We just need to ensure the palette contains valid data, which will happen
automatically as long as the plane 1 is large enough.
Sponsored-by: Sovereign Tech Fund
Signed-off-by: Niklas Haas <git@haasn.dev>
The only noticable changes in benchmarks are for
the x2 horizontal no_rnd case where SSE2 and movhps
are beneficial:
Old benchmarks:
avg_pixels_tab[1][1]_c: 42.2 ( 1.00x)
avg_pixels_tab[1][1]_mmxext: 10.8 ( 3.89x)
avg_pixels_tab[1][2]_c: 18.0 ( 1.00x)
avg_pixels_tab[1][2]_mmxext: 6.1 ( 2.96x)
put_no_rnd_pixels_tab[1][1]_c: 29.7 ( 1.00x)
put_no_rnd_pixels_tab[1][1]_mmxext: 12.3 ( 2.41x)
put_no_rnd_pixels_tab[1][2]_c: 20.4 ( 1.00x)
put_no_rnd_pixels_tab[1][2]_mmxext: 12.2 ( 1.67x)
put_pixels_tab[1][1]_c: 29.9 ( 1.00x)
put_pixels_tab[1][1]_mmxext: 7.6 ( 3.92x)
put_pixels_tab[1][2]_c: 16.8 ( 1.00x)
put_pixels_tab[1][2]_mmxext: 6.4 ( 2.63x)
New benchmarks:
avg_pixels_tab[1][1]_c: 42.3 ( 1.00x)
avg_pixels_tab[1][1]_sse2: 10.7 ( 3.95x)
avg_pixels_tab[1][2]_c: 17.8 ( 1.00x)
avg_pixels_tab[1][2]_sse2: 6.3 ( 2.83x)
put_no_rnd_pixels_tab[1][1]_c: 29.6 ( 1.00x)
put_no_rnd_pixels_tab[1][1]_sse2: 10.5 ( 2.81x)
put_no_rnd_pixels_tab[1][2]_c: 20.4 ( 1.00x)
put_no_rnd_pixels_tab[1][2]_sse2: 12.3 ( 1.67x)
put_pixels_tab[1][1]_c: 30.1 ( 1.00x)
put_pixels_tab[1][1]_sse2: 7.6 ( 3.93x)
put_pixels_tab[1][2]_c: 16.8 ( 1.00x)
put_pixels_tab[1][2]_sse2: 6.4 ( 2.64x)
Switching to SSE2 unfortunately increased codesize of the relevant
functions by 160B.
This makes these functions ABI compatible, i.e. they no longer
rely on others calling emms_c to fix the fpu state. It also
implies that many mpegvideo decoders (the exceptions are MPEG-4,
RV30, RV40 and the VC-1 family) now no longer use any mmx registers
at all. So one can remove the emms_c from the MPEG-1/2 decoder.
The same is true for VP3.
Reviewed-by: James Almer <jamrial@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
No change in benchmarks here; this already allows
to remove an emms_c from cavsdec.c.
Reviewed-by: James Almer <jamrial@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
I want to start adding more data layouts, like semiplanar formats (nv12), or
palette formats. I made an effort to distinguish existing checks for rw.packed
into "mode != PLANAR" and "mode == PACKED", based on the intent of the
surrounding code, in anticipation of these new layouts.
Sponsored-by: Sovereign Tech Fund
Signed-off-by: Niklas Haas <git@haasn.dev>
See previous commit for justification. I decided to split these
refactors up into several independent commits to make it easier
to review and bisect, since they are all independent atomic changes.
Sponsored-by: Sovereign Tech Fund
Signed-off-by: Niklas Haas <git@haasn.dev>
Instead of hard-coding SWS_PIXEL_F32 here. This is not really useful
yet, but I wanted to clean up the semantics here regardless.
Sponsored-by: Sovereign Tech Fund
Signed-off-by: Niklas Haas <git@haasn.dev>
This is a minor cosmetic improvement that allows me to use more
convenient names for a filter-related metadata fields, without
confusion.
Sponsored-by: Sovereign Tech Fund
Signed-off-by: Niklas Haas <git@haasn.dev>
This ensures 100% coverage of all uop primitives by generating the set of
tests exactly from the list of seen primitives, using the uops macros.
There are some annoying quirks still because of the fact that we have to
essentially "untranslate" the UOPs back to SwsOps that result back in the
intended uop after the translation, but overall it's not too bad and still
much better than the status quo of hand-rolling the list of test cases.
Signed-off-by: Niklas Haas <git@haasn.dev>
Replace plain memcmp+fail() with checkasm_check_pixel_padded() for
DC, planar, and angular prediction tests. Use PIXEL_RECT for output
buffers instead of flat arrays.
