[x265] [PATCH] asm: avx2 code for satd for all chroma i422 for 10 bpp

[sumalatha at multicorewareinc.com]

# HG changeset patch
# User Sumalatha Polureddy
# Date 1432534660 -19800
#      Mon May 25 11:47:40 2015 +0530
# Node ID 9934ac4ab4572583b1c47ea3e07f3e0968c4a903
# Parent  a7bf7a150a705489cb63d0454c59ec599bad8c93
asm: avx2 code for  satd for all chroma i422 for 10 bpp

avx2:
[i422] satd[  4x8]  5.32x    290.14          1542.77
[i422] satd[  4x4]  4.32x    179.90          777.19
[i422] satd[ 8x16]  4.44x    990.86          4402.91
[i422] satd[  8x8]  5.03x    480.01          2415.82
[i422] satd[ 4x16]  5.20x    547.30          2845.97
[i422] satd[  8x4]  4.00x    310.37          1240.04
[i422] satd[ 8x12]  5.16x    820.36          4229.93
[i422] satd[16x32]  9.39x    1994.98         18729.04
[i422] satd[16x16]  9.16x    1021.66         9358.59
[i422] satd[ 8x32]  4.53x    2024.80         9175.30
[i422] satd[ 16x8]  8.93x    530.12          4735.74
[i422] satd[16x24]  4.89x    2921.85         14279.47
[i422] satd[ 4x32]  4.67x    1213.40         5662.05
[i422] satd[12x32]  4.82x    3632.33         17507.59
[i422] satd[32x64]  9.62x    7672.94         73822.49
[i422] satd[32x32]  10.75x   3891.23         41826.38
[i422] satd[16x64]  9.66x    3821.15         36931.40
[i422] satd[32x16]  8.97x    1932.95         17338.38
[i422] satd[32x48]  4.72x    11329.97        53510.81
[i422] satd[ 8x64]  4.45x    3972.60         17696.34
[i422] satd[24x64]  5.06x    11323.39        57300.73

diff -r a7bf7a150a70 -r 9934ac4ab457 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Fri May 22 14:29:35 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon May 25 11:47:40 2015 +0530
@@ -1070,6 +1070,13 @@
         p.chroma[X265_CSP_I422].pu[CHROMA_422_8x12].satd = x265_pixel_satd_8x12_avx;
         p.chroma[X265_CSP_I422].pu[CHROMA_422_12x32].satd = x265_pixel_satd_12x32_avx;
         p.chroma[X265_CSP_I422].pu[CHROMA_422_4x32].satd = x265_pixel_satd_4x32_avx;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_4x8].satd = x265_pixel_satd_4x8_avx;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_8x16].satd = x265_pixel_satd_8x16_avx;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_4x4].satd = x265_pixel_satd_4x4_avx;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_8x8].satd = x265_pixel_satd_8x8_avx;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_4x16].satd = x265_pixel_satd_4x16_avx;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_8x32].satd = x265_pixel_satd_8x32_avx;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_8x4].satd = x265_pixel_satd_8x4_avx;
 
         ALL_LUMA_PU(satd, pixel_satd, avx);
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x8].satd = x265_pixel_satd_8x8_avx;
@@ -1273,6 +1280,13 @@
         p.chroma[X265_CSP_I420].pu[CHROMA_420_32x16].satd = x265_pixel_satd_32x16_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_32x24].satd = x265_pixel_satd_32x24_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_32x8].satd = x265_pixel_satd_32x8_avx2;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x32].satd = x265_pixel_satd_16x32_avx2;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_32x64].satd = x265_pixel_satd_32x64_avx2;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x16].satd = x265_pixel_satd_16x16_avx2;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_32x32].satd = x265_pixel_satd_32x32_avx2;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x64].satd = x265_pixel_satd_16x64_avx2;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x8].satd = x265_pixel_satd_16x8_avx2;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_32x16].satd = x265_pixel_satd_32x16_avx2;
 
         p.cu[BLOCK_16x16].ssd_s = x265_pixel_ssd_s_16_avx2;
         p.cu[BLOCK_32x32].ssd_s = x265_pixel_ssd_s_32_avx2;