4 * Copyright (C) 2012 - 2013 Guillaume Martres
6 * This file is part of Libav.
8 * Libav 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 * Libav 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 Libav; 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"
27 #include "bit_depth_template.c"
29 #define POS(x, y) src[(x) + stride * (y)]
31 static av_always_inline void FUNC(intra_pred)(HEVCContext *s, int x0, int y0,
32 int log2_size, int c_idx)
35 ((x) >> s->sps->log2_min_pu_size)
37 (s->ref->tab_mvf[(x) + (y) * min_pu_width])
38 #define MVF_PU(x, y) \
39 MVF(PU(x0 + ((x) << hshift)), PU(y0 + ((y) << vshift)))
40 #define IS_INTRA(x, y) \
42 #define MIN_TB_ADDR_ZS(x, y) \
43 s->pps->min_tb_addr_zs[(y) * s->sps->min_tb_width + (x)]
45 #define EXTEND(ptr, val, len) \
47 pixel4 pix = PIXEL_SPLAT_X4(val); \
48 for (i = 0; i < (len); i += 4) \
49 AV_WN4P(ptr + i, pix); \
52 #define EXTEND_LEFT_CIP(ptr, start, length) \
53 for (i = (start); i > (start) - (length); i--) \
54 if (!IS_INTRA(i - 1, -1)) \
56 #define EXTEND_RIGHT_CIP(ptr, start, length) \
57 for (i = (start); i < (start) + (length); i++) \
58 if (!IS_INTRA(i, -1)) \
60 #define EXTEND_UP_CIP(ptr, start, length) \
61 for (i = (start); i > (start) - (length); i--) \
62 if (!IS_INTRA(-1, i - 1)) \
64 #define EXTEND_UP_CIP_0(ptr, start, length) \
65 for (i = (start); i > (start) - (length); i--) \
67 #define EXTEND_DOWN_CIP(ptr, start, length) \
68 for (i = (start); i < (start) + (length); i++) \
69 if (!IS_INTRA(-1, i)) \
71 HEVCLocalContext *lc = &s->HEVClc;
73 int hshift = s->sps->hshift[c_idx];
74 int vshift = s->sps->vshift[c_idx];
75 int size = (1 << log2_size);
76 int size_in_luma = size << hshift;
77 int size_in_tbs = size_in_luma >> s->sps->log2_min_tb_size;
80 int x_tb = x0 >> s->sps->log2_min_tb_size;
81 int y_tb = y0 >> s->sps->log2_min_tb_size;
82 int cur_tb_addr = MIN_TB_ADDR_ZS(x_tb, y_tb);
84 ptrdiff_t stride = s->frame->linesize[c_idx] / sizeof(pixel);
85 pixel *src = (pixel*)s->frame->data[c_idx] + x + y * stride;
87 int min_pu_width = s->sps->min_pu_width;
89 enum IntraPredMode mode = c_idx ? lc->pu.intra_pred_mode_c :
90 lc->tu.cur_intra_pred_mode;
92 pixel left_array[2 * MAX_TB_SIZE + 1];
93 pixel filtered_left_array[2 * MAX_TB_SIZE + 1];
94 pixel top_array[2 * MAX_TB_SIZE + 1];
95 pixel filtered_top_array[2 * MAX_TB_SIZE + 1];
97 pixel *left = left_array + 1;
98 pixel *top = top_array + 1;
99 pixel *filtered_left = filtered_left_array + 1;
100 pixel *filtered_top = filtered_top_array + 1;
102 int cand_bottom_left = lc->na.cand_bottom_left && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb - 1, y_tb + size_in_tbs);
103 int cand_left = lc->na.cand_left;
104 int cand_up_left = lc->na.cand_up_left;
105 int cand_up = lc->na.cand_up;
106 int cand_up_right = lc->na.cand_up_right && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb + size_in_tbs, y_tb - 1);
108 int bottom_left_size = (FFMIN(y0 + 2 * size_in_luma, s->sps->height) -
