4 * Copyright (C) 2012 - 2013 Guillaume Martres
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "libavutil/pixdesc.h"
25 #include "bit_depth_template.c"
28 #define POS(x, y) src[(x) + stride * (y)]
30 static av_always_inline void FUNC(intra_pred)(HEVCContext *s, int x0, int y0,
31 int log2_size, int c_idx)
34 ((x) >> s->sps->log2_min_pu_size)
36 (s->ref->tab_mvf[(x) + (y) * min_pu_width])
37 #define MVF_PU(x, y) \
38 MVF(PU(x0 + ((x) << hshift)), PU(y0 + ((y) << vshift)))
39 #define IS_INTRA(x, y) \
40 (MVF_PU(x, y).pred_flag == PF_INTRA)
41 #define MIN_TB_ADDR_ZS(x, y) \
42 s->pps->min_tb_addr_zs[(y) * s->sps->min_tb_width + (x)]
44 #define EXTEND(ptr, val, len) \
46 pixel4 pix = PIXEL_SPLAT_X4(val); \
47 for (i = 0; i < (len); i += 4) \
48 AV_WN4P(ptr + i, pix); \
51 #define EXTEND_RIGHT_CIP(ptr, start, length) \
52 for (i = start; i < (start) + (length); i += 4) \
53 if (!IS_INTRA(i, -1)) \
54 AV_WN4P(&ptr[i], a); \
56 a = PIXEL_SPLAT_X4(ptr[i+3])
57 #define EXTEND_LEFT_CIP(ptr, start, length) \
58 for (i = start; i > (start) - (length); i--) \
59 if (!IS_INTRA(i - 1, -1)) \
61 #define EXTEND_UP_CIP(ptr, start, length) \
62 for (i = (start); i > (start) - (length); i -= 4) \
63 if (!IS_INTRA(-1, i - 3)) \
64 AV_WN4P(&ptr[i - 3], a); \
66 a = PIXEL_SPLAT_X4(ptr[i - 3])
67 #define EXTEND_DOWN_CIP(ptr, start, length) \
68 for (i = start; i < (start) + (length); i += 4) \
69 if (!IS_INTRA(-1, i)) \
70 AV_WN4P(&ptr[i], a); \
72 a = PIXEL_SPLAT_X4(ptr[i + 3])
74 HEVCLocalContext *lc = s->HEVClc;
76 int hshift = s->sps->hshift[c_idx];
77 int vshift = s->sps->vshift[c_idx];
78 int size = (1 << log2_size);
79 int size_in_luma_h = size << hshift;
80 int size_in_tbs_h = size_in_luma_h >> s->sps->log2_min_tb_size;
81 int size_in_luma_v = size << vshift;
82 int size_in_tbs_v = size_in_luma_v >> s->sps->log2_min_tb_size;
85 int x_tb = x0 >> s->sps->log2_min_tb_size;
86 int y_tb = y0 >> s->sps->log2_min_tb_size;
87 int cur_tb_addr = MIN_TB_ADDR_ZS(x_tb, y_tb);
89 ptrdiff_t stride = s->frame->linesize[c_idx] / sizeof(pixel);
90 pixel *src = (pixel*)s->frame->data[c_idx] + x + y * stride;
92 int min_pu_width = s->sps->min_pu_width;
94 enum IntraPredMode mode = c_idx ? lc->pu.intra_pred_mode_c :
95 lc->tu.cur_intra_pred_mode;
97 pixel left_array[2 * MAX_TB_SIZE + 1];
98 pixel filtered_left_array[2 * MAX_TB_SIZE + 1];
99 pixel top_array[2 * MAX_TB_SIZE + 1];
100 pixel filtered_top_array[2 * MAX_TB_SIZE + 1];
102 pixel *left = left_array + 1;
103 pixel *top = top_array + 1;
104 pixel *filtered_left = filtered_left_array + 1;
105 pixel *filtered_top = filtered_top_array + 1;
107 int cand_bottom_left = lc->na.cand_bottom_left && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb - 1, y_tb + size_in_tbs_v);
108 int cand_left = lc->na.cand_left;
109 int cand_up_left = lc->na.cand_up_left;
110 int cand_up = lc->na.cand_up;
