Merge remote branch 'internal/upstream-experimental' into HEAD

Conflicts:
	vp8/decoder/onyxd_int.h

Change-Id: Icf445b589c2bc61d93d8c977379bbd84387d0488

11 files changed, 144 insertions, 178 deletions

diff --git a/vp8/common/blockd.h b/vp8/common/blockd.h
index 906e05520..3c3592aab 100644
--- a/vp8/common/blockd.h
+++ b/vp8/common/blockd.h
@@ -175,8 +175,6 @@ typedef struct
     unsigned char need_to_clamp_mvs;
 
     unsigned char segment_id;                  /* Which set of segmentation parameters should be used for this MB */
-
-    unsigned char force_no_skip; /* encoder only */
 } MB_MODE_INFO;
 
 
diff --git a/vp8/common/reconinter.c b/vp8/common/reconinter.c
index 7cfab4140..6862bae11 100644
--- a/vp8/common/reconinter.c
+++ b/vp8/common/reconinter.c
@@ -207,12 +207,12 @@ static void build_inter_predictors2b(MACROBLOCKD *x, BLOCKD *d, int pitch)
 }
 
 
+/*encoder only*/
 void vp8_build_inter_predictors_mbuv(MACROBLOCKD *x)
 {
     int i;
 
-    if (x->mode_info_context->mbmi.ref_frame != INTRA_FRAME &&
-        x->mode_info_context->mbmi.mode != SPLITMV)
+    if (x->mode_info_context->mbmi.mode != SPLITMV)
     {
         unsigned char *uptr, *vptr;
         unsigned char *upred_ptr = &x->predictor[256];
@@ -257,69 +257,32 @@ void vp8_build_inter_predictors_mbuv(MACROBLOCKD *x)
 }
 
 /*encoder only*/
-void vp8_build_inter_predictors_mby(MACROBLOCKD *x)
+void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x)
 {
+    unsigned char *ptr_base;
+    unsigned char *ptr;
+    unsigned char *pred_ptr = x->predictor;
+    int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
+    int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
+    int pre_stride = x->block[0].pre_stride;
 
-  if (x->mode_info_context->mbmi.ref_frame != INTRA_FRAME &&
-      x->mode_info_context->mbmi.mode != SPLITMV)
-    {
-        unsigned char *ptr_base;
-        unsigned char *ptr;
-        unsigned char *pred_ptr = x->predictor;
-        int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
-        int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
-        int pre_stride = x->block[0].pre_stride;
-
-        ptr_base = x->pre.y_buffer;
-        ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
+    ptr_base = x->pre.y_buffer;
+    ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
 
-        if ((mv_row | mv_col) & 7)
-        {
-            x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, pred_ptr, 16);
-        }
-        else
-        {
-            RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, pred_ptr, 16);
-        }
+    if ((mv_row | mv_col) & 7)
+    {
+        x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, pred_ptr, 16);
     }
     else
     {
-        int i;
-
-        if (x->mode_info_context->mbmi.partitioning < 3)
-        {
-            for (i = 0; i < 4; i++)
-            {
-                BLOCKD *d = &x->block[bbb[i]];
-                build_inter_predictors4b(x, d, 16);
-            }
-
-        }
-        else
-        {
-            for (i = 0; i < 16; i += 2)
-            {
-                BLOCKD *d0 = &x->block[i];
-                BLOCKD *d1 = &x->block[i+1];
-
-                if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
-                    build_inter_predictors2b(x, d0, 16);
-                else
-                {
-                    vp8_build_inter_predictors_b(d0, 16, x->subpixel_predict);
-                    vp8_build_inter_predictors_b(d1, 16, x->subpixel_predict);
-                }
-
-            }
-        }
+        RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, pred_ptr, 16);
     }
 }
 
 void vp8_build_inter_predictors_mb(MACROBLOCKD *x)
 {
 
-    if (x->mode_info_context->mbmi.ref_frame != INTRA_FRAME &&
-        x->mode_info_context->mbmi.mode != SPLITMV)
+    if (x->mode_info_context->mbmi.mode != SPLITMV)
     {
         int offset;
         unsigned char *ptr_base;
@@ -535,58 +498,58 @@ static void vp8_build_inter_predictors_b_s(BLOCKD *d, unsigned char *dst_ptr, vp
 
 
 
