Go to most recent revision | Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 4349 | Serge | 1 | /* |
| 2 | * HEVC video Decoder |
||
| 3 | * |
||
| 4 | * Copyright (C) 2012 - 2013 Guillaume Martres |
||
| 5 | * |
||
| 6 | * This file is part of FFmpeg. |
||
| 7 | * |
||
| 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. |
||
| 12 | * |
||
| 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. |
||
| 17 | * |
||
| 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 |
||
| 21 | */ |
||
| 22 | |||
| 23 | #include "libavutil/pixdesc.h" |
||
| 24 | #include "bit_depth_template.c" |
||
| 25 | #include "hevcpred.h" |
||
| 26 | |||
| 27 | #define POS(x, y) src[(x) + stride * (y)] |
||
| 28 | |||
| 29 | static void FUNC(intra_pred)(HEVCContext *s, int x0, int y0, int log2_size, int c_idx) |
||
| 30 | { |
||
| 31 | #define PU(x) \ |
||
| 32 | ((x) >> s->sps->log2_min_pu_size) |
||
| 33 | #define MVF(x, y) \ |
||
| 34 | (s->ref->tab_mvf[(x) + (y) * min_pu_width]) |
||
| 35 | #define MVF_PU(x, y) \ |
||
| 36 | MVF(PU(x0 + ((x) << hshift)), PU(y0 + ((y) << vshift))) |
||
| 37 | #define IS_INTRA(x, y) \ |
||
| 38 | MVF_PU(x, y).is_intra |
||
| 39 | #define MIN_TB_ADDR_ZS(x, y) \ |
||
| 40 | s->pps->min_tb_addr_zs[(y) * s->sps->min_tb_width + (x)] |
||
| 41 | #define EXTEND_LEFT(ptr, start, length) \ |
||
| 42 | for (i = (start); i > (start) - (length); i--) \ |
||
| 43 | ptr[i - 1] = ptr[i] |
||
| 44 | #define EXTEND_RIGHT(ptr, start, length) \ |
||
| 45 | for (i = (start); i < (start) + (length); i++) \ |
||
| 46 | ptr[i] = ptr[i - 1] |
||
| 47 | #define EXTEND_UP(ptr, start, length) EXTEND_LEFT(ptr, start, length) |
||
| 48 | #define EXTEND_DOWN(ptr, start, length) EXTEND_RIGHT(ptr, start, length) |
||
| 49 | #define EXTEND_LEFT_CIP(ptr, start, length) \ |
||
| 50 | for (i = (start); i > (start) - (length); i--) \ |
||
| 51 | if (!IS_INTRA(i - 1, -1)) \ |
||
| 52 | ptr[i - 1] = ptr[i] |
||
| 53 | #define EXTEND_RIGHT_CIP(ptr, start, length) \ |
||
| 54 | for (i = (start); i < (start) + (length); i++) \ |
||
| 55 | if (!IS_INTRA(i, -1)) \ |
||
| 56 | ptr[i] = ptr[i - 1] |
||
| 57 | #define EXTEND_UP_CIP(ptr, start, length) \ |
||
| 58 | for (i = (start); i > (start) - (length); i--) \ |
||
| 59 | if (!IS_INTRA(-1, i - 1)) \ |
||
| 60 | ptr[i - 1] = ptr[i] |
||
| 61 | #define EXTEND_UP_CIP_0(ptr, start, length) \ |
||
| 62 | for (i = (start); i > (start) - (length); i--) \ |
||
| 63 | ptr[i - 1] = ptr[i] |
||
| 64 | #define EXTEND_DOWN_CIP(ptr, start, length) \ |
||
| 65 | for (i = (start); i < (start) + (length); i++) \ |
||
| 66 | if (!IS_INTRA(-1, i)) \ |
||
| 67 | ptr[i] = ptr[i - 1] |
||
| 68 | HEVCLocalContext *lc = s->HEVClc; |
||
| 69 | int i; |
||
| 70 | int hshift = s->sps->hshift[c_idx]; |
||
| 71 | int vshift = s->sps->vshift[c_idx]; |
||
