[x265] [PATCH] asm: modified satd and sad asm functions in 16bpp to avoid overflow

[yuvaraj at multicorewareinc.com]

# HG changeset patch
# User Yuvaraj Venkatesh <yuvaraj at multicorewareinc.com>
# Date 1391592757 -19800
#      Wed Feb 05 15:02:37 2014 +0530
# Node ID b14a8528c478bf3068ed95aeef68c050014785cd
# Parent  1374f1168c5cbb97a893172e37bd9f5c6ed5690c
asm: modified satd and sad asm functions in 16bpp to avoid overflow

diff -r 1374f1168c5c -r b14a8528c478 source/common/x86/pixel-a.asm
--- a/source/common/x86/pixel-a.asm	Wed Feb 05 12:59:57 2014 +0530
+++ b/source/common/x86/pixel-a.asm	Wed Feb 05 15:02:37 2014 +0530
@@ -511,7 +511,7 @@
 %endif
 %endmacro
 
-%macro SATD_4x8_SSE 3
+%macro SATD_4x8_SSE 3-4
 %if HIGH_BIT_DEPTH
     movh    m0, [r0+0*r1]
     movh    m4, [r2+0*r3]
@@ -577,7 +577,11 @@
     DIFFOP 2, 6, 3, 5, 7
 %endif
 %endif ; HIGH_BIT_DEPTH
+%if %0 == 4
+    SATD_8x4_1_SSE %1, 0, 1, 2, 3, 4, 5, 7, %3, %4
+%else
     SATD_8x4_SSE %1, 0, 1, 2, 3, 4, 5, 7, %3
+%endif
 %endmacro
 
 ;-----------------------------------------------------------------------------
@@ -2391,56 +2395,66 @@
     SATD_START_MMX
     mov r6, r0
     mov r7, r2
+    pxor m7, m7
 %if vertical==0
     mova m7, [hmul_4p]
 %endif
-    SATD_4x8_SSE vertical, 0, swap
+    SATD_4x8_SSE vertical, 0, 4, 5
     lea r0, [r0 + r1*2*SIZEOF_PIXEL]
     lea r2, [r2 + r3*2*SIZEOF_PIXEL]
-    SATD_4x8_SSE vertical, 1, add
+    SATD_4x8_SSE vertical, 1, 4, 5
     lea r0, [r6 + 4*SIZEOF_PIXEL]
     lea r2, [r7 + 4*SIZEOF_PIXEL]
-    SATD_4x8_SSE vertical, 1, add
+    SATD_4x8_SSE vertical, 1, 4, 5
     lea r0, [r0 + r1*2*SIZEOF_PIXEL]
     lea r2, [r2 + r3*2*SIZEOF_PIXEL]
-    SATD_4x8_SSE vertical, 1, add
+    SATD_4x8_SSE vertical, 1, 4, 5
     lea r0, [r6 + 8*SIZEOF_PIXEL]
     lea r2, [r7 + 8*SIZEOF_PIXEL]
-    SATD_4x8_SSE vertical, 1, add
+    SATD_4x8_SSE vertical, 1, 4, 5
     lea r0, [r0 + r1*2*SIZEOF_PIXEL]
     lea r2, [r2 + r3*2*SIZEOF_PIXEL]
-    SATD_4x8_SSE vertical, 1, add
-    HADDW m7, m1
-    movd eax, m7
+    SATD_4x8_SSE vertical, 1, 4, 5
+    pxor    m1, m1
+    movhlps m1, m7
+    paddd   m7, m1
+    pshufd  m1, m7, 1
+    paddd   m7, m1
+    movd   eax, m7
     RET
 %else
 cglobal pixel_satd_12x16, 4,7,8,0-gprsize
     SATD_START_MMX
     mov r6, r0
     mov [rsp], r2
+    pxor m7, m7
 %if vertical==0
     mova m7, [hmul_4p]
 %endif
-    SATD_4x8_SSE vertical, 0, swap
+    SATD_4x8_SSE vertical, 0, 4, 5
     lea r0, [r0 + r1*2*SIZEOF_PIXEL]
     lea r2, [r2 + r3*2*SIZEOF_PIXEL]
-    SATD_4x8_SSE vertical, 1, add
+    SATD_4x8_SSE vertical, 1, 4, 5
     lea r0, [r6 + 4*SIZEOF_PIXEL]
     mov r2, [rsp]
     add r2, 4*SIZEOF_PIXEL
-    SATD_4x8_SSE vertical, 1, add
+    SATD_4x8_SSE vertical, 1, 4, 5
     lea r0, [r0 + r1*2*SIZEOF_PIXEL]
     lea r2, [r2 + r3*2*SIZEOF_PIXEL]
-    SATD_4x8_SSE vertical, 1, add
+    SATD_4x8_SSE vertical, 1, 4, 5
     lea r0, [r6 + 8*SIZEOF_PIXEL]
     mov r2, [rsp]
     add r2, 8*SIZEOF_PIXEL
-    SATD_4x8_SSE vertical, 1, add
+    SATD_4x8_SSE vertical, 1, 4, 5
     lea r0, [r0 + r1*2*SIZEOF_PIXEL]
     lea r2, [r2 + r3*2*SIZEOF_PIXEL]
-    SATD_4x8_SSE vertical, 1, add
-    HADDW m7, m1
-    movd eax, m7
+    SATD_4x8_SSE vertical, 1, 4, 5
+    pxor    m1, m1
+    movhlps m1, m7
+    paddd   m7, m1
+    pshufd  m1, m7, 1
+    paddd   m7, m1
+    movd   eax, m7
     RET
 %endif
 
diff -r 1374f1168c5c -r b14a8528c478 source/common/x86/sad16-a.asm
--- a/source/common/x86/sad16-a.asm	Wed Feb 05 12:59:57 2014 +0530
+++ b/source/common/x86/sad16-a.asm	Wed Feb 05 15:02:37 2014 +0530
@@ -274,9 +274,10 @@
     lea     r0, [r0+4*r1]
     lea     r2, [r2+4*r3]
     ABSW2   m3, m4, m3, m4, m7, m5
-    paddd   m1, m2
-    paddd   m3, m4
-    paddd   m0, m1
+    paddw   m1, m2
+    paddw   m3, m4
+    paddw   m3, m1
+    pmaddwd m3, [pw_1]
     paddd   m0, m3
 %else
     movu    m1, [r2]
@@ -286,8 +287,9 @@
     ABSW2   m1, m2, m1, m2, m3, m4
     lea     r0, [r0+4*r1]
     lea     r2, [r2+4*r3]
-    paddw   m0, m1
-    paddw   m0, m2
+    paddw   m2, m1
+    pmaddwd m2, [pw_1]
+    paddd   m0, m2
 %endif
 %endmacro
 
@@ -308,7 +310,7 @@
     jg .loop
 %endif
 
-    HADDW   m0, m1
+    HADDD   m0, m1
     movd    eax, xm0
     RET
 %endmacro