-void vp8_build_inter_predictors_mb_s(MACROBLOCKD *x)
+void vp8_build_inter16x16_predictors_mb_s(MACROBLOCKD *x)
 {
-    /*unsigned char *pred_ptr = x->block[0].predictor;
-    unsigned char *dst_ptr = *(x->block[0].base_dst) + x->block[0].dst;*/
-    unsigned char *pred_ptr = x->predictor;
     unsigned char *dst_ptr = x->dst.y_buffer;
 
-    if (x->mode_info_context->mbmi.mode != SPLITMV)
-    {
-        int offset;
-        unsigned char *ptr_base;
-        unsigned char *ptr;
-        unsigned char *uptr, *vptr;
-        /*unsigned char *pred_ptr = x->predictor;
-        unsigned char *upred_ptr = &x->predictor[256];
-        unsigned char *vpred_ptr = &x->predictor[320];*/
-        unsigned char *udst_ptr = x->dst.u_buffer;
-        unsigned char *vdst_ptr = x->dst.v_buffer;
+    int offset;
+    unsigned char *ptr_base;
+    unsigned char *ptr;
+    unsigned char *uptr, *vptr;
+    unsigned char *udst_ptr = x->dst.u_buffer;
+    unsigned char *vdst_ptr = x->dst.v_buffer;
 
-        int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
-        int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
-        int pre_stride = x->dst.y_stride; /*x->block[0].pre_stride;*/
+    int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
+    int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
+    int pre_stride = x->dst.y_stride; /*x->block[0].pre_stride;*/
 
-        ptr_base = x->pre.y_buffer;
-        ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
+    ptr_base = x->pre.y_buffer;
+    ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
 
-        if ((mv_row | mv_col) & 7)
-        {
-            x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
-        }
-        else
-        {
-            RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
-        }
+    if ((mv_row | mv_col) & 7)
+    {
+        x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
+    }
+    else
+    {
+        RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
+    }
 
-        mv_row = x->block[16].bmi.mv.as_mv.row;
-        mv_col = x->block[16].bmi.mv.as_mv.col;
-        pre_stride >>= 1;
-        offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
-        uptr = x->pre.u_buffer + offset;
-        vptr = x->pre.v_buffer + offset;
+    mv_row = x->block[16].bmi.mv.as_mv.row;
+    mv_col = x->block[16].bmi.mv.as_mv.col;
+    pre_stride >>= 1;
+    offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
+    uptr = x->pre.u_buffer + offset;
+    vptr = x->pre.v_buffer + offset;
 
-        if ((mv_row | mv_col) & 7)
-        {
-            x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, udst_ptr, x->dst.uv_stride);
-            x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vdst_ptr, x->dst.uv_stride);
-        }
-        else
-        {
-            RECON_INVOKE(&x->rtcd->recon, copy8x8)(uptr, pre_stride, udst_ptr, x->dst.uv_stride);
-            RECON_INVOKE(&x->rtcd->recon, copy8x8)(vptr, pre_stride, vdst_ptr, x->dst.uv_stride);
-        }
+    if ((mv_row | mv_col) & 7)
+    {
+        x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, udst_ptr, x->dst.uv_stride);
+        x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vdst_ptr, x->dst.uv_stride);
+    }
+    else
+    {
+        RECON_INVOKE(&x->rtcd->recon, copy8x8)(uptr, pre_stride, udst_ptr, x->dst.uv_stride);
+        RECON_INVOKE(&x->rtcd->recon, copy8x8)(vptr, pre_stride, vdst_ptr, x->dst.uv_stride);
+    }
+}
+void vp8_build_inter_predictors_mb_s(MACROBLOCKD *x)
+{
+    unsigned char *dst_ptr = x->dst.y_buffer;
+
+    if (x->mode_info_context->mbmi.mode != SPLITMV)
+    {
+        vp8_build_inter16x16_predictors_mb_s(x);
     }
     else
     {
@@ -599,25 +562,20 @@ void vp8_build_inter_predictors_mb_s(MACROBLOCKD *x)
         {
             for (i = 0; i < 4; i++)
             {
+                unsigned char *ptr_base;
+                unsigned char *ptr;
                 BLOCKD *d = &x->block[bbb[i]];
-                /*build_inter_predictors4b(x, d, 16);*/
 
-                {
-                    unsigned char *ptr_base;
-                    unsigned char *ptr;
-                    unsigned char *pred_ptr = d->predictor;
+                ptr_base = *(d->base_pre);
+                ptr = ptr_base + d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
 
