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 "./vpx_config.h"
12 #include "./vpx_dsp_rtcd.h"
13
14 #if CONFIG_VP9_HIGHBITDEPTH
15 #include "vpx_dsp/vpx_dsp_common.h"
16 #endif // CONFIG_VP9_HIGHBITDEPTH
17 #include "vpx_mem/vpx_mem.h"
18 #include "vpx_ports/mem.h"
19 #include "vpx_ports/vpx_once.h"
20
21 #include "vp9/common/vp9_reconintra.h"
22 #include "vp9/common/vp9_onyxc_int.h"
23
24 const TX_TYPE intra_mode_to_tx_type_lookup[INTRA_MODES] = {
25 DCT_DCT, // DC
26 ADST_DCT, // V
27 DCT_ADST, // H
28 DCT_DCT, // D45
29 ADST_ADST, // D135
30 ADST_DCT, // D117
31 DCT_ADST, // D153
32 DCT_ADST, // D207
33 ADST_DCT, // D63
34 ADST_ADST, // TM
35 };
36
37 enum {
38 NEED_LEFT = 1 << 1,
39 NEED_ABOVE = 1 << 2,
40 NEED_ABOVERIGHT = 1 << 3,
41 };
42
43 static const uint8_t extend_modes[INTRA_MODES] = {
44 NEED_ABOVE | NEED_LEFT, // DC
45 NEED_ABOVE, // V
46 NEED_LEFT, // H
47 NEED_ABOVERIGHT, // D45
48 NEED_LEFT | NEED_ABOVE, // D135
49 NEED_LEFT | NEED_ABOVE, // D117
50 NEED_LEFT | NEED_ABOVE, // D153
51 NEED_LEFT, // D207
52 NEED_ABOVERIGHT, // D63
53 NEED_LEFT | NEED_ABOVE, // TM
54 };
55
56 typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride,
57 const uint8_t *above, const uint8_t *left);
58
59 static intra_pred_fn pred[INTRA_MODES][TX_SIZES];
60 static intra_pred_fn dc_pred[2][2][TX_SIZES];
61
62 #if CONFIG_VP9_HIGHBITDEPTH
63 typedef void (*intra_high_pred_fn)(uint16_t *dst, ptrdiff_t stride,
64 const uint16_t *above, const uint16_t *left,
65 int bd);
66 static intra_high_pred_fn pred_high[INTRA_MODES][4];
67 static intra_high_pred_fn dc_pred_high[2][2][4];
68 #endif // CONFIG_VP9_HIGHBITDEPTH
69
vp9_init_intra_predictors_internal(void)70 static void vp9_init_intra_predictors_internal(void) {
71 #define INIT_ALL_SIZES(p, type) \
72 p[TX_4X4] = vpx_##type##_predictor_4x4; \
73 p[TX_8X8] = vpx_##type##_predictor_8x8; \
74 p[TX_16X16] = vpx_##type##_predictor_16x16; \
75 p[TX_32X32] = vpx_##type##_predictor_32x32
76
77 INIT_ALL_SIZES(pred[V_PRED], v);
78 INIT_ALL_SIZES(pred[H_PRED], h);
79 INIT_ALL_SIZES(pred[D207_PRED], d207);
80 INIT_ALL_SIZES(pred[D45_PRED], d45);
81 INIT_ALL_SIZES(pred[D63_PRED], d63);
82 INIT_ALL_SIZES(pred[D117_PRED], d117);
83 INIT_ALL_SIZES(pred[D135_PRED], d135);
84 INIT_ALL_SIZES(pred[D153_PRED], d153);
85 INIT_ALL_SIZES(pred[TM_PRED], tm);
86
87 INIT_ALL_SIZES(dc_pred[0][0], dc_128);
88 INIT_ALL_SIZES(dc_pred[0][1], dc_top);
89 INIT_ALL_SIZES(dc_pred[1][0], dc_left);
90 INIT_ALL_SIZES(dc_pred[1][1], dc);
91
92 #if CONFIG_VP9_HIGHBITDEPTH
93 INIT_ALL_SIZES(pred_high[V_PRED], highbd_v);
94 INIT_ALL_SIZES(pred_high[H_PRED], highbd_h);
95 INIT_ALL_SIZES(pred_high[D207_PRED], highbd_d207);
96 INIT_ALL_SIZES(pred_high[D45_PRED], highbd_d45);
97 INIT_ALL_SIZES(pred_high[D63_PRED], highbd_d63);
98 INIT_ALL_SIZES(pred_high[D117_PRED], highbd_d117);
99 INIT_ALL_SIZES(pred_high[D135_PRED], highbd_d135);
100 INIT_ALL_SIZES(pred_high[D153_PRED], highbd_d153);
101 INIT_ALL_SIZES(pred_high[TM_PRED], highbd_tm);
102
103 INIT_ALL_SIZES(dc_pred_high[0][0], highbd_dc_128);
104 INIT_ALL_SIZES(dc_pred_high[0][1], highbd_dc_top);
105 INIT_ALL_SIZES(dc_pred_high[1][0], highbd_dc_left);
106 INIT_ALL_SIZES(dc_pred_high[1][1], highbd_dc);
107 #endif // CONFIG_VP9_HIGHBITDEPTH
108
109 #undef intra_pred_allsizes
110 }
111
112 #if CONFIG_VP9_HIGHBITDEPTH
