This doesn't have a big performance impact, and it's rather simpler
just having one version of everything.

Change-Id: I5fa5e7640a63d0ccb0c371f266c6eee99d9520f9

Change-Id: Ifac3a3bf620061865b82b986d6b16bcabd96a187

This fixes some Valgrind errors caused by reads from x_by_xplus1 that
used tainted data as an address (see the comments in selfguided_sse4.c
for what's going on).

It also rewrites the algorithm to use an integral image approach
instead of the handwritten filters that the code was using. The end
result is roughly the same efficiency (I think that there's one more
memory load per group of pixels, but this seems not to be measurable)
and I've done some performance optimisation with perf too. Several
32-bit multiplications have been replaced by madd instructions which
do 16-bit multiplications and add adjacent lanes. This is equivalent
to a 32-bit multiplication when the 32-bit lanes contain numbers below
2^15, but runs significantly faster.

Change-Id: I3d0f3043c7861707a56e2fd1849574dc73897d6c

BUG=b/69488541

Change-Id: I2113bba4589f61a09d0dd07c64a522f4d0ae304b
(cherry picked from commit cccda0db727c2282375b174104294b40911d1447)

BUG=b/69445855,b/69441422

Change-Id: Iaf5aba78dc39f01c87fb726611e674d34af6bffe
(cherry picked from commit 75ff22f309de2e25477d336a6a8e9e58d3bb2272)

under visual studio c4334:
result of 32-bit shift implicitly converted to 64 bits (was 64-bit shift
intended?)

Change-Id: If06793116ddfbe3265a17a17a2bcaa6ee8cf9e2d
(cherry picked from commit 535ecf6b31fe97f704f6725989cffad88ad960d8)

BUG=b/69238080,b/69288165

Change-Id: Ia761d4b77049a55bd8040b5ed76063b2fac750ee
(cherry picked from commit c9762668a3f25c2dfe31c426871450fbfd44b9e0)

BUG=b/69205191

Change-Id: I79e404dc2cd6db06e71a64338b74eb4b575ba431
(cherry picked from commit 85f2a5ae4c15de5dd530766eb3933b9de976d9cf)

BUG=69073461

Change-Id: Ib28b41adfa2738681357903a81a89bcab01c87b3
(cherry picked from commit 08b26a8a257e54210d8bbdba799980bc291f368e)

Change-Id: Ief1bedd68de55c29de15f56d805e242d932ff359

Change-Id: Ifb295cbcde5474d33c4eca008d89c9dda68d327e

Reason for revert: nightly test failures due to imcompatibility with lv-map.

BUG=aomedia:1052
BUG=aomedia:1058
BUG=aomedia:1059

Change-Id: Ifbe9cf4542b1b023b8b9e0a2f780e0075914bee0

To silence asan failures in fuzzing tests.

BUG=:68825590,68825594,68825599

Change-Id: Ib2c713dc19af223da5e5fc5cec4652d71856f830
(cherry picked from commit e43ea91055133baaf3b691170a097a456c032e23)

Unit tests for:
aom_jnt_comp_avg_pred_c
aom_jnt_sadmxn_avg_c
aom_sadmxh_sse2
aom_jnt_comp_avg_pred_ssse3
aom_jnt_sadmxn_avg_ssse3

Change-Id: I22463f143fd6513fd68f4bb82203e1b4fe3bbab7

Previous assumption on reduced_tx_set_used=0 led to many assertion
failures and prevented signalling reduced_tx_set_used equal to 1.

BUG=aomedia:1053

Change-Id: If9a9dff8d01ba3ec942e06559c153f06d34555f9

Add ssse3 implementations for the sad_avg c function at low bit-depth.
With this, aom_jnt_sad c functions can all have simd implementations.
This CL follows existing MACRO definitions for multiple combinations
of block sizes.

Change-Id: I882343684026525f5589a239337cfac2dd411e11

The code was assuming that an mi was always 4 samples wide and
high. For chroma planes with subsampling, this is wrong and the size
and position of the sb in the plane was over-estimated by a factor of
two. This meant that we sent all the coefficients in the top-left hand
quarter of the tile. Since the encoder and decoder made the same
mistake, this worked fine, but it's clearly not what we're supposed to
do!