This enables:
- Detailed per-pixel difference output when run with 'checkasm -v'
- Detection of out-of-bounds writes beyond the NxN block area
- Padding violation reporting (writes past block boundary)
Previously, a test failure would only report "FAILED" with no
information about which pixels were wrong, making assembly debugging
difficult. Follows the pattern established in 4d4b301e4a (checkasm:
hevc_pel: Use helpers for checking for writes out of bounds).
Suggested-by: Martin Storsjö <martin@martin.st>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>
The current approach of re-testing the C reference for every backend
separately leads to both confusing output (e.g. having an extra redundant
`memcpy_c` line for every op, even those not implemented by the memcpy
backend), as well as a lot of unnecessary wasted time re-testing and
re-benching the same C variant for every backend.
This new API function lets us test the C function only a single time, while
simultaneously having all of the other backends implicitly compare themselves
against the C reference.
Signed-off-by: Niklas Haas <git@haasn.dev>
Commit 4569ab7eaa tried to set this
only on the object files for the checkasm library itself, but
missed that EXT_CHECKASMOBJS lacks the path prefix, thus this
wasn't set at all.
Alternatively, for simplicity, we could keep passing this for
all checkasm object files, not only the checkasm library objects;
the other object files don't use it in any case.
Fixes stack overflow on Windows when by default we have 1 MB.
Individually those functions fit, but when they are all inlined, it's
too much.
Signed-off-by: Kacper Michajłow <kasper93@gmail.com>
This is required for overriding defines that exist in the public
headers of checkasm, when e.g. building with assembly disabled
for an architecture where we normally would use the checked_call
wrapper.
This fixes a leftover in how checkasm is integrated into the
ffmpeg build system; there were many different approaches
considered for fixing --disable-asm, and the ffmpeg configure
integration didn't end up matching the final solution.
This fixes building with --disable-asm.
The checkasm tool originated in x264. It was later rewritten and
modernized for FFmpeg (and relicensed to LGPL). For the dav1d
project, it was relicensed again to 2-clause BSD (with permission
from the relevant authors).
The FFmpeg and dav1d implementations of checkasm have since evolved
independently (with some amount of ported code between the two,
with relicensing permission where relevant).
To synchronize the development, and to make it possible to easily
adopt checkasm in other projects, it has been split out into a
standalone project/library on its own, developed at
https://code.videolan.org/videolan/checkasm/.
That version has all the features of checkasm in both FFmpeg and
dav1d, and has got a number of extra improvements on top:
- More/fixed tests (e.g. properly clobbering high bits of 32-bit registers
on most platforms),
- Vastly improved overall performance / runtime for benchmarking, due
primarily to the ability to scale the runtime of each test to that test's
complexity.
- Much more robust statistical analysis of benchmarking results; including
robust outlier rejection, an estimation of the histogram, and the ability
to report the variance / stddev in addition to the (trimmed) mean.
- Interactive HTML and JSON output formats in addition to CSV/TSV.
- More readable and user-friendly output across the board, especially for
failures and data dumps (e.g. also showing errors inside padding bytes).
- Better cross-platform support, including dynamic fallback of timer
implementations on ARM platforms, a better RISC-V and AArch64 harness,
and more.
On AArch64, it tests which timer out of pmccntr_el0, linux perf,
macos kperf, cntvct_el0 is available, without the user needing to
configure things, and falling back on clock_gettime if neither of
them can be used. This means one automatically gets the best
available timer, if userspace access to pmccntr_el0 has been
unlocked with a kernel module, or if one has permission to use
the perf API, or if the cntvct_el0 is exact enough to be useful.
On AArch64 macOS, there is now a test harness that catches clobbered
registers and stack clobbering, like on other platforms.
- An option for setting affinity, for benchmarking on heterogenous
core systems. (On Linux, this is already easily done through
taskset, but on Windows, the checkasm built in option makes it
possible there as well, and portable.)
- Printing of the tested CPU core name, where possible.
To integrate this external implementation of checkasm into FFmpeg,
without having to build libcheckasm as an external library, the upstream
sources are added as a git subtree, and integrated into the FFmpeg
build system as a foreign source.
For the long and storied history of how we arrived at this solution,
see: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/22546
The relevant config headers for checkasm are generated by configure,
and the sources are built as part of the main ffmpeg build. The
upstream sources, while they use meson as primary build system,
are structured to make it easy to build as part of a foreign build
system.
The existing testcases are mostly kept untouched (only three minor
changes are required, in crc.c, sw_ops.c and vp8dsp.c), while the
majority of the logic from checkasm.c, checkasm.h and the arch
specific assembly files are removed, replaced with the external
implementation.
Co-Authored-By: Martin Storsjö <martin@martin.st>
Signed-off-by: Niklas Haas <git@haasn.dev>
To reproduce this commit, run:
$ git subtree add --squash --prefix=tests/checkasm/ext \
https://code.ffmpeg.org/FFmpeg/checkasm.git master
To update at a later point in time, replace `add` by `pull`
Not only is this duplicating code, but it also hard-codes a reference to
`checkasm_lfg`, which I want to eliminate in the interest of being able to
switch out the checkasm implementation.