111 int cand_up_right = lc->na.cand_up_right && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb + size_in_tbs_h, y_tb - 1);
113 int bottom_left_size = (FFMIN(y0 + 2 * size_in_luma_v, s->sps->height) -
114 (y0 + size_in_luma_v)) >> vshift;
115 int top_right_size = (FFMIN(x0 + 2 * size_in_luma_h, s->sps->width) -
116 (x0 + size_in_luma_h)) >> hshift;
118 if (s->pps->constrained_intra_pred_flag == 1) {
119 int size_in_luma_pu_v = PU(size_in_luma_v);
120 int size_in_luma_pu_h = PU(size_in_luma_h);
121 int on_pu_edge_x = !(x0 & ((1 << s->sps->log2_min_pu_size) - 1));
122 int on_pu_edge_y = !(y0 & ((1 << s->sps->log2_min_pu_size) - 1));
123 if (!size_in_luma_pu_h)
125 if (cand_bottom_left == 1 && on_pu_edge_x) {
126 int x_left_pu = PU(x0 - 1);
127 int y_bottom_pu = PU(y0 + size_in_luma_v);
128 int max = FFMIN(size_in_luma_pu_v, s->sps->min_pu_height - y_bottom_pu);
129 cand_bottom_left = 0;
130 for (i = 0; i < max; i += 2)
131 cand_bottom_left |= (MVF(x_left_pu, y_bottom_pu + i).pred_flag == PF_INTRA);
133 if (cand_left == 1 && on_pu_edge_x) {
134 int x_left_pu = PU(x0 - 1);
135 int y_left_pu = PU(y0);
136 int max = FFMIN(size_in_luma_pu_v, s->sps->min_pu_height - y_left_pu);
138 for (i = 0; i < max; i += 2)
139 cand_left |= (MVF(x_left_pu, y_left_pu + i).pred_flag == PF_INTRA);
141 if (cand_up_left == 1) {
142 int x_left_pu = PU(x0 - 1);
143 int y_top_pu = PU(y0 - 1);
144 cand_up_left = MVF(x_left_pu, y_top_pu).pred_flag == PF_INTRA;
146 if (cand_up == 1 && on_pu_edge_y) {
147 int x_top_pu = PU(x0);
148 int y_top_pu = PU(y0 - 1);
149 int max = FFMIN(size_in_luma_pu_h, s->sps->min_pu_width - x_top_pu);
151 for (i = 0; i < max; i += 2)
152 cand_up |= (MVF(x_top_pu + i, y_top_pu).pred_flag == PF_INTRA);
154 if (cand_up_right == 1 && on_pu_edge_y) {
155 int y_top_pu = PU(y0 - 1);
156 int x_right_pu = PU(x0 + size_in_luma_h);
157 int max = FFMIN(size_in_luma_pu_h, s->sps->min_pu_width - x_right_pu);
159 for (i = 0; i < max; i += 2)
160 cand_up_right |= (MVF(x_right_pu + i, y_top_pu).pred_flag == PF_INTRA);
162 memset(left, 128, 2 * MAX_TB_SIZE*sizeof(pixel));
163 memset(top , 128, 2 * MAX_TB_SIZE*sizeof(pixel));
167 left[-1] = POS(-1, -1);
171 memcpy(top, src - stride, size * sizeof(pixel));
173 memcpy(top + size, src - stride + size, size * sizeof(pixel));
174 EXTEND(top + size + top_right_size, POS(size + top_right_size - 1, -1),
175 size - top_right_size);
178 for (i = 0; i < size; i++)
179 left[i] = POS(-1, i);
180 if (cand_bottom_left) {
181 for (i = size; i < size + bottom_left_size; i++)
182 left[i] = POS(-1, i);
183 EXTEND(left + size + bottom_left_size, POS(-1, size + bottom_left_size - 1),
184 size - bottom_left_size);
187 if (s->pps->constrained_intra_pred_flag == 1) {
188 if (cand_bottom_left || cand_left || cand_up_left || cand_up || cand_up_right) {
189 int size_max_x = x0 + ((2 * size) << hshift) < s->sps->width ?