Change-Id: I0da8ada1d76639ad476ad84491658bc25ef3a43f

Both c function and ssse3 have passed unit tests.

Change-Id: I48cff97ebf2735b43256b83f3b41ce7ccdf27393

Change function names and add SIMD implementation for two c functions:
(1) var_filter_block2d_bil_first_pass
(2) var_filter_block2d_bil_second_pass

This CL allows aom_jnt_sub_pixel_avg_variance now in SIMD.

Change-Id: Ib41ef13d62ae91a0ca481bcebb24568dcd4722c4

Change-Id: I91dd5d3351d5dcc70ffcdb883d1e7cbd054d1a27

In order to address hardware concerns regarding luma averaging,
this patch caps the maximum length of the sides of the luma
averaging area to 32.

This proposal was accepted by the hardware working group on
November 20th 2017.

Regression on Subset1:
  PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
0.0078 |  0.0572 | -0.0705 |  -0.0272 | -0.0202 | -0.0391 |     0.0375

Change-Id: I875a6f2114df4d857ed66c4690ee08da2df426e4

BUG=aomedia:1026

Change-Id: If1b62c728bb77ca733aaa5b9100608f6b9d33db1

Change-Id: I9854d5ec193dadaa455c209b1482ead895982cb7

- Reuse scan order
- Truncate to max eob of 32*32
- Quantization and entropy coding same as done for TX_32X32
- Reuse quantization matrices of TX_32X32

Compression performance is roughly neutral:
https://arewecompressedyet.com/?job=tx64x64_oldscans%402017-11-06T03%3A11%3A53.868Z&job=tx64x64_reusescans%402017-11-06T03%3A12%3A55.738Z

Change-Id: Ie9182c1c69a42a3c1ab4fc980abbd6000c64f179

Change-Id: I6c08f4aaba10f89fd17f3100dc5f14bb2087e76b

Very little impact on comression quality.

Change-Id: I3b0fbebe7c6e53f299a764aba49a22b931bb8bd0

Provisionally adopted on 11/17.

Also, some related tweaks to fix build errors.

Change-Id: I7d5592450e9284d489b46adc274cd0cfccd04b3c

This bug was causing a 14% drop in performance on the lowres set,
this fix reduces the drop to 0.2%.

Change-Id: I166efc9cc5735c3366a0989e924d2c9fae9e706b

Change-Id: Ia07e909c89eda3742682531dca068ecf63a6281e

A few bits of data were not zeroed because aom_wb_overwrite_literal
was being used instead of aom_wb_write_literal.

BUG=aomedia:1041

Change-Id: If9196551e99b6e5eeaefc3fe022088088f1dcd51

The fixes in rdopt.c improves the coding performance of
4:1 transforms significantly.

Change-Id: I0e8db93e3f6d9bf0b2de01f2ce83c305d78d2262

Change-Id: Ia06a6f2ac6473f10f09df3c0cc9cd45c2b781146

Change-Id: If6a6aad09d5773f8858d7c163c15e9fcefccc9cb

BUG=aomedia:1031

Change-Id: I44007d418dba65cda9bd5fe44f8bfa66c080c7bc

Change-Id: I2960f6d6d6bbdf747a32ed1cdd2b5a3c4e51ba8b

Abstract the inter block transform coefficient writing unit.

Change-Id: I8e7a83d2d92941258f7250fee4c96f5ddfc4572e

Previously lv_map bins (cdf2) were treated specially in that the probability
was rounded to 7bit before passed to the arithmetic engine.

Change-Id: I75d8437a6185e529e42e9867e3df18384447f2fd

The horizontal delay is specified in pixels to work for SB64 and SB128.
The wavefront gradient is changed so the above block is available.

Change-Id: I24cc426bded6904925930f6d431f5737070f9e17

BUG=aomedia:1036

Change-Id: I8444658957c6e19f0525e21c3918000a4097a729

It seemed the bug got fixed in current TOT, and the unit tests failed
before passed now. Re-enabled LOOPFILTERING_ACROSS_TILES.

BUG=aomedia:1023

Change-Id: I11efe82d6c9232b702409e69750490fd3456c320