Improve SSE2 half-pixel filter funtions

Rewrote these functions to process 16 pixels once instead of 8.

Change-Id: Ic67e80124467a446a3df4cfecfb76a4248602adb

3 files changed, 391 insertions, 114 deletions

diff --git a/vp8/encoder/x86/variance_impl_sse2.asm b/vp8/encoder/x86/variance_impl_sse2.asm
index 5d1a17d44..c2c30deb2 100644
--- a/vp8/encoder/x86/variance_impl_sse2.asm
+++ b/vp8/encoder/x86/variance_impl_sse2.asm
@@ -790,7 +790,7 @@ filter_block2d_bil_variance:
     ret
 
 
-;void vp8_half_horiz_vert_variance16x_h_sse2
+;void vp8_half_horiz_vert_variance8x_h_sse2
 ;(
 ;    unsigned char *ref_ptr,
 ;    int ref_pixels_per_line,
@@ -800,8 +800,8 @@ filter_block2d_bil_variance:
 ;    int *sum,
 ;    unsigned int *sumsquared
 ;)
-global sym(vp8_half_horiz_vert_variance16x_h_sse2)
-sym(vp8_half_horiz_vert_variance16x_h_sse2):
+global sym(vp8_half_horiz_vert_variance8x_h_sse2)
+sym(vp8_half_horiz_vert_variance8x_h_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 7
@@ -835,7 +835,7 @@ sym(vp8_half_horiz_vert_variance16x_h_sse2):
         add             rsi, r8
 %endif
 
-vp8_half_horiz_vert_variance16x_h_1:
+vp8_half_horiz_vert_variance8x_h_1:
 
         movq            xmm1,           QWORD PTR [rsi]     ;
         movq            xmm2,           QWORD PTR [rsi+1]   ;
@@ -863,7 +863,7 @@ vp8_half_horiz_vert_variance16x_h_1:
 %endif
 
         sub             rcx,            1                   ;
-        jnz             vp8_half_horiz_vert_variance16x_h_1     ;
+        jnz             vp8_half_horiz_vert_variance8x_h_1     ;
 
         movdq2q         mm6,            xmm6                ;
         movdq2q         mm7,            xmm7                ;
@@ -910,8 +910,123 @@ vp8_half_horiz_vert_variance16x_h_1:
     pop         rbp
     ret
 
+;void vp8_half_horiz_vert_variance16x_h_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp8_half_horiz_vert_variance16x_h_sse2)
+sym(vp8_half_horiz_vert_variance16x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
 
-;void vp8_half_vert_variance16x_h_sse2
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0) ;ref_ptr              ;
+
+        mov             rdi,            arg(2) ;src_ptr              ;
+        movsxd          rcx,            dword ptr arg(4) ;Height              ;
+        movsxd          rax,            dword ptr arg(1) ;ref_pixels_per_line
+        movsxd          rdx,            dword ptr arg(3)    ;src_pixels_per_line
+
+        pxor            xmm0,           xmm0                ;
+
+        movdqu          xmm5,           XMMWORD PTR [rsi]
+        movdqu          xmm3,           XMMWORD PTR [rsi+1]
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3) horizontal line 1
+
+        lea             rsi,            [rsi + rax]
+
+vp8_half_horiz_vert_variance16x_h_1:
+        movdqu          xmm1,           XMMWORD PTR [rsi]     ;
+        movdqu          xmm2,           XMMWORD PTR [rsi+1]   ;
+        pavgb           xmm1,           xmm2                ;  xmm1 = avg(xmm1,xmm3) horizontal line i+1
+
+        pavgb           xmm5,           xmm1                ;  xmm = vertical average of the above
+
+        movdqa          xmm4,           xmm5
+        punpcklbw       xmm5,           xmm0                ;  xmm5 = words of above
+        punpckhbw       xmm4,           xmm0
+
+        movq            xmm3,           QWORD PTR [rdi]     ;  xmm3 = d0,d1,d2..d7
+        punpcklbw       xmm3,           xmm0                ;  xmm3 = words of above
+        psubw           xmm5,           xmm3                ;  xmm5 -= xmm3
+
+        movq            xmm3,           QWORD PTR [rdi+8]
+        punpcklbw       xmm3,           xmm0
+        psubw           xmm4,           xmm3
+
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        paddw           xmm6,           xmm4
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        pmaddwd         xmm4,           xmm4
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+        paddd           xmm7,           xmm4
+
+        movdqa          xmm5,           xmm1                ;  save xmm1 for use on the next row
+
+        lea             rsi,            [rsi + rax]
+        lea             rdi,            [rdi + rdx]
+
+        sub             rcx,            1                   ;
+        jnz             vp8_half_horiz_vert_variance16x_h_1     ;
+
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(5) ;[Sum]
+        mov         rdi,            arg(6) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp8_half_vert_variance8x_h_sse2
 ;(
 ;    unsigned char *ref_ptr,
 ;    int ref_pixels_per_line,
@@ -921,8 +1036,8 @@ vp8_half_horiz_vert_variance16x_h_1:
 ;    int *sum,
 ;    unsigned int *sumsquared
 ;)
-global sym(vp8_half_vert_variance16x_h_sse2)
-sym(vp8_half_vert_variance16x_h_sse2):
+global sym(vp8_half_vert_variance8x_h_sse2)
+sym(vp8_half_vert_variance8x_h_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 7
@@ -945,7 +1060,7 @@ sym(vp8_half_vert_variance16x_h_sse2):
         movsxd          rax,            dword ptr arg(1) ;ref_pixels_per_line
 
