The ld1r is a leftover from the arm version, where this trick is
beneficial on some cores.
Use a single-lane load where we don't need the semantics of ld1r.
This is cherrypicked from libav commit
ed8d293306.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
This avoids loading and calculating coefficients that we know will
be zero, and avoids filling the temp buffer with zeros in places
where we know the second pass won't read.
This gives a pretty substantial speedup for the smaller subpartitions.
The code size increases from 14740 bytes to 24292 bytes.
The idct16/32_end macros are moved above the individual functions; the
instructions themselves are unchanged, but since new functions are added
at the same place where the code is moved from, the diff looks rather
messy.
Before:
vp9_inv_dct_dct_16x16_sub1_add_neon: 236.7
vp9_inv_dct_dct_16x16_sub2_add_neon: 1051.0
vp9_inv_dct_dct_16x16_sub4_add_neon: 1051.0
vp9_inv_dct_dct_16x16_sub8_add_neon: 1051.0
vp9_inv_dct_dct_16x16_sub12_add_neon: 1387.4
vp9_inv_dct_dct_16x16_sub16_add_neon: 1387.6
vp9_inv_dct_dct_32x32_sub1_add_neon: 554.1
vp9_inv_dct_dct_32x32_sub2_add_neon: 5198.5
vp9_inv_dct_dct_32x32_sub4_add_neon: 5198.6
vp9_inv_dct_dct_32x32_sub8_add_neon: 5196.3
vp9_inv_dct_dct_32x32_sub12_add_neon: 6183.4
vp9_inv_dct_dct_32x32_sub16_add_neon: 6174.3
vp9_inv_dct_dct_32x32_sub20_add_neon: 7151.4
vp9_inv_dct_dct_32x32_sub24_add_neon: 7145.3
vp9_inv_dct_dct_32x32_sub28_add_neon: 8119.3
vp9_inv_dct_dct_32x32_sub32_add_neon: 8118.7
After:
vp9_inv_dct_dct_16x16_sub1_add_neon: 236.7
vp9_inv_dct_dct_16x16_sub2_add_neon: 640.8
vp9_inv_dct_dct_16x16_sub4_add_neon: 639.0
vp9_inv_dct_dct_16x16_sub8_add_neon: 842.0
vp9_inv_dct_dct_16x16_sub12_add_neon: 1388.3
vp9_inv_dct_dct_16x16_sub16_add_neon: 1389.3
vp9_inv_dct_dct_32x32_sub1_add_neon: 554.1
vp9_inv_dct_dct_32x32_sub2_add_neon: 3685.5
vp9_inv_dct_dct_32x32_sub4_add_neon: 3685.1
vp9_inv_dct_dct_32x32_sub8_add_neon: 3684.4
vp9_inv_dct_dct_32x32_sub12_add_neon: 5312.2
vp9_inv_dct_dct_32x32_sub16_add_neon: 5315.4
vp9_inv_dct_dct_32x32_sub20_add_neon: 7154.9
vp9_inv_dct_dct_32x32_sub24_add_neon: 7154.5
vp9_inv_dct_dct_32x32_sub28_add_neon: 8126.6
vp9_inv_dct_dct_32x32_sub32_add_neon: 8127.2
This is cherrypicked from libav commit
a63da4511d.
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows reusing the macro for a separate implementation of the
pass2 function.
This is cherrypicked from libav commit
79d332ebbd.
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows reusing the macro for a separate implementation of the
pass2 function.
This is cherrypicked from libav commit
47b3c2c18d.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
This reduces the code size of libavcodec/aarch64/vp9itxfm_neon.o from
19496 to 14740 bytes.
This gives a small slowdown of a couple of tens of cycles, but makes
it more feasible to add more optimized versions of these transforms.
Before:
vp9_inv_dct_dct_16x16_sub4_add_neon: 1036.7
vp9_inv_dct_dct_16x16_sub16_add_neon: 1372.2
vp9_inv_dct_dct_32x32_sub4_add_neon: 5180.0
vp9_inv_dct_dct_32x32_sub32_add_neon: 8095.7
After:
vp9_inv_dct_dct_16x16_sub4_add_neon: 1051.0
vp9_inv_dct_dct_16x16_sub16_add_neon: 1390.1
vp9_inv_dct_dct_32x32_sub4_add_neon: 5199.9
vp9_inv_dct_dct_32x32_sub32_add_neon: 8125.8
This is cherrypicked from libav commit
115476018d.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
This reduces the code size of libavcodec/arm/vp9itxfm_neon.o from
15324 to 12388 bytes.
This gives a small slowdown of a couple tens of cycles, up to around
150 cycles for the full case of the largest transform, but makes
it more feasible to add more optimized versions of these transforms.
Before: Cortex A7 A8 A9 A53
vp9_inv_dct_dct_16x16_sub4_add_neon: 2063.4 1516.0 1719.5 1245.1
vp9_inv_dct_dct_16x16_sub16_add_neon: 3279.3 2454.5 2525.2 1982.3
vp9_inv_dct_dct_32x32_sub4_add_neon: 10750.0 7955.4 8525.6 6754.2
vp9_inv_dct_dct_32x32_sub32_add_neon: 18574.0 17108.4 14216.7 12010.2
After:
vp9_inv_dct_dct_16x16_sub4_add_neon: 2060.8 1608.5 1735.7 1262.0
vp9_inv_dct_dct_16x16_sub16_add_neon: 3211.2 2443.5 2546.1 1999.5
vp9_inv_dct_dct_32x32_sub4_add_neon: 10682.0 8043.8 8581.3 6810.1
vp9_inv_dct_dct_32x32_sub32_add_neon: 18522.4 17277.4 14286.7 12087.9
This is cherrypicked from libav commit
0331c3f5e8.
