| File: | root/firefox-clang/media/ffvpx/libavcodec/vp8.c |
| Warning: | line 1894, column 9 Branch condition evaluates to a garbage value |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | /* | |||
| 2 | * VP7/VP8 compatible video decoder | |||
| 3 | * | |||
| 4 | * Copyright (C) 2010 David Conrad | |||
| 5 | * Copyright (C) 2010 Ronald S. Bultje | |||
| 6 | * Copyright (C) 2010 Fiona Glaser | |||
| 7 | * Copyright (C) 2012 Daniel Kang | |||
| 8 | * Copyright (C) 2014 Peter Ross | |||
| 9 | * | |||
| 10 | * This file is part of FFmpeg. | |||
| 11 | * | |||
| 12 | * FFmpeg is free software; you can redistribute it and/or | |||
| 13 | * modify it under the terms of the GNU Lesser General Public | |||
| 14 | * License as published by the Free Software Foundation; either | |||
| 15 | * version 2.1 of the License, or (at your option) any later version. | |||
| 16 | * | |||
| 17 | * FFmpeg is distributed in the hope that it will be useful, | |||
| 18 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |||
| 19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |||
| 20 | * Lesser General Public License for more details. | |||
| 21 | * | |||
| 22 | * You should have received a copy of the GNU Lesser General Public | |||
| 23 | * License along with FFmpeg; if not, write to the Free Software | |||
| 24 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |||
| 25 | */ | |||
| 26 | ||||
| 27 | #include "config_components.h" | |||
| 28 | ||||
| 29 | #include "libavutil/mem.h" | |||
| 30 | #include "libavutil/mem_internal.h" | |||
| 31 | ||||
| 32 | #include "avcodec.h" | |||
| 33 | #include "codec_internal.h" | |||
| 34 | #include "decode.h" | |||
| 35 | #include "hwaccel_internal.h" | |||
| 36 | #include "hwconfig.h" | |||
| 37 | #include "mathops.h" | |||
| 38 | #include "progressframe.h" | |||
| 39 | #include "libavutil/refstruct.h" | |||
| 40 | #include "thread.h" | |||
| 41 | #include "vp8.h" | |||
| 42 | #include "vp89_rac.h" | |||
| 43 | #include "vp8data.h" | |||
| 44 | #include "vpx_rac.h" | |||
| 45 | ||||
| 46 | #if ARCH_ARM0 | |||
| 47 | # include "arm/vp8.h" | |||
| 48 | #endif | |||
| 49 | ||||
| 50 | // fixme: add 1 bit to all the calls to this? | |||
| 51 | static int vp8_rac_get_sint(VPXRangeCoder *c, int bits) | |||
| 52 | { | |||
| 53 | int v; | |||
| 54 | ||||
| 55 | if (!vp89_rac_get(c)) | |||
| 56 | return 0; | |||
| 57 | ||||
| 58 | v = vp89_rac_get_uint(c, bits); | |||
| 59 | ||||
| 60 | if (vp89_rac_get(c)) | |||
| 61 | v = -v; | |||
| 62 | ||||
| 63 | return v; | |||
| 64 | } | |||
| 65 | ||||
| 66 | static int vp8_rac_get_nn(VPXRangeCoder *c) | |||
| 67 | { | |||
| 68 | int v = vp89_rac_get_uint(c, 7) << 1; | |||
| 69 | return v + !v; | |||
| 70 | } | |||
| 71 | ||||
| 72 | // DCTextra | |||
| 73 | static int vp8_rac_get_coeff(VPXRangeCoder *c, const uint8_t *prob) | |||
| 74 | { | |||
| 75 | int v = 0; | |||
| 76 | ||||
| 77 | do { | |||
| 78 | v = (v<<1) + vpx_rac_get_probvpx_rac_get_prob(c, *prob++); | |||
| 79 | } while (*prob); | |||
| 80 | ||||
| 81 | return v; | |||
| 82 | } | |||
| 83 | ||||
| 84 | static void free_buffers(VP8Context *s) | |||
| 85 | { | |||
| 86 | int i; | |||
| 87 | if (s->thread_data) | |||
| 88 | for (i = 0; i < MAX_THREADS8; i++) { | |||
| 89 | #if HAVE_THREADS1 | |||
| 90 | pthread_cond_destroystrict_pthread_cond_destroy(&s->thread_data[i].cond); | |||
| 91 | pthread_mutex_destroystrict_pthread_mutex_destroy(&s->thread_data[i].lock); | |||
| 92 | #endif | |||
| 93 | av_freep(&s->thread_data[i].filter_strength); | |||
| 94 | } | |||
| 95 | av_freep(&s->thread_data); | |||
| 96 | av_freep(&s->macroblocks_base); | |||
| 97 | av_freep(&s->intra4x4_pred_mode_top); | |||
| 98 | av_freep(&s->top_nnz); | |||
| 99 | av_freep(&s->top_border); | |||
| 100 | ||||
| 101 | s->macroblocks = NULL((void*)0); | |||
| 102 | } | |||
| 103 | ||||
| 104 | static int vp8_alloc_frame(VP8Context *s, VP8Frame *f, int ref) | |||
| 105 | { | |||
| 106 | int ret = ff_progress_frame_get_buffer(s->avctx, &f->tf, | |||
| 107 | ref ? AV_GET_BUFFER_FLAG_REF(1 << 0) : 0); | |||
| 108 | if (ret < 0) | |||
| 109 | return ret; | |||
| 110 | f->seg_map = av_refstruct_allocz(s->mb_width * s->mb_height); | |||
| 111 | if (!f->seg_map) { | |||
| 112 | ret = AVERROR(ENOMEM)(-(12)); | |||
| 113 | goto fail; | |||
| 114 | } | |||
| 115 | ret = ff_hwaccel_frame_priv_alloc(s->avctx, &f->hwaccel_picture_private); | |||
| 116 | if (ret < 0) | |||
| 117 | goto fail; | |||
| 118 | ||||
| 119 | return 0; | |||
| 120 | ||||
| 121 | fail: | |||
| 122 | av_refstruct_unref(&f->seg_map); | |||
| 123 | ff_progress_frame_unref(&f->tf); | |||
| 124 | return ret; | |||
| 125 | } | |||
| 126 | ||||
| 127 | static void vp8_release_frame(VP8Frame *f) | |||
| 128 | { | |||
| 129 | av_refstruct_unref(&f->seg_map); | |||
| 130 | av_refstruct_unref(&f->hwaccel_picture_private); | |||
| 131 | ff_progress_frame_unref(&f->tf); | |||
| 132 | } | |||
| 133 | ||||
| 134 | static av_cold__attribute__((cold)) void vp8_decode_flush_impl(AVCodecContext *avctx, int free_mem) | |||
| 135 | { | |||
| 136 | VP8Context *s = avctx->priv_data; | |||
| 137 | int i; | |||
| 138 | ||||
| 139 | for (i = 0; i < FF_ARRAY_ELEMS(s->frames)(sizeof(s->frames) / sizeof((s->frames)[0])); i++) | |||
| 140 | vp8_release_frame(&s->frames[i]); | |||
| 141 | memset(s->framep, 0, sizeof(s->framep)); | |||
| 142 | ||||
| 143 | if (free_mem) | |||
| 144 | free_buffers(s); | |||
| 145 | ||||
| 146 | if (FF_HW_HAS_CB(avctx, flush)((avctx)->hwaccel && ffhwaccel((avctx)->hwaccel )->flush)) | |||
| 147 | FF_HW_SIMPLE_CALL(avctx, flush)(ffhwaccel((avctx)->hwaccel)->flush(avctx)); | |||
| 148 | } | |||
| 149 | ||||
| 150 | static av_cold__attribute__((cold)) void vp8_decode_flush(AVCodecContext *avctx) | |||
| 151 | { | |||
| 152 | vp8_decode_flush_impl(avctx, 0); | |||
| 153 | } | |||
| 154 | ||||
| 155 | static VP8Frame *vp8_find_free_buffer(VP8Context *s) | |||
| 156 | { | |||
| 157 | VP8Frame *frame = NULL((void*)0); | |||
| 158 | int i; | |||
| 159 | ||||
| 160 | // find a free buffer | |||
| 161 | for (i = 0; i < 5; i++) | |||
| 162 | if (&s->frames[i] != s->framep[VP8_FRAME_CURRENT] && | |||
| 163 | &s->frames[i] != s->framep[VP8_FRAME_PREVIOUS] && | |||
| 164 | &s->frames[i] != s->framep[VP8_FRAME_GOLDEN] && | |||
| 165 | &s->frames[i] != s->framep[VP8_FRAME_ALTREF]) { | |||
| 166 | frame = &s->frames[i]; | |||
| 167 | break; | |||
| 168 | } | |||
| 169 | if (i == 5) { | |||
| 170 | av_log(s->avctx, AV_LOG_FATAL8, "Ran out of free frames!\n"); | |||
| 171 | abort(); | |||
| 172 | } | |||
| 173 | if (frame->tf.f) | |||
| 174 | vp8_release_frame(frame); | |||
| 175 | ||||
| 176 | return frame; | |||
| 177 | } | |||
| 178 | ||||
| 179 | static enum AVPixelFormat get_pixel_format(VP8Context *s) | |||
| 180 | { | |||
| 181 | enum AVPixelFormat pix_fmts[] = { | |||
| 182 | #if CONFIG_VP8_VAAPI_HWACCEL1 | |||
| 183 | AV_PIX_FMT_VAAPI, | |||
| 184 | #endif | |||
| 185 | #if CONFIG_VP8_NVDEC_HWACCEL0 | |||
| 186 | AV_PIX_FMT_CUDA, | |||
| 187 | #endif | |||
| 188 | AV_PIX_FMT_YUV420P, | |||
| 189 | AV_PIX_FMT_NONE, | |||
| 190 | }; | |||
| 191 | ||||
| 192 | return ff_get_format(s->avctx, pix_fmts); | |||
| 193 | } | |||
| 194 | ||||
| 195 | static av_always_inline__attribute__((always_inline)) inline | |||
| 196 | int update_dimensions(VP8Context *s, int width, int height, int is_vp7) | |||
| 197 | { | |||
| 198 | AVCodecContext *avctx = s->avctx; | |||
| 199 | int i, ret, dim_reset = 0; | |||
| 200 | ||||
| 201 | if (width != s->avctx->width || ((width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height) && s->macroblocks_base || | |||
| 202 | height != s->avctx->height) { | |||
| 203 | vp8_decode_flush_impl(s->avctx, 1); | |||
| 204 | ||||
| 205 | ret = ff_set_dimensions(s->avctx, width, height); | |||
| 206 | if (ret < 0) | |||
| 207 | return ret; | |||
| 208 | ||||
| 209 | dim_reset = (s->macroblocks_base != NULL((void*)0)); | |||
| 210 | } | |||
| 211 | ||||
| 212 | if ((s->pix_fmt == AV_PIX_FMT_NONE || dim_reset) && | |||
| 213 | !s->actually_webp && !is_vp7) { | |||
| 214 | s->pix_fmt = get_pixel_format(s); | |||
| 215 | if (s->pix_fmt < 0) | |||
| 216 | return AVERROR(EINVAL)(-(22)); | |||
| 217 | avctx->pix_fmt = s->pix_fmt; | |||
| 218 | } | |||
| 219 | ||||
| 220 | s->mb_width = (s->avctx->coded_width + 15) / 16; | |||
| 221 | s->mb_height = (s->avctx->coded_height + 15) / 16; | |||
| 222 | ||||
| 223 | s->mb_layout = is_vp7 || avctx->active_thread_type == FF_THREAD_SLICE2 && | |||
| 224 | avctx->thread_count > 1; | |||
| 225 | if (!s->mb_layout) { // Frame threading and one thread | |||
| 226 | s->macroblocks_base = av_mallocz((s->mb_width + s->mb_height * 2 + 1) * | |||
| 227 | sizeof(*s->macroblocks)); | |||
| 228 | s->intra4x4_pred_mode_top = av_mallocz(s->mb_width * 4); | |||
| 229 | } else // Sliced threading | |||
| 230 | s->macroblocks_base = av_mallocz((s->mb_width + 2) * (s->mb_height + 2) * | |||
| 231 | sizeof(*s->macroblocks)); | |||
| 232 | s->top_nnz = av_mallocz(s->mb_width * sizeof(*s->top_nnz)); | |||
| 233 | s->top_border = av_mallocz((s->mb_width + 1) * sizeof(*s->top_border)); | |||
| 234 | s->thread_data = av_mallocz(MAX_THREADS8 * sizeof(VP8ThreadData)); | |||
| 235 | ||||
| 236 | if (!s->macroblocks_base || !s->top_nnz || !s->top_border || | |||
| 237 | !s->thread_data || (!s->intra4x4_pred_mode_top && !s->mb_layout)) { | |||
| 238 | free_buffers(s); | |||
| 239 | return AVERROR(ENOMEM)(-(12)); | |||
| 240 | } | |||
| 241 | ||||
| 242 | for (i = 0; i < MAX_THREADS8; i++) { | |||
| 243 | s->thread_data[i].filter_strength = | |||
| 244 | av_mallocz(s->mb_width * sizeof(*s->thread_data[0].filter_strength)); | |||
| 245 | if (!s->thread_data[i].filter_strength) { | |||
| 246 | free_buffers(s); | |||
| 247 | return AVERROR(ENOMEM)(-(12)); | |||
| 248 | } | |||
| 249 | #if HAVE_THREADS1 | |||
| 250 | ret = pthread_mutex_initstrict_pthread_mutex_init(&s->thread_data[i].lock, NULL((void*)0)); | |||
| 251 | if (ret) { | |||
| 252 | free_buffers(s); | |||
| 253 | return AVERROR(ret)(-(ret)); | |||
| 254 | } | |||
| 255 | ret = pthread_cond_initstrict_pthread_cond_init(&s->thread_data[i].cond, NULL((void*)0)); | |||
| 256 | if (ret) { | |||
| 257 | free_buffers(s); | |||
| 258 | return AVERROR(ret)(-(ret)); | |||
| 259 | } | |||
| 260 | #endif | |||
| 261 | } | |||
| 262 | ||||
| 263 | s->macroblocks = s->macroblocks_base + 1; | |||
| 264 | ||||
| 265 | return 0; | |||
| 266 | } | |||
| 267 | ||||
| 268 | static int vp7_update_dimensions(VP8Context *s, int width, int height) | |||
| 269 | { | |||
| 270 | return update_dimensions(s, width, height, IS_VP71); | |||
| 271 | } | |||
| 272 | ||||
| 273 | static int vp8_update_dimensions(VP8Context *s, int width, int height) | |||
| 274 | { | |||
| 275 | return update_dimensions(s, width, height, IS_VP80); | |||
| 276 | } | |||
| 277 | ||||
| 278 | ||||
| 279 | static void parse_segment_info(VP8Context *s) | |||
| 280 | { | |||
