libvpx: Pull from upstream

source/libvpx/vp9/decoder/vp9_decodeframe.c

 /*
  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 178 matching lines @@
     xd->plane[i].dequant = cm->uv_dequant[q_index];
 }
 
 static void inverse_transform_block(MACROBLOCKD* xd, int plane, int block,
                                     TX_SIZE tx_size, uint8_t *dst, int stride,
                                     int eob) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
   if (eob > 0) {
     TX_TYPE tx_type = DCT_DCT;
     tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+#if CONFIG_VP9_HIGHBITDEPTH
+    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+      if (xd->lossless) {
+        tx_type = DCT_DCT;
+        vp9_highbd_iwht4x4_add(dqcoeff, dst, stride, eob, xd->bd);
+      } else {
+        const PLANE_TYPE plane_type = pd->plane_type;
+        switch (tx_size) {
+          case TX_4X4:
+            tx_type = get_tx_type_4x4(plane_type, xd, block);
+            vp9_highbd_iht4x4_add(tx_type, dqcoeff, dst, stride, eob, xd->bd);
+            break;
+          case TX_8X8:
+            tx_type = get_tx_type(plane_type, xd);
+            vp9_highbd_iht8x8_add(tx_type, dqcoeff, dst, stride, eob, xd->bd);
+            break;
+          case TX_16X16:
+            tx_type = get_tx_type(plane_type, xd);
+            vp9_highbd_iht16x16_add(tx_type, dqcoeff, dst, stride, eob, xd->bd);
+            break;
+          case TX_32X32:
+            tx_type = DCT_DCT;
+            vp9_highbd_idct32x32_add(dqcoeff, dst, stride, eob, xd->bd);
+            break;
+          default:
+            assert(0 && "Invalid transform size");
+        }
+      }
+    } else {
+      if (xd->lossless) {
+        tx_type = DCT_DCT;
+        vp9_iwht4x4_add(dqcoeff, dst, stride, eob);
+      } else {
+        const PLANE_TYPE plane_type = pd->plane_type;
+        switch (tx_size) {
+          case TX_4X4:
+            tx_type = get_tx_type_4x4(plane_type, xd, block);
+            vp9_iht4x4_add(tx_type, dqcoeff, dst, stride, eob);
+            break;
+          case TX_8X8:
+            tx_type = get_tx_type(plane_type, xd);
+            vp9_iht8x8_add(tx_type, dqcoeff, dst, stride, eob);
+            break;
+          case TX_16X16:
+            tx_type = get_tx_type(plane_type, xd);
+            vp9_iht16x16_add(tx_type, dqcoeff, dst, stride, eob);
+            break;
+          case TX_32X32:
+            tx_type = DCT_DCT;
+            vp9_idct32x32_add(dqcoeff, dst, stride, eob);
+            break;
+          default:
+            assert(0 && "Invalid transform size");
+            return;
+        }
+      }
+    }
+#else
     if (xd->lossless) {
       tx_type = DCT_DCT;
       vp9_iwht4x4_add(dqcoeff, dst, stride, eob);
     } else {
       const PLANE_TYPE plane_type = pd->plane_type;
       switch (tx_size) {
         case TX_4X4:
           tx_type = get_tx_type_4x4(plane_type, xd, block);
           vp9_iht4x4_add(tx_type, dqcoeff, dst, stride, eob);
           break;
         case TX_8X8:
           tx_type = get_tx_type(plane_type, xd);
           vp9_iht8x8_add(tx_type, dqcoeff, dst, stride, eob);
           break;
         case TX_16X16:
           tx_type = get_tx_type(plane_type, xd);
           vp9_iht16x16_add(tx_type, dqcoeff, dst, stride, eob);
           break;
         case TX_32X32:
           tx_type = DCT_DCT;
           vp9_idct32x32_add(dqcoeff, dst, stride, eob);
           break;
         default:
           assert(0 && "Invalid transform size");
+          return;
       }
     }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
     if (eob == 1) {
       vpx_memset(dqcoeff, 0, 2 * sizeof(dqcoeff[0]));
     } else {
       if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10)
         vpx_memset(dqcoeff, 0, 4 * (4 << tx_size) * sizeof(dqcoeff[0]));
       else if (tx_size == TX_32X32 && eob <= 34)
         vpx_memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0]));
       else
         vpx_memset(dqcoeff, 0, (16 << (tx_size << 1)) * sizeof(dqcoeff[0]));
@@ skipping 16 matching lines @@
   struct macroblockd_plane *const pd = &xd->plane[plane];
   MODE_INFO *const mi = xd->mi[0].src_mi;
   const PREDICTION_MODE mode = (plane == 0) ? get_y_mode(mi, block)
                                             : mi->mbmi.uv_mode;
   int x, y;
   uint8_t *dst;
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y);
   dst = &pd->dst.buf[4 * y * pd->dst.stride + 4 * x];
 