109 (y0 + size_in_luma)) >> vshift;
110 int top_right_size = (FFMIN(x0 + 2 * size_in_luma, s->sps->width) -
111 (x0 + size_in_luma)) >> hshift;
113 if (s->pps->constrained_intra_pred_flag == 1) {
114 int size_in_luma_pu = PU(size_in_luma);
115 int on_pu_edge_x = !(x0 & ((1 << s->sps->log2_min_pu_size) - 1));
116 int on_pu_edge_y = !(y0 & ((1 << s->sps->log2_min_pu_size) - 1));
117 if (!size_in_luma_pu)
119 if (cand_bottom_left == 1 && on_pu_edge_x) {
120 int x_left_pu = PU(x0 - 1);
121 int y_bottom_pu = PU(y0 + size_in_luma);
122 int max = FFMIN(size_in_luma_pu, s->sps->min_pu_height - y_bottom_pu);
123 cand_bottom_left = 0;
124 for (i = 0; i < max; i++)
125 cand_bottom_left |= MVF(x_left_pu, y_bottom_pu + i).is_intra;
127 if (cand_left == 1 && on_pu_edge_x) {
128 int x_left_pu = PU(x0 - 1);
129 int y_left_pu = PU(y0);
130 int max = FFMIN(size_in_luma_pu, s->sps->min_pu_height - y_left_pu);
132 for (i = 0; i < max; i++)
133 cand_left |= MVF(x_left_pu, y_left_pu + i).is_intra;
135 if (cand_up_left == 1) {
136 int x_left_pu = PU(x0 - 1);
137 int y_top_pu = PU(y0 - 1);
138 cand_up_left = MVF(x_left_pu, y_top_pu).is_intra;
140 if (cand_up == 1 && on_pu_edge_y) {
141 int x_top_pu = PU(x0);
142 int y_top_pu = PU(y0 - 1);
143 int max = FFMIN(size_in_luma_pu, s->sps->min_pu_width - x_top_pu);
145 for (i = 0; i < max; i++)
146 cand_up |= MVF(x_top_pu + i, y_top_pu).is_intra;
148 if (cand_up_right == 1 && on_pu_edge_y) {
149 int y_top_pu = PU(y0 - 1);
150 int x_right_pu = PU(x0 + size_in_luma);
151 int max = FFMIN(size_in_luma_pu, s->sps->min_pu_width - x_right_pu);
153 for (i = 0; i < max; i++)
154 cand_up_right |= MVF(x_right_pu + i, y_top_pu).is_intra;
156 for (i = 0; i < 2 * MAX_TB_SIZE; i++) {
161 if (cand_bottom_left) {
162 for (i = size; i < size + bottom_left_size; i++)
163 left[i] = POS(-1, i);
164 EXTEND(left + size + bottom_left_size, POS(-1, size + bottom_left_size - 1),
165 size - bottom_left_size);
168 for (i = size - 1; i >= 0; i--)
169 left[i] = POS(-1, i);
171 left[-1] = POS(-1, -1);
175 memcpy(top, src - stride, size * sizeof(pixel));
177 memcpy(top + size, src - stride + size, size * sizeof(pixel));
178 EXTEND(top + size + top_right_size, POS(size + top_right_size - 1, -1),
179 size - top_right_size);
182 if (s->pps->constrained_intra_pred_flag == 1) {
183 if (cand_bottom_left || cand_left || cand_up_left || cand_up || cand_up_right) {
184 int size_max_x = x0 + ((2 * size) << hshift) < s->sps->width ?
185 2 * size : (s->sps->width - x0) >> hshift;
186 int size_max_y = y0 + ((2 * size) << vshift) < s->sps->height ?
187 2 * size : (s->sps->height - y0) >> vshift;
188 int j = size + (cand_bottom_left? bottom_left_size: 0) -1;
189 if (!cand_up_right) {
190 size_max_x = x0 + ((size) << hshift) < s->sps->width ?
191 size : (s->sps->width - x0) >> hshift;
193 if (!cand_bottom_left) {
194 size_max_y = y0 + (( size) << vshift) < s->sps->height ?
