with LLVM -fsanitize=undefined to
assert. Using unsigned integers instead has defined behavior and
produces identical results, and makes rgb64ToUV_c_template match
rgb64ToY_c_template.
Fixes: signed integer overflow
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
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>
Rémi Denis-Courmont
Niklas Haas
Michael Niedermayer
Martin Storsjö
Paul B Mahol
Andreas Rheinhardt
L. E. Segovia
Khem Raj
Lynne
Jin Bo
Lu Wang
James Almer
Tomas Härdin
Jeremy Dorfman
Hubert Mazur