| 281 | VPXRangeCoder *c = &s->c; | |||
| 282 | int i; | |||
| 283 | ||||
| 284 | s->segmentation.update_map = vp89_rac_get(c); | |||
| 285 | s->segmentation.update_feature_data = vp89_rac_get(c); | |||
| 286 | ||||
| 287 | if (s->segmentation.update_feature_data) { | |||
| 288 | s->segmentation.absolute_vals = vp89_rac_get(c); | |||
| 289 | ||||
| 290 | for (i = 0; i < 4; i++) | |||
| 291 | s->segmentation.base_quant[i] = vp8_rac_get_sint(c, 7); | |||
| 292 | ||||
| 293 | for (i = 0; i < 4; i++) | |||
| 294 | s->segmentation.filter_level[i] = vp8_rac_get_sint(c, 6); | |||
| 295 | } | |||
| 296 | if (s->segmentation.update_map) | |||
| 297 | for (i = 0; i < 3; i++) | |||
| 298 | s->prob->segmentid[i] = vp89_rac_get(c) ? vp89_rac_get_uint(c, 8) : 255; | |||
| 299 | } | |||
| 300 | ||||
| 301 | static void update_lf_deltas(VP8Context *s) | |||
| 302 | { | |||
| 303 | VPXRangeCoder *c = &s->c; | |||
| 304 | int i; | |||
| 305 | ||||
| 306 | for (i = 0; i < 4; i++) { | |||
| 307 | if (vp89_rac_get(c)) { | |||
| 308 | s->lf_delta.ref[i] = vp89_rac_get_uint(c, 6); | |||
| 309 | ||||
| 310 | if (vp89_rac_get(c)) | |||
| 311 | s->lf_delta.ref[i] = -s->lf_delta.ref[i]; | |||
| 312 | } | |||
| 313 | } | |||
| 314 | ||||
| 315 | for (i = MODE_I4x44; i <= VP8_MVMODE_SPLIT; i++) { | |||
| 316 | if (vp89_rac_get(c)) { | |||
| 317 | s->lf_delta.mode[i] = vp89_rac_get_uint(c, 6); | |||
| 318 | ||||
| 319 | if (vp89_rac_get(c)) | |||
| 320 | s->lf_delta.mode[i] = -s->lf_delta.mode[i]; | |||
| 321 | } | |||
| 322 | } | |||
| 323 | } | |||
| 324 | ||||
| 325 | static int setup_partitions(VP8Context *s, const uint8_t *buf, int buf_size) | |||
| 326 | { | |||
| 327 | const uint8_t *sizes = buf; | |||
| 328 | int i; | |||
| 329 | int ret; | |||
| 330 | ||||
| 331 | s->num_coeff_partitions = 1 << vp89_rac_get_uint(&s->c, 2); | |||
| 332 | ||||
| 333 | buf += 3 * (s->num_coeff_partitions - 1); | |||
| 334 | buf_size -= 3 * (s->num_coeff_partitions - 1); | |||
| 335 | if (buf_size < 0) | |||
| 336 | return -1; | |||
| 337 | ||||
| 338 | for (i = 0; i < s->num_coeff_partitions - 1; i++) { | |||
| 339 | int size = AV_RL24(sizes + 3 * i)((((const uint8_t*)(sizes + 3 * i))[2] << 16) | (((const uint8_t*)(sizes + 3 * i))[1] << 8) | ((const uint8_t*) (sizes + 3 * i))[0]); | |||
| 340 | if (buf_size - size < 0) | |||
| 341 | return -1; | |||
| 342 | s->coeff_partition_size[i] = size; | |||
| 343 | ||||
| 344 | ret = ff_vpx_init_range_decoder(&s->coeff_partition[i], buf, size); | |||
| 345 | if (ret < 0) | |||
| 346 | return ret; | |||
| 347 | buf += size; | |||
| 348 | buf_size -= size; | |||
| 349 | } | |||
| 350 | ||||
| 351 | s->coeff_partition_size[i] = buf_size; | |||
| 352 | ||||
| 353 | return ff_vpx_init_range_decoder(&s->coeff_partition[i], buf, buf_size); | |||
| 354 | } | |||
| 355 | ||||
| 356 | static void vp7_get_quants(VP8Context *s) | |||
| 357 | { | |||
| 358 | VPXRangeCoder *c = &s->c; | |||
| 359 | ||||
| 360 | int yac_qi = vp89_rac_get_uint(c, 7); | |||
| 361 | int ydc_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi; | |||
| 362 | int y2dc_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi; | |||
| 363 | int y2ac_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi; | |||
| 364 | int uvdc_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi; | |||
| 365 | int uvac_qi = vp89_rac_get(c) ? vp89_rac_get_uint(c, 7) : yac_qi; | |||
| 366 | ||||
| 367 | s->qmat[0].luma_qmul[0] = vp7_ydc_qlookup[ydc_qi]; | |||
| 368 | s->qmat[0].luma_qmul[1] = vp7_yac_qlookup[yac_qi]; | |||
| 369 | s->qmat[0].luma_dc_qmul[0] = vp7_y2dc_qlookup[y2dc_qi]; | |||
| 370 | s->qmat[0].luma_dc_qmul[1] = vp7_y2ac_qlookup[y2ac_qi]; | |||
| 371 | s->qmat[0].chroma_qmul[0] = FFMIN(vp7_ydc_qlookup[uvdc_qi], 132)((vp7_ydc_qlookup[uvdc_qi]) > (132) ? (132) : (vp7_ydc_qlookup [uvdc_qi])); | |||
| 372 | s->qmat[0].chroma_qmul[1] = vp7_yac_qlookup[uvac_qi]; | |||
| 373 | } | |||
| 374 | ||||
| 375 | static void vp8_get_quants(VP8Context *s) | |||
| 376 | { | |||
| 377 | VPXRangeCoder *c = &s->c; | |||
| 378 | int i, base_qi; | |||
| 379 | ||||
| 380 | s->quant.yac_qi = vp89_rac_get_uint(c, 7); | |||
| 381 | s->quant.ydc_delta = vp8_rac_get_sint(c, 4); | |||
| 382 | s->quant.y2dc_delta = vp8_rac_get_sint(c, 4); | |||
| 383 | s->quant.y2ac_delta = vp8_rac_get_sint(c, 4); | |||
| 384 | s->quant.uvdc_delta = vp8_rac_get_sint(c, 4); | |||
| 385 | s->quant.uvac_delta = vp8_rac_get_sint(c, 4); | |||
| 386 | ||||
| 387 | for (i = 0; i < 4; i++) { | |||
| 388 | if (s->segmentation.enabled) { | |||
| 389 | base_qi = s->segmentation.base_quant[i]; | |||
| 390 | if (!s->segmentation.absolute_vals) | |||
| 391 | base_qi += s->quant.yac_qi; | |||
| 392 | } else | |||
| 393 | base_qi = s->quant.yac_qi; | |||
| 394 | ||||
| 395 | s->qmat[i].luma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2av_clip_uintp2_c(base_qi + s->quant.ydc_delta, 7)]; | |||
| 396 | s->qmat[i].luma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2av_clip_uintp2_c(base_qi, 7)]; | |||
| 397 | s->qmat[i].luma_dc_qmul[0] = vp8_dc_qlookup[av_clip_uintp2av_clip_uintp2_c(base_qi + s->quant.y2dc_delta, 7)] * 2; | |||
| 398 | /* 101581>>16 is equivalent to 155/100 */ | |||
| 399 | s->qmat[i].luma_dc_qmul[1] = vp8_ac_qlookup[av_clip_uintp2av_clip_uintp2_c(base_qi + s->quant.y2ac_delta, 7)] * 101581 >> 16; | |||
| 400 | s->qmat[i].chroma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2av_clip_uintp2_c(base_qi + s->quant.uvdc_delta, 7)]; | |||
| 401 | s->qmat[i].chroma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2av_clip_uintp2_c(base_qi + s->quant.uvac_delta, 7)]; | |||
| 402 | ||||
| 403 | s->qmat[i].luma_dc_qmul[1] = FFMAX(s->qmat[i].luma_dc_qmul[1], 8)((s->qmat[i].luma_dc_qmul[1]) > (8) ? (s->qmat[i].luma_dc_qmul [1]) : (8)); | |||
| 404 | s->qmat[i].chroma_qmul[0] = FFMIN(s->qmat[i].chroma_qmul[0], 132)((s->qmat[i].chroma_qmul[0]) > (132) ? (132) : (s->qmat [i].chroma_qmul[0])); | |||
| 405 | } | |||
| 406 | } | |||
| 407 | ||||
| 408 | /** | |||
| 409 | * Determine which buffers golden and altref should be updated with after this frame. | |||
| 410 | * The spec isn't clear here, so I'm going by my understanding of what libvpx does | |||
| 411 | * | |||
| 412 | * Intra frames update all 3 references | |||
| 413 | * Inter frames update VP8_FRAME_PREVIOUS if the update_last flag is set | |||
| 414 | * If the update (golden|altref) flag is set, it's updated with the current frame | |||
| 415 | * if update_last is set, and VP8_FRAME_PREVIOUS otherwise. | |||
| 416 | * If the flag is not set, the number read means: | |||
| 417 | * 0: no update | |||
| 418 | * 1: VP8_FRAME_PREVIOUS | |||
| 419 | * 2: update golden with altref, or update altref with golden | |||
| 420 | */ | |||
| 421 | static VP8FrameType ref_to_update(VP8Context *s, int update, VP8FrameType ref) | |||
| 422 | { | |||
| 423 | VPXRangeCoder *c = &s->c; | |||
| 424 | ||||
| 425 | if (update) | |||
| 426 | return VP8_FRAME_CURRENT; | |||
| 427 | ||||
| 428 | switch (vp89_rac_get_uint(c, 2)) { | |||
| 429 | case 1: | |||
| 430 | return VP8_FRAME_PREVIOUS; | |||
| 431 | case 2: | |||
| 432 | return (ref == VP8_FRAME_GOLDEN) ? VP8_FRAME_ALTREF : VP8_FRAME_GOLDEN; | |||
| 433 | } | |||
| 434 | return VP8_FRAME_NONE; | |||
| 435 | } | |||
| 436 | ||||
| 437 | static void vp78_reset_probability_tables(VP8Context *s) | |||
| 438 | { | |||
| 439 | int i, j; | |||
| 440 | for (i = 0; i < 4; i++) | |||
| 441 | for (j = 0; j < 16; j++) | |||
| 442 | memcpy(s->prob->token[i][j], vp8_token_default_probs[i][vp8_coeff_band[j]], | |||
| 443 | sizeof(s->prob->token[i][j])); | |||
| 444 | } | |||
| 445 | ||||
| 446 | static void vp78_update_probability_tables(VP8Context *s) | |||
| 447 | { | |||
| 448 | VPXRangeCoder *c = &s->c; | |||
| 449 | int i, j, k, l, m; | |||
| 450 | ||||
| 451 | for (i = 0; i < 4; i++) | |||
| 452 | for (j = 0; j < 8; j++) | |||
| 453 | for (k = 0; k < 3; k++) | |||
| 454 | for (l = 0; l < NUM_DCT_TOKENS-1; l++) | |||
| 455 | if (vpx_rac_get_prob_branchy(c, ff_vp8_token_update_probs[i][j][k][l])) { | |||
| 456 | int prob = vp89_rac_get_uint(c, 8); | |||
| 457 | for (m = 0; vp8_coeff_band_indexes[j][m] >= 0; m++) | |||
| 458 | s->prob->token[i][vp8_coeff_band_indexes[j][m]][k][l] = prob; | |||
| 459 | } | |||
| 460 | } | |||
| 461 | ||||
| 462 | #define VP7_MVC_SIZE17 17 | |||
| 463 | #define VP8_MVC_SIZE19 19 | |||
| 464 | ||||
| 465 | static void vp78_update_pred16x16_pred8x8_mvc_probabilities(VP8Context *s, | |||
| 466 | int mvc_size) | |||
| 467 | { | |||
| 468 | VPXRangeCoder *c = &s->c; | |||
| 469 | int i, j; | |||
| 470 | ||||
| 471 | if (vp89_rac_get(c)) | |||
| 472 | for (i = 0; i < 4; i++) | |||
| 473 | s->prob->pred16x16[i] = vp89_rac_get_uint(c, 8); | |||
| 474 | if (vp89_rac_get(c)) | |||
| 475 | for (i = 0; i < 3; i++) | |||
| 476 | s->prob->pred8x8c[i] = vp89_rac_get_uint(c, 8); | |||
| 477 | ||||
| 478 | // 17.2 MV probability update | |||
| 479 | for (i = 0; i < 2; i++) | |||
| 480 | for (j = 0; j < mvc_size; j++) | |||
| 481 | if (vpx_rac_get_prob_branchy(c, vp8_mv_update_prob[i][j])) | |||
| 482 | s->prob->mvc[i][j] = vp8_rac_get_nn(c); | |||
| 483 | } | |||
| 484 | ||||
| 485 | static void update_refs(VP8Context *s) | |||
| 486 | { | |||
| 487 | VPXRangeCoder *c = &s->c; | |||
| 488 | ||||
| 489 | int update_golden = vp89_rac_get(c); | |||
| 490 | int update_altref = vp89_rac_get(c); | |||
| 491 | ||||
| 492 | s->update_golden = ref_to_update(s, update_golden, VP8_FRAME_GOLDEN); | |||
| 493 | s->update_altref = ref_to_update(s, update_altref, VP8_FRAME_ALTREF); | |||
| 494 | } | |||
| 495 | ||||
| 496 | static void copy_chroma(AVFrame *dst, const AVFrame *src, int width, int height) | |||
| 497 | { | |||
| 498 | int i, j; | |||
| 499 | ||||
| 500 | for (j = 1; j < 3; j++) { | |||
| 501 | for (i = 0; i < height / 2; i++) | |||
| 502 | memcpy(dst->data[j] + i * dst->linesize[j], | |||
| 503 | src->data[j] + i * src->linesize[j], width / 2); | |||
| 504 | } | |||
| 505 | } | |||
| 506 | ||||
| 507 | static void fade(uint8_t *dst, ptrdiff_t dst_linesize, | |||
| 508 | const uint8_t *src, ptrdiff_t src_linesize, | |||
| 509 | int width, int height, | |||
| 510 | int alpha, int beta) | |||
| 511 | { | |||
| 512 | int i, j; | |||
| 513 | for (j = 0; j < height; j++) { | |||
| 514 | const uint8_t *src2 = src + j * src_linesize; | |||
| 515 | uint8_t *dst2 = dst + j * dst_linesize; | |||
| 516 | for (i = 0; i < width; i++) { | |||
| 517 | uint8_t y = src2[i]; | |||
| 518 | dst2[i] = av_clip_uint8av_clip_uint8_c(y + ((y * beta) >> 8) + alpha); | |||
| 519 | } | |||
| 520 | } | |||
| 521 | } | |||
| 522 | ||||
| 523 | static int vp7_fade_frame(VP8Context *s, int alpha, int beta) | |||
| 524 | { | |||
| 525 | int ret; | |||
| 526 | ||||
| 527 | if (!s->keyframe && (alpha || beta)) { | |||
| 528 | int width = s->mb_width * 16; | |||
| 529 | int height = s->mb_height * 16; | |||
| 530 | const AVFrame *src; | |||
| 531 | AVFrame *dst; | |||
| 532 | ||||
