out[lut[i]] = in[i] lookups were 4.04 times(!) slower than
out[i] = in[lut[i]] lookups for an out-of-place FFT of length 4096.
The permutes remain unchanged for anything but out-of-place monolithic
FFT, as those benefit quite a lot from the current order (it means
there's only 1 lookup necessary to add to an offset, rather than
a full gather).
The code was based around non-power-of-two FFTs, so this wasn't
benchmarked early on.
No buffer will be fetched from the pool after it's uninitialized, so there's
no benefit from waiting until every single buffer has been returned to it
before freeing them all.
This should free some memory in certain scenarios, which can be beneficial in
low memory systems.
Based on a patch by Jonas Karlman.
Reviewed-by: Anton Khirnov <anton@khirnov.net>
Signed-off-by: James Almer <jamrial@gmail.com>
This is much less precise than the cycle counter register, but
the cycle counter register is not available on apple platforms
(and on linux, it requires a kernel module for allowing user mode
access).
Signed-off-by: Martin Storsjö <martin@martin.st>
This commit adds support for in-place FFT transforms. Since our
internal transforms were all in-place anyway, this only changes
the permutation on the input.
Unfortunately, research papers were of no help here. All focused
on dry hardware implementations, where permutes are free, or on
software implementations where binary bloat is of no concern so
storing dozen times the transforms for each permutation and version
is not considered bad practice.
Still, for a pure C implementation, it's only around 28% slower
than the multi-megabyte FFTW3 in unaligned mode.
Unlike a closed permutation like with PFA, split-radix FFT bit-reversals
contain multiple NOPs, multiple simple swaps, and a few chained swaps,
so regular single-loop single-state permute loops were not possible.
Instead, we filter out parts of the input indices which are redundant.
This allows for a single branch, and with some clever AVX512 asm,
could possibly be SIMD'd without refactoring.
The inplace_idx array is guaranteed to never be larger than the
revtab array, and in practice only requires around log2(len) entries.
The power-of-two MDCTs can be done in-place as well. And it's
possible to eliminate a copy in the compound MDCTs too, however
it'll be slower than doing them out of place, and we'd need to dirty
the input array.
This patch also fixes a -Wtautological-constant-out-of-range-compare
warning from Clang and a -Wtype-limits warning from GCC on systems
where size_t is 64bits and unsigned 32bits. The reason for this seems
to be that variable (whose value derives from sizeof() and can therefore
be known at compile-time) is used instead of using sizeof() directly in
the comparison.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
libavutil/common.h is a public header that provides generic math
functions whereas libavutil/intmath.h is a private header that contains
plattform-specific optimized versions of said math functions. common.h
includes intmath.h (when building the FFmpeg libraries) so that the
optimized versions are used for them.
This interdependency sometimes causes trouble: intmath.h once contained
an inlined ff_sqrt function that relied upon av_log2_16bit. In case there
was no optimized logarithm available on this plattform, intmath.h needed
to include common.h to get the generic implementation and this has been
done after the optimized versions (if any) have been provided so that
common.h used the optimized versions; it also needed to be done before
ff_sqrt. Yet when intmath.h was included from common.h and if an ordinary
inclusion guard was used by common.h, the #include "common.h" in intmath.h
was a no-op and therefore av_log2_16bit was still unknown at the end of
intmath.h (and also in ff_sqrt) if no optimized version was available.
Before a955b59658 this was solved by
duplicating the #ifndef av_log2_16bit check after the inclusion of
common.h in intmath.h; said commit instead moved these checks to the
end of common.h, outside the inclusion guards and made common.h include
itself to get these unguarded defines. This is still the current
state of affairs.
Yet this is unnecessary since 9734b8ba56
as said commit removed ff_sqrt as well as the #include "common.h" from
intmath.h. Therefore this commit moves everything inside the inclusion
guards and makes common.h not include itself.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Fixes: negation of -9223372036854775808 cannot be represented in type 'int64_t' (aka 'long'); cast to an unsigned type to negate this value to itself
Fixes: 29437/clusterfuzz-testcase-minimized-ffmpeg_dem_MOV_fuzzer-4748510022991872
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: division by zero
Fixes: 29555/clusterfuzz-testcase-minimized-ffmpeg_dem_VIVO_fuzzer-5149951447400448
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This reverts commit 6696a07ac6.
