tree 95646875d00e0ed7d7818e9d9c3aec32cdbb5af1
parent a7f472b0eabf3dfd800578a9a1c76d6cb044b555
author chiyotsai <chiyotsai@google.com> 1662581814 -0700
committer Chi Yo Tsai <chiyotsai@google.com> 1663782547 +0000

AVX2: Port av1_convolve_(x|y|2d)_sr from SVT-AV1

This commit also adds the following optimizations:
 4 tap filter: width 8, 16, 32, 64, 128
 2 tap filter: width 2, 4

Fixed a typo: convolve_8tap_unapck_avx2 -> convolve_8tap_unpack_avx2

The new convolve functions are 30% to 100% faster than the old code,
but the old code is kept in place to support encoder-side only 12-tap
filtering.  As a result, there's some redundancy that can be removed
in later commits.

RTC Performance:
| SPD_SET | TESTSET  | AVG_PSNR | OVR_PSNR |  SSIM   | ENC_T |
|---------|----------|----------|----------|---------|-------|
|    7    |   rtc    | +0.000%  | +0.000%  | +0.000% | -1.6% |
|    7    | rtc_derf | +0.000%  | +0.000%  | +0.000% | -1.5% |
|---------|----------|----------|----------|---------|-------|
|    8    |   rtc    | +0.000%  | +0.000%  | +0.000% | -1.1% |
|    8    | rtc_derf | +0.000%  | +0.000%  | +0.000% | -0.8% |
|---------|----------|----------|----------|---------|-------|
|    9    |   rtc    | +0.000%  | +0.000%  | +0.000% | -1.1% |
|    9    | rtc_derf | +0.000%  | +0.000%  | +0.000% | -0.8% |
|---------|----------|----------|----------|---------|-------|
|   10    |   rtc    | +0.000%  | +0.000%  | +0.000% | -0.9% |
|   10    | rtc_derf | +0.000%  | +0.000%  | +0.000% | -0.7% |

VOD Performance:
| SPD_SET | TESTSET  | AVG_PSNR | OVR_PSNR |  SSIM   | ENC_T |
|---------|----------|----------|----------|---------|-------|
|    1    | lowres2  | +0.000%  | +0.000%  | +0.000% | -0.4% |
|    1    | midres2  | +0.000%  | +0.000%  | +0.000% | -0.4% |
|---------|----------|----------|----------|---------|-------|
|    6    |  hdres2  | +0.000%  | +0.000%  | +0.000% | -0.5% |
|    6    | lowres2  | +0.000%  | +0.000%  | +0.000% | -0.5% |
|    6    | midres2  | +0.000%  | +0.000%  | +0.000% | -0.5% |

Change-Id: Ia75a3f286de712ec8152404fde847aaea4bc1f47