GitLab

Calling get_uniform(max=1) results in a read_bits(n=0),
In get_uniform, the n param is renamed to max to clarify the
semantics. Asserts are added to detect calls to get_uniform()
and get_bits() that don't actually read anything.

Closes #76

Ronald S. Bultje authored Oct 23, 2018 and Jean-Baptiste Kempf committed Oct 23, 2018

This is identical to what libaom does. Fixes #65.

Fixes #69.

Also prevent using warped motion when force_integer_mv=1. Fixes #84.

Fixes #83.

Fixes #82.

armasm64 fails to assemble this:
error A2173: syntax error in expression
        sub             x7,  x7,  w4, uxth

This clearly is a bug in armasm64, and will be reported. For now,
this workaround should be harmless though, as we've just loaded
the register with ldrh, so the upper parts of the register should
be zeroed.

Fixes #60.

These can be used in intrabc coding. Fixes #74.

Fixes #80.

checkasm --bench on a Qualcomm Kryo (Sanpdragon 820):
nop: 33.0
avg_w4_8bpc_c: 450.5
avg_w4_8bpc_neon: 20.1
avg_w8_8bpc_c: 438.6
avg_w8_8bpc_neon: 45.2
avg_w16_8bpc_c: 1003.7
avg_w16_8bpc_neon: 112.8
avg_w32_8bpc_c: 3249.6
avg_w32_8bpc_neon: 429.9
avg_w64_8bpc_c: 7213.3
avg_w64_8bpc_neon: 1299.4
avg_w128_8bpc_c: 16791.3
avg_w128_8bpc_neon: 2978.4
w_avg_w4_8bpc_c: 605.7
w_avg_w4_8bpc_neon: 30.9
w_avg_w8_8bpc_c: 545.8
w_avg_w8_8bpc_neon: 72.9
w_avg_w16_8bpc_c: 1430.1
w_avg_w16_8bpc_neon: 193.5
w_avg_w32_8bpc_c: 4876.3
w_avg_w32_8bpc_neon: 715.3
w_avg_w64_8bpc_c: 11338.0
w_avg_w64_8bpc_neon: 2147.0
w_avg_w128_8bpc_c: 26822.0
w_avg_w128_8bpc_neon: 4596.3
mask_w4_8bpc_c: 604.6
mask_w4_8bpc_neon: 37.2
mask_w8_8bpc_c: 654.8
mask_w8_8bpc_neon: 96.0
mask_w16_8bpc_c: 1663.0
mask_w16_8bpc_neon: 272.4
mask_w32_8bpc_c: 5707.6
mask_w32_8bpc_neon: 1028.9
mask_w64_8bpc_c: 12735.3
mask_w64_8bpc_neon: 2533.2
mask_w128_8bpc_c: 31027.6
mask_w128_8bpc_neon: 6247.2

Ordering the elements this way is more SIMD-friendly.

Janne Grunau authored Oct 18, 2018 and Ronald S. Bultje committed Oct 20, 2018

The simplified versions catches tg_start != 0 for the first tile group
too. Fixes a heap-buffer-overflow in setup_tile when the first tile index is
larger than 0 but the number of tiles is correct. Fixes #61, #73.

Helped-by: Henrik Gramner <gramner@twoorioles.com>

This will help when writing x86_64 assembly.

All the functions are public, and the only prototype in this header
is a duplicate.

Janne Grunau authored Oct 17, 2018 and Ronald S. Bultje committed Oct 18, 2018

Fix following ubsan error in #68:
../src/env.h:296:24: runtime error: shift exponent -1 is negative
[Detaching after fork from child process 22253]
    #0 0x7ffff76ad6f9 in get_poc_diff /home/janne/src/dav1d/build-usan/../src/env.h:296:24
    #1 0x7ffff76ad6f9 in parse_frame_hdr /home/janne/src/dav1d/build-usan/../src/obu.c:757
    #2 0x7ffff7696491 in dav1d_parse_obus /home/janne/src/dav1d/build-usan/../src/obu.c:1023:20
    #3 0x7ffff7921c7d in dav1d_decode /home/janne/src/dav1d/build-usan/../src/lib.c:193:20
    #4 0x424869 in main /home/janne/src/dav1d/build-usan/../tools/dav1d.c:108:20
    #5 0x7ffff63dfae6 in __libc_start_main (/lib64/libc.so.6+0x21ae6)
    #6 0x403489 in _start (/home/janne/src/dav1d/build-usan/tools/dav1d+0x403489)

I can't reproduce the ubsan error in the issue.

Janne Grunau authored Oct 17, 2018 and Ronald S. Bultje committed Oct 18, 2018

Reject out of range values as errors and avoid undefined shifts. Fixes #67.

Janne Grunau authored Oct 17, 2018 and Ronald S. Bultje committed Oct 18, 2018

Fixes #66.

Janne Grunau authored Oct 17, 2018 and Ronald S. Bultje committed Oct 18, 2018

Fixes #62.

Janne Grunau authored Oct 15, 2018 and Ronald S. Bultje committed Oct 18, 2018

Reverts commit f17c5e08 "check IntraBC
displacement vector".
Fixes #63 and an overlapping memcpy in
clusterfuzz-testcase-minimized-dav1d_fuzzer-5675967192760320. Credits to
oss-fuzz and Tyson Smith.

wiener_luma_8bpc_c: 326272.1
wiener_luma_8bpc_avx2: 19841.5

Decoding time of first 1000 frames of Chimera-8bit-1920x1080.ivf goes
from 27.471 to 23.558 seconds.

Also copy 4 pixels so SIMD can use a padded write (movd).

Deriving uvtx for chroma-from-luma is the same for lossless.
The availability of CfL is constrained to when they match.

Removes the splat of pixels that are then overwritten.
Rename cfl_pred_1 to cfl_pred now that there is only one.

Janne Grunau authored Oct 13, 2018 and Jean-Baptiste Kempf committed Oct 13, 2018

Fixes emmory leak with asan seen with 'dav1d --tilethreads 2 ...'