-                    ptr_base = *(d->base_pre);
-                    ptr = ptr_base + d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
-
-                    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
-                    {
-                        x->subpixel_predict8x8(ptr, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
-                    }
-                    else
-                    {
-                        RECON_INVOKE(&x->rtcd->recon, copy8x8)(ptr, d->pre_stride, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
-                    }
+                if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
+                {
+                    x->subpixel_predict8x8(ptr, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
+                }
+                else
+                {
+                    RECON_INVOKE(&x->rtcd->recon, copy8x8)(ptr, d->pre_stride, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
                 }
             }
         }
@@ -633,7 +591,6 @@ void vp8_build_inter_predictors_mb_s(MACROBLOCKD *x)
                     /*build_inter_predictors2b(x, d0, 16);*/
                     unsigned char *ptr_base;
                     unsigned char *ptr;
-                    unsigned char *pred_ptr = d0->predictor;
 
                     ptr_base = *(d0->base_pre);
                     ptr = ptr_base + d0->pre + (d0->bmi.mv.as_mv.row >> 3) * d0->pre_stride + (d0->bmi.mv.as_mv.col >> 3);
@@ -665,7 +622,6 @@ void vp8_build_inter_predictors_mb_s(MACROBLOCKD *x)
                 /*build_inter_predictors2b(x, d0, 8);*/
                 unsigned char *ptr_base;
                 unsigned char *ptr;
-                unsigned char *pred_ptr = d0->predictor;
 
                 ptr_base = *(d0->base_pre);
                 ptr = ptr_base + d0->pre + (d0->bmi.mv.as_mv.row >> 3) * d0->pre_stride + (d0->bmi.mv.as_mv.col >> 3);
diff --git a/vp8/common/reconinter.h b/vp8/common/reconinter.h
index 7c1dee431..688bebe96 100644
--- a/vp8/common/reconinter.h
+++ b/vp8/common/reconinter.h
@@ -14,8 +14,9 @@
 
 extern void vp8_build_inter_predictors_mb(MACROBLOCKD *x);
 extern void vp8_build_inter_predictors_mb_s(MACROBLOCKD *x);
+extern void vp8_build_inter16x16_predictors_mb_s(MACROBLOCKD *x);
 
-extern void vp8_build_inter_predictors_mby(MACROBLOCKD *x);
+extern void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x);
 extern void vp8_build_uvmvs(MACROBLOCKD *x, int fullpixel);
 extern void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, vp8_subpix_fn_t sppf);
 extern void vp8_build_inter_predictors_mbuv(MACROBLOCKD *x);
diff --git a/vp8/common/threading.h b/vp8/common/threading.h
index 44eaf0800..b7542b306 100644
--- a/vp8/common/threading.h
+++ b/vp8/common/threading.h
@@ -12,8 +12,6 @@
 #ifndef _PTHREAD_EMULATION
 #define _PTHREAD_EMULATION
 
-#define VPXINFINITE 10000       /* 10second. */
-
 #if CONFIG_OS_SUPPORT && CONFIG_MULTITHREAD
 
 /* Thread management macros */
@@ -28,7 +26,7 @@
 #define pthread_t HANDLE
 #define pthread_attr_t DWORD
 #define pthread_create(thhandle,attr,thfunc,tharg) (int)((*thhandle=(HANDLE)_beginthreadex(NULL,0,(unsigned int (__stdcall *)(void *))thfunc,tharg,0,NULL))==NULL)
-#define pthread_join(thread, result) ((WaitForSingleObject((thread),VPXINFINITE)!=WAIT_OBJECT_0) || !CloseHandle(thread))
+#define pthread_join(thread, result) ((WaitForSingleObject((thread),INFINITE)!=WAIT_OBJECT_0) || !CloseHandle(thread))
 #define pthread_detach(thread) if(thread!=NULL)CloseHandle(thread)
 #define thread_sleep(nms) Sleep(nms)
 #define pthread_cancel(thread) terminate_thread(thread,0)
@@ -62,7 +60,7 @@
 #define sem_t HANDLE
 #define pause(voidpara) __asm PAUSE
 #define sem_init(sem, sem_attr1, sem_init_value) (int)((*sem = CreateEvent(NULL,FALSE,FALSE,NULL))==NULL)
-#define sem_wait(sem) (int)(WAIT_OBJECT_0 != WaitForSingleObject(*sem,VPXINFINITE))
+#define sem_wait(sem) (int)(WAIT_OBJECT_0 != WaitForSingleObject(*sem,INFINITE))
 #define sem_post(sem) SetEvent(*sem)
 #define sem_destroy(sem) if(*sem)((int)(CloseHandle(*sem))==TRUE)
 #define thread_sleep(nms) Sleep(nms)
diff --git a/vp8/common/x86/idctllm_sse2.asm b/vp8/common/x86/idctllm_sse2.asm
index edee1578e..34a7e18ae 100644
--- a/vp8/common/x86/idctllm_sse2.asm
+++ b/vp8/common/x86/idctllm_sse2.asm
@@ -32,9 +32,6 @@ sym(idct_dequant_0_2x_sse2):
         mov         rdx,            arg(1) ; dequant
         mov         rax,            arg(0) ; qcoeff
 