It is wrong to restrict timecodes to always contain leading zeros or for hours
or frames to be 2 chars only.
Signed-off-by: Marton Balint <cus@passwd.hu>
fixes http://trac.ffmpeg.org/ticket/9055
The hw decoder may allocate a large frame from AVHWFramesContext, and adjust width and height based on bitstream.
We need to use resolution from src frame instead of AVHWFramesContext.
test command:
ffmpeg -loglevel debug -hide_banner -hwaccel vaapi -init_hw_device vaapi=va:/dev/dri/renderD128 -hwaccel_device va -hwaccel_output_format vaapi -init_hw_device vulkan=vulk -filter_hw_device vulk -i 1920x1080.264 -c:v libx264 -r:v 30 -profile:v high -preset veryfast -vf "hwmap,chromaber_vulkan=0:0,hwdownload,format=nv12" -map 0 -y vaapiouts.mkv
expected:
No green bar at bottom.
Fixes: signed integer overflow: 2147462079 + 2149596 cannot be represented in type 'int'
Fixes: 27565/clusterfuzz-testcase-minimized-ffmpeg_DEMUXER_fuzzer-5091972813160448
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This patch adds support for arbitrary-point FFTs and all even MDCT
transforms.
Odd MDCTs are not supported yet as they're based on the DCT-II and DCT-III
and they're very niche.
With this we can now write tests.
Fixes: division by zero
Fixes: 26451/clusterfuzz-testcase-minimized-ffmpeg_dem_VIVO_fuzzer-4756955832516608
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
INT32_MAX (2147483647) isn't exactly representable by a floating point
value, with the closest being 2147483648.0. So when rescaling a value
of 1.0, this could overflow when casting the 64-bit value returned from
lrintf() into 32 bits.
Unfortunately the properties of integer overflows don't match up well
with how a Fourier Transform operates. So clip the value before
casting to a 32-bit int.
Should be noted we don't have overflows with the table values we're
currently using. However, converting a Kaiser-Bessel window function
with a length of 256 and a parameter of 5.0 to fixed point did create
overflows. So this is more of insurance to save debugging time
in case something changes in the future.
The macro is only used during init, so it being a little slower is
not a problem.
Do it only when requested with the AV_CODEC_EXPORT_DATA_VIDEO_ENC_PARAMS
flag.
Drop previous code using the long-deprecated AV_FRAME_DATA_QP_TABLE*
API. Temporarily disable fate-filter-pp, fate-filter-pp7,
fate-filter-spp. They will be reenabled once these filters are converted
in following commits.
This was introduced in version 4.6. And may not exist all without an
optional package. So to prevent a hard dependency on needing the Linux
kernel headers to compile, make this optional.
Also ignore the status of the ioctl, since it may fail on older kernels
which don't support it. It's okay to ignore as its not fatal and any
serious errors will be caught later by the mmap call.
When targeting a recent enough macOS/iOS version that has clock_gettime
it won't be a weak symbol, in which case clang warns for this check
as it's always true:
warning: address of function 'clock_gettime' will always
evaluate to 'true'
This warning is silenced by using the address-of operator to make
the intent explicit.
These two extensions and two features are both optionally used by
libplacebo to speed up rendering, so it makes sense for libavutil to
automatically enable them as well.
Vulkan formats with a PACK suffix define native endianess.
Vulkan formats without a PACK suffix are in bytestream order.
Pixel formats with a LE/BE suffix define endianess.
Pixel formats without LE/BE suffix are in bytestream order.
This relies on the fact that host memory is always going to be required
to be aligned to the platform's page size, which means we can adjust
the pointers when we map them to buffers and therefore skip an entire
copy. This has already had extensive testing in libplacebo without
problems, so its safe to use here as well.
Speeds up downloads and uploads on platforms which do not pool their
memory hugely, but less so on platforms that do.
We can pool the buffers ourselves, but that can come as a later patch
if necessary.