| 533 | if (!s->framep[VP8_FRAME_PREVIOUS] || | |||
| 534 | !s->framep[VP8_FRAME_GOLDEN]) { | |||
| 535 | av_log(s->avctx, AV_LOG_WARNING24, "Discarding interframe without a prior keyframe!\n"); | |||
| 536 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 537 | } | |||
| 538 | ||||
| 539 | src = | |||
| 540 | dst = s->framep[VP8_FRAME_PREVIOUS]->tf.f; | |||
| 541 | ||||
| 542 | /* preserve the golden frame, write a new previous frame */ | |||
| 543 | if (s->framep[VP8_FRAME_GOLDEN] == s->framep[VP8_FRAME_PREVIOUS]) { | |||
| 544 | VP8Frame *prev_frame = vp8_find_free_buffer(s); | |||
| 545 | ||||
| 546 | ret = vp8_alloc_frame(s, prev_frame, 1); | |||
| 547 | if (ret < 0) | |||
| 548 | return ret; | |||
| 549 | s->framep[VP8_FRAME_PREVIOUS] = prev_frame; | |||
| 550 | ||||
| 551 | dst = s->framep[VP8_FRAME_PREVIOUS]->tf.f; | |||
| 552 | ||||
| 553 | copy_chroma(dst, src, width, height); | |||
| 554 | } | |||
| 555 | ||||
| 556 | fade(dst->data[0], dst->linesize[0], | |||
| 557 | src->data[0], src->linesize[0], | |||
| 558 | width, height, alpha, beta); | |||
| 559 | } | |||
| 560 | ||||
| 561 | return 0; | |||
| 562 | } | |||
| 563 | ||||
| 564 | static int vp7_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size) | |||
| 565 | { | |||
| 566 | VPXRangeCoder *c = &s->c; | |||
| 567 | int part1_size, hscale, vscale, i, j, ret; | |||
| 568 | int width = s->avctx->width; | |||
| 569 | int height = s->avctx->height; | |||
| 570 | int alpha = 0; | |||
| 571 | int beta = 0; | |||
| 572 | int fade_present = 1; | |||
| 573 | ||||
| 574 | if (buf_size < 4) { | |||
| 575 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 576 | } | |||
| 577 | ||||
| 578 | s->profile = (buf[0] >> 1) & 7; | |||
| 579 | if (s->profile > 1) { | |||
| 580 | avpriv_request_sample(s->avctx, "Unknown profile %d", s->profile); | |||
| 581 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 582 | } | |||
| 583 | ||||
| 584 | s->keyframe = !(buf[0] & 1); | |||
| 585 | s->invisible = 0; | |||
| 586 | part1_size = AV_RL24(buf)((((const uint8_t*)(buf))[2] << 16) | (((const uint8_t* )(buf))[1] << 8) | ((const uint8_t*)(buf))[0]) >> 4; | |||
| 587 | ||||
| 588 | if (buf_size < 4 - s->profile + part1_size) { | |||
| 589 | av_log(s->avctx, AV_LOG_ERROR16, "Buffer size %d is too small, needed : %d\n", buf_size, 4 - s->profile + part1_size); | |||
| 590 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 591 | } | |||
| 592 | ||||
| 593 | buf += 4 - s->profile; | |||
| 594 | buf_size -= 4 - s->profile; | |||
| 595 | ||||
| 596 | memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab, sizeof(s->put_pixels_tab)); | |||
| 597 | ||||
| 598 | ret = ff_vpx_init_range_decoder(c, buf, part1_size); | |||
| 599 | if (ret < 0) | |||
| 600 | return ret; | |||
| 601 | buf += part1_size; | |||
| 602 | buf_size -= part1_size; | |||
| 603 | ||||
| 604 | /* A. Dimension information (keyframes only) */ | |||
| 605 | if (s->keyframe) { | |||
| 606 | width = vp89_rac_get_uint(c, 12); | |||
| 607 | height = vp89_rac_get_uint(c, 12); | |||
| 608 | hscale = vp89_rac_get_uint(c, 2); | |||
| 609 | vscale = vp89_rac_get_uint(c, 2); | |||
| 610 | if (hscale || vscale) | |||
| 611 | avpriv_request_sample(s->avctx, "Upscaling"); | |||
| 612 | ||||
| 613 | s->update_golden = s->update_altref = VP8_FRAME_CURRENT; | |||
| 614 | vp78_reset_probability_tables(s); | |||
| 615 | memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter, | |||
| 616 | sizeof(s->prob->pred16x16)); | |||
| 617 | memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter, | |||
| 618 | sizeof(s->prob->pred8x8c)); | |||
| 619 | for (i = 0; i < 2; i++) | |||
| 620 | memcpy(s->prob->mvc[i], vp7_mv_default_prob[i], | |||
| 621 | sizeof(vp7_mv_default_prob[i])); | |||
| 622 | memset(&s->segmentation, 0, sizeof(s->segmentation)); | |||
| 623 | memset(&s->lf_delta, 0, sizeof(s->lf_delta)); | |||
| 624 | memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan)); | |||
| 625 | } | |||
| 626 | ||||
| 627 | if (s->keyframe || s->profile > 0) | |||
| 628 | memset(s->inter_dc_pred, 0 , sizeof(s->inter_dc_pred)); | |||
| 629 | ||||
| 630 | /* B. Decoding information for all four macroblock-level features */ | |||
| 631 | for (i = 0; i < 4; i++) { | |||
| 632 | s->feature_enabled[i] = vp89_rac_get(c); | |||
| 633 | if (s->feature_enabled[i]) { | |||
| 634 | s->feature_present_prob[i] = vp89_rac_get_uint(c, 8); | |||
| 635 | ||||
| 636 | for (j = 0; j < 3; j++) | |||
| 637 | s->feature_index_prob[i][j] = | |||
| 638 | vp89_rac_get(c) ? vp89_rac_get_uint(c, 8) : 255; | |||
| 639 | ||||
| 640 | if (vp7_feature_value_size[s->profile][i]) | |||
| 641 | for (j = 0; j < 4; j++) | |||
| 642 | s->feature_value[i][j] = | |||
| 643 | vp89_rac_get(c) ? vp89_rac_get_uint(c, vp7_feature_value_size[s->profile][i]) : 0; | |||
| 644 | } | |||
| 645 | } | |||
| 646 | ||||
| 647 | s->segmentation.enabled = 0; | |||
| 648 | s->segmentation.update_map = 0; | |||
| 649 | s->lf_delta.enabled = 0; | |||
| 650 | ||||
| 651 | s->num_coeff_partitions = 1; | |||
| 652 | ret = ff_vpx_init_range_decoder(&s->coeff_partition[0], buf, buf_size); | |||
| 653 | if (ret < 0) | |||
| 654 | return ret; | |||
| 655 | ||||
| 656 | if (!s->macroblocks_base || /* first frame */ | |||
| 657 | width != s->avctx->width || height != s->avctx->height || | |||
| 658 | (width + 15) / 16 != s->mb_width || (height + 15) / 16 != s->mb_height) { | |||
| 659 | if ((ret = vp7_update_dimensions(s, width, height)) < 0) | |||
| 660 | return ret; | |||
| 661 | } | |||
| 662 | ||||
| 663 | /* C. Dequantization indices */ | |||
| 664 | vp7_get_quants(s); | |||
| 665 | ||||
| 666 | /* D. Golden frame update flag (a Flag) for interframes only */ | |||
| 667 | if (!s->keyframe) { | |||
| 668 | s->update_golden = vp89_rac_get(c) ? VP8_FRAME_CURRENT : VP8_FRAME_NONE; | |||
| 669 | s->sign_bias[VP8_FRAME_GOLDEN] = 0; | |||
| 670 | } | |||
| 671 | ||||
| 672 | s->update_last = 1; | |||
| 673 | s->update_probabilities = 1; | |||
| 674 | ||||
| 675 | if (s->profile > 0) { | |||
| 676 | s->update_probabilities = vp89_rac_get(c); | |||
| 677 | if (!s->update_probabilities) | |||
| 678 | s->prob[1] = s->prob[0]; | |||
| 679 | ||||
| 680 | if (!s->keyframe) | |||
| 681 | fade_present = vp89_rac_get(c); | |||
| 682 | } | |||
| 683 | ||||
| 684 | if (vpx_rac_is_end(c)) | |||
| 685 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 686 | /* E. Fading information for previous frame */ | |||
| 687 | if (fade_present && vp89_rac_get(c)) { | |||
| 688 | alpha = (int8_t) vp89_rac_get_uint(c, 8); | |||
| 689 | beta = (int8_t) vp89_rac_get_uint(c, 8); | |||
| 690 | } | |||
| 691 | ||||
| 692 | /* F. Loop filter type */ | |||
| 693 | if (!s->profile) | |||
| 694 | s->filter.simple = vp89_rac_get(c); | |||
| 695 | ||||
| 696 | /* G. DCT coefficient ordering specification */ | |||
| 697 | if (vp89_rac_get(c)) | |||
| 698 | for (i = 1; i < 16; i++) | |||
| 699 | s->prob[0].scan[i] = ff_zigzag_scan[vp89_rac_get_uint(c, 4)]; | |||
| 700 | ||||
| 701 | /* H. Loop filter levels */ | |||
| 702 | if (s->profile > 0) | |||
| 703 | s->filter.simple = vp89_rac_get(c); | |||
| 704 | s->filter.level = vp89_rac_get_uint(c, 6); | |||
| 705 | s->filter.sharpness = vp89_rac_get_uint(c, 3); | |||
| 706 | ||||
| 707 | /* I. DCT coefficient probability update; 13.3 Token Probability Updates */ | |||
| 708 | vp78_update_probability_tables(s); | |||
| 709 | ||||
| 710 | s->mbskip_enabled = 0; | |||
| 711 | ||||
| 712 | /* J. The remaining frame header data occurs ONLY FOR INTERFRAMES */ | |||
| 713 | if (!s->keyframe) { | |||
| 714 | s->prob->intra = vp89_rac_get_uint(c, 8); | |||
| 715 | s->prob->last = vp89_rac_get_uint(c, 8); | |||
| 716 | vp78_update_pred16x16_pred8x8_mvc_probabilities(s, VP7_MVC_SIZE17); | |||
| 717 | } | |||
| 718 | ||||
| 719 | if (vpx_rac_is_end(c)) | |||
| 720 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 721 | ||||
| 722 | if ((ret = vp7_fade_frame(s, alpha, beta)) < 0) | |||
| 723 | return ret; | |||
| 724 | ||||
| 725 | return 0; | |||
| 726 | } | |||
| 727 | ||||
| 728 | static int vp8_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size) | |||
| 729 | { | |||
| 730 | VPXRangeCoder *c = &s->c; | |||
| 731 | int header_size, hscale, vscale, ret; | |||
| 732 | int width = s->avctx->width; | |||
| 733 | int height = s->avctx->height; | |||
| 734 | ||||
| 735 | if (buf_size < 3) { | |||
| 736 | av_log(s->avctx, AV_LOG_ERROR16, "Insufficient data (%d) for header\n", buf_size); | |||
| 737 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 738 | } | |||
| 739 | ||||
| 740 | s->keyframe = !(buf[0] & 1); | |||
| 741 | s->profile = (buf[0]>>1) & 7; | |||
| 742 | s->invisible = !(buf[0] & 0x10); | |||
| 743 | header_size = AV_RL24(buf)((((const uint8_t*)(buf))[2] << 16) | (((const uint8_t* )(buf))[1] << 8) | ((const uint8_t*)(buf))[0]) >> 5; | |||
| 744 | buf += 3; | |||
| 745 | buf_size -= 3; | |||
| 746 | ||||
| 747 | s->header_partition_size = header_size; | |||
| 748 | ||||
| 749 | if (s->profile > 3) | |||
| 750 | av_log(s->avctx, AV_LOG_WARNING24, "Unknown profile %d\n", s->profile); | |||
| 751 | ||||
| 752 | if (!s->profile) | |||
| 753 | memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab, | |||
| 754 | sizeof(s->put_pixels_tab)); | |||
| 755 | else // profile 1-3 use bilinear, 4+ aren't defined so whatever | |||
| 756 | memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_bilinear_pixels_tab, | |||
| 757 | sizeof(s->put_pixels_tab)); | |||
| 758 | ||||
| 759 | if (header_size > buf_size - 7 * s->keyframe) { | |||
| 760 | av_log(s->avctx, AV_LOG_ERROR16, "Header size larger than data provided\n"); | |||
| 761 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 762 | } | |||
| 763 | ||||
| 764 | if (s->keyframe) { | |||
| 765 | if (AV_RL24(buf)((((const uint8_t*)(buf))[2] << 16) | (((const uint8_t* )(buf))[1] << 8) | ((const uint8_t*)(buf))[0]) != 0x2a019d) { | |||
| 766 | av_log(s->avctx, AV_LOG_ERROR16, | |||
| 767 | "Invalid start code 0x%x\n", AV_RL24(buf)((((const uint8_t*)(buf))[2] << 16) | (((const uint8_t* )(buf))[1] << 8) | ((const uint8_t*)(buf))[0])); | |||
| 768 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 769 | } | |||
| 770 | width = AV_RL16(buf + 3)(((const union unaligned_16 *) (buf + 3))->l) & 0x3fff; | |||
| 771 | height = AV_RL16(buf + 5)(((const union unaligned_16 *) (buf + 5))->l) & 0x3fff; | |||
| 772 | hscale = buf[4] >> 6; | |||
| 773 | vscale = buf[6] >> 6; | |||
| 774 | buf += 7; | |||
| 775 | buf_size -= 7; | |||
| 776 | ||||
| 777 | if (hscale || vscale) | |||
| 778 | avpriv_request_sample(s->avctx, "Upscaling"); | |||
| 779 | ||||
| 780 | s->update_golden = s->update_altref = VP8_FRAME_CURRENT; | |||
| 781 | vp78_reset_probability_tables(s); | |||
| 782 | memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter, | |||
| 783 | sizeof(s->prob->pred16x16)); | |||
| 784 | memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter, | |||
| 785 | sizeof(s->prob->pred8x8c)); | |||