-    ; Zero out xmm7, for use unpacking
-        pxor        xmm7,           xmm7
-
         movd        xmm4,           [rax]
         movd        xmm5,           [rdx]
 
@@ -43,9 +40,12 @@ sym(idct_dequant_0_2x_sse2):
 
         pmullw      xmm4,           xmm5
 
+    ; Zero out xmm5, for use unpacking
+        pxor        xmm5,           xmm5
+
     ; clear coeffs
-        movd        [rax],          xmm7
-        movd        [rax+32],       xmm7
+        movd        [rax],          xmm5
+        movd        [rax+32],       xmm5
 ;pshufb
         pshuflw     xmm4,           xmm4,       00000000b
         pshufhw     xmm4,           xmm4,       00000000b
@@ -62,10 +62,10 @@ sym(idct_dequant_0_2x_sse2):
         lea         rcx,            [3*rcx]
         movq        xmm3,           [rax+rcx]
 
-        punpcklbw   xmm0,           xmm7
-        punpcklbw   xmm1,           xmm7
-        punpcklbw   xmm2,           xmm7
-        punpcklbw   xmm3,           xmm7
+        punpcklbw   xmm0,           xmm5
+        punpcklbw   xmm1,           xmm5
+        punpcklbw   xmm2,           xmm5
+        punpcklbw   xmm3,           xmm5
 
         mov         rax,            arg(3) ; dst
         movsxd      rdx,            dword ptr arg(4) ; dst_stride
@@ -77,10 +77,10 @@ sym(idct_dequant_0_2x_sse2):
         paddw       xmm3,           xmm4
 
     ; pack up before storing
-        packuswb    xmm0,           xmm7
-        packuswb    xmm1,           xmm7
-        packuswb    xmm2,           xmm7
-        packuswb    xmm3,           xmm7
+        packuswb    xmm0,           xmm5
+        packuswb    xmm1,           xmm5
+        packuswb    xmm2,           xmm5
+        packuswb    xmm3,           xmm5
 
     ; store blocks back out
         movq        [rax],          xmm0
@@ -102,6 +102,7 @@ sym(idct_dequant_full_2x_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -347,6 +348,7 @@ sym(idct_dequant_full_2x_sse2):
     pop         rdi
     pop         rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
@@ -377,8 +379,8 @@ sym(idct_dequant_dc_0_2x_sse2):
         mov         rdi,            arg(3) ; dst
         mov         rdx,            arg(5) ; dc
 
-    ; Zero out xmm7, for use unpacking
-        pxor        xmm7,           xmm7
+    ; Zero out xmm5, for use unpacking
+        pxor        xmm5,           xmm5
 
     ; load up 2 dc words here == 2*16 = doubleword
         movd        xmm4,           [rdx]
@@ -398,10 +400,10 @@ sym(idct_dequant_dc_0_2x_sse2):
         psraw       xmm4,           3
 
     ; Predict buffer needs to be expanded from bytes to words
-        punpcklbw   xmm0,           xmm7
-        punpcklbw   xmm1,           xmm7
-        punpcklbw   xmm2,           xmm7
-        punpcklbw   xmm3,           xmm7
+        punpcklbw   xmm0,           xmm5
+        punpcklbw   xmm1,           xmm5
+        punpcklbw   xmm2,           xmm5
+        punpcklbw   xmm3,           xmm5
 
     ; Add to predict buffer
         paddw       xmm0,           xmm4
@@ -410,10 +412,10 @@ sym(idct_dequant_dc_0_2x_sse2):
         paddw       xmm3,           xmm4
 
