[x265] [PATCH] asm: avx2 code for pixel_avg[16xN, 32xN] for 10 bpp

[sumalatha at multicorewareinc.com]

# HG changeset patch
# User Sumalatha Polureddy<sumalatha at multicorewareinc.com>
# Date 1434524988 -19800
#      Wed Jun 17 12:39:48 2015 +0530
# Node ID 8774c63a4047fa5e54679237aeb88518740efa8d
# Parent  be0ed447922cc81e809d296e75424bb71822aea7
asm: avx2 code for pixel_avg[16xN, 32xN] for 10 bpp

sse2:
avg_pp[32x32]      1662.52         3823.17
avg_pp[32x16]      874.92          2046.83
avg_pp[ 32x8]      479.38          1126.35
avg_pp[32x24]      1245.13         3170.44
avg_pp[32x64]      3187.47         7329.66
avx2:
avg_pp[32x32]      1584.80         3699.90
avg_pp[32x16]      820.54          2264.90
avg_pp[ 32x8]      450.56          975.42
avg_pp[32x24]      1104.99         2727.65
avg_pp[32x64]      3046.60         7310.93

diff -r be0ed447922c -r 8774c63a4047 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Tue Jun 16 11:15:03 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Wed Jun 17 12:39:48 2015 +0530
@@ -1568,6 +1568,18 @@
         p.pu[LUMA_64x48].sad_x4 = PFX(pixel_sad_x4_64x48_avx2);
         p.pu[LUMA_64x64].sad_x4 = PFX(pixel_sad_x4_64x64_avx2);
 
+        p.pu[LUMA_16x4].pixelavg_pp = PFX(pixel_avg_16x4_avx2);
+        p.pu[LUMA_16x8].pixelavg_pp = PFX(pixel_avg_16x8_avx2);
+        p.pu[LUMA_16x12].pixelavg_pp = PFX(pixel_avg_16x12_avx2);
+        p.pu[LUMA_16x16].pixelavg_pp = PFX(pixel_avg_16x16_avx2);
+        p.pu[LUMA_16x32].pixelavg_pp = PFX(pixel_avg_16x32_avx2);
+        p.pu[LUMA_16x64].pixelavg_pp = PFX(pixel_avg_16x64_avx2);
+        p.pu[LUMA_32x8].pixelavg_pp = PFX(pixel_avg_32x8_avx2);
+        p.pu[LUMA_32x16].pixelavg_pp = PFX(pixel_avg_32x16_avx2);
+        p.pu[LUMA_32x24].pixelavg_pp = PFX(pixel_avg_32x24_avx2);
+        p.pu[LUMA_32x32].pixelavg_pp = PFX(pixel_avg_32x32_avx2);
+        p.pu[LUMA_32x64].pixelavg_pp = PFX(pixel_avg_32x64_avx2);
+
         p.pu[LUMA_16x4].convert_p2s = PFX(filterPixelToShort_16x4_avx2);
         p.pu[LUMA_16x8].convert_p2s = PFX(filterPixelToShort_16x8_avx2);
         p.pu[LUMA_16x12].convert_p2s = PFX(filterPixelToShort_16x12_avx2);
diff -r be0ed447922c -r 8774c63a4047 source/common/x86/mc-a.asm
--- a/source/common/x86/mc-a.asm	Tue Jun 16 11:15:03 2015 +0530
+++ b/source/common/x86/mc-a.asm	Wed Jun 17 12:39:48 2015 +0530
@@ -3528,6 +3528,49 @@
 AVG_WEIGHT 32, 8
 AVG_WEIGHT 48, 8
 AVG_WEIGHT 64, 8
+
+%define BIWEIGHT BIWEIGHT_SSSE3
+%define BIWEIGHT_START BIWEIGHT_START_SSSE3
+INIT_YMM avx2
+cglobal pixel_avg_weight_w16
+    BIWEIGHT_START
+    AVG_START 5
+.height_loop:
+    movu     m0, [t2]
+    movu     m1, [t4]
+    SBUTTERFLY wd, 0, 1, 2
+    pmaddwd   m0, m3
+    pmaddwd   m1, m3
+    pmulld  m0, m4
+    pmulld  m1, m4
+    packusdw  m0, m1
+    mova    [t0], m0
+    AVG_END
+
+cglobal pixel_avg_weight_w32
+    BIWEIGHT_START
+    AVG_START 5
+.height_loop:
+    movu     m0, [t2]
+    movu     m1, [t4]
+    SBUTTERFLY wd, 0, 1, 2
+    pmaddwd   m0, m3
+    pmaddwd   m1, m3
+    pmulld  m0, m4
+    pmulld  m1, m4
+    packuswb  m0, m1
+    mova    [t0], m0
+    movu     m0, [t2 + 32]
+    movu     m1, [t4 + 32]
+    SBUTTERFLY wd, 0, 1, 2
+    pmaddwd   m0, m3
+    pmaddwd   m1, m3
+    pmulld  m0, m4
+    pmulld  m1, m4
+    packusdw  m0, m1
+    mova    [t0 + 32], m0
+    AVG_END
+
 %else ;!HIGH_BIT_DEPTH
 INIT_XMM sse2
 AVG_WEIGHT 8,  7
@@ -4096,6 +4139,22 @@
 AVG_FUNC 12, movdqu, movdqa, movq
 AVGH 12, 16
 
+INIT_XMM avx2
+AVG_FUNC 16, movdqu, movdqa
+AVGH 16, 64
+AVGH 16, 32
+AVGH 16, 16
+AVGH 16, 12
+AVGH 16, 8
+AVGH 16, 4
+
+AVG_FUNC 32, movdqu, movdqa
+AVGH 32, 64
+AVGH 32, 32
+AVGH 32, 24
+AVGH 32, 16
+AVGH 32, 8
+
 %else ;!HIGH_BIT_DEPTH
 
 INIT_MMX mmx2