| 1 | /* |
| 2 | * The simplest mpeg encoder (well, it was the simplest!) |
| 3 | * Copyright (c) 2000,2001 Fabrice Bellard |
| 4 | * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> |
| 5 | * |
| 6 | * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at> |
| 7 | * |
| 8 | * This file is part of FFmpeg. |
| 9 | * |
| 10 | * FFmpeg is free software; you can redistribute it and/or |
| 11 | * modify it under the terms of the GNU Lesser General Public |
| 12 | * License as published by the Free Software Foundation; either |
| 13 | * version 2.1 of the License, or (at your option) any later version. |
| 14 | * |
| 15 | * FFmpeg is distributed in the hope that it will be useful, |
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 18 | * Lesser General Public License for more details. |
| 19 | * |
| 20 | * You should have received a copy of the GNU Lesser General Public |
| 21 | * License along with FFmpeg; if not, write to the Free Software |
| 22 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 23 | */ |
| 24 | |
| 25 | /** |
| 26 | * @file |
| 27 | * The simplest mpeg encoder (well, it was the simplest!). |
| 28 | */ |
| 29 | |
| 30 | #include "libavutil/attributes.h" |
| 31 | #include "libavutil/avassert.h" |
| 32 | #include "libavutil/imgutils.h" |
| 33 | #include "libavutil/internal.h" |
| 34 | #include "libavutil/motion_vector.h" |
| 35 | #include "libavutil/timer.h" |
| 36 | #include "avcodec.h" |
| 37 | #include "blockdsp.h" |
| 38 | #include "h264chroma.h" |
| 39 | #include "idctdsp.h" |
| 40 | #include "internal.h" |
| 41 | #include "mathops.h" |
| 42 | #include "mpegutils.h" |
| 43 | #include "mpegvideo.h" |
| 44 | #include "mjpegenc.h" |
| 45 | #include "msmpeg4.h" |
| 46 | #include "qpeldsp.h" |
| 47 | #include "thread.h" |
| 48 | #include <limits.h> |
| 49 | |
| 50 | static const uint8_t ff_default_chroma_qscale_table[32] = { |
| 51 | // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
| 52 | 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, |
| 53 | 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 |
| 54 | }; |
| 55 | |
| 56 | const uint8_t ff_mpeg1_dc_scale_table[128] = { |
| 57 | // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
| 58 | 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 59 | 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 60 | 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 61 | 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 62 | 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 63 | 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 64 | 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 65 | 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 66 | }; |
| 67 | |
| 68 | static const uint8_t mpeg2_dc_scale_table1[128] = { |
| 69 | // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
| 70 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, |
| 71 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, |
| 72 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, |
| 73 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, |
| 74 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, |
| 75 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, |
| 76 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, |
| 77 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, |
| 78 | }; |
| 79 | |
| 80 | static const uint8_t mpeg2_dc_scale_table2[128] = { |
| 81 | // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
| 82 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 83 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 84 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 85 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 86 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 87 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 88 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 89 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 90 | }; |
| 91 | |
| 92 | static const uint8_t mpeg2_dc_scale_table3[128] = { |
| 93 | // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
| 94 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, |
| 95 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, |
| 96 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, |
| 97 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, |
| 98 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, |
| 99 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, |
| 100 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, |
| 101 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, |
| 102 | }; |
| 103 | |
| 104 | const uint8_t *const ff_mpeg2_dc_scale_table[4] = { |
| 105 | ff_mpeg1_dc_scale_table, |
| 106 | mpeg2_dc_scale_table1, |
| 107 | mpeg2_dc_scale_table2, |
| 108 | mpeg2_dc_scale_table3, |
| 109 | }; |
| 110 | |
| 111 | const uint8_t ff_alternate_horizontal_scan[64] = { |
| 112 | 0, 1, 2, 3, 8, 9, 16, 17, |
| 113 | 10, 11, 4, 5, 6, 7, 15, 14, |
| 114 | 13, 12, 19, 18, 24, 25, 32, 33, |
| 115 | 26, 27, 20, 21, 22, 23, 28, 29, |
| 116 | 30, 31, 34, 35, 40, 41, 48, 49, |
| 117 | 42, 43, 36, 37, 38, 39, 44, 45, |
| 118 | 46, 47, 50, 51, 56, 57, 58, 59, |
| 119 | 52, 53, 54, 55, 60, 61, 62, 63, |
| 120 | }; |
| 121 | |
| 122 | const uint8_t ff_alternate_vertical_scan[64] = { |
| 123 | 0, 8, 16, 24, 1, 9, 2, 10, |
| 124 | 17, 25, 32, 40, 48, 56, 57, 49, |
| 125 | 41, 33, 26, 18, 3, 11, 4, 12, |
| 126 | 19, 27, 34, 42, 50, 58, 35, 43, |
| 127 | 51, 59, 20, 28, 5, 13, 6, 14, |
| 128 | 21, 29, 36, 44, 52, 60, 37, 45, |
| 129 | 53, 61, 22, 30, 7, 15, 23, 31, |
| 130 | 38, 46, 54, 62, 39, 47, 55, 63, |
| 131 | }; |
| 132 | |
| 133 | static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s, |
| 134 | int16_t *block, int n, int qscale) |
| 135 | { |
| 136 | int i, level, nCoeffs; |
| 137 | const uint16_t *quant_matrix; |
| 138 | |
| 139 | nCoeffs= s->block_last_index[n]; |
| 140 | |
| 141 | block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale; |
| 142 | /* XXX: only mpeg1 */ |
| 143 | quant_matrix = s->intra_matrix; |
| 144 | for(i=1;i<=nCoeffs;i++) { |
| 145 | int j= s->intra_scantable.permutated[i]; |
| 146 | level = block[j]; |
| 147 | if (level) { |
| 148 | if (level < 0) { |
| 149 | level = -level; |
| 150 | level = (int)(level * qscale * quant_matrix[j]) >> 3; |
| 151 | level = (level - 1) | 1; |
| 152 | level = -level; |
| 153 | } else { |
| 154 | level = (int)(level * qscale * quant_matrix[j]) >> 3; |
| 155 | level = (level - 1) | 1; |
| 156 | } |
| 157 | block[j] = level; |
| 158 | } |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s, |
| 163 | int16_t *block, int n, int qscale) |
| 164 | { |
| 165 | int i, level, nCoeffs; |
| 166 | const uint16_t *quant_matrix; |
| 167 | |
| 168 | nCoeffs= s->block_last_index[n]; |
| 169 | |
| 170 | quant_matrix = s->inter_matrix; |
| 171 | for(i=0; i<=nCoeffs; i++) { |
| 172 | int j= s->intra_scantable.permutated[i]; |
| 173 | level = block[j]; |
| 174 | if (level) { |
| 175 | if (level < 0) { |
| 176 | level = -level; |
| 177 | level = (((level << 1) + 1) * qscale * |
| 178 | ((int) (quant_matrix[j]))) >> 4; |
| 179 | level = (level - 1) | 1; |
| 180 | level = -level; |
| 181 | } else { |
| 182 | level = (((level << 1) + 1) * qscale * |
| 183 | ((int) (quant_matrix[j]))) >> 4; |
| 184 | level = (level - 1) | 1; |
| 185 | } |
| 186 | block[j] = level; |
| 187 | } |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s, |
| 192 | int16_t *block, int n, int qscale) |
| 193 | { |
| 194 | int i, level, nCoeffs; |
| 195 | const uint16_t *quant_matrix; |
| 196 | |
| 197 | if(s->alternate_scan) nCoeffs= 63; |
| 198 | else nCoeffs= s->block_last_index[n]; |
| 199 | |
| 200 | block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale; |
| 201 | quant_matrix = s->intra_matrix; |
| 202 | for(i=1;i<=nCoeffs;i++) { |
| 203 | int j= s->intra_scantable.permutated[i]; |
| 204 | level = block[j]; |
| 205 | if (level) { |
| 206 | if (level < 0) { |
| 207 | level = -level; |
| 208 | level = (int)(level * qscale * quant_matrix[j]) >> 3; |
| 209 | level = -level; |
| 210 | } else { |
| 211 | level = (int)(level * qscale * quant_matrix[j]) >> 3; |
| 212 | } |
| 213 | block[j] = level; |
| 214 | } |
| 215 | } |
| 216 | } |
| 217 | |
| 218 | static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s, |
| 219 | int16_t *block, int n, int qscale) |
| 220 | { |
| 221 | int i, level, nCoeffs; |
| 222 | const uint16_t *quant_matrix; |
| 223 | int sum=-1; |
| 224 | |
| 225 | if(s->alternate_scan) nCoeffs= 63; |
| 226 | else nCoeffs= s->block_last_index[n]; |
| 227 | |
| 228 | block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale; |
| 229 | sum += block[0]; |
| 230 | quant_matrix = s->intra_matrix; |
| 231 | for(i=1;i<=nCoeffs;i++) { |
| 232 | int j= s->intra_scantable.permutated[i]; |
| 233 | level = block[j]; |
| 234 | if (level) { |
| 235 | if (level < 0) { |
| 236 | level = -level; |
| 237 | level = (int)(level * qscale * quant_matrix[j]) >> 3; |
| 238 | level = -level; |
| 239 | } else { |
| 240 | level = (int)(level * qscale * quant_matrix[j]) >> 3; |
| 241 | } |
| 242 | block[j] = level; |
| 243 | sum+=level; |
| 244 | } |
| 245 | } |
| 246 | block[63]^=sum&1; |
| 247 | } |
| 248 | |
| 249 | static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s, |
| 250 | int16_t *block, int n, int qscale) |
| 251 | { |
| 252 | int i, level, nCoeffs; |
| 253 | const uint16_t *quant_matrix; |
| 254 | int sum=-1; |
| 255 | |
| 256 | if(s->alternate_scan) nCoeffs= 63; |
| 257 | else nCoeffs= s->block_last_index[n]; |
| 258 | |
| 259 | quant_matrix = s->inter_matrix; |
| 260 | for(i=0; i<=nCoeffs; i++) { |
| 261 | int j= s->intra_scantable.permutated[i]; |
| 262 | level = block[j]; |
| 263 | if (level) { |
| 264 | if (level < 0) { |
| 265 | level = -level; |
| 266 | level = (((level << 1) + 1) * qscale * |
| 267 | ((int) (quant_matrix[j]))) >> 4; |
| 268 | level = -level; |
| 269 | } else { |
| 270 | level = (((level << 1) + 1) * qscale * |
| 271 | ((int) (quant_matrix[j]))) >> 4; |
| 272 | } |
| 273 | block[j] = level; |
| 274 | sum+=level; |
| 275 | } |
| 276 | } |
| 277 | block[63]^=sum&1; |
| 278 | } |
| 279 | |
| 280 | static void dct_unquantize_h263_intra_c(MpegEncContext *s, |
| 281 | int16_t *block, int n, int qscale) |
| 282 | { |
| 283 | int i, level, qmul, qadd; |
| 284 | int nCoeffs; |
| 285 | |
| 286 | av_assert2(s->block_last_index[n]>=0 || s->h263_aic); |
| 287 | |
| 288 | qmul = qscale << 1; |
| 289 | |
| 290 | if (!s->h263_aic) { |
| 291 | block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale; |
| 292 | qadd = (qscale - 1) | 1; |
| 293 | }else{ |
| 294 | qadd = 0; |
| 295 | } |
| 296 | if(s->ac_pred) |
| 297 | nCoeffs=63; |
| 298 | else |
| 299 | nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ]; |
| 300 | |
| 301 | for(i=1; i<=nCoeffs; i++) { |
| 302 | level = block[i]; |
| 303 | if (level) { |
| 304 | if (level < 0) { |
| 305 | level = level * qmul - qadd; |
| 306 | } else { |
| 307 | level = level * qmul + qadd; |
| 308 | } |
| 309 | block[i] = level; |
| 310 | } |
| 311 | } |
| 312 | } |
| 313 | |
| 314 | static void dct_unquantize_h263_inter_c(MpegEncContext *s, |
| 315 | int16_t *block, int n, int qscale) |
| 316 | { |
| 317 | int i, level, qmul, qadd; |
| 318 | int nCoeffs; |
| 319 | |
| 320 | av_assert2(s->block_last_index[n]>=0); |
| 321 | |
| 322 | qadd = (qscale - 1) | 1; |
| 323 | qmul = qscale << 1; |
| 324 | |
| 325 | nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ]; |
| 326 | |
| 327 | for(i=0; i<=nCoeffs; i++) { |
| 328 | level = block[i]; |
| 329 | if (level) { |
| 330 | if (level < 0) { |
| 331 | level = level * qmul - qadd; |
| 332 | } else { |
| 333 | level = level * qmul + qadd; |
| 334 | } |
| 335 | block[i] = level; |
| 336 | } |
| 337 | } |
| 338 | } |
| 339 | |
| 340 | static void mpeg_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type, |
| 341 | int (*mv)[2][4][2], |
| 342 | int mb_x, int mb_y, int mb_intra, int mb_skipped) |
| 343 | { |
| 344 | MpegEncContext *s = opaque; |
| 345 | |
| 346 | s->mv_dir = mv_dir; |
| 347 | s->mv_type = mv_type; |
| 348 | s->mb_intra = mb_intra; |
| 349 | s->mb_skipped = mb_skipped; |
| 350 | s->mb_x = mb_x; |
| 351 | s->mb_y = mb_y; |
| 352 | memcpy(s->mv, mv, sizeof(*mv)); |
| 353 | |
| 354 | ff_init_block_index(s); |
| 355 | ff_update_block_index(s); |
| 356 | |
| 357 | s->bdsp.clear_blocks(s->block[0]); |
| 358 | |
| 359 | s->dest[0] = s->current_picture.f->data[0] + (s->mb_y * 16 * s->linesize) + s->mb_x * 16; |
| 360 | s->dest[1] = s->current_picture.f->data[1] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift); |
| 361 | s->dest[2] = s->current_picture.f->data[2] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift); |
| 362 | |
| 363 | if (ref) |
| 364 | av_log(s->avctx, AV_LOG_DEBUG, "Interlaced error concealment is not fully implemented\n"); |
| 365 | ff_mpv_decode_mb(s, s->block); |
| 366 | } |
| 367 | |
| 368 | static void gray16(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h) |
| 369 | { |
| 370 | while(h--) |
| 371 | memset(dst + h*linesize, 128, 16); |
| 372 | } |
| 373 | |
| 374 | static void gray8(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h) |
| 375 | { |
| 376 | while(h--) |
| 377 | memset(dst + h*linesize, 128, 8); |
| 378 | } |
| 379 | |
| 380 | /* init common dct for both encoder and decoder */ |
| 381 | static av_cold int dct_init(MpegEncContext *s) |
| 382 | { |
| 383 | ff_blockdsp_init(&s->bdsp, s->avctx); |
| 384 | ff_h264chroma_init(&s->h264chroma, 8); //for lowres |
| 385 | ff_hpeldsp_init(&s->hdsp, s->avctx->flags); |
| 386 | ff_me_cmp_init(&s->mecc, s->avctx); |
| 387 | ff_mpegvideodsp_init(&s->mdsp); |
| 388 | ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample); |
| 389 | |
| 390 | if (s->avctx->debug & FF_DEBUG_NOMC) { |
| 391 | int i; |
| 392 | for (i=0; i<4; i++) { |
| 393 | s->hdsp.avg_pixels_tab[0][i] = gray16; |
| 394 | s->hdsp.put_pixels_tab[0][i] = gray16; |
| 395 | s->hdsp.put_no_rnd_pixels_tab[0][i] = gray16; |
| 396 | |
| 397 | s->hdsp.avg_pixels_tab[1][i] = gray8; |
| 398 | s->hdsp.put_pixels_tab[1][i] = gray8; |
| 399 | s->hdsp.put_no_rnd_pixels_tab[1][i] = gray8; |
| 400 | } |
| 401 | } |
| 402 | |
| 403 | s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c; |
| 404 | s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c; |
| 405 | s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c; |
| 406 | s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c; |
| 407 | s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c; |
| 408 | if (s->flags & CODEC_FLAG_BITEXACT) |
| 409 | s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact; |
| 410 | s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c; |
| 411 | |
| 412 | if (HAVE_INTRINSICS_NEON) |
| 413 | ff_mpv_common_init_neon(s); |
| 414 | |
| 415 | if (ARCH_ALPHA) |
| 416 | ff_mpv_common_init_axp(s); |
| 417 | if (ARCH_ARM) |
| 418 | ff_mpv_common_init_arm(s); |
| 419 | if (ARCH_PPC) |
| 420 | ff_mpv_common_init_ppc(s); |
| 421 | if (ARCH_X86) |
| 422 | ff_mpv_common_init_x86(s); |
| 423 | |
| 424 | return 0; |
| 425 | } |
| 426 | |
| 427 | av_cold void ff_mpv_idct_init(MpegEncContext *s) |
| 428 | { |
| 429 | ff_idctdsp_init(&s->idsp, s->avctx); |
| 430 | |
| 431 | /* load & permutate scantables |
| 432 | * note: only wmv uses different ones |
| 433 | */ |
| 434 | if (s->alternate_scan) { |
| 435 | ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_alternate_vertical_scan); |
| 436 | ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_alternate_vertical_scan); |
| 437 | } else { |
| 438 | ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_zigzag_direct); |
| 439 | ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_zigzag_direct); |
| 440 | } |
| 441 | ff_init_scantable(s->idsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan); |
| 442 | ff_init_scantable(s->idsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan); |
| 443 | } |
| 444 | |
| 445 | static int frame_size_alloc(MpegEncContext *s, int linesize) |
| 446 | { |
| 447 | int alloc_size = FFALIGN(FFABS(linesize) + 64, 32); |
| 448 | |