         pxor            xmm0,           xmm0                ;
-vp8_half_vert_variance16x_h_1:
+vp8_half_vert_variance8x_h_1:
         movq            xmm5,           QWORD PTR [rsi]     ;  xmm5 = s0,s1,s2..s8
         movq            xmm3,           QWORD PTR [rsi+rax] ;  xmm3 = s1,s2,s3..s9
 
@@ -969,7 +1084,7 @@ vp8_half_vert_variance16x_h_1:
 %endif
 
         sub             rcx,            1                   ;
-        jnz             vp8_half_vert_variance16x_h_1          ;
+        jnz             vp8_half_vert_variance8x_h_1          ;
 
         movdq2q         mm6,            xmm6                ;
         movdq2q         mm7,            xmm7                ;
@@ -1016,8 +1131,115 @@ vp8_half_vert_variance16x_h_1:
     pop         rbp
     ret
 
+;void vp8_half_vert_variance16x_h_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp8_half_vert_variance16x_h_sse2)
+sym(vp8_half_vert_variance16x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
 
-;void vp8_half_horiz_variance16x_h_sse2
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0)              ;ref_ptr
+
+        mov             rdi,            arg(2)              ;src_ptr
+        movsxd          rcx,            dword ptr arg(4)    ;Height
+        movsxd          rax,            dword ptr arg(1)    ;ref_pixels_per_line
+        movsxd          rdx,            dword ptr arg(3)    ;src_pixels_per_line
+
+        movdqu          xmm5,           XMMWORD PTR [rsi]
+        lea             rsi,            [rsi + rax          ]
+        pxor            xmm0,           xmm0
+
+vp8_half_vert_variance16x_h_1:
+        movdqu          xmm3,           XMMWORD PTR [rsi]
+
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3)
+        movdqa          xmm4,           xmm5
+        punpcklbw       xmm5,           xmm0
+        punpckhbw       xmm4,           xmm0
+
+        movq            xmm2,           QWORD PTR [rdi]
+        punpcklbw       xmm2,           xmm0
+        psubw           xmm5,           xmm2
+        movq            xmm2,           QWORD PTR [rdi+8]
+        punpcklbw       xmm2,           xmm0
+        psubw           xmm4,           xmm2
+
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        paddw           xmm6,           xmm4
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        pmaddwd         xmm4,           xmm4
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+        paddd           xmm7,           xmm4
+
+        movdqa          xmm5,           xmm3
+
+        lea             rsi,            [rsi + rax]
+        lea             rdi,            [rdi + rdx]
+
+        sub             rcx,            1
+        jnz             vp8_half_vert_variance16x_h_1
+
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(5) ;[Sum]
+        mov         rdi,            arg(6) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vp8_half_horiz_variance8x_h_sse2
 ;(
 ;    unsigned char *ref_ptr,
 ;    int ref_pixels_per_line,
@@ -1027,8 +1249,8 @@ vp8_half_vert_variance16x_h_1:
 ;    int *sum,
 ;    unsigned int *sumsquared
 ;)
-global sym(vp8_half_horiz_variance16x_h_sse2)
-sym(vp8_half_horiz_variance16x_h_sse2):
+global sym(vp8_half_horiz_variance8x_h_sse2)
+sym(vp8_half_horiz_variance8x_h_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 7
@@ -1050,7 +1272,7 @@ sym(vp8_half_horiz_variance16x_h_sse2):
         movsxd          rcx,            dword ptr arg(4) ;Height              ;
 
