1 /*
2  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3  *
4  *  Use of this source code is governed by a BSD-style license
5  *  that can be found in the LICENSE file in the root of the source
6  *  tree. An additional intellectual property rights grant can be found
7  *  in the file PATENTS.  All contributing project authors may
8  *  be found in the AUTHORS file in the root of the source tree.
9  */
10 
11 #include <assert.h>
12 
13 #include "./vpx_scale_rtcd.h"
14 #include "./vpx_config.h"
15 
16 #include "vpx/vpx_integer.h"
17 
18 #include "vp9/common/vp9_blockd.h"
19 #include "vp9/common/vp9_reconinter.h"
20 #include "vp9/common/vp9_reconintra.h"
21 
22 #if CONFIG_VP9_HIGHBITDEPTH
high_inter_predictor(const uint8_t * src,int src_stride,uint8_t * dst,int dst_stride,const int subpel_x,const int subpel_y,const struct scale_factors * sf,int w,int h,int ref,const InterpKernel * kernel,int xs,int ys,int bd)23 void high_inter_predictor(const uint8_t *src, int src_stride,
24                                  uint8_t *dst, int dst_stride,
25                                  const int subpel_x,
26                                  const int subpel_y,
27                                  const struct scale_factors *sf,
28                                  int w, int h, int ref,
29                                  const InterpKernel *kernel,
30                                  int xs, int ys, int bd) {
31   sf->highbd_predict[subpel_x != 0][subpel_y != 0][ref](
32       src, src_stride, dst, dst_stride,
33       kernel[subpel_x], xs, kernel[subpel_y], ys, w, h, bd);
34 }
35 
vp9_highbd_build_inter_predictor(const uint8_t * src,int src_stride,uint8_t * dst,int dst_stride,const MV * src_mv,const struct scale_factors * sf,int w,int h,int ref,const InterpKernel * kernel,enum mv_precision precision,int x,int y,int bd)36 void vp9_highbd_build_inter_predictor(const uint8_t *src, int src_stride,
37                                       uint8_t *dst, int dst_stride,
38                                       const MV *src_mv,
39                                       const struct scale_factors *sf,
40                                       int w, int h, int ref,
41                                       const InterpKernel *kernel,
42                                       enum mv_precision precision,
43                                       int x, int y, int bd) {
44   const int is_q4 = precision == MV_PRECISION_Q4;
45   const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
46                      is_q4 ? src_mv->col : src_mv->col * 2 };
47   MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf);
48   const int subpel_x = mv.col & SUBPEL_MASK;
49   const int subpel_y = mv.row & SUBPEL_MASK;
50 
51   src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);
52 
53   high_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
54                        sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4, bd);
55 }
56 #endif  // CONFIG_VP9_HIGHBITDEPTH
57 
vp9_build_inter_predictor(const uint8_t * src,int src_stride,uint8_t * dst,int dst_stride,const MV * src_mv,const struct scale_factors * sf,int w,int h,int ref,const InterpKernel * kernel,enum mv_precision precision,int x,int y)58 void vp9_build_inter_predictor(const uint8_t *src, int src_stride,
59                                uint8_t *dst, int dst_stride,
60                                const MV *src_mv,
61                                const struct scale_factors *sf,
62                                int w, int h, int ref,
63                                const InterpKernel *kernel,
64                                enum mv_precision precision,
65                                int x, int y) {
66   const int is_q4 = precision == MV_PRECISION_Q4;
67   const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
68                      is_q4 ? src_mv->col : src_mv->col * 2 };
69   MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf);
70   const int subpel_x = mv.col & SUBPEL_MASK;
71   const int subpel_y = mv.row & SUBPEL_MASK;
72 
73   src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);
74 
75   inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
76                   sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4);
77 }
78 
round_mv_comp_q4(int value)79 static INLINE int round_mv_comp_q4(int value) {
80   return (value < 0 ? value - 2 : value + 2) / 4;
81 }
82 
mi_mv_pred_q4(const MODE_INFO * mi,int idx)83 static MV mi_mv_pred_q4(const MODE_INFO *mi, int idx) {
84   MV res = { round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.row +
85                               mi->bmi[1].as_mv[idx].as_mv.row +
86                               mi->bmi[2].as_mv[idx].as_mv.row +
87                               mi->bmi[3].as_mv[idx].as_mv.row),
88              round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.col +
89                               mi->bmi[1].as_mv[idx].as_mv.col +
90                               mi->bmi[2].as_mv[idx].as_mv.col +
91                               mi->bmi[3].as_mv[idx].as_mv.col) };
92   return res;
93 }
94 
round_mv_comp_q2(int value)95 static INLINE int round_mv_comp_q2(int value) {
96   return (value < 0 ? value - 1 : value + 1) / 2;
97 }
98 
mi_mv_pred_q2(const MODE_INFO * mi,int idx,int block0,int block1)99 static MV mi_mv_pred_q2(const MODE_INFO *mi, int idx, int block0, int block1) {
100   MV res = { round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.row +
101                               mi->bmi[block1].as_mv[idx].as_mv.row),
102              round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.col +
103                               mi->bmi[block1].as_mv[idx].as_mv.col) };
104   return res;
105 }
106 
107 // TODO(jkoleszar): yet another mv clamping function :-(
clamp_mv_to_umv_border_sb(const MACROBLOCKD * xd,const MV * src_mv,int bw,int bh,int ss_x,int ss_y)108 MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd, const MV *src_mv,
109                              int bw, int bh, int ss_x, int ss_y) {
110   // If the MV points so far into the UMV border that no visible pixels
111   // are used for reconstruction, the subpel part of the MV can be
112   // discarded and the MV limited to 16 pixels with equivalent results.