| 449 | if (s->avctx->hwaccel || s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) |
| 450 | return 0; |
| 451 | |
| 452 | if (linesize < 24) { |
| 453 | av_log(s->avctx, AV_LOG_ERROR, "Image too small, temporary buffers cannot function\n"); |
| 454 | return AVERROR_PATCHWELCOME; |
| 455 | } |
| 456 | |
| 457 | // edge emu needs blocksize + filter length - 1 |
| 458 | // (= 17x17 for halfpel / 21x21 for h264) |
| 459 | // VC1 computes luma and chroma simultaneously and needs 19X19 + 9x9 |
| 460 | // at uvlinesize. It supports only YUV420 so 24x24 is enough |
| 461 | // linesize * interlaced * MBsize |
| 462 | // we also use this buffer for encoding in encode_mb_internal() needig an additional 32 lines |
| 463 | FF_ALLOCZ_ARRAY_OR_GOTO(s->avctx, s->edge_emu_buffer, alloc_size, 4 * 68, |
| 464 | fail); |
| 465 | |
| 466 | FF_ALLOCZ_ARRAY_OR_GOTO(s->avctx, s->me.scratchpad, alloc_size, 4 * 16 * 2, |
| 467 | fail) |
| 468 | s->me.temp = s->me.scratchpad; |
| 469 | s->rd_scratchpad = s->me.scratchpad; |
| 470 | s->b_scratchpad = s->me.scratchpad; |
| 471 | s->obmc_scratchpad = s->me.scratchpad + 16; |
| 472 | |
| 473 | return 0; |
| 474 | fail: |
| 475 | av_freep(&s->edge_emu_buffer); |
| 476 | return AVERROR(ENOMEM); |
| 477 | } |
| 478 | |
| 479 | /** |
| 480 | * Allocate a frame buffer |
| 481 | */ |
| 482 | static int alloc_frame_buffer(MpegEncContext *s, Picture *pic) |
| 483 | { |
| 484 | int edges_needed = av_codec_is_encoder(s->avctx->codec); |
| 485 | int r, ret; |
| 486 | |
| 487 | pic->tf.f = pic->f; |
| 488 | if (s->codec_id != AV_CODEC_ID_WMV3IMAGE && |
| 489 | s->codec_id != AV_CODEC_ID_VC1IMAGE && |
| 490 | s->codec_id != AV_CODEC_ID_MSS2) { |
| 491 | if (edges_needed) { |
| 492 | pic->f->width = s->avctx->width + 2 * EDGE_WIDTH; |
| 493 | pic->f->height = s->avctx->height + 2 * EDGE_WIDTH; |
| 494 | } |
| 495 | |
| 496 | r = ff_thread_get_buffer(s->avctx, &pic->tf, |
| 497 | pic->reference ? AV_GET_BUFFER_FLAG_REF : 0); |
| 498 | } else { |
| 499 | pic->f->width = s->avctx->width; |
| 500 | pic->f->height = s->avctx->height; |
| 501 | pic->f->format = s->avctx->pix_fmt; |
| 502 | r = avcodec_default_get_buffer2(s->avctx, pic->f, 0); |
| 503 | } |
| 504 | |
| 505 | if (r < 0 || !pic->f->buf[0]) { |
| 506 | av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %p)\n", |
| 507 | r, pic->f->data[0]); |
| 508 | return -1; |
| 509 | } |
| 510 | |
| 511 | if (edges_needed) { |
| 512 | int i; |
| 513 | for (i = 0; pic->f->data[i]; i++) { |
| 514 | int offset = (EDGE_WIDTH >> (i ? s->chroma_y_shift : 0)) * |
| 515 | pic->f->linesize[i] + |
| 516 | (EDGE_WIDTH >> (i ? s->chroma_x_shift : 0)); |
| 517 | pic->f->data[i] += offset; |
| 518 | } |
| 519 | pic->f->width = s->avctx->width; |
| 520 | pic->f->height = s->avctx->height; |
| 521 | } |
| 522 | |
| 523 | if (s->avctx->hwaccel) { |
| 524 | assert(!pic->hwaccel_picture_private); |
| 525 | if (s->avctx->hwaccel->frame_priv_data_size) { |
| 526 | pic->hwaccel_priv_buf = av_buffer_allocz(s->avctx->hwaccel->frame_priv_data_size); |
| 527 | if (!pic->hwaccel_priv_buf) { |
| 528 | av_log(s->avctx, AV_LOG_ERROR, "alloc_frame_buffer() failed (hwaccel private data allocation)\n"); |
| 529 | return -1; |
| 530 | } |
| 531 | pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data; |
| 532 | } |
| 533 | } |
| 534 | |
| 535 | if (s->linesize && (s->linesize != pic->f->linesize[0] || |
| 536 | s->uvlinesize != pic->f->linesize[1])) { |
| 537 | av_log(s->avctx, AV_LOG_ERROR, |
| 538 | "get_buffer() failed (stride changed)\n"); |
| 539 | ff_mpeg_unref_picture(s, pic); |
| 540 | return -1; |
| 541 | } |
| 542 | |
| 543 | if (pic->f->linesize[1] != pic->f->linesize[2]) { |
| 544 | av_log(s->avctx, AV_LOG_ERROR, |
| 545 | "get_buffer() failed (uv stride mismatch)\n"); |
| 546 | ff_mpeg_unref_picture(s, pic); |
| 547 | return -1; |
| 548 | } |
| 549 | |
| 550 | if (!s->edge_emu_buffer && |
| 551 | (ret = frame_size_alloc(s, pic->f->linesize[0])) < 0) { |
| 552 | av_log(s->avctx, AV_LOG_ERROR, |
| 553 | "get_buffer() failed to allocate context scratch buffers.\n"); |
| 554 | ff_mpeg_unref_picture(s, pic); |
| 555 | return ret; |
| 556 | } |
| 557 | |
| 558 | return 0; |
| 559 | } |
| 560 | |
| 561 | void ff_free_picture_tables(Picture *pic) |
| 562 | { |
| 563 | int i; |
| 564 | |
| 565 | pic->alloc_mb_width = |
| 566 | pic->alloc_mb_height = 0; |
| 567 | |
| 568 | av_buffer_unref(&pic->mb_var_buf); |
| 569 | av_buffer_unref(&pic->mc_mb_var_buf); |
| 570 | av_buffer_unref(&pic->mb_mean_buf); |
| 571 | av_buffer_unref(&pic->mbskip_table_buf); |
| 572 | av_buffer_unref(&pic->qscale_table_buf); |
| 573 | av_buffer_unref(&pic->mb_type_buf); |
| 574 | |
| 575 | for (i = 0; i < 2; i++) { |
| 576 | av_buffer_unref(&pic->motion_val_buf[i]); |
| 577 | av_buffer_unref(&pic->ref_index_buf[i]); |
| 578 | } |
| 579 | } |
| 580 | |
| 581 | static int alloc_picture_tables(MpegEncContext *s, Picture *pic) |
| 582 | { |
| 583 | const int big_mb_num = s->mb_stride * (s->mb_height + 1) + 1; |
| 584 | const int mb_array_size = s->mb_stride * s->mb_height; |
| 585 | const int b8_array_size = s->b8_stride * s->mb_height * 2; |
| 586 | int i; |
| 587 | |
| 588 | |
| 589 | pic->mbskip_table_buf = av_buffer_allocz(mb_array_size + 2); |
| 590 | pic->qscale_table_buf = av_buffer_allocz(big_mb_num + s->mb_stride); |
| 591 | pic->mb_type_buf = av_buffer_allocz((big_mb_num + s->mb_stride) * |
| 592 | sizeof(uint32_t)); |
| 593 | if (!pic->mbskip_table_buf || !pic->qscale_table_buf || !pic->mb_type_buf) |
| 594 | return AVERROR(ENOMEM); |
| 595 | |
| 596 | if (s->encoding) { |
| 597 | pic->mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t)); |
| 598 | pic->mc_mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t)); |
| 599 | pic->mb_mean_buf = av_buffer_allocz(mb_array_size); |
| 600 | if (!pic->mb_var_buf || !pic->mc_mb_var_buf || !pic->mb_mean_buf) |
| 601 | return AVERROR(ENOMEM); |
| 602 | } |
| 603 | |
| 604 | if (s->out_format == FMT_H263 || s->encoding || s->avctx->debug_mv || |
| 605 | (s->avctx->flags2 & CODEC_FLAG2_EXPORT_MVS)) { |
| 606 | int mv_size = 2 * (b8_array_size + 4) * sizeof(int16_t); |
| 607 | int ref_index_size = 4 * mb_array_size; |
| 608 | |
| 609 | for (i = 0; mv_size && i < 2; i++) { |
| 610 | pic->motion_val_buf[i] = av_buffer_allocz(mv_size); |
| 611 | pic->ref_index_buf[i] = av_buffer_allocz(ref_index_size); |
| 612 | if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) |
| 613 | return AVERROR(ENOMEM); |
| 614 | } |
| 615 | } |
| 616 | |
| 617 | pic->alloc_mb_width = s->mb_width; |
| 618 | pic->alloc_mb_height = s->mb_height; |
| 619 | |
| 620 | return 0; |
| 621 | } |
| 622 | |
| 623 | static int make_tables_writable(Picture *pic) |
| 624 | { |
| 625 | int ret, i; |
| 626 | #define MAKE_WRITABLE(table) \ |
| 627 | do {\ |
| 628 | if (pic->table &&\ |
| 629 | (ret = av_buffer_make_writable(&pic->table)) < 0)\ |
| 630 | return ret;\ |
| 631 | } while (0) |
| 632 | |
| 633 | MAKE_WRITABLE(mb_var_buf); |
| 634 | MAKE_WRITABLE(mc_mb_var_buf); |
| 635 | MAKE_WRITABLE(mb_mean_buf); |
| 636 | MAKE_WRITABLE(mbskip_table_buf); |
| 637 | MAKE_WRITABLE(qscale_table_buf); |
| 638 | MAKE_WRITABLE(mb_type_buf); |
| 639 | |
| 640 | for (i = 0; i < 2; i++) { |
| 641 | MAKE_WRITABLE(motion_val_buf[i]); |
| 642 | MAKE_WRITABLE(ref_index_buf[i]); |
| 643 | } |
| 644 | |
| 645 | return 0; |
| 646 | } |
| 647 | |
| 648 | /** |
| 649 | * Allocate a Picture. |
| 650 | * The pixels are allocated/set by calling get_buffer() if shared = 0 |
| 651 | */ |
| 652 | int ff_alloc_picture(MpegEncContext *s, Picture *pic, int shared) |
| 653 | { |
| 654 | int i, ret; |
| 655 | |
| 656 | if (pic->qscale_table_buf) |
| 657 | if ( pic->alloc_mb_width != s->mb_width |
| 658 | || pic->alloc_mb_height != s->mb_height) |
| 659 | ff_free_picture_tables(pic); |
| 660 | |
| 661 | if (shared) { |
| 662 | av_assert0(pic->f->data[0]); |
| 663 | pic->shared = 1; |
| 664 | } else { |
| 665 | av_assert0(!pic->f->buf[0]); |
| 666 | |
| 667 | if (alloc_frame_buffer(s, pic) < 0) |
| 668 | return -1; |
| 669 | |
| 670 | s->linesize = pic->f->linesize[0]; |
| 671 | s->uvlinesize = pic->f->linesize[1]; |
| 672 | } |
| 673 | |
| 674 | if (!pic->qscale_table_buf) |
| 675 | ret = alloc_picture_tables(s, pic); |
| 676 | else |
| 677 | ret = make_tables_writable(pic); |
| 678 | if (ret < 0) |
| 679 | goto fail; |
| 680 | |
| 681 | if (s->encoding) { |
| 682 | pic->mb_var = (uint16_t*)pic->mb_var_buf->data; |
| 683 | pic->mc_mb_var = (uint16_t*)pic->mc_mb_var_buf->data; |
| 684 | pic->mb_mean = pic->mb_mean_buf->data; |
| 685 | } |
| 686 | |
| 687 | pic->mbskip_table = pic->mbskip_table_buf->data; |
| 688 | pic->qscale_table = pic->qscale_table_buf->data + 2 * s->mb_stride + 1; |
| 689 | pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * s->mb_stride + 1; |
| 690 | |
| 691 | if (pic->motion_val_buf[0]) { |
| 692 | for (i = 0; i < 2; i++) { |
| 693 | pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4; |
| 694 | pic->ref_index[i] = pic->ref_index_buf[i]->data; |
| 695 | } |
| 696 | } |
| 697 | |
| 698 | return 0; |
| 699 | fail: |
| 700 | av_log(s->avctx, AV_LOG_ERROR, "Error allocating a picture.\n"); |
| 701 | ff_mpeg_unref_picture(s, pic); |
| 702 | ff_free_picture_tables(pic); |
| 703 | return AVERROR(ENOMEM); |
| 704 | } |
| 705 | |
| 706 | /** |
| 707 | * Deallocate a picture. |
| 708 | */ |
| 709 | void ff_mpeg_unref_picture(MpegEncContext *s, Picture *pic) |
| 710 | { |
| 711 | int off = offsetof(Picture, mb_mean) + sizeof(pic->mb_mean); |
| 712 | |
| 713 | pic->tf.f = pic->f; |
| 714 | /* WM Image / Screen codecs allocate internal buffers with different |
| 715 | * dimensions / colorspaces; ignore user-defined callbacks for these. */ |
| 716 | if (s->codec_id != AV_CODEC_ID_WMV3IMAGE && |
| 717 | s->codec_id != AV_CODEC_ID_VC1IMAGE && |
| 718 | s->codec_id != AV_CODEC_ID_MSS2) |
| 719 | ff_thread_release_buffer(s->avctx, &pic->tf); |
| 720 | else if (pic->f) |
| 721 | av_frame_unref(pic->f); |
| 722 | |
| 723 | av_buffer_unref(&pic->hwaccel_priv_buf); |
| 724 | |
| 725 | if (pic->needs_realloc) |
| 726 | ff_free_picture_tables(pic); |
| 727 | |
| 728 | memset((uint8_t*)pic + off, 0, sizeof(*pic) - off); |
| 729 | } |
| 730 | |
| 731 | static int update_picture_tables(Picture *dst, Picture *src) |
| 732 | { |
| 733 | int i; |
| 734 | |
| 735 | #define UPDATE_TABLE(table)\ |
| 736 | do {\ |
| 737 | if (src->table &&\ |
| 738 | (!dst->table || dst->table->buffer != src->table->buffer)) {\ |
| 739 | av_buffer_unref(&dst->table);\ |
| 740 | dst->table = av_buffer_ref(src->table);\ |
| 741 | if (!dst->table) {\ |
| 742 | ff_free_picture_tables(dst);\ |
| 743 | return AVERROR(ENOMEM);\ |
| 744 | }\ |
| 745 | }\ |
| 746 | } while (0) |
| 747 | |
| 748 | UPDATE_TABLE(mb_var_buf); |
| 749 | UPDATE_TABLE(mc_mb_var_buf); |
| 750 | UPDATE_TABLE(mb_mean_buf); |
| 751 | UPDATE_TABLE(mbskip_table_buf); |
| 752 | UPDATE_TABLE(qscale_table_buf); |
| 753 | UPDATE_TABLE(mb_type_buf); |
| 754 | for (i = 0; i < 2; i++) { |
| 755 | UPDATE_TABLE(motion_val_buf[i]); |
| 756 | UPDATE_TABLE(ref_index_buf[i]); |
| 757 | } |
| 758 | |
| 759 | dst->mb_var = src->mb_var; |
| 760 | dst->mc_mb_var = src->mc_mb_var; |
| 761 | dst->mb_mean = src->mb_mean; |
| 762 | dst->mbskip_table = src->mbskip_table; |
| 763 | dst->qscale_table = src->qscale_table; |
| 764 | dst->mb_type = src->mb_type; |
| 765 | for (i = 0; i < 2; i++) { |
| 766 | dst->motion_val[i] = src->motion_val[i]; |
| 767 | dst->ref_index[i] = src->ref_index[i]; |
| 768 | } |
| 769 | |
| 770 | dst->alloc_mb_width = src->alloc_mb_width; |
| 771 | dst->alloc_mb_height = src->alloc_mb_height; |
| 772 | |
| 773 | return 0; |
| 774 | } |
| 775 | |
| 776 | int ff_mpeg_ref_picture(MpegEncContext *s, Picture *dst, Picture *src) |
| 777 | { |
| 778 | int ret; |
| 779 | |
| 780 | av_assert0(!dst->f->buf[0]); |
| 781 | av_assert0(src->f->buf[0]); |
| 782 | |
| 783 | src->tf.f = src->f; |
| 784 | dst->tf.f = dst->f; |
| 785 | ret = ff_thread_ref_frame(&dst->tf, &src->tf); |
| 786 | if (ret < 0) |
| 787 | goto fail; |
| 788 | |
| 789 | ret = update_picture_tables(dst, src); |
| 790 | if (ret < 0) |
| 791 | goto fail; |
| 792 | |
| 793 | if (src->hwaccel_picture_private) { |
| 794 | dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf); |
| 795 | if (!dst->hwaccel_priv_buf) |
| 796 | goto fail; |
| 797 | dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data; |
| 798 | } |
| 799 | |
| 800 | dst->field_picture = src->field_picture; |
| 801 | dst->mb_var_sum = src->mb_var_sum; |
| 802 | dst->mc_mb_var_sum = src->mc_mb_var_sum; |
| 803 | dst->b_frame_score = src->b_frame_score; |
| 804 | dst->needs_realloc = src->needs_realloc; |
| 805 | dst->reference = src->reference; |
| 806 | dst->shared = src->shared; |
| 807 | |
| 808 | return 0; |
| 809 | fail: |
| 810 | ff_mpeg_unref_picture(s, dst); |
| 811 | return ret; |
| 812 | } |
| 813 | |
| 814 | static void exchange_uv(MpegEncContext *s) |
| 815 | { |
| 816 | int16_t (*tmp)[64]; |
| 817 | |
| 818 | tmp = s->pblocks[4]; |
| 819 | s->pblocks[4] = s->pblocks[5]; |
| 820 | s->pblocks[5] = tmp; |
| 821 | } |
| 822 | |
| 823 | static int init_duplicate_context(MpegEncContext *s) |
| 824 | { |
| 825 | int y_size = s->b8_stride * (2 * s->mb_height + 1); |
| 826 | int c_size = s->mb_stride * (s->mb_height + 1); |
| 827 | int yc_size = y_size + 2 * c_size; |
| 828 | int i; |
| 829 | |
| 830 | if (s->mb_height & 1) |
| 831 | yc_size += 2*s->b8_stride + 2*s->mb_stride; |
| 832 | |
| 833 | s->edge_emu_buffer = |
| 834 | s->me.scratchpad = |
| 835 | s->me.temp = |
| 836 | s->rd_scratchpad = |
| 837 | s->b_scratchpad = |
| 838 | s->obmc_scratchpad = NULL; |
| 839 | |
| 840 | if (s->encoding) { |
| 841 | FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map, |
| 842 | ME_MAP_SIZE * sizeof(uint32_t), fail) |
| 843 | FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map, |
| 844 | ME_MAP_SIZE * sizeof(uint32_t), fail) |
| 845 | if (s->avctx->noise_reduction) { |
| 846 | FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum, |
| 847 | 2 * 64 * sizeof(int), fail) |
| 848 | } |
| 849 | } |
| 850 | FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(int16_t), fail) |
| 851 | s->block = s->blocks[0]; |
| 852 | |
| 853 | for (i = 0; i < 12; i++) { |
| 854 | s->pblocks[i] = &s->block[i]; |
| 855 | } |
| 856 | if (s->avctx->codec_tag == AV_RL32("VCR2")) |
| 857 | exchange_uv(s); |
| 858 | |
| 859 | if (s->out_format == FMT_H263) { |
| 860 | /* ac values */ |
| 861 | FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base, |
| 862 | yc_size * sizeof(int16_t) * 16, fail); |
| 863 | s->ac_val[0] = s->ac_val_base + s->b8_stride + 1; |
| 864 | s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1; |
| 865 | s->ac_val[2] = s->ac_val[1] + c_size; |
| 866 | } |
| 867 | |
| 868 | return 0; |
| 869 | fail: |
| 870 | return -1; // free() through ff_mpv_common_end() |
| 871 | } |
| 872 | |
| 873 | static void free_duplicate_context(MpegEncContext *s) |
| 874 | { |
| 875 | if (!s) |
| 876 | return; |
| 877 | |
| 878 | av_freep(&s->edge_emu_buffer); |
| 879 | av_freep(&s->me.scratchpad); |
| 880 | s->me.temp = |
| 881 | s->rd_scratchpad = |
| 882 | s->b_scratchpad = |
| 883 | s->obmc_scratchpad = NULL; |
| 884 | |
| 885 | av_freep(&s->dct_error_sum); |
| 886 | av_freep(&s->me.map); |
| 887 | av_freep(&s->me.score_map); |
| 888 | av_freep(&s->blocks); |
| 889 | av_freep(&s->ac_val_base); |
| 890 | s->block = NULL; |
| 891 | } |
| 892 | |
| 893 | static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src) |
| 894 | { |
| 895 | #define COPY(a) bak->a = src->a |
| 896 | COPY(edge_emu_buffer); |
| 897 | COPY(me.scratchpad); |
| 898 | COPY(me.temp); |
| 899 | COPY(rd_scratchpad); |
| 900 | COPY(b_scratchpad); |
| 901 | COPY(obmc_scratchpad); |
| 902 | COPY(me.map); |
| 903 | COPY(me.score_map); |
| 904 | COPY(blocks); |
| 905 | COPY(block); |
| 906 | COPY(start_mb_y); |
| 907 | COPY(end_mb_y); |
| 908 | COPY(me.map_generation); |
| 909 | COPY(pb); |
| 910 | COPY(dct_error_sum); |
| 911 | COPY(dct_count[0]); |
| 912 | COPY(dct_count[1]); |
| 913 | COPY(ac_val_base); |
| 914 | COPY(ac_val[0]); |