         pxor            xmm0,           xmm0                ;
-vp8_half_horiz_variance16x16_1:
+vp8_half_horiz_variance8x_h_1:
         movq            xmm5,           QWORD PTR [rsi]     ;  xmm5 = s0,s1,s2..s8
         movq            xmm3,           QWORD PTR [rsi+1]   ;  xmm3 = s1,s2,s3..s9
 
@@ -1073,7 +1295,7 @@ vp8_half_horiz_variance16x16_1:
         add             rdi, r9
 %endif
         sub             rcx,            1                   ;
-        jnz             vp8_half_horiz_variance16x16_1        ;
+        jnz             vp8_half_horiz_variance8x_h_1        ;
 
         movdq2q         mm6,            xmm6                ;
         movdq2q         mm7,            xmm7                ;
@@ -1120,6 +1342,109 @@ vp8_half_horiz_variance16x16_1:
     pop         rbp
     ret
 
+;void vp8_half_horiz_variance16x_h_sse2
+;(
+;    unsigned char *ref_ptr,
+;    int ref_pixels_per_line,
+;    unsigned char *src_ptr,
+;    int src_pixels_per_line,
+;    unsigned int Height,
+;    int *sum,
+;    unsigned int *sumsquared
+;)
+global sym(vp8_half_horiz_variance16x_h_sse2)
+sym(vp8_half_horiz_variance16x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0) ;ref_ptr              ;
+
+        mov             rdi,            arg(2) ;src_ptr              ;
+        movsxd          rcx,            dword ptr arg(4) ;Height              ;
+        movsxd          rax,            dword ptr arg(1) ;ref_pixels_per_line
+        movsxd          rdx,            dword ptr arg(3)    ;src_pixels_per_line
+
+        pxor            xmm0,           xmm0                ;
+
+vp8_half_horiz_variance16x_h_1:
+        movdqu          xmm5,           XMMWORD PTR [rsi]     ;  xmm5 = s0,s1,s2..s15
+        movdqu          xmm3,           XMMWORD PTR [rsi+1]   ;  xmm3 = s1,s2,s3..s16
+
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3)
+        movdqa          xmm1,           xmm5
+        punpcklbw       xmm5,           xmm0                ;  xmm5 = words of above
+        punpckhbw       xmm1,           xmm0
+
+        movq            xmm3,           QWORD PTR [rdi]     ;  xmm3 = d0,d1,d2..d7
+        punpcklbw       xmm3,           xmm0                ;  xmm3 = words of above
+        movq            xmm2,           QWORD PTR [rdi+8]
+        punpcklbw       xmm2,           xmm0
+
+        psubw           xmm5,           xmm3                ;  xmm5 -= xmm3
+        psubw           xmm1,           xmm2
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        paddw           xmm6,           xmm1
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        pmaddwd         xmm1,           xmm1
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+        paddd           xmm7,           xmm1
+
+        lea             rsi,            [rsi + rax]
+        lea             rdi,            [rdi + rdx]
+
+        sub             rcx,            1                   ;
+        jnz             vp8_half_horiz_variance16x_h_1        ;
+
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(5) ;[Sum]
+        mov         rdi,            arg(6) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
 