     ; pack up before storing
-        packuswb    xmm0,           xmm7
-        packuswb    xmm1,           xmm7
-        packuswb    xmm2,           xmm7
-        packuswb    xmm3,           xmm7
+        packuswb    xmm0,           xmm5
+        packuswb    xmm1,           xmm5
+        packuswb    xmm2,           xmm5
+        packuswb    xmm3,           xmm5
 
     ; Load destination stride before writing out,
     ;   doesn't need to persist
@@ -441,6 +443,7 @@ sym(idct_dequant_dc_full_2x_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -692,6 +695,7 @@ sym(idct_dequant_dc_full_2x_sse2):
     pop         rdi
     pop         rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
diff --git a/vp8/common/x86/iwalsh_sse2.asm b/vp8/common/x86/iwalsh_sse2.asm
index 83c97df7d..1da4fd8da 100644
--- a/vp8/common/x86/iwalsh_sse2.asm
+++ b/vp8/common/x86/iwalsh_sse2.asm
@@ -17,7 +17,7 @@ sym(vp8_short_inv_walsh4x4_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 2
-    SAVE_XMM
+    SAVE_XMM 6
     push        rsi
     push        rdi
     ; end prolog
@@ -41,7 +41,7 @@ sym(vp8_short_inv_walsh4x4_sse2):
     movdqa    xmm4, xmm0
     punpcklqdq  xmm0, xmm3          ;d1 a1
     punpckhqdq  xmm4, xmm3          ;c1 b1
-    movd    xmm7, eax
+    movd    xmm6, eax
 
     movdqa    xmm1, xmm4          ;c1 b1
     paddw   xmm4, xmm0          ;dl+cl a1+b1 aka op[4] op[0]
@@ -66,7 +66,7 @@ sym(vp8_short_inv_walsh4x4_sse2):
     pshufd    xmm2, xmm1, 4eh       ;ip[8] ip[12]
     movdqa    xmm3, xmm4          ;ip[4] ip[0]
 
-    pshufd    xmm7, xmm7, 0       ;03 03 03 03 03 03 03 03
+    pshufd    xmm6, xmm6, 0       ;03 03 03 03 03 03 03 03
 
     paddw   xmm4, xmm2          ;ip[4]+ip[8] ip[0]+ip[12] aka b1 a1
     psubw   xmm3, xmm2          ;ip[4]-ip[8] ip[0]-ip[12] aka c1 d1
@@ -90,8 +90,8 @@ sym(vp8_short_inv_walsh4x4_sse2):
     punpcklwd xmm5, xmm0          ; 31 21 11 01 30 20 10 00
     punpckhwd xmm1, xmm0          ; 33 23 13 03 32 22 12 02
 ;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-    paddw   xmm5, xmm7
-    paddw   xmm1, xmm7
+    paddw   xmm5, xmm6
+    paddw   xmm1, xmm6
 