| 72 | int size = (1 << log2_size); |
||
| 73 | int size_in_luma = size << hshift; |
||
| 74 | int size_in_tbs = size_in_luma >> s->sps->log2_min_tb_size; |
||
| 75 | int x = x0 >> hshift; |
||
| 76 | int y = y0 >> vshift; |
||
| 77 | int x_tb = x0 >> s->sps->log2_min_tb_size; |
||
| 78 | int y_tb = y0 >> s->sps->log2_min_tb_size; |
||
| 79 | int cur_tb_addr = MIN_TB_ADDR_ZS(x_tb, y_tb); |
||
| 80 | |||
| 81 | ptrdiff_t stride = s->frame->linesize[c_idx] / sizeof(pixel); |
||
| 82 | pixel *src = (pixel*)s->frame->data[c_idx] + x + y * stride; |
||
| 83 | |||
| 84 | int min_pu_width = s->sps->min_pu_width; |
||
| 85 | |||
| 86 | enum IntraPredMode mode = c_idx ? lc->pu.intra_pred_mode_c : |
||
| 87 | lc->tu.cur_intra_pred_mode; |
||
| 88 | |||
| 89 | pixel left_array[2 * MAX_TB_SIZE + 1]; |
||
| 90 | pixel filtered_left_array[2 * MAX_TB_SIZE + 1]; |
||
| 91 | pixel top_array[2 * MAX_TB_SIZE + 1]; |
||
| 92 | pixel filtered_top_array[2 * MAX_TB_SIZE + 1]; |
||
| 93 | |||
| 94 | pixel *left = left_array + 1; |
||
| 95 | pixel *top = top_array + 1; |
||
| 96 | pixel *filtered_left = filtered_left_array + 1; |
||
| 97 | pixel *filtered_top = filtered_top_array + 1; |
||
| 98 | |||
| 99 | int cand_bottom_left = lc->na.cand_bottom_left && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb - 1, y_tb + size_in_tbs); |
||
| 100 | int cand_left = lc->na.cand_left; |
||
| 101 | int cand_up_left = lc->na.cand_up_left; |
||
| 102 | int cand_up = lc->na.cand_up; |
||
| 103 | int cand_up_right = lc->na.cand_up_right && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb + size_in_tbs, y_tb - 1); |
||
| 104 | |||
| 105 | int bottom_left_size = (FFMIN(y0 + 2 * size_in_luma, s->sps->height) - |
||
| 106 | (y0 + size_in_luma)) >> vshift; |
||
| 107 | int top_right_size = (FFMIN(x0 + 2 * size_in_luma, s->sps->width) - |
||
| 108 | (x0 + size_in_luma)) >> hshift; |
||
| 109 | |||
| 110 | if (s->pps->constrained_intra_pred_flag == 1) { |
||
| 111 | int size_in_luma_pu = PU(size_in_luma); |
||
| 112 | int on_pu_edge_x = !(x0 & ((1 << s->sps->log2_min_pu_size) - 1)); |
||
| 113 | int on_pu_edge_y = !(y0 & ((1 << s->sps->log2_min_pu_size) - 1)); |
||
| 114 | if(!size_in_luma_pu) |
||
| 115 | size_in_luma_pu++; |
||
| 116 | if (cand_bottom_left == 1 && on_pu_edge_x) { |
||
| 117 | int x_left_pu = PU(x0 - 1); |
||
| 118 | int y_bottom_pu = PU(y0 + size_in_luma); |
||
| 119 | int max = FFMIN(size_in_luma_pu, s->sps->min_pu_height - y_bottom_pu); |
||
| 120 | cand_bottom_left = 0; |
||
| 121 | for (i = 0; i < max; i++) |
||
| 122 | cand_bottom_left |= MVF(x_left_pu, y_bottom_pu + i).is_intra; |
||
| 123 | } |
||
| 124 | if (cand_left == 1 && on_pu_edge_x) { |
||
| 125 | int x_left_pu = PU(x0 - 1); |
||
| 126 | int y_left_pu = PU(y0); |
||
| 127 | int max = FFMIN(size_in_luma_pu, s->sps->min_pu_height - y_left_pu); |
