OpenVisualCloud / SVT-AV1 / 2532
0%

DEFAULT BRANCH: master
Ran
Jobs 2
Files 386
Run time 5561min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
2532

push

travis-ci-com

mafonsoN 
Update convolve avx2 functions (#530)

* Cherry pick new convolve avx2 code from branch m0-test

Including commits f02be5f2, 85a12515, 66a2e90a, bb65697a, 1e2d2e1c and f88f737d.

Add 4-tap add 2-tap convolve AVX2 code.

4-tap convolve AVX2 code is ported from aom.

In 2-tap case (bilinear), the speed comparison vs using 8-tap filter is

Blk Size | SR  X    Y    2D  | JNT  X    Y    2D
  4x  4  |    1.22 1.54 1.32 |     1.06 1.60 1.14
  4x  8  |    1.31 1.40 1.29 |     1.04 1.43 1.12
  8x  4  |    1.24 1.53 1.34 |     1.04 1.60 1.16
  8x  8  |    1.39 1.38 1.29 |     1.04 1.38 1.19
  8x 16  |    1.33 1.27 1.41 |     1.09 1.26 1.30
 16x  8  |    1.29 1.35 1.44 |     1.14 1.37 1.28
 16x 16  |    1.33 1.25 1.54 |     1.18 1.25 1.39
 16x 32  |    1.29 1.17 1.61 |     1.18 1.27 1.42
 32x 16  |    1.45 1.26 1.64 |     1.25 1.23 1.41
 32x 32  |    1.53 1.30 1.64 |     1.23 1.21 1.42
 32x 64  |    1.60 1.16 1.67 |     1.23 1.17 1.44
 64x 32  |    1.62 1.22 1.67 |     1.25 1.18 1.44
 64x 64  |    1.63 1.17 1.67 |     1.22 1.16 1.46
 64x128  |    1.50 1.16 1.66 |     1.26 1.15 1.41
128x 64  |    1.61 1.18 1.69 |     1.25 1.14 1.45
128x128  |    1.61 1.16 1.68 |     1.25 1.13 1.40
  4x 16  |    1.32 1.26 1.39 |     1.10 1.26 1.24
 16x  4  |    1.60 1.48 1.46 |     1.10 1.50 1.24
  8x 32  |    1.32 1.18 1.49 |     1.14 1.26 1.33
 32x  8  |    1.52 1.37 1.53 |     1.19 1.35 1.35
 16x 64  |    1.41 1.14 1.62 |     1.22 1.26 1.44
 64x 16  |    1.70 1.27 1.68 |     1.32 1.24 1.46

Update 2-tap av1_convolve_x_sr_avx2()

Speed gains:
 4x  4: 1.25
 4x  8: 1.40
 8x  4: 1.20
 8x  8: 1.18
 8x 16: 0.98
16x  8: 1.82
16x 16: 1.95
16x 32: 2.20
32x 16: 2.24
32x 32: 2.73
32x 64: 2.61
64x 32: 2.33
64x 64: 2.43
64x128: 2.63
28x 64: 2.53
28x128: 2.31
 4x 16: 1.41
16x  4: 1.76
 8x 32: 0.97
32x  8: 1.92
16x 64: 2.24
64x 16: 2.13

* Update eb_av1_convolve_x_sr_avx2()

Separate all widths.
Add tap 6 case.
Use average in tap... (continued)

32146 of 86405 relevant lines covered (37.2%)

2564734.16 hits per line

Jobs
ID Job ID Ran Files Coverage
10 2532.10 (COVERALLS_PARALLEL=true build_type=debug CMAKE_EFLAGS="-DCOVERAGE=ON") 0
33.52
 Travis Job 2532.10
11 2532.11 (COVERALLS_PARALLEL=true build_type=debug CMAKE_EFLAGS="-DCOVERAGE=ON") 0
28.12
 Travis Job 2532.11
Source Files on build 2532
Detailed source file information is not available for this build.