NEON ports of svt_av1_inv_txfm2d_add_16x16_c and svt_av1_inv_txfm2d_add_32x32_c and svt_av1_inv_txfm2d_add_64x64_c
Description
Issue
Adds NEON port of the svt_av1_inv_txfm2d_add_16x16_c and svt_av1_inv_txfm2d_add_32x32_c and svt_av1_inv_txfm2d_add_64x64_c functions based on SSE4.1.
svt_av1_inv_txfm2d_add_16x16_neon based on this commit
svt_av1_inv_txfm2d_add_32x32_neon based on this commit
svt_av1_inv_txfm2d_add_64x64_neon based on this commit
The system under test is c7g.4xlarge AWS graviton instance.
The combined effect of both functions were obtained as follows:
wget wget http://ultravideo.fi/video/Bosphorus_3840x2160_120fps_420_8bit_YUV_Y4M.7z
7z x Bosphorus_3840x2160_120fps_420_8bit_YUV_Y4M.7z
SvtAv1EncApp -i Bosphorus_3840x2160.y4m --crf [C] --preset [P]After
| filename | preset | crf | User time (seconds) | System time (seconds) | Percent of CPU this job got | Maximum resident set size (kbytes) | Average Speed | Loaded runtime |
|---|---|---|---|---|---|---|---|---|
| Bosphorus_3840x2160.y4m' | 4 | 30 | 726.67 | 5.43 | 1146.00% | 5279508 | 2.388 | 146.81 |
| Bosphorus_3840x2160.y4m' | 6 | 30 | 250.52 | 2.45 | 1049.00% | 5063080 | 6.507 | 51.46 |
| Bosphorus_3840x2160.y4m' | 8 | 30 | 116.82 | 1.64 | 1023.00% | 4278196 | 13.999 | 24.32 |
| Bosphorus_3840x2160.y4m' | 10 | 30 | 52.41 | 1.57 | 1205.00% | 3602100 | 39.897 | 11.31 |
| Bosphorus_3840x2160.y4m' | 12 | 30 | 33.97 | 1.32 | 1078.00% | 3334380 | 57.25 | 7.6 |
| Bosphorus_3840x2160.y4m' | 13 | 30 | 34.05 | 1.24 | 1079.00% | 3332672 | 57.315 | 7.56 |
| Bosphorus_1920x1080.y4m' | 4 | 30 | 349.93 | 1.96 | 1155.00% | 2144340 | 9.997 | 70.15 |
| Bosphorus_1920x1080.y4m' | 6 | 30 | 125.86 | 1.17 | 1119.00% | 2102704 | 27.535 | 25.62 |
| Bosphorus_1920x1080.y4m' | 8 | 30 | 62.68 | 0.94 | 962.00% | 1924628 | 48.308 | 13.04 |
| Bosphorus_1920x1080.y4m' | 10 | 30 | 26.99 | 0.71 | 1216.00% | 1146116 | 148.407 | 5.69 |
| Bosphorus_1920x1080.y4m' | 12 | 30 | 12.74 | 0.7 | 1019.00% | 989744 | 273.317 | 2.84 |
| Bosphorus_1920x1080.y4m' | 13 | 30 | 10.40 | 0.59 | 982.00% | 977680 | 329.371 | 2.43 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 4 | 30 | 482.32 | 0.86 | 734.00% | 1387080 | 2.331 | 98.96 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 6 | 30 | 482.04 | 0.84 | 734.00% | 1414476 | 2.333 | 97.87 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 8 | 30 | 482.46 | 0.74 | 735.00% | 1396044 | 2.334 | 98.61 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 10 | 30 | 125.80 | 0.73 | 575.00% | 1169184 | 6.995 | 26.22 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 12 | 30 | 52.49 | 0.56 | 408.00% | 1137756 | 11.906 | 14.33 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 13 | 30 | 49.30 | 0.57 | 349.00% | 1162180 | 10.817 | 15.01 |