 SECTION_RODATA
 ;    short xmm_bi_rd[8] = { 64, 64, 64, 64,64, 64, 64, 64};
diff --git a/vp8/encoder/x86/variance_sse2.c b/vp8/encoder/x86/variance_sse2.c
index 7cf6a6308..4612a6711 100644
--- a/vp8/encoder/x86/variance_sse2.c
+++ b/vp8/encoder/x86/variance_sse2.c
@@ -81,6 +81,16 @@ void vp8_filter_block2d_bil_var_sse2
     int *sum,
     unsigned int *sumsquared
 );
+void vp8_half_horiz_vert_variance8x_h_sse2
+(
+    const unsigned char *ref_ptr,
+    int ref_pixels_per_line,
+    const unsigned char *src_ptr,
+    int src_pixels_per_line,
+    unsigned int Height,
+    int *sum,
+    unsigned int *sumsquared
+);
 void vp8_half_horiz_vert_variance16x_h_sse2
 (
     const unsigned char *ref_ptr,
@@ -91,6 +101,16 @@ void vp8_half_horiz_vert_variance16x_h_sse2
     int *sum,
     unsigned int *sumsquared
 );
+void vp8_half_horiz_variance8x_h_sse2
+(
+    const unsigned char *ref_ptr,
+    int ref_pixels_per_line,
+    const unsigned char *src_ptr,
+    int src_pixels_per_line,
+    unsigned int Height,
+    int *sum,
+    unsigned int *sumsquared
+);
 void vp8_half_horiz_variance16x_h_sse2
 (
     const unsigned char *ref_ptr,
@@ -101,6 +121,16 @@ void vp8_half_horiz_variance16x_h_sse2
     int *sum,
     unsigned int *sumsquared
 );
+void vp8_half_vert_variance8x_h_sse2
+(
+    const unsigned char *ref_ptr,
+    int ref_pixels_per_line,
+    const unsigned char *src_ptr,
+    int src_pixels_per_line,
+    unsigned int Height,
+    int *sum,
+    unsigned int *sumsquared
+);
 void vp8_half_vert_variance16x_h_sse2
 (
     const unsigned char *ref_ptr,
@@ -262,21 +292,21 @@ unsigned int vp8_sub_pixel_variance8x8_wmt
 
     if (xoffset == 4 && yoffset == 0)
     {
-        vp8_half_horiz_variance16x_h_sse2(
+        vp8_half_horiz_variance8x_h_sse2(
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 8,
             &xsum, &xxsum);
     }
     else if (xoffset == 0 && yoffset == 4)
     {
-        vp8_half_vert_variance16x_h_sse2(
+        vp8_half_vert_variance8x_h_sse2(
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 8,
             &xsum, &xxsum);
     }
     else if (xoffset == 4 && yoffset == 4)
     {
-        vp8_half_horiz_vert_variance16x_h_sse2(
+        vp8_half_horiz_vert_variance8x_h_sse2(
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 8,
             &xsum, &xxsum);
@@ -317,11 +347,6 @@ unsigned int vp8_sub_pixel_variance16x16_wmt
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 16,
             &xsum0, &xxsum0);
-
-        vp8_half_horiz_variance16x_h_sse2(
-            src_ptr + 8, src_pixels_per_line,
-            dst_ptr + 8, dst_pixels_per_line, 16,
-            &xsum1, &xxsum1);
     }
     else if (xoffset == 0 && yoffset == 4)
     {
@@ -329,11 +354,6 @@ unsigned int vp8_sub_pixel_variance16x16_wmt
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 16,
             &xsum0, &xxsum0);
-
-        vp8_half_vert_variance16x_h_sse2(
-            src_ptr + 8, src_pixels_per_line,
-            dst_ptr + 8, dst_pixels_per_line, 16,
-            &xsum1, &xxsum1);
     }
     else if (xoffset == 4 && yoffset == 4)
     {
@@ -341,11 +361,6 @@ unsigned int vp8_sub_pixel_variance16x16_wmt
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 16,
             &xsum0, &xxsum0);
-
-        vp8_half_horiz_vert_variance16x_h_sse2(
-            src_ptr + 8, src_pixels_per_line,
-            dst_ptr + 8, dst_pixels_per_line, 16,
-            &xsum1, &xxsum1);
     }
     else
     {
@@ -356,17 +371,16 @@ unsigned int vp8_sub_pixel_variance16x16_wmt
             &xsum0, &xxsum0
         );
 
