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
vp9_highbd_build_inter_predictor(const uint16_t * src,int src_stride,uint16_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)23 void vp9_highbd_build_inter_predictor(
24 const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride,
25 const MV *src_mv, const struct scale_factors *sf, int w, int h, int ref,
26 const InterpKernel *kernel, enum mv_precision precision, int x, int y,
27 int bd) {
28 const int is_q4 = precision == MV_PRECISION_Q4;
29 const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
30 is_q4 ? src_mv->col : src_mv->col * 2 };
31 MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf);
32 const int subpel_x = mv.col & SUBPEL_MASK;
33 const int subpel_y = mv.row & SUBPEL_MASK;
34
35 src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);
36
37 highbd_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
38 sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4,
39 bd);
40 }
41 #endif // CONFIG_VP9_HIGHBITDEPTH
42
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)43 void vp9_build_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
44 int dst_stride, const MV *src_mv,
45 const struct scale_factors *sf, int w, int h,
46 int ref, const InterpKernel *kernel,
47 enum mv_precision precision, int x, int y) {
48 const int is_q4 = precision == MV_PRECISION_Q4;
49 const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
50 is_q4 ? src_mv->col : src_mv->col * 2 };
51 MV32 mv = vp9_scale_mv(&mv_q4, x, y, sf);
52 const int subpel_x = mv.col & SUBPEL_MASK;
53 const int subpel_y = mv.row & SUBPEL_MASK;
54
55 src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);
56
57 inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y, sf, w,
58 h, ref, kernel, sf->x_step_q4, sf->y_step_q4);
59 }
60
round_mv_comp_q4(int value)61 static INLINE int round_mv_comp_q4(int value) {
62 return (value < 0 ? value - 2 : value + 2) / 4;
63 }
64
mi_mv_pred_q4(const MODE_INFO * mi,int idx)65 static MV mi_mv_pred_q4(const MODE_INFO *mi, int idx) {
66 MV res = { round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.row +
67 mi->bmi[1].as_mv[idx].as_mv.row +
68 mi->bmi[2].as_mv[idx].as_mv.row +
69 mi->bmi[3].as_mv[idx].as_mv.row),
70 round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.col +
71 mi->bmi[1].as_mv[idx].as_mv.col +
72 mi->bmi[2].as_mv[idx].as_mv.col +
73 mi->bmi[3].as_mv[idx].as_mv.col) };
74 return res;
75 }
76
round_mv_comp_q2(int value)77 static INLINE int round_mv_comp_q2(int value) {
78 return (value < 0 ? value - 1 : value + 1) / 2;
79 }
80
mi_mv_pred_q2(const MODE_INFO * mi,int idx,int block0,int block1)81 static MV mi_mv_pred_q2(const MODE_INFO *mi, int idx, int block0, int block1) {
82 MV res = { round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.row +
83 mi->bmi[block1].as_mv[idx].as_mv.row),
84 round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.col +
85 mi->bmi[block1].as_mv[idx].as_mv.col) };
86 return res;
87 }
88
89 // 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)90 MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd, const MV *src_mv, int bw,
91 int bh, int ss_x, int ss_y) {
92 // If the MV points so far into the UMV border that no visible pixels
93 // are used for reconstruction, the subpel part of the MV can be
94 // discarded and the MV limited to 16 pixels with equivalent results.