113   const int spel_left = (VP9_INTERP_EXTEND + bw) << SUBPEL_BITS;
114   const int spel_right = spel_left - SUBPEL_SHIFTS;
115   const int spel_top = (VP9_INTERP_EXTEND + bh) << SUBPEL_BITS;
116   const int spel_bottom = spel_top - SUBPEL_SHIFTS;
117   MV clamped_mv = {
118     src_mv->row * (1 << (1 - ss_y)),
119     src_mv->col * (1 << (1 - ss_x))
120   };
121   assert(ss_x <= 1);
122   assert(ss_y <= 1);
123 
124   clamp_mv(&clamped_mv,
125            xd->mb_to_left_edge * (1 << (1 - ss_x)) - spel_left,
126            xd->mb_to_right_edge * (1 << (1 - ss_x)) + spel_right,
127            xd->mb_to_top_edge * (1 << (1 - ss_y)) - spel_top,
128            xd->mb_to_bottom_edge * (1 << (1 - ss_y)) + spel_bottom);
129 
130   return clamped_mv;
131 }
132 
average_split_mvs(const struct macroblockd_plane * pd,const MODE_INFO * mi,int ref,int block)133 MV average_split_mvs(const struct macroblockd_plane *pd,
134                      const MODE_INFO *mi, int ref, int block) {
135   const int ss_idx = ((pd->subsampling_x > 0) << 1) | (pd->subsampling_y > 0);
136   MV res = {0, 0};
137   switch (ss_idx) {
138     case 0:
139       res = mi->bmi[block].as_mv[ref].as_mv;
140       break;
141     case 1:
142       res = mi_mv_pred_q2(mi, ref, block, block + 2);
143       break;
144     case 2:
145       res = mi_mv_pred_q2(mi, ref, block, block + 1);
146       break;
147     case 3:
148       res = mi_mv_pred_q4(mi, ref);
149       break;
150     default:
151       assert(ss_idx <= 3 && ss_idx >= 0);
152   }
153   return res;
154 }
155 
build_inter_predictors(MACROBLOCKD * xd,int plane,int block,int bw,int bh,int x,int y,int w,int h,int mi_x,int mi_y)156 static void build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
157                                    int bw, int bh,
158                                    int x, int y, int w, int h,
159                                    int mi_x, int mi_y) {
160   struct macroblockd_plane *const pd = &xd->plane[plane];
161   const MODE_INFO *mi = xd->mi[0];
162   const int is_compound = has_second_ref(&mi->mbmi);
163   const InterpKernel *kernel = vp9_filter_kernels[mi->mbmi.interp_filter];
164   int ref;
165 
166   for (ref = 0; ref < 1 + is_compound; ++ref) {
167     const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
168     struct buf_2d *const pre_buf = &pd->pre[ref];
169     struct buf_2d *const dst_buf = &pd->dst;
170     uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
171     const MV mv = mi->mbmi.sb_type < BLOCK_8X8
172                ? average_split_mvs(pd, mi, ref, block)
173                : mi->mbmi.mv[ref].as_mv;
174 
175     // TODO(jkoleszar): This clamping is done in the incorrect place for the
176     // scaling case. It needs to be done on the scaled MV, not the pre-scaling
177     // MV. Note however that it performs the subsampling aware scaling so
178     // that the result is always q4.
179     // mv_precision precision is MV_PRECISION_Q4.
180     const MV mv_q4 = clamp_mv_to_umv_border_sb(xd, &mv, bw, bh,
181                                                pd->subsampling_x,
182                                                pd->subsampling_y);
183 
184     uint8_t *pre;
185     MV32 scaled_mv;
186     int xs, ys, subpel_x, subpel_y;
187     const int is_scaled = vp9_is_scaled(sf);
188 
189     if (is_scaled) {
190       // Co-ordinate of containing block to pixel precision.