| 915 | COPY(ac_val[1]); |
| 916 | COPY(ac_val[2]); |
| 917 | #undef COPY |
| 918 | } |
| 919 | |
| 920 | int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src) |
| 921 | { |
| 922 | MpegEncContext bak; |
| 923 | int i, ret; |
| 924 | // FIXME copy only needed parts |
| 925 | // START_TIMER |
| 926 | backup_duplicate_context(&bak, dst); |
| 927 | memcpy(dst, src, sizeof(MpegEncContext)); |
| 928 | backup_duplicate_context(dst, &bak); |
| 929 | for (i = 0; i < 12; i++) { |
| 930 | dst->pblocks[i] = &dst->block[i]; |
| 931 | } |
| 932 | if (dst->avctx->codec_tag == AV_RL32("VCR2")) |
| 933 | exchange_uv(dst); |
| 934 | if (!dst->edge_emu_buffer && |
| 935 | (ret = frame_size_alloc(dst, dst->linesize)) < 0) { |
| 936 | av_log(dst->avctx, AV_LOG_ERROR, "failed to allocate context " |
| 937 | "scratch buffers.\n"); |
| 938 | return ret; |
| 939 | } |
| 940 | // STOP_TIMER("update_duplicate_context") |
| 941 | // about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads |
| 942 | return 0; |
| 943 | } |
| 944 | |
| 945 | int ff_mpeg_update_thread_context(AVCodecContext *dst, |
| 946 | const AVCodecContext *src) |
| 947 | { |
| 948 | int i, ret; |
| 949 | MpegEncContext *s = dst->priv_data, *s1 = src->priv_data; |
| 950 | |
| 951 | if (dst == src) |
| 952 | return 0; |
| 953 | |
| 954 | av_assert0(s != s1); |
| 955 | |
| 956 | // FIXME can parameters change on I-frames? |
| 957 | // in that case dst may need a reinit |
| 958 | if (!s->context_initialized) { |
| 959 | memcpy(s, s1, sizeof(MpegEncContext)); |
| 960 | |
| 961 | s->avctx = dst; |
| 962 | s->bitstream_buffer = NULL; |
| 963 | s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0; |
| 964 | |
| 965 | if (s1->context_initialized){ |
| 966 | // s->picture_range_start += MAX_PICTURE_COUNT; |
| 967 | // s->picture_range_end += MAX_PICTURE_COUNT; |
| 968 | ff_mpv_idct_init(s); |
| 969 | if((ret = ff_mpv_common_init(s)) < 0){ |
| 970 | memset(s, 0, sizeof(MpegEncContext)); |
| 971 | s->avctx = dst; |
| 972 | return ret; |
| 973 | } |
| 974 | } |
| 975 | } |
| 976 | |
| 977 | if (s->height != s1->height || s->width != s1->width || s->context_reinit) { |
| 978 | s->context_reinit = 0; |
| 979 | s->height = s1->height; |
| 980 | s->width = s1->width; |
| 981 | if ((ret = ff_mpv_common_frame_size_change(s)) < 0) |
| 982 | return ret; |
| 983 | } |
| 984 | |
| 985 | s->avctx->coded_height = s1->avctx->coded_height; |
| 986 | s->avctx->coded_width = s1->avctx->coded_width; |
| 987 | s->avctx->width = s1->avctx->width; |
| 988 | s->avctx->height = s1->avctx->height; |
| 989 | |
| 990 | s->coded_picture_number = s1->coded_picture_number; |
| 991 | s->picture_number = s1->picture_number; |
| 992 | |
| 993 | av_assert0(!s->picture || s->picture != s1->picture); |
| 994 | if(s->picture) |
| 995 | for (i = 0; i < MAX_PICTURE_COUNT; i++) { |
| 996 | ff_mpeg_unref_picture(s, &s->picture[i]); |
| 997 | if (s1->picture[i].f->buf[0] && |
| 998 | (ret = ff_mpeg_ref_picture(s, &s->picture[i], &s1->picture[i])) < 0) |
| 999 | return ret; |
| 1000 | } |
| 1001 | |
| 1002 | #define UPDATE_PICTURE(pic)\ |
| 1003 | do {\ |
| 1004 | ff_mpeg_unref_picture(s, &s->pic);\ |
| 1005 | if (s1->pic.f && s1->pic.f->buf[0])\ |
| 1006 | ret = ff_mpeg_ref_picture(s, &s->pic, &s1->pic);\ |
| 1007 | else\ |
| 1008 | ret = update_picture_tables(&s->pic, &s1->pic);\ |
| 1009 | if (ret < 0)\ |
| 1010 | return ret;\ |
| 1011 | } while (0) |
| 1012 | |
| 1013 | UPDATE_PICTURE(current_picture); |
| 1014 | UPDATE_PICTURE(last_picture); |
| 1015 | UPDATE_PICTURE(next_picture); |
| 1016 | |
| 1017 | s->last_picture_ptr = REBASE_PICTURE(s1->last_picture_ptr, s, s1); |
| 1018 | s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1); |
| 1019 | s->next_picture_ptr = REBASE_PICTURE(s1->next_picture_ptr, s, s1); |
| 1020 | |
| 1021 | // Error/bug resilience |
| 1022 | s->next_p_frame_damaged = s1->next_p_frame_damaged; |
| 1023 | s->workaround_bugs = s1->workaround_bugs; |
| 1024 | s->padding_bug_score = s1->padding_bug_score; |
| 1025 | |
| 1026 | // MPEG4 timing info |
| 1027 | memcpy(&s->last_time_base, &s1->last_time_base, |
| 1028 | (char *) &s1->pb_field_time + sizeof(s1->pb_field_time) - |
| 1029 | (char *) &s1->last_time_base); |
| 1030 | |
| 1031 | // B-frame info |
| 1032 | s->max_b_frames = s1->max_b_frames; |
| 1033 | s->low_delay = s1->low_delay; |
| 1034 | s->droppable = s1->droppable; |
| 1035 | |
| 1036 | // DivX handling (doesn't work) |
| 1037 | s->divx_packed = s1->divx_packed; |
| 1038 | |
| 1039 | if (s1->bitstream_buffer) { |
| 1040 | if (s1->bitstream_buffer_size + |
| 1041 | FF_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size) |
| 1042 | av_fast_malloc(&s->bitstream_buffer, |
| 1043 | &s->allocated_bitstream_buffer_size, |
| 1044 | s1->allocated_bitstream_buffer_size); |
| 1045 | s->bitstream_buffer_size = s1->bitstream_buffer_size; |
| 1046 | memcpy(s->bitstream_buffer, s1->bitstream_buffer, |
| 1047 | s1->bitstream_buffer_size); |
| 1048 | memset(s->bitstream_buffer + s->bitstream_buffer_size, 0, |
| 1049 | FF_INPUT_BUFFER_PADDING_SIZE); |
| 1050 | } |
| 1051 | |
| 1052 | // linesize dependend scratch buffer allocation |
| 1053 | if (!s->edge_emu_buffer) |
| 1054 | if (s1->linesize) { |
| 1055 | if (frame_size_alloc(s, s1->linesize) < 0) { |
| 1056 | av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context " |
| 1057 | "scratch buffers.\n"); |
| 1058 | return AVERROR(ENOMEM); |
| 1059 | } |
| 1060 | } else { |
| 1061 | av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not " |
| 1062 | "be allocated due to unknown size.\n"); |
| 1063 | } |
| 1064 | |
| 1065 | // MPEG2/interlacing info |
| 1066 | memcpy(&s->progressive_sequence, &s1->progressive_sequence, |
| 1067 | (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence); |
| 1068 | |
| 1069 | if (!s1->first_field) { |
| 1070 | s->last_pict_type = s1->pict_type; |
| 1071 | if (s1->current_picture_ptr) |
| 1072 | s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f->quality; |
| 1073 | } |
| 1074 | |
| 1075 | return 0; |
| 1076 | } |
| 1077 | |
| 1078 | /** |
| 1079 | * Set the given MpegEncContext to common defaults |
| 1080 | * (same for encoding and decoding). |
| 1081 | * The changed fields will not depend upon the |
| 1082 | * prior state of the MpegEncContext. |
| 1083 | */ |
| 1084 | void ff_mpv_common_defaults(MpegEncContext *s) |
| 1085 | { |
| 1086 | s->y_dc_scale_table = |
| 1087 | s->c_dc_scale_table = ff_mpeg1_dc_scale_table; |
| 1088 | s->chroma_qscale_table = ff_default_chroma_qscale_table; |
| 1089 | s->progressive_frame = 1; |
| 1090 | s->progressive_sequence = 1; |
| 1091 | s->picture_structure = PICT_FRAME; |
| 1092 | |
| 1093 | s->coded_picture_number = 0; |
| 1094 | s->picture_number = 0; |
| 1095 | |
| 1096 | s->f_code = 1; |
| 1097 | s->b_code = 1; |
| 1098 | |
| 1099 | s->slice_context_count = 1; |
| 1100 | } |
| 1101 | |
| 1102 | /** |
| 1103 | * Set the given MpegEncContext to defaults for decoding. |
| 1104 | * the changed fields will not depend upon |
| 1105 | * the prior state of the MpegEncContext. |
| 1106 | */ |
| 1107 | void ff_mpv_decode_defaults(MpegEncContext *s) |
| 1108 | { |
| 1109 | ff_mpv_common_defaults(s); |
| 1110 | } |
| 1111 | |
| 1112 | void ff_mpv_decode_init(MpegEncContext *s, AVCodecContext *avctx) |
| 1113 | { |
| 1114 | s->avctx = avctx; |
| 1115 | s->width = avctx->coded_width; |
| 1116 | s->height = avctx->coded_height; |
| 1117 | s->codec_id = avctx->codec->id; |
| 1118 | s->workaround_bugs = avctx->workaround_bugs; |
| 1119 | s->flags = avctx->flags; |
| 1120 | s->flags2 = avctx->flags2; |
| 1121 | |
| 1122 | /* convert fourcc to upper case */ |
| 1123 | s->codec_tag = avpriv_toupper4(avctx->codec_tag); |
| 1124 | |
| 1125 | s->stream_codec_tag = avpriv_toupper4(avctx->stream_codec_tag); |
| 1126 | } |
| 1127 | |
| 1128 | static int init_er(MpegEncContext *s) |
| 1129 | { |
| 1130 | ERContext *er = &s->er; |
| 1131 | int mb_array_size = s->mb_height * s->mb_stride; |
| 1132 | int i; |
| 1133 | |
| 1134 | er->avctx = s->avctx; |
| 1135 | er->mecc = &s->mecc; |
| 1136 | |
| 1137 | er->mb_index2xy = s->mb_index2xy; |
| 1138 | er->mb_num = s->mb_num; |
| 1139 | er->mb_width = s->mb_width; |
| 1140 | er->mb_height = s->mb_height; |
| 1141 | er->mb_stride = s->mb_stride; |
| 1142 | er->b8_stride = s->b8_stride; |
| 1143 | |
| 1144 | er->er_temp_buffer = av_malloc(s->mb_height * s->mb_stride); |
| 1145 | er->error_status_table = av_mallocz(mb_array_size); |
| 1146 | if (!er->er_temp_buffer || !er->error_status_table) |
| 1147 | goto fail; |
| 1148 | |
| 1149 | er->mbskip_table = s->mbskip_table; |
| 1150 | er->mbintra_table = s->mbintra_table; |
| 1151 | |
| 1152 | for (i = 0; i < FF_ARRAY_ELEMS(s->dc_val); i++) |
| 1153 | er->dc_val[i] = s->dc_val[i]; |
| 1154 | |
| 1155 | er->decode_mb = mpeg_er_decode_mb; |
| 1156 | er->opaque = s; |
| 1157 | |
| 1158 | return 0; |
| 1159 | fail: |
| 1160 | av_freep(&er->er_temp_buffer); |
| 1161 | av_freep(&er->error_status_table); |
| 1162 | return AVERROR(ENOMEM); |
| 1163 | } |
| 1164 | |
| 1165 | /** |
| 1166 | * Initialize and allocates MpegEncContext fields dependent on the resolution. |
| 1167 | */ |
| 1168 | static int init_context_frame(MpegEncContext *s) |
| 1169 | { |
| 1170 | int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y; |
| 1171 | |
| 1172 | s->mb_width = (s->width + 15) / 16; |
| 1173 | s->mb_stride = s->mb_width + 1; |
| 1174 | s->b8_stride = s->mb_width * 2 + 1; |
| 1175 | mb_array_size = s->mb_height * s->mb_stride; |
| 1176 | mv_table_size = (s->mb_height + 2) * s->mb_stride + 1; |
| 1177 | |
| 1178 | /* set default edge pos, will be overridden |
| 1179 | * in decode_header if needed */ |
| 1180 | s->h_edge_pos = s->mb_width * 16; |
| 1181 | s->v_edge_pos = s->mb_height * 16; |
| 1182 | |
| 1183 | s->mb_num = s->mb_width * s->mb_height; |
| 1184 | |
| 1185 | s->block_wrap[0] = |
| 1186 | s->block_wrap[1] = |
| 1187 | s->block_wrap[2] = |
| 1188 | s->block_wrap[3] = s->b8_stride; |
| 1189 | s->block_wrap[4] = |
| 1190 | s->block_wrap[5] = s->mb_stride; |
| 1191 | |
| 1192 | y_size = s->b8_stride * (2 * s->mb_height + 1); |
| 1193 | c_size = s->mb_stride * (s->mb_height + 1); |
| 1194 | yc_size = y_size + 2 * c_size; |
| 1195 | |
| 1196 | if (s->mb_height & 1) |
| 1197 | yc_size += 2*s->b8_stride + 2*s->mb_stride; |
| 1198 | |
| 1199 | FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int), fail); // error ressilience code looks cleaner with this |
| 1200 | for (y = 0; y < s->mb_height; y++) |
| 1201 | for (x = 0; x < s->mb_width; x++) |
| 1202 | s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride; |
| 1203 | |
| 1204 | s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed? |
| 1205 | |
| 1206 | if (s->encoding) { |
| 1207 | /* Allocate MV tables */ |
| 1208 | FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail) |
| 1209 | FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail) |
| 1210 | FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail) |
| 1211 | FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail) |
| 1212 | FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail) |
| 1213 | FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail) |
| 1214 | s->p_mv_table = s->p_mv_table_base + s->mb_stride + 1; |
| 1215 | s->b_forw_mv_table = s->b_forw_mv_table_base + s->mb_stride + 1; |
| 1216 | s->b_back_mv_table = s->b_back_mv_table_base + s->mb_stride + 1; |
| 1217 | s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1; |
| 1218 | s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1; |
| 1219 | s->b_direct_mv_table = s->b_direct_mv_table_base + s->mb_stride + 1; |
| 1220 | |
| 1221 | /* Allocate MB type table */ |
| 1222 | FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size * sizeof(uint16_t), fail) // needed for encoding |
| 1223 | |
| 1224 | FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail) |
| 1225 | |
| 1226 | FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab, |
| 1227 | mb_array_size * sizeof(float), fail); |
| 1228 | FF_ALLOC_OR_GOTO(s->avctx, s->bits_tab, |
| 1229 | mb_array_size * sizeof(float), fail); |
| 1230 | |
| 1231 | } |
| 1232 | |
| 1233 | if (s->codec_id == AV_CODEC_ID_MPEG4 || |
| 1234 | (s->flags & CODEC_FLAG_INTERLACED_ME)) { |
| 1235 | /* interlaced direct mode decoding tables */ |
| 1236 | for (i = 0; i < 2; i++) { |
| 1237 | int j, k; |
| 1238 | for (j = 0; j < 2; j++) { |
| 1239 | for (k = 0; k < 2; k++) { |
| 1240 | FF_ALLOCZ_OR_GOTO(s->avctx, |
| 1241 | s->b_field_mv_table_base[i][j][k], |
| 1242 | mv_table_size * 2 * sizeof(int16_t), |
| 1243 | fail); |
| 1244 | s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] + |
| 1245 | s->mb_stride + 1; |
| 1246 | } |
| 1247 | FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail) |
| 1248 | FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail) |
| 1249 | s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1; |
| 1250 | } |
| 1251 | FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail) |
| 1252 | } |
| 1253 | } |
| 1254 | if (s->out_format == FMT_H263) { |
| 1255 | /* cbp values */ |
| 1256 | FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride, fail); |
| 1257 | s->coded_block = s->coded_block_base + s->b8_stride + 1; |
| 1258 | |
| 1259 | /* cbp, ac_pred, pred_dir */ |
| 1260 | FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table , mb_array_size * sizeof(uint8_t), fail); |
| 1261 | FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail); |
| 1262 | } |
| 1263 | |
| 1264 | if (s->h263_pred || s->h263_plus || !s->encoding) { |
| 1265 | /* dc values */ |
| 1266 | // MN: we need these for error resilience of intra-frames |
| 1267 | FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail); |
| 1268 | s->dc_val[0] = s->dc_val_base + s->b8_stride + 1; |
| 1269 | s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1; |
| 1270 | s->dc_val[2] = s->dc_val[1] + c_size; |
| 1271 | for (i = 0; i < yc_size; i++) |
| 1272 | s->dc_val_base[i] = 1024; |
| 1273 | } |
| 1274 | |
| 1275 | /* which mb is a intra block */ |
| 1276 | FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail); |
| 1277 | memset(s->mbintra_table, 1, mb_array_size); |
| 1278 | |
| 1279 | /* init macroblock skip table */ |
| 1280 | FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail); |
| 1281 | // Note the + 1 is for a quicker mpeg4 slice_end detection |
| 1282 | |
| 1283 | return init_er(s); |
| 1284 | fail: |
| 1285 | return AVERROR(ENOMEM); |
| 1286 | } |
| 1287 | |
| 1288 | /** |
| 1289 | * init common structure for both encoder and decoder. |
| 1290 | * this assumes that some variables like width/height are already set |
| 1291 | */ |
| 1292 | av_cold int ff_mpv_common_init(MpegEncContext *s) |
| 1293 | { |
| 1294 | int i; |
| 1295 | int nb_slices = (HAVE_THREADS && |
| 1296 | s->avctx->active_thread_type & FF_THREAD_SLICE) ? |
| 1297 | s->avctx->thread_count : 1; |
| 1298 | |
| 1299 | if (s->encoding && s->avctx->slices) |
| 1300 | nb_slices = s->avctx->slices; |
| 1301 | |
| 1302 | if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence) |
| 1303 | s->mb_height = (s->height + 31) / 32 * 2; |
| 1304 | else |
| 1305 | s->mb_height = (s->height + 15) / 16; |
| 1306 | |
| 1307 | if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) { |
| 1308 | av_log(s->avctx, AV_LOG_ERROR, |
| 1309 | "decoding to AV_PIX_FMT_NONE is not supported.\n"); |
| 1310 | return -1; |
| 1311 | } |
| 1312 | |
| 1313 | if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) { |
| 1314 | int max_slices; |
| 1315 | if (s->mb_height) |
| 1316 | max_slices = FFMIN(MAX_THREADS, s->mb_height); |
| 1317 | else |
| 1318 | max_slices = MAX_THREADS; |
| 1319 | av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d)," |
| 1320 | " reducing to %d\n", nb_slices, max_slices); |
| 1321 | nb_slices = max_slices; |
| 1322 | } |
| 1323 | |
| 1324 | if ((s->width || s->height) && |
| 1325 | av_image_check_size(s->width, s->height, 0, s->avctx)) |
| 1326 | return -1; |
| 1327 | |
| 1328 | dct_init(s); |
| 1329 | |
| 1330 | s->flags = s->avctx->flags; |
| 1331 | s->flags2 = s->avctx->flags2; |