95 const int spel_left = (VP9_INTERP_EXTEND + bw) << SUBPEL_BITS;
96 const int spel_right = spel_left - SUBPEL_SHIFTS;
97 const int spel_top = (VP9_INTERP_EXTEND + bh) << SUBPEL_BITS;
98 const int spel_bottom = spel_top - SUBPEL_SHIFTS;
99 MV clamped_mv = { (short)(src_mv->row * (1 << (1 - ss_y))),
100 (short)(src_mv->col * (1 << (1 - ss_x))) };
101 assert(ss_x <= 1);
102 assert(ss_y <= 1);
103
104 clamp_mv(&clamped_mv, xd->mb_to_left_edge * (1 << (1 - ss_x)) - spel_left,
105 xd->mb_to_right_edge * (1 << (1 - ss_x)) + spel_right,
106 xd->mb_to_top_edge * (1 << (1 - ss_y)) - spel_top,
107 xd->mb_to_bottom_edge * (1 << (1 - ss_y)) + spel_bottom);
108
109 return clamped_mv;
110 }
111
average_split_mvs(const struct macroblockd_plane * pd,const MODE_INFO * mi,int ref,int block)112 MV average_split_mvs(const struct macroblockd_plane *pd, const MODE_INFO *mi,
113 int ref, int block) {
114 const int ss_idx = ((pd->subsampling_x > 0) << 1) | (pd->subsampling_y > 0);
115 MV res = { 0, 0 };
116 switch (ss_idx) {
117 case 0: res = mi->bmi[block].as_mv[ref].as_mv; break;
118 case 1: res = mi_mv_pred_q2(mi, ref, block, block + 2); break;
119 case 2: res = mi_mv_pred_q2(mi, ref, block, block + 1); break;
120 case 3: res = mi_mv_pred_q4(mi, ref); break;
121 default: assert(ss_idx <= 3 && ss_idx >= 0);
122 }
123 return res;
124 }
125
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)126 static void build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
127 int bw, int bh, int x, int y, int w, int h,
128 int mi_x, int mi_y) {
129 struct macroblockd_plane *const pd = &xd->plane[plane];
130 const MODE_INFO *mi = xd->mi[0];
131 const int is_compound = has_second_ref(mi);
132 const InterpKernel *kernel = vp9_filter_kernels[mi->interp_filter];
133 int ref;
134
135 for (ref = 0; ref < 1 + is_compound; ++ref) {
136 const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
137 struct buf_2d *const pre_buf = &pd->pre[ref];
138 struct buf_2d *const dst_buf = &pd->dst;
139 uint8_t *const dst = dst_buf->buf + (int64_t)dst_buf->stride * y + x;
140 const MV mv = mi->sb_type < BLOCK_8X8
141 ? average_split_mvs(pd, mi, ref, block)
142 : mi->mv[ref].as_mv;
143
144 // TODO(jkoleszar): This clamping is done in the incorrect place for the
145 // scaling case. It needs to be done on the scaled MV, not the pre-scaling
146 // MV. Note however that it performs the subsampling aware scaling so
147 // that the result is always q4.
148 // mv_precision precision is MV_PRECISION_Q4.
149 const MV mv_q4 = clamp_mv_to_umv_border_sb(
150 xd, &mv, bw, bh, pd->subsampling_x, pd->subsampling_y);
151
152 uint8_t *pre;
153 MV32 scaled_mv;
154 int xs, ys, subpel_x, subpel_y;
155 const int is_scaled = vp9_is_scaled(sf);
156
157 if (is_scaled) {
158 // Co-ordinate of containing block to pixel precision.
159 const int x_start = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x));
160 const int y_start = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y));
161 #if 0 // CONFIG_BETTER_HW_COMPATIBILITY
162 assert(xd->mi[0]->sb_type != BLOCK_4X8 &&
163 xd->mi[0]->sb_type != BLOCK_8X4);
164 assert(mv_q4.row == mv.row * (1 << (1 - pd->subsampling_y)) &&
165 mv_q4.col == mv.col * (1 << (1 - pd->subsampling_x)));
166 #endif
167 if (plane == 0)
168 pre_buf->buf = xd->block_refs[ref]->buf->y_buffer;
169 else if (plane == 1)
170 pre_buf->buf = xd->block_refs[ref]->buf->u_buffer;
171 else
172 pre_buf->buf = xd->block_refs[ref]->buf->v_buffer;
173
174 pre_buf->buf +=
175 scaled_buffer_offset(x_start + x, y_start + y, pre_buf->stride, sf);
176 pre = pre_buf->buf;
177 scaled_mv = vp9_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
178 xs = sf->x_step_q4;
179 ys = sf->y_step_q4;
180 } else {
181 pre = pre_buf->buf + ((int64_t)y * pre_buf->stride + x);
182 scaled_mv.row = mv_q4.row;
183 scaled_mv.col = mv_q4.col;
184 xs = ys = 16;
185 }
186 subpel_x = scaled_mv.col & SUBPEL_MASK;
187 subpel_y = scaled_mv.row & SUBPEL_MASK;
188 pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride +
189 (scaled_mv.col >> SUBPEL_BITS);
190
191 #if CONFIG_VP9_HIGHBITDEPTH