   vp9_predict_intra_block(xd, block >> (tx_size << 1),
-                          b_width_log2(plane_bsize), tx_size, mode,
+                          b_width_log2_lookup[plane_bsize], tx_size, mode,
                           dst, pd->dst.stride, dst, pd->dst.stride,
                           x, y, plane);
 
   if (!mi->mbmi.skip) {
     const int eob = vp9_decode_block_tokens(cm, xd, plane, block,
                                             plane_bsize, x, y, tx_size,
                                             args->r);
     inverse_transform_block(xd, plane, block, tx_size, dst, pd->dst.stride,
                             eob);
   }
@@ skipping 45 matching lines @@
   set_skip_context(xd, mi_row, mi_col);
 
   // Distance of Mb to the various image edges. These are specified to 8th pel
   // as they are always compared to values that are in 1/8th pel units
   set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
 
   vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
   return &xd->mi[0].mbmi;
 }
 
-static void set_ref(VP9_COMMON *const cm, MACROBLOCKD *const xd,
-                    int idx, int mi_row, int mi_col) {
-  MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
-  RefBuffer *ref_buffer = &cm->frame_refs[mbmi->ref_frame[idx] - LAST_FRAME];
-  xd->block_refs[idx] = ref_buffer;
-  if (!vp9_is_valid_scale(&ref_buffer->sf))
-    vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
-                       "Invalid scale factors");
-  if (ref_buffer->buf->corrupted)
-    vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
-                       "Block reference is corrupt");
-  vp9_setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col,
-                       &ref_buffer->sf);
-  xd->corrupted |= ref_buffer->buf->corrupted;
-}
-
 static void decode_block(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                          const TileInfo *const tile,
                          int mi_row, int mi_col,
                          vp9_reader *r, BLOCK_SIZE bsize) {
   const int less8x8 = bsize < BLOCK_8X8;
   MB_MODE_INFO *mbmi = set_offsets(cm, xd, tile, bsize, mi_row, mi_col);
   vp9_read_mode_info(cm, xd, tile, mi_row, mi_col, r);
 
   if (less8x8)
     bsize = BLOCK_8X8;
 
   if (mbmi->skip) {
     reset_skip_context(xd, bsize);
   } else {
     if (cm->seg.enabled)
       setup_plane_dequants(cm, xd, vp9_get_qindex(&cm->seg, mbmi->segment_id,
                                                   cm->base_qindex));
   }
 
   if (!is_inter_block(mbmi)) {
     struct intra_args arg = { cm, xd, r };
     vp9_foreach_transformed_block(xd, bsize,
                                   predict_and_reconstruct_intra_block, &arg);
   } else {
-    // Setup
-    set_ref(cm, xd, 0, mi_row, mi_col);
-    if (has_second_ref(mbmi))
-      set_ref(cm, xd, 1, mi_row, mi_col);
-
     // Prediction
     vp9_dec_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
 
     // Reconstruction
     if (!mbmi->skip) {
       int eobtotal = 0;
       struct inter_args arg = { cm, xd, r, &eobtotal };
       vp9_foreach_transformed_block(xd, bsize, reconstruct_inter_block, &arg);
       if (!less8x8 && eobtotal == 0)
         mbmi->skip = 1;  // skip loopfilter
@@ skipping 203 matching lines @@
 }
 
 static void setup_quantization(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                                struct vp9_read_bit_buffer *rb) {
   int update = 0;
 
   cm->base_qindex = vp9_rb_read_literal(rb, QINDEX_BITS);
   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 &&
                  cm->uv_dc_delta_q == 0 &&
                  cm->uv_ac_delta_q == 0;
+#if CONFIG_VP9_HIGHBITDEPTH
+  xd->bd = (int)cm->bit_depth;
+#endif
 }
 
