These signals were in the uncompressed frame header (as a temporary
hack), which caused two problems:
* We don't want that header to be duplicated in the slice header
* It was necessary to signal the number of bits to transmit up front
However, the filter size can be 128x128 which is greater than the SB
size, and a decoder wouldn't be able to know whether to read a bit or
not until the final SB of that 128x128 block has been decoded
(depending on whether the 128x128 is all skip or not). Therefore the
signalling was changed for 128x128 blocks so that every top left SB of
a 128x128 filter block contains a signal regardless of whether the
block is all skip or not. Also, all the MB's of 128x128 block are
filtered even if they are skip MB's. This gives the signal a purpose
even when the 128x128 block is all skip, and it also gives a slight
coding gain as it leaves a way to filter skip blocks, which was
previously forbidden.
Low latency:
PSNR YCbCr: -0.19% -0.14% -0.06%
PSNRHVS: -0.15%
SSIM: -0.13%
MSSSIM: -0.15%
CIEDE2000: -0.19%
High latency:
PSNR YCbCr: -0.03% -0.01% -0.09%
PSNRHVS: 0.04%
SSIM: 0.00%
MSSSIM: 0.02%
CIEDE2000: -0.02%
Change-Id: I69ba7144d07d388b4f0968f6a53558f480979171
To get ready for pulling AV1 to nextgenv2
Replace the experimental flag by MOTION_VAR. Rename major variables.
Change-Id: If6cf4f37b9319c46d8f90df551cc7295d66ca205
* changes:
Deringing cleanup: remove DERING_REFINEMENT (always on now)
Don't run the deringing filter on skipped blocks within a superblock
Don't dering skipped superblocks
On x86 use _mm_set_epi32 when _mm_cvtsi64_si128 isn't available
Currently the patch does not have any impact on the RD performance. The
fix could however potentially help on the next step of work, especially
when the extra altref frames allow non-zero temporal filtering strength
and their corresponding OVERLAY frames, i.e. the INTNL_OVERLAY frames
are being added.
Change-Id: I2e07fb3d0aa547a0b5dd05bb4ba865cd46309076
* changes:
Remove custom rans types
Remove add_token_no_extra.
Remove unused aom_rans_build_cdf_from_pdf
Add the tool used to generate the constrained tokenset.
Remove the starting zero from ANS CDFs.
Import the aom_read/write_symbol abstractions from aom/master
(cherry picked from aom/master commit 11206c60d930be9d29100567aa67f2a65463852a)
Includes renames in a bunch of places not handled by the original
due to differing tree states.
Change-Id: Ic74d9d8850b8c80a51e55e425bbf472a67e2653f
It was a fairly small production optimization for VP9.
Change-Id: Ie93b474ea5b7e63384a7c0b3a56b135462d1471b
(cherry picked from aom/master commit df9bb76b1330de42fe13827df4c72010adb51429)
The code that generates the raw distribution is based on a MATLAB
program by Debargha Mukherjee, and the algorithm used to quantize the
distribution comes from the ANS Toolkit by Jarek Duda.
Change-Id: Ic273f7d9e43e3ecd999e9e7e04cde57e8559375a
(cherry picked from aom/master commit ef446026aeafa318f9bee182b8c80eb4f1ef5a0a)
This brings it in line with the Daala CDFs and will make it easier to
share code.
Change-Id: Idfd2d2b33c3b9b2c4e72ce72fb3d8039013448b9
(cherry picked from aom/master commit af98507ca928afe33e9f88fdd2ca168379528d6a)
* changes:
Unfork ANS decode_coefs
Remove ZERO_TOKEN from the ANS tokenset
Drop costing ANS tokens from derived probabilities
Unfork ANS pack_mb_tokens
Fix a bug in the C implementation of the ihalfright32
transform, in the case that its input and output buffers are the same.
This occurs when it is called by av1_iht32x16_512_add_c.
Change-Id: I61c652e2662178520c0639a2879ae128a9c7ec3f
- av1_fht32x32 AVX2 function level time reduction ~89% compared to C.
- av1_fht32x32_avx2() on DCT_DCT improves 42.62% over aom_fdct32x32_avx2()
But function replacement must go with the corresponding inverse txfm.
- No obvious user level time reduction due to 32x32 TX_TYPE selection.
- Zero high 128b YMM to avoid AVX-SSE transition penalties
(fix 16x16 case).
- Added 32x32 AVX2 unit tests to verify bitexact.
- AVX2 optimization summary:
On CPU i7-6700, based on 16x16/32x32 fwd txfm optimization results:
C to AVX2: function level time reduction, ~86-89%.
SSE2 to AVX2: function level time reduction, ~51%.
Change-Id: Idd0cd8bf066a61c7117140ef15ab6c1f8eb4b036
This can be re-added after aligning AOM's ANS with nextgenv2's ANS.
This partially reverts commit 3829cd2f2f9904572019aa047d068baeee843767.
Change-Id: I78afc587f1abfe33ffcd53b3262910cfae135534
This mimics what's currently done in aom/master. This can be re-added
after aligning AOM's ANS with nextgenv2's ANS.
Change-Id: I3ae62181dd4803694204a234c717a86a15ca8a40
