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
13 #include "vpx_mem/vpx_mem.h"
14 #include "vpx_ports/vpx_once.h"
15
16 #include "./vp9_rtcd.h"
17
18 #include "vp9/common/vp9_reconintra.h"
19 #include "vp9/common/vp9_onyxc_int.h"
20
21 const TX_TYPE intra_mode_to_tx_type_lookup[INTRA_MODES] = {
22 DCT_DCT, // DC
23 ADST_DCT, // V
24 DCT_ADST, // H
25 DCT_DCT, // D45
26 ADST_ADST, // D135
27 ADST_DCT, // D117
28 DCT_ADST, // D153
29 DCT_ADST, // D207
30 ADST_DCT, // D63
31 ADST_ADST, // TM
32 };
33
34 #define intra_pred_sized(type, size) \
35 void vp9_##type##_predictor_##size##x##size##_c(uint8_t *dst, \
36 ptrdiff_t stride, \
37 const uint8_t *above, \
38 const uint8_t *left) { \
39 type##_predictor(dst, stride, size, above, left); \
40 }
41
42 #define intra_pred_allsizes(type) \
43 intra_pred_sized(type, 4) \
44 intra_pred_sized(type, 8) \
45 intra_pred_sized(type, 16) \
46 intra_pred_sized(type, 32)
47
d207_predictor(uint8_t * dst,ptrdiff_t stride,int bs,const uint8_t * above,const uint8_t * left)48 static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
49 const uint8_t *above, const uint8_t *left) {
50 int r, c;
51 (void)above;
52 // first column
53 for (r = 0; r < bs - 1; ++r)
54 dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1);
55 dst[(bs - 1) * stride] = left[bs - 1];
56 dst++;
57
58 // second column
59 for (r = 0; r < bs - 2; ++r)
60 dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1] * 2 +
61 left[r + 2], 2);
62 dst[(bs - 2) * stride] = ROUND_POWER_OF_TWO(left[bs - 2] +
63 left[bs - 1] * 3, 2);
64 dst[(bs - 1) * stride] = left[bs - 1];
65 dst++;
66
67 // rest of last row
68 for (c = 0; c < bs - 2; ++c)
69 dst[(bs - 1) * stride + c] = left[bs - 1];
70
71 for (r = bs - 2; r >= 0; --r)
72 for (c = 0; c < bs - 2; ++c)
73 dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
74 }
intra_pred_allsizes(d207)75 intra_pred_allsizes(d207)
76
77 static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
78 const uint8_t *above, const uint8_t *left) {
79 int r, c;
80 (void)left;
81 for (r = 0; r < bs; ++r) {
82 for (c = 0; c < bs; ++c)
83 dst[c] = r & 1 ? ROUND_POWER_OF_TWO(above[r/2 + c] +
84 above[r/2 + c + 1] * 2 +
85 above[r/2 + c + 2], 2)
86 : ROUND_POWER_OF_TWO(above[r/2 + c] +
87 above[r/2 + c + 1], 1);
88 dst += stride;
89 }
90 }
intra_pred_allsizes(d63)91 intra_pred_allsizes(d63)
92
93 static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
94 const uint8_t *above, const uint8_t *left) {
95 int r, c;
96 (void)left;
97 for (r = 0; r < bs; ++r) {
98 for (c = 0; c < bs; ++c)
99 dst[c] = r + c + 2 < bs * 2 ? ROUND_POWER_OF_TWO(above[r + c] +
100 above[r + c + 1] * 2 +
101 above[r + c + 2], 2)
102 : above[bs * 2 - 1];
103 dst += stride;
104 }
105 }
intra_pred_allsizes(d45)106 intra_pred_allsizes(d45)
107
108 static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
109 const uint8_t *above, const uint8_t *left) {
110 int r, c;
111 // first row
112 for (c = 0; c < bs; c++)
113 dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1);
114 dst += stride;
115
116 // second row
117 dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
118 for (c = 1; c < bs; c++)
119 dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
120 dst += stride;
121
122 // the rest of first col
123 dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
124 for (r = 3; r < bs; ++r)
125 dst[(r - 2) * stride] = ROUND_POWER_OF_TWO(left[r - 3] + left[r - 2] * 2 +
126 left[r - 1], 2);
127