| 786 | memcpy(s->prob->mvc, vp8_mv_default_prob, | |||
| 787 | sizeof(s->prob->mvc)); | |||
| 788 | memset(&s->segmentation, 0, sizeof(s->segmentation)); | |||
| 789 | memset(&s->lf_delta, 0, sizeof(s->lf_delta)); | |||
| 790 | } | |||
| 791 | ||||
| 792 | ret = ff_vpx_init_range_decoder(c, buf, header_size); | |||
| 793 | if (ret < 0) | |||
| 794 | return ret; | |||
| 795 | buf += header_size; | |||
| 796 | buf_size -= header_size; | |||
| 797 | ||||
| 798 | if (s->keyframe) { | |||
| 799 | s->colorspace = vp89_rac_get(c); | |||
| 800 | if (s->colorspace) | |||
| 801 | av_log(s->avctx, AV_LOG_WARNING24, "Unspecified colorspace\n"); | |||
| 802 | s->fullrange = vp89_rac_get(c); | |||
| 803 | } | |||
| 804 | ||||
| 805 | if ((s->segmentation.enabled = vp89_rac_get(c))) | |||
| 806 | parse_segment_info(s); | |||
| 807 | else | |||
| 808 | s->segmentation.update_map = 0; // FIXME: move this to some init function? | |||
| 809 | ||||
| 810 | s->filter.simple = vp89_rac_get(c); | |||
| 811 | s->filter.level = vp89_rac_get_uint(c, 6); | |||
| 812 | s->filter.sharpness = vp89_rac_get_uint(c, 3); | |||
| 813 | ||||
| 814 | if ((s->lf_delta.enabled = vp89_rac_get(c))) { | |||
| 815 | s->lf_delta.update = vp89_rac_get(c); | |||
| 816 | if (s->lf_delta.update) | |||
| 817 | update_lf_deltas(s); | |||
| 818 | } | |||
| 819 | ||||
| 820 | if (setup_partitions(s, buf, buf_size)) { | |||
| 821 | av_log(s->avctx, AV_LOG_ERROR16, "Invalid partitions\n"); | |||
| 822 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 823 | } | |||
| 824 | ||||
| 825 | if (!s->macroblocks_base || /* first frame */ | |||
| 826 | width != s->avctx->width || height != s->avctx->height || | |||
| 827 | (width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height) | |||
| 828 | if ((ret = vp8_update_dimensions(s, width, height)) < 0) | |||
| 829 | return ret; | |||
| 830 | ||||
| 831 | vp8_get_quants(s); | |||
| 832 | ||||
| 833 | if (!s->keyframe) { | |||
| 834 | update_refs(s); | |||
| 835 | s->sign_bias[VP8_FRAME_GOLDEN] = vp89_rac_get(c); | |||
| 836 | s->sign_bias[VP8_FRAME_ALTREF] = vp89_rac_get(c); | |||
| 837 | } | |||
| 838 | ||||
| 839 | // if we aren't saving this frame's probabilities for future frames, | |||
| 840 | // make a copy of the current probabilities | |||
| 841 | if (!(s->update_probabilities = vp89_rac_get(c))) | |||
| 842 | s->prob[1] = s->prob[0]; | |||
| 843 | ||||
| 844 | s->update_last = s->keyframe || vp89_rac_get(c); | |||
| 845 | ||||
| 846 | vp78_update_probability_tables(s); | |||
| 847 | ||||
| 848 | if ((s->mbskip_enabled = vp89_rac_get(c))) | |||
| 849 | s->prob->mbskip = vp89_rac_get_uint(c, 8); | |||
| 850 | ||||
| 851 | if (!s->keyframe) { | |||
| 852 | s->prob->intra = vp89_rac_get_uint(c, 8); | |||
| 853 | s->prob->last = vp89_rac_get_uint(c, 8); | |||
| 854 | s->prob->golden = vp89_rac_get_uint(c, 8); | |||
| 855 | vp78_update_pred16x16_pred8x8_mvc_probabilities(s, VP8_MVC_SIZE19); | |||
| 856 | } | |||
| 857 | ||||
| 858 | // Record the entropy coder state here so that hwaccels can use it. | |||
| 859 | s->c.code_word = vpx_rac_renorm(&s->c); | |||
| 860 | s->coder_state_at_header_end.input = s->c.buffer - (-s->c.bits / 8); | |||
| 861 | s->coder_state_at_header_end.range = s->c.high; | |||
| 862 | s->coder_state_at_header_end.value = s->c.code_word >> 16; | |||
| 863 | s->coder_state_at_header_end.bit_count = -s->c.bits % 8; | |||
| 864 | ||||
| 865 | return 0; | |||
| 866 | } | |||
| 867 | ||||
| 868 | static av_always_inline__attribute__((always_inline)) inline | |||
| 869 | void clamp_mv(const VP8mvbounds *s, VP8mv *dst, const VP8mv *src) | |||
| 870 | { | |||
| 871 | dst->x = av_clipav_clip_c(src->x, av_clipav_clip_c(s->mv_min.x, INT16_MIN(-32767-1), INT16_MAX(32767)), | |||
| 872 | av_clipav_clip_c(s->mv_max.x, INT16_MIN(-32767-1), INT16_MAX(32767))); | |||
| 873 | dst->y = av_clipav_clip_c(src->y, av_clipav_clip_c(s->mv_min.y, INT16_MIN(-32767-1), INT16_MAX(32767)), | |||
| 874 | av_clipav_clip_c(s->mv_max.y, INT16_MIN(-32767-1), INT16_MAX(32767))); | |||
| 875 | } | |||
| 876 | ||||
| 877 | /** | |||
| 878 | * Motion vector coding, 17.1. | |||
| 879 | */ | |||
| 880 | static av_always_inline__attribute__((always_inline)) inline int read_mv_component(VPXRangeCoder *c, const uint8_t *p, int vp7) | |||
| 881 | { | |||
| 882 | int bit, x = 0; | |||
| 883 | ||||
| 884 | if (vpx_rac_get_prob_branchy(c, p[0])) { | |||
| 885 | int i; | |||
| 886 | ||||
| 887 | for (i = 0; i < 3; i++) | |||
| 888 | x += vpx_rac_get_probvpx_rac_get_prob(c, p[9 + i]) << i; | |||
| 889 | for (i = (vp7 ? 7 : 9); i > 3; i--) | |||
| 890 | x += vpx_rac_get_probvpx_rac_get_prob(c, p[9 + i]) << i; | |||
| 891 | if (!(x & (vp7 ? 0xF0 : 0xFFF0)) || vpx_rac_get_probvpx_rac_get_prob(c, p[12])) | |||
| 892 | x += 8; | |||
| 893 | } else { | |||
| 894 | // small_mvtree | |||
| 895 | const uint8_t *ps = p + 2; | |||
| 896 | bit = vpx_rac_get_probvpx_rac_get_prob(c, *ps); | |||
| 897 | ps += 1 + 3 * bit; | |||
| 898 | x += 4 * bit; | |||
| 899 | bit = vpx_rac_get_probvpx_rac_get_prob(c, *ps); | |||
| 900 | ps += 1 + bit; | |||
| 901 | x += 2 * bit; | |||
| 902 | x += vpx_rac_get_probvpx_rac_get_prob(c, *ps); | |||
| 903 | } | |||
| 904 | ||||
| 905 | return (x && vpx_rac_get_probvpx_rac_get_prob(c, p[1])) ? -x : x; | |||
| 906 | } | |||
| 907 | ||||
| 908 | static int vp7_read_mv_component(VPXRangeCoder *c, const uint8_t *p) | |||
| 909 | { | |||
| 910 | return read_mv_component(c, p, 1); | |||
| 911 | } | |||
| 912 | ||||
| 913 | static int vp8_read_mv_component(VPXRangeCoder *c, const uint8_t *p) | |||
| 914 | { | |||
| 915 | return read_mv_component(c, p, 0); | |||
| 916 | } | |||
| 917 | ||||
| 918 | static av_always_inline__attribute__((always_inline)) inline | |||
| 919 | const uint8_t *get_submv_prob(uint32_t left, uint32_t top, int is_vp7) | |||
| 920 | { | |||
| 921 | if (is_vp7) | |||
| 922 | return vp7_submv_prob; | |||
| 923 | ||||
| 924 | if (left == top) | |||
| 925 | return vp8_submv_prob[4 - !!left]; | |||
| 926 | if (!top) | |||
| 927 | return vp8_submv_prob[2]; | |||
| 928 | return vp8_submv_prob[1 - !!left]; | |||
| 929 | } | |||
| 930 | ||||
| 931 | /** | |||
| 932 | * Split motion vector prediction, 16.4. | |||
| 933 | * @returns the number of motion vectors parsed (2, 4 or 16) | |||
| 934 | */ | |||
| 935 | static av_always_inline__attribute__((always_inline)) inline | |||
| 936 | int decode_splitmvs(const VP8Context *s, VPXRangeCoder *c, VP8Macroblock *mb, | |||
| 937 | int layout, int is_vp7) | |||
| 938 | { | |||
| 939 | int part_idx; | |||
| 940 | int n, num; | |||
| 941 | const VP8Macroblock *top_mb; | |||
| 942 | const VP8Macroblock *left_mb = &mb[-1]; | |||
| 943 | const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning]; | |||
| 944 | const uint8_t *mbsplits_top, *mbsplits_cur, *firstidx; | |||
| 945 | const VP8mv *top_mv; | |||
| 946 | const VP8mv *left_mv = left_mb->bmv; | |||
| 947 | const VP8mv *cur_mv = mb->bmv; | |||
| 948 | ||||
| 949 | if (!layout) // layout is inlined, s->mb_layout is not | |||
| 950 | top_mb = &mb[2]; | |||
| 951 | else | |||
| 952 | top_mb = &mb[-s->mb_width - 1]; | |||
| 953 | mbsplits_top = vp8_mbsplits[top_mb->partitioning]; | |||
| 954 | top_mv = top_mb->bmv; | |||
| 955 | ||||
| 956 | if (vpx_rac_get_prob_branchy(c, vp8_mbsplit_prob[0])) { | |||
| 957 | if (vpx_rac_get_prob_branchy(c, vp8_mbsplit_prob[1])) | |||
| 958 | part_idx = VP8_SPLITMVMODE_16x8 + vpx_rac_get_probvpx_rac_get_prob(c, vp8_mbsplit_prob[2]); | |||
| 959 | else | |||
| 960 | part_idx = VP8_SPLITMVMODE_8x8; | |||
| 961 | } else { | |||
| 962 | part_idx = VP8_SPLITMVMODE_4x4; | |||
| 963 | } | |||
| 964 | ||||
| 965 | num = vp8_mbsplit_count[part_idx]; | |||
| 966 | mbsplits_cur = vp8_mbsplits[part_idx], | |||
| 967 | firstidx = vp8_mbfirstidx[part_idx]; | |||
| 968 | mb->partitioning = part_idx; | |||
| 969 | ||||
| 970 | for (n = 0; n < num; n++) { | |||
| 971 | int k = firstidx[n]; | |||
| 972 | uint32_t left, above; | |||
| 973 | const uint8_t *submv_prob; | |||
| 974 | ||||
| 975 | if (!(k & 3)) | |||
| 976 | left = AV_RN32A(&left_mv[mbsplits_left[k + 3]])(((const av_alias32*)(&left_mv[mbsplits_left[k + 3]]))-> u32); | |||
| 977 | else | |||
| 978 | left = AV_RN32A(&cur_mv[mbsplits_cur[k - 1]])(((const av_alias32*)(&cur_mv[mbsplits_cur[k - 1]]))-> u32); | |||
| 979 | if (k <= 3) | |||
| 980 | above = AV_RN32A(&top_mv[mbsplits_top[k + 12]])(((const av_alias32*)(&top_mv[mbsplits_top[k + 12]]))-> u32); | |||
| 981 | else | |||
| 982 | above = AV_RN32A(&cur_mv[mbsplits_cur[k - 4]])(((const av_alias32*)(&cur_mv[mbsplits_cur[k - 4]]))-> u32); | |||
| 983 | ||||
| 984 | submv_prob = get_submv_prob(left, above, is_vp7); | |||
| 985 | ||||
| 986 | if (vpx_rac_get_prob_branchy(c, submv_prob[0])) { | |||
| 987 | if (vpx_rac_get_prob_branchy(c, submv_prob[1])) { | |||
| 988 | if (vpx_rac_get_prob_branchy(c, submv_prob[2])) { | |||
| 989 | mb->bmv[n].y = mb->mv.y + | |||
| 990 | read_mv_component(c, s->prob->mvc[0], is_vp7); | |||
| 991 | mb->bmv[n].x = mb->mv.x + | |||
| 992 | read_mv_component(c, s->prob->mvc[1], is_vp7); | |||
| 993 | } else { | |||
| 994 | AV_ZERO32(&mb->bmv[n])(((av_alias32*)(&mb->bmv[n]))->u32 = 0); | |||
| 995 | } | |||
| 996 | } else { | |||
| 997 | AV_WN32A(&mb->bmv[n], above)(((av_alias32*)(&mb->bmv[n]))->u32 = (above)); | |||
| 998 | } | |||
| 999 | } else { | |||
| 1000 | AV_WN32A(&mb->bmv[n], left)(((av_alias32*)(&mb->bmv[n]))->u32 = (left)); | |||
| 1001 | } | |||
| 1002 | } | |||
| 1003 | ||||
| 1004 | return num; | |||
| 1005 | } | |||
| 1006 | ||||
| 1007 | /** | |||
| 1008 | * The vp7 reference decoder uses a padding macroblock column (added to right | |||
| 1009 | * edge of the frame) to guard against illegal macroblock offsets. The | |||
| 1010 | * algorithm has bugs that permit offsets to straddle the padding column. | |||
| 1011 | * This function replicates those bugs. | |||
| 1012 | * | |||
| 1013 | * @param[out] edge_x macroblock x address | |||
| 1014 | * @param[out] edge_y macroblock y address | |||
| 1015 | * | |||
| 1016 | * @return macroblock offset legal (boolean) | |||
| 1017 | */ | |||
| 1018 | static int vp7_calculate_mb_offset(int mb_x, int mb_y, int mb_width, | |||
| 1019 | int xoffset, int yoffset, int boundary, | |||
| 1020 | int *edge_x, int *edge_y) | |||
| 1021 | { | |||
| 1022 | int vwidth = mb_width + 1; | |||
| 1023 | int new = (mb_y + yoffset) * vwidth + mb_x + xoffset; | |||
| 1024 | if (new < boundary || new % vwidth == vwidth - 1) | |||
| 1025 | return 0; | |||
| 1026 | *edge_y = new / vwidth; | |||
| 1027 | *edge_x = new % vwidth; | |||
| 1028 | return 1; | |||
| 1029 | } | |||
| 1030 | ||||
| 1031 | static const VP8mv *get_bmv_ptr(const VP8Macroblock *mb, int subblock) | |||
| 1032 | { | |||
| 1033 | return &mb->bmv[mb->mode == VP8_MVMODE_SPLIT ? vp8_mbsplits[mb->partitioning][subblock] : 0]; | |||
| 1034 | } | |||
| 1035 | ||||