||
| 128 | cand_left = 0; |
||
| 129 | for (i = 0; i < max; i++) |
||
| 130 | cand_left |= MVF(x_left_pu, y_left_pu + i).is_intra; |
||
| 131 | } |
||
| 132 | if (cand_up_left == 1) { |
||
| 133 | int x_left_pu = PU(x0 - 1); |
||
| 134 | int y_top_pu = PU(y0 - 1); |
||
| 135 | cand_up_left = MVF(x_left_pu, y_top_pu).is_intra; |
||
| 136 | } |
||
| 137 | if (cand_up == 1 && on_pu_edge_y) { |
||
| 138 | int x_top_pu = PU(x0); |
||
| 139 | int y_top_pu = PU(y0 - 1); |
||
| 140 | int max = FFMIN(size_in_luma_pu, s->sps->min_pu_width - x_top_pu); |
||
| 141 | cand_up = 0; |
||
| 142 | for (i = 0; i < max; i++) |
||
| 143 | cand_up |= MVF(x_top_pu + i, y_top_pu).is_intra; |
||
| 144 | } |
||
| 145 | if (cand_up_right == 1 && on_pu_edge_y) { |
||
| 146 | int y_top_pu = PU(y0 - 1); |
||
| 147 | int x_right_pu = PU(x0 + size_in_luma); |
||
| 148 | int max = FFMIN(size_in_luma_pu, s->sps->min_pu_width - x_right_pu); |
||
| 149 | cand_up_right = 0; |
||
| 150 | for (i = 0; i < max; i++) |
||
| 151 | cand_up_right |= MVF(x_right_pu + i, y_top_pu).is_intra; |
||
| 152 | } |
||
| 153 | for (i = 0; i < 2 * MAX_TB_SIZE; i++) { |
||
| 154 | left[i] = 128; |
||
| 155 | top[i] = 128; |
||
| 156 | } |
||
| 157 | } |
||
| 158 | if (cand_bottom_left) { |
||
| 159 | for (i = size + bottom_left_size; i < (size << 1); i++) |
||
| 160 | if (IS_INTRA(-1, size + bottom_left_size - 1) || !s->pps->constrained_intra_pred_flag) |
||
| 161 | left[i] = POS(-1, size + bottom_left_size - 1); |
||
| 162 | for (i = size + bottom_left_size - 1; i >= size; i--) |
||
| 163 | if (IS_INTRA(-1, i) || !s->pps->constrained_intra_pred_flag) |
||
| 164 | left[i] = POS(-1, i); |
||
| 165 | } |
||
| 166 | if (cand_left) |
||
| 167 | for (i = size - 1; i >= 0; i--) |
||
| 168 | if (IS_INTRA(-1, i) || !s->pps->constrained_intra_pred_flag) |
||
| 169 | left[i] = POS(-1, i); |
||
| 170 | if (cand_up_left) |
||
| 171 | if (IS_INTRA(-1, -1) || !s->pps->constrained_intra_pred_flag) { |
||
| 172 | left[-1] = POS(-1, -1); |
||
| 173 | top[-1] = left[-1]; |
||
| 174 | } |
||
| 175 | if (cand_up) |
||
| 176 | for (i = size - 1; i >= 0; i--) |
||
| 177 | if (IS_INTRA(i, -1) || !s->pps->constrained_intra_pred_flag) |
||
| 178 | top[i] = POS(i, -1); |
||
| 179 | if (cand_up_right) { |
||
| 180 | for (i = size + top_right_size; i < (size << 1); i++) |
||
| 181 | if (IS_INTRA(size + top_right_size - 1, -1) || !s->pps->constrained_intra_pred_flag) |
||
| 182 | top[i] = POS(size + top_right_size - 1, -1); |
||
| 183 | for (i = size + top_right_size - 1; i >= size; i--) |
||
| 184 | if (IS_INTRA(i, -1) || !s->pps->constrained_intra_pred_flag) |
||
| 185 | top[i] = POS(i, -1); |
||
| 186 | } |
||
| 187 | |||
| 188 | if (s->pps->constrained_intra_pred_flag == 1) { |
||