The test data size is quite large, so re-setting up unused data is eating up
quite a significant amount of CPU time.
This commit cuts execution time of sw_ops in half.
Signed-off-by: Niklas Haas <git@haasn.dev>
Should fix buffer overflows as reported by clang-asan and use of uninitialized
values as reported by valgrind.
Signed-off-by: James Almer <jamrial@gmail.com>
The previous chroma stride formula (width >> log2_chroma_w) is correct
for planar yuv but wrong for semi-planar nv12/nv21, where the UV plane
is interleaved at width bytes per row (width/2 UV pairs of 2 bytes
each). Use av_image_get_linesize() so the test feeds a valid stride to
libswscale regardless of input format; for the existing planar suites
the value is unchanged.
With the stride fixed, add nv12 and nv21 to check_yuv2rgb() so the
upcoming NEON 16bpp paths get bench coverage. ff_get_unscaled_swscale
does not wire a C yuv2rgb fast path for these inputs, so the suites
report bench-only (no correctness reference); they still run clobber
detection and cycle counts.
Signed-off-by: DROOdotFOO <drew@axol.io>
Unfortunately a bit slower than the MMX version due to
the impossibility to use memory operands in paddw.
The situation would reverse if ff_dctB_mmx() would have
to issue emms.
dctB_c: 3.7 ( 1.00x)
dctB_mmx: 3.3 ( 1.13x)
dctB_sse2: 3.6 ( 1.03x)
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Test 3-tap for 8x8/16x16/32x32 (both filtered_left and
filtered_top outputs). Test strong smoothing for filtered_top
and in-place left modification.
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>
This change should improve performance on Skylake and later
Intel CPUs (which have only half the ports for saturated adds/subs
for mmx register compared to xmm register): llvm-mca predicts
a 25% performance improvement on Skylake.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Due to a discrepancy between SSE2 and the C version coefficients
for idct_put and idct_add are restricted to a range not causing
overflows.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This allows to use pavgb to reduce the amount of instructions used
to calculate the average; processing two rows via movhps allows
to reduce the amount of pxor and pavgb even further and turned
out to be beneficial.
This patch also avoids a load as the constant used here can be easily
generated at runtime.
Old benchmarks:
put_no_rnd_pixels_l2_c: 13.3 ( 1.00x)
put_no_rnd_pixels_l2_mmx: 11.6 ( 1.15x)
New benchmarks:
put_no_rnd_pixels_l2_c: 13.4 ( 1.00x)
put_no_rnd_pixels_l2_sse2: 7.5 ( 1.77x)
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Instead of implicitly testing for NaN values. This is mostly a straightforward
translation, but we need some slight extra boilerplate to ensure the mask
is correctly updated when e.g. commuting past a swizzle.
Signed-off-by: Niklas Haas <git@haasn.dev>
These can randomly trigger the alpha/zero fast paths, resulting in spurious
tests or randomly diverging performance if the backend happens to implement
that particular fast path.
Signed-off-by: Niklas Haas <git@haasn.dev>
This was not actually testing integer path. Additionally, for integer
scales, there is a special fast path for expansion from bits to full range,
which we should separate from the random value test.
Most of these filters don't test anything meaningfully different relative to
each other; the only filters that really have special significant are POINT
(for now) and maybe BILINEAR down the line.
Apart from that, SINC, combined with the src size loop, already tests both
extreme cases (large and small filters), with large, oscillating unwindonwed
weights.
The other filters are not adding anything of substance to this, while massively
slowing down the runtime of this test. We can, of course, change this if the
backends ever get more nuanced handling.
checkasm: all 855 tests passed (down from 1575)
Signed-off-by: Niklas Haas <git@haasn.dev>
The current code was a bit clumsy in that it always picked the first
available backend when choosing the new function. This meant that some x86
paths were not being tested at all, whenever the memcpy backend (which has
higher priority) could serve the request.
This change makes it so that each backend is explicitly tested against only
implementations provided by that same backend.
checkasm: all 1575 tests passed (up from 1305)
As an aside, it also lets us benchmark the memcpy backend directly against
the C reference backend.
Signed-off-by: Niklas Haas <git@haasn.dev>
These don't actually exist at runtime, and will soon be removed from the
backends as well.
This commit is intentionally a bit incomplete; as I will rewrite this
based on the auto-generated macros in the upcoming ops_micro series.
Signed-off-by: Niklas Haas <git@haasn.dev>
They have been superseded by SSSE3; the SSE2 version was even disabled
(and segfaults if enabled).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Niklas HaasandNiklas Haas
Ramiro Polla
Zhao ZhiliandNiklas Haas
Andreas Rheinhardt
DROOdotFOOandRamiro Polla
Martin Storsjö
Jun ZhaoandJun Zhao
Kacper Michajłow
James Almer