195 size : (s->sps->height - y0) >> vshift;
197 if (cand_bottom_left || cand_left || cand_up_left) {
198 while (j > -1 && !IS_INTRA(-1, j))
200 if (!IS_INTRA(-1, j)) {
202 while (j < size_max_x && !IS_INTRA(j, -1))
204 EXTEND_LEFT_CIP(top, j, j + 1);
210 while (j < size_max_x && !IS_INTRA(j, -1))
214 EXTEND_LEFT_CIP(top, j, j + 1);
216 EXTEND_LEFT_CIP(top, j, j);
222 if (cand_bottom_left || cand_left) {
223 EXTEND_DOWN_CIP(left, j, size_max_y - j);
226 EXTEND(left, left[-1], size);
227 if (!cand_bottom_left)
228 EXTEND(left + size, left[size - 1], size);
229 if (x0 != 0 && y0 != 0) {
230 EXTEND_UP_CIP(left, size_max_y - 1, size_max_y);
231 } else if (x0 == 0) {
232 EXTEND_UP_CIP_0(left, size_max_y - 1, size_max_y);
234 EXTEND_UP_CIP(left, size_max_y - 1, size_max_y - 1);
238 EXTEND_RIGHT_CIP(top, 0, size_max_x);
242 // Infer the unavailable samples
243 if (!cand_bottom_left) {
245 EXTEND(left + size, left[size - 1], size);
246 } else if (cand_up_left) {
247 EXTEND(left, left[-1], 2 * size);
249 } else if (cand_up) {
251 EXTEND(left, left[-1], 2 * size);
254 } else if (cand_up_right) {
255 EXTEND(top, top[size], size);
256 left[-1] = top[size];
257 EXTEND(left, left[-1], 2 * size);
261 } else { // No samples available
262 left[-1] = (1 << (BIT_DEPTH - 1));
263 EXTEND(top, left[-1], 2 * size);
264 EXTEND(left, left[-1], 2 * size);
269 EXTEND(left, left[size], size);
274 EXTEND(top, left[-1], size);
276 EXTEND(top + size, top[size - 1], size);
281 if (c_idx == 0 && mode != INTRA_DC && size != 4) {
282 int intra_hor_ver_dist_thresh[] = { 7, 1, 0 };
283 int min_dist_vert_hor = FFMIN(FFABS((int)mode - 26),
284 FFABS((int)mode - 10));
285 if (min_dist_vert_hor > intra_hor_ver_dist_thresh[log2_size - 3]) {
286 int threshold = 1 << (BIT_DEPTH - 5);
287 if (s->sps->sps_strong_intra_smoothing_enable_flag &&
289 FFABS(top[-1] + top[63] - 2 * top[31]) < threshold &&
290 FFABS(left[-1] + left[63] - 2 * left[31]) < threshold) {
291 // We can't just overwrite values in top because it could be
292 // a pointer into src
293 filtered_top[-1] = top[-1];
294 filtered_top[63] = top[63];
295 for (i = 0; i < 63; i++)
296 filtered_top[i] = ((64 - (i + 1)) * top[-1] +
297 (i + 1) * top[63] + 32) >> 6;
298 for (i = 0; i < 63; i++)
299 left[i] = ((64 - (i + 1)) * left[-1] +
300 (i + 1) * left[63] + 32) >> 6;
303 filtered_left[2 * size - 1] = left[2 * size - 1];
304 filtered_top[2 * size - 1] = top[2 * size - 1];
305 for (i = 2 * size - 2; i >= 0; i--)
306 filtered_left[i] = (left[i + 1] + 2 * left[i] +
307 left[i - 1] + 2) >> 2;
309 filtered_left[-1] = (left[0] + 2 * left[-1] + top[0] + 2) >> 2;
310 for (i = 2 * size - 2; i >= 0; i--)
311 filtered_top[i] = (top[i + 1] + 2 * top[i] +
312 top[i - 1] + 2) >> 2;
313 left = filtered_left;
321 s->hpc.pred_planar[log2_size - 2]((uint8_t *)src, (uint8_t *)top,
322 (uint8_t *)left, stride);
325 s->hpc.pred_dc((uint8_t *)src, (uint8_t *)top,
326 (uint8_t *)left, stride, log2_size, c_idx);
329 s->hpc.pred_angular[log2_size - 2]((uint8_t *)src, (uint8_t *)top,
330 (uint8_t *)left, stride, c_idx,
336 #define INTRA_PRED(size) \
337 static void FUNC(intra_pred_ ## size)(HEVCContext *s, int x0, int y0, int c_idx) \
339 FUNC(intra_pred)(s, x0, y0, size, c_idx); \
349 static av_always_inline void FUNC(pred_planar)(uint8_t *_src, const uint8_t *_top,
350 const uint8_t *_left, ptrdiff_t stride,
354 pixel *src = (pixel *)_src;
355 const pixel *top = (const pixel *)_top;
356 const pixel *left = (const pixel *)_left;
357 int size = 1 << trafo_size;
358 for (y = 0; y < size; y++)
359 for (x = 0; x < size; x++)
360 POS(x, y) = ((size - 1 - x) * left[y] + (x + 1) * top[size] +
361 (size - 1 - y) * top[x] + (y + 1) * left[size] + size) >> (trafo_size + 1);
364 #define PRED_PLANAR(size)\
