Fix pulsing issue with scaling

Updates the YV12_BUFFER_CONFIG structure to be crop-aware. The
exiting width/height parameters are left unchanged, storing the
width and height algined to a 16 byte boundary. The cropped
dimensions are added as new fields.

This fixes a nasty visual pulse when switching between scaled and
unscaled frame dimensions due to a mismatch between the scaling
ratio and the 16-byte aligned sizes.

Change-Id: Id4a3f6aea6b9b9ae38bdfa1b87b7eb2cfcdd57b6

3 files changed, 80 insertions, 174 deletions

diff --git a/vpx_scale/generic/yv12config.c b/vpx_scale/generic/yv12config.c
index 267d55f40..fc7f82881 100644
--- a/vpx_scale/generic/yv12config.c
+++ b/vpx_scale/generic/yv12config.c
@@ -38,10 +38,12 @@ vp8_yv12_de_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf) {
 int vp8_yv12_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
                                   int width, int height, int border) {
   if (ybf) {
-    int y_stride = ((width + 2 * border) + 31) & ~31;
-    int yplane_size = (height + 2 * border) * y_stride;
-    int uv_width = width >> 1;
-    int uv_height = height >> 1;
+    int aligned_width = (width + 15) & ~15;
+    int aligned_height = (height + 15) & ~15;
+    int y_stride = ((aligned_width + 2 * border) + 31) & ~31;
+    int yplane_size = (aligned_height + 2 * border) * y_stride;
+    int uv_width = aligned_width >> 1;
+    int uv_height = aligned_height >> 1;
     /** There is currently a bunch of code which assumes
       *  uv_stride == y_stride/2, so enforce this here. */
     int uv_stride = y_stride >> 1;
@@ -56,17 +58,18 @@ int vp8_yv12_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
     if (!ybf->buffer_alloc || ybf->buffer_alloc_sz < frame_size)
       return -1;
 
-    /** Only support allocating buffers that have a height and width that
-      *  are multiples of 16, and a border that's a multiple of 32.
-      * The border restriction is required to get 16-byte alignment of the
-      *  start of the chroma rows without intoducing an arbitrary gap
-      *  between planes, which would break the semantics of things like
-      *  vpx_img_set_rect(). */
-    if ((width & 0xf) | (height & 0xf) | (border & 0x1f))
+    /* Only support allocating buffers that have a border that's a multiple
+     * of 32. The border restriction is required to get 16-byte alignment of
+     * the start of the chroma rows without intoducing an arbitrary gap
+     * between planes, which would break the semantics of things like
+     * vpx_img_set_rect(). */
+    if (border & 0x1f)
       return -3;
 
-    ybf->y_width  = width;
-    ybf->y_height = height;
+    ybf->y_crop_width = width;
+    ybf->y_crop_height = height;
+    ybf->y_width  = aligned_width;
+    ybf->y_height = aligned_height;
     ybf->y_stride = y_stride;
 
     ybf->uv_width = uv_width;
diff --git a/vpx_scale/generic/yv12extend.c b/vpx_scale/generic/yv12extend.c
index d733bd49d..49d7e8e56 100644
--- a/vpx_scale/generic/yv12extend.c
+++ b/vpx_scale/generic/yv12extend.c
@@ -20,180 +20,81 @@
 /****************************************************************************
  *
  ****************************************************************************/
-void
-vp8_yv12_extend_frame_borders_c(YV12_BUFFER_CONFIG *ybf) {
+static void extend_plane(uint8_t *s,       /* source */
+                         int sp,           /* source pitch */
+                         int w,            /* width */
+                         int h,            /* height */
+                         int et,           /* extend top border */
+                         int el,           /* extend left border */
+                         int eb,           /* extend bottom border */
+                         int er) {         /* extend right border */
   int i;
-  unsigned char *src_ptr1, *src_ptr2;
-  unsigned char *dest_ptr1, *dest_ptr2;
-
-  unsigned int Border;
-  int plane_stride;
-  int plane_height;
-  int plane_width;
-
-  /***********/
-  /* Y Plane */
-  /***********/
-  Border = ybf->border;
-  plane_stride = ybf->y_stride;
-  plane_height = ybf->y_height;
-  plane_width = ybf->y_width;
+  uint8_t *src_ptr1, *src_ptr2;
+  uint8_t *dest_ptr1, *dest_ptr2;
+  int linesize;
 
   /* copy the left and right most columns out */
-  src_ptr1 = ybf->y_buffer;
-  src_ptr2 = src_ptr1 + plane_width - 1;
-  dest_ptr1 = src_ptr1 - Border;
-  dest_ptr2 = src_ptr2 + 1;
-
-  for (i = 0; i < plane_height; i++) {
-    vpx_memset(dest_ptr1, src_ptr1[0], Border);
-    vpx_memset(dest_ptr2, src_ptr2[0], Border);
-    src_ptr1  += plane_stride;
-    src_ptr2  += plane_stride;
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
+  src_ptr1 = s;
+  src_ptr2 = s + w - 1;
+  dest_ptr1 = s - el;
+  dest_ptr2 = s + w;
+
+  for (i = 0; i < h; i++) {
+    vpx_memset(dest_ptr1, src_ptr1[0], el);
+    vpx_memset(dest_ptr2, src_ptr2[0], er);
+    src_ptr1  += sp;
+    src_ptr2  += sp;
+    dest_ptr1 += sp;
+    dest_ptr2 += sp;
   }
 
-  /* Now copy the top and bottom source lines into each line of the respective borders */
-  src_ptr1 = ybf->y_buffer - Border;
-  src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride;
-  dest_ptr1 = src_ptr1 - (Border * plane_stride);
-  dest_ptr2 = src_ptr2 + plane_stride;
-
-  for (i = 0; i < (int)Border; i++) {
-    vpx_memcpy(dest_ptr1, src_ptr1, plane_stride);
-    vpx_memcpy(dest_ptr2, src_ptr2, plane_stride);
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
+  /* Now copy the top and bottom lines into each line of the respective
+   * borders
+   */
+  src_ptr1 = s - el;
+  src_ptr2 = s + sp * (h - 1) - el;
+  dest_ptr1 = s + sp * (-et) - el;
+  dest_ptr2 = s + sp * (h) - el;
+  linesize = el + er + w;
+
+  for (i = 0; i < et; i++) {
+    vpx_memcpy(dest_ptr1, src_ptr1, linesize);
+    dest_ptr1 += sp;
   }
 
-
-  /***********/
-  /* U Plane */
-  /***********/
-  plane_stride = ybf->uv_stride;
-  plane_height = ybf->uv_height;
-  plane_width = ybf->uv_width;
-  Border /= 2;
-
-  /* copy the left and right most columns out */
-  src_ptr1 = ybf->u_buffer;
-  src_ptr2 = src_ptr1 + plane_width - 1;
-  dest_ptr1 = src_ptr1 - Border;
-  dest_ptr2 = src_ptr2 + 1;
-
-  for (i = 0; i < plane_height; i++) {
-    vpx_memset(dest_ptr1, src_ptr1[0], Border);
-    vpx_memset(dest_ptr2, src_ptr2[0], Border);
-    src_ptr1  += plane_stride;
-    src_ptr2  += plane_stride;
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  /* Now copy the top and bottom source lines into each line of the respective borders */
-  src_ptr1 = ybf->u_buffer - Border;
-  src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride;
-  dest_ptr1 = src_ptr1 - (Border * plane_stride);
-  dest_ptr2 = src_ptr2 + plane_stride;
-
-  for (i = 0; i < (int)(Border); i++) {
-    vpx_memcpy(dest_ptr1, src_ptr1, plane_stride);
-    vpx_memcpy(dest_ptr2, src_ptr2, plane_stride);
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  /***********/
-  /* V Plane */
-  /***********/
-
-  /* copy the left and right most columns out */
-  src_ptr1 = ybf->v_buffer;
-  src_ptr2 = src_ptr1 + plane_width - 1;
-  dest_ptr1 = src_ptr1 - Border;
-  dest_ptr2 = src_ptr2 + 1;
-
-  for (i = 0; i < plane_height; i++) {
-    vpx_memset(dest_ptr1, src_ptr1[0], Border);
-    vpx_memset(dest_ptr2, src_ptr2[0], Border);
-    src_ptr1  += plane_stride;
-    src_ptr2  += plane_stride;
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  /* Now copy the top and bottom source lines into each line of the respective borders */
-  src_ptr1 = ybf->v_buffer - Border;
-  src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride;
-  dest_ptr1 = src_ptr1 - (Border * plane_stride);
-  dest_ptr2 = src_ptr2 + plane_stride;
-
-  for (i = 0; i < (int)(Border); i++) {
-    vpx_memcpy(dest_ptr1, src_ptr1, plane_stride);
-    vpx_memcpy(dest_ptr2, src_ptr2, plane_stride);
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
+  for (i = 0; i < eb; i++) {
+    vpx_memcpy(dest_ptr2, src_ptr2, linesize);
+    dest_ptr2 += sp;
   }
 }
 
-
-static void
-extend_frame_borders_yonly_c(YV12_BUFFER_CONFIG *ybf) {
-  int i;
-  unsigned char *src_ptr1, *src_ptr2;
-  unsigned char *dest_ptr1, *dest_ptr2;
-
-  unsigned int Border;
-  int plane_stride;
-  int plane_height;
-  int plane_width;
-
-  /***********/
-  /* Y Plane */
-  /***********/
-  Border = ybf->border;
-  plane_stride = ybf->y_stride;
-  plane_height = ybf->y_height;
-  plane_width = ybf->y_width;
-
-  /* copy the left and right most columns out */
-  src_ptr1 = ybf->y_buffer;
-  src_ptr2 = src_ptr1 + plane_width - 1;
-  dest_ptr1 = src_ptr1 - Border;
-  dest_ptr2 = src_ptr2 + 1;
-
-  for (i = 0; i < plane_height; i++) {
-    vpx_memset(dest_ptr1, src_ptr1[0], Border);
-    vpx_memset(dest_ptr2, src_ptr2[0], Border);
-    src_ptr1  += plane_stride;
-    src_ptr2  += plane_stride;
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  /* Now copy the top and bottom source lines into each line of the respective borders */
-  src_ptr1 = ybf->y_buffer - Border;
-  src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride;
-  dest_ptr1 = src_ptr1 - (Border * plane_stride);
-  dest_ptr2 = src_ptr2 + plane_stride;
-
-  for (i = 0; i < (int)Border; i++) {
-    vpx_memcpy(dest_ptr1, src_ptr1, plane_stride);
-    vpx_memcpy(dest_ptr2, src_ptr2, plane_stride);
-    dest_ptr1 += plane_stride;
-    dest_ptr2 += plane_stride;
-  }
-
-  plane_stride /= 2;
-  plane_height /= 2;
-  plane_width /= 2;
-  Border /= 2;
-
+void
+vp8_yv12_extend_frame_borders_c(YV12_BUFFER_CONFIG *ybf) {
+  assert(ybf->y_height - ybf->y_crop_height < 16);
+  assert(ybf->y_width - ybf->y_crop_width < 16);
+  assert(ybf->y_height - ybf->y_crop_height >= 0);
+  assert(ybf->y_width - ybf->y_crop_width >= 0);
+
+  extend_plane(ybf->y_buffer, ybf->y_stride,
+               ybf->y_crop_width, ybf->y_crop_height,
+               ybf->border, ybf->border,
+               ybf->border + ybf->y_height - ybf->y_crop_height,
+               ybf->border + ybf->y_width - ybf->y_crop_width);
+
+  extend_plane(ybf->u_buffer, ybf->uv_stride,
+               (ybf->y_crop_width + 1) / 2, (ybf->y_crop_height + 1) / 2,
+               ybf->border / 2, ybf->border / 2,
+               (ybf->border + ybf->y_height - ybf->y_crop_height + 1) / 2,
+               (ybf->border + ybf->y_width - ybf->y_crop_width + 1) / 2);
+
+  extend_plane(ybf->v_buffer, ybf->uv_stride,
+               (ybf->y_crop_width + 1) / 2, (ybf->y_crop_height + 1) / 2,
+               ybf->border / 2, ybf->border / 2,
+               (ybf->border + ybf->y_height - ybf->y_crop_height + 1) / 2,
+               (ybf->border + ybf->y_width - ybf->y_crop_width + 1) / 2);
 }
 
 
-
 /****************************************************************************
  *
  *  ROUTINE       : vp8_yv12_copy_frame
diff --git a/vpx_scale/yv12config.h b/vpx_scale/yv12config.h
index 45e57f401..14b6e278b 100644
--- a/vpx_scale/yv12config.h
+++ b/vpx_scale/yv12config.h
@@ -42,6 +42,8 @@ extern "C" {
   typedef struct yv12_buffer_config {
     int   y_width;
     int   y_height;
+    int   y_crop_width;
+    int   y_crop_height;
     int   y_stride;
     /*    int   yinternal_width; */