Line data Source code
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 :
70 0 : 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 0 : INIT_ALL_SIZES(pred[V_PRED], v);
78 0 : INIT_ALL_SIZES(pred[H_PRED], h);
79 0 : INIT_ALL_SIZES(pred[D207_PRED], d207);
80 0 : INIT_ALL_SIZES(pred[D45_PRED], d45);
81 0 : INIT_ALL_SIZES(pred[D63_PRED], d63);
82 0 : INIT_ALL_SIZES(pred[D117_PRED], d117);
83 0 : INIT_ALL_SIZES(pred[D135_PRED], d135);
84 0 : INIT_ALL_SIZES(pred[D153_PRED], d153);
85 0 : INIT_ALL_SIZES(pred[TM_PRED], tm);
86 :
87 0 : INIT_ALL_SIZES(dc_pred[0][0], dc_128);
88 0 : INIT_ALL_SIZES(dc_pred[0][1], dc_top);
89 0 : INIT_ALL_SIZES(dc_pred[1][0], dc_left);
90 0 : 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 0 : }
111 :
112 : #if CONFIG_VP9_HIGHBITDEPTH
113 : static void build_intra_predictors_high(
114 : const MACROBLOCKD *xd, const uint8_t *ref8, int ref_stride, uint8_t *dst8,
115 : int dst_stride, PREDICTION_MODE mode, TX_SIZE tx_size, int up_available,
116 : int left_available, int right_available, int x, int y, int plane, int bd) {
117 : int i;
118 : uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
119 : uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
120 : DECLARE_ALIGNED(16, uint16_t, left_col[32]);
121 : DECLARE_ALIGNED(16, uint16_t, above_data[64 + 16]);
122 : uint16_t *above_row = above_data + 16;
123 : const uint16_t *const_above_row = above_row;
124 : const int bs = 4 << tx_size;
125 : int frame_width, frame_height;
126 : int x0, y0;
127 : const struct macroblockd_plane *const pd = &xd->plane[plane];
128 : const int need_left = extend_modes[mode] & NEED_LEFT;
129 : const int need_above = extend_modes[mode] & NEED_ABOVE;
130 : const int need_aboveright = extend_modes[mode] & NEED_ABOVERIGHT;
131 : int base = 128 << (bd - 8);
132 : // 127 127 127 .. 127 127 127 127 127 127
133 : // 129 A B .. Y Z
134 : // 129 C D .. W X
135 : // 129 E F .. U V
136 : // 129 G H .. S T T T T T
137 : // For 10 bit and 12 bit, 127 and 129 are replaced by base -1 and base + 1.
138 :
139 : // Get current frame pointer, width and height.
140 : if (plane == 0) {
141 : frame_width = xd->cur_buf->y_width;
142 : frame_height = xd->cur_buf->y_height;
143 : } else {
144 : frame_width = xd->cur_buf->uv_width;
145 : frame_height = xd->cur_buf->uv_height;
146 : }
147 :
148 : // Get block position in current frame.
149 : x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
150 : y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
151 :
152 : // NEED_LEFT
153 : if (need_left) {
154 : if (left_available) {
155 : if (xd->mb_to_bottom_edge < 0) {
156 : /* slower path if the block needs border extension */
157 : if (y0 + bs <= frame_height) {
158 : for (i = 0; i < bs; ++i) left_col[i] = ref[i * ref_stride - 1];
159 : } else {
160 : const int extend_bottom = frame_height - y0;
161 : for (i = 0; i < extend_bottom; ++i)
162 : left_col[i] = ref[i * ref_stride - 1];
163 : for (; i < bs; ++i)
164 : left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
165 : }
166 : } else {
167 : /* faster path if the block does not need extension */
168 : for (i = 0; i < bs; ++i) left_col[i] = ref[i * ref_stride - 1];
169 : }
170 : } else {
171 : vpx_memset16(left_col, base + 1, bs);
172 : }
173 : }
174 :
175 : // NEED_ABOVE
176 : if (need_above) {
177 : if (up_available) {
178 : const uint16_t *above_ref = ref - ref_stride;
179 : if (xd->mb_to_right_edge < 0) {
180 : /* slower path if the block needs border extension */
181 : if (x0 + bs <= frame_width) {
182 : memcpy(above_row, above_ref, bs * sizeof(above_row[0]));
183 : } else if (x0 <= frame_width) {
184 : const int r = frame_width - x0;
185 : memcpy(above_row, above_ref, r * sizeof(above_row[0]));