| 189 | if (cand_bottom_left || cand_left || cand_up_left || cand_up || cand_up_right) { |
||
| 190 | int size_max_x = x0 + ((2 * size) << hshift) < s->sps->width ? |
||
| 191 | 2 * size : (s->sps->width - x0) >> hshift; |
||
| 192 | int size_max_y = y0 + ((2 * size) << vshift) < s->sps->height ? |
||
| 193 | 2 * size : (s->sps->height - y0) >> vshift; |
||
| 194 | int j = size + (cand_bottom_left? bottom_left_size: 0) -1; |
||
| 195 | if (!cand_up_right) { |
||
| 196 | size_max_x = x0 + ((size) << hshift) < s->sps->width ? |
||
| 197 | size : (s->sps->width - x0) >> hshift; |
||
| 198 | } |
||
| 199 | if (!cand_bottom_left) { |
||
| 200 | size_max_y = y0 + (( size) << vshift) < s->sps->height ? |
||
| 201 | size : (s->sps->height - y0) >> vshift; |
||
| 202 | } |
||
| 203 | if (cand_bottom_left || cand_left || cand_up_left) { |
||
| 204 | while (j>-1 && !IS_INTRA(-1, j)) j--; |
||
| 205 | if (!IS_INTRA(-1, j)) { |
||
| 206 | j = 0; |
||
| 207 | while(j < size_max_x && !IS_INTRA(j, -1)) j++; |
||
| 208 | EXTEND_LEFT_CIP(top, j, j+1); |
||
| 209 | left[-1] = top[-1]; |
||
| 210 | j = 0; |
||
| 211 | } |
||
| 212 | } else { |
||
| 213 | j = 0; |
||
| 214 | while (j < size_max_x && !IS_INTRA(j, -1)) j++; |
||
| 215 | if (j > 0) |
||
| 216 | if (x0 > 0) { |
||
| 217 | EXTEND_LEFT_CIP(top, j, j+1); |
||
| 218 | } else { |
||
| 219 | EXTEND_LEFT_CIP(top, j, j); |
||
| 220 | top[-1] = top[0]; |
||
| 221 | } |
||
| 222 | left[-1] = top[-1]; |
||
| 223 | j = 0; |
||
| 224 | } |
||
| 225 | if (cand_bottom_left || cand_left) { |
||
| 226 | EXTEND_DOWN_CIP(left, j, size_max_y-j); |
||
| 227 | } |
||
| 228 | if (!cand_left) { |
||
| 229 | EXTEND_DOWN(left, 0, size); |
||
| 230 | } |
||
| 231 | if (!cand_bottom_left) { |
||
| 232 | EXTEND_DOWN(left, size, size); |
||
| 233 | } |
||
| 234 | if (x0 != 0 && y0 != 0) { |
||
| 235 | EXTEND_UP_CIP(left, size_max_y - 1, size_max_y); |
||
| 236 | } else if( x0 == 0) { |
||
| 237 | EXTEND_UP_CIP_0(left, size_max_y - 1, size_max_y); |
||
| 238 | } else{ |
||
| 239 | EXTEND_UP_CIP(left, size_max_y - 1, size_max_y-1); |
||
| 240 | } |
||
| 241 | top[-1] = left[-1]; |
||
| 242 | if (y0 != 0) { |
||
| 243 | EXTEND_RIGHT_CIP(top, 0, size_max_x); |
||
| 244 | } |
||
| 245 | } |
||
| 246 | } |
||
| 247 | // Infer the unavailable samples |
||
| 248 | if (!cand_bottom_left) { |
||
| 249 | if (cand_left) { |
||
| 250 | EXTEND_DOWN(left, size, size); |
||
| 251 | } else if (cand_up_left) { |
||
| 252 | EXTEND_DOWN(left, 0, 2 * size); |
||
| 253 | cand_left = 1; |
||
| 254 | } else if (cand_up) { |
||
| 255 | left[-1] = top[0]; |
||
| 256 | EXTEND_DOWN(left, 0, 2 * size); |
||
| 257 | cand_up_left = 1; |
||
| 258 | cand_left = 1; |
||
| 259 | } else if (cand_up_right) { |
||
| 260 | EXTEND_LEFT(top, size, size); |
||
| 261 | left[-1] = top[0]; |
||