-
         vp8_filter_block2d_bil_var_sse2(
             src_ptr + 8, src_pixels_per_line,
             dst_ptr + 8, dst_pixels_per_line, 16,
             xoffset, yoffset,
             &xsum1, &xxsum1
         );
+        xsum0 += xsum1;
+        xxsum0 += xxsum1;
     }
 
-    xsum0 += xsum1;
-    xxsum0 += xxsum1;
     *sse = xxsum0;
     return (xxsum0 - ((xsum0 * xsum0) >> 8));
 }
@@ -406,11 +420,6 @@ unsigned int vp8_sub_pixel_variance16x8_wmt
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 8,
             &xsum0, &xxsum0);
-
-        vp8_half_horiz_variance16x_h_sse2(
-            src_ptr + 8, src_pixels_per_line,
-            dst_ptr + 8, dst_pixels_per_line, 8,
-            &xsum1, &xxsum1);
     }
     else if (xoffset == 0 && yoffset == 4)
     {
@@ -418,11 +427,6 @@ unsigned int vp8_sub_pixel_variance16x8_wmt
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 8,
             &xsum0, &xxsum0);
-
-        vp8_half_vert_variance16x_h_sse2(
-            src_ptr + 8, src_pixels_per_line,
-            dst_ptr + 8, dst_pixels_per_line, 8,
-            &xsum1, &xxsum1);
     }
     else if (xoffset == 4 && yoffset == 4)
     {
@@ -430,11 +434,6 @@ unsigned int vp8_sub_pixel_variance16x8_wmt
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 8,
             &xsum0, &xxsum0);
-
-        vp8_half_horiz_vert_variance16x_h_sse2(
-            src_ptr + 8, src_pixels_per_line,
-            dst_ptr + 8, dst_pixels_per_line, 8,
-            &xsum1, &xxsum1);
     }
     else
     {
@@ -449,11 +448,10 @@ unsigned int vp8_sub_pixel_variance16x8_wmt
             dst_ptr + 8, dst_pixels_per_line, 8,
             xoffset, yoffset,
             &xsum1, &xxsum1);
+        xsum0 += xsum1;
+        xxsum0 += xxsum1;
     }
 
-    xsum0 += xsum1;
-    xxsum0 += xxsum1;
-
     *sse = xxsum0;
     return (xxsum0 - ((xsum0 * xsum0) >> 7));
 }
@@ -474,21 +472,21 @@ unsigned int vp8_sub_pixel_variance8x16_wmt
 
     if (xoffset == 4 && yoffset == 0)
     {
-        vp8_half_horiz_variance16x_h_sse2(
+        vp8_half_horiz_variance8x_h_sse2(
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 16,
             &xsum, &xxsum);
     }
     else if (xoffset == 0 && yoffset == 4)
     {
-        vp8_half_vert_variance16x_h_sse2(
+        vp8_half_vert_variance8x_h_sse2(
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 16,
             &xsum, &xxsum);
     }
     else if (xoffset == 4 && yoffset == 4)
     {
-        vp8_half_horiz_vert_variance16x_h_sse2(
+        vp8_half_horiz_vert_variance8x_h_sse2(
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 16,
             &xsum, &xxsum);
@@ -589,21 +587,14 @@ unsigned int vp8_variance_halfpixvar16x16_h_wmt(
     int  dst_pixels_per_line,
     unsigned int *sse)
 {
-    int xsum0, xsum1;
-    unsigned int xxsum0, xxsum1;
+    int xsum0;
+    unsigned int xxsum0;
 
     vp8_half_horiz_variance16x_h_sse2(
         src_ptr, src_pixels_per_line,
         dst_ptr, dst_pixels_per_line, 16,
         &xsum0, &xxsum0);
 
-    vp8_half_horiz_variance16x_h_sse2(
-        src_ptr + 8, src_pixels_per_line,
-        dst_ptr + 8, dst_pixels_per_line, 16,
-        &xsum1, &xxsum1);
-
-    xsum0 += xsum1;
-    xxsum0 += xxsum1;
     *sse = xxsum0;
     return (xxsum0 - ((xsum0 * xsum0) >> 8));
 }
@@ -616,21 +607,13 @@ unsigned int vp8_variance_halfpixvar16x16_v_wmt(
     int  dst_pixels_per_line,
     unsigned int *sse)
 {
-    int xsum0, xsum1;
-    unsigned int xxsum0, xxsum1;
-
+    int xsum0;
+    unsigned int xxsum0;
     vp8_half_vert_variance16x_h_sse2(
         src_ptr, src_pixels_per_line,
         dst_ptr, dst_pixels_per_line, 16,
         &xsum0, &xxsum0);
 