| 1332 | |
| 1333 | /* set chroma shifts */ |
| 1334 | avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, |
| 1335 | &s->chroma_x_shift, |
| 1336 | &s->chroma_y_shift); |
| 1337 | |
| 1338 | |
| 1339 | FF_ALLOCZ_OR_GOTO(s->avctx, s->picture, |
| 1340 | MAX_PICTURE_COUNT * sizeof(Picture), fail); |
| 1341 | for (i = 0; i < MAX_PICTURE_COUNT; i++) { |
| 1342 | s->picture[i].f = av_frame_alloc(); |
| 1343 | if (!s->picture[i].f) |
| 1344 | goto fail; |
| 1345 | } |
| 1346 | memset(&s->next_picture, 0, sizeof(s->next_picture)); |
| 1347 | memset(&s->last_picture, 0, sizeof(s->last_picture)); |
| 1348 | memset(&s->current_picture, 0, sizeof(s->current_picture)); |
| 1349 | memset(&s->new_picture, 0, sizeof(s->new_picture)); |
| 1350 | s->next_picture.f = av_frame_alloc(); |
| 1351 | if (!s->next_picture.f) |
| 1352 | goto fail; |
| 1353 | s->last_picture.f = av_frame_alloc(); |
| 1354 | if (!s->last_picture.f) |
| 1355 | goto fail; |
| 1356 | s->current_picture.f = av_frame_alloc(); |
| 1357 | if (!s->current_picture.f) |
| 1358 | goto fail; |
| 1359 | s->new_picture.f = av_frame_alloc(); |
| 1360 | if (!s->new_picture.f) |
| 1361 | goto fail; |
| 1362 | |
| 1363 | if (init_context_frame(s)) |
| 1364 | goto fail; |
| 1365 | |
| 1366 | s->parse_context.state = -1; |
| 1367 | |
| 1368 | s->context_initialized = 1; |
| 1369 | s->thread_context[0] = s; |
| 1370 | |
| 1371 | // if (s->width && s->height) { |
| 1372 | if (nb_slices > 1) { |
| 1373 | for (i = 1; i < nb_slices; i++) { |
| 1374 | s->thread_context[i] = av_malloc(sizeof(MpegEncContext)); |
| 1375 | memcpy(s->thread_context[i], s, sizeof(MpegEncContext)); |
| 1376 | } |
| 1377 | |
| 1378 | for (i = 0; i < nb_slices; i++) { |
| 1379 | if (init_duplicate_context(s->thread_context[i]) < 0) |
| 1380 | goto fail; |
| 1381 | s->thread_context[i]->start_mb_y = |
| 1382 | (s->mb_height * (i) + nb_slices / 2) / nb_slices; |
| 1383 | s->thread_context[i]->end_mb_y = |
| 1384 | (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices; |
| 1385 | } |
| 1386 | } else { |
| 1387 | if (init_duplicate_context(s) < 0) |
| 1388 | goto fail; |
| 1389 | s->start_mb_y = 0; |
| 1390 | s->end_mb_y = s->mb_height; |
| 1391 | } |
| 1392 | s->slice_context_count = nb_slices; |
| 1393 | // } |
| 1394 | |
| 1395 | return 0; |
| 1396 | fail: |
| 1397 | ff_mpv_common_end(s); |
| 1398 | return -1; |
| 1399 | } |
| 1400 | |
| 1401 | /** |
| 1402 | * Frees and resets MpegEncContext fields depending on the resolution. |
| 1403 | * Is used during resolution changes to avoid a full reinitialization of the |
| 1404 | * codec. |
| 1405 | */ |
| 1406 | static void free_context_frame(MpegEncContext *s) |
| 1407 | { |
| 1408 | int i, j, k; |
| 1409 | |
| 1410 | av_freep(&s->mb_type); |
| 1411 | av_freep(&s->p_mv_table_base); |
| 1412 | av_freep(&s->b_forw_mv_table_base); |
| 1413 | av_freep(&s->b_back_mv_table_base); |
| 1414 | av_freep(&s->b_bidir_forw_mv_table_base); |
| 1415 | av_freep(&s->b_bidir_back_mv_table_base); |
| 1416 | av_freep(&s->b_direct_mv_table_base); |
| 1417 | s->p_mv_table = NULL; |
| 1418 | s->b_forw_mv_table = NULL; |
| 1419 | s->b_back_mv_table = NULL; |
| 1420 | s->b_bidir_forw_mv_table = NULL; |
| 1421 | s->b_bidir_back_mv_table = NULL; |
| 1422 | s->b_direct_mv_table = NULL; |
| 1423 | for (i = 0; i < 2; i++) { |
| 1424 | for (j = 0; j < 2; j++) { |
| 1425 | for (k = 0; k < 2; k++) { |
| 1426 | av_freep(&s->b_field_mv_table_base[i][j][k]); |
| 1427 | s->b_field_mv_table[i][j][k] = NULL; |
| 1428 | } |
| 1429 | av_freep(&s->b_field_select_table[i][j]); |
| 1430 | av_freep(&s->p_field_mv_table_base[i][j]); |
| 1431 | s->p_field_mv_table[i][j] = NULL; |
| 1432 | } |
| 1433 | av_freep(&s->p_field_select_table[i]); |
| 1434 | } |
| 1435 | |
| 1436 | av_freep(&s->dc_val_base); |
| 1437 | av_freep(&s->coded_block_base); |
| 1438 | av_freep(&s->mbintra_table); |
| 1439 | av_freep(&s->cbp_table); |
| 1440 | av_freep(&s->pred_dir_table); |
| 1441 | |
| 1442 | av_freep(&s->mbskip_table); |
| 1443 | |
| 1444 | av_freep(&s->er.error_status_table); |
| 1445 | av_freep(&s->er.er_temp_buffer); |
| 1446 | av_freep(&s->mb_index2xy); |
| 1447 | av_freep(&s->lambda_table); |
| 1448 | |
| 1449 | av_freep(&s->cplx_tab); |
| 1450 | av_freep(&s->bits_tab); |
| 1451 | |
| 1452 | s->linesize = s->uvlinesize = 0; |
| 1453 | } |
| 1454 | |
| 1455 | int ff_mpv_common_frame_size_change(MpegEncContext *s) |
| 1456 | { |
| 1457 | int i, err = 0; |
| 1458 | |
| 1459 | if (!s->context_initialized) |
| 1460 | return AVERROR(EINVAL); |
| 1461 | |
| 1462 | if (s->slice_context_count > 1) { |
| 1463 | for (i = 0; i < s->slice_context_count; i++) { |
| 1464 | free_duplicate_context(s->thread_context[i]); |
| 1465 | } |
| 1466 | for (i = 1; i < s->slice_context_count; i++) { |
| 1467 | av_freep(&s->thread_context[i]); |
| 1468 | } |
| 1469 | } else |
| 1470 | free_duplicate_context(s); |
| 1471 | |
| 1472 | free_context_frame(s); |
| 1473 | |
| 1474 | if (s->picture) |
| 1475 | for (i = 0; i < MAX_PICTURE_COUNT; i++) { |
| 1476 | s->picture[i].needs_realloc = 1; |
| 1477 | } |
| 1478 | |
| 1479 | s->last_picture_ptr = |
| 1480 | s->next_picture_ptr = |
| 1481 | s->current_picture_ptr = NULL; |
| 1482 | |
| 1483 | // init |
| 1484 | if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence) |
| 1485 | s->mb_height = (s->height + 31) / 32 * 2; |
| 1486 | else |
| 1487 | s->mb_height = (s->height + 15) / 16; |
| 1488 | |
| 1489 | if ((s->width || s->height) && |
| 1490 | (err = av_image_check_size(s->width, s->height, 0, s->avctx)) < 0) |
| 1491 | goto fail; |
| 1492 | |
| 1493 | if ((err = init_context_frame(s))) |
| 1494 | goto fail; |
| 1495 | |
| 1496 | s->thread_context[0] = s; |
| 1497 | |
| 1498 | if (s->width && s->height) { |
| 1499 | int nb_slices = s->slice_context_count; |
| 1500 | if (nb_slices > 1) { |
| 1501 | for (i = 1; i < nb_slices; i++) { |
| 1502 | s->thread_context[i] = av_malloc(sizeof(MpegEncContext)); |
| 1503 | memcpy(s->thread_context[i], s, sizeof(MpegEncContext)); |
| 1504 | } |
| 1505 | |
| 1506 | for (i = 0; i < nb_slices; i++) { |
| 1507 | if ((err = init_duplicate_context(s->thread_context[i])) < 0) |
| 1508 | goto fail; |
| 1509 | s->thread_context[i]->start_mb_y = |
| 1510 | (s->mb_height * (i) + nb_slices / 2) / nb_slices; |
| 1511 | s->thread_context[i]->end_mb_y = |
| 1512 | (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices; |
| 1513 | } |
| 1514 | } else { |
| 1515 | err = init_duplicate_context(s); |
| 1516 | if (err < 0) |
| 1517 | goto fail; |
| 1518 | s->start_mb_y = 0; |
| 1519 | s->end_mb_y = s->mb_height; |
| 1520 | } |
| 1521 | s->slice_context_count = nb_slices; |
| 1522 | } |
| 1523 | |
| 1524 | return 0; |
| 1525 | fail: |
| 1526 | ff_mpv_common_end(s); |
| 1527 | return err; |
| 1528 | } |
| 1529 | |
| 1530 | /* init common structure for both encoder and decoder */ |
| 1531 | void ff_mpv_common_end(MpegEncContext *s) |
| 1532 | { |
| 1533 | int i; |
| 1534 | |
| 1535 | if (s->slice_context_count > 1) { |
| 1536 | for (i = 0; i < s->slice_context_count; i++) { |
| 1537 | free_duplicate_context(s->thread_context[i]); |
| 1538 | } |
| 1539 | for (i = 1; i < s->slice_context_count; i++) { |
| 1540 | av_freep(&s->thread_context[i]); |
| 1541 | } |
| 1542 | s->slice_context_count = 1; |
| 1543 | } else free_duplicate_context(s); |
| 1544 | |
| 1545 | av_freep(&s->parse_context.buffer); |
| 1546 | s->parse_context.buffer_size = 0; |
| 1547 | |
| 1548 | av_freep(&s->bitstream_buffer); |
| 1549 | s->allocated_bitstream_buffer_size = 0; |
| 1550 | |
| 1551 | if (s->picture) { |
| 1552 | for (i = 0; i < MAX_PICTURE_COUNT; i++) { |
| 1553 | ff_free_picture_tables(&s->picture[i]); |
| 1554 | ff_mpeg_unref_picture(s, &s->picture[i]); |
| 1555 | av_frame_free(&s->picture[i].f); |
| 1556 | } |
| 1557 | } |
| 1558 | av_freep(&s->picture); |
| 1559 | ff_free_picture_tables(&s->last_picture); |
| 1560 | ff_mpeg_unref_picture(s, &s->last_picture); |
| 1561 | av_frame_free(&s->last_picture.f); |
| 1562 | ff_free_picture_tables(&s->current_picture); |
| 1563 | ff_mpeg_unref_picture(s, &s->current_picture); |
| 1564 | av_frame_free(&s->current_picture.f); |
| 1565 | ff_free_picture_tables(&s->next_picture); |
| 1566 | ff_mpeg_unref_picture(s, &s->next_picture); |
| 1567 | av_frame_free(&s->next_picture.f); |
| 1568 | ff_free_picture_tables(&s->new_picture); |
| 1569 | ff_mpeg_unref_picture(s, &s->new_picture); |
| 1570 | av_frame_free(&s->new_picture.f); |
| 1571 | |
| 1572 | free_context_frame(s); |
| 1573 | |
| 1574 | s->context_initialized = 0; |
| 1575 | s->last_picture_ptr = |
| 1576 | s->next_picture_ptr = |
| 1577 | s->current_picture_ptr = NULL; |
| 1578 | s->linesize = s->uvlinesize = 0; |
| 1579 | } |
| 1580 | |
| 1581 | av_cold void ff_init_rl(RLTable *rl, |
| 1582 | uint8_t static_store[2][2 * MAX_RUN + MAX_LEVEL + 3]) |
| 1583 | { |
| 1584 | int8_t max_level[MAX_RUN + 1], max_run[MAX_LEVEL + 1]; |
| 1585 | uint8_t index_run[MAX_RUN + 1]; |
| 1586 | int last, run, level, start, end, i; |
| 1587 | |
| 1588 | /* If table is static, we can quit if rl->max_level[0] is not NULL */ |
| 1589 | if (static_store && rl->max_level[0]) |
| 1590 | return; |
| 1591 | |
| 1592 | /* compute max_level[], max_run[] and index_run[] */ |
| 1593 | for (last = 0; last < 2; last++) { |
| 1594 | if (last == 0) { |
| 1595 | start = 0; |
| 1596 | end = rl->last; |
| 1597 | } else { |
| 1598 | start = rl->last; |
| 1599 | end = rl->n; |
| 1600 | } |
| 1601 | |
| 1602 | memset(max_level, 0, MAX_RUN + 1); |
| 1603 | memset(max_run, 0, MAX_LEVEL + 1); |
| 1604 | memset(index_run, rl->n, MAX_RUN + 1); |
| 1605 | for (i = start; i < end; i++) { |
| 1606 | run = rl->table_run[i]; |
| 1607 | level = rl->table_level[i]; |
| 1608 | if (index_run[run] == rl->n) |
| 1609 | index_run[run] = i; |
| 1610 | if (level > max_level[run]) |
| 1611 | max_level[run] = level; |
| 1612 | if (run > max_run[level]) |
| 1613 | max_run[level] = run; |
| 1614 | } |
| 1615 | if (static_store) |
| 1616 | rl->max_level[last] = static_store[last]; |
| 1617 | else |
| 1618 | rl->max_level[last] = av_malloc(MAX_RUN + 1); |
| 1619 | memcpy(rl->max_level[last], max_level, MAX_RUN + 1); |
| 1620 | if (static_store) |
| 1621 | rl->max_run[last] = static_store[last] + MAX_RUN + 1; |
| 1622 | else |
| 1623 | rl->max_run[last] = av_malloc(MAX_LEVEL + 1); |
| 1624 | memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1); |
| 1625 | if (static_store) |
| 1626 | rl->index_run[last] = static_store[last] + MAX_RUN + MAX_LEVEL + 2; |
| 1627 | else |
| 1628 | rl->index_run[last] = av_malloc(MAX_RUN + 1); |
| 1629 | memcpy(rl->index_run[last], index_run, MAX_RUN + 1); |
| 1630 | } |
| 1631 | } |
| 1632 | |
| 1633 | av_cold void ff_init_vlc_rl(RLTable *rl, unsigned static_size) |
| 1634 | { |
| 1635 | int i, q; |
| 1636 | VLC_TYPE table[1500][2] = {{0}}; |
| 1637 | VLC vlc = { .table = table, .table_allocated = static_size }; |
| 1638 | av_assert0(static_size <= FF_ARRAY_ELEMS(table)); |
| 1639 | init_vlc(&vlc, 9, rl->n + 1, &rl->table_vlc[0][1], 4, 2, &rl->table_vlc[0][0], 4, 2, INIT_VLC_USE_NEW_STATIC); |
| 1640 | |
| 1641 | for (q = 0; q < 32; q++) { |
| 1642 | int qmul = q * 2; |
| 1643 | int qadd = (q - 1) | 1; |
| 1644 | |
| 1645 | if (q == 0) { |
| 1646 | qmul = 1; |
| 1647 | qadd = 0; |
| 1648 | } |
| 1649 | for (i = 0; i < vlc.table_size; i++) { |
| 1650 | int code = vlc.table[i][0]; |
| 1651 | int len = vlc.table[i][1]; |
| 1652 | int level, run; |
| 1653 | |
| 1654 | if (len == 0) { // illegal code |
| 1655 | run = 66; |
| 1656 | level = MAX_LEVEL; |
| 1657 | } else if (len < 0) { // more bits needed |
| 1658 | run = 0; |
| 1659 | level = code; |
| 1660 | } else { |
| 1661 | if (code == rl->n) { // esc |
| 1662 | run = 66; |
| 1663 | level = 0; |
| 1664 | } else { |
| 1665 | run = rl->table_run[code] + 1; |
| 1666 | level = rl->table_level[code] * qmul + qadd; |
| 1667 | if (code >= rl->last) run += 192; |
| 1668 | } |
| 1669 | } |
| 1670 | rl->rl_vlc[q][i].len = len; |
| 1671 | rl->rl_vlc[q][i].level = level; |
| 1672 | rl->rl_vlc[q][i].run = run; |
| 1673 | } |
| 1674 | } |
| 1675 | } |
| 1676 | |
| 1677 | static void release_unused_pictures(MpegEncContext *s) |
| 1678 | { |
| 1679 | int i; |
| 1680 | |
| 1681 | /* release non reference frames */ |
| 1682 | for (i = 0; i < MAX_PICTURE_COUNT; i++) { |
| 1683 | if (!s->picture[i].reference) |
| 1684 | ff_mpeg_unref_picture(s, &s->picture[i]); |
| 1685 | } |
| 1686 | } |
| 1687 | |
| 1688 | static inline int pic_is_unused(MpegEncContext *s, Picture *pic) |
| 1689 | { |
| 1690 | if (pic == s->last_picture_ptr) |
| 1691 | return 0; |
| 1692 | if (!pic->f->buf[0]) |
| 1693 | return 1; |
| 1694 | if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF)) |
| 1695 | return 1; |
| 1696 | return 0; |
| 1697 | } |
| 1698 | |
| 1699 | static int find_unused_picture(MpegEncContext *s, int shared) |
| 1700 | { |
| 1701 | int i; |
| 1702 | |
| 1703 | if (shared) { |
| 1704 | for (i = 0; i < MAX_PICTURE_COUNT; i++) { |
| 1705 | if (!s->picture[i].f->buf[0] && &s->picture[i] != s->last_picture_ptr) |
| 1706 | return i; |
| 1707 | } |
| 1708 | } else { |
| 1709 | for (i = 0; i < MAX_PICTURE_COUNT; i++) { |
| 1710 | if (pic_is_unused(s, &s->picture[i])) |
| 1711 | return i; |
| 1712 | } |
| 1713 | } |
| 1714 | |
| 1715 | av_log(s->avctx, AV_LOG_FATAL, |
| 1716 | "Internal error, picture buffer overflow\n"); |
| 1717 | /* We could return -1, but the codec would crash trying to draw into a |
| 1718 | * non-existing frame anyway. This is safer than waiting for a random crash. |
| 1719 | * Also the return of this is never useful, an encoder must only allocate |
| 1720 | * as much as allowed in the specification. This has no relationship to how |
| 1721 | * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large |
| 1722 | * enough for such valid streams). |
| 1723 | * Plus, a decoder has to check stream validity and remove frames if too |
| 1724 | * many reference frames are around. Waiting for "OOM" is not correct at |
| 1725 | * all. Similarly, missing reference frames have to be replaced by |
| 1726 | * interpolated/MC frames, anything else is a bug in the codec ... |
| 1727 | */ |
| 1728 | abort(); |
| 1729 | return -1; |
| 1730 | } |
| 1731 | |
| 1732 | int ff_find_unused_picture(MpegEncContext *s, int shared) |
| 1733 | { |
| 1734 | int ret = find_unused_picture(s, shared); |
| 1735 | |
| 1736 | if (ret >= 0 && ret < MAX_PICTURE_COUNT) { |
| 1737 | if (s->picture[ret].needs_realloc) { |
| 1738 | s->picture[ret].needs_realloc = 0; |
| 1739 | ff_free_picture_tables(&s->picture[ret]); |
| 1740 | ff_mpeg_unref_picture(s, &s->picture[ret]); |
| 1741 | } |
| 1742 | } |
| 1743 | return ret; |
| 1744 | } |
| 1745 | |
| 1746 | static void gray_frame(AVFrame *frame) |
| 1747 | { |
| 1748 | int i, h_chroma_shift, v_chroma_shift; |
| 1749 | |
| 1750 | av_pix_fmt_get_chroma_sub_sample(frame->format, &h_chroma_shift, &v_chroma_shift); |
| 1751 | |
| 1752 | for(i=0; i<frame->height; i++) |
| 1753 | memset(frame->data[0] + frame->linesize[0]*i, 0x80, frame->width); |
| 1754 | for(i=0; i<FF_CEIL_RSHIFT(frame->height, v_chroma_shift); i++) { |
| 1755 | memset(frame->data[1] + frame->linesize[1]*i, |
| 1756 | 0x80, FF_CEIL_RSHIFT(frame->width, h_chroma_shift)); |
| 1757 | memset(frame->data[2] + frame->linesize[2]*i, |
| 1758 | 0x80, FF_CEIL_RSHIFT(frame->width, h_chroma_shift)); |
| 1759 | } |
| 1760 | } |
| 1761 | |
| 1762 | /** |
| 1763 | * generic function called after decoding |
| 1764 | * the header and before a frame is decoded. |
| 1765 | */ |
| 1766 | int ff_mpv_frame_start(MpegEncContext *s, AVCodecContext *avctx) |
| 1767 | { |
| 1768 | int i, ret; |
| 1769 | Picture *pic; |
| 1770 | s->mb_skipped = 0; |
| 1771 | |
| 1772 | if (!ff_thread_can_start_frame(avctx)) { |
| 1773 | av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n"); |
| 1774 | return -1; |
| 1775 | } |
| 1776 | |
| 1777 | /* mark & release old frames */ |
| 1778 | if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr && |
| 1779 | s->last_picture_ptr != s->next_picture_ptr && |
| 1780 | s->last_picture_ptr->f->buf[0]) { |