| 262 | EXTEND_DOWN(left ,0 , 2 * size); |
||
| 263 | cand_up = 1; |
||
| 264 | cand_up_left = 1; |
||
| 265 | cand_left = 1; |
||
| 266 | } else { // No samples available |
||
| 267 | top[0] = left[-1] = (1 << (BIT_DEPTH - 1)); |
||
| 268 | EXTEND_RIGHT(top, 1, 2 * size - 1); |
||
| 269 | EXTEND_DOWN(left, 0, 2 * size); |
||
| 270 | } |
||
| 271 | } |
||
| 272 | |||
| 273 | if (!cand_left) { |
||
| 274 | EXTEND_UP(left, size, size); |
||
| 275 | } |
||
| 276 | if (!cand_up_left) { |
||
| 277 | left[-1] = left[0]; |
||
| 278 | } |
||
| 279 | if (!cand_up) { |
||
| 280 | top[0] = left[-1]; |
||
| 281 | EXTEND_RIGHT(top, 1, size-1); |
||
| 282 | } |
||
| 283 | if (!cand_up_right) { |
||
| 284 | EXTEND_RIGHT(top, size, size); |
||
| 285 | } |
||
| 286 | |||
| 287 | top[-1] = left[-1]; |
||
| 288 | |||
| 289 | #undef EXTEND_LEFT_CIP |
||
| 290 | #undef EXTEND_RIGHT_CIP |
||
| 291 | #undef EXTEND_UP_CIP |
||
| 292 | #undef EXTEND_DOWN_CIP |
||
| 293 | #undef IS_INTRA |
||
| 294 | #undef MVF_PU |
||
| 295 | #undef MVF |
||
| 296 | #undef PU |
||
| 297 | #undef EXTEND_LEFT |
||
| 298 | #undef EXTEND_RIGHT |
||
| 299 | #undef EXTEND_UP |
||
| 300 | #undef EXTEND_DOWN |
||
| 301 | #undef MIN_TB_ADDR_ZS |
||
| 302 | |||
| 303 | // Filtering process |
||
| 304 | if (c_idx == 0 && mode != INTRA_DC && size != 4) { |
||
| 305 | int intra_hor_ver_dist_thresh[] = { 7, 1, 0 }; |
||
| 306 | int min_dist_vert_hor = FFMIN(FFABS((int)(mode - 26U)), |
||
| 307 | FFABS((int)(mode - 10U))); |
||
| 308 | if (min_dist_vert_hor > intra_hor_ver_dist_thresh[log2_size - 3]) { |
||
| 309 | int threshold = 1 << (BIT_DEPTH - 5); |
||
| 310 | if (s->sps->sps_strong_intra_smoothing_enable_flag && |
||
| 311 | log2_size == 5 && |
||
| 312 | FFABS(top[-1] + top[63] - 2 * top[31]) < threshold && |
||
| 313 | FFABS(left[-1] + left[63] - 2 * left[31]) < threshold) { |
||
| 314 | // We can't just overwrite values in top because it could be |
||
| 315 | // a pointer into src |
||
| 316 | filtered_top[-1] = top[-1]; |
||
| 317 | filtered_top[63] = top[63]; |
||
| 318 | for (i = 0; i < 63; i++) |
||
| 319 | filtered_top[i] = ((64 - (i + 1)) * top[-1] + |
||
| 320 | (i + 1) * top[63] + 32) >> 6; |
||
| 321 | for (i = 0; i < 63; i++) |
||
| 322 | left[i] = ((64 - (i + 1)) * left[-1] + |
||
| 323 | (i + 1) * left[63] + 32) >> 6; |
||
| 324 | top = filtered_top; |
||
| 325 | } else { |
||
| 326 | filtered_left[2 * size - 1] = left[2 * size - 1]; |
||
| 327 | filtered_top[2 * size - 1] = top[2 * size - 1]; |
||
| 328 | for (i = 2 * size - 2; i >= 0; i--) |
||
| 329 | filtered_left[i] = (left[i + 1] + 2 * left[i] + |
||
| 330 | left[i - 1] + 2) >> 2; |
||
| 331 | filtered_top[-1] = |
||
| 332 | filtered_left[-1] = (left[0] + 2 * left[-1] + |
||
| 333 | top[0] + 2) >> 2; |
||
| 334 | for (i = 2 * size - 2; i >= 0; i--) |