| 1036 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1037 | void vp7_decode_mvs(VP8Context *s, VP8Macroblock *mb, | |||
| 1038 | int mb_x, int mb_y, int layout) | |||
| 1039 | { | |||
| 1040 | enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR }; | |||
| 1041 | enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT }; | |||
| 1042 | int idx = CNT_ZERO; | |||
| 1043 | VP8mv near_mv[3]; | |||
| 1044 | uint8_t cnt[3] = { 0 }; | |||
| 1045 | VPXRangeCoder *c = &s->c; | |||
| 1046 | int i; | |||
| 1047 | ||||
| 1048 | AV_ZERO32(&near_mv[0])(((av_alias32*)(&near_mv[0]))->u32 = 0); | |||
| 1049 | AV_ZERO32(&near_mv[1])(((av_alias32*)(&near_mv[1]))->u32 = 0); | |||
| 1050 | AV_ZERO32(&near_mv[2])(((av_alias32*)(&near_mv[2]))->u32 = 0); | |||
| 1051 | ||||
| 1052 | for (i = 0; i < VP7_MV_PRED_COUNT12; i++) { | |||
| 1053 | const VP7MVPred * pred = &vp7_mv_pred[i]; | |||
| 1054 | int edge_x, edge_y; | |||
| 1055 | ||||
| 1056 | if (vp7_calculate_mb_offset(mb_x, mb_y, s->mb_width, pred->xoffset, | |||
| 1057 | pred->yoffset, !s->profile, &edge_x, &edge_y)) { | |||
| 1058 | const VP8Macroblock *edge = (s->mb_layout == 1) | |||
| 1059 | ? s->macroblocks_base + 1 + edge_x + | |||
| 1060 | (s->mb_width + 1) * (edge_y + 1) | |||
| 1061 | : s->macroblocks + edge_x + | |||
| 1062 | (s->mb_height - edge_y - 1) * 2; | |||
| 1063 | uint32_t mv = AV_RN32A(get_bmv_ptr(edge, vp7_mv_pred[i].subblock))(((const av_alias32*)(get_bmv_ptr(edge, vp7_mv_pred[i].subblock )))->u32); | |||
| 1064 | if (mv) { | |||
| 1065 | if (AV_RN32A(&near_mv[CNT_NEAREST])(((const av_alias32*)(&near_mv[CNT_NEAREST]))->u32)) { | |||
| 1066 | if (mv == AV_RN32A(&near_mv[CNT_NEAREST])(((const av_alias32*)(&near_mv[CNT_NEAREST]))->u32)) { | |||
| 1067 | idx = CNT_NEAREST; | |||
| 1068 | } else if (AV_RN32A(&near_mv[CNT_NEAR])(((const av_alias32*)(&near_mv[CNT_NEAR]))->u32)) { | |||
| 1069 | if (mv != AV_RN32A(&near_mv[CNT_NEAR])(((const av_alias32*)(&near_mv[CNT_NEAR]))->u32)) | |||
| 1070 | continue; | |||
| 1071 | idx = CNT_NEAR; | |||
| 1072 | } else { | |||
| 1073 | AV_WN32A(&near_mv[CNT_NEAR], mv)(((av_alias32*)(&near_mv[CNT_NEAR]))->u32 = (mv)); | |||
| 1074 | idx = CNT_NEAR; | |||
| 1075 | } | |||
| 1076 | } else { | |||
| 1077 | AV_WN32A(&near_mv[CNT_NEAREST], mv)(((av_alias32*)(&near_mv[CNT_NEAREST]))->u32 = (mv)); | |||
| 1078 | idx = CNT_NEAREST; | |||
| 1079 | } | |||
| 1080 | } else { | |||
| 1081 | idx = CNT_ZERO; | |||
| 1082 | } | |||
| 1083 | } else { | |||
| 1084 | idx = CNT_ZERO; | |||
| 1085 | } | |||
| 1086 | cnt[idx] += vp7_mv_pred[i].score; | |||
| 1087 | } | |||
| 1088 | ||||
| 1089 | mb->partitioning = VP8_SPLITMVMODE_NONE; | |||
| 1090 | ||||
| 1091 | if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_ZERO]][0])) { | |||
| 1092 | mb->mode = VP8_MVMODE_MV; | |||
| 1093 | ||||
| 1094 | if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAREST]][1])) { | |||
| 1095 | ||||
| 1096 | if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][2])) { | |||
| 1097 | ||||
| 1098 | if (cnt[CNT_NEAREST] > cnt[CNT_NEAR]) | |||
| 1099 | AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAREST] ? 0 : AV_RN32A(&near_mv[CNT_NEAREST]))(((av_alias32*)(&mb->mv))->u32 = (cnt[CNT_ZERO] > cnt[CNT_NEAREST] ? 0 : (((const av_alias32*)(&near_mv[CNT_NEAREST ]))->u32))); | |||
| 1100 | else | |||
| 1101 | AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAR] ? 0 : AV_RN32A(&near_mv[CNT_NEAR]))(((av_alias32*)(&mb->mv))->u32 = (cnt[CNT_ZERO] > cnt[CNT_NEAR] ? 0 : (((const av_alias32*)(&near_mv[CNT_NEAR ]))->u32))); | |||
| 1102 | ||||
| 1103 | if (vpx_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][3])) { | |||
| 1104 | mb->mode = VP8_MVMODE_SPLIT; | |||
| 1105 | mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP71) - 1]; | |||
| 1106 | } else { | |||
| 1107 | mb->mv.y += vp7_read_mv_component(c, s->prob->mvc[0]); | |||
| 1108 | mb->mv.x += vp7_read_mv_component(c, s->prob->mvc[1]); | |||
| 1109 | mb->bmv[0] = mb->mv; | |||
| 1110 | } | |||
| 1111 | } else { | |||
| 1112 | mb->mv = near_mv[CNT_NEAR]; | |||
| 1113 | mb->bmv[0] = mb->mv; | |||
| 1114 | } | |||
| 1115 | } else { | |||
| 1116 | mb->mv = near_mv[CNT_NEAREST]; | |||
| 1117 | mb->bmv[0] = mb->mv; | |||
| 1118 | } | |||
| 1119 | } else { | |||
| 1120 | mb->mode = VP8_MVMODE_ZERO; | |||
| 1121 | AV_ZERO32(&mb->mv)(((av_alias32*)(&mb->mv))->u32 = 0); | |||
| 1122 | mb->bmv[0] = mb->mv; | |||
| 1123 | } | |||
| 1124 | } | |||
| 1125 | ||||
| 1126 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1127 | void vp8_decode_mvs(VP8Context *s, const VP8mvbounds *mv_bounds, VP8Macroblock *mb, | |||
| 1128 | int mb_x, int mb_y, int layout) | |||
| 1129 | { | |||
| 1130 | VP8Macroblock *mb_edge[3] = { 0 /* top */, | |||
| 1131 | mb - 1 /* left */, | |||
| 1132 | 0 /* top-left */ }; | |||
| 1133 | enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV }; | |||
| 1134 | enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT }; | |||
| 1135 | int idx = CNT_ZERO; | |||
| 1136 | int cur_sign_bias = s->sign_bias[mb->ref_frame]; | |||
| 1137 | const int8_t *sign_bias = s->sign_bias; | |||
| 1138 | VP8mv near_mv[4]; | |||
| 1139 | uint8_t cnt[4] = { 0 }; | |||
| 1140 | VPXRangeCoder *c = &s->c; | |||
| 1141 | ||||
| 1142 | if (!layout) { // layout is inlined (s->mb_layout is not) | |||
| 1143 | mb_edge[0] = mb + 2; | |||
| 1144 | mb_edge[2] = mb + 1; | |||
| 1145 | } else { | |||
| 1146 | mb_edge[0] = mb - s->mb_width - 1; | |||
| 1147 | mb_edge[2] = mb - s->mb_width - 2; | |||
| 1148 | } | |||
| 1149 | ||||
| 1150 | AV_ZERO32(&near_mv[0])(((av_alias32*)(&near_mv[0]))->u32 = 0); | |||
| 1151 | AV_ZERO32(&near_mv[1])(((av_alias32*)(&near_mv[1]))->u32 = 0); | |||
| 1152 | AV_ZERO32(&near_mv[2])(((av_alias32*)(&near_mv[2]))->u32 = 0); | |||
| 1153 | ||||
| 1154 | /* Process MB on top, left and top-left */ | |||
| 1155 | #define MV_EDGE_CHECK(n){ const VP8Macroblock *edge = mb_edge[n]; int edge_ref = edge ->ref_frame; if (edge_ref != VP8_FRAME_CURRENT) { uint32_t mv = (((const av_alias32*)(&edge->mv))->u32); if ( mv) { if (cur_sign_bias != sign_bias[edge_ref]) { mv = ~mv; mv = ((mv & 0x7fff7fff) + 0x00010001) ^ (mv & 0x80008000 ); } if (!n || mv != (((const av_alias32*)(&near_mv[idx]) )->u32)) (((av_alias32*)(&near_mv[++idx]))->u32 = ( mv)); cnt[idx] += 1 + (n != 2); } else cnt[CNT_ZERO] += 1 + ( n != 2); } } \ | |||
| 1156 | { \ | |||
| 1157 | const VP8Macroblock *edge = mb_edge[n]; \ | |||
| 1158 | int edge_ref = edge->ref_frame; \ | |||
| 1159 | if (edge_ref != VP8_FRAME_CURRENT) { \ | |||
| 1160 | uint32_t mv = AV_RN32A(&edge->mv)(((const av_alias32*)(&edge->mv))->u32); \ | |||
| 1161 | if (mv) { \ | |||
| 1162 | if (cur_sign_bias != sign_bias[edge_ref]) { \ | |||
| 1163 | /* SWAR negate of the values in mv. */ \ | |||
| 1164 | mv = ~mv; \ | |||
| 1165 | mv = ((mv & 0x7fff7fff) + \ | |||
| 1166 | 0x00010001) ^ (mv & 0x80008000); \ | |||
| 1167 | } \ | |||
| 1168 | if (!n || mv != AV_RN32A(&near_mv[idx])(((const av_alias32*)(&near_mv[idx]))->u32)) \ | |||
| 1169 | AV_WN32A(&near_mv[++idx], mv)(((av_alias32*)(&near_mv[++idx]))->u32 = (mv)); \ | |||
| 1170 | cnt[idx] += 1 + (n != 2); \ | |||
| 1171 | } else \ | |||
| 1172 | cnt[CNT_ZERO] += 1 + (n != 2); \ | |||
| 1173 | } \ | |||
| 1174 | } | |||
| 1175 | ||||
| 1176 | MV_EDGE_CHECK(0){ const VP8Macroblock *edge = mb_edge[0]; int edge_ref = edge ->ref_frame; if (edge_ref != VP8_FRAME_CURRENT) { uint32_t mv = (((const av_alias32*)(&edge->mv))->u32); if ( mv) { if (cur_sign_bias != sign_bias[edge_ref]) { mv = ~mv; mv = ((mv & 0x7fff7fff) + 0x00010001) ^ (mv & 0x80008000 ); } if (!0 || mv != (((const av_alias32*)(&near_mv[idx]) )->u32)) (((av_alias32*)(&near_mv[++idx]))->u32 = ( mv)); cnt[idx] += 1 + (0 != 2); } else cnt[CNT_ZERO] += 1 + ( 0 != 2); } } | |||
| 1177 | MV_EDGE_CHECK(1){ const VP8Macroblock *edge = mb_edge[1]; int edge_ref = edge ->ref_frame; if (edge_ref != VP8_FRAME_CURRENT) { uint32_t mv = (((const av_alias32*)(&edge->mv))->u32); if ( mv) { if (cur_sign_bias != sign_bias[edge_ref]) { mv = ~mv; mv = ((mv & 0x7fff7fff) + 0x00010001) ^ (mv & 0x80008000 ); } if (!1 || mv != (((const av_alias32*)(&near_mv[idx]) )->u32)) (((av_alias32*)(&near_mv[++idx]))->u32 = ( mv)); cnt[idx] += 1 + (1 != 2); } else cnt[CNT_ZERO] += 1 + ( 1 != 2); } } | |||
| 1178 | MV_EDGE_CHECK(2){ const VP8Macroblock *edge = mb_edge[2]; int edge_ref = edge ->ref_frame; if (edge_ref != VP8_FRAME_CURRENT) { uint32_t mv = (((const av_alias32*)(&edge->mv))->u32); if ( mv) { if (cur_sign_bias != sign_bias[edge_ref]) { mv = ~mv; mv = ((mv & 0x7fff7fff) + 0x00010001) ^ (mv & 0x80008000 ); } if (!2 || mv != (((const av_alias32*)(&near_mv[idx]) )->u32)) (((av_alias32*)(&near_mv[++idx]))->u32 = ( mv)); cnt[idx] += 1 + (2 != 2); } else cnt[CNT_ZERO] += 1 + ( 2 != 2); } } | |||
| 1179 | ||||
| 1180 | mb->partitioning = VP8_SPLITMVMODE_NONE; | |||
| 1181 | if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_ZERO]][0])) { | |||
| 1182 | mb->mode = VP8_MVMODE_MV; | |||
| 1183 | ||||
| 1184 | /* If we have three distinct MVs, merge first and last if they're the same */ | |||
| 1185 | if (cnt[CNT_SPLITMV] && | |||
| 1186 | AV_RN32A(&near_mv[1 + VP8_EDGE_TOP])(((const av_alias32*)(&near_mv[1 + VP8_EDGE_TOP]))->u32 ) == AV_RN32A(&near_mv[1 + VP8_EDGE_TOPLEFT])(((const av_alias32*)(&near_mv[1 + VP8_EDGE_TOPLEFT]))-> u32)) | |||
| 1187 | cnt[CNT_NEAREST] += 1; | |||
| 1188 | ||||
| 1189 | /* Swap near and nearest if necessary */ | |||
| 1190 | if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) { | |||
| 1191 | FFSWAP(uint8_t, cnt[CNT_NEAREST], cnt[CNT_NEAR])do{uint8_t SWAP_tmp= cnt[CNT_NEAR]; cnt[CNT_NEAR]= cnt[CNT_NEAREST ]; cnt[CNT_NEAREST]= SWAP_tmp;}while(0); | |||
| 1192 | FFSWAP(VP8mv, near_mv[CNT_NEAREST], near_mv[CNT_NEAR])do{VP8mv SWAP_tmp= near_mv[CNT_NEAR]; near_mv[CNT_NEAR]= near_mv [CNT_NEAREST]; near_mv[CNT_NEAREST]= SWAP_tmp;}while(0); | |||
| 1193 | } | |||
| 1194 | ||||
| 1195 | if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAREST]][1])) { | |||
| 1196 | if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAR]][2])) { | |||
| 1197 | /* Choose the best mv out of 0,0 and the nearest mv */ | |||
| 1198 | clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]); | |||
| 1199 | cnt[CNT_SPLITMV] = ((mb_edge[VP8_EDGE_LEFT]->mode == VP8_MVMODE_SPLIT) + | |||
| 1200 | (mb_edge[VP8_EDGE_TOP]->mode == VP8_MVMODE_SPLIT)) * 2 + | |||
| 1201 | (mb_edge[VP8_EDGE_TOPLEFT]->mode == VP8_MVMODE_SPLIT); | |||
| 1202 | ||||
| 1203 | if (vpx_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_SPLITMV]][3])) { | |||
| 1204 | mb->mode = VP8_MVMODE_SPLIT; | |||
| 1205 | mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP80) - 1]; | |||
| 1206 | } else { | |||
| 1207 | mb->mv.y += vp8_read_mv_component(c, s->prob->mvc[0]); | |||
| 1208 | mb->mv.x += vp8_read_mv_component(c, s->prob->mvc[1]); | |||
| 1209 | mb->bmv[0] = mb->mv; | |||
| 1210 | } | |||
| 1211 | } else { | |||
| 1212 | clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAR]); | |||