192 if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
193 highbd_inter_predictor(CONVERT_TO_SHORTPTR(pre), pre_buf->stride,
194 CONVERT_TO_SHORTPTR(dst), dst_buf->stride,
195 subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys,
196 xd->bd);
197 } else {
198 inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride, subpel_x,
199 subpel_y, sf, w, h, ref, kernel, xs, ys);
200 }
201 #else
202 inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride, subpel_x,
203 subpel_y, sf, w, h, ref, kernel, xs, ys);
204 #endif // CONFIG_VP9_HIGHBITDEPTH
205 }
206 }
207
build_inter_predictors_for_planes(MACROBLOCKD * xd,BLOCK_SIZE bsize,int mi_row,int mi_col,int plane_from,int plane_to)208 static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
209 int mi_row, int mi_col,
210 int plane_from, int plane_to) {
211 int plane;
212 const int mi_x = mi_col * MI_SIZE;
213 const int mi_y = mi_row * MI_SIZE;
214 for (plane = plane_from; plane <= plane_to; ++plane) {
215 const BLOCK_SIZE plane_bsize =
216 get_plane_block_size(bsize, &xd->plane[plane]);
217 const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
218 const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
219 const int bw = 4 * num_4x4_w;
220 const int bh = 4 * num_4x4_h;
221
222 if (xd->mi[0]->sb_type < BLOCK_8X8) {
223 int i = 0, x, y;
224 assert(bsize == BLOCK_8X8);
225 for (y = 0; y < num_4x4_h; ++y)
226 for (x = 0; x < num_4x4_w; ++x)
227 build_inter_predictors(xd, plane, i++, bw, bh, 4 * x, 4 * y, 4, 4,
228 mi_x, mi_y);
229 } else {
230 build_inter_predictors(xd, plane, 0, bw, bh, 0, 0, bw, bh, mi_x, mi_y);
231 }
232 }
233 }
234
vp9_build_inter_predictors_sby(MACROBLOCKD * xd,int mi_row,int mi_col,BLOCK_SIZE bsize)235 void vp9_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
236 BLOCK_SIZE bsize) {
237 build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0, 0);
238 }
239
vp9_build_inter_predictors_sbp(MACROBLOCKD * xd,int mi_row,int mi_col,BLOCK_SIZE bsize,int plane)240 void vp9_build_inter_predictors_sbp(MACROBLOCKD *xd, int mi_row, int mi_col,
241 BLOCK_SIZE bsize, int plane) {
242 build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, plane, plane);
243 }
244
vp9_build_inter_predictors_sbuv(MACROBLOCKD * xd,int mi_row,int mi_col,BLOCK_SIZE bsize)245 void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
246 BLOCK_SIZE bsize) {
247 build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 1,
248 MAX_MB_PLANE - 1);
249 }
250
vp9_build_inter_predictors_sb(MACROBLOCKD * xd,int mi_row,int mi_col,BLOCK_SIZE bsize)251 void vp9_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
252 BLOCK_SIZE bsize) {
253 build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0,
254 MAX_MB_PLANE - 1);
255 }
256
vp9_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],const YV12_BUFFER_CONFIG * src,int mi_row,int mi_col)257 void vp9_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],
258 const YV12_BUFFER_CONFIG *src, int mi_row,
259 int mi_col) {
260 uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer,
261 src->v_buffer };
262 const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride,
263 src->uv_stride };
264 int i;
265
266 for (i = 0; i < MAX_MB_PLANE; ++i) {
267 struct macroblockd_plane *const pd = &planes[i];
268 setup_pred_plane(&pd->dst, buffers[i], strides[i], mi_row, mi_col, NULL,
269 pd->subsampling_x, pd->subsampling_y);
270 }
271 }
272
vp9_setup_pre_planes(MACROBLOCKD * xd,int idx,const YV12_BUFFER_CONFIG * src,int mi_row,int mi_col,const struct scale_factors * sf)273 void vp9_setup_pre_planes(MACROBLOCKD *xd, int idx,
274 const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
275 const struct scale_factors *sf) {
276 if (src != NULL) {
277 int i;
278 uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer,
279 src->v_buffer };
280 const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride,
281 src->uv_stride };
282 for (i = 0; i < MAX_MB_PLANE; ++i) {
283 struct macroblockd_plane *const pd = &xd->plane[i];
284 setup_pred_plane(&pd->pre[idx], buffers[i], strides[i], mi_row, mi_col,
285 sf, pd->subsampling_x, pd->subsampling_y);
286 }
287 }
288 }
289