[x265] [PATCH] asm: improve old avx2 code for sad[64x48]

[sumalatha at multicorewareinc.com]

# HG changeset patch
# User Sumalatha Polureddy
# Date 1428315710 -19800
#      Mon Apr 06 15:51:50 2015 +0530
# Node ID 6fd0c740b7c417b7b54627ead1edd288f33718f4
# Parent  ee87bb231905d5d8462e763d7e25c7957f52af2a
asm: improve old avx2 code for sad[64x48]

old:
sad[64x48]  16.79x   1504.65         25267.23

new:
sad[64x48]  20.18x   1260.99         25451.33

diff -r ee87bb231905 -r 6fd0c740b7c4 source/common/x86/sad-a.asm
--- a/source/common/x86/sad-a.asm	Mon Apr 06 14:53:36 2015 +0530
+++ b/source/common/x86/sad-a.asm	Mon Apr 06 15:51:50 2015 +0530
@@ -4306,10 +4306,12 @@
     RET
 
 INIT_YMM avx2
-cglobal pixel_sad_64x48, 4,5,6
+cglobal pixel_sad_64x48, 4,7,6
     xorps           m0, m0
     xorps           m5, m5
-    mov             r4d, 24
+    mov             r4d, 12
+    lea             r5, [r1 * 3]
+    lea             r6, [r3 * 3]
 .loop
     movu           m1, [r0]               ; first 32 of row 0 of pix0
     movu           m2, [r2]               ; first 32 of row 0 of pix1
@@ -4331,8 +4333,28 @@
     paddd          m0, m1
     paddd          m5, m3
 
-    lea     r2,     [r2 + 2 * r3]
-    lea     r0,     [r0 + 2 * r1]
+    movu           m1, [r0 + 2 * r1]      ; first 32 of row 0 of pix0
+    movu           m2, [r2 + 2 * r3]      ; first 32 of row 0 of pix1
+    movu           m3, [r0 + 2 * r1 + 32] ; second 32 of row 0 of pix0
+    movu           m4, [r2 + 2 * r3 + 32] ; second 32 of row 0 of pix1
+
+    psadbw         m1, m2
+    psadbw         m3, m4
+    paddd          m0, m1
+    paddd          m5, m3
+
+    movu           m1, [r0 + r5]          ; first 32 of row 1 of pix0
+    movu           m2, [r2 + r6]          ; first 32 of row 1 of pix1
+    movu           m3, [r0 + 32 + r5]     ; second 32 of row 1 of pix0
+    movu           m4, [r2 + 32 + r6]     ; second 32 of row 1 of pix1
+
+    psadbw         m1, m2
+    psadbw         m3, m4
+    paddd          m0, m1
+    paddd          m5, m3
+
+    lea     r2,     [r2 + 4 * r3]
+    lea     r0,     [r0 + 4 * r1]
 
     dec         r4d
     jnz         .loop