| 1213 | mb->bmv[0] = mb->mv; | |||
| 1214 | } | |||
| 1215 | } else { | |||
| 1216 | clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAREST]); | |||
| 1217 | mb->bmv[0] = mb->mv; | |||
| 1218 | } | |||
| 1219 | } else { | |||
| 1220 | mb->mode = VP8_MVMODE_ZERO; | |||
| 1221 | AV_ZERO32(&mb->mv)(((av_alias32*)(&mb->mv))->u32 = 0); | |||
| 1222 | mb->bmv[0] = mb->mv; | |||
| 1223 | } | |||
| 1224 | } | |||
| 1225 | ||||
| 1226 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1227 | void decode_intra4x4_modes(VP8Context *s, VPXRangeCoder *c, VP8Macroblock *mb, | |||
| 1228 | int mb_x, int keyframe, int layout) | |||
| 1229 | { | |||
| 1230 | uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb; | |||
| 1231 | ||||
| 1232 | if (layout) { | |||
| 1233 | VP8Macroblock *mb_top = mb - s->mb_width - 1; | |||
| 1234 | memcpy(mb->intra4x4_pred_mode_top, mb_top->intra4x4_pred_mode_top, 4); | |||
| 1235 | } | |||
| 1236 | if (keyframe) { | |||
| 1237 | int x, y; | |||
| 1238 | uint8_t *top; | |||
| 1239 | uint8_t *const left = s->intra4x4_pred_mode_left; | |||
| 1240 | if (layout) | |||
| 1241 | top = mb->intra4x4_pred_mode_top; | |||
| 1242 | else | |||
| 1243 | top = s->intra4x4_pred_mode_top + 4 * mb_x; | |||
| 1244 | for (y = 0; y < 4; y++) { | |||
| 1245 | for (x = 0; x < 4; x++) { | |||
| 1246 | const uint8_t *ctx; | |||
| 1247 | ctx = vp8_pred4x4_prob_intra[top[x]][left[y]]; | |||
| 1248 | *intra4x4 = vp89_rac_get_tree(c, vp8_pred4x4_tree, ctx); | |||
| 1249 | left[y] = top[x] = *intra4x4; | |||
| 1250 | intra4x4++; | |||
| 1251 | } | |||
| 1252 | } | |||
| 1253 | } else { | |||
| 1254 | int i; | |||
| 1255 | for (i = 0; i < 16; i++) | |||
| 1256 | intra4x4[i] = vp89_rac_get_tree(c, vp8_pred4x4_tree, | |||
| 1257 | vp8_pred4x4_prob_inter); | |||
| 1258 | } | |||
| 1259 | } | |||
| 1260 | ||||
| 1261 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1262 | void decode_mb_mode(VP8Context *s, const VP8mvbounds *mv_bounds, | |||
| 1263 | VP8Macroblock *mb, int mb_x, int mb_y, | |||
| 1264 | uint8_t *segment, const uint8_t *ref, int layout, int is_vp7) | |||
| 1265 | { | |||
| 1266 | VPXRangeCoder *c = &s->c; | |||
| 1267 | static const char * const vp7_feature_name[] = { "q-index", | |||
| 1268 | "lf-delta", | |||
| 1269 | "partial-golden-update", | |||
| 1270 | "blit-pitch" }; | |||
| 1271 | if (is_vp7) { | |||
| 1272 | int i; | |||
| 1273 | *segment = 0; | |||
| 1274 | for (i = 0; i < 4; i++) { | |||
| 1275 | if (s->feature_enabled[i]) { | |||
| 1276 | if (vpx_rac_get_prob_branchy(c, s->feature_present_prob[i])) { | |||
| 1277 | int index = vp89_rac_get_tree(c, vp7_feature_index_tree, | |||
| 1278 | s->feature_index_prob[i]); | |||
| 1279 | av_log(s->avctx, AV_LOG_WARNING24, | |||
| 1280 | "Feature %s present in macroblock (value 0x%x)\n", | |||
| 1281 | vp7_feature_name[i], s->feature_value[i][index]); | |||
| 1282 | } | |||
| 1283 | } | |||
| 1284 | } | |||
| 1285 | } else if (s->segmentation.update_map) { | |||
| 1286 | int bit = vpx_rac_get_probvpx_rac_get_prob(c, s->prob->segmentid[0]); | |||
| 1287 | *segment = vpx_rac_get_probvpx_rac_get_prob(c, s->prob->segmentid[1+bit]) + 2*bit; | |||
| 1288 | } else if (s->segmentation.enabled) | |||
| 1289 | *segment = ref ? *ref : *segment; | |||
| 1290 | mb->segment = *segment; | |||
| 1291 | ||||
| 1292 | mb->skip = s->mbskip_enabled ? vpx_rac_get_probvpx_rac_get_prob(c, s->prob->mbskip) : 0; | |||
| 1293 | ||||
| 1294 | if (s->keyframe) { | |||
| 1295 | mb->mode = vp89_rac_get_tree(c, vp8_pred16x16_tree_intra, | |||
| 1296 | vp8_pred16x16_prob_intra); | |||
| 1297 | ||||
| 1298 | if (mb->mode == MODE_I4x44) { | |||
| 1299 | decode_intra4x4_modes(s, c, mb, mb_x, 1, layout); | |||
| 1300 | } else { | |||
| 1301 | const uint32_t modes = (is_vp7 ? vp7_pred4x4_mode | |||
| 1302 | : vp8_pred4x4_mode)[mb->mode] * 0x01010101u; | |||
| 1303 | if (s->mb_layout) | |||
| 1304 | AV_WN32A(mb->intra4x4_pred_mode_top, modes)(((av_alias32*)(mb->intra4x4_pred_mode_top))->u32 = (modes )); | |||
| 1305 | else | |||
| 1306 | AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes)(((av_alias32*)(s->intra4x4_pred_mode_top + 4 * mb_x))-> u32 = (modes)); | |||
| 1307 | AV_WN32A(s->intra4x4_pred_mode_left, modes)(((av_alias32*)(s->intra4x4_pred_mode_left))->u32 = (modes )); | |||
| 1308 | } | |||
| 1309 | ||||
| 1310 | mb->chroma_pred_mode = vp89_rac_get_tree(c, vp8_pred8x8c_tree, | |||
| 1311 | vp8_pred8x8c_prob_intra); | |||
| 1312 | mb->ref_frame = VP8_FRAME_CURRENT; | |||
| 1313 | } else if (vpx_rac_get_prob_branchy(c, s->prob->intra)) { | |||
| 1314 | // inter MB, 16.2 | |||
| 1315 | if (vpx_rac_get_prob_branchy(c, s->prob->last)) | |||
| 1316 | mb->ref_frame = | |||
| 1317 | (!is_vp7 && vpx_rac_get_probvpx_rac_get_prob(c, s->prob->golden)) ? VP8_FRAME_ALTREF | |||
| 1318 | : VP8_FRAME_GOLDEN; | |||
| 1319 | else | |||
| 1320 | mb->ref_frame = VP8_FRAME_PREVIOUS; | |||
| 1321 | s->ref_count[mb->ref_frame - 1]++; | |||
| 1322 | ||||
| 1323 | // motion vectors, 16.3 | |||
| 1324 | if (is_vp7) | |||
| 1325 | vp7_decode_mvs(s, mb, mb_x, mb_y, layout); | |||
| 1326 | else | |||
| 1327 | vp8_decode_mvs(s, mv_bounds, mb, mb_x, mb_y, layout); | |||
| 1328 | } else { | |||
| 1329 | // intra MB, 16.1 | |||
| 1330 | mb->mode = vp89_rac_get_tree(c, vp8_pred16x16_tree_inter, | |||
| 1331 | s->prob->pred16x16); | |||
| 1332 | ||||
| 1333 | if (mb->mode == MODE_I4x44) | |||
| 1334 | decode_intra4x4_modes(s, c, mb, mb_x, 0, layout); | |||
| 1335 | ||||
| 1336 | mb->chroma_pred_mode = vp89_rac_get_tree(c, vp8_pred8x8c_tree, | |||
| 1337 | s->prob->pred8x8c); | |||
| 1338 | mb->ref_frame = VP8_FRAME_CURRENT; | |||
| 1339 | mb->partitioning = VP8_SPLITMVMODE_NONE; | |||
| 1340 | AV_ZERO32(&mb->bmv[0])(((av_alias32*)(&mb->bmv[0]))->u32 = 0); | |||
| 1341 | } | |||
| 1342 | } | |||
| 1343 | ||||
| 1344 | /** | |||
| 1345 | * @param r arithmetic bitstream reader context | |||
| 1346 | * @param block destination for block coefficients | |||
| 1347 | * @param probs probabilities to use when reading trees from the bitstream | |||
| 1348 | * @param i initial coeff index, 0 unless a separate DC block is coded | |||
| 1349 | * @param qmul array holding the dc/ac dequant factor at position 0/1 | |||
| 1350 | * | |||
| 1351 | * @return 0 if no coeffs were decoded | |||
| 1352 | * otherwise, the index of the last coeff decoded plus one | |||
| 1353 | */ | |||
| 1354 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1355 | int decode_block_coeffs_internal(VPXRangeCoder *r, int16_t block[16], | |||
| 1356 | uint8_t probs[16][3][NUM_DCT_TOKENS - 1], | |||
| 1357 | int i, const uint8_t *token_prob, const int16_t qmul[2], | |||
| 1358 | const uint8_t scan[16], int vp7) | |||
| 1359 | { | |||
| 1360 | VPXRangeCoder c = *r; | |||
| 1361 | goto skip_eob; | |||
| 1362 | do { | |||
| 1363 | int coeff; | |||
| 1364 | restart: | |||
| 1365 | if (!vpx_rac_get_prob_branchy(&c, token_prob[0])) // DCT_EOB | |||
| 1366 | break; | |||
| 1367 | ||||
| 1368 | skip_eob: | |||
| 1369 | if (!vpx_rac_get_prob_branchy(&c, token_prob[1])) { // DCT_0 | |||
| 1370 | if (++i == 16) | |||
| 1371 | break; // invalid input; blocks should end with EOB | |||
| 1372 | token_prob = probs[i][0]; | |||
| 1373 | if (vp7) | |||
| 1374 | goto restart; | |||
| 1375 | goto skip_eob; | |||
| 1376 | } | |||
| 1377 | ||||
| 1378 | if (!vpx_rac_get_prob_branchy(&c, token_prob[2])) { // DCT_1 | |||
| 1379 | coeff = 1; | |||
| 1380 | token_prob = probs[i + 1][1]; | |||
| 1381 | } else { | |||
| 1382 | if (!vpx_rac_get_prob_branchy(&c, token_prob[3])) { // DCT 2,3,4 | |||
| 1383 | coeff = vpx_rac_get_prob_branchy(&c, token_prob[4]); | |||
| 1384 | if (coeff) | |||
| 1385 | coeff += vpx_rac_get_probvpx_rac_get_prob(&c, token_prob[5]); | |||
| 1386 | coeff += 2; | |||
| 1387 | } else { | |||
| 1388 | // DCT_CAT* | |||
| 1389 | if (!vpx_rac_get_prob_branchy(&c, token_prob[6])) { | |||
| 1390 | if (!vpx_rac_get_prob_branchy(&c, token_prob[7])) { // DCT_CAT1 | |||
| 1391 | coeff = 5 + vpx_rac_get_probvpx_rac_get_prob(&c, vp8_dct_cat1_prob[0]); | |||
| 1392 | } else { // DCT_CAT2 | |||
| 1393 | coeff = 7; | |||
| 1394 | coeff += vpx_rac_get_probvpx_rac_get_prob(&c, vp8_dct_cat2_prob[0]) << 1; | |||
| 1395 | coeff += vpx_rac_get_probvpx_rac_get_prob(&c, vp8_dct_cat2_prob[1]); | |||
| 1396 | } | |||
| 1397 | } else { // DCT_CAT3 and up | |||
| 1398 | int a = vpx_rac_get_probvpx_rac_get_prob(&c, token_prob[8]); | |||
| 1399 | int b = vpx_rac_get_probvpx_rac_get_prob(&c, token_prob[9 + a]); | |||
| 1400 | int cat = (a << 1) + b; | |||
| 1401 | coeff = 3 + (8 << cat); | |||
| 1402 | coeff += vp8_rac_get_coeff(&c, ff_vp8_dct_cat_prob[cat]); | |||
| 1403 | } | |||
| 1404 | } | |||
| 1405 | token_prob = probs[i + 1][2]; | |||
| 1406 | } | |||
| 1407 | block[scan[i]] = (vp89_rac_get(&c) ? -coeff : coeff) * qmul[!!i]; | |||
| 1408 | } while (++i < 16); | |||
| 1409 | ||||
| 1410 | *r = c; | |||
| 1411 | return i; | |||
| 1412 | } | |||
| 1413 | ||||
| 1414 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1415 | int inter_predict_dc(int16_t block[16], int16_t pred[2]) | |||
| 1416 | { | |||
| 1417 | int16_t dc = block[0]; | |||
| 1418 | int ret = 0; | |||
| 1419 | ||||
| 1420 | if (pred[1] > 3) { | |||
| 1421 | dc += pred[0]; | |||
| 1422 | ret = 1; | |||
| 1423 | } | |||
| 1424 | ||||
| 1425 | if (!pred[0] | !dc | ((int32_t)pred[0] ^ (int32_t)dc) >> 31) { | |||
| 1426 | block[0] = pred[0] = dc; | |||
| 1427 | pred[1] = 0; | |||
| 1428 | } else { | |||
| 1429 | if (pred[0] == dc) | |||
| 1430 | pred[1]++; | |||
| 1431 | block[0] = pred[0] = dc; | |||
| 1432 | } | |||
| 1433 | ||||
| 1434 | return ret; | |||
| 1435 | } | |||
| 1436 | ||||
| 1437 | static int vp7_decode_block_coeffs_internal(VPXRangeCoder *r, | |||
| 1438 | int16_t block[16], | |||
| 1439 | uint8_t probs[16][3][NUM_DCT_TOKENS - 1], | |||
| 1440 | int i, const uint8_t *token_prob, | |||
| 1441 | const int16_t qmul[2], | |||
| 1442 | const uint8_t scan[16]) | |||
| 1443 | { | |||
| 1444 | return decode_block_coeffs_internal(r, block, probs, i, | |||
| 1445 | token_prob, qmul, scan, IS_VP71); | |||
| 1446 | } | |||
| 1447 | ||||
| 1448 | #ifndef vp8_decode_block_coeffs_internal | |||
| 1449 | static int vp8_decode_block_coeffs_internal(VPXRangeCoder *r, | |||
| 1450 | int16_t block[16], | |||
| 1451 | uint8_t probs[16][3][NUM_DCT_TOKENS - 1], | |||
| 1452 | int i, const uint8_t *token_prob, | |||
| 1453 | const int16_t qmul[2]) | |||
| 1454 | { | |||
| 1455 | return decode_block_coeffs_internal(r, block, probs, i, | |||
| 1456 | token_prob, qmul, ff_zigzag_scan, IS_VP80); | |||
| 1457 | } | |||
| 1458 | #endif | |||
| 1459 | ||||
| 1460 | /** | |||
| 1461 | * @param c arithmetic bitstream reader context | |||
| 1462 | * @param block destination for block coefficients | |||
| 1463 | * @param probs probabilities to use when reading trees from the bitstream | |||
| 1464 | * @param i initial coeff index, 0 unless a separate DC block is coded | |||