| 1781 | ff_mpeg_unref_picture(s, s->last_picture_ptr); |
| 1782 | } |
| 1783 | |
| 1784 | /* release forgotten pictures */ |
| 1785 | /* if (mpeg124/h263) */ |
| 1786 | for (i = 0; i < MAX_PICTURE_COUNT; i++) { |
| 1787 | if (&s->picture[i] != s->last_picture_ptr && |
| 1788 | &s->picture[i] != s->next_picture_ptr && |
| 1789 | s->picture[i].reference && !s->picture[i].needs_realloc) { |
| 1790 | if (!(avctx->active_thread_type & FF_THREAD_FRAME)) |
| 1791 | av_log(avctx, AV_LOG_ERROR, |
| 1792 | "releasing zombie picture\n"); |
| 1793 | ff_mpeg_unref_picture(s, &s->picture[i]); |
| 1794 | } |
| 1795 | } |
| 1796 | |
| 1797 | ff_mpeg_unref_picture(s, &s->current_picture); |
| 1798 | |
| 1799 | release_unused_pictures(s); |
| 1800 | |
| 1801 | if (s->current_picture_ptr && !s->current_picture_ptr->f->buf[0]) { |
| 1802 | // we already have a unused image |
| 1803 | // (maybe it was set before reading the header) |
| 1804 | pic = s->current_picture_ptr; |
| 1805 | } else { |
| 1806 | i = ff_find_unused_picture(s, 0); |
| 1807 | if (i < 0) { |
| 1808 | av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n"); |
| 1809 | return i; |
| 1810 | } |
| 1811 | pic = &s->picture[i]; |
| 1812 | } |
| 1813 | |
| 1814 | pic->reference = 0; |
| 1815 | if (!s->droppable) { |
| 1816 | if (s->pict_type != AV_PICTURE_TYPE_B) |
| 1817 | pic->reference = 3; |
| 1818 | } |
| 1819 | |
| 1820 | pic->f->coded_picture_number = s->coded_picture_number++; |
| 1821 | |
| 1822 | if (ff_alloc_picture(s, pic, 0) < 0) |
| 1823 | return -1; |
| 1824 | |
| 1825 | s->current_picture_ptr = pic; |
| 1826 | // FIXME use only the vars from current_pic |
| 1827 | s->current_picture_ptr->f->top_field_first = s->top_field_first; |
| 1828 | if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO || |
| 1829 | s->codec_id == AV_CODEC_ID_MPEG2VIDEO) { |
| 1830 | if (s->picture_structure != PICT_FRAME) |
| 1831 | s->current_picture_ptr->f->top_field_first = |
| 1832 | (s->picture_structure == PICT_TOP_FIELD) == s->first_field; |
| 1833 | } |
| 1834 | s->current_picture_ptr->f->interlaced_frame = !s->progressive_frame && |
| 1835 | !s->progressive_sequence; |
| 1836 | s->current_picture_ptr->field_picture = s->picture_structure != PICT_FRAME; |
| 1837 | |
| 1838 | s->current_picture_ptr->f->pict_type = s->pict_type; |
| 1839 | // if (s->flags && CODEC_FLAG_QSCALE) |
| 1840 | // s->current_picture_ptr->quality = s->new_picture_ptr->quality; |
| 1841 | s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I; |
| 1842 | |
| 1843 | if ((ret = ff_mpeg_ref_picture(s, &s->current_picture, |
| 1844 | s->current_picture_ptr)) < 0) |
| 1845 | return ret; |
| 1846 | |
| 1847 | if (s->pict_type != AV_PICTURE_TYPE_B) { |
| 1848 | s->last_picture_ptr = s->next_picture_ptr; |
| 1849 | if (!s->droppable) |
| 1850 | s->next_picture_ptr = s->current_picture_ptr; |
| 1851 | } |
| 1852 | av_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n", |
| 1853 | s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr, |
| 1854 | s->last_picture_ptr ? s->last_picture_ptr->f->data[0] : NULL, |
| 1855 | s->next_picture_ptr ? s->next_picture_ptr->f->data[0] : NULL, |
| 1856 | s->current_picture_ptr ? s->current_picture_ptr->f->data[0] : NULL, |
| 1857 | s->pict_type, s->droppable); |
| 1858 | |
| 1859 | if ((!s->last_picture_ptr || !s->last_picture_ptr->f->buf[0]) && |
| 1860 | (s->pict_type != AV_PICTURE_TYPE_I || |
| 1861 | s->picture_structure != PICT_FRAME)) { |
| 1862 | int h_chroma_shift, v_chroma_shift; |
| 1863 | av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt, |
| 1864 | &h_chroma_shift, &v_chroma_shift); |
| 1865 | if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0]) |
| 1866 | av_log(avctx, AV_LOG_DEBUG, |
| 1867 | "allocating dummy last picture for B frame\n"); |
| 1868 | else if (s->pict_type != AV_PICTURE_TYPE_I) |
| 1869 | av_log(avctx, AV_LOG_ERROR, |
| 1870 | "warning: first frame is no keyframe\n"); |
| 1871 | else if (s->picture_structure != PICT_FRAME) |
| 1872 | av_log(avctx, AV_LOG_DEBUG, |
| 1873 | "allocate dummy last picture for field based first keyframe\n"); |
| 1874 | |
| 1875 | /* Allocate a dummy frame */ |
| 1876 | i = ff_find_unused_picture(s, 0); |
| 1877 | if (i < 0) { |
| 1878 | av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n"); |
| 1879 | return i; |
| 1880 | } |
| 1881 | s->last_picture_ptr = &s->picture[i]; |
| 1882 | |
| 1883 | s->last_picture_ptr->reference = 3; |
| 1884 | s->last_picture_ptr->f->key_frame = 0; |
| 1885 | s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P; |
| 1886 | |
| 1887 | if (ff_alloc_picture(s, s->last_picture_ptr, 0) < 0) { |
| 1888 | s->last_picture_ptr = NULL; |
| 1889 | return -1; |
| 1890 | } |
| 1891 | |
| 1892 | if (!avctx->hwaccel && !(avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)) { |
| 1893 | for(i=0; i<avctx->height; i++) |
| 1894 | memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, |
| 1895 | 0x80, avctx->width); |
| 1896 | for(i=0; i<FF_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) { |
| 1897 | memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i, |
| 1898 | 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift)); |
| 1899 | memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i, |
| 1900 | 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift)); |
| 1901 | } |
| 1902 | |
| 1903 | if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){ |
| 1904 | for(i=0; i<avctx->height; i++) |
| 1905 | memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width); |
| 1906 | } |
| 1907 | } |
| 1908 | |
| 1909 | ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0); |
| 1910 | ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1); |
| 1911 | } |
| 1912 | if ((!s->next_picture_ptr || !s->next_picture_ptr->f->buf[0]) && |
| 1913 | s->pict_type == AV_PICTURE_TYPE_B) { |
| 1914 | /* Allocate a dummy frame */ |
| 1915 | i = ff_find_unused_picture(s, 0); |
| 1916 | if (i < 0) { |
| 1917 | av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n"); |
| 1918 | return i; |
| 1919 | } |
| 1920 | s->next_picture_ptr = &s->picture[i]; |
| 1921 | |
| 1922 | s->next_picture_ptr->reference = 3; |
| 1923 | s->next_picture_ptr->f->key_frame = 0; |
| 1924 | s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P; |
| 1925 | |
| 1926 | if (ff_alloc_picture(s, s->next_picture_ptr, 0) < 0) { |
| 1927 | s->next_picture_ptr = NULL; |
| 1928 | return -1; |
| 1929 | } |
| 1930 | ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0); |
| 1931 | ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1); |
| 1932 | } |
| 1933 | |
| 1934 | #if 0 // BUFREF-FIXME |
| 1935 | memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data)); |
| 1936 | memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data)); |
| 1937 | #endif |
| 1938 | if (s->last_picture_ptr) { |
| 1939 | ff_mpeg_unref_picture(s, &s->last_picture); |
| 1940 | if (s->last_picture_ptr->f->buf[0] && |
| 1941 | (ret = ff_mpeg_ref_picture(s, &s->last_picture, |
| 1942 | s->last_picture_ptr)) < 0) |
| 1943 | return ret; |
| 1944 | } |
| 1945 | if (s->next_picture_ptr) { |
| 1946 | ff_mpeg_unref_picture(s, &s->next_picture); |
| 1947 | if (s->next_picture_ptr->f->buf[0] && |
| 1948 | (ret = ff_mpeg_ref_picture(s, &s->next_picture, |
| 1949 | s->next_picture_ptr)) < 0) |
| 1950 | return ret; |
| 1951 | } |
| 1952 | |
| 1953 | av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr && |
| 1954 | s->last_picture_ptr->f->buf[0])); |
| 1955 | |
| 1956 | if (s->picture_structure!= PICT_FRAME) { |
| 1957 | int i; |
| 1958 | for (i = 0; i < 4; i++) { |
| 1959 | if (s->picture_structure == PICT_BOTTOM_FIELD) { |
| 1960 | s->current_picture.f->data[i] += |
| 1961 | s->current_picture.f->linesize[i]; |
| 1962 | } |
| 1963 | s->current_picture.f->linesize[i] *= 2; |
| 1964 | s->last_picture.f->linesize[i] *= 2; |
| 1965 | s->next_picture.f->linesize[i] *= 2; |
| 1966 | } |
| 1967 | } |
| 1968 | |
| 1969 | s->err_recognition = avctx->err_recognition; |
| 1970 | |
| 1971 | /* set dequantizer, we can't do it during init as |
| 1972 | * it might change for mpeg4 and we can't do it in the header |
| 1973 | * decode as init is not called for mpeg4 there yet */ |
| 1974 | if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) { |
| 1975 | s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra; |
| 1976 | s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter; |
| 1977 | } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) { |
| 1978 | s->dct_unquantize_intra = s->dct_unquantize_h263_intra; |
| 1979 | s->dct_unquantize_inter = s->dct_unquantize_h263_inter; |
| 1980 | } else { |
| 1981 | s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra; |
| 1982 | s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter; |
| 1983 | } |
| 1984 | |
| 1985 | if (s->avctx->debug & FF_DEBUG_NOMC) { |
| 1986 | gray_frame(s->current_picture_ptr->f); |
| 1987 | } |
| 1988 | |
| 1989 | return 0; |
| 1990 | } |
| 1991 | |
| 1992 | /* called after a frame has been decoded. */ |
| 1993 | void ff_mpv_frame_end(MpegEncContext *s) |
| 1994 | { |
| 1995 | emms_c(); |
| 1996 | |
| 1997 | if (s->current_picture.reference) |
| 1998 | ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0); |
| 1999 | } |
| 2000 | |
| 2001 | |
| 2002 | #if FF_API_VISMV |
| 2003 | static int clip_line(int *sx, int *sy, int *ex, int *ey, int maxx) |
| 2004 | { |
| 2005 | if(*sx > *ex) |
| 2006 | return clip_line(ex, ey, sx, sy, maxx); |
| 2007 | |
| 2008 | if (*sx < 0) { |
| 2009 | if (*ex < 0) |
| 2010 | return 1; |
| 2011 | *sy = *ey + (*sy - *ey) * (int64_t)*ex / (*ex - *sx); |
| 2012 | *sx = 0; |
| 2013 | } |
| 2014 | |
| 2015 | if (*ex > maxx) { |
| 2016 | if (*sx > maxx) |
| 2017 | return 1; |
| 2018 | *ey = *sy + (*ey - *sy) * (int64_t)(maxx - *sx) / (*ex - *sx); |
| 2019 | *ex = maxx; |
| 2020 | } |
| 2021 | return 0; |
| 2022 | } |
| 2023 | |
| 2024 | |
| 2025 | /** |
| 2026 | * Draw a line from (ex, ey) -> (sx, sy). |
| 2027 | * @param w width of the image |
| 2028 | * @param h height of the image |
| 2029 | * @param stride stride/linesize of the image |
| 2030 | * @param color color of the arrow |
| 2031 | */ |
| 2032 | static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, |
| 2033 | int w, int h, int stride, int color) |
| 2034 | { |
| 2035 | int x, y, fr, f; |
| 2036 | |
| 2037 | if (clip_line(&sx, &sy, &ex, &ey, w - 1)) |
| 2038 | return; |
| 2039 | if (clip_line(&sy, &sx, &ey, &ex, h - 1)) |
| 2040 | return; |
| 2041 | |
| 2042 | sx = av_clip(sx, 0, w - 1); |
| 2043 | sy = av_clip(sy, 0, h - 1); |
| 2044 | ex = av_clip(ex, 0, w - 1); |
| 2045 | ey = av_clip(ey, 0, h - 1); |
| 2046 | |
| 2047 | buf[sy * stride + sx] += color; |
| 2048 | |
| 2049 | if (FFABS(ex - sx) > FFABS(ey - sy)) { |
| 2050 | if (sx > ex) { |
| 2051 | FFSWAP(int, sx, ex); |
| 2052 | FFSWAP(int, sy, ey); |
| 2053 | } |
| 2054 | buf += sx + sy * stride; |
| 2055 | ex -= sx; |
| 2056 | f = ((ey - sy) << 16) / ex; |
| 2057 | for (x = 0; x <= ex; x++) { |
| 2058 | y = (x * f) >> 16; |
| 2059 | fr = (x * f) & 0xFFFF; |
| 2060 | buf[y * stride + x] += (color * (0x10000 - fr)) >> 16; |
| 2061 | if(fr) buf[(y + 1) * stride + x] += (color * fr ) >> 16; |
| 2062 | } |
| 2063 | } else { |
| 2064 | if (sy > ey) { |
| 2065 | FFSWAP(int, sx, ex); |
| 2066 | FFSWAP(int, sy, ey); |
| 2067 | } |
| 2068 | buf += sx + sy * stride; |
| 2069 | ey -= sy; |
| 2070 | if (ey) |
| 2071 | f = ((ex - sx) << 16) / ey; |
| 2072 | else |
| 2073 | f = 0; |
| 2074 | for(y= 0; y <= ey; y++){ |
| 2075 | x = (y*f) >> 16; |
| 2076 | fr = (y*f) & 0xFFFF; |
| 2077 | buf[y * stride + x] += (color * (0x10000 - fr)) >> 16; |
| 2078 | if(fr) buf[y * stride + x + 1] += (color * fr ) >> 16; |
| 2079 | } |
| 2080 | } |
| 2081 | } |
| 2082 | |
| 2083 | /** |
| 2084 | * Draw an arrow from (ex, ey) -> (sx, sy). |
| 2085 | * @param w width of the image |
| 2086 | * @param h height of the image |
| 2087 | * @param stride stride/linesize of the image |
| 2088 | * @param color color of the arrow |
| 2089 | */ |
| 2090 | static void draw_arrow(uint8_t *buf, int sx, int sy, int ex, |
| 2091 | int ey, int w, int h, int stride, int color, int tail, int direction) |
| 2092 | { |
| 2093 | int dx,dy; |
| 2094 | |
| 2095 | if (direction) { |
| 2096 | FFSWAP(int, sx, ex); |
| 2097 | FFSWAP(int, sy, ey); |
| 2098 | } |
| 2099 | |
| 2100 | sx = av_clip(sx, -100, w + 100); |
| 2101 | sy = av_clip(sy, -100, h + 100); |
| 2102 | ex = av_clip(ex, -100, w + 100); |
| 2103 | ey = av_clip(ey, -100, h + 100); |
| 2104 | |
| 2105 | dx = ex - sx; |
| 2106 | dy = ey - sy; |
| 2107 | |
| 2108 | if (dx * dx + dy * dy > 3 * 3) { |
| 2109 | int rx = dx + dy; |
| 2110 | int ry = -dx + dy; |
| 2111 | int length = ff_sqrt((rx * rx + ry * ry) << 8); |
| 2112 | |
| 2113 | // FIXME subpixel accuracy |
| 2114 | rx = ROUNDED_DIV(rx * 3 << 4, length); |
| 2115 | ry = ROUNDED_DIV(ry * 3 << 4, length); |
| 2116 | |
| 2117 | if (tail) { |
| 2118 | rx = -rx; |
| 2119 | ry = -ry; |
| 2120 | } |
| 2121 | |
| 2122 | draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color); |
| 2123 | draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color); |
| 2124 | } |
| 2125 | draw_line(buf, sx, sy, ex, ey, w, h, stride, color); |
| 2126 | } |
| 2127 | #endif |
| 2128 | |
| 2129 | static int add_mb(AVMotionVector *mb, uint32_t mb_type, |
| 2130 | int dst_x, int dst_y, |
| 2131 | int src_x, int src_y, |
| 2132 | int direction) |
| 2133 | { |
| 2134 | if (dst_x == src_x && dst_y == src_y) |
| 2135 | return 0; |
| 2136 | mb->w = IS_8X8(mb_type) || IS_8X16(mb_type) ? 8 : 16; |
| 2137 | mb->h = IS_8X8(mb_type) || IS_16X8(mb_type) ? 8 : 16; |
| 2138 | mb->src_x = src_x; |
| 2139 | mb->src_y = src_y; |
| 2140 | mb->dst_x = dst_x; |
| 2141 | mb->dst_y = dst_y; |
| 2142 | mb->source = direction ? 1 : -1; |
| 2143 | mb->flags = 0; // XXX: does mb_type contain extra information that could be exported here? |
| 2144 | return 1; |
| 2145 | } |
| 2146 | |
| 2147 | /** |
| 2148 | * Print debugging info for the given picture. |
| 2149 | */ |
| 2150 | void ff_print_debug_info2(AVCodecContext *avctx, AVFrame *pict, uint8_t *mbskip_table, |
| 2151 | uint32_t *mbtype_table, int8_t *qscale_table, int16_t (*motion_val[2])[2], |
| 2152 | int *low_delay, |
| 2153 | int mb_width, int mb_height, int mb_stride, int quarter_sample) |
| 2154 | { |
| 2155 | if ((avctx->flags2 & CODEC_FLAG2_EXPORT_MVS) && mbtype_table && motion_val[0]) { |
| 2156 | const int shift = 1 + quarter_sample; |
| 2157 | const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1; |
| 2158 | const int mv_stride = (mb_width << mv_sample_log2) + |
| 2159 | (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1); |
| 2160 | int mb_x, mb_y, mbcount = 0; |
| 2161 | |
| 2162 | /* size is width * height * 2 * 4 where 2 is for directions and 4 is |
| 2163 | * for the maximum number of MB (4 MB in case of IS_8x8) */ |
| 2164 | AVMotionVector *mvs = av_malloc_array(mb_width * mb_height, 2 * 4 * sizeof(AVMotionVector)); |
| 2165 | if (!mvs) |
| 2166 | return; |
| 2167 | |
| 2168 | for (mb_y = 0; mb_y < mb_height; mb_y++) { |
| 2169 | for (mb_x = 0; mb_x < mb_width; mb_x++) { |
| 2170 | int i, direction, mb_type = mbtype_table[mb_x + mb_y * mb_stride]; |
| 2171 | for (direction = 0; direction < 2; direction++) { |
| 2172 | if (!USES_LIST(mb_type, direction)) |
| 2173 | continue; |
| 2174 | if (IS_8X8(mb_type)) { |
| 2175 | for (i = 0; i < 4; i++) { |
| 2176 | int sx = mb_x * 16 + 4 + 8 * (i & 1); |
| 2177 | int sy = mb_y * 16 + 4 + 8 * (i >> 1); |
| 2178 | int xy = (mb_x * 2 + (i & 1) + |
| 2179 | (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1); |
| 2180 | int mx = (motion_val[direction][xy][0] >> shift) + sx; |
| 2181 | int my = (motion_val[direction][xy][1] >> shift) + sy; |
| 2182 | mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx, my, direction); |
| 2183 | } |
| 2184 | } else if (IS_16X8(mb_type)) { |
| 2185 | for (i = 0; i < 2; i++) { |
| 2186 | int sx = mb_x * 16 + 8; |
| 2187 | int sy = mb_y * 16 + 4 + 8 * i; |
| 2188 | int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1); |
| 2189 | int mx = (motion_val[direction][xy][0] >> shift); |