build_intra_predictors_high(const MACROBLOCKD * xd,const uint8_t * ref8,int ref_stride,uint8_t * dst8,int dst_stride,PREDICTION_MODE mode,TX_SIZE tx_size,int up_available,int left_available,int right_available,int x,int y,int plane,int bd)113 static void build_intra_predictors_high(const MACROBLOCKD *xd,
114 const uint8_t *ref8,
115 int ref_stride,
116 uint8_t *dst8,
117 int dst_stride,
118 PREDICTION_MODE mode,
119 TX_SIZE tx_size,
120 int up_available,
121 int left_available,
122 int right_available,
123 int x, int y,
124 int plane, int bd) {
125 int i;
126 uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
127 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
128 DECLARE_ALIGNED(16, uint16_t, left_col[32]);
129 DECLARE_ALIGNED(16, uint16_t, above_data[64 + 16]);
130 uint16_t *above_row = above_data + 16;
131 const uint16_t *const_above_row = above_row;
132 const int bs = 4 << tx_size;
133 int frame_width, frame_height;
134 int x0, y0;
135 const struct macroblockd_plane *const pd = &xd->plane[plane];
136 const int need_left = extend_modes[mode] & NEED_LEFT;
137 const int need_above = extend_modes[mode] & NEED_ABOVE;
138 const int need_aboveright = extend_modes[mode] & NEED_ABOVERIGHT;
139 int base = 128 << (bd - 8);
140 // 127 127 127 .. 127 127 127 127 127 127
141 // 129 A B .. Y Z
142 // 129 C D .. W X
143 // 129 E F .. U V
144 // 129 G H .. S T T T T T
145 // For 10 bit and 12 bit, 127 and 129 are replaced by base -1 and base + 1.
146
147 // Get current frame pointer, width and height.
148 if (plane == 0) {
149 frame_width = xd->cur_buf->y_width;
150 frame_height = xd->cur_buf->y_height;
151 } else {
152 frame_width = xd->cur_buf->uv_width;
153 frame_height = xd->cur_buf->uv_height;
154 }
155
156 // Get block position in current frame.
157 x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
158 y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
159
160 // NEED_LEFT
161 if (need_left) {
162 if (left_available) {
163 if (xd->mb_to_bottom_edge < 0) {
164 /* slower path if the block needs border extension */
165 if (y0 + bs <= frame_height) {
166 for (i = 0; i < bs; ++i)
167 left_col[i] = ref[i * ref_stride - 1];
168 } else {
169 const int extend_bottom = frame_height - y0;
170 for (i = 0; i < extend_bottom; ++i)
171 left_col[i] = ref[i * ref_stride - 1];
172 for (; i < bs; ++i)
173 left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
174 }
175 } else {
176 /* faster path if the block does not need extension */
177 for (i = 0; i < bs; ++i)
178 left_col[i] = ref[i * ref_stride - 1];
179 }
180 } else {
181 vpx_memset16(left_col, base + 1, bs);
182 }
183 }
184
185 // NEED_ABOVE
186 if (need_above) {
187 if (up_available) {
188 const uint16_t *above_ref = ref - ref_stride;
189 if (xd->mb_to_right_edge < 0) {
190 /* slower path if the block needs border extension */
191 if (x0 + bs <= frame_width) {
192 memcpy(above_row, above_ref, bs * sizeof(above_row[0]));
193 } else if (x0 <= frame_width) {
194 const int r = frame_width - x0;
195 memcpy(above_row, above_ref, r * sizeof(above_row[0]));
196 vpx_memset16(above_row + r, above_row[r - 1], x0 + bs - frame_width);
197 }
198 } else {
199 /* faster path if the block does not need extension */
200 if (bs == 4 && right_available && left_available) {
201 const_above_row = above_ref;
202 } else {
203 memcpy(above_row, above_ref, bs * sizeof(above_row[0]));
204 }
205 }
206 above_row[-1] = left_available ? above_ref[-1] : (base + 1);
207 } else {
208 vpx_memset16(above_row, base - 1, bs);