 static INTERP_FILTER read_interp_filter(struct vp9_read_bit_buffer *rb) {
   const INTERP_FILTER literal_to_filter[] = { EIGHTTAP_SMOOTH,
                                               EIGHTTAP,
                                               EIGHTTAP_SHARP,
                                               BILINEAR };
   return vp9_rb_read_bit(rb) ? SWITCHABLE
                              : literal_to_filter[vp9_rb_read_literal(rb, 2)];
 }
 
 void vp9_read_frame_size(struct vp9_read_bit_buffer *rb,
                          int *width, int *height) {
-  const int w = vp9_rb_read_literal(rb, 16) + 1;
-  const int h = vp9_rb_read_literal(rb, 16) + 1;
-  *width = w;
-  *height = h;
+  *width = vp9_rb_read_literal(rb, 16) + 1;
+  *height = vp9_rb_read_literal(rb, 16) + 1;
 }
 
 static void setup_display_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
   cm->display_width = cm->width;
   cm->display_height = cm->height;
   if (vp9_rb_read_bit(rb))
     vp9_read_frame_size(rb, &cm->display_width, &cm->display_height);
 }
 
 static void resize_context_buffers(VP9_COMMON *cm, int width, int height) {
@@ skipping 34 matching lines @@
           cm->subsampling_x, cm->subsampling_y,
 #if CONFIG_VP9_HIGHBITDEPTH
           cm->use_highbitdepth,
 #endif
           VP9_DEC_BORDER_IN_PIXELS,
           &cm->frame_bufs[cm->new_fb_idx].raw_frame_buffer, cm->get_fb_cb,
           cm->cb_priv)) {
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate frame buffer");
   }
+  cm->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x;
+  cm->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y;
   cm->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
 }
 
 static INLINE int valid_ref_frame_img_fmt(vpx_bit_depth_t ref_bit_depth,
                                           int ref_xss, int ref_yss,
                                           vpx_bit_depth_t this_bit_depth,
                                           int this_xss, int this_yss) {
   return ref_bit_depth == this_bit_depth && ref_xss == this_xss &&
          ref_yss == this_yss;
 }
@@ skipping 13 matching lines @@
                            "Frame reference is corrupt");
       }
       found = 1;
       break;
     }
   }
 
   if (!found)
     vp9_read_frame_size(rb, &width, &height);
 
-  if (width <=0 || height <= 0)
+  if (width <= 0 || height <= 0)
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Invalid frame size");
 
   // Check to make sure at least one of frames that this frame references
   // has valid dimensions.
   for (i = 0; i < REFS_PER_FRAME; ++i) {
     RefBuffer *const ref_frame = &cm->frame_refs[i];
     has_valid_ref_frame |= valid_ref_frame_size(ref_frame->buf->y_crop_width,
                                                 ref_frame->buf->y_crop_height,
                                                 width, height);
   }
   if (!has_valid_ref_frame)
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Referenced frame has invalid size");
   for (i = 0; i < REFS_PER_FRAME; ++i) {
     RefBuffer *const ref_frame = &cm->frame_refs[i];
     if (!valid_ref_frame_img_fmt(
             ref_frame->buf->bit_depth,
-            ref_frame->buf->uv_crop_width < ref_frame->buf->y_crop_width,
-            ref_frame->buf->uv_crop_height < ref_frame->buf->y_crop_height,
+            ref_frame->buf->subsampling_x,
+            ref_frame->buf->subsampling_y,
             cm->bit_depth,
             cm->subsampling_x,
             cm->subsampling_y))
       vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                          "Referenced frame has incompatible color space");
   }
 
   resize_context_buffers(cm, width, height);
   setup_display_size(cm, rb);
 
   if (vp9_realloc_frame_buffer(
           get_frame_new_buffer(cm), cm->width, cm->height,
           cm->subsampling_x, cm->subsampling_y,
 #if CONFIG_VP9_HIGHBITDEPTH
           cm->use_highbitdepth,
 #endif
           VP9_DEC_BORDER_IN_PIXELS,
           &cm->frame_bufs[cm->new_fb_idx].raw_frame_buffer, cm->get_fb_cb,
           cm->cb_priv)) {
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate frame buffer");
   }
+  cm->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x;
+  cm->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y;
   cm->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth;
 }
 
 static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
   int min_log2_tile_cols, max_log2_tile_cols, max_ones;
   vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
 
