Help msvc 2017 compiler to generate correct code

This loop is wrongly vectorized by the MSVC2017 compiler, this change
is a work-around for the compiler bug.

Change-Id: Ie4c8403965c3e4cd6d70eb3dbc92148f5272f0ab

aom_dsp/sad.c

diff --git a/aom_dsp/sad.c b/aom_dsp/sad.c
index 6adb69c..e500fc4 100644
--- a/aom_dsp/sad.c
+++ b/aom_dsp/sad.c
@@ -341,27 +341,24 @@
 #endif  // CONFIG_HIGHBITDEPTH
 
 #if CONFIG_AV1 && CONFIG_EXT_INTER
-                                    static INLINE
+static INLINE
     unsigned int masked_sad(const uint8_t *src, int src_stride,
                             const uint8_t *a, int a_stride, const uint8_t *b,
                             int b_stride, const uint8_t *m, int m_stride,
                             int width, int height) {
   int y, x;
   unsigned int sad = 0;
-
   for (y = 0; y < height; y++) {
     for (x = 0; x < width; x++) {
-      const uint8_t pred = AOM_BLEND_A64(m[x], a[x], b[x]);
+      const int16_t pred = AOM_BLEND_A64(m[x], a[x], b[x]);
       sad += abs(pred - src[x]);
     }
-
     src += src_stride;
     a += a_stride;
     b += b_stride;
     m += m_stride;
   }
   sad = (sad + 31) >> 6;
-
   return sad;
 }