209 above_row[-1] = base - 1;
210 }
211 }
212
213 // NEED_ABOVERIGHT
214 if (need_aboveright) {
215 if (up_available) {
216 const uint16_t *above_ref = ref - ref_stride;
217 if (xd->mb_to_right_edge < 0) {
218 /* slower path if the block needs border extension */
219 if (x0 + 2 * bs <= frame_width) {
220 if (right_available && bs == 4) {
221 memcpy(above_row, above_ref, 2 * bs * sizeof(above_row[0]));
222 } else {
223 memcpy(above_row, above_ref, bs * sizeof(above_row[0]));
224 vpx_memset16(above_row + bs, above_row[bs - 1], bs);
225 }
226 } else if (x0 + bs <= frame_width) {
227 const int r = frame_width - x0;
228 if (right_available && bs == 4) {
229 memcpy(above_row, above_ref, r * sizeof(above_row[0]));
230 vpx_memset16(above_row + r, above_row[r - 1],
231 x0 + 2 * bs - frame_width);
232 } else {
233 memcpy(above_row, above_ref, bs * sizeof(above_row[0]));
234 vpx_memset16(above_row + bs, above_row[bs - 1], bs);
235 }
236 } else if (x0 <= frame_width) {
237 const int r = frame_width - x0;
238 memcpy(above_row, above_ref, r * sizeof(above_row[0]));
239 vpx_memset16(above_row + r, above_row[r - 1],
240 x0 + 2 * bs - frame_width);
241 }
242 above_row[-1] = left_available ? above_ref[-1] : (base + 1);
243 } else {
244 /* faster path if the block does not need extension */
245 if (bs == 4 && right_available && left_available) {
246 const_above_row = above_ref;
247 } else {
248 memcpy(above_row, above_ref, bs * sizeof(above_row[0]));
249 if (bs == 4 && right_available)
250 memcpy(above_row + bs, above_ref + bs, bs * sizeof(above_row[0]));
251 else
252 vpx_memset16(above_row + bs, above_row[bs - 1], bs);
253 above_row[-1] = left_available ? above_ref[-1] : (base + 1);
254 }
255 }
256 } else {
257 vpx_memset16(above_row, base - 1, bs * 2);
258 above_row[-1] = base - 1;
259 }
260 }
261
262 // predict
263 if (mode == DC_PRED) {
264 dc_pred_high[left_available][up_available][tx_size](dst, dst_stride,
265 const_above_row,
266 left_col, xd->bd);
267 } else {
268 pred_high[mode][tx_size](dst, dst_stride, const_above_row, left_col,
269 xd->bd);
270 }
271 }
272 #endif // CONFIG_VP9_HIGHBITDEPTH
273
build_intra_predictors(const MACROBLOCKD * xd,const uint8_t * ref,int ref_stride,uint8_t * dst,int dst_stride,PREDICTION_MODE mode,TX_SIZE tx_size,int up_available,int left_available,int right_available,int x,int y,int plane)274 static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
275 int ref_stride, uint8_t *dst, int dst_stride,
276 PREDICTION_MODE mode, TX_SIZE tx_size,
277 int up_available, int left_available,
278 int right_available, int x, int y,
279 int plane) {
280 int i;
281 DECLARE_ALIGNED(16, uint8_t, left_col[32]);
282 DECLARE_ALIGNED(16, uint8_t, above_data[64 + 16]);
283 uint8_t *above_row = above_data + 16;
284 const uint8_t *const_above_row = above_row;
285 const int bs = 4 << tx_size;
286 int frame_width, frame_height;
287 int x0, y0;
288 const struct macroblockd_plane *const pd = &xd->plane[plane];
289
290 // 127 127 127 .. 127 127 127 127 127 127
291 // 129 A B .. Y Z
292 // 129 C D .. W X
293 // 129 E F .. U V
294 // 129 G H .. S T T T T T
295 // ..
296
297 // Get current frame pointer, width and height.
298 if (plane == 0) {
299 frame_width = xd->cur_buf->y_width;
300 frame_height = xd->cur_buf->y_height;
301 } else {
302 frame_width = xd->cur_buf->uv_width;
303 frame_height = xd->cur_buf->uv_height;
304 }
305
306 // Get block position in current frame.