| 1465 | * @param zero_nhood the initial prediction context for number of surrounding | |||
| 1466 | * all-zero blocks (only left/top, so 0-2) | |||
| 1467 | * @param qmul array holding the dc/ac dequant factor at position 0/1 | |||
| 1468 | * @param scan scan pattern (VP7 only) | |||
| 1469 | * | |||
| 1470 | * @return 0 if no coeffs were decoded | |||
| 1471 | * otherwise, the index of the last coeff decoded plus one | |||
| 1472 | */ | |||
| 1473 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1474 | int decode_block_coeffs(VPXRangeCoder *c, int16_t block[16], | |||
| 1475 | uint8_t probs[16][3][NUM_DCT_TOKENS - 1], | |||
| 1476 | int i, int zero_nhood, const int16_t qmul[2], | |||
| 1477 | const uint8_t scan[16], int vp7) | |||
| 1478 | { | |||
| 1479 | const uint8_t *token_prob = probs[i][zero_nhood]; | |||
| 1480 | if (!vpx_rac_get_prob_branchy(c, token_prob[0])) // DCT_EOB | |||
| 1481 | return 0; | |||
| 1482 | return vp7 ? vp7_decode_block_coeffs_internal(c, block, probs, i, | |||
| 1483 | token_prob, qmul, scan) | |||
| 1484 | : vp8_decode_block_coeffs_internal(c, block, probs, i, | |||
| 1485 | token_prob, qmul); | |||
| 1486 | } | |||
| 1487 | ||||
| 1488 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1489 | void decode_mb_coeffs(VP8Context *s, VP8ThreadData *td, VPXRangeCoder *c, | |||
| 1490 | VP8Macroblock *mb, uint8_t t_nnz[9], uint8_t l_nnz[9], | |||
| 1491 | int is_vp7) | |||
| 1492 | { | |||
| 1493 | int i, x, y, luma_start = 0, luma_ctx = 3; | |||
| 1494 | int nnz_pred, nnz, nnz_total = 0; | |||
| 1495 | int segment = mb->segment; | |||
| 1496 | int block_dc = 0; | |||
| 1497 | ||||
| 1498 | if (mb->mode != MODE_I4x44 && (is_vp7 || mb->mode != VP8_MVMODE_SPLIT)) { | |||
| 1499 | nnz_pred = t_nnz[8] + l_nnz[8]; | |||
| 1500 | ||||
| 1501 | // decode DC values and do hadamard | |||
| 1502 | nnz = decode_block_coeffs(c, td->block_dc, s->prob->token[1], 0, | |||
| 1503 | nnz_pred, s->qmat[segment].luma_dc_qmul, | |||
| 1504 | ff_zigzag_scan, is_vp7); | |||
| 1505 | l_nnz[8] = t_nnz[8] = !!nnz; | |||
| 1506 | ||||
| 1507 | if (is_vp7 && mb->mode > MODE_I4x44) { | |||
| 1508 | nnz |= inter_predict_dc(td->block_dc, | |||
| 1509 | s->inter_dc_pred[mb->ref_frame - 1]); | |||
| 1510 | } | |||
| 1511 | ||||
| 1512 | if (nnz) { | |||
| 1513 | nnz_total += nnz; | |||
| 1514 | block_dc = 1; | |||
| 1515 | if (nnz == 1) | |||
| 1516 | s->vp8dsp.vp8_luma_dc_wht_dc(td->block, td->block_dc); | |||
| 1517 | else | |||
| 1518 | s->vp8dsp.vp8_luma_dc_wht(td->block, td->block_dc); | |||
| 1519 | } | |||
| 1520 | luma_start = 1; | |||
| 1521 | luma_ctx = 0; | |||
| 1522 | } | |||
| 1523 | ||||
| 1524 | // luma blocks | |||
| 1525 | for (y = 0; y < 4; y++) | |||
| 1526 | for (x = 0; x < 4; x++) { | |||
| 1527 | nnz_pred = l_nnz[y] + t_nnz[x]; | |||
| 1528 | nnz = decode_block_coeffs(c, td->block[y][x], | |||
| 1529 | s->prob->token[luma_ctx], | |||
| 1530 | luma_start, nnz_pred, | |||
| 1531 | s->qmat[segment].luma_qmul, | |||
| 1532 | s->prob[0].scan, is_vp7); | |||
| 1533 | /* nnz+block_dc may be one more than the actual last index, | |||
| 1534 | * but we don't care */ | |||
| 1535 | td->non_zero_count_cache[y][x] = nnz + block_dc; | |||
| 1536 | t_nnz[x] = l_nnz[y] = !!nnz; | |||
| 1537 | nnz_total += nnz; | |||
| 1538 | } | |||
| 1539 | ||||
| 1540 | // chroma blocks | |||
| 1541 | // TODO: what to do about dimensions? 2nd dim for luma is x, | |||
| 1542 | // but for chroma it's (y<<1)|x | |||
| 1543 | for (i = 4; i < 6; i++) | |||
| 1544 | for (y = 0; y < 2; y++) | |||
| 1545 | for (x = 0; x < 2; x++) { | |||
| 1546 | nnz_pred = l_nnz[i + 2 * y] + t_nnz[i + 2 * x]; | |||
| 1547 | nnz = decode_block_coeffs(c, td->block[i][(y << 1) + x], | |||
| 1548 | s->prob->token[2], 0, nnz_pred, | |||
| 1549 | s->qmat[segment].chroma_qmul, | |||
| 1550 | s->prob[0].scan, is_vp7); | |||
| 1551 | td->non_zero_count_cache[i][(y << 1) + x] = nnz; | |||
| 1552 | t_nnz[i + 2 * x] = l_nnz[i + 2 * y] = !!nnz; | |||
| 1553 | nnz_total += nnz; | |||
| 1554 | } | |||
| 1555 | ||||
| 1556 | // if there were no coded coeffs despite the macroblock not being marked skip, | |||
| 1557 | // we MUST not do the inner loop filter and should not do IDCT | |||
| 1558 | // Since skip isn't used for bitstream prediction, just manually set it. | |||
| 1559 | if (!nnz_total) | |||
| 1560 | mb->skip = 1; | |||
| 1561 | } | |||
| 1562 | ||||
| 1563 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1564 | void backup_mb_border(uint8_t *top_border, const uint8_t *src_y, | |||
| 1565 | const uint8_t *src_cb, const uint8_t *src_cr, | |||
| 1566 | ptrdiff_t linesize, ptrdiff_t uvlinesize, int simple) | |||
| 1567 | { | |||
| 1568 | AV_COPY128(top_border, src_y + 15 * linesize)do { (((av_alias64*)(top_border))->u64 = ((const av_alias64 *)(src_y + 15 * linesize))->u64); (((av_alias64*)((char*)( top_border)+8))->u64 = ((const av_alias64*)((char*)(src_y + 15 * linesize)+8))->u64); } while(0); | |||
| 1569 | if (!simple) { | |||
| 1570 | AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize)(((av_alias64*)(top_border + 16))->u64 = ((const av_alias64 *)(src_cb + 7 * uvlinesize))->u64); | |||
| 1571 | AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize)(((av_alias64*)(top_border + 24))->u64 = ((const av_alias64 *)(src_cr + 7 * uvlinesize))->u64); | |||
| 1572 | } | |||
| 1573 | } | |||
| 1574 | ||||
| 1575 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1576 | void xchg_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb, | |||
| 1577 | uint8_t *src_cr, ptrdiff_t linesize, ptrdiff_t uvlinesize, int mb_x, | |||
| 1578 | int mb_y, int mb_width, int simple, int xchg) | |||
| 1579 | { | |||
| 1580 | uint8_t *top_border_m1 = top_border - 32; // for TL prediction | |||
| 1581 | src_y -= linesize; | |||
| 1582 | src_cb -= uvlinesize; | |||
| 1583 | src_cr -= uvlinesize; | |||
| 1584 | ||||
| 1585 | #define XCHG(a, b, xchg)do { if (xchg) do{av_alias64 SWAP_tmp= *(av_alias64*)(a); *(av_alias64 *)(a)= *(av_alias64*)(b); *(av_alias64*)(b)= SWAP_tmp;}while( 0); else (((av_alias64*)(b))->u64 = ((const av_alias64*)(a ))->u64); } while (0) \ | |||
| 1586 | do { \ | |||
| 1587 | if (xchg) \ | |||
| 1588 | AV_SWAP64(b, a)do{av_alias64 SWAP_tmp= *(av_alias64*)(a); *(av_alias64*)(a)= *(av_alias64*)(b); *(av_alias64*)(b)= SWAP_tmp;}while(0); \ | |||
| 1589 | else \ | |||
| 1590 | AV_COPY64(b, a)(((av_alias64*)(b))->u64 = ((const av_alias64*)(a))->u64 ); \ | |||
| 1591 | } while (0) | |||
| 1592 | ||||
| 1593 | XCHG(top_border_m1 + 8, src_y - 8, xchg)do { if (xchg) do{av_alias64 SWAP_tmp= *(av_alias64*)(top_border_m1 + 8); *(av_alias64*)(top_border_m1 + 8)= *(av_alias64*)(src_y - 8); *(av_alias64*)(src_y - 8)= SWAP_tmp;}while(0); else (( (av_alias64*)(src_y - 8))->u64 = ((const av_alias64*)(top_border_m1 + 8))->u64); } while (0); | |||
| 1594 | XCHG(top_border, src_y, xchg)do { if (xchg) do{av_alias64 SWAP_tmp= *(av_alias64*)(top_border ); *(av_alias64*)(top_border)= *(av_alias64*)(src_y); *(av_alias64 *)(src_y)= SWAP_tmp;}while(0); else (((av_alias64*)(src_y))-> u64 = ((const av_alias64*)(top_border))->u64); } while (0); | |||
| 1595 | XCHG(top_border + 8, src_y + 8, 1)do { if (1) do{av_alias64 SWAP_tmp= *(av_alias64*)(top_border + 8); *(av_alias64*)(top_border + 8)= *(av_alias64*)(src_y + 8); *(av_alias64*)(src_y + 8)= SWAP_tmp;}while(0); else (((av_alias64 *)(src_y + 8))->u64 = ((const av_alias64*)(top_border + 8) )->u64); } while (0); | |||
| 1596 | if (mb_x < mb_width - 1) | |||
| 1597 | XCHG(top_border + 32, src_y + 16, 1)do { if (1) do{av_alias64 SWAP_tmp= *(av_alias64*)(top_border + 32); *(av_alias64*)(top_border + 32)= *(av_alias64*)(src_y + 16); *(av_alias64*)(src_y + 16)= SWAP_tmp;}while(0); else ( ((av_alias64*)(src_y + 16))->u64 = ((const av_alias64*)(top_border + 32))->u64); } while (0); | |||
| 1598 | ||||
| 1599 | // only copy chroma for normal loop filter | |||
| 1600 | // or to initialize the top row to 127 | |||
| 1601 | if (!simple || !mb_y) { | |||
| 1602 | XCHG(top_border_m1 + 16, src_cb - 8, xchg)do { if (xchg) do{av_alias64 SWAP_tmp= *(av_alias64*)(top_border_m1 + 16); *(av_alias64*)(top_border_m1 + 16)= *(av_alias64*)(src_cb - 8); *(av_alias64*)(src_cb - 8)= SWAP_tmp;}while(0); else ( ((av_alias64*)(src_cb - 8))->u64 = ((const av_alias64*)(top_border_m1 + 16))->u64); } while (0); | |||
| 1603 | XCHG(top_border_m1 + 24, src_cr - 8, xchg)do { if (xchg) do{av_alias64 SWAP_tmp= *(av_alias64*)(top_border_m1 + 24); *(av_alias64*)(top_border_m1 + 24)= *(av_alias64*)(src_cr - 8); *(av_alias64*)(src_cr - 8)= SWAP_tmp;}while(0); else ( ((av_alias64*)(src_cr - 8))->u64 = ((const av_alias64*)(top_border_m1 + 24))->u64); } while (0); | |||
| 1604 | XCHG(top_border + 16, src_cb, 1)do { if (1) do{av_alias64 SWAP_tmp= *(av_alias64*)(top_border + 16); *(av_alias64*)(top_border + 16)= *(av_alias64*)(src_cb ); *(av_alias64*)(src_cb)= SWAP_tmp;}while(0); else (((av_alias64 *)(src_cb))->u64 = ((const av_alias64*)(top_border + 16))-> u64); } while (0); | |||
| 1605 | XCHG(top_border + 24, src_cr, 1)do { if (1) do{av_alias64 SWAP_tmp= *(av_alias64*)(top_border + 24); *(av_alias64*)(top_border + 24)= *(av_alias64*)(src_cr ); *(av_alias64*)(src_cr)= SWAP_tmp;}while(0); else (((av_alias64 *)(src_cr))->u64 = ((const av_alias64*)(top_border + 24))-> u64); } while (0); | |||
| 1606 | } | |||
| 1607 | } | |||
| 1608 | ||||
| 1609 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1610 | int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y) | |||
| 1611 | { | |||
| 1612 | if (!mb_x) | |||
| 1613 | return mb_y ? TOP_DC_PRED8x85 : DC_128_PRED8x86; | |||
| 1614 | else | |||
| 1615 | return mb_y ? mode : LEFT_DC_PRED8x84; | |||
| 1616 | } | |||
| 1617 | ||||
| 1618 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1619 | int check_tm_pred8x8_mode(int mode, int mb_x, int mb_y, int vp7) | |||
| 1620 | { | |||
| 1621 | if (!mb_x) | |||
| 1622 | return mb_y ? VERT_PRED8x82 : (vp7 ? DC_128_PRED8x86 : DC_129_PRED8x88); | |||
| 1623 | else | |||
| 1624 | return mb_y ? mode : HOR_PRED8x81; | |||
| 1625 | } | |||
| 1626 | ||||
| 1627 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1628 | int check_intra_pred8x8_mode_emuedge(int mode, int mb_x, int mb_y, int vp7) | |||
| 1629 | { | |||
| 1630 | switch (mode) { | |||
| 1631 | case DC_PRED8x80: | |||
| 1632 | return check_dc_pred8x8_mode(mode, mb_x, mb_y); | |||
| 1633 | case VERT_PRED8x82: | |||
| 1634 | return !mb_y ? (vp7 ? DC_128_PRED8x86 : DC_127_PRED8x87) : mode; | |||
| 1635 | case HOR_PRED8x81: | |||
| 1636 | return !mb_x ? (vp7 ? DC_128_PRED8x86 : DC_129_PRED8x88) : mode; | |||
| 1637 | case PLANE_PRED8x83: /* TM */ | |||
| 1638 | return check_tm_pred8x8_mode(mode, mb_x, mb_y, vp7); | |||
| 1639 | } | |||
| 1640 | return mode; | |||
| 1641 | } | |||
| 1642 | ||||
| 1643 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1644 | int check_tm_pred4x4_mode(int mode, int mb_x, int mb_y, int vp7) | |||
| 1645 | { | |||
| 1646 | if (!mb_x) { | |||
