swscale: extend mmx padding.

Fixes a crash when forcing libc to strictly adhere to malloc sizes.

libswscale/utils.c

@@ -1049,19 +1049,20 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
     //Note we need at least one pixel more at the end because of the MMX code (just in case someone wanna replace the 4000/8000)
     /* align at 16 bytes for AltiVec */
     for (i=0; i<c->vLumBufSize; i++) {
-        FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
+        FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
     }
-    c->uv_off_px   = dst_stride_px;
-    c->uv_off_byte = dst_stride;
+    // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
+    c->uv_off_px   = dst_stride_px + 64 / c->scalingBpp;
+    c->uv_off_byte = dst_stride + 16;
     for (i=0; i<c->vChrBufSize; i++) {
-        FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+1, fail);
+        FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
-        c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1);
+        c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
     }
     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
         for (i=0; i<c->vLumBufSize; i++) {
-            FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
+            FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
         }