190 2 * size : (s->sps->width - x0) >> hshift;
191 int size_max_y = y0 + ((2 * size) << vshift) < s->sps->height ?
192 2 * size : (s->sps->height - y0) >> vshift;
193 int j = size + (cand_bottom_left? bottom_left_size: 0) -1;
194 if (!cand_up_right) {
195 size_max_x = x0 + ((size) << hshift) < s->sps->width ?
196 size : (s->sps->width - x0) >> hshift;
198 if (!cand_bottom_left) {
199 size_max_y = y0 + (( size) << vshift) < s->sps->height ?
200 size : (s->sps->height - y0) >> vshift;
202 if (cand_bottom_left || cand_left || cand_up_left) {
203 while (j > -1 && !IS_INTRA(-1, j))
205 if (!IS_INTRA(-1, j)) {
207 while (j < size_max_x && !IS_INTRA(j, -1))
209 EXTEND_LEFT_CIP(top, j, j + 1);
215 while (j < size_max_x && !IS_INTRA(j, -1))
219 EXTEND_LEFT_CIP(top, j, j + 1);
221 EXTEND_LEFT_CIP(top, j, j);
228 if (cand_bottom_left || cand_left) {
229 a = PIXEL_SPLAT_X4(left[-1]);
230 EXTEND_DOWN_CIP(left, 0, size_max_y);
233 EXTEND(left, left[-1], size);
234 if (!cand_bottom_left)
235 EXTEND(left + size, left[size - 1], size);
236 if (x0 != 0 && y0 != 0) {
237 a = PIXEL_SPLAT_X4(left[size_max_y - 1]);
238 EXTEND_UP_CIP(left, size_max_y - 1, size_max_y);
239 if (!IS_INTRA(-1, - 1))
241 } else if (x0 == 0) {
242 a = PIXEL_SPLAT_X4(left[size_max_y - 1]);
243 EXTEND(left, 0, size_max_y);
245 a = PIXEL_SPLAT_X4(left[size_max_y - 1]);
246 EXTEND_UP_CIP(left, size_max_y - 1, size_max_y);
250 a = PIXEL_SPLAT_X4(left[-1]);
251 EXTEND_RIGHT_CIP(top, 0, size_max_x);
255 // Infer the unavailable samples
256 if (!cand_bottom_left) {
258 EXTEND(left + size, left[size - 1], size);
259 } else if (cand_up_left) {
260 EXTEND(left, left[-1], 2 * size);
262 } else if (cand_up) {
264 EXTEND(left, left[-1], 2 * size);
267 } else if (cand_up_right) {
268 EXTEND(top, top[size], size);
269 left[-1] = top[size];
270 EXTEND(left, left[-1], 2 * size);
274 } else { // No samples available
275 left[-1] = (1 << (BIT_DEPTH - 1));
276 EXTEND(top, left[-1], 2 * size);
277 EXTEND(left, left[-1], 2 * size);
282 EXTEND(left, left[size], size);
287 EXTEND(top, left[-1], size);
289 EXTEND(top + size, top[size - 1], size);
295 if (mode != INTRA_DC && size != 4){
296 int intra_hor_ver_dist_thresh[] = { 7, 1, 0 };
297 int min_dist_vert_hor = FFMIN(FFABS((int)(mode - 26U)),
298 FFABS((int)(mode - 10U)));
299 if (min_dist_vert_hor > intra_hor_ver_dist_thresh[log2_size - 3]) {
300 int threshold = 1 << (BIT_DEPTH - 5);
301 if (s->sps->sps_strong_intra_smoothing_enable_flag && c_idx == 0 &&
303 FFABS(top[-1] + top[63] - 2 * top[31]) < threshold &&
304 FFABS(left[-1] + left[63] - 2 * left[31]) < threshold) {
305 // We can't just overwrite values in top because it could be
306 // a pointer into src
307 filtered_top[-1] = top[-1];
308 filtered_top[63] = top[63];
309 for (i = 0; i < 63; i++)
310 filtered_top[i] = ((64 - (i + 1)) * top[-1] +
311 (i + 1) * top[63] + 32) >> 6;
312 for (i = 0; i < 63; i++)
313 left[i] = ((64 - (i + 1)) * left[-1] +