| 2190 | int my = (motion_val[direction][xy][1] >> shift); |
| 2191 | |
| 2192 | if (IS_INTERLACED(mb_type)) |
| 2193 | my *= 2; |
| 2194 | |
| 2195 | mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx + sx, my + sy, direction); |
| 2196 | } |
| 2197 | } else if (IS_8X16(mb_type)) { |
| 2198 | for (i = 0; i < 2; i++) { |
| 2199 | int sx = mb_x * 16 + 4 + 8 * i; |
| 2200 | int sy = mb_y * 16 + 8; |
| 2201 | int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1); |
| 2202 | int mx = motion_val[direction][xy][0] >> shift; |
| 2203 | int my = motion_val[direction][xy][1] >> shift; |
| 2204 | |
| 2205 | if (IS_INTERLACED(mb_type)) |
| 2206 | my *= 2; |
| 2207 | |
| 2208 | mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx + sx, my + sy, direction); |
| 2209 | } |
| 2210 | } else { |
| 2211 | int sx = mb_x * 16 + 8; |
| 2212 | int sy = mb_y * 16 + 8; |
| 2213 | int xy = (mb_x + mb_y * mv_stride) << mv_sample_log2; |
| 2214 | int mx = (motion_val[direction][xy][0]>>shift) + sx; |
| 2215 | int my = (motion_val[direction][xy][1]>>shift) + sy; |
| 2216 | mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx, my, direction); |
| 2217 | } |
| 2218 | } |
| 2219 | } |
| 2220 | } |
| 2221 | |
| 2222 | if (mbcount) { |
| 2223 | AVFrameSideData *sd; |
| 2224 | |
| 2225 | av_log(avctx, AV_LOG_DEBUG, "Adding %d MVs info to frame %d\n", mbcount, avctx->frame_number); |
| 2226 | sd = av_frame_new_side_data(pict, AV_FRAME_DATA_MOTION_VECTORS, mbcount * sizeof(AVMotionVector)); |
| 2227 | if (!sd) |
| 2228 | return; |
| 2229 | memcpy(sd->data, mvs, mbcount * sizeof(AVMotionVector)); |
| 2230 | } |
| 2231 | |
| 2232 | av_freep(&mvs); |
| 2233 | } |
| 2234 | |
| 2235 | /* TODO: export all the following to make them accessible for users (and filters) */ |
| 2236 | if (avctx->hwaccel || !mbtype_table |
| 2237 | || (avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)) |
| 2238 | return; |
| 2239 | |
| 2240 | |
| 2241 | if (avctx->debug & (FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)) { |
| 2242 | int x,y; |
| 2243 | |
| 2244 | av_log(avctx, AV_LOG_DEBUG, "New frame, type: %c\n", |
| 2245 | av_get_picture_type_char(pict->pict_type)); |
| 2246 | for (y = 0; y < mb_height; y++) { |
| 2247 | for (x = 0; x < mb_width; x++) { |
| 2248 | if (avctx->debug & FF_DEBUG_SKIP) { |
| 2249 | int count = mbskip_table[x + y * mb_stride]; |
| 2250 | if (count > 9) |
| 2251 | count = 9; |
| 2252 | av_log(avctx, AV_LOG_DEBUG, "%1d", count); |
| 2253 | } |
| 2254 | if (avctx->debug & FF_DEBUG_QP) { |
| 2255 | av_log(avctx, AV_LOG_DEBUG, "%2d", |
| 2256 | qscale_table[x + y * mb_stride]); |
| 2257 | } |
| 2258 | if (avctx->debug & FF_DEBUG_MB_TYPE) { |
| 2259 | int mb_type = mbtype_table[x + y * mb_stride]; |
| 2260 | // Type & MV direction |
| 2261 | if (IS_PCM(mb_type)) |
| 2262 | av_log(avctx, AV_LOG_DEBUG, "P"); |
| 2263 | else if (IS_INTRA(mb_type) && IS_ACPRED(mb_type)) |
| 2264 | av_log(avctx, AV_LOG_DEBUG, "A"); |
| 2265 | else if (IS_INTRA4x4(mb_type)) |
| 2266 | av_log(avctx, AV_LOG_DEBUG, "i"); |
| 2267 | else if (IS_INTRA16x16(mb_type)) |
| 2268 | av_log(avctx, AV_LOG_DEBUG, "I"); |
| 2269 | else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type)) |
| 2270 | av_log(avctx, AV_LOG_DEBUG, "d"); |
| 2271 | else if (IS_DIRECT(mb_type)) |
| 2272 | av_log(avctx, AV_LOG_DEBUG, "D"); |
| 2273 | else if (IS_GMC(mb_type) && IS_SKIP(mb_type)) |
| 2274 | av_log(avctx, AV_LOG_DEBUG, "g"); |
| 2275 | else if (IS_GMC(mb_type)) |
| 2276 | av_log(avctx, AV_LOG_DEBUG, "G"); |
| 2277 | else if (IS_SKIP(mb_type)) |
| 2278 | av_log(avctx, AV_LOG_DEBUG, "S"); |
| 2279 | else if (!USES_LIST(mb_type, 1)) |
| 2280 | av_log(avctx, AV_LOG_DEBUG, ">"); |
| 2281 | else if (!USES_LIST(mb_type, 0)) |
| 2282 | av_log(avctx, AV_LOG_DEBUG, "<"); |
| 2283 | else { |
| 2284 | av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1)); |
| 2285 | av_log(avctx, AV_LOG_DEBUG, "X"); |
| 2286 | } |
| 2287 | |
| 2288 | // segmentation |
| 2289 | if (IS_8X8(mb_type)) |
| 2290 | av_log(avctx, AV_LOG_DEBUG, "+"); |
| 2291 | else if (IS_16X8(mb_type)) |
| 2292 | av_log(avctx, AV_LOG_DEBUG, "-"); |
| 2293 | else if (IS_8X16(mb_type)) |
| 2294 | av_log(avctx, AV_LOG_DEBUG, "|"); |
| 2295 | else if (IS_INTRA(mb_type) || IS_16X16(mb_type)) |
| 2296 | av_log(avctx, AV_LOG_DEBUG, " "); |
| 2297 | else |
| 2298 | av_log(avctx, AV_LOG_DEBUG, "?"); |
| 2299 | |
| 2300 | |
| 2301 | if (IS_INTERLACED(mb_type)) |
| 2302 | av_log(avctx, AV_LOG_DEBUG, "="); |
| 2303 | else |
| 2304 | av_log(avctx, AV_LOG_DEBUG, " "); |
| 2305 | } |
| 2306 | } |
| 2307 | av_log(avctx, AV_LOG_DEBUG, "\n"); |
| 2308 | } |
| 2309 | } |
| 2310 | |
| 2311 | if ((avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) || |
| 2312 | (avctx->debug_mv)) { |
| 2313 | int mb_y; |
| 2314 | int i; |
| 2315 | int h_chroma_shift, v_chroma_shift, block_height; |
| 2316 | #if FF_API_VISMV |
| 2317 | const int shift = 1 + quarter_sample; |
| 2318 | uint8_t *ptr; |
| 2319 | const int width = avctx->width; |
| 2320 | const int height = avctx->height; |
| 2321 | #endif |
| 2322 | const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1; |
| 2323 | const int mv_stride = (mb_width << mv_sample_log2) + |
| 2324 | (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1); |
| 2325 | |
| 2326 | *low_delay = 0; // needed to see the vectors without trashing the buffers |
| 2327 | |
| 2328 | avcodec_get_chroma_sub_sample(avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift); |
| 2329 | |
| 2330 | av_frame_make_writable(pict); |
| 2331 | |
| 2332 | pict->opaque = NULL; |
| 2333 | #if FF_API_VISMV |
| 2334 | ptr = pict->data[0]; |
| 2335 | #endif |
| 2336 | block_height = 16 >> v_chroma_shift; |
| 2337 | |
| 2338 | for (mb_y = 0; mb_y < mb_height; mb_y++) { |
| 2339 | int mb_x; |
| 2340 | for (mb_x = 0; mb_x < mb_width; mb_x++) { |
| 2341 | const int mb_index = mb_x + mb_y * mb_stride; |
| 2342 | #if FF_API_VISMV |
| 2343 | if ((avctx->debug_mv) && motion_val[0]) { |
| 2344 | int type; |
| 2345 | for (type = 0; type < 3; type++) { |
| 2346 | int direction = 0; |
| 2347 | switch (type) { |
| 2348 | case 0: |
| 2349 | if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_P_FOR)) || |
| 2350 | (pict->pict_type!= AV_PICTURE_TYPE_P)) |
| 2351 | continue; |
| 2352 | direction = 0; |
| 2353 | break; |
| 2354 | case 1: |
| 2355 | if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_FOR)) || |
| 2356 | (pict->pict_type!= AV_PICTURE_TYPE_B)) |
| 2357 | continue; |
| 2358 | direction = 0; |
| 2359 | break; |
| 2360 | case 2: |
| 2361 | if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_BACK)) || |
| 2362 | (pict->pict_type!= AV_PICTURE_TYPE_B)) |
| 2363 | continue; |
| 2364 | direction = 1; |
| 2365 | break; |
| 2366 | } |
| 2367 | if (!USES_LIST(mbtype_table[mb_index], direction)) |
| 2368 | continue; |
| 2369 | |
| 2370 | if (IS_8X8(mbtype_table[mb_index])) { |
| 2371 | int i; |
| 2372 | for (i = 0; i < 4; i++) { |
| 2373 | int sx = mb_x * 16 + 4 + 8 * (i & 1); |
| 2374 | int sy = mb_y * 16 + 4 + 8 * (i >> 1); |
| 2375 | int xy = (mb_x * 2 + (i & 1) + |
| 2376 | (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1); |
| 2377 | int mx = (motion_val[direction][xy][0] >> shift) + sx; |
| 2378 | int my = (motion_val[direction][xy][1] >> shift) + sy; |
| 2379 | draw_arrow(ptr, sx, sy, mx, my, width, |
| 2380 | height, pict->linesize[0], 100, 0, direction); |
| 2381 | } |
| 2382 | } else if (IS_16X8(mbtype_table[mb_index])) { |
| 2383 | int i; |
| 2384 | for (i = 0; i < 2; i++) { |
| 2385 | int sx = mb_x * 16 + 8; |
| 2386 | int sy = mb_y * 16 + 4 + 8 * i; |
| 2387 | int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1); |
| 2388 | int mx = (motion_val[direction][xy][0] >> shift); |
| 2389 | int my = (motion_val[direction][xy][1] >> shift); |
| 2390 | |
| 2391 | if (IS_INTERLACED(mbtype_table[mb_index])) |
| 2392 | my *= 2; |
| 2393 | |
| 2394 | draw_arrow(ptr, sx, sy, mx + sx, my + sy, width, |
| 2395 | height, pict->linesize[0], 100, 0, direction); |
| 2396 | } |
| 2397 | } else if (IS_8X16(mbtype_table[mb_index])) { |
| 2398 | int i; |
| 2399 | for (i = 0; i < 2; i++) { |
| 2400 | int sx = mb_x * 16 + 4 + 8 * i; |
| 2401 | int sy = mb_y * 16 + 8; |
| 2402 | int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1); |
| 2403 | int mx = motion_val[direction][xy][0] >> shift; |
| 2404 | int my = motion_val[direction][xy][1] >> shift; |
| 2405 | |
| 2406 | if (IS_INTERLACED(mbtype_table[mb_index])) |
| 2407 | my *= 2; |
| 2408 | |
| 2409 | draw_arrow(ptr, sx, sy, mx + sx, my + sy, width, |
| 2410 | height, pict->linesize[0], 100, 0, direction); |
| 2411 | } |
| 2412 | } else { |
| 2413 | int sx= mb_x * 16 + 8; |
| 2414 | int sy= mb_y * 16 + 8; |
| 2415 | int xy= (mb_x + mb_y * mv_stride) << mv_sample_log2; |
| 2416 | int mx= (motion_val[direction][xy][0]>>shift) + sx; |
| 2417 | int my= (motion_val[direction][xy][1]>>shift) + sy; |
| 2418 | draw_arrow(ptr, sx, sy, mx, my, width, height, pict->linesize[0], 100, 0, direction); |
| 2419 | } |
| 2420 | } |
| 2421 | } |
| 2422 | #endif |
| 2423 | if ((avctx->debug & FF_DEBUG_VIS_QP)) { |
| 2424 | uint64_t c = (qscale_table[mb_index] * 128 / 31) * |
| 2425 | 0x0101010101010101ULL; |
| 2426 | int y; |
| 2427 | for (y = 0; y < block_height; y++) { |
| 2428 | *(uint64_t *)(pict->data[1] + 8 * mb_x + |
| 2429 | (block_height * mb_y + y) * |
| 2430 | pict->linesize[1]) = c; |
| 2431 | *(uint64_t *)(pict->data[2] + 8 * mb_x + |
| 2432 | (block_height * mb_y + y) * |
| 2433 | pict->linesize[2]) = c; |
| 2434 | } |
| 2435 | } |
| 2436 | if ((avctx->debug & FF_DEBUG_VIS_MB_TYPE) && |
| 2437 | motion_val[0]) { |
| 2438 | int mb_type = mbtype_table[mb_index]; |
| 2439 | uint64_t u,v; |
| 2440 | int y; |
| 2441 | #define COLOR(theta, r) \ |
| 2442 | u = (int)(128 + r * cos(theta * 3.141592 / 180)); \ |
| 2443 | v = (int)(128 + r * sin(theta * 3.141592 / 180)); |
| 2444 | |
| 2445 | |
| 2446 | u = v = 128; |
| 2447 | if (IS_PCM(mb_type)) { |
| 2448 | COLOR(120, 48) |
| 2449 | } else if ((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) || |
| 2450 | IS_INTRA16x16(mb_type)) { |
| 2451 | COLOR(30, 48) |
| 2452 | } else if (IS_INTRA4x4(mb_type)) { |
| 2453 | COLOR(90, 48) |
| 2454 | } else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type)) { |
| 2455 | // COLOR(120, 48) |
| 2456 | } else if (IS_DIRECT(mb_type)) { |
| 2457 | COLOR(150, 48) |
| 2458 | } else if (IS_GMC(mb_type) && IS_SKIP(mb_type)) { |
| 2459 | COLOR(170, 48) |
| 2460 | } else if (IS_GMC(mb_type)) { |
| 2461 | COLOR(190, 48) |
| 2462 | } else if (IS_SKIP(mb_type)) { |
| 2463 | // COLOR(180, 48) |
| 2464 | } else if (!USES_LIST(mb_type, 1)) { |
| 2465 | COLOR(240, 48) |
| 2466 | } else if (!USES_LIST(mb_type, 0)) { |
| 2467 | COLOR(0, 48) |
| 2468 | } else { |
| 2469 | av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1)); |
| 2470 | COLOR(300,48) |
| 2471 | } |
| 2472 | |
| 2473 | u *= 0x0101010101010101ULL; |
| 2474 | v *= 0x0101010101010101ULL; |
| 2475 | for (y = 0; y < block_height; y++) { |
| 2476 | *(uint64_t *)(pict->data[1] + 8 * mb_x + |
| 2477 | (block_height * mb_y + y) * pict->linesize[1]) = u; |
| 2478 | *(uint64_t *)(pict->data[2] + 8 * mb_x + |
| 2479 | (block_height * mb_y + y) * pict->linesize[2]) = v; |
| 2480 | } |
| 2481 | |
| 2482 | // segmentation |
| 2483 | if (IS_8X8(mb_type) || IS_16X8(mb_type)) { |
| 2484 | *(uint64_t *)(pict->data[0] + 16 * mb_x + 0 + |
| 2485 | (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL; |
| 2486 | *(uint64_t *)(pict->data[0] + 16 * mb_x + 8 + |
| 2487 | (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL; |
| 2488 | } |
| 2489 | if (IS_8X8(mb_type) || IS_8X16(mb_type)) { |
| 2490 | for (y = 0; y < 16; y++) |
| 2491 | pict->data[0][16 * mb_x + 8 + (16 * mb_y + y) * |
| 2492 | pict->linesize[0]] ^= 0x80; |
| 2493 | } |
| 2494 | if (IS_8X8(mb_type) && mv_sample_log2 >= 2) { |
| 2495 | int dm = 1 << (mv_sample_log2 - 2); |
| 2496 | for (i = 0; i < 4; i++) { |
| 2497 | int sx = mb_x * 16 + 8 * (i & 1); |
| 2498 | int sy = mb_y * 16 + 8 * (i >> 1); |
| 2499 | int xy = (mb_x * 2 + (i & 1) + |
| 2500 | (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1); |
| 2501 | // FIXME bidir |
| 2502 | int32_t *mv = (int32_t *) &motion_val[0][xy]; |
| 2503 | if (mv[0] != mv[dm] || |
| 2504 | mv[dm * mv_stride] != mv[dm * (mv_stride + 1)]) |
| 2505 | for (y = 0; y < 8; y++) |
| 2506 | pict->data[0][sx + 4 + (sy + y) * pict->linesize[0]] ^= 0x80; |
| 2507 | if (mv[0] != mv[dm * mv_stride] || mv[dm] != mv[dm * (mv_stride + 1)]) |
| 2508 | *(uint64_t *)(pict->data[0] + sx + (sy + 4) * |
| 2509 | pict->linesize[0]) ^= 0x8080808080808080ULL; |
| 2510 | } |
| 2511 | } |
| 2512 | |
| 2513 | if (IS_INTERLACED(mb_type) && |
| 2514 | avctx->codec->id == AV_CODEC_ID_H264) { |
| 2515 | // hmm |
| 2516 | } |
| 2517 | } |
| 2518 | mbskip_table[mb_index] = 0; |
| 2519 | } |
| 2520 | } |
| 2521 | } |
| 2522 | } |
| 2523 | |
| 2524 | void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict) |
| 2525 | { |
| 2526 | ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type, |
| 2527 | p->qscale_table, p->motion_val, &s->low_delay, |
| 2528 | s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample); |
| 2529 | } |
| 2530 | |
| 2531 | int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type) |
| 2532 | { |
| 2533 | AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf); |
| 2534 | int offset = 2*s->mb_stride + 1; |
| 2535 | if(!ref) |
| 2536 | return AVERROR(ENOMEM); |
| 2537 | av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16)); |
| 2538 | ref->size -= offset; |
| 2539 | ref->data += offset; |
| 2540 | return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type); |
| 2541 | } |
| 2542 | |
| 2543 | static inline int hpel_motion_lowres(MpegEncContext *s, |
| 2544 | uint8_t *dest, uint8_t *src, |
| 2545 | int field_based, int field_select, |
| 2546 | int src_x, int src_y, |
| 2547 | int width, int height, ptrdiff_t stride, |
| 2548 | int h_edge_pos, int v_edge_pos, |
| 2549 | int w, int h, h264_chroma_mc_func *pix_op, |
| 2550 | int motion_x, int motion_y) |
| 2551 | { |
| 2552 | const int lowres = s->avctx->lowres; |
| 2553 | const int op_index = FFMIN(lowres, 3); |
| 2554 | const int s_mask = (2 << lowres) - 1; |
| 2555 | int emu = 0; |
| 2556 | int sx, sy; |
| 2557 | |
| 2558 | if (s->quarter_sample) { |
| 2559 | motion_x /= 2; |
| 2560 | motion_y /= 2; |
| 2561 | } |
| 2562 | |
| 2563 | sx = motion_x & s_mask; |
| 2564 | sy = motion_y & s_mask; |
| 2565 | src_x += motion_x >> lowres + 1; |
| 2566 | src_y += motion_y >> lowres + 1; |
| 2567 | |
| 2568 | src += src_y * stride + src_x; |
| 2569 | |
| 2570 | if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w, 0) || |
| 2571 | (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) { |
| 2572 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src, |
| 2573 | s->linesize, s->linesize, |
| 2574 | w + 1, (h + 1) << field_based, |
| 2575 | src_x, src_y << field_based, |
| 2576 | h_edge_pos, v_edge_pos); |
| 2577 | src = s->edge_emu_buffer; |
| 2578 | emu = 1; |
| 2579 | } |
| 2580 | |
| 2581 | sx = (sx << 2) >> lowres; |
| 2582 | sy = (sy << 2) >> lowres; |
| 2583 | if (field_select) |
| 2584 | src += s->linesize; |
| 2585 | pix_op[op_index](dest, src, stride, h, sx, sy); |
| 2586 | return emu; |
| 2587 | } |
| 2588 | |
| 2589 | /* apply one mpeg motion vector to the three components */ |
| 2590 | static av_always_inline void mpeg_motion_lowres(MpegEncContext *s, |
| 2591 | uint8_t *dest_y, |
| 2592 | uint8_t *dest_cb, |
| 2593 | uint8_t *dest_cr, |
| 2594 | int field_based, |
| 2595 | int bottom_field, |
| 2596 | int field_select, |
| 2597 | uint8_t **ref_picture, |
| 2598 | h264_chroma_mc_func *pix_op, |
| 2599 | int motion_x, int motion_y, |
| 2600 | int h, int mb_y) |
| 2601 | { |
| 2602 | uint8_t *ptr_y, *ptr_cb, *ptr_cr; |
| 2603 | int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy; |
| 2604 | ptrdiff_t uvlinesize, linesize; |
| 2605 | const int lowres = s->avctx->lowres; |
| 2606 | const int op_index = FFMIN(lowres-1+s->chroma_x_shift, 3); |
| 2607 | const int block_s = 8>>lowres; |
