[x265] [PATCH 009 of 307] x86: AVX512 pixel_avg_weight 8x32, 16x4, 16x12, 16x32, 16x64

[mythreyi at multicorewareinc.com]

# HG changeset patch
# User Vignesh Vijayakumar
# Date 1499231221 -19800
#      Wed Jul 05 10:37:01 2017 +0530
# Node ID 84757e275a5427f0875fc4fd651bd1c48d534e8f
# Parent  69b61721fa2ffdf1a0f6609a299c4e0104b48628
x86: AVX512 pixel_avg_weight 8x32, 16x4, 16x12, 16x32, 16x64

diff -r 69b61721fa2f -r 84757e275a54 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Jun 26 16:31:52 2017 +0530
+++ b/source/common/x86/asm-primitives.cpp	Wed Jul 05 10:37:01 2017 +0530
@@ -3757,8 +3757,13 @@
 
         p.cu[BLOCK_8x8].var = PFX(pixel_var_8x8_avx512);
         p.cu[BLOCK_16x16].var = PFX(pixel_var_16x16_avx512);
+        p.pu[LUMA_16x64].pixelavg_pp = PFX(pixel_avg_16x64_avx512);
+        p.pu[LUMA_16x32].pixelavg_pp = PFX(pixel_avg_16x32_avx512);
         p.pu[LUMA_16x16].pixelavg_pp = PFX(pixel_avg_16x16_avx512);
+        p.pu[LUMA_16x12].pixelavg_pp = PFX(pixel_avg_16x12_avx512);
         p.pu[LUMA_16x8].pixelavg_pp = PFX(pixel_avg_16x8_avx512);
+        p.pu[LUMA_16x4].pixelavg_pp = PFX(pixel_avg_16x4_avx512);
+        p.pu[LUMA_8x32].pixelavg_pp = PFX(pixel_avg_8x32_avx512);
         p.pu[LUMA_8x16].pixelavg_pp = PFX(pixel_avg_8x16_avx512);
         p.pu[LUMA_8x8].pixelavg_pp = PFX(pixel_avg_8x8_avx512);
         p.pu[LUMA_8x4].pixelavg_pp = PFX(pixel_avg_8x4_avx512);
diff -r 69b61721fa2f -r 84757e275a54 source/common/x86/mc-a.asm
--- a/source/common/x86/mc-a.asm	Mon Jun 26 16:31:52 2017 +0530
+++ b/source/common/x86/mc-a.asm	Wed Jul 05 10:37:01 2017 +0530
@@ -4413,8 +4413,13 @@
 AVGH 16, 4
 
 INIT_XMM avx512
+AVGH 16, 64
+AVGH 16, 32
 AVGH 16, 16
+AVGH 16, 12
 AVGH 16,  8
+AVGH 16,  4
+AVGH  8, 32
 AVGH  8, 16
 AVGH  8,  8
 AVGH  8,  4