-    vp8_half_vert_variance16x_h_sse2(
-        src_ptr + 8, src_pixels_per_line,
-        dst_ptr + 8, dst_pixels_per_line, 16,
-        &xsum1, &xxsum1);
-
-    xsum0 += xsum1;
-    xxsum0 += xxsum1;
     *sse = xxsum0;
     return (xxsum0 - ((xsum0 * xsum0) >> 8));
 }
@@ -643,21 +626,14 @@ unsigned int vp8_variance_halfpixvar16x16_hv_wmt(
     int  dst_pixels_per_line,
     unsigned int *sse)
 {
-    int xsum0, xsum1;
-    unsigned int xxsum0, xxsum1;
+    int xsum0;
+    unsigned int xxsum0;
 
     vp8_half_horiz_vert_variance16x_h_sse2(
         src_ptr, src_pixels_per_line,
         dst_ptr, dst_pixels_per_line, 16,
         &xsum0, &xxsum0);
 
-    vp8_half_horiz_vert_variance16x_h_sse2(
-        src_ptr + 8, src_pixels_per_line,
-        dst_ptr + 8, dst_pixels_per_line, 16,
-        &xsum1, &xxsum1);
-
-    xsum0 += xsum1;
-    xxsum0 += xxsum1;
     *sse = xxsum0;
     return (xxsum0 - ((xsum0 * xsum0) >> 8));
 }
diff --git a/vp8/encoder/x86/variance_ssse3.c b/vp8/encoder/x86/variance_ssse3.c
index 750ae8b86..d50ae3ade 100644
--- a/vp8/encoder/x86/variance_ssse3.c
+++ b/vp8/encoder/x86/variance_ssse3.c
@@ -87,14 +87,6 @@ unsigned int vp8_sub_pixel_variance16x16_ssse3
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 16,
             &xsum0, &xxsum0);
-
-        vp8_half_horiz_variance16x_h_sse2(
-            src_ptr + 8, src_pixels_per_line,
-            dst_ptr + 8, dst_pixels_per_line, 16,
-            &xsum1, &xxsum1);
-
-        xsum0 += xsum1;
-        xxsum0 += xxsum1;
     }
     else if (xoffset == 0 && yoffset == 4)
     {
@@ -102,14 +94,6 @@ unsigned int vp8_sub_pixel_variance16x16_ssse3
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 16,
             &xsum0, &xxsum0);
-
-        vp8_half_vert_variance16x_h_sse2(
-            src_ptr + 8, src_pixels_per_line,
-            dst_ptr + 8, dst_pixels_per_line, 16,
-            &xsum1, &xxsum1);
-
-        xsum0 += xsum1;
-        xxsum0 += xxsum1;
     }
     else if (xoffset == 4 && yoffset == 4)
     {
@@ -117,22 +101,14 @@ unsigned int vp8_sub_pixel_variance16x16_ssse3
             src_ptr, src_pixels_per_line,
             dst_ptr, dst_pixels_per_line, 16,
             &xsum0, &xxsum0);
-
-        vp8_half_horiz_vert_variance16x_h_sse2(
-            src_ptr + 8, src_pixels_per_line,
-            dst_ptr + 8, dst_pixels_per_line, 16,
-            &xsum1, &xxsum1);
-
-        xsum0 += xsum1;
-        xxsum0 += xxsum1;
     }
     else
     {
-      vp8_filter_block2d_bil_var_ssse3(
-          src_ptr, src_pixels_per_line,
-          dst_ptr, dst_pixels_per_line, 16,
-          xoffset, yoffset,
-          &xsum0, &xxsum0);
+        vp8_filter_block2d_bil_var_ssse3(
+            src_ptr, src_pixels_per_line,
+            dst_ptr, dst_pixels_per_line, 16,
+            xoffset, yoffset,
+            &xsum0, &xxsum0);
     }
 
     *sse = xxsum0;