| jellyfish-1080p-hevc-10bit.yuv' | 4 | 30 | 386.96 | 0.74 | 646.00% | 457700 | 7.515 | 77.74 |
| jellyfish-1080p-hevc-10bit.yuv' | 6 | 30 | 386.94 | 0.6 | 647.00% | 471080 | 7.522 | 77.83 |
| jellyfish-1080p-hevc-10bit.yuv' | 8 | 30 | 387.26 | 0.69 | 647.00% | 458920 | 7.515 | 78.31 |
| jellyfish-1080p-hevc-10bit.yuv' | 10 | 30 | 107.94 | 0.84 | 613.00% | 382488 | 25.433 | 22.08 |
| jellyfish-1080p-hevc-10bit.yuv' | 12 | 30 | 50.94 | 0.74 | 377.00% | 363032 | 32.947 | 14.43 |
| jellyfish-1080p-hevc-10bit.yuv' | 13 | 30 | 43.99 | 0.67 | 364.00% | 357908 | 36.883 | 12.75 |
Before
| filename | preset | crf | User time (seconds) | System time (seconds) | Percent of CPU this job got | Maximum resident set size (kbytes) | Average Speed | Loaded runtime |
|---|---|---|---|---|---|---|---|---|
| Bosphorus_3840x2160.y4m' | 4 | 30 | 722.04 | 5.68 | 1145.00% | 5265168 | 2.4 | 146.4 |
| Bosphorus_3840x2160.y4m' | 6 | 30 | 250.26 | 2.75 | 1050.00% | 4999256 | 6.512 | 51.06 |
| Bosphorus_3840x2160.y4m' | 8 | 30 | 116.70 | 1.56 | 1024.00% | 4273956 | 14.02 | 24.48 |
| Bosphorus_3840x2160.y4m' | 10 | 30 | 52.52 | 1.49 | 1206.00% | 3618116 | 39.747 | 11.28 |
| Bosphorus_3840x2160.y4m' | 12 | 30 | 33.78 | 1.49 | 1075.00% | 3324704 | 57.04 | 7.58 |
| Bosphorus_3840x2160.y4m' | 13 | 30 | 33.84 | 1.39 | 1074.00% | 3336404 | 57.224 | 7.57 |
| Bosphorus_1920x1080.y4m' | 4 | 30 | 348.91 | 1.98 | 1160.00% | 2150908 | 10.066 | 70.27 |
| Bosphorus_1920x1080.y4m' | 6 | 30 | 125.61 | 1.32 | 1118.00% | 2077612 | 27.569 | 25.74 |
| Bosphorus_1920x1080.y4m' | 8 | 30 | 62.53 | 1.08 | 961.00% | 1921316 | 48.208 | 13.02 |
| Bosphorus_1920x1080.y4m' | 10 | 30 | 26.85 | 0.8 | 1201.00% | 1132408 | 146.775 | 5.71 |
| Bosphorus_1920x1080.y4m' | 12 | 30 | 12.74 | 0.59 | 1018.00% | 985352 | 276.139 | 2.83 |
| Bosphorus_1920x1080.y4m' | 13 | 30 | 10.32 | 0.65 | 984.00% | 971432 | 331.657 | 2.4 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 4 | 30 | 540.85 | 0.87 | 748.00% | 1414512 | 2.119 | 109.68 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 6 | 30 | 541.04 | 0.74 | 751.00% | 1378556 | 2.125 | 109.74 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 8 | 30 | 541.09 | 0.76 | 751.00% | 1409128 | 2.126 | 109.68 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 10 | 30 | 159.72 | 0.68 | 625.00% | 1141892 | 5.99 | 32.79 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 12 | 30 | 77.25 | 0.44 | 496.00% | 1138108 | 9.846 | 18.05 |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 13 | 30 | 67.44 | 0.77 | 422.00% | 1166720 | 9.535 | 17.4 |
| jellyfish-1080p-hevc-10bit.yuv' | 4 | 30 | 431.06 | 0.57 | 654.00% | 469708 | 6.832 | 87 |
| jellyfish-1080p-hevc-10bit.yuv' | 6 | 30 | 430.56 | 0.66 | 653.00% | 459800 | 6.83 | 87.12 |
| jellyfish-1080p-hevc-10bit.yuv' | 8 | 30 | 430.51 | 0.71 | 654.00% | 470820 | 6.84 | 87.37 |