| 2608 | const int s_mask = (2 << lowres) - 1; |
| 2609 | const int h_edge_pos = s->h_edge_pos >> lowres; |
| 2610 | const int v_edge_pos = s->v_edge_pos >> lowres; |
| 2611 | linesize = s->current_picture.f->linesize[0] << field_based; |
| 2612 | uvlinesize = s->current_picture.f->linesize[1] << field_based; |
| 2613 | |
| 2614 | // FIXME obviously not perfect but qpel will not work in lowres anyway |
| 2615 | if (s->quarter_sample) { |
| 2616 | motion_x /= 2; |
| 2617 | motion_y /= 2; |
| 2618 | } |
| 2619 | |
| 2620 | if(field_based){ |
| 2621 | motion_y += (bottom_field - field_select)*((1 << lowres)-1); |
| 2622 | } |
| 2623 | |
| 2624 | sx = motion_x & s_mask; |
| 2625 | sy = motion_y & s_mask; |
| 2626 | src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1); |
| 2627 | src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1); |
| 2628 | |
| 2629 | if (s->out_format == FMT_H263) { |
| 2630 | uvsx = ((motion_x >> 1) & s_mask) | (sx & 1); |
| 2631 | uvsy = ((motion_y >> 1) & s_mask) | (sy & 1); |
| 2632 | uvsrc_x = src_x >> 1; |
| 2633 | uvsrc_y = src_y >> 1; |
| 2634 | } else if (s->out_format == FMT_H261) { |
| 2635 | // even chroma mv's are full pel in H261 |
| 2636 | mx = motion_x / 4; |
| 2637 | my = motion_y / 4; |
| 2638 | uvsx = (2 * mx) & s_mask; |
| 2639 | uvsy = (2 * my) & s_mask; |
| 2640 | uvsrc_x = s->mb_x * block_s + (mx >> lowres); |
| 2641 | uvsrc_y = mb_y * block_s + (my >> lowres); |
| 2642 | } else { |
| 2643 | if(s->chroma_y_shift){ |
| 2644 | mx = motion_x / 2; |
| 2645 | my = motion_y / 2; |
| 2646 | uvsx = mx & s_mask; |
| 2647 | uvsy = my & s_mask; |
| 2648 | uvsrc_x = s->mb_x * block_s + (mx >> lowres + 1); |
| 2649 | uvsrc_y = (mb_y * block_s >> field_based) + (my >> lowres + 1); |
| 2650 | } else { |
| 2651 | if(s->chroma_x_shift){ |
| 2652 | //Chroma422 |
| 2653 | mx = motion_x / 2; |
| 2654 | uvsx = mx & s_mask; |
| 2655 | uvsy = motion_y & s_mask; |
| 2656 | uvsrc_y = src_y; |
| 2657 | uvsrc_x = s->mb_x*block_s + (mx >> (lowres+1)); |
| 2658 | } else { |
| 2659 | //Chroma444 |
| 2660 | uvsx = motion_x & s_mask; |
| 2661 | uvsy = motion_y & s_mask; |
| 2662 | uvsrc_x = src_x; |
| 2663 | uvsrc_y = src_y; |
| 2664 | } |
| 2665 | } |
| 2666 | } |
| 2667 | |
| 2668 | ptr_y = ref_picture[0] + src_y * linesize + src_x; |
| 2669 | ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x; |
| 2670 | ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x; |
| 2671 | |
| 2672 | if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s, 0) || uvsrc_y<0 || |
| 2673 | (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) { |
| 2674 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, |
| 2675 | linesize >> field_based, linesize >> field_based, |
| 2676 | 17, 17 + field_based, |
| 2677 | src_x, src_y << field_based, h_edge_pos, |
| 2678 | v_edge_pos); |
| 2679 | ptr_y = s->edge_emu_buffer; |
| 2680 | if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) { |
| 2681 | uint8_t *ubuf = s->edge_emu_buffer + 18 * s->linesize; |
| 2682 | uint8_t *vbuf =ubuf + 9 * s->uvlinesize; |
| 2683 | s->vdsp.emulated_edge_mc(ubuf, ptr_cb, |
| 2684 | uvlinesize >> field_based, uvlinesize >> field_based, |
| 2685 | 9, 9 + field_based, |
| 2686 | uvsrc_x, uvsrc_y << field_based, |
| 2687 | h_edge_pos >> 1, v_edge_pos >> 1); |
| 2688 | s->vdsp.emulated_edge_mc(vbuf, ptr_cr, |
| 2689 | uvlinesize >> field_based,uvlinesize >> field_based, |
| 2690 | 9, 9 + field_based, |
| 2691 | uvsrc_x, uvsrc_y << field_based, |
| 2692 | h_edge_pos >> 1, v_edge_pos >> 1); |
| 2693 | ptr_cb = ubuf; |
| 2694 | ptr_cr = vbuf; |
| 2695 | } |
| 2696 | } |
| 2697 | |
| 2698 | // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data |
| 2699 | if (bottom_field) { |
| 2700 | dest_y += s->linesize; |
| 2701 | dest_cb += s->uvlinesize; |
| 2702 | dest_cr += s->uvlinesize; |
| 2703 | } |
| 2704 | |
| 2705 | if (field_select) { |
| 2706 | ptr_y += s->linesize; |
| 2707 | ptr_cb += s->uvlinesize; |
| 2708 | ptr_cr += s->uvlinesize; |
| 2709 | } |
| 2710 | |
| 2711 | sx = (sx << 2) >> lowres; |
| 2712 | sy = (sy << 2) >> lowres; |
| 2713 | pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy); |
| 2714 | |
| 2715 | if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) { |
| 2716 | int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h; |
| 2717 | uvsx = (uvsx << 2) >> lowres; |
| 2718 | uvsy = (uvsy << 2) >> lowres; |
| 2719 | if (hc) { |
| 2720 | pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy); |
| 2721 | pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy); |
| 2722 | } |
| 2723 | } |
| 2724 | // FIXME h261 lowres loop filter |
| 2725 | } |
| 2726 | |
| 2727 | static inline void chroma_4mv_motion_lowres(MpegEncContext *s, |
| 2728 | uint8_t *dest_cb, uint8_t *dest_cr, |
| 2729 | uint8_t **ref_picture, |
| 2730 | h264_chroma_mc_func * pix_op, |
| 2731 | int mx, int my) |
| 2732 | { |
| 2733 | const int lowres = s->avctx->lowres; |
| 2734 | const int op_index = FFMIN(lowres, 3); |
| 2735 | const int block_s = 8 >> lowres; |
| 2736 | const int s_mask = (2 << lowres) - 1; |
| 2737 | const int h_edge_pos = s->h_edge_pos >> lowres + 1; |
| 2738 | const int v_edge_pos = s->v_edge_pos >> lowres + 1; |
| 2739 | int emu = 0, src_x, src_y, sx, sy; |
| 2740 | ptrdiff_t offset; |
| 2741 | uint8_t *ptr; |
| 2742 | |
| 2743 | if (s->quarter_sample) { |
| 2744 | mx /= 2; |
| 2745 | my /= 2; |
| 2746 | } |
| 2747 | |
| 2748 | /* In case of 8X8, we construct a single chroma motion vector |
| 2749 | with a special rounding */ |
| 2750 | mx = ff_h263_round_chroma(mx); |
| 2751 | my = ff_h263_round_chroma(my); |
| 2752 | |
| 2753 | sx = mx & s_mask; |
| 2754 | sy = my & s_mask; |
| 2755 | src_x = s->mb_x * block_s + (mx >> lowres + 1); |
| 2756 | src_y = s->mb_y * block_s + (my >> lowres + 1); |
| 2757 | |
| 2758 | offset = src_y * s->uvlinesize + src_x; |
| 2759 | ptr = ref_picture[1] + offset; |
| 2760 | if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) || |
| 2761 | (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) { |
| 2762 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, |
| 2763 | s->uvlinesize, s->uvlinesize, |
| 2764 | 9, 9, |
| 2765 | src_x, src_y, h_edge_pos, v_edge_pos); |
| 2766 | ptr = s->edge_emu_buffer; |
| 2767 | emu = 1; |
| 2768 | } |
| 2769 | sx = (sx << 2) >> lowres; |
| 2770 | sy = (sy << 2) >> lowres; |
| 2771 | pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy); |
| 2772 | |
| 2773 | ptr = ref_picture[2] + offset; |
| 2774 | if (emu) { |
| 2775 | s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, |
| 2776 | s->uvlinesize, s->uvlinesize, |
| 2777 | 9, 9, |
| 2778 | src_x, src_y, h_edge_pos, v_edge_pos); |
| 2779 | ptr = s->edge_emu_buffer; |
| 2780 | } |
| 2781 | pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy); |
| 2782 | } |
| 2783 | |
| 2784 | /** |
| 2785 | * motion compensation of a single macroblock |
| 2786 | * @param s context |
| 2787 | * @param dest_y luma destination pointer |
| 2788 | * @param dest_cb chroma cb/u destination pointer |
| 2789 | * @param dest_cr chroma cr/v destination pointer |
| 2790 | * @param dir direction (0->forward, 1->backward) |
| 2791 | * @param ref_picture array[3] of pointers to the 3 planes of the reference picture |
| 2792 | * @param pix_op halfpel motion compensation function (average or put normally) |
| 2793 | * the motion vectors are taken from s->mv and the MV type from s->mv_type |
| 2794 | */ |
| 2795 | static inline void MPV_motion_lowres(MpegEncContext *s, |
| 2796 | uint8_t *dest_y, uint8_t *dest_cb, |
| 2797 | uint8_t *dest_cr, |
| 2798 | int dir, uint8_t **ref_picture, |
| 2799 | h264_chroma_mc_func *pix_op) |
| 2800 | { |
| 2801 | int mx, my; |
| 2802 | int mb_x, mb_y, i; |
| 2803 | const int lowres = s->avctx->lowres; |
| 2804 | const int block_s = 8 >>lowres; |
| 2805 | |
| 2806 | mb_x = s->mb_x; |
| 2807 | mb_y = s->mb_y; |
| 2808 | |
| 2809 | switch (s->mv_type) { |
| 2810 | case MV_TYPE_16X16: |
| 2811 | mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
| 2812 | 0, 0, 0, |
| 2813 | ref_picture, pix_op, |
| 2814 | s->mv[dir][0][0], s->mv[dir][0][1], |
| 2815 | 2 * block_s, mb_y); |
| 2816 | break; |
| 2817 | case MV_TYPE_8X8: |
| 2818 | mx = 0; |
| 2819 | my = 0; |
| 2820 | for (i = 0; i < 4; i++) { |
| 2821 | hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) * |
| 2822 | s->linesize) * block_s, |
| 2823 | ref_picture[0], 0, 0, |
| 2824 | (2 * mb_x + (i & 1)) * block_s, |
| 2825 | (2 * mb_y + (i >> 1)) * block_s, |
| 2826 | s->width, s->height, s->linesize, |
| 2827 | s->h_edge_pos >> lowres, s->v_edge_pos >> lowres, |
| 2828 | block_s, block_s, pix_op, |
| 2829 | s->mv[dir][i][0], s->mv[dir][i][1]); |
| 2830 | |
| 2831 | mx += s->mv[dir][i][0]; |
| 2832 | my += s->mv[dir][i][1]; |
| 2833 | } |
| 2834 | |
| 2835 | if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) |
| 2836 | chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture, |
| 2837 | pix_op, mx, my); |
| 2838 | break; |
| 2839 | case MV_TYPE_FIELD: |
| 2840 | if (s->picture_structure == PICT_FRAME) { |
| 2841 | /* top field */ |
| 2842 | mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
| 2843 | 1, 0, s->field_select[dir][0], |
| 2844 | ref_picture, pix_op, |
| 2845 | s->mv[dir][0][0], s->mv[dir][0][1], |
| 2846 | block_s, mb_y); |
| 2847 | /* bottom field */ |
| 2848 | mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
| 2849 | 1, 1, s->field_select[dir][1], |
| 2850 | ref_picture, pix_op, |
| 2851 | s->mv[dir][1][0], s->mv[dir][1][1], |
| 2852 | block_s, mb_y); |
| 2853 | } else { |
| 2854 | if (s->picture_structure != s->field_select[dir][0] + 1 && |
| 2855 | s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) { |
| 2856 | ref_picture = s->current_picture_ptr->f->data; |
| 2857 | |
| 2858 | } |
| 2859 | mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
| 2860 | 0, 0, s->field_select[dir][0], |
| 2861 | ref_picture, pix_op, |
| 2862 | s->mv[dir][0][0], |
| 2863 | s->mv[dir][0][1], 2 * block_s, mb_y >> 1); |
| 2864 | } |
| 2865 | break; |
| 2866 | case MV_TYPE_16X8: |
| 2867 | for (i = 0; i < 2; i++) { |
| 2868 | uint8_t **ref2picture; |
| 2869 | |
| 2870 | if (s->picture_structure == s->field_select[dir][i] + 1 || |
| 2871 | s->pict_type == AV_PICTURE_TYPE_B || s->first_field) { |
| 2872 | ref2picture = ref_picture; |
| 2873 | } else { |
| 2874 | ref2picture = s->current_picture_ptr->f->data; |
| 2875 | } |
| 2876 | |
| 2877 | mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
| 2878 | 0, 0, s->field_select[dir][i], |
| 2879 | ref2picture, pix_op, |
| 2880 | s->mv[dir][i][0], s->mv[dir][i][1] + |
| 2881 | 2 * block_s * i, block_s, mb_y >> 1); |
| 2882 | |
| 2883 | dest_y += 2 * block_s * s->linesize; |
| 2884 | dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize; |
| 2885 | dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize; |
| 2886 | } |
| 2887 | break; |
| 2888 | case MV_TYPE_DMV: |
| 2889 | if (s->picture_structure == PICT_FRAME) { |
| 2890 | for (i = 0; i < 2; i++) { |
| 2891 | int j; |
| 2892 | for (j = 0; j < 2; j++) { |
| 2893 | mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
| 2894 | 1, j, j ^ i, |
| 2895 | ref_picture, pix_op, |
| 2896 | s->mv[dir][2 * i + j][0], |
| 2897 | s->mv[dir][2 * i + j][1], |
| 2898 | block_s, mb_y); |
| 2899 | } |
| 2900 | pix_op = s->h264chroma.avg_h264_chroma_pixels_tab; |
| 2901 | } |
| 2902 | } else { |
| 2903 | for (i = 0; i < 2; i++) { |
| 2904 | mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr, |
| 2905 | 0, 0, s->picture_structure != i + 1, |
| 2906 | ref_picture, pix_op, |
| 2907 | s->mv[dir][2 * i][0],s->mv[dir][2 * i][1], |
| 2908 | 2 * block_s, mb_y >> 1); |
| 2909 | |
| 2910 | // after put we make avg of the same block |
| 2911 | pix_op = s->h264chroma.avg_h264_chroma_pixels_tab; |
| 2912 | |
| 2913 | // opposite parity is always in the same |
| 2914 | // frame if this is second field |
| 2915 | if (!s->first_field) { |
| 2916 | ref_picture = s->current_picture_ptr->f->data; |
| 2917 | } |
| 2918 | } |
| 2919 | } |
| 2920 | break; |
| 2921 | default: |
| 2922 | av_assert2(0); |
| 2923 | } |
| 2924 | } |
| 2925 | |
| 2926 | /** |
| 2927 | * find the lowest MB row referenced in the MVs |
| 2928 | */ |
| 2929 | int ff_mpv_lowest_referenced_row(MpegEncContext *s, int dir) |
| 2930 | { |
| 2931 | int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample; |
| 2932 | int my, off, i, mvs; |
| 2933 | |
| 2934 | if (s->picture_structure != PICT_FRAME || s->mcsel) |
| 2935 | goto unhandled; |
| 2936 | |
| 2937 | switch (s->mv_type) { |
| 2938 | case MV_TYPE_16X16: |
| 2939 | mvs = 1; |
| 2940 | break; |
| 2941 | case MV_TYPE_16X8: |
| 2942 | mvs = 2; |
| 2943 | break; |
| 2944 | case MV_TYPE_8X8: |
| 2945 | mvs = 4; |
| 2946 | break; |
| 2947 | default: |
| 2948 | goto unhandled; |
| 2949 | } |
| 2950 | |
| 2951 | for (i = 0; i < mvs; i++) { |
| 2952 | my = s->mv[dir][i][1]<<qpel_shift; |
| 2953 | my_max = FFMAX(my_max, my); |
| 2954 | my_min = FFMIN(my_min, my); |
| 2955 | } |
| 2956 | |
| 2957 | off = (FFMAX(-my_min, my_max) + 63) >> 6; |
| 2958 | |
| 2959 | return FFMIN(FFMAX(s->mb_y + off, 0), s->mb_height-1); |
| 2960 | unhandled: |
| 2961 | return s->mb_height-1; |
| 2962 | } |
| 2963 | |
| 2964 | /* put block[] to dest[] */ |
| 2965 | static inline void put_dct(MpegEncContext *s, |
| 2966 | int16_t *block, int i, uint8_t *dest, int line_size, int qscale) |
| 2967 | { |
| 2968 | s->dct_unquantize_intra(s, block, i, qscale); |
| 2969 | s->idsp.idct_put(dest, line_size, block); |
| 2970 | } |
| 2971 | |
| 2972 | /* add block[] to dest[] */ |
| 2973 | static inline void add_dct(MpegEncContext *s, |
| 2974 | int16_t *block, int i, uint8_t *dest, int line_size) |
| 2975 | { |
| 2976 | if (s->block_last_index[i] >= 0) { |
| 2977 | s->idsp.idct_add(dest, line_size, block); |
| 2978 | } |
| 2979 | } |
| 2980 | |
| 2981 | static inline void add_dequant_dct(MpegEncContext *s, |
| 2982 | int16_t *block, int i, uint8_t *dest, int line_size, int qscale) |
| 2983 | { |
| 2984 | if (s->block_last_index[i] >= 0) { |
| 2985 | s->dct_unquantize_inter(s, block, i, qscale); |
| 2986 | |
| 2987 | s->idsp.idct_add(dest, line_size, block); |
| 2988 | } |
| 2989 | } |
| 2990 | |
| 2991 | /** |
| 2992 | * Clean dc, ac, coded_block for the current non-intra MB. |
| 2993 | */ |
| 2994 | void ff_clean_intra_table_entries(MpegEncContext *s) |
| 2995 | { |
| 2996 | int wrap = s->b8_stride; |
| 2997 | int xy = s->block_index[0]; |
| 2998 | |
| 2999 | s->dc_val[0][xy ] = |
| 3000 | s->dc_val[0][xy + 1 ] = |
| 3001 | s->dc_val[0][xy + wrap] = |
| 3002 | s->dc_val[0][xy + 1 + wrap] = 1024; |
| 3003 | /* ac pred */ |
| 3004 | memset(s->ac_val[0][xy ], 0, 32 * sizeof(int16_t)); |
| 3005 | memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t)); |
| 3006 | if (s->msmpeg4_version>=3) { |
| 3007 | s->coded_block[xy ] = |
| 3008 | s->coded_block[xy + 1 ] = |
| 3009 | s->coded_block[xy + wrap] = |
| 3010 | s->coded_block[xy + 1 + wrap] = 0; |
| 3011 | } |
| 3012 | /* chroma */ |
| 3013 | wrap = s->mb_stride; |
| 3014 | xy = s->mb_x + s->mb_y * wrap; |
| 3015 | s->dc_val[1][xy] = |
| 3016 | s->dc_val[2][xy] = 1024; |
| 3017 | /* ac pred */ |
| 3018 | memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t)); |
| 3019 | memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t)); |
| 3020 | |
| 3021 | s->mbintra_table[xy]= 0; |
| 3022 | } |
| 3023 | |
| 3024 | /* generic function called after a macroblock has been parsed by the |
| 3025 | decoder or after it has been encoded by the encoder. |
| 3026 | |
| 3027 | Important variables used: |
| 3028 | s->mb_intra : true if intra macroblock |
| 3029 | s->mv_dir : motion vector direction |
| 3030 | s->mv_type : motion vector type |
| 3031 | s->mv : motion vector |
| 3032 | s->interlaced_dct : true if interlaced dct used (mpeg2) |
| 3033 | */ |
| 3034 | static av_always_inline |
| 3035 | void mpv_decode_mb_internal(MpegEncContext *s, int16_t block[12][64], |
| 3036 | int lowres_flag, int is_mpeg12) |
| 3037 | { |
| 3038 | const int mb_xy = s->mb_y * s->mb_stride + s->mb_x; |
| 3039 | |
| 3040 | if (CONFIG_XVMC && |