||
| 335 | filtered_top[i] = (top[i + 1] + 2 * top[i] + |
||
| 336 | top[i - 1] + 2) >> 2; |
||
| 337 | left = filtered_left; |
||
| 338 | top = filtered_top; |
||
| 339 | } |
||
| 340 | } |
||
| 341 | } |
||
| 342 | |||
| 343 | switch (mode) { |
||
| 344 | case INTRA_PLANAR: |
||
| 345 | s->hpc.pred_planar[log2_size - 2]((uint8_t*)src, (uint8_t*)top, |
||
| 346 | (uint8_t*)left, stride); |
||
| 347 | break; |
||
| 348 | case INTRA_DC: |
||
| 349 | s->hpc.pred_dc((uint8_t*)src, (uint8_t*)top, |
||
| 350 | (uint8_t*)left, stride, log2_size, c_idx); |
||
| 351 | break; |
||
| 352 | default: |
||
| 353 | s->hpc.pred_angular[log2_size - 2]((uint8_t*)src, (uint8_t*)top, |
||
| 354 | (uint8_t*)left, stride, c_idx, mode); |
||
| 355 | break; |
||
| 356 | } |
||
| 357 | } |
||
| 358 | |||
| 359 | static void FUNC(pred_planar_0)(uint8_t *_src, const uint8_t *_top, |
||
| 360 | const uint8_t *_left, |
||
| 361 | ptrdiff_t stride) |
||
| 362 | { |
||
| 363 | int x, y; |
||
| 364 | pixel *src = (pixel*)_src; |
||
| 365 | const pixel *top = (const pixel*)_top; |
||
| 366 | const pixel *left = (const pixel*)_left; |
||
| 367 | for (y = 0; y < 4; y++) |
||
| 368 | for (x = 0; x < 4; x++) |
||
| 369 | POS(x, y) = ((3 - x) * left[y] + (x + 1) * top[4] + |
||
| 370 | (3 - y) * top[x] + (y + 1) * left[4] + 4) >> 3; |
||
| 371 | } |
||
| 372 | |||
| 373 | static void FUNC(pred_planar_1)(uint8_t *_src, const uint8_t *_top, |
||
| 374 | const uint8_t *_left, ptrdiff_t stride) |
||
| 375 | { |
||
| 376 | int x, y; |
||
| 377 | pixel *src = (pixel*)_src; |
||
| 378 | const pixel *top = (const pixel*)_top; |
||
| 379 | const pixel *left = (const pixel*)_left; |
||
| 380 | for (y = 0; y < 8; y++) |
||
| 381 | for (x = 0; x < 8; x++) |
||
| 382 | POS(x, y) = ((7 - x) * left[y] + (x + 1) * top[8] + |
||
| 383 | (7 - y) * top[x] + (y + 1) * left[8] + 8) >> 4; |
||
| 384 | } |
||
| 385 | |||
| 386 | static void FUNC(pred_planar_2)(uint8_t *_src, const uint8_t *_top, |
||
| 387 | const uint8_t *_left, ptrdiff_t stride) |
||
| 388 | { |
||
| 389 | int x, y; |
||
| 390 | pixel *src = (pixel*)_src; |
||
| 391 | const pixel *top = (const pixel*)_top; |
||
| 392 | const pixel *left = (const pixel*)_left; |
||
| 393 | for (y = 0; y < 16; y++) |
||
| 394 | for (x = 0; x < 16; x++) |
||
| 395 | POS(x, y) = ((15 - x) * left[y] + (x + 1) * top[16] + |
||
| 396 | (15 - y) * top[x] + (y + 1) * left[16] + 16) >> 5; |
||
| 397 | } |
||
| 398 | |||
| 399 | static void FUNC(pred_planar_3)(uint8_t *_src, const uint8_t *_top, |
||
| 400 | const uint8_t *_left, ptrdiff_t stride) |
||
| 401 | { |
||
| 402 | int x, y; |
||
| 403 | pixel *src = (pixel*)_src; |
||
| 404 | const pixel *top = (const pixel*)_top; |
||
| 405 | const pixel *left = (const pixel*)_left; |
||
| 406 | for (y = 0; y < 32; y++) |