128 // the rest of the block
129 for (r = 2; r < bs; ++r) {
130 for (c = 1; c < bs; c++)
131 dst[c] = dst[-2 * stride + c - 1];
132 dst += stride;
133 }
134 }
intra_pred_allsizes(d117)135 intra_pred_allsizes(d117)
136
137 static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
138 const uint8_t *above, const uint8_t *left) {
139 int r, c;
140 dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
141 for (c = 1; c < bs; c++)
142 dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
143
144 dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
145 for (r = 2; r < bs; ++r)
146 dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
147 left[r], 2);
148
149 dst += stride;
150 for (r = 1; r < bs; ++r) {
151 for (c = 1; c < bs; c++)
152 dst[c] = dst[-stride + c - 1];
153 dst += stride;
154 }
155 }
intra_pred_allsizes(d135)156 intra_pred_allsizes(d135)
157
158 static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
159 const uint8_t *above, const uint8_t *left) {
160 int r, c;
161 dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1);
162 for (r = 1; r < bs; r++)
163 dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1);
164 dst++;
165
166 dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
167 dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
168 for (r = 2; r < bs; r++)
169 dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
170 left[r], 2);
171 dst++;
172
173 for (c = 0; c < bs - 2; c++)
174 dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c] * 2 + above[c + 1], 2);
175 dst += stride;
176
177 for (r = 1; r < bs; ++r) {
178 for (c = 0; c < bs - 2; c++)
179 dst[c] = dst[-stride + c - 2];
180 dst += stride;
181 }
182 }
intra_pred_allsizes(d153)183 intra_pred_allsizes(d153)
184
185 static INLINE void v_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
186 const uint8_t *above, const uint8_t *left) {
187 int r;
188 (void)left;
189 for (r = 0; r < bs; r++) {
190 vpx_memcpy(dst, above, bs);
191 dst += stride;
192 }
193 }
intra_pred_allsizes(v)194 intra_pred_allsizes(v)
195
196 static INLINE void h_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
197 const uint8_t *above, const uint8_t *left) {
198 int r;
199 (void)above;
200 for (r = 0; r < bs; r++) {
201 vpx_memset(dst, left[r], bs);
202 dst += stride;
203 }
204 }
intra_pred_allsizes(h)205 intra_pred_allsizes(h)
206
207 static INLINE void tm_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
208 const uint8_t *above, const uint8_t *left) {
209 int r, c;
210 int ytop_left = above[-1];
211
212 for (r = 0; r < bs; r++) {
213 for (c = 0; c < bs; c++)
214 dst[c] = clip_pixel(left[r] + above[c] - ytop_left);
215 dst += stride;
216 }
217 }
intra_pred_allsizes(tm)218 intra_pred_allsizes(tm)
219
220 static INLINE void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
221 const uint8_t *above, const uint8_t *left) {
222 int r;
223 (void)above;
224 (void)left;
225 for (r = 0; r < bs; r++) {
226 vpx_memset(dst, 128, bs);
227 dst += stride;
228 }
229 }
intra_pred_allsizes(dc_128)230 intra_pred_allsizes(dc_128)
231
232 static INLINE void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
233 const uint8_t *above,
234 const uint8_t *left) {
235 int i, r, expected_dc, sum = 0;
236 (void)above;
237 for (i = 0; i < bs; i++)
238 sum += left[i];
239 expected_dc = (sum + (bs >> 1)) / bs;
240
241 for (r = 0; r < bs; r++) {
242 vpx_memset(dst, expected_dc, bs);
243 dst += stride;
244 }
245 }
intra_pred_allsizes(dc_left)246 intra_pred_allsizes(dc_left)