   // columns
   max_ones = max_log2_tile_cols - min_log2_tile_cols;
   cm->log2_tile_cols = min_log2_tile_cols;
@@ skipping 373 matching lines @@
 BITSTREAM_PROFILE vp9_read_profile(struct vp9_read_bit_buffer *rb) {
   int profile = vp9_rb_read_bit(rb);
   profile |= vp9_rb_read_bit(rb) << 1;
   if (profile > 2)
     profile += vp9_rb_read_bit(rb);
   return (BITSTREAM_PROFILE) profile;
 }
 
 static void read_bitdepth_colorspace_sampling(
     VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
-  if (cm->profile >= PROFILE_2)
+  if (cm->profile >= PROFILE_2) {
     cm->bit_depth = vp9_rb_read_bit(rb) ? VPX_BITS_12 : VPX_BITS_10;
+#if CONFIG_VP9_HIGHBITDEPTH
+    cm->use_highbitdepth = 1;
+#endif
+  } else {
+    cm->bit_depth = VPX_BITS_8;
+#if CONFIG_VP9_HIGHBITDEPTH
+    cm->use_highbitdepth = 0;
+#endif
+  }
   cm->color_space = (COLOR_SPACE)vp9_rb_read_literal(rb, 3);
   if (cm->color_space != SRGB) {
     vp9_rb_read_bit(rb);  // [16,235] (including xvycc) vs [0,255] range
     if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) {
       cm->subsampling_x = vp9_rb_read_bit(rb);
       cm->subsampling_y = vp9_rb_read_bit(rb);
       if (cm->subsampling_x == 1 && cm->subsampling_y == 1)
         vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
                            "4:2:0 color not supported in profile 1 or 3");
       if (vp9_rb_read_bit(rb))
@@ skipping 83 matching lines @@
                            "Invalid frame sync code");
       if (cm->profile > PROFILE_0) {
         read_bitdepth_colorspace_sampling(cm, rb);
       } else {
         // NOTE: The intra-only frame header does not include the specification
         // of either the color format or color sub-sampling in profile 0. VP9
         // specifies that the default color space should be YUV 4:2:0 in this
         // case (normative).
         cm->color_space = BT_601;
         cm->subsampling_y = cm->subsampling_x = 1;
+        cm->bit_depth = VPX_BITS_8;
+#if CONFIG_VP9_HIGHBITDEPTH
+        cm->use_highbitdepth = 0;
+#endif
       }
 
       pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
       setup_frame_size(cm, rb);
       pbi->need_resync = 0;
     } else {
       pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
       for (i = 0; i < REFS_PER_FRAME; ++i) {
         const int ref = vp9_rb_read_literal(rb, REF_FRAMES_LOG2);
         const int idx = cm->ref_frame_map[ref];
@@ skipping 20 matching lines @@
         vp9_setup_scale_factors_for_frame(&ref_buf->sf,
                                           ref_buf->buf->y_crop_width,
                                           ref_buf->buf->y_crop_height,
                                           cm->width, cm->height);
 #endif
         if (vp9_is_scaled(&ref_buf->sf))
           vp9_extend_frame_borders(ref_buf->buf);
       }
     }
   }
+#if CONFIG_VP9_HIGHBITDEPTH
+  get_frame_new_buffer(cm)->bit_depth = cm->bit_depth;
+#endif
 
   if (pbi->need_resync) {
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Keyframe / intra-only frame required to reset decoder"
                        " state");
   }
 
   if (!cm->error_resilient_mode) {
     cm->refresh_frame_context = vp9_rb_read_bit(rb);
     cm->frame_parallel_decoding_mode = vp9_rb_read_bit(rb);
@@ skipping 216 matching lines @@
       debug_check_frame_counts(cm);
     }
   } else {
     vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                        "Decode failed. Frame data is corrupted.");
   }
 
   if (cm->refresh_frame_context)
     cm->frame_contexts[cm->frame_context_idx] = cm->fc;
 }