sepia: clobber XMM registers correctly

(cherry picked from commit a865ced4888701e8caf0137672a4857c2b0d47d7)

Conflicts:
	modules/video_filter/sepia.c

modules/video_filter/sepia.c

@@ -206,6 +206,7 @@ static picture_t *Filter( filter_t *p_filter, picture_t *p_pic )
  * instructions. It copies those 8 bytes to 128b register and fills the gaps
  * with zeroes and following operations are made with word-operating instructs.
  *****************************************************************************/
+VLC_SSE
 static inline void Sepia8ySSE2(uint8_t * dst, const uint8_t * src,
                          int i_intensity_spread)
 {
@@ -225,72 +226,51 @@ static inline void Sepia8ySSE2(uint8_t * dst, const uint8_t * src,
         "movq          %%xmm1, (%0)  \n"    // load to dest
         :
         :"r" (dst), "r"(src), "r"(i_intensity_spread)
-        :"memory");
+        :"memory", "xmm1", "xmm2", "xmm3");
 }
-#endif
 
-/*****************************************************************************
- * PlanarI420Sepia: Applies sepia to one frame of the planar I420 video
- *****************************************************************************
- * This function applies sepia effect to one frame of the video by iterating
- * through video lines. We iterate for every two lines and for every two pixels
- * in line to calculate new sepia values for four y components as well for u
- * and v components.
- *****************************************************************************/
-static void PlanarI420Sepia( picture_t *p_pic, picture_t *p_outpic,
+VLC_SSE
+static void PlanarI420SepiaSSE( picture_t *p_pic, picture_t *p_outpic,
                                 int i_intensity )
 {
-    // prepared values to copy for U and V channels
+    /* prepared values to copy for U and V channels */
     const uint8_t filling_const_8u = 128 - i_intensity / 6;
     const uint8_t filling_const_8v = 128 + i_intensity / 14;
-
-#if defined(CAN_COMPILE_SSE2)
-    if (vlc_CPU() & CPU_CAPABILITY_SSE2)
-    {
     /* prepared value for faster broadcasting in xmm register */
     int i_intensity_spread = 0x10001 * (uint8_t) i_intensity;
 
     __asm__ volatile(
         "pxor      %%xmm7, %%xmm7\n"
-        ::);
+        ::: "xmm7");
 
     /* iterate for every two visible line in the frame */
     for (int y = 0; y < p_pic->p[Y_PLANE].i_visible_lines - 1; y += 2)
     {
         const int i_dy_line1_start = y * p_outpic->p[Y_PLANE].i_pitch;
-            const int i_dy_line2_start =
-            (y + 1) * p_outpic->p[Y_PLANE].i_pitch;
-            const int i_du_line_start =
-            (y / 2) * p_outpic->p[U_PLANE].i_pitch;
-            const int i_dv_line_start =
-            (y / 2) * p_outpic->p[V_PLANE].i_pitch;
+        const int i_dy_line2_start = (y + 1) * p_outpic->p[Y_PLANE].i_pitch;
+        const int i_du_line_start =  (y / 2) * p_outpic->p[U_PLANE].i_pitch;
+        const int i_dv_line_start =  (y / 2) * p_outpic->p[V_PLANE].i_pitch;
         int x = 0;
         /* iterate for every visible line in the frame (eight values at once) */
         for ( ; x < p_pic->p[Y_PLANE].i_visible_pitch - 15; x += 16 )
         {
             /* Compute yellow channel values with asm function */
-                Sepia8ySSE2(
-                    &p_outpic->p[Y_PLANE].p_pixels[i_dy_line1_start + x],
+            Sepia8ySSE2(&p_outpic->p[Y_PLANE].p_pixels[i_dy_line1_start + x],
                         &p_pic->p[Y_PLANE].p_pixels[i_dy_line1_start + x],
                         i_intensity_spread );
-                Sepia8ySSE2(
-                    &p_outpic->p[Y_PLANE].p_pixels[i_dy_line2_start + x],
+            Sepia8ySSE2(&p_outpic->p[Y_PLANE].p_pixels[i_dy_line2_start + x],
                         &p_pic->p[Y_PLANE].p_pixels[i_dy_line2_start + x],
                         i_intensity_spread );
-                Sepia8ySSE2(
-                    &p_outpic->p[Y_PLANE].p_pixels[i_dy_line1_start + x + 8],
+            Sepia8ySSE2(&p_outpic->p[Y_PLANE].p_pixels[i_dy_line1_start + x + 8],
                         &p_pic->p[Y_PLANE].p_pixels[i_dy_line1_start + x + 8],
                         i_intensity_spread );
-                Sepia8ySSE2(
-                    &p_outpic->p[Y_PLANE].p_pixels[i_dy_line2_start + x + 8],
+            Sepia8ySSE2(&p_outpic->p[Y_PLANE].p_pixels[i_dy_line2_start + x + 8],
                         &p_pic->p[Y_PLANE].p_pixels[i_dy_line2_start + x + 8],
                         i_intensity_spread );
             /* Copy precomputed values to destination memory location */
-                vlc_memset(
-                    &p_outpic->p[U_PLANE].p_pixels[i_du_line_start + (x / 2)],
+            memset(&p_outpic->p[U_PLANE].p_pixels[i_du_line_start + (x / 2)],
                    filling_const_8u, 8 );
-                vlc_memset(
-                    &p_outpic->p[V_PLANE].p_pixels[i_dv_line_start + (x / 2)],
+            memset(&p_outpic->p[V_PLANE].p_pixels[i_dv_line_start + (x / 2)],
                    filling_const_8v, 8 );
         }
         /* Completing the job, the cycle above takes really big chunks, so
@@ -322,10 +302,29 @@ static void PlanarI420Sepia( picture_t *p_pic, picture_t *p_outpic,
                 filling_const_8v;
         }
     }
-    }
-    else
+}
 #endif
-    {
+
+/*****************************************************************************
+ * PlanarI420Sepia: Applies sepia to one frame of the planar I420 video
+ *****************************************************************************
+ * This function applies sepia effect to one frame of the video by iterating
+ * through video lines. We iterate for every two lines and for every two pixels
+ * in line to calculate new sepia values for four y components as well for u
+ * and v components.
+ *****************************************************************************/
+static void PlanarI420Sepia( picture_t *p_pic, picture_t *p_outpic,
+                               int i_intensity )
+{
+#if defined(CAN_COMPILE_SSE2)
+    if (vlc_CPU() & CPU_CAPABILITY_SSE2)
+        return PlanarI420SepiaSSE( p_pic, p_outpic, i_intensity );
+#endif
+
+    // prepared values to copy for U and V channels
+    const uint8_t filling_const_8u = 128 - i_intensity / 6;
+    const uint8_t filling_const_8v = 128 + i_intensity / 14;
+
     /* iterate for every two visible line in the frame */
     for( int y = 0; y < p_pic->p[Y_PLANE].i_visible_lines - 1; y += 2)
     {
@@ -366,7 +365,6 @@ static void PlanarI420Sepia( picture_t *p_pic, picture_t *p_outpic,
                 filling_const_8v;
         }
     }
-    }
 }
 
 /*****************************************************************************