247
248 static INLINE void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
249 const uint8_t *above, const uint8_t *left) {
250 int i, r, expected_dc, sum = 0;
251 (void)left;
252 for (i = 0; i < bs; i++)
253 sum += above[i];
254 expected_dc = (sum + (bs >> 1)) / bs;
255
256 for (r = 0; r < bs; r++) {
257 vpx_memset(dst, expected_dc, bs);
258 dst += stride;
259 }
260 }
intra_pred_allsizes(dc_top)261 intra_pred_allsizes(dc_top)
262
263 static INLINE void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
264 const uint8_t *above, const uint8_t *left) {
265 int i, r, expected_dc, sum = 0;
266 const int count = 2 * bs;
267
268 for (i = 0; i < bs; i++) {
269 sum += above[i];
270 sum += left[i];
271 }
272
273 expected_dc = (sum + (count >> 1)) / count;
274
275 for (r = 0; r < bs; r++) {
276 vpx_memset(dst, expected_dc, bs);
277 dst += stride;
278 }
279 }
280 intra_pred_allsizes(dc)
281 #undef intra_pred_allsizes
282
283 typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride,
284 const uint8_t *above, const uint8_t *left);
285
286 static intra_pred_fn pred[INTRA_MODES][4];
287 static intra_pred_fn dc_pred[2][2][4];
288
init_intra_pred_fn_ptrs(void)289 static void init_intra_pred_fn_ptrs(void) {
290 #define intra_pred_allsizes(l, type) \
291 l[0] = vp9_##type##_predictor_4x4; \
292 l[1] = vp9_##type##_predictor_8x8; \
293 l[2] = vp9_##type##_predictor_16x16; \
294 l[3] = vp9_##type##_predictor_32x32
295
296 intra_pred_allsizes(pred[V_PRED], v);
297 intra_pred_allsizes(pred[H_PRED], h);
298 intra_pred_allsizes(pred[D207_PRED], d207);
299 intra_pred_allsizes(pred[D45_PRED], d45);
300 intra_pred_allsizes(pred[D63_PRED], d63);
301 intra_pred_allsizes(pred[D117_PRED], d117);
302 intra_pred_allsizes(pred[D135_PRED], d135);
303 intra_pred_allsizes(pred[D153_PRED], d153);
304 intra_pred_allsizes(pred[TM_PRED], tm);
305
306 intra_pred_allsizes(dc_pred[0][0], dc_128);
307 intra_pred_allsizes(dc_pred[0][1], dc_top);
308 intra_pred_allsizes(dc_pred[1][0], dc_left);
309 intra_pred_allsizes(dc_pred[1][1], dc);
310
311 #undef intra_pred_allsizes
312 }
313
build_intra_predictors(const MACROBLOCKD * xd,const uint8_t * ref,int ref_stride,uint8_t * dst,int dst_stride,MB_PREDICTION_MODE mode,TX_SIZE tx_size,int up_available,int left_available,int right_available,int x,int y,int plane)314 static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
315 int ref_stride, uint8_t *dst, int dst_stride,
316 MB_PREDICTION_MODE mode, TX_SIZE tx_size,
317 int up_available, int left_available,
318 int right_available, int x, int y,
319 int plane) {
320 int i;
321 DECLARE_ALIGNED_ARRAY(16, uint8_t, left_col, 64);
322 DECLARE_ALIGNED_ARRAY(16, uint8_t, above_data, 128 + 16);
323 uint8_t *above_row = above_data + 16;
324 const uint8_t *const_above_row = above_row;
325 const int bs = 4 << tx_size;
326 int frame_width, frame_height;
327 int x0, y0;
328 const struct macroblockd_plane *const pd = &xd->plane[plane];
329
330 // 127 127 127 .. 127 127 127 127 127 127
331 // 129 A B .. Y Z
332 // 129 C D .. W X
333 // 129 E F .. U V
334 // 129 G H .. S T T T T T
335 // ..
336
337 once(init_intra_pred_fn_ptrs);
338
339 // Get current frame pointer, width and height.
340 if (plane == 0) {
341 frame_width = xd->cur_buf->y_width;
342 frame_height = xd->cur_buf->y_height;
343 } else {
344 frame_width = xd->cur_buf->uv_width;
345 frame_height = xd->cur_buf->uv_height;
346 }
347
348 // Get block position in current frame.
