diff --git a/libavformat/matroskadec.c b/libavformat/matroskadec.c
index 9cf3e956fc..43eb58f46a 100644
--- a/libavformat/matroskadec.c
+++ b/libavformat/matroskadec.c
@@ -1191,7 +1191,7 @@ static int matroska_parse_seekhead_entry(MatroskaDemuxContext *matroska, int idx
             matroska->num_levels++;
             matroska->current_id = 0;
 
-            ebml_parse(matroska, matroska_segment, matroska);
+            ret = ebml_parse(matroska, matroska_segment, matroska);
 
             /* remove dummy level */
             while (matroska->num_levels) {
@@ -1383,7 +1383,7 @@ static int matroska_read_header(AVFormatContext *s, AVFormatParameters *ap)
         }
         if (encodings_list->nb_elem > 1) {
             av_log(matroska->ctx, AV_LOG_ERROR,
-                   "Multiple combined encodings no supported");
+                   "Multiple combined encodings not supported");
         } else if (encodings_list->nb_elem == 1) {
             if (encodings[0].type ||
                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
diff --git a/libswscale/utils.c b/libswscale/utils.c
index c112fa3264..8b793f7f84 100644
--- a/libswscale/utils.c
+++ b/libswscale/utils.c
@@ -186,8 +186,8 @@ static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSi
 
     emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
 
-    // NOTE: the +1 is for the MMX scaler which reads over the end
-    FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
+    // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
+    FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
 
     if (FFABS(xInc - 0x10000) <10) { // unscaled
         int i;
@@ -473,7 +473,7 @@ static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSi
 
     // Note the +1 is for the MMX scaler which reads over the end
     /* align at 16 for AltiVec (needed by hScale_altivec_real) */
-    FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
+    FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
 
     /* normalize & store in outFilter */
     for (i=0; i<dstW; i++) {
@@ -493,10 +493,14 @@ static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSi
         }
     }
 
-    (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
+    (*filterPos)[dstW+0] =
+    (*filterPos)[dstW+1] =
+    (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
     for (i=0; i<*outFilterSize; i++) {
-        int j= dstW*(*outFilterSize);
-        (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
+        int k= (dstW - 1) * (*outFilterSize) + i;
+        (*outFilter)[k + 1 * (*outFilterSize)] =
+        (*outFilter)[k + 2 * (*outFilterSize)] =
+        (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
     }
 
     ret=0;