365 static void FUNC(pred_planar_ ## size)(uint8_t *src, const uint8_t *top, \
366 const uint8_t *left, ptrdiff_t stride) \
368 FUNC(pred_planar)(src, top, left, stride, size + 2); \
378 static void FUNC(pred_dc)(uint8_t *_src, const uint8_t *_top,
379 const uint8_t *_left,
380 ptrdiff_t stride, int log2_size, int c_idx)
383 int size = (1 << log2_size);
384 pixel *src = (pixel *)_src;
385 const pixel *top = (const pixel *)_top;
386 const pixel *left = (const pixel *)_left;
389 for (i = 0; i < size; i++)
390 dc += left[i] + top[i];
392 dc >>= log2_size + 1;
394 a = PIXEL_SPLAT_X4(dc);
396 for (i = 0; i < size; i++)
397 for (j = 0; j < size / 4; j++)
398 AV_WN4PA(&POS(j * 4, i), a);
400 if (c_idx == 0 && size < 32) {
401 POS(0, 0) = (left[0] + 2 * dc + top[0] + 2) >> 2;
402 for (x = 1; x < size; x++)
403 POS(x, 0) = (top[x] + 3 * dc + 2) >> 2;
404 for (y = 1; y < size; y++)
405 POS(0, y) = (left[y] + 3 * dc + 2) >> 2;
409 static av_always_inline void FUNC(pred_angular)(uint8_t *_src,
411 const uint8_t *_left,
412 ptrdiff_t stride, int c_idx,
416 pixel *src = (pixel *)_src;
417 const pixel *top = (const pixel *)_top;
418 const pixel *left = (const pixel *)_left;
420 static const int intra_pred_angle[] = {
421 32, 26, 21, 17, 13, 9, 5, 2, 0, -2, -5, -9, -13, -17, -21, -26, -32,
422 -26, -21, -17, -13, -9, -5, -2, 0, 2, 5, 9, 13, 17, 21, 26, 32
424 static const int inv_angle[] = {
425 -4096, -1638, -910, -630, -482, -390, -315, -256, -315, -390, -482,
426 -630, -910, -1638, -4096
429 int angle = intra_pred_angle[mode - 2];
430 pixel ref_array[3 * MAX_TB_SIZE + 1];
431 pixel *ref_tmp = ref_array + size;
433 int last = (size * angle) >> 5;
437 if (angle < 0 && last < -1) {
438 for (x = 0; x <= size; x++)
439 ref_tmp[x] = top[x - 1];
440 for (x = last; x <= -1; x++)
441 ref_tmp[x] = left[-1 + ((x * inv_angle[mode - 11] + 128) >> 8)];
445 for (y = 0; y < size; y++) {
446 int idx = ((y + 1) * angle) >> 5;
447 int fact = ((y + 1) * angle) & 31;
449 for (x = 0; x < size; x++) {
450 POS(x, y) = ((32 - fact) * ref[x + idx + 1] +
451 fact * ref[x + idx + 2] + 16) >> 5;
454 for (x = 0; x < size; x++)
455 POS(x, y) = ref[x + idx + 1];
458 if (mode == 26 && c_idx == 0 && size < 32) {
459 for (y = 0; y < size; y++)
460 POS(0, y) = av_clip_pixel(top[0] + ((left[y] - left[-1]) >> 1));
464 if (angle < 0 && last < -1) {
465 for (x = 0; x <= size; x++)
466 ref_tmp[x] = left[x - 1];
467 for (x = last; x <= -1; x++)
468 ref_tmp[x] = top[-1 + ((x * inv_angle[mode - 11] + 128) >> 8)];
472 for (x = 0; x < size; x++) {
473 int idx = ((x + 1) * angle) >> 5;
474 int fact = ((x + 1) * angle) & 31;
476 for (y = 0; y < size; y++) {
477 POS(x, y) = ((32 - fact) * ref[y + idx + 1] +
478 fact * ref[y + idx + 2] + 16) >> 5;
481 for (y = 0; y < size; y++)
482 POS(x, y) = ref[y + idx + 1];
485 if (mode == 10 && c_idx == 0 && size < 32) {
486 for (x = 0; x < size; x++)
487 POS(x, 0) = av_clip_pixel(left[0] + ((top[x] - top[-1]) >> 1));
492 static void FUNC(pred_angular_0)(uint8_t *src, const uint8_t *top,
494 ptrdiff_t stride, int c_idx, int mode)
496 FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 2);
499 static void FUNC(pred_angular_1)(uint8_t *src, const uint8_t *top,
501 ptrdiff_t stride, int c_idx, int mode)
503 FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 3);
506 static void FUNC(pred_angular_2)(uint8_t *src, const uint8_t *top,
508 ptrdiff_t stride, int c_idx, int mode)
510 FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 4);
513 static void FUNC(pred_angular_3)(uint8_t *src, const uint8_t *top,
515 ptrdiff_t stride, int c_idx, int mode)
517 FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 5);
520 #undef EXTEND_LEFT_CIP
521 #undef EXTEND_RIGHT_CIP
523 #undef EXTEND_DOWN_CIP
529 #undef MIN_TB_ADDR_ZS