349 x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
350 y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
351
352 vpx_memset(left_col, 129, 64);
353
354 // left
355 if (left_available) {
356 if (xd->mb_to_bottom_edge < 0) {
357 /* slower path if the block needs border extension */
358 if (y0 + bs <= frame_height) {
359 for (i = 0; i < bs; ++i)
360 left_col[i] = ref[i * ref_stride - 1];
361 } else {
362 const int extend_bottom = frame_height - y0;
363 for (i = 0; i < extend_bottom; ++i)
364 left_col[i] = ref[i * ref_stride - 1];
365 for (; i < bs; ++i)
366 left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
367 }
368 } else {
369 /* faster path if the block does not need extension */
370 for (i = 0; i < bs; ++i)
371 left_col[i] = ref[i * ref_stride - 1];
372 }
373 }
374
375 // TODO(hkuang) do not extend 2*bs pixels for all modes.
376 // above
377 if (up_available) {
378 const uint8_t *above_ref = ref - ref_stride;
379 if (xd->mb_to_right_edge < 0) {
380 /* slower path if the block needs border extension */
381 if (x0 + 2 * bs <= frame_width) {
382 if (right_available && bs == 4) {
383 vpx_memcpy(above_row, above_ref, 2 * bs);
384 } else {
385 vpx_memcpy(above_row, above_ref, bs);
386 vpx_memset(above_row + bs, above_row[bs - 1], bs);
387 }
388 } else if (x0 + bs <= frame_width) {
389 const int r = frame_width - x0;
390 if (right_available && bs == 4) {
391 vpx_memcpy(above_row, above_ref, r);
392 vpx_memset(above_row + r, above_row[r - 1],
393 x0 + 2 * bs - frame_width);
394 } else {
395 vpx_memcpy(above_row, above_ref, bs);
396 vpx_memset(above_row + bs, above_row[bs - 1], bs);
397 }
398 } else if (x0 <= frame_width) {
399 const int r = frame_width - x0;
400 if (right_available && bs == 4) {
401 vpx_memcpy(above_row, above_ref, r);
402 vpx_memset(above_row + r, above_row[r - 1],
403 x0 + 2 * bs - frame_width);
404 } else {
405 vpx_memcpy(above_row, above_ref, r);
406 vpx_memset(above_row + r, above_row[r - 1],
407 x0 + 2 * bs - frame_width);
408 }
409 }
410 above_row[-1] = left_available ? above_ref[-1] : 129;
411 } else {
412 /* faster path if the block does not need extension */
413 if (bs == 4 && right_available && left_available) {
414 const_above_row = above_ref;
415 } else {
416 vpx_memcpy(above_row, above_ref, bs);
417 if (bs == 4 && right_available)
418 vpx_memcpy(above_row + bs, above_ref + bs, bs);
419 else
420 vpx_memset(above_row + bs, above_row[bs - 1], bs);
421 above_row[-1] = left_available ? above_ref[-1] : 129;
422 }
423 }
424 } else {
425 vpx_memset(above_row, 127, bs * 2);
426 above_row[-1] = 127;
427 }
428
429 // predict
430 if (mode == DC_PRED) {
431 dc_pred[left_available][up_available][tx_size](dst, dst_stride,
432 const_above_row, left_col);
433 } else {
434 pred[mode][tx_size](dst, dst_stride, const_above_row, left_col);
435 }
436 }
437
vp9_predict_intra_block(const MACROBLOCKD * xd,int block_idx,int bwl_in,TX_SIZE tx_size,MB_PREDICTION_MODE mode,const uint8_t * ref,int ref_stride,uint8_t * dst,int dst_stride,int aoff,int loff,int plane)438 void vp9_predict_intra_block(const MACROBLOCKD *xd, int block_idx, int bwl_in,
439 TX_SIZE tx_size, MB_PREDICTION_MODE mode,
440 const uint8_t *ref, int ref_stride,
441 uint8_t *dst, int dst_stride,
442 int aoff, int loff, int plane) {
443 const int bwl = bwl_in - tx_size;
444 const int wmask = (1 << bwl) - 1;
445 const int have_top = (block_idx >> bwl) || xd->up_available;
446 const int have_left = (block_idx & wmask) || xd->left_available;
447 const int have_right = ((block_idx & wmask) != wmask);
448 const int x = aoff * 4;
449 const int y = loff * 4;
450
451 assert(bwl >= 0);
452 build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size,
453 have_top, have_left, have_right, x, y, plane);
454 }
455