| 3041 | s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) { |
| 3042 | s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks |
| 3043 | return; |
| 3044 | } |
| 3045 | |
| 3046 | if(s->avctx->debug&FF_DEBUG_DCT_COEFF) { |
| 3047 | /* print DCT coefficients */ |
| 3048 | int i,j; |
| 3049 | av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y); |
| 3050 | for(i=0; i<6; i++){ |
| 3051 | for(j=0; j<64; j++){ |
| 3052 | av_log(s->avctx, AV_LOG_DEBUG, "%5d", |
| 3053 | block[i][s->idsp.idct_permutation[j]]); |
| 3054 | } |
| 3055 | av_log(s->avctx, AV_LOG_DEBUG, "\n"); |
| 3056 | } |
| 3057 | } |
| 3058 | |
| 3059 | s->current_picture.qscale_table[mb_xy] = s->qscale; |
| 3060 | |
| 3061 | /* update DC predictors for P macroblocks */ |
| 3062 | if (!s->mb_intra) { |
| 3063 | if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) { |
| 3064 | if(s->mbintra_table[mb_xy]) |
| 3065 | ff_clean_intra_table_entries(s); |
| 3066 | } else { |
| 3067 | s->last_dc[0] = |
| 3068 | s->last_dc[1] = |
| 3069 | s->last_dc[2] = 128 << s->intra_dc_precision; |
| 3070 | } |
| 3071 | } |
| 3072 | else if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) |
| 3073 | s->mbintra_table[mb_xy]=1; |
| 3074 | |
| 3075 | if ( (s->flags&CODEC_FLAG_PSNR) |
| 3076 | || s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor |
| 3077 | || !(s->encoding && (s->intra_only || s->pict_type==AV_PICTURE_TYPE_B) && s->avctx->mb_decision != FF_MB_DECISION_RD)) { //FIXME precalc |
| 3078 | uint8_t *dest_y, *dest_cb, *dest_cr; |
| 3079 | int dct_linesize, dct_offset; |
| 3080 | op_pixels_func (*op_pix)[4]; |
| 3081 | qpel_mc_func (*op_qpix)[16]; |
| 3082 | const int linesize = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics |
| 3083 | const int uvlinesize = s->current_picture.f->linesize[1]; |
| 3084 | const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag; |
| 3085 | const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8; |
| 3086 | |
| 3087 | /* avoid copy if macroblock skipped in last frame too */ |
| 3088 | /* skip only during decoding as we might trash the buffers during encoding a bit */ |
| 3089 | if(!s->encoding){ |
| 3090 | uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy]; |
| 3091 | |
| 3092 | if (s->mb_skipped) { |
| 3093 | s->mb_skipped= 0; |
| 3094 | av_assert2(s->pict_type!=AV_PICTURE_TYPE_I); |
| 3095 | *mbskip_ptr = 1; |
| 3096 | } else if(!s->current_picture.reference) { |
| 3097 | *mbskip_ptr = 1; |
| 3098 | } else{ |
| 3099 | *mbskip_ptr = 0; /* not skipped */ |
| 3100 | } |
| 3101 | } |
| 3102 | |
| 3103 | dct_linesize = linesize << s->interlaced_dct; |
| 3104 | dct_offset = s->interlaced_dct ? linesize : linesize * block_size; |
| 3105 | |
| 3106 | if(readable){ |
| 3107 | dest_y= s->dest[0]; |
| 3108 | dest_cb= s->dest[1]; |
| 3109 | dest_cr= s->dest[2]; |
| 3110 | }else{ |
| 3111 | dest_y = s->b_scratchpad; |
| 3112 | dest_cb= s->b_scratchpad+16*linesize; |
| 3113 | dest_cr= s->b_scratchpad+32*linesize; |
| 3114 | } |
| 3115 | |
| 3116 | if (!s->mb_intra) { |
| 3117 | /* motion handling */ |
| 3118 | /* decoding or more than one mb_type (MC was already done otherwise) */ |
| 3119 | if(!s->encoding){ |
| 3120 | |
| 3121 | if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) { |
| 3122 | if (s->mv_dir & MV_DIR_FORWARD) { |
| 3123 | ff_thread_await_progress(&s->last_picture_ptr->tf, |
| 3124 | ff_mpv_lowest_referenced_row(s, 0), |
| 3125 | 0); |
| 3126 | } |
| 3127 | if (s->mv_dir & MV_DIR_BACKWARD) { |
| 3128 | ff_thread_await_progress(&s->next_picture_ptr->tf, |
| 3129 | ff_mpv_lowest_referenced_row(s, 1), |
| 3130 | 0); |
| 3131 | } |
| 3132 | } |
| 3133 | |
| 3134 | if(lowres_flag){ |
| 3135 | h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab; |
| 3136 | |
| 3137 | if (s->mv_dir & MV_DIR_FORWARD) { |
| 3138 | MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix); |
| 3139 | op_pix = s->h264chroma.avg_h264_chroma_pixels_tab; |
| 3140 | } |
| 3141 | if (s->mv_dir & MV_DIR_BACKWARD) { |
| 3142 | MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix); |
| 3143 | } |
| 3144 | }else{ |
| 3145 | op_qpix = s->me.qpel_put; |
| 3146 | if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){ |
| 3147 | op_pix = s->hdsp.put_pixels_tab; |
| 3148 | }else{ |
| 3149 | op_pix = s->hdsp.put_no_rnd_pixels_tab; |
| 3150 | } |
| 3151 | if (s->mv_dir & MV_DIR_FORWARD) { |
| 3152 | ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix); |
| 3153 | op_pix = s->hdsp.avg_pixels_tab; |
| 3154 | op_qpix= s->me.qpel_avg; |
| 3155 | } |
| 3156 | if (s->mv_dir & MV_DIR_BACKWARD) { |
| 3157 | ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix); |
| 3158 | } |
| 3159 | } |
| 3160 | } |
| 3161 | |
| 3162 | /* skip dequant / idct if we are really late ;) */ |
| 3163 | if(s->avctx->skip_idct){ |
| 3164 | if( (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B) |
| 3165 | ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I) |
| 3166 | || s->avctx->skip_idct >= AVDISCARD_ALL) |
| 3167 | goto skip_idct; |
| 3168 | } |
| 3169 | |
| 3170 | /* add dct residue */ |
| 3171 | if(s->encoding || !( s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO |
| 3172 | || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){ |
| 3173 | add_dequant_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale); |
| 3174 | add_dequant_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale); |
| 3175 | add_dequant_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale); |
| 3176 | add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale); |
| 3177 | |
| 3178 | if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){ |
| 3179 | if (s->chroma_y_shift){ |
| 3180 | add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale); |
| 3181 | add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale); |
| 3182 | }else{ |
| 3183 | dct_linesize >>= 1; |
| 3184 | dct_offset >>=1; |
| 3185 | add_dequant_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale); |
| 3186 | add_dequant_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale); |
| 3187 | add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale); |
| 3188 | add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale); |
| 3189 | } |
| 3190 | } |
| 3191 | } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){ |
| 3192 | add_dct(s, block[0], 0, dest_y , dct_linesize); |
| 3193 | add_dct(s, block[1], 1, dest_y + block_size, dct_linesize); |
| 3194 | add_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize); |
| 3195 | add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize); |
| 3196 | |
| 3197 | if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){ |
| 3198 | if(s->chroma_y_shift){//Chroma420 |
| 3199 | add_dct(s, block[4], 4, dest_cb, uvlinesize); |
| 3200 | add_dct(s, block[5], 5, dest_cr, uvlinesize); |
| 3201 | }else{ |
| 3202 | //chroma422 |
| 3203 | dct_linesize = uvlinesize << s->interlaced_dct; |
| 3204 | dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size; |
| 3205 | |
| 3206 | add_dct(s, block[4], 4, dest_cb, dct_linesize); |
| 3207 | add_dct(s, block[5], 5, dest_cr, dct_linesize); |
| 3208 | add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize); |
| 3209 | add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize); |
| 3210 | if(!s->chroma_x_shift){//Chroma444 |
| 3211 | add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize); |
| 3212 | add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize); |
| 3213 | add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize); |
| 3214 | add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize); |
| 3215 | } |
| 3216 | } |
| 3217 | }//fi gray |
| 3218 | } |
| 3219 | else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) { |
| 3220 | ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr); |
| 3221 | } |
| 3222 | } else { |
| 3223 | /* dct only in intra block */ |
| 3224 | if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){ |
| 3225 | put_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale); |
| 3226 | put_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale); |
| 3227 | put_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale); |
| 3228 | put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale); |
| 3229 | |
| 3230 | if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){ |
| 3231 | if(s->chroma_y_shift){ |
| 3232 | put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale); |
| 3233 | put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale); |
| 3234 | }else{ |
| 3235 | dct_offset >>=1; |
| 3236 | dct_linesize >>=1; |
| 3237 | put_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale); |
| 3238 | put_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale); |
| 3239 | put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale); |
| 3240 | put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale); |
| 3241 | } |
| 3242 | } |
| 3243 | }else{ |
| 3244 | s->idsp.idct_put(dest_y, dct_linesize, block[0]); |
| 3245 | s->idsp.idct_put(dest_y + block_size, dct_linesize, block[1]); |
| 3246 | s->idsp.idct_put(dest_y + dct_offset, dct_linesize, block[2]); |
| 3247 | s->idsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]); |
| 3248 | |
| 3249 | if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){ |
| 3250 | if(s->chroma_y_shift){ |
| 3251 | s->idsp.idct_put(dest_cb, uvlinesize, block[4]); |
| 3252 | s->idsp.idct_put(dest_cr, uvlinesize, block[5]); |
| 3253 | }else{ |
| 3254 | |
| 3255 | dct_linesize = uvlinesize << s->interlaced_dct; |
| 3256 | dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size; |
| 3257 | |
| 3258 | s->idsp.idct_put(dest_cb, dct_linesize, block[4]); |
| 3259 | s->idsp.idct_put(dest_cr, dct_linesize, block[5]); |
| 3260 | s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]); |
| 3261 | s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]); |
| 3262 | if(!s->chroma_x_shift){//Chroma444 |
| 3263 | s->idsp.idct_put(dest_cb + block_size, dct_linesize, block[8]); |
| 3264 | s->idsp.idct_put(dest_cr + block_size, dct_linesize, block[9]); |
| 3265 | s->idsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]); |
| 3266 | s->idsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]); |
| 3267 | } |
| 3268 | } |
| 3269 | }//gray |
| 3270 | } |
| 3271 | } |
| 3272 | skip_idct: |
| 3273 | if(!readable){ |
| 3274 | s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y , linesize,16); |
| 3275 | s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift); |
| 3276 | s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift); |
| 3277 | } |
| 3278 | } |
| 3279 | } |
| 3280 | |
| 3281 | void ff_mpv_decode_mb(MpegEncContext *s, int16_t block[12][64]) |
| 3282 | { |
| 3283 | #if !CONFIG_SMALL |
| 3284 | if(s->out_format == FMT_MPEG1) { |
| 3285 | if(s->avctx->lowres) mpv_decode_mb_internal(s, block, 1, 1); |
| 3286 | else mpv_decode_mb_internal(s, block, 0, 1); |
| 3287 | } else |
| 3288 | #endif |
| 3289 | if(s->avctx->lowres) mpv_decode_mb_internal(s, block, 1, 0); |
| 3290 | else mpv_decode_mb_internal(s, block, 0, 0); |
| 3291 | } |
| 3292 | |
| 3293 | void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h) |
| 3294 | { |
| 3295 | ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f, |
| 3296 | s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure, |
| 3297 | s->first_field, s->low_delay); |
| 3298 | } |
| 3299 | |
| 3300 | void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename |
| 3301 | const int linesize = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics |
| 3302 | const int uvlinesize = s->current_picture.f->linesize[1]; |
| 3303 | const int mb_size= 4 - s->avctx->lowres; |
| 3304 | |
| 3305 | s->block_index[0]= s->b8_stride*(s->mb_y*2 ) - 2 + s->mb_x*2; |
| 3306 | s->block_index[1]= s->b8_stride*(s->mb_y*2 ) - 1 + s->mb_x*2; |
| 3307 | s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2; |
| 3308 | s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2; |
| 3309 | s->block_index[4]= s->mb_stride*(s->mb_y + 1) + s->b8_stride*s->mb_height*2 + s->mb_x - 1; |
| 3310 | s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x - 1; |
| 3311 | //block_index is not used by mpeg2, so it is not affected by chroma_format |
| 3312 | |
| 3313 | s->dest[0] = s->current_picture.f->data[0] + ((s->mb_x - 1) << mb_size); |
| 3314 | s->dest[1] = s->current_picture.f->data[1] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift)); |
| 3315 | s->dest[2] = s->current_picture.f->data[2] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift)); |
| 3316 | |
| 3317 | if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME)) |
| 3318 | { |
| 3319 | if(s->picture_structure==PICT_FRAME){ |
| 3320 | s->dest[0] += s->mb_y * linesize << mb_size; |
| 3321 | s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift); |
| 3322 | s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift); |
| 3323 | }else{ |
| 3324 | s->dest[0] += (s->mb_y>>1) * linesize << mb_size; |
| 3325 | s->dest[1] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift); |
| 3326 | s->dest[2] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift); |
| 3327 | av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD)); |
| 3328 | } |
| 3329 | } |
| 3330 | } |
| 3331 | |
| 3332 | /** |
| 3333 | * Permute an 8x8 block. |
| 3334 | * @param block the block which will be permuted according to the given permutation vector |
| 3335 | * @param permutation the permutation vector |
| 3336 | * @param last the last non zero coefficient in scantable order, used to speed the permutation up |
| 3337 | * @param scantable the used scantable, this is only used to speed the permutation up, the block is not |
| 3338 | * (inverse) permutated to scantable order! |
| 3339 | */ |
| 3340 | void ff_block_permute(int16_t *block, uint8_t *permutation, const uint8_t *scantable, int last) |
| 3341 | { |
| 3342 | int i; |
| 3343 | int16_t temp[64]; |
| 3344 | |
| 3345 | if(last<=0) return; |
| 3346 | //if(permutation[1]==1) return; //FIXME it is ok but not clean and might fail for some permutations |
| 3347 | |
| 3348 | for(i=0; i<=last; i++){ |
| 3349 | const int j= scantable[i]; |
| 3350 | temp[j]= block[j]; |
| 3351 | block[j]=0; |
| 3352 | } |
| 3353 | |
| 3354 | for(i=0; i<=last; i++){ |
| 3355 | const int j= scantable[i]; |
| 3356 | const int perm_j= permutation[j]; |
| 3357 | block[perm_j]= temp[j]; |
| 3358 | } |
| 3359 | } |
| 3360 | |
| 3361 | void ff_mpeg_flush(AVCodecContext *avctx){ |
| 3362 | int i; |
| 3363 | MpegEncContext *s = avctx->priv_data; |
| 3364 | |
| 3365 | if (!s || !s->picture) |
| 3366 | return; |
| 3367 | |
| 3368 | for (i = 0; i < MAX_PICTURE_COUNT; i++) |
| 3369 | ff_mpeg_unref_picture(s, &s->picture[i]); |
| 3370 | s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL; |
| 3371 | |
| 3372 | ff_mpeg_unref_picture(s, &s->current_picture); |
| 3373 | ff_mpeg_unref_picture(s, &s->last_picture); |
| 3374 | ff_mpeg_unref_picture(s, &s->next_picture); |
| 3375 | |
| 3376 | s->mb_x= s->mb_y= 0; |
| 3377 | s->closed_gop= 0; |
| 3378 | |
| 3379 | s->parse_context.state= -1; |
| 3380 | s->parse_context.frame_start_found= 0; |
| 3381 | s->parse_context.overread= 0; |
| 3382 | s->parse_context.overread_index= 0; |
| 3383 | s->parse_context.index= 0; |
| 3384 | s->parse_context.last_index= 0; |
| 3385 | s->bitstream_buffer_size=0; |
| 3386 | s->pp_time=0; |
| 3387 | } |
| 3388 | |
| 3389 | /** |
| 3390 | * set qscale and update qscale dependent variables. |
| 3391 | */ |
| 3392 | void ff_set_qscale(MpegEncContext * s, int qscale) |
| 3393 | { |
| 3394 | if (qscale < 1) |
| 3395 | qscale = 1; |
| 3396 | else if (qscale > 31) |
| 3397 | qscale = 31; |
| 3398 | |
| 3399 | s->qscale = qscale; |
| 3400 | s->chroma_qscale= s->chroma_qscale_table[qscale]; |
| 3401 | |
| 3402 | s->y_dc_scale= s->y_dc_scale_table[ qscale ]; |
| 3403 | s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ]; |
| 3404 | } |
| 3405 | |
| 3406 | void ff_mpv_report_decode_progress(MpegEncContext *s) |
| 3407 | { |
| 3408 | if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred) |
| 3409 | ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0); |
| 3410 | } |