314 (i + 1) * left[63] + 32) >> 6;
317 filtered_left[2 * size - 1] = left[2 * size - 1];
318 filtered_top[2 * size - 1] = top[2 * size - 1];
319 for (i = 2 * size - 2; i >= 0; i--)
320 filtered_left[i] = (left[i + 1] + 2 * left[i] +
321 left[i - 1] + 2) >> 2;
323 filtered_left[-1] = (left[0] + 2 * left[-1] + top[0] + 2) >> 2;
324 for (i = 2 * size - 2; i >= 0; i--)
325 filtered_top[i] = (top[i + 1] + 2 * top[i] +
326 top[i - 1] + 2) >> 2;
327 left = filtered_left;
336 s->hpc.pred_planar[log2_size - 2]((uint8_t *)src, (uint8_t *)top,
337 (uint8_t *)left, stride);
340 s->hpc.pred_dc((uint8_t *)src, (uint8_t *)top,
341 (uint8_t *)left, stride, log2_size, c_idx);
344 s->hpc.pred_angular[log2_size - 2]((uint8_t *)src, (uint8_t *)top,
345 (uint8_t *)left, stride, c_idx,
351 #define INTRA_PRED(size) \
352 static void FUNC(intra_pred_ ## size)(HEVCContext *s, int x0, int y0, int c_idx) \
354 FUNC(intra_pred)(s, x0, y0, size, c_idx); \
364 static av_always_inline void FUNC(pred_planar)(uint8_t *_src, const uint8_t *_top,
365 const uint8_t *_left, ptrdiff_t stride,
369 pixel *src = (pixel *)_src;
370 const pixel *top = (const pixel *)_top;
371 const pixel *left = (const pixel *)_left;
372 int size = 1 << trafo_size;
373 for (y = 0; y < size; y++)
374 for (x = 0; x < size; x++)
375 POS(x, y) = ((size - 1 - x) * left[y] + (x + 1) * top[size] +
376 (size - 1 - y) * top[x] + (y + 1) * left[size] + size) >> (trafo_size + 1);
379 #define PRED_PLANAR(size)\
380 static void FUNC(pred_planar_ ## size)(uint8_t *src, const uint8_t *top, \
381 const uint8_t *left, ptrdiff_t stride) \
383 FUNC(pred_planar)(src, top, left, stride, size + 2); \
393 static void FUNC(pred_dc)(uint8_t *_src, const uint8_t *_top,
394 const uint8_t *_left,
395 ptrdiff_t stride, int log2_size, int c_idx)
398 int size = (1 << log2_size);
399 pixel *src = (pixel *)_src;
400 const pixel *top = (const pixel *)_top;
401 const pixel *left = (const pixel *)_left;
404 for (i = 0; i < size; i++)
405 dc += left[i] + top[i];
407 dc >>= log2_size + 1;
409 a = PIXEL_SPLAT_X4(dc);
411 for (i = 0; i < size; i++)
412 for (j = 0; j < size; j+=4)
413 AV_WN4P(&POS(j, i), a);
415 if (c_idx == 0 && size < 32) {
416 POS(0, 0) = (left[0] + 2 * dc + top[0] + 2) >> 2;
417 for (x = 1; x < size; x++)
418 POS(x, 0) = (top[x] + 3 * dc + 2) >> 2;
419 for (y = 1; y < size; y++)
420 POS(0, y) = (left[y] + 3 * dc + 2) >> 2;
424 static av_always_inline void FUNC(pred_angular)(uint8_t *_src,
426 const uint8_t *_left,
427 ptrdiff_t stride, int c_idx,
431 pixel *src = (pixel *)_src;
432 const pixel *top = (const pixel *)_top;
433 const pixel *left = (const pixel *)_left;
435 static const int intra_pred_angle[] = {
436 32, 26, 21, 17, 13, 9, 5, 2, 0, -2, -5, -9, -13, -17, -21, -26, -32,
437 -26, -21, -17, -13, -9, -5, -2, 0, 2, 5, 9, 13, 17, 21, 26, 32
439 static const int inv_angle[] = {
440 -4096, -1638, -910, -630, -482, -390, -315, -256, -315, -390, -482,