||
| 407 | for (x = 0; x < 32; x++) |
||
| 408 | POS(x, y) = ((31 - x) * left[y] + (x + 1) * top[32] + |
||
| 409 | (31 - y) * top[x] + (y + 1) * left[32] + 32) >> 6; |
||
| 410 | } |
||
| 411 | |||
| 412 | static void FUNC(pred_dc)(uint8_t *_src, const uint8_t *_top, |
||
| 413 | const uint8_t *_left, |
||
| 414 | ptrdiff_t stride, int log2_size, int c_idx) |
||
| 415 | { |
||
| 416 | int i, j, x, y; |
||
| 417 | int size = (1 << log2_size); |
||
| 418 | pixel *src = (pixel*)_src; |
||
| 419 | const pixel *top = (const pixel*)_top; |
||
| 420 | const pixel *left = (const pixel*)_left; |
||
| 421 | int dc = size; |
||
| 422 | pixel4 a; |
||
| 423 | for (i = 0; i < size; i++) |
||
| 424 | dc += left[i] + top[i]; |
||
| 425 | |||
| 426 | dc >>= log2_size + 1; |
||
| 427 | |||
| 428 | a = PIXEL_SPLAT_X4(dc); |
||
| 429 | |||
| 430 | for (i = 0; i < size; i++) |
||
| 431 | for (j = 0; j < size / 4; j++) |
||
| 432 | AV_WN4PA(&POS(j * 4, i), a); |
||
| 433 | |||
| 434 | if (c_idx == 0 && size < 32) { |
||
| 435 | POS(0, 0) = (left[0] + 2 * dc + top[0] + 2) >> 2; |
||
| 436 | for (x = 1; x < size; x++) |
||
| 437 | POS(x, 0) = (top[x] + 3 * dc + 2) >> 2; |
||
| 438 | for (y = 1; y < size; y++) |
||
| 439 | POS(0, y) = (left[y] + 3 * dc + 2) >> 2; |
||
| 440 | } |
||
| 441 | } |
||
| 442 | |||
| 443 | static av_always_inline void FUNC(pred_angular)(uint8_t *_src, |
||
| 444 | const uint8_t *_top, |
||
| 445 | const uint8_t *_left, |
||
| 446 | ptrdiff_t stride, int c_idx, |
||
| 447 | int mode, int size) |
||
| 448 | { |
||
| 449 | int x, y; |
||
| 450 | pixel *src = (pixel*)_src; |
||
| 451 | const pixel *top = (const pixel*)_top; |
||
| 452 | const pixel *left = (const pixel*)_left; |
||
| 453 | |||
| 454 | static const int intra_pred_angle[] = { |
||
| 455 | 32, 26, 21, 17, 13, 9, 5, 2, 0, -2, -5, -9, -13, -17, -21, -26, -32, |
||
| 456 | -26, -21, -17, -13, -9, -5, -2, 0, 2, 5, 9, 13, 17, 21, 26, 32 |
||
| 457 | }; |
||
| 458 | static const int inv_angle[] = { |
||
| 459 | -4096, -1638, -910, -630, -482, -390, -315, -256, -315, -390, -482, |
||
| 460 | -630, -910, -1638, -4096 |
||
| 461 | }; |
||
| 462 | |||
| 463 | int angle = intra_pred_angle[mode - 2]; |
||
| 464 | pixel ref_array[3 * MAX_TB_SIZE + 1]; |
||
| 465 | pixel *ref_tmp = ref_array + size; |
||
| 466 | const pixel *ref; |
||
| 467 | int last = (size * angle) >> 5; |
||
| 468 | |||
| 469 | if (mode >= 18) { |
||
| 470 | ref = top - 1; |
||
| 471 | if (angle < 0 && last < -1) { |
||
| 472 | for (x = 0; x <= size; x++) |
||
| 473 | ref_tmp[x] = top[x - 1]; |
||
| 474 | for (x = last; x <= -1; x++) |
||
| 475 | ref_tmp[x] = left[-1 + ((x * inv_angle[mode-11] + 128) >> 8)]; |
||
| 476 | ref = ref_tmp; |
||
| 477 | } |
||
| 478 | |||
| 479 | for (y = 0; y < size; y++) { |
||
