Revert r97185 "Update libvpx snapshot to v0.9.7-p1 (Cayuga)."

libvpx/source/libvpx/vp8/encoder/x86/variance_impl_mmx.asm

Index: libvpx/source/libvpx/vp8/encoder/x86/variance_impl_mmx.asm
diff --git a/libvpx/source/libvpx/vp8/encoder/x86/variance_impl_mmx.asm b/libvpx/source/libvpx/vp8/encoder/x86/variance_impl_mmx.asm
index 13b76ea91fd62dfc5d32dfd254d0ab4b3ac488bb..67a9b4d3efa2e624acb41754045e601b6b10da3d 100644
--- a/libvpx/source/libvpx/vp8/encoder/x86/variance_impl_mmx.asm
+++ b/libvpx/source/libvpx/vp8/encoder/x86/variance_impl_mmx.asm
@@ -843,6 +843,136 @@ filter_block2d_bil_var_mmx_loop:
     pop         rbp
     ret
 
+;unsigned int vp8_get16x16pred_error_mmx
+;(
+;    unsigned char *src_ptr,
+;    int src_stride,
+;    unsigned char *ref_ptr,
+;    int ref_stride
+;)
+global sym(vp8_get16x16pred_error_mmx)
+sym(vp8_get16x16pred_error_mmx):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 4
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    sub         rsp, 16
+    ; end prolog
+
+        mov         rsi,            arg(0) ;DWORD PTR [src_ptr]
+        mov         rdi,            arg(2) ;DWORD PTR [ref_ptr]
+
+        movsxd      rax,            DWORD PTR arg(1) ;[src_stride]
+        movsxd      rdx,            DWORD PTR arg(3) ;[ref_stride]
+
+        pxor        mm0,            mm0                     ; clear xmm0 for unpack
+        pxor        mm7,            mm7                     ; clear xmm7 for accumulating diffs
+
+        pxor        mm6,            mm6                     ; clear xmm6 for accumulating sse
+        mov         rcx,            16
+
+var16loop:
+
+        movq        mm1,            [rsi]
+        movq        mm2,            [rdi]
+
+        movq        mm3,            mm1
+        movq        mm4,            mm2
+
+        punpcklbw   mm1,            mm0
+        punpckhbw   mm3,            mm0
+
+        punpcklbw   mm2,            mm0
+        punpckhbw   mm4,            mm0
+
+        psubw       mm1,            mm2
+        psubw       mm3,            mm4
+
+        paddw       mm7,            mm1
+        pmaddwd     mm1,            mm1
+
+        paddw       mm7,            mm3
+        pmaddwd     mm3,            mm3
+
+        paddd       mm6,            mm1
+        paddd       mm6,            mm3
+
+
+        movq        mm1,            [rsi+8]
+        movq        mm2,            [rdi+8]
+
+        movq        mm3,            mm1
+        movq        mm4,            mm2
+
+        punpcklbw   mm1,            mm0
+        punpckhbw   mm3,            mm0
+
+        punpcklbw   mm2,            mm0
+        punpckhbw   mm4,            mm0
+
+        psubw       mm1,            mm2
+        psubw       mm3,            mm4
+
+        paddw       mm7,            mm1
+        pmaddwd     mm1,            mm1
+
+        paddw       mm7,            mm3
+        pmaddwd     mm3,            mm3
+
+        paddd       mm6,            mm1
+        paddd       mm6,            mm3
+
+        add         rsi,            rax
+        add         rdi,            rdx
+
+        sub         rcx,            1
+        jnz         var16loop
+
+
+        movq        mm1,            mm6
+        pxor        mm6,            mm6
+
+        pxor        mm5,            mm5
+        punpcklwd   mm6,            mm7
+
+        punpckhwd   mm5,            mm7
+        psrad       mm5,            16
+
+        psrad       mm6,            16
+        paddd       mm6,            mm5
+
+        movq        mm2,            mm1
+        psrlq       mm1,            32
+
+        paddd       mm2,            mm1
+        movq        mm7,            mm6
+
+        psrlq       mm6,            32
+        paddd       mm6,            mm7
+
+        movd DWORD PTR [rsp],       mm6  ;Sum
+        movd DWORD PTR [rsp+4],     mm2  ;SSE
+
+        ; return (SSE-((Sum*Sum)>>8));
+        movsxd      rdx, dword ptr [rsp]
+        imul        rdx, rdx
+        sar         rdx, 8
+        movsxd      rax, dword ptr [rsp + 4]
+        sub         rax, rdx
+
+
+    ; begin epilog
+    add rsp, 16
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
 
 SECTION_RODATA
 ;short mmx_bi_rd[4] = { 64, 64, 64, 64};