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids concatenation, which can't be used if the whole macro
is wrapped within another macro.
This is also arguably more readable.
This is cherrypicked from libav commit
58d87e0f49.
Signed-off-by: Martin Storsjö <martin@martin.st>
1. limit to single layer, as there is no current support for setting distortion/quality of multiple layers
2. encoder mode should be kept at default setting (0)
3. remove fixed_alloc parameter from context : seldom if ever used, and no way of properly configuring at the moment
4. add irreversible setting, to allow for lossless encoding. Set to OpenJPEG default (enabled)
5. set numresolution max to 33, which is the maximum number of allowed resolutions according the J2K spec
Signed-off-by: Michael Bradshaw <mjbshaw@google.com>
Make it clear that there is no timing-dependent behavior. In particular,
there is no state in which both input and output are denied, and where
you have to wait for a while yourself to make progress (apparently some
hardware decoders like to do this).
Avoid wording that makes references to time. It shouldn't be mistaken
for some kind of asynchronous API (like POSIX read() can return EAGAIN
if there is no new input yet). It's a state machine, so try to use
appropriate terms.
Signed-off-by: Diego Biurrun <diego@biurrun.de>
Merges Libav commit 8a60bba0ae.
Apparently the demuxer outputs the wrong padding for HE-AAC (based on
the raw sample rate, or so). aacdec contains a hack to adjust the muxer
padding accordingly before it's used to trim the decoder output. This
modified the packet side data, which in combination with the old
decoding API would change the packet the user passed to the decoder.
This is clearly not allowed, and it breaks running some gapless fate
tests with "-fflags +keepside" applied (without keepside, the packet
metadata is typically newly allocated, essentially making a copy and not
modifying the user's input packet).
This should probably be fixed in the demuxer (and consequently also the
muxer), but for now only fix the immediate problem.
Regression since 946ed78f5f (2012).
Fixes: timeout in 730/clusterfuzz-testcase-5265113739165696 (part 2 of 2)
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/targets/ffmpeg
Reviewed-by: BBB
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: timeout in 730/clusterfuzz-testcase-5265113739165696 (part 1 of 2)
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/targets/ffmpeg
Reviewed-by: BBB
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The constants used in the decoder used floating point precision,
and this caused different values to be generated on different
architectures.
So, eradicate floating point numbers and use fixed point (32.32)
arithmetics everywhere, replacing constants with precomputed integer
values.
Signed-off-by: Vittorio Giovara <vittorio.giovara at gmail.com>
Signed-off-by: Paul B Mahol <onemda@gmail.com>
The way videotoolbox hooks in as a hwaccel is pretty hacky. The VT decode
API is not invoked until end_frame(), so alloc_frame() returns a dummy
frame with a 1-byte buffer. When end_frame() is eventually called, the
dummy buffer is replaced with the actual decoded data from
VTDecompressionSessionDecodeFrame().
When the VT decoder fails, the frame returned to the h264 decoder from
alloc_frame() remains invalid and should not be used. Before
9747219958, it was accidentally being
returned all the way up to the API user. After that commit, the dummy
frame was unref'd so the user received an error.
However, since that commit, VT hwaccel failures started causing random
segfaults in the h264 decoder. This happened more often on iOS where the
VT implementation is more likely to throw errors on bitstream anomolies.
A recent report of this issue can be see in
http://ffmpeg.org/pipermail/libav-user/2016-November/009831.html
The issue here is that the dummy frame is still referenced internally by the
h264 decoder, as part of the reflist and cur_pic_ptr. Deallocating the
frame causes assertions like this one to trip later on during decoding:
Assertion h->cur_pic_ptr->f->buf[0] failed at src/libavcodec/h264_slice.c:1340
With this commit, we leave the dummy 1-byte frame intact, but avoid returning it
to the user.
This reverts commit 9747219958.
Signed-off-by: wm4 <nfxjfg@googlemail.com>
Without this the FPU state becomes trashed and causes mysterious
fate failures with cpuflags=0
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
There is no reason that draining couldn't return an error or two. But
some decoders don't handle this very well, and might always return an
error. This can lead to API users getting into an infinite loop and
burning CPU, because no progress is made and EOF is never returned.
In fact, ffmpeg.c contains a hack against such a case. It is made
unnecessary with this commit, and removed with the next one. (This
particular error case seems to have been fixed since the hack was
added, though.)
This might lose frames if decoding returns errors during draining.
This checks the sprite delta intermediates for overflow
Fixes: 716/clusterfuzz-testcase-4890287480504320
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/targets/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Also clear the state on errors
Fixes integer overflows in 701/clusterfuzz-testcase-6594719951880192
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/targets/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes timeout with 700/clusterfuzz-testcase-5660909504561152
Fixes timeout with 702/clusterfuzz-testcase-4553541576294400
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/targets/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