441 -630, -910, -1638, -4096
444 int angle = intra_pred_angle[mode - 2];
445 pixel ref_array[3 * MAX_TB_SIZE + 4];
446 pixel *ref_tmp = ref_array + size;
448 int last = (size * angle) >> 5;
452 if (angle < 0 && last < -1) {
453 for (x = 0; x <= size; x += 4)
454 AV_WN4P(&ref_tmp[x], AV_RN4P(&top[x - 1]));
455 for (x = last; x <= -1; x++)
456 ref_tmp[x] = left[-1 + ((x * inv_angle[mode - 11] + 128) >> 8)];
460 for (y = 0; y < size; y++) {
461 int idx = ((y + 1) * angle) >> 5;
462 int fact = ((y + 1) * angle) & 31;
464 for (x = 0; x < size; x += 4) {
465 POS(x , y) = ((32 - fact) * ref[x + idx + 1] +
466 fact * ref[x + idx + 2] + 16) >> 5;
467 POS(x + 1, y) = ((32 - fact) * ref[x + 1 + idx + 1] +
468 fact * ref[x + 1 + idx + 2] + 16) >> 5;
469 POS(x + 2, y) = ((32 - fact) * ref[x + 2 + idx + 1] +
470 fact * ref[x + 2 + idx + 2] + 16) >> 5;
471 POS(x + 3, y) = ((32 - fact) * ref[x + 3 + idx + 1] +
472 fact * ref[x + 3 + idx + 2] + 16) >> 5;
475 for (x = 0; x < size; x += 4)
476 AV_WN4P(&POS(x, y), AV_RN4P(&ref[x + idx + 1]));
479 if (mode == 26 && c_idx == 0 && size < 32) {
480 for (y = 0; y < size; y++)
481 POS(0, y) = av_clip_pixel(top[0] + ((left[y] - left[-1]) >> 1));
485 if (angle < 0 && last < -1) {
486 for (x = 0; x <= size; x += 4)
487 AV_WN4P(&ref_tmp[x], AV_RN4P(&left[x - 1]));
488 for (x = last; x <= -1; x++)
489 ref_tmp[x] = top[-1 + ((x * inv_angle[mode - 11] + 128) >> 8)];
493 for (x = 0; x < size; x++) {
494 int idx = ((x + 1) * angle) >> 5;
495 int fact = ((x + 1) * angle) & 31;
497 for (y = 0; y < size; y++) {
498 POS(x, y) = ((32 - fact) * ref[y + idx + 1] +
499 fact * ref[y + idx + 2] + 16) >> 5;
502 for (y = 0; y < size; y++)
503 POS(x, y) = ref[y + idx + 1];
506 if (mode == 10 && c_idx == 0 && size < 32) {
507 for (x = 0; x < size; x += 4) {
508 POS(x, 0) = av_clip_pixel(left[0] + ((top[x ] - top[-1]) >> 1));
509 POS(x + 1, 0) = av_clip_pixel(left[0] + ((top[x + 1] - top[-1]) >> 1));
510 POS(x + 2, 0) = av_clip_pixel(left[0] + ((top[x + 2] - top[-1]) >> 1));
511 POS(x + 3, 0) = av_clip_pixel(left[0] + ((top[x + 3] - top[-1]) >> 1));
517 static void FUNC(pred_angular_0)(uint8_t *src, const uint8_t *top,
519 ptrdiff_t stride, int c_idx, int mode)
521 FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 2);
524 static void FUNC(pred_angular_1)(uint8_t *src, const uint8_t *top,
526 ptrdiff_t stride, int c_idx, int mode)
528 FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 3);
531 static void FUNC(pred_angular_2)(uint8_t *src, const uint8_t *top,
533 ptrdiff_t stride, int c_idx, int mode)
535 FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 4);
538 static void FUNC(pred_angular_3)(uint8_t *src, const uint8_t *top,
540 ptrdiff_t stride, int c_idx, int mode)
542 FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 5);
545 #undef EXTEND_LEFT_CIP
546 #undef EXTEND_RIGHT_CIP
548 #undef EXTEND_DOWN_CIP
554 #undef MIN_TB_ADDR_ZS