186 : vpx_memset16(above_row + r, above_row[r - 1], x0 + bs - frame_width);
187 : }
188 : } else {
189 : /* faster path if the block does not need extension */
190 : if (bs == 4 && right_available && left_available) {
191 : const_above_row = above_ref;
192 : } else {
193 : memcpy(above_row, above_ref, bs * sizeof(above_row[0]));
194 : }
195 : }
196 : above_row[-1] = left_available ? above_ref[-1] : (base + 1);
197 : } else {
198 : vpx_memset16(above_row, base - 1, bs);
199 : above_row[-1] = base - 1;
200 : }
201 : }
202 :
203 : // NEED_ABOVERIGHT
204 : if (need_aboveright) {
205 : if (up_available) {
206 : const uint16_t *above_ref = ref - ref_stride;
207 : if (xd->mb_to_right_edge < 0) {
208 : /* slower path if the block needs border extension */
209 : if (x0 + 2 * bs <= frame_width) {
210 : if (right_available && bs == 4) {
211 : memcpy(above_row, above_ref, 2 * bs * sizeof(above_row[0]));
212 : } else {
213 : memcpy(above_row, above_ref, bs * sizeof(above_row[0]));
214 : vpx_memset16(above_row + bs, above_row[bs - 1], bs);
215 : }
216 : } else if (x0 + bs <= frame_width) {
217 : const int r = frame_width - x0;
218 : if (right_available && bs == 4) {
219 : memcpy(above_row, above_ref, r * sizeof(above_row[0]));
220 : vpx_memset16(above_row + r, above_row[r - 1],
221 : x0 + 2 * bs - frame_width);
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 <= frame_width) {
227 : const int r = frame_width - x0;
228 : memcpy(above_row, above_ref, r * sizeof(above_row[0]));
229 : vpx_memset16(above_row + r, above_row[r - 1],
230 : x0 + 2 * bs - frame_width);
231 : }
232 : above_row[-1] = left_available ? above_ref[-1] : (base + 1);
233 : } else {
234 : /* faster path if the block does not need extension */
235 : if (bs == 4 && right_available && left_available) {
236 : const_above_row = above_ref;
237 : } else {
238 : memcpy(above_row, above_ref, bs * sizeof(above_row[0]));
239 : if (bs == 4 && right_available)
240 : memcpy(above_row + bs, above_ref + bs, bs * sizeof(above_row[0]));
241 : else
242 : vpx_memset16(above_row + bs, above_row[bs - 1], bs);
243 : above_row[-1] = left_available ? above_ref[-1] : (base + 1);
244 : }
245 : }
246 : } else {
247 : vpx_memset16(above_row, base - 1, bs * 2);
248 : above_row[-1] = base - 1;
249 : }
250 : }
251 :
252 : // predict
253 : if (mode == DC_PRED) {
254 : dc_pred_high[left_available][up_available][tx_size](
255 : dst, dst_stride, const_above_row, left_col, xd->bd);
256 : } else {
257 : pred_high[mode][tx_size](dst, dst_stride, const_above_row, left_col,
258 : xd->bd);
259 : }
260 : }
261 : #endif // CONFIG_VP9_HIGHBITDEPTH
262 :
263 0 : static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
264 : int ref_stride, uint8_t *dst, int dst_stride,
265 : PREDICTION_MODE mode, TX_SIZE tx_size,
266 : int up_available, int left_available,
267 : int right_available, int x, int y,
268 : int plane) {
269 : int i;
270 : DECLARE_ALIGNED(16, uint8_t, left_col[32]);
271 : DECLARE_ALIGNED(16, uint8_t, above_data[64 + 16]);
272 0 : uint8_t *above_row = above_data + 16;
273 0 : const uint8_t *const_above_row = above_row;
274 0 : const int bs = 4 << tx_size;
275 : int frame_width, frame_height;
276 : int x0, y0;
277 0 : const struct macroblockd_plane *const pd = &xd->plane[plane];
278 :
279 : // 127 127 127 .. 127 127 127 127 127 127
280 : // 129 A B .. Y Z
281 : // 129 C D .. W X
282 : // 129 E F .. U V
283 : // 129 G H .. S T T T T T
284 : // ..
285 :
286 : // Get current frame pointer, width and height.
287 0 : if (plane == 0) {
288 0 : frame_width = xd->cur_buf->y_width;
289 0 : frame_height = xd->cur_buf->y_height;
290 : } else {
291 0 : frame_width = xd->cur_buf->uv_width;
292 0 : frame_height = xd->cur_buf->uv_height;
293 : }
294 :
295 : // Get block position in current frame.