| jellyfish-1080p-hevc-10bit.yuv' | 10 | 30 | 139.99 | 0.60 | 648.00% | 383564 | 20.805 | 28.66 |
| jellyfish-1080p-hevc-10bit.yuv' | 12 | 30 | 67.20 | 0.44 | 425.00% | 362644 | 28.373 | 17.19 |
| jellyfish-1080p-hevc-10bit.yuv' | 13 | 30 | 55.27 | 0.46 | 418.00% | 358100 | 33.878 | 14.36 |
Speedup
| filename | preset | crf | User time (seconds) | ||
|---|---|---|---|---|---|
| After | Before | Improv. | |||
| Bosphorus_3840x2160.y4m' | 4 | 30 | 726.67 | 722.04 | -0.64% |
| Bosphorus_3840x2160.y4m' | 6 | 30 | 250.52 | 250.26 | -0.10% |
| Bosphorus_3840x2160.y4m' | 8 | 30 | 116.82 | 116.70 | -0.10% |
| Bosphorus_3840x2160.y4m' | 10 | 30 | 52.41 | 52.52 | 0.21% |
| Bosphorus_3840x2160.y4m' | 12 | 30 | 33.97 | 33.78 | -0.56% |
| Bosphorus_3840x2160.y4m' | 13 | 30 | 34.05 | 33.84 | -0.62% |
| Bosphorus_1920x1080.y4m' | 4 | 30 | 349.93 | 348.91 | -0.29% |
| Bosphorus_1920x1080.y4m' | 6 | 30 | 125.86 | 125.61 | -0.20% |
| Bosphorus_1920x1080.y4m' | 8 | 30 | 62.68 | 62.53 | -0.24% |
| Bosphorus_1920x1080.y4m' | 10 | 30 | 26.99 | 26.85 | -0.52% |
| Bosphorus_1920x1080.y4m' | 12 | 30 | 12.74 | 12.74 | 0.00% |
| Bosphorus_1920x1080.y4m' | 13 | 30 | 10.40 | 10.32 | -0.77% |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 4 | 30 | 482.32 | 540.85 | 12.14% |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 6 | 30 | 482.04 | 541.04 | 12.24% |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 8 | 30 | 482.46 | 541.09 | 12.15% |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 10 | 30 | 125.80 | 159.72 | 26.96% |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 12 | 30 | 52.49 | 77.25 | 47.17% |
| jellyfish-400-mbps-4k-uhd-hevc-10bit.yuv' | 13 | 30 | 49.30 | 67.44 | 36.80% |
| jellyfish-1080p-hevc-10bit.yuv' | 4 | 30 | 386.96 | 431.06 | 11.40% |
| jellyfish-1080p-hevc-10bit.yuv' | 6 | 30 | 386.94 | 430.56 | 11.27% |
| jellyfish-1080p-hevc-10bit.yuv' | 8 | 30 | 387.26 | 430.51 | 11.17% |
| jellyfish-1080p-hevc-10bit.yuv' | 10 | 30 | 107.94 | 139.99 | 29.69% |
| jellyfish-1080p-hevc-10bit.yuv' | 12 | 30 | 50.94 | 67.20 | 31.92% |
| jellyfish-1080p-hevc-10bit.yuv' | 13 | 30 | 43.99 | 55.27 | 25.64% |
Author(s)
Rodrigo Causarano ( @rjcausarano) Gerardo Puga ( @glpuga)
Performance impact
-
quality -
memory -
speed -
8 bit -
10 bit -
N/A
Test set
-
obj-1-fast can be found here -
other -
N/A
Merge method
-
Allow the maintainer to squash and merge when PR is ready to create a 1-commit to the master branch. The maintainer will be able to fix typos / combine commit messages to create a more readable 1-commit message or use whatever is stated in the 'Description' section -
I will clean up my commits and the maintainer shall use 'rebase and merge' to the master branch
Edited by Rodrigo Causarano