     psraw   xmm5, 3
     psraw   xmm1, 3
diff --git a/vp8/common/x86/loopfilter_sse2.asm b/vp8/common/x86/loopfilter_sse2.asm
index 849133dc4..c2ce1a106 100644
--- a/vp8/common/x86/loopfilter_sse2.asm
+++ b/vp8/common/x86/loopfilter_sse2.asm
@@ -288,7 +288,7 @@ sym(vp8_loop_filter_horizontal_edge_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -338,7 +338,7 @@ sym(vp8_loop_filter_horizontal_edge_uv_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -584,7 +584,7 @@ sym(vp8_mbloop_filter_horizontal_edge_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -634,7 +634,7 @@ sym(vp8_mbloop_filter_horizontal_edge_uv_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -1024,7 +1024,7 @@ sym(vp8_loop_filter_vertical_edge_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -1091,7 +1091,7 @@ sym(vp8_loop_filter_vertical_edge_uv_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -1249,7 +1249,7 @@ sym(vp8_mbloop_filter_vertical_edge_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -1318,7 +1318,7 @@ sym(vp8_mbloop_filter_vertical_edge_uv_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -1386,7 +1386,7 @@ sym(vp8_loop_filter_simple_horizontal_edge_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -1503,7 +1503,7 @@ sym(vp8_loop_filter_simple_vertical_edge_sse2):
     push        rbp         ; save old base pointer value.
     mov         rbp, rsp    ; set new base pointer value.
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx         ; save callee-saved reg
     push        rsi
     push        rdi
diff --git a/vp8/common/x86/postproc_sse2.asm b/vp8/common/x86/postproc_sse2.asm
index 30b4bf53a..06d51ec6f 100644
--- a/vp8/common/x86/postproc_sse2.asm
+++ b/vp8/common/x86/postproc_sse2.asm
@@ -26,7 +26,7 @@ sym(vp8_post_proc_down_and_across_xmm):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 7
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -256,7 +256,7 @@ sym(vp8_mbpost_proc_down_xmm):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 5
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -456,7 +456,7 @@ sym(vp8_mbpost_proc_across_ip_xmm):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 5
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
diff --git a/vp8/common/x86/recon_sse2.asm b/vp8/common/x86/recon_sse2.asm
index 4ad3973ec..67b6420a9 100644
--- a/vp8/common/x86/recon_sse2.asm
+++ b/vp8/common/x86/recon_sse2.asm
@@ -67,7 +67,7 @@ sym(vp8_recon4b_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 4
-    SAVE_XMM
+    SAVE_XMM 7
     push        rsi
     push        rdi
     ; end prolog
diff --git a/vp8/common/x86/subpixel_sse2.asm b/vp8/common/x86/subpixel_sse2.asm
index b87cad259..83e3b1479 100644
--- a/vp8/common/x86/subpixel_sse2.asm
+++ b/vp8/common/x86/subpixel_sse2.asm
@@ -37,7 +37,7 @@ sym(vp8_filter_block1d8_h6_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 7
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -157,7 +157,7 @@ sym(vp8_filter_block1d16_h6_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 7
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -333,7 +333,7 @@ sym(vp8_filter_block1d8_v6_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 8
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -428,7 +428,7 @@ sym(vp8_filter_block1d16_v6_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 8
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -538,7 +538,7 @@ sym(vp8_filter_block1d8_h6_only_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -651,7 +651,7 @@ sym(vp8_filter_block1d16_h6_only_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -816,7 +816,7 @@ sym(vp8_filter_block1d8_v6_only_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -908,7 +908,6 @@ sym(vp8_unpack_block1d16_h6_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 5
-    ;SAVE_XMM                          ;xmm6, xmm7 are not used here.
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -948,7 +947,6 @@ unpack_block1d16_h6_sse2_rowloop:
     pop rdi
     pop rsi
     RESTORE_GOT
-    ;RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
@@ -969,7 +967,7 @@ sym(vp8_bilinear_predict16x16_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -1238,7 +1236,7 @@ sym(vp8_bilinear_predict8x8_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
diff --git a/vp8/common/x86/subpixel_ssse3.asm b/vp8/common/x86/subpixel_ssse3.asm
index 0ec18de76..1ddbc54bd 100644
--- a/vp8/common/x86/subpixel_ssse3.asm
+++ b/vp8/common/x86/subpixel_ssse3.asm
@@ -39,6 +39,7 @@ sym(vp8_filter_block1d8_h6_ssse3):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -107,6 +108,7 @@ filter_block1d8_h6_rowloop_ssse3:
     pop rdi
     pop rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
@@ -162,6 +164,7 @@ filter_block1d8_h4_rowloop_ssse3:
     pop rdi
     pop rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
@@ -179,7 +182,7 @@ sym(vp8_filter_block1d16_h6_ssse3):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -286,6 +289,7 @@ sym(vp8_filter_block1d4_h6_ssse3):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -393,6 +397,7 @@ filter_block1d4_h4_rowloop_ssse3:
     pop rdi
     pop rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
@@ -413,6 +418,7 @@ sym(vp8_filter_block1d16_v6_ssse3):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -508,6 +514,7 @@ vp8_filter_block1d16_v6_ssse3_loop:
     pop rdi
     pop rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
@@ -580,6 +587,7 @@ vp8_filter_block1d16_v4_ssse3_loop:
     pop rdi
     pop rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
@@ -598,6 +606,7 @@ sym(vp8_filter_block1d8_v6_ssse3):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -670,6 +679,7 @@ vp8_filter_block1d8_v6_ssse3_loop:
     pop rdi
     pop rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
@@ -718,6 +728,7 @@ vp8_filter_block1d8_v4_ssse3_loop:
     pop rdi
     pop rsi
     RESTORE_GOT
+    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
@@ -874,7 +885,7 @@ sym(vp8_bilinear_predict16x16_ssse3):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
@@ -1137,7 +1148,7 @@ sym(vp8_bilinear_predict8x8_ssse3):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
-    SAVE_XMM
+    SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi