Miscellaneous decoder changes for high bitdepth

Also includes yv12 config changes.

Change-Id: Iacf40d8bf486815b54c32a127ce3cd4516b7e44f

diff --git a/tools_common.c b/tools_common.c
index 2ec1711..0d4774f 100644 (file)
--- a/tools_common.c
+++ b/tools_common.c
@@ -224,7 +224,8 @@ void vpx_img_write(const vpx_image_t *img, FILE *file) {
   for (plane = 0; plane < 3; ++plane) {
     const unsigned char *buf = img->planes[plane];
     const int stride = img->stride[plane];
-    const int w = vpx_img_plane_width(img, plane);
+    const int w = vpx_img_plane_width(img, plane) *
+        ((img->fmt & VPX_IMG_FMT_HIGHBITDEPTH) ? 2 : 1);
     const int h = vpx_img_plane_height(img, plane);
     int y;
 

diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h
index 792e9d9..3954c45 100644 (file)
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -180,6 +180,7 @@ typedef struct VP9Common {
 
   // VPX_BITS_8 in profile 0 or 1, VPX_BITS_10 or VPX_BITS_12 in profile 2 or 3.
   vpx_bit_depth_t bit_depth;
+  vpx_bit_depth_t dequant_bit_depth;  // bit_depth of current dequantizer
 
 #if CONFIG_VP9_POSTPROC
   struct postproc_state  postproc_state;

diff --git a/vp9/common/vp9_reconinter.c b/vp9/common/vp9_reconinter.c
index 9c4e323..1a46b39 100644 (file)
--- a/vp9/common/vp9_reconinter.c
+++ b/vp9/common/vp9_reconinter.c
@@ -467,13 +467,13 @@ static void dec_build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
       int y1 = ((y0_16 + (h - 1) * ys) >> SUBPEL_BITS) + 1;
       int x_pad = 0, y_pad = 0;
 
-      if (subpel_x || (sf->x_step_q4 != 16)) {
+      if (subpel_x || (sf->x_step_q4 != SUBPEL_SHIFTS)) {
         x0 -= VP9_INTERP_EXTEND - 1;
         x1 += VP9_INTERP_EXTEND;
         x_pad = 1;
       }
 
-      if (subpel_y || (sf->y_step_q4 != 16)) {
+      if (subpel_y || (sf->y_step_q4 != SUBPEL_SHIFTS)) {
         y0 -= VP9_INTERP_EXTEND - 1;
         y1 += VP9_INTERP_EXTEND;
         y_pad = 1;

diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index 0ec2710..db55de1 100644 (file)
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -652,8 +652,10 @@ static void setup_quantization(VP9_COMMON *const cm, MACROBLOCKD *const xd,
   update |= read_delta_q(rb, &cm->y_dc_delta_q);
   update |= read_delta_q(rb, &cm->uv_dc_delta_q);
   update |= read_delta_q(rb, &cm->uv_ac_delta_q);
-  if (update)
+  if (update || cm->bit_depth != cm->dequant_bit_depth) {
     vp9_init_dequantizer(cm);
+    cm->dequant_bit_depth = cm->bit_depth;
+  }
 
   xd->lossless = cm->base_qindex == 0 &&
                  cm->y_dc_delta_q == 0 &&

diff --git a/vp9/decoder/vp9_decoder.c b/vp9/decoder/vp9_decoder.c
index 6ee3d70..b7e1a00 100644 (file)
--- a/vp9/decoder/vp9_decoder.c
+++ b/vp9/decoder/vp9_decoder.c
@@ -69,6 +69,7 @@ VP9Decoder *vp9_decoder_create() {
   cm->current_video_frame = 0;
   pbi->ready_for_new_data = 1;
   cm->bit_depth = VPX_BITS_8;
+  cm->dequant_bit_depth = VPX_BITS_8;
 
   // vp9_init_dequantizer() is first called here. Add check in
   // frame_init_dequantizer() to avoid unnecessary calling of

diff --git a/vp9/vp9_dx_iface.c b/vp9/vp9_dx_iface.c
index 393c66e..85e32d3 100644 (file)
--- a/vp9/vp9_dx_iface.c
+++ b/vp9/vp9_dx_iface.c
@@ -437,7 +437,6 @@ static vpx_image_t *decoder_get_frame(vpx_codec_alg_priv_t *ctx,
     // call to get_frame.
     if (!(*iter)) {
       img = &ctx->img;
-      img->bit_depth = (int)ctx->pbi->common.bit_depth;
       *iter = img;
     }
   }

diff --git a/vp9/vp9_iface_common.h b/vp9/vp9_iface_common.h
index fc98b62..0398b40 100644 (file)
--- a/vp9/vp9_iface_common.h
+++ b/vp9/vp9_iface_common.h
@@ -46,6 +46,22 @@ static void yuvconfig2image(vpx_image_t *img, const YV12_BUFFER_CONFIG  *yv12,
   img->stride[VPX_PLANE_U] = yv12->uv_stride;
   img->stride[VPX_PLANE_V] = yv12->uv_stride;
   img->stride[VPX_PLANE_ALPHA] = yv12->y_stride;
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (yv12->flags & YV12_FLAG_HIGHBITDEPTH) {
+    // vpx_image_t uses byte strides and a pointer to the first byte
+    // of the image.
+    img->fmt |= VPX_IMG_FMT_HIGHBITDEPTH;
+    img->bit_depth = yv12->bit_depth;
+    img->planes[VPX_PLANE_Y] = (uint8_t*)CONVERT_TO_SHORTPTR(yv12->y_buffer);
+    img->planes[VPX_PLANE_U] = (uint8_t*)CONVERT_TO_SHORTPTR(yv12->u_buffer);
+    img->planes[VPX_PLANE_V] = (uint8_t*)CONVERT_TO_SHORTPTR(yv12->v_buffer);
+    img->planes[VPX_PLANE_ALPHA] = NULL;
+    img->stride[VPX_PLANE_Y] = 2 * yv12->y_stride;
+    img->stride[VPX_PLANE_U] = 2 * yv12->uv_stride;
+    img->stride[VPX_PLANE_V] = 2 * yv12->uv_stride;
+    img->stride[VPX_PLANE_ALPHA] = 2 * yv12->y_stride;
+  }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
   img->bps = bps;
   img->user_priv = user_priv;
   img->img_data = yv12->buffer_alloc;
@@ -72,6 +88,32 @@ static vpx_codec_err_t image2yuvconfig(const vpx_image_t *img,
   yv12->y_stride = img->stride[VPX_PLANE_Y];
   yv12->uv_stride = img->stride[VPX_PLANE_U];
 
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (img->fmt & VPX_IMG_FMT_HIGHBITDEPTH) {
+    // In vpx_image_t
+    //     planes point to uint8 address of start of data
+    //     stride counts uint8s to reach next row
+    // In YV12_BUFFER_CONFIG
+    //     y_buffer, u_buffer, v_buffer point to uint16 address of data
+    //     stride and border counts in uint16s
+    // This means that all the address calculations in the main body of code
+    // should work correctly.
+    // However, before we do any pixel operations we need to cast the address
+    // to a uint16 ponter and double its value.
+    yv12->y_buffer = CONVERT_TO_BYTEPTR(yv12->y_buffer);
+    yv12->u_buffer = CONVERT_TO_BYTEPTR(yv12->u_buffer);
+    yv12->v_buffer = CONVERT_TO_BYTEPTR(yv12->v_buffer);
+    yv12->y_stride >>= 1;
+    yv12->uv_stride >>= 1;
+    yv12->flags = YV12_FLAG_HIGHBITDEPTH;
+  } else {
+    yv12->flags = 0;
+  }
+  yv12->border  = (yv12->y_stride - img->w) / 2;
+#else
+  yv12->border  = (img->stride[VPX_PLANE_Y] - img->w) / 2;
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
   yv12->border  = (img->stride[VPX_PLANE_Y] - img->w) / 2;
   return VPX_CODEC_OK;
 }

diff --git a/vpxdec.c b/vpxdec.c
index cf23c29..f5c945c 100644 (file)
--- a/vpxdec.c
+++ b/vpxdec.c
@@ -551,8 +551,8 @@ static void high_img_upshift(vpx_image_t *dst, vpx_image_t *src,
     int h = src->d_h;
     int x, y;
     if (plane) {
-      w >>= src->x_chroma_shift;
-      h >>= src->y_chroma_shift;
+      w = (w + src->x_chroma_shift) >> src->x_chroma_shift;
+      h = (h + src->y_chroma_shift) >> src->y_chroma_shift;
     }
     for (y = 0; y < h; y++) {
       uint16_t *p_src = (uint16_t *)(src->planes[plane] +
@@ -590,8 +590,8 @@ static void low_img_upshift(vpx_image_t *dst, vpx_image_t *src,
     int h = src->d_h;
     int x, y;
     if (plane) {
-      w >>= src->x_chroma_shift;
-      h >>= src->y_chroma_shift;
+      w = (w + src->x_chroma_shift) >> src->x_chroma_shift;
+      h = (h + src->y_chroma_shift) >> src->y_chroma_shift;
     }
     for (y = 0; y < h; y++) {
       uint8_t *p_src = src->planes[plane] + y * src->stride[plane];
@@ -636,8 +636,8 @@ static void high_img_downshift(vpx_image_t *dst, vpx_image_t *src,
     int h = src->d_h;
     int x, y;
     if (plane) {
-      w >>= src->x_chroma_shift;
-      h >>= src->y_chroma_shift;
+      w = (w + src->x_chroma_shift) >> src->x_chroma_shift;
+      h = (h + src->y_chroma_shift) >> src->y_chroma_shift;
     }
     for (y = 0; y < h; y++) {
       uint16_t *p_src = (uint16_t *)(src->planes[plane] +
@@ -674,8 +674,8 @@ static void low_img_downshift(vpx_image_t *dst, vpx_image_t *src,
     int h = src->d_h;
     int x, y;
     if (plane) {
-      w >>= src->x_chroma_shift;
-      h >>= src->y_chroma_shift;
+      w = (w + src->x_chroma_shift) >> src->x_chroma_shift;
+      h = (h + src->y_chroma_shift) >> src->y_chroma_shift;
     }
     for (y = 0; y < h; y++) {
       uint16_t *p_src = (uint16_t *)(src->planes[plane] +
@@ -696,6 +696,14 @@ static void img_downshift(vpx_image_t *dst, vpx_image_t *src,
     low_img_downshift(dst, src, down_shift);
   }
 }
+
+static int img_shifted_realloc_required(const vpx_image_t *img,
+                                        const vpx_image_t *shifted,
+                                        vpx_img_fmt_t required_fmt) {
+  return img->d_w != shifted->d_w ||
+         img->d_h != shifted->d_h ||
+         required_fmt != shifted->fmt;
+}
 #endif
 
 int main_loop(int argc, const char **argv_) {
@@ -1130,16 +1138,17 @@ int main_loop(int argc, const char **argv_) {
       }
       // Shift up or down if necessary
       if (output_bit_depth != img->bit_depth) {
+        const vpx_img_fmt_t shifted_fmt = output_bit_depth == 8 ?
+            img->fmt ^ (img->fmt & VPX_IMG_FMT_HIGHBITDEPTH) :
+            img->fmt | VPX_IMG_FMT_HIGHBITDEPTH;
+        if (img_shifted &&
+            img_shifted_realloc_required(img, img_shifted, shifted_fmt)) {
+          vpx_img_free(img_shifted);
+          img_shifted = NULL;
+        }
         if (!img_shifted) {
-          if (output_bit_depth == 8) {
-            img_shifted = vpx_img_alloc(
-                NULL, img->fmt - VPX_IMG_FMT_HIGHBITDEPTH,
-                img->d_w, img->d_h, 16);
-          } else {
-            img_shifted = vpx_img_alloc(
-                NULL, img->fmt | VPX_IMG_FMT_HIGHBITDEPTH,
-                img->d_w, img->d_h, 16);
-          }
+          img_shifted = vpx_img_alloc(NULL, shifted_fmt,
+                                      img->d_w, img->d_h, 16);
           img_shifted->bit_depth = output_bit_depth;
         }
         if (output_bit_depth > img->bit_depth) {