296 0 : x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
297 0 : y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
298 :
299 : // NEED_LEFT
300 0 : if (extend_modes[mode] & NEED_LEFT) {
301 0 : if (left_available) {
302 0 : if (xd->mb_to_bottom_edge < 0) {
303 : /* slower path if the block needs border extension */
304 0 : if (y0 + bs <= frame_height) {
305 0 : for (i = 0; i < bs; ++i) left_col[i] = ref[i * ref_stride - 1];
306 : } else {
307 0 : const int extend_bottom = frame_height - y0;
308 0 : for (i = 0; i < extend_bottom; ++i)
309 0 : left_col[i] = ref[i * ref_stride - 1];
310 0 : for (; i < bs; ++i)
311 0 : left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
312 : }
313 : } else {
314 : /* faster path if the block does not need extension */
315 0 : for (i = 0; i < bs; ++i) left_col[i] = ref[i * ref_stride - 1];
316 : }
317 : } else {
318 0 : memset(left_col, 129, bs);
319 : }
320 : }
321 :
322 : // NEED_ABOVE
323 0 : if (extend_modes[mode] & NEED_ABOVE) {
324 0 : if (up_available) {
325 0 : const uint8_t *above_ref = ref - ref_stride;
326 0 : if (xd->mb_to_right_edge < 0) {
327 : /* slower path if the block needs border extension */
328 0 : if (x0 + bs <= frame_width) {
329 0 : memcpy(above_row, above_ref, bs);
330 0 : } else if (x0 <= frame_width) {
331 0 : const int r = frame_width - x0;
332 0 : memcpy(above_row, above_ref, r);
333 0 : memset(above_row + r, above_row[r - 1], x0 + bs - frame_width);
334 : }
335 : } else {
336 : /* faster path if the block does not need extension */
337 0 : if (bs == 4 && right_available && left_available) {
338 0 : const_above_row = above_ref;
339 : } else {
340 0 : memcpy(above_row, above_ref, bs);
341 : }
342 : }
343 0 : above_row[-1] = left_available ? above_ref[-1] : 129;
344 : } else {
345 0 : memset(above_row, 127, bs);
346 0 : above_row[-1] = 127;
347 : }
348 : }
349 :
350 : // NEED_ABOVERIGHT
351 0 : if (extend_modes[mode] & NEED_ABOVERIGHT) {
352 0 : if (up_available) {
353 0 : const uint8_t *above_ref = ref - ref_stride;
354 0 : if (xd->mb_to_right_edge < 0) {
355 : /* slower path if the block needs border extension */
356 0 : if (x0 + 2 * bs <= frame_width) {
357 0 : if (right_available && bs == 4) {
358 0 : memcpy(above_row, above_ref, 2 * bs);
359 : } else {
360 0 : memcpy(above_row, above_ref, bs);
361 0 : memset(above_row + bs, above_row[bs - 1], bs);
362 : }
363 0 : } else if (x0 + bs <= frame_width) {
364 0 : const int r = frame_width - x0;
365 0 : if (right_available && bs == 4) {
366 0 : memcpy(above_row, above_ref, r);
367 0 : memset(above_row + r, above_row[r - 1], x0 + 2 * bs - frame_width);
368 : } else {
369 0 : memcpy(above_row, above_ref, bs);
370 0 : memset(above_row + bs, above_row[bs - 1], bs);
371 : }
372 0 : } else if (x0 <= frame_width) {
373 0 : const int r = frame_width - x0;
374 0 : memcpy(above_row, above_ref, r);
375 0 : memset(above_row + r, above_row[r - 1], x0 + 2 * bs - frame_width);
376 : }
377 : } else {
378 : /* faster path if the block does not need extension */
379 0 : if (bs == 4 && right_available && left_available) {
380 0 : const_above_row = above_ref;
381 : } else {
382 0 : memcpy(above_row, above_ref, bs);
383 0 : if (bs == 4 && right_available)
384 0 : memcpy(above_row + bs, above_ref + bs, bs);
385 : else
386 0 : memset(above_row + bs, above_row[bs - 1], bs);
387 : }
388 : }
389 0 : above_row[-1] = left_available ? above_ref[-1] : 129;
390 : } else {
391 0 : memset(above_row, 127, bs * 2);
392 0 : above_row[-1] = 127;
393 : }
394 : }
395 :
396 : // predict
397 0 : if (mode == DC_PRED) {
398 0 : dc_pred[left_available][up_available][tx_size](dst, dst_stride,
399 : const_above_row, left_col);
400 : } else {
401 0 : pred[mode][tx_size](dst, dst_stride, const_above_row, left_col);
402 : }
403 0 : }
404 :
405 0 : void vp9_predict_intra_block(const MACROBLOCKD *xd, int bwl_in, TX_SIZE tx_size,
406 : PREDICTION_MODE mode, const uint8_t *ref,
407 : int ref_stride, uint8_t *dst, int dst_stride,
408 : int aoff, int loff, int plane) {
409 0 : const int bw = (1 << bwl_in);
410 0 : const int txw = (1 << tx_size);
411 0 : const int have_top = loff || (xd->above_mi != NULL);
412 0 : const int have_left = aoff || (xd->left_mi != NULL);
413 0 : const int have_right = (aoff + txw) < bw;
414 0 : const int x = aoff * 4;
415 0 : const int y = loff * 4;
416 :
417 : #if CONFIG_VP9_HIGHBITDEPTH
418 : if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
419 : build_intra_predictors_high(xd, ref, ref_stride, dst, dst_stride, mode,
420 : tx_size, have_top, have_left, have_right, x, y,
421 : plane, xd->bd);
422 : return;
423 : }
424 : #endif
425 0 : build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size,
426 : have_top, have_left, have_right, x, y, plane);
427 0 : }
428 :
429 0 : void vp9_init_intra_predictors(void) {
430 0 : once(vp9_init_intra_predictors_internal);
431 0 : }
|