| 480 | int idx = ((y + 1) * angle) >> 5; |
||
| 481 | int fact = ((y + 1) * angle) & 31; |
||
| 482 | if (fact) { |
||
| 483 | for (x = 0; x < size; x++) { |
||
| 484 | POS(x, y) = ((32 - fact) * ref[x + idx + 1] + |
||
| 485 | fact * ref[x + idx + 2] + 16) >> 5; |
||
| 486 | } |
||
| 487 | } else { |
||
| 488 | for (x = 0; x < size; x++) { |
||
| 489 | POS(x, y) = ref[x + idx + 1]; |
||
| 490 | } |
||
| 491 | } |
||
| 492 | } |
||
| 493 | if (mode == 26 && c_idx == 0 && size < 32) { |
||
| 494 | for (y = 0; y < size; y++) |
||
| 495 | POS(0, y) = av_clip_pixel(top[0] + ((left[y] - left[-1]) >> 1)); |
||
| 496 | } |
||
| 497 | } else { |
||
| 498 | ref = left - 1; |
||
| 499 | if (angle < 0 && last < -1) { |
||
| 500 | for (x = 0; x <= size; x++) |
||
| 501 | ref_tmp[x] = left[x - 1]; |
||
| 502 | for (x = last; x <= -1; x++) |
||
| 503 | ref_tmp[x] = top[-1 + ((x * inv_angle[mode-11] + 128) >> 8)]; |
||
| 504 | ref = ref_tmp; |
||
| 505 | } |
||
| 506 | |||
| 507 | for (x = 0; x < size; x++) { |
||
| 508 | int idx = ((x + 1) * angle) >> 5; |
||
| 509 | int fact = ((x + 1) * angle) & 31; |
||
| 510 | if (fact) { |
||
| 511 | for (y = 0; y < size; y++) { |
||
| 512 | POS(x, y) = ((32 - fact) * ref[y + idx + 1] + |
||
| 513 | fact * ref[y + idx + 2] + 16) >> 5; |
||
| 514 | } |
||
| 515 | } else { |
||
| 516 | for (y = 0; y < size; y++) { |
||
| 517 | POS(x, y) = ref[y + idx + 1]; |
||
| 518 | } |
||
| 519 | } |
||
| 520 | } |
||
| 521 | if (mode == 10 && c_idx == 0 && size < 32) { |
||
| 522 | for (x = 0; x < size; x++) |
||
| 523 | POS(x, 0) = av_clip_pixel(left[0] + ((top[x] - top[-1]) >> 1)); |
||
| 524 | } |
||
| 525 | } |
||
| 526 | } |
||
| 527 | |||
| 528 | static void FUNC(pred_angular_0)(uint8_t *src, const uint8_t *top, |
||
| 529 | const uint8_t *left, |
||
| 530 | ptrdiff_t stride, int c_idx, int mode) |
||
| 531 | { |
||
| 532 | FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 2); |
||
| 533 | } |
||
| 534 | |||
| 535 | static void FUNC(pred_angular_1)(uint8_t *src, const uint8_t *top, |
||
| 536 | const uint8_t *left, |
||
| 537 | ptrdiff_t stride, int c_idx, int mode) |
||
| 538 | { |
||
| 539 | FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 3); |
||
| 540 | } |
||
| 541 | |||
| 542 | static void FUNC(pred_angular_2)(uint8_t *src, const uint8_t *top, |
||
| 543 | const uint8_t *left, |
||
| 544 | ptrdiff_t stride, int c_idx, int mode) |
||
| 545 | { |
||
| 546 | FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 4); |
||
| 547 | } |
||
| 548 | |||
| 549 | static void FUNC(pred_angular_3)(uint8_t *src, const uint8_t *top, |
||
| 550 | const uint8_t *left, |
||
| 551 | ptrdiff_t stride, int c_idx, int mode) |
||
| 552 | { |
||
| 553 | FUNC(pred_angular)(src, top, left, stride, c_idx, mode, 1 << 5); |
||
| 554 | } |
||
| 555 | #undef POS |