191       const int x_start = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x));
192       const int y_start = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y));
193       if (plane == 0)
194         pre_buf->buf = xd->block_refs[ref]->buf->y_buffer;
195       else if (plane == 1)
196         pre_buf->buf = xd->block_refs[ref]->buf->u_buffer;
197       else
198         pre_buf->buf = xd->block_refs[ref]->buf->v_buffer;
199 
200       pre_buf->buf += scaled_buffer_offset(x_start + x, y_start + y,
201                                            pre_buf->stride, sf);
202       pre = pre_buf->buf;
203       scaled_mv = vp9_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
204       xs = sf->x_step_q4;
205       ys = sf->y_step_q4;
206     } else {
207       pre = pre_buf->buf + (y * pre_buf->stride + x);
208       scaled_mv.row = mv_q4.row;
209       scaled_mv.col = mv_q4.col;
210       xs = ys = 16;
211     }
212     subpel_x = scaled_mv.col & SUBPEL_MASK;
213     subpel_y = scaled_mv.row & SUBPEL_MASK;
214     pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride
215            + (scaled_mv.col >> SUBPEL_BITS);
216 
217 #if CONFIG_VP9_HIGHBITDEPTH
218     if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
219       high_inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
220                            subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys,
221                            xd->bd);
222     } else {
223       inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
224                       subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
225     }
226 #else
227     inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
228                     subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
229 #endif  // CONFIG_VP9_HIGHBITDEPTH
230   }
231 }
232 
build_inter_predictors_for_planes(MACROBLOCKD * xd,BLOCK_SIZE bsize,int mi_row,int mi_col,int plane_from,int plane_to)233 static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
234                                               int mi_row, int mi_col,
235                                               int plane_from, int plane_to) {
236   int plane;
237   const int mi_x = mi_col * MI_SIZE;
238   const int mi_y = mi_row * MI_SIZE;
239   for (plane = plane_from; plane <= plane_to; ++plane) {
240     const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize,
241                                                         &xd->plane[plane]);
242     const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
243     const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
244     const int bw = 4 * num_4x4_w;
245     const int bh = 4 * num_4x4_h;
246 
247     if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8) {
248       int i = 0, x, y;
249       assert(bsize == BLOCK_8X8);
250       for (y = 0; y < num_4x4_h; ++y)
251         for (x = 0; x < num_4x4_w; ++x)
252            build_inter_predictors(xd, plane, i++, bw, bh,
253                                   4 * x, 4 * y, 4, 4, mi_x, mi_y);
254     } else {
255       build_inter_predictors(xd, plane, 0, bw, bh,
256                              0, 0, bw, bh, mi_x, mi_y);
257     }
258   }
259 }
260 
vp9_build_inter_predictors_sby(MACROBLOCKD * xd,int mi_row,int mi_col,BLOCK_SIZE bsize)261 void vp9_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
262                                     BLOCK_SIZE bsize) {
263   build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0, 0);
264 }
265 
vp9_build_inter_predictors_sbp(MACROBLOCKD * xd,int mi_row,int mi_col,BLOCK_SIZE bsize,int plane)266 void vp9_build_inter_predictors_sbp(MACROBLOCKD *xd, int mi_row, int mi_col,
267                                     BLOCK_SIZE bsize, int plane) {
268   build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, plane, plane);
269 }
270 
vp9_build_inter_predictors_sbuv(MACROBLOCKD * xd,int mi_row,int mi_col,BLOCK_SIZE bsize)271 void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
272                                      BLOCK_SIZE bsize) {
273   build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 1,
274                                     MAX_MB_PLANE - 1);
275 }
276 
vp9_build_inter_predictors_sb(MACROBLOCKD * xd,int mi_row,int mi_col,BLOCK_SIZE bsize)277 void vp9_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
278                                    BLOCK_SIZE bsize) {
279   build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0,
280                                     MAX_MB_PLANE - 1);
281 }
282 
vp9_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],const YV12_BUFFER_CONFIG * src,int mi_row,int mi_col)283 void vp9_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],
284                           const YV12_BUFFER_CONFIG *src,
285                           int mi_row, int mi_col) {
286   uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer,
287       src->v_buffer};
288   const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride,
289       src->uv_stride};
290   int i;
291 
292   for (i = 0; i < MAX_MB_PLANE; ++i) {
293     struct macroblockd_plane *const pd = &planes[i];
294     setup_pred_plane(&pd->dst, buffers[i], strides[i], mi_row, mi_col, NULL,
295                      pd->subsampling_x, pd->subsampling_y);
296   }
297 }
298 
vp9_setup_pre_planes(MACROBLOCKD * xd,int idx,const YV12_BUFFER_CONFIG * src,int mi_row,int mi_col,const struct scale_factors * sf)299 void vp9_setup_pre_planes(MACROBLOCKD *xd, int idx,
300                           const YV12_BUFFER_CONFIG *src,
301                           int mi_row, int mi_col,
302                           const struct scale_factors *sf) {
303   if (src != NULL) {
304     int i;
305     uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer,
306         src->v_buffer};
307     const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride,
308         src->uv_stride};
309     for (i = 0; i < MAX_MB_PLANE; ++i) {
310       struct macroblockd_plane *const pd = &xd->plane[i];
311       setup_pred_plane(&pd->pre[idx], buffers[i], strides[i], mi_row, mi_col,
312                        sf, pd->subsampling_x, pd->subsampling_y);
313     }
314   }
315 }
316