h264dec: be more explicit in handling container cropping

The current condition can trigger in cases where it shouldn't, with
unexpected results.
Make sure that:
- container cropping is really based on the original dimensions from the
  caller
- those dimenions are discarded on size change

The code is still quite hacky and eventually should be deprecated and
removed, with the decision about which cropping is used delegated to the
caller.

libavcodec/h264_slice.c

@@ -378,6 +378,8 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     h->avctx->coded_width   = h1->avctx->coded_width;
     h->avctx->width         = h1->avctx->width;
     h->avctx->height        = h1->avctx->height;
+    h->width_from_caller    = h1->width_from_caller;
+    h->height_from_caller   = h1->height_from_caller;
     h->coded_picture_number = h1->coded_picture_number;
     h->first_field          = h1->first_field;
     h->picture_structure    = h1->picture_structure;
@@ -797,10 +799,15 @@ static int init_dimensions(H264Context *h)
     int height = h->height - (sps->crop_top   + sps->crop_bottom);
 
     /* handle container cropping */
-    if (FFALIGN(h->avctx->width,  16) == FFALIGN(width,  16) &&
-        FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16)) {
-        width  = h->avctx->width;
-        height = h->avctx->height;
+    if (h->width_from_caller > 0 && h->height_from_caller > 0     &&
+        !sps->crop_top && !sps->crop_left                         &&
+        FFALIGN(h->width_from_caller,  16) == FFALIGN(width,  16) &&
+        FFALIGN(h->height_from_caller, 16) == FFALIGN(height, 16)) {
+        width  = h->width_from_caller;
+        height = h->height_from_caller;
+    } else {
+        h->width_from_caller  = 0;
+        h->height_from_caller = 0;
     }
 
     h->avctx->coded_width  = h->width;

libavcodec/h264dec.c

@@ -285,6 +285,9 @@ static int h264_init_context(AVCodecContext *avctx, H264Context *h)
 
     h->avctx                 = avctx;
 
+    h->width_from_caller     = avctx->width;
+    h->height_from_caller    = avctx->height;
+
     h->picture_structure     = PICT_FRAME;
     h->workaround_bugs       = avctx->workaround_bugs;
     h->flags                 = avctx->flags;

libavcodec/h264dec.h

@@ -514,6 +514,11 @@ typedef struct H264Context {
      * the slice data */
     int field_started;
 
+    /* original AVCodecContext dimensions, used to handle container
+     * cropping */
+    int width_from_caller;
+    int height_from_caller;
+
     AVFrame *output_frame;
 
     int enable_er;