These inclusions are not necessary, as cpu.h is already included
wherever it is needed (via direct inclusion or via the arch-specific
headers).
Also remove other unnecessary cpu.h inclusions from ordinary
non-headers.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Dither none is only implemented in full chroma interpolation for these rgb formats
Its also a obscure choice (producing less nice images) that implementing it in the
other code-paths makes no sense
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Call the scaler function directly rather than through a function
pointer. Drop the now-unused return value from ff_getSwsFunc() and
rename the function to reflect its new role.
This will be useful in the following commits, where it will become
important that the amount of output is different for scaled vs unscaled
case.
Call ff_sws_rgb2rgb_init via ff_thread_once instead of checking one of the
variables it updates.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Some files currently rely on libavutil/cpu.h to include it for them;
yet said file won't use include it any more after the currently
deprecated functions are removed, so include attributes.h directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The last user of g15Mask, r15Mask, g16Mask and r16Mask was disabled
in 77a416e8aa and finally removed in
36e8de07ed62609df45d064b56501e3084d25723; b15Mask and b16Mask were
apparently always unused (except for in_asm_used_var_warning_killer,
a function that only existed to make the compiler not optimize ASM
constants away).
w10 is unused since d604bab901, w02
since ef423a6618.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
mask24hh etc. are unused since f099fbf5f3,
mask32b and mask32r since 296609f859,
mask32g since b38d487466 and mask32 since
f8a138be52.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Add inline function for vec_xl if VSX is not supported. vec_xl intrinsic
is only available on POWER 7 or higher.
Fixes ticket #8750.
Signed-off-by: Andriy Gelman <andriy.gelman@gmail.com>
Fixes fate-qtrle-32bit on big-endian.
The macro does a simple byte swap on uint8 array without any casts, so
it's valid on big-endian arches.
The mentioned test was failing because the byteswap function
shuffle_bytes_3210_c() is used in the pixel format conversion
(argb->bgra).
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andriy Gelman <andriy.gelman@gmail.com>
Fixes vf_scale outputting RGB AVFrames with limited range flagged
in case either input or output specifically sets the range.
This is the reverse of the logic utilized for RGB and PAL8 content
in sws_setColorspaceDetails.
These conversion appears to be exhibiting the same rounding error as the rgbf32 formats where.
I seperated the rounding value from the 16 and 128 offsets, I think it makes it a little more clear.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
changes since v1:
- made into fate test
- fixed c90 warnings
- tests more intermediate formats
- tested on BE mips too
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Regression since: 3adffab073
-1 is consistent what other error paths return
Reviewed-by: Martin Storsjö <martin@martin.st>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
We return 0 for this particular architecture but should instead be
returning the number of lines.
Fixes users who check the return value matches what they expect.
256 bits is just wide enough to fit all the operands needed to vectorize
the software implementation, but AVX2 is needed to for a couple of
instructions like cross-lane permutation.
Output is bit-for-bit identical to C.
Signed-off-by: Nelson Gomez <nelson.gomez@microsoft.com>
Extracting information from SwsContext in assembly is difficult, and
rearranging SwsContext just for asm access didn't look good. These
functions only need a couple of fields from it anyway, so just make
them parameters in their own right.
Signed-off-by: Nelson Gomez <nelson.gomez@microsoft.com>
The NEON hscale function only supports X8 filter sizes and should only
be selected when these are being used. At the moment filterAlign is
set to 8 but in the future when extra NEON assembly for specific sizes is
added they will need to have checks here too.
The immediate usecase for this change is making the hscale checkasm
test easier and without NEON specific edge-cases (x86 already has these
guards).
This applies the same fix from 718c8f9aa5
on the 32 bit arm version of the function, fixing fate-checkasm-sw_scale
there.
Signed-off-by: Martin Storsjö <martin@martin.st>
The NEON hscale function only supports X8 filter sizes and should only
be selected when these are being used. At the moment filterAlign is
set to 8 but in the future when extra NEON assembly for specific sizes is
added they will need to have checks here too.
The immediate usecase for this change is making the hscale checkasm
test easier and without NEON specific edge-cases (x86 already has these
guards).
Signed-off-by: Josh de Kock <josh@itanimul.li>
libswscale/vscale.c makes extensive use of function pointers and in
doing so it converts these function pointers to and from a pointer to
void. Yet this is actually against the C standard:
C90 only guarantees that one can convert a pointer to any incomplete
type or object type to void* and back with the result comparing equal
to the original which makes pointers to void generic pointers to
incomplete or object type. Yet C90 lacks a generic function pointer
type.
C99 additionally guarantees that a pointer to a function of one type may
be converted to a pointer to a function of another type with the result
and the original comparing equal when converting back.
This makes any function pointer type a generic function pointer type.
Yet even this does not make pointers to void generic function pointers.
Both GCC and Clang emit warnings for this when in pedantic mode.
This commit fixes this by using a union that can hold one member of any
of the required function pointer types to store the function pointer.
This works even for C90.
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
The x18 is a reserved platform register on Darwin and Windows.
x8/w8 seems to be unused in this function though (and same about
x10 and x14), so there's really no reason to use x18 here - just change
the uses of x18/w18 into x8/w8 instead without any further rewrites.
Signed-off-by: Martin Storsjö <martin@martin.st>
Fixes: signed integer overflow: 1169365504 + 981452800 cannot be represented in type 'int'
Fixes: ticket8293
Found-by: Suhwan
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: signed integer overflow: 524280 * 4432 cannot be represented in type 'int'
Fixes: ticket8322
Found-by: Suhwan
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Add swscale input support for Y210LE, output support and fate
test could be added later if there is requirement for software
CSC to this packed format.
Signed-off-by: Linjie Fu <linjie.fu@intel.com>
Tested using this command:
/ffmpeg -pix_fmt yuv420p -s 1920*1080 -i ArashRawYuv420.yuv \
-vcodec rawvideo -s 1920*1080 -pix_fmt rgb24 -f null /dev/null
The fps increase from 389 to 640 on Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz
Signed-off-by: Ting Fu <ting.fu@intel.com>
Bug #8255 points out a double free error in libwscale/utils.c file.
The double free is because the pointer to cascaded_context of an
sw_context is not set to NULL after freeing it. When the sw_context
is later freed, sws_freeContext is called on the cascaded_context,
causing a double free.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The original inline assembly and nasm code have the same fps when called by command.
NASM code almost has no impact on the perfromance.
Signed-off-by: Ting Fu <ting.fu@intel.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
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>