307 x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
308 y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
309
310 // NEED_LEFT
311 if (extend_modes[mode] & NEED_LEFT) {
312 if (left_available) {
313 if (xd->mb_to_bottom_edge < 0) {
314 /* slower path if the block needs border extension */
315 if (y0 + bs <= frame_height) {
316 for (i = 0; i < bs; ++i)
317 left_col[i] = ref[i * ref_stride - 1];
318 } else {
319 const int extend_bottom = frame_height - y0;
320 for (i = 0; i < extend_bottom; ++i)
321 left_col[i] = ref[i * ref_stride - 1];
322 for (; i < bs; ++i)
323 left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
324 }
325 } else {
326 /* faster path if the block does not need extension */
327 for (i = 0; i < bs; ++i)
328 left_col[i] = ref[i * ref_stride - 1];
329 }
330 } else {
331 memset(left_col, 129, bs);
332 }
333 }
334
335 // NEED_ABOVE
336 if (extend_modes[mode] & NEED_ABOVE) {
337 if (up_available) {
338 const uint8_t *above_ref = ref - ref_stride;
339 if (xd->mb_to_right_edge < 0) {
340 /* slower path if the block needs border extension */
341 if (x0 + bs <= frame_width) {
342 memcpy(above_row, above_ref, bs);
343 } else if (x0 <= frame_width) {
344 const int r = frame_width - x0;
345 memcpy(above_row, above_ref, r);
346 memset(above_row + r, above_row[r - 1], x0 + bs - frame_width);
347 }
348 } else {
349 /* faster path if the block does not need extension */
350 if (bs == 4 && right_available && left_available) {
351 const_above_row = above_ref;
352 } else {
353 memcpy(above_row, above_ref, bs);
354 }
355 }
356 above_row[-1] = left_available ? above_ref[-1] : 129;
357 } else {
358 memset(above_row, 127, bs);
359 above_row[-1] = 127;
360 }
361 }
362
363 // NEED_ABOVERIGHT
364 if (extend_modes[mode] & NEED_ABOVERIGHT) {
365 if (up_available) {
366 const uint8_t *above_ref = ref - ref_stride;
367 if (xd->mb_to_right_edge < 0) {
368 /* slower path if the block needs border extension */
369 if (x0 + 2 * bs <= frame_width) {
370 if (right_available && bs == 4) {
371 memcpy(above_row, above_ref, 2 * bs);
372 } else {
373 memcpy(above_row, above_ref, bs);
374 memset(above_row + bs, above_row[bs - 1], bs);
375 }
376 } else if (x0 + bs <= frame_width) {
377 const int r = frame_width - x0;
378 if (right_available && bs == 4) {
379 memcpy(above_row, above_ref, r);
380 memset(above_row + r, above_row[r - 1], x0 + 2 * bs - frame_width);
381 } else {
382 memcpy(above_row, above_ref, bs);
383 memset(above_row + bs, above_row[bs - 1], bs);
384 }
385 } else if (x0 <= frame_width) {
386 const int r = frame_width - x0;
387 memcpy(above_row, above_ref, r);
388 memset(above_row + r, above_row[r - 1], x0 + 2 * bs - frame_width);
389 }
390 } else {
391 /* faster path if the block does not need extension */
392 if (bs == 4 && right_available && left_available) {
393 const_above_row = above_ref;
394 } else {
395 memcpy(above_row, above_ref, bs);
396 if (bs == 4 && right_available)
397 memcpy(above_row + bs, above_ref + bs, bs);
398 else
399 memset(above_row + bs, above_row[bs - 1], bs);
400 }
401 }
402 above_row[-1] = left_available ? above_ref[-1] : 129;
403 } else {
404 memset(above_row, 127, bs * 2);
405 above_row[-1] = 127;
406 }
407 }
408
409 // predict
410 if (mode == DC_PRED) {
411 dc_pred[left_available][up_available][tx_size](dst, dst_stride,
412 const_above_row, left_col);
413 } else {
414 pred[mode][tx_size](dst, dst_stride, const_above_row, left_col);
415 }
416 }
417
vp9_predict_intra_block(const MACROBLOCKD * xd,int bwl_in,TX_SIZE tx_size,PREDICTION_MODE mode,const uint8_t * ref,int ref_stride,uint8_t * dst,int dst_stride,int aoff,int loff,int plane)418 void vp9_predict_intra_block(const MACROBLOCKD *xd, int bwl_in,
419 TX_SIZE tx_size, PREDICTION_MODE mode,
420 const uint8_t *ref, int ref_stride,
421 uint8_t *dst, int dst_stride,
422 int aoff, int loff, int plane) {
423 const int bw = (1 << bwl_in);
424 const int txw = (1 << tx_size);
425 const int have_top = loff || (xd->above_mi != NULL);
426 const int have_left = aoff || (xd->left_mi != NULL);
427 const int have_right = (aoff + txw) < bw;
428 const int x = aoff * 4;
429 const int y = loff * 4;
430
431 #if CONFIG_VP9_HIGHBITDEPTH
432 if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
433 build_intra_predictors_high(xd, ref, ref_stride, dst, dst_stride, mode,
434 tx_size, have_top, have_left, have_right,
435 x, y, plane, xd->bd);
436 return;
437 }
438 #endif
439 build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size,
440 have_top, have_left, have_right, x, y, plane);
441 }
442
vp9_init_intra_predictors(void)443 void vp9_init_intra_predictors(void) {
444 once(vp9_init_intra_predictors_internal);
445 }
446