[x265] [PATCH] asm: fix sad[48x64] avx2 code for main12

[rajesh at multicorewareinc.com]

# HG changeset patch
# User Rajesh Paulraj<rajesh at multicorewareinc.com>
# Date 1443515696 -19800
#      Tue Sep 29 14:04:56 2015 +0530
# Node ID 8411b3d5b07c65bfcf2813133bb8daa665526ac4
# Parent  d7cde5dbec838f53e87faceac989d6cd987bfc72
asm: fix sad[48x64] avx2 code for main12

diff -r d7cde5dbec83 -r 8411b3d5b07c source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Tue Sep 29 14:00:27 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Tue Sep 29 14:04:56 2015 +0530
@@ -1630,8 +1630,8 @@
         p.pu[LUMA_32x24].sad = PFX(pixel_sad_32x24_avx2);
         p.pu[LUMA_32x32].sad = PFX(pixel_sad_32x32_avx2);
         p.pu[LUMA_32x64].sad = PFX(pixel_sad_32x64_avx2);
+        p.pu[LUMA_48x64].sad = PFX(pixel_sad_48x64_avx2);
 #if X265_DEPTH <= 10
-        p.pu[LUMA_48x64].sad = PFX(pixel_sad_48x64_avx2);
         p.pu[LUMA_64x16].sad = PFX(pixel_sad_64x16_avx2);
         p.pu[LUMA_64x32].sad = PFX(pixel_sad_64x32_avx2);
         p.pu[LUMA_64x48].sad = PFX(pixel_sad_64x48_avx2);
diff -r d7cde5dbec83 -r 8411b3d5b07c source/common/x86/sad16-a.asm
--- a/source/common/x86/sad16-a.asm	Tue Sep 29 14:00:27 2015 +0530
+++ b/source/common/x86/sad16-a.asm	Tue Sep 29 14:04:56 2015 +0530
@@ -769,9 +769,8 @@
 INIT_YMM avx2
 cglobal pixel_sad_48x64, 4, 5, 7
     pxor    m0, m0
-    pxor    m5, m5
-    pxor    m6, m6
     mov     r4d, 64/2
+    mova    m6, [pw_1]
     add     r3d, r3d
     add     r1d, r1d
 .loop:
@@ -784,9 +783,8 @@
     pabsw   m1, m1
     pabsw   m2, m2
     pabsw   m3, m3
-    paddw   m0, m1
-    paddw   m5, m2
-    paddw   m6, m3
+    paddw   m1, m2
+    paddw   m5, m3, m1
 
     movu    m1, [r2 + r3 + 0 * mmsize]
     movu    m2, [r2 + r3 + 1 * mmsize]
@@ -797,22 +795,20 @@
     pabsw   m1, m1
     pabsw   m2, m2
     pabsw   m3, m3
-    paddw   m0, m1
-    paddw   m5, m2
-    paddw   m6, m3
+    paddw   m1, m2
+    paddw   m3, m1
 
+    pmaddwd m5, m6
+    paddd   m0, m5
+    pmaddwd m3, m6
+    paddd   m0, m3
     lea     r0, [r0 + 2 * r1]
     lea     r2, [r2 + 2 * r3]
 
     dec     r4d
     jg      .loop
 
-    HADDUWD m0, m1
-    HADDUWD m5, m1
-    HADDUWD m6, m1
-    paddd   m0, m5
-    paddd   m0, m6
-    HADDD   m0, m1
+    HADDD   m0, m3
     movd    eax, xm0
     RET