| 1647 | return mb_y ? VERT_VP8_PRED10 : (vp7 ? DC_128_PRED11 : DC_129_PRED13); | |||
| 1648 | } else { | |||
| 1649 | return mb_y ? mode : HOR_VP8_PRED14; | |||
| 1650 | } | |||
| 1651 | } | |||
| 1652 | ||||
| 1653 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1654 | int check_intra_pred4x4_mode_emuedge(int mode, int mb_x, int mb_y, | |||
| 1655 | int *copy_buf, int vp7) | |||
| 1656 | { | |||
| 1657 | switch (mode) { | |||
| 1658 | case VERT_PRED0: | |||
| 1659 | if (!mb_x && mb_y) { | |||
| 1660 | *copy_buf = 1; | |||
| 1661 | return mode; | |||
| 1662 | } | |||
| 1663 | /* fall-through */ | |||
| 1664 | case DIAG_DOWN_LEFT_PRED3: | |||
| 1665 | case VERT_LEFT_PRED7: | |||
| 1666 | return !mb_y ? (vp7 ? DC_128_PRED11 : DC_127_PRED12) : mode; | |||
| 1667 | case HOR_PRED1: | |||
| 1668 | if (!mb_y) { | |||
| 1669 | *copy_buf = 1; | |||
| 1670 | return mode; | |||
| 1671 | } | |||
| 1672 | /* fall-through */ | |||
| 1673 | case HOR_UP_PRED8: | |||
| 1674 | return !mb_x ? (vp7 ? DC_128_PRED11 : DC_129_PRED13) : mode; | |||
| 1675 | case TM_VP8_PRED9: | |||
| 1676 | return check_tm_pred4x4_mode(mode, mb_x, mb_y, vp7); | |||
| 1677 | case DC_PRED2: /* 4x4 DC doesn't use the same "H.264-style" exceptions | |||
| 1678 | * as 16x16/8x8 DC */ | |||
| 1679 | case DIAG_DOWN_RIGHT_PRED4: | |||
| 1680 | case VERT_RIGHT_PRED5: | |||
| 1681 | case HOR_DOWN_PRED6: | |||
| 1682 | if (!mb_y || !mb_x) | |||
| 1683 | *copy_buf = 1; | |||
| 1684 | return mode; | |||
| 1685 | } | |||
| 1686 | return mode; | |||
| 1687 | } | |||
| 1688 | ||||
| 1689 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1690 | void intra_predict(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], | |||
| 1691 | VP8Macroblock *mb, int mb_x, int mb_y, int is_vp7) | |||
| 1692 | { | |||
| 1693 | int x, y, mode, nnz; | |||
| 1694 | uint32_t tr; | |||
| 1695 | ||||
| 1696 | /* for the first row, we need to run xchg_mb_border to init the top edge | |||
| 1697 | * to 127 otherwise, skip it if we aren't going to deblock */ | |||
| 1698 | if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0) | |||
| 1699 | xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2], | |||
| 1700 | s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width, | |||
| 1701 | s->filter.simple, 1); | |||
| 1702 | ||||
| 1703 | if (mb->mode < MODE_I4x44) { | |||
| 1704 | mode = check_intra_pred8x8_mode_emuedge(mb->mode, mb_x, mb_y, is_vp7); | |||
| 1705 | s->hpc.pred16x16[mode](dst[0], s->linesize); | |||
| 1706 | } else { | |||
| 1707 | uint8_t *ptr = dst[0]; | |||
| 1708 | const uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb; | |||
| 1709 | const uint8_t lo = is_vp7 ? 128 : 127; | |||
| 1710 | const uint8_t hi = is_vp7 ? 128 : 129; | |||
| 1711 | const uint8_t tr_top[4] = { lo, lo, lo, lo }; | |||
| 1712 | ||||
| 1713 | // all blocks on the right edge of the macroblock use bottom edge | |||
| 1714 | // the top macroblock for their topright edge | |||
| 1715 | const uint8_t *tr_right = ptr - s->linesize + 16; | |||
| 1716 | ||||
| 1717 | // if we're on the right edge of the frame, said edge is extended | |||
| 1718 | // from the top macroblock | |||
| 1719 | if (mb_y && mb_x == s->mb_width - 1) { | |||
| 1720 | tr = tr_right[-1] * 0x01010101u; | |||
| 1721 | tr_right = (uint8_t *) &tr; | |||
| 1722 | } | |||
| 1723 | ||||
| 1724 | if (mb->skip) | |||
| 1725 | AV_ZERO128(td->non_zero_count_cache)do { (((av_alias64*)(td->non_zero_count_cache))->u64 = 0 ); (((av_alias64*)((char*)(td->non_zero_count_cache)+8))-> u64 = 0); } while(0); | |||
| 1726 | ||||
| 1727 | for (y = 0; y < 4; y++) { | |||
| 1728 | const uint8_t *topright = ptr + 4 - s->linesize; | |||
| 1729 | for (x = 0; x < 4; x++) { | |||
| 1730 | int copy = 0; | |||
| 1731 | ptrdiff_t linesize = s->linesize; | |||
| 1732 | uint8_t *dst = ptr + 4 * x; | |||
| 1733 | LOCAL_ALIGNED(4, uint8_t, copy_dst, [5 * 8])_Alignas(4) uint8_t la_copy_dst [5 * 8] ; uint8_t (*copy_dst) = la_copy_dst; | |||
| 1734 | ||||
| 1735 | if ((y == 0 || x == 3) && mb_y == 0) { | |||
| 1736 | topright = tr_top; | |||
| 1737 | } else if (x == 3) | |||
| 1738 | topright = tr_right; | |||
| 1739 | ||||
| 1740 | mode = check_intra_pred4x4_mode_emuedge(intra4x4[x], mb_x + x, | |||
| 1741 | mb_y + y, ©, is_vp7); | |||
| 1742 | if (copy) { | |||
| 1743 | dst = copy_dst + 12; | |||
| 1744 | linesize = 8; | |||
| 1745 | if (!(mb_y + y)) { | |||
| 1746 | copy_dst[3] = lo; | |||
| 1747 | AV_WN32A(copy_dst + 4, lo * 0x01010101U)(((av_alias32*)(copy_dst + 4))->u32 = (lo * 0x01010101U)); | |||
| 1748 | } else { | |||
| 1749 | AV_COPY32(copy_dst + 4, ptr + 4 * x - s->linesize)(((av_alias32*)(copy_dst + 4))->u32 = ((const av_alias32*) (ptr + 4 * x - s->linesize))->u32); | |||
| 1750 | if (!(mb_x + x)) { | |||
| 1751 | copy_dst[3] = hi; | |||
| 1752 | } else { | |||
| 1753 | copy_dst[3] = ptr[4 * x - s->linesize - 1]; | |||
| 1754 | } | |||
| 1755 | } | |||
| 1756 | if (!(mb_x + x)) { | |||
| 1757 | copy_dst[11] = | |||
| 1758 | copy_dst[19] = | |||
| 1759 | copy_dst[27] = | |||
| 1760 | copy_dst[35] = hi; | |||
| 1761 | } else { | |||
| 1762 | copy_dst[11] = ptr[4 * x - 1]; | |||
| 1763 | copy_dst[19] = ptr[4 * x + s->linesize - 1]; | |||
| 1764 | copy_dst[27] = ptr[4 * x + s->linesize * 2 - 1]; | |||
| 1765 | copy_dst[35] = ptr[4 * x + s->linesize * 3 - 1]; | |||
| 1766 | } | |||
| 1767 | } | |||
| 1768 | s->hpc.pred4x4[mode](dst, topright, linesize); | |||
| 1769 | if (copy) { | |||
| 1770 | AV_COPY32(ptr + 4 * x, copy_dst + 12)(((av_alias32*)(ptr + 4 * x))->u32 = ((const av_alias32*)( copy_dst + 12))->u32); | |||
| 1771 | AV_COPY32(ptr + 4 * x + s->linesize, copy_dst + 20)(((av_alias32*)(ptr + 4 * x + s->linesize))->u32 = ((const av_alias32*)(copy_dst + 20))->u32); | |||
| 1772 | AV_COPY32(ptr + 4 * x + s->linesize * 2, copy_dst + 28)(((av_alias32*)(ptr + 4 * x + s->linesize * 2))->u32 = ( (const av_alias32*)(copy_dst + 28))->u32); | |||
| 1773 | AV_COPY32(ptr + 4 * x + s->linesize * 3, copy_dst + 36)(((av_alias32*)(ptr + 4 * x + s->linesize * 3))->u32 = ( (const av_alias32*)(copy_dst + 36))->u32); | |||
| 1774 | } | |||
| 1775 | ||||
| 1776 | nnz = td->non_zero_count_cache[y][x]; | |||
| 1777 | if (nnz) { | |||
| 1778 | if (nnz == 1) | |||
| 1779 | s->vp8dsp.vp8_idct_dc_add(ptr + 4 * x, | |||
| 1780 | td->block[y][x], s->linesize); | |||
| 1781 | else | |||
| 1782 | s->vp8dsp.vp8_idct_add(ptr + 4 * x, | |||
| 1783 | td->block[y][x], s->linesize); | |||
| 1784 | } | |||
| 1785 | topright += 4; | |||
| 1786 | } | |||
| 1787 | ||||
| 1788 | ptr += 4 * s->linesize; | |||
| 1789 | intra4x4 += 4; | |||
| 1790 | } | |||
| 1791 | } | |||
| 1792 | ||||
| 1793 | mode = check_intra_pred8x8_mode_emuedge(mb->chroma_pred_mode, | |||
| 1794 | mb_x, mb_y, is_vp7); | |||
| 1795 | s->hpc.pred8x8[mode](dst[1], s->uvlinesize); | |||
| 1796 | s->hpc.pred8x8[mode](dst[2], s->uvlinesize); | |||
| 1797 | ||||
| 1798 | if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0) | |||
| 1799 | xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2], | |||
| 1800 | s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width, | |||
| 1801 | s->filter.simple, 0); | |||
| 1802 | } | |||
| 1803 | ||||
| 1804 | static const uint8_t subpel_idx[3][8] = { | |||
| 1805 | { 0, 1, 2, 1, 2, 1, 2, 1 }, // nr. of left extra pixels, | |||
| 1806 | // also function pointer index | |||
| 1807 | { 0, 3, 5, 3, 5, 3, 5, 3 }, // nr. of extra pixels required | |||
| 1808 | { 0, 2, 3, 2, 3, 2, 3, 2 }, // nr. of right extra pixels | |||
| 1809 | }; | |||
| 1810 | ||||
| 1811 | /** | |||
| 1812 | * luma MC function | |||
| 1813 | * | |||
| 1814 | * @param s VP8 decoding context | |||
| 1815 | * @param dst target buffer for block data at block position | |||
| 1816 | * @param ref reference picture buffer at origin (0, 0) | |||
| 1817 | * @param mv motion vector (relative to block position) to get pixel data from | |||
| 1818 | * @param x_off horizontal position of block from origin (0, 0) | |||
| 1819 | * @param y_off vertical position of block from origin (0, 0) | |||
| 1820 | * @param block_w width of block (16, 8 or 4) | |||
| 1821 | * @param block_h height of block (always same as block_w) | |||
| 1822 | * @param width width of src/dst plane data | |||
| 1823 | * @param height height of src/dst plane data | |||
| 1824 | * @param linesize size of a single line of plane data, including padding | |||
| 1825 | * @param mc_func motion compensation function pointers (bilinear or sixtap MC) | |||
| 1826 | */ | |||
| 1827 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1828 | void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst, | |||
| 1829 | const ProgressFrame *ref, const VP8mv *mv, | |||
| 1830 | int x_off, int y_off, int block_w, int block_h, | |||
| 1831 | int width, int height, ptrdiff_t linesize, | |||
| 1832 | vp8_mc_func mc_func[3][3]) | |||
| 1833 | { | |||
| 1834 | const uint8_t *src = ref->f->data[0]; | |||
| 1835 | ||||
| 1836 | if (AV_RN32A(mv)(((const av_alias32*)(mv))->u32)) { | |||
| 1837 | ptrdiff_t src_linesize = linesize; | |||
| 1838 | ||||
| 1839 | int mx = (mv->x * 2) & 7, mx_idx = subpel_idx[0][mx]; | |||
| 1840 | int my = (mv->y * 2) & 7, my_idx = subpel_idx[0][my]; | |||
| 1841 | ||||
| 1842 | x_off += mv->x >> 2; | |||
| 1843 | y_off += mv->y >> 2; | |||
| 1844 | ||||
| 1845 | // edge emulation | |||
| 1846 | ff_progress_frame_await(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 4); | |||
| 1847 | src += y_off * linesize + x_off; | |||
| 1848 | if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] || | |||
| 1849 | y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) { | |||
| 1850 | s->vdsp.emulated_edge_mc(td->edge_emu_buffer, | |||
| 1851 | src - my_idx * linesize - mx_idx, | |||
| 1852 | EDGE_EMU_LINESIZE32, linesize, | |||
| 1853 | block_w + subpel_idx[1][mx], | |||
| 1854 | block_h + subpel_idx[1][my], | |||
| 1855 | x_off - mx_idx, y_off - my_idx, | |||
| 1856 | width, height); | |||
| 1857 | src = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE32 * my_idx; | |||
| 1858 | src_linesize = EDGE_EMU_LINESIZE32; | |||
| 1859 | } | |||
| 1860 | mc_func[my_idx][mx_idx](dst, linesize, src, src_linesize, block_h, mx, my); | |||
| 1861 | } else { | |||
| 1862 | ff_progress_frame_await(ref, (3 + y_off + block_h) >> 4); | |||
| 1863 | mc_func[0][0](dst, linesize, src + y_off * linesize + x_off, | |||
| 1864 | linesize, block_h, 0, 0); | |||
| 1865 | } | |||
| 1866 | } | |||
| 1867 | ||||
| 1868 | /** | |||
| 1869 | * chroma MC function | |||
| 1870 | * | |||
| 1871 | * @param s VP8 decoding context | |||
| 1872 | * @param dst1 target buffer for block data at block position (U plane) | |||
| 1873 | * @param dst2 target buffer for block data at block position (V plane) | |||
| 1874 | * @param ref reference picture buffer at origin (0, 0) | |||
| 1875 | * @param mv motion vector (relative to block position) to get pixel data from | |||
| 1876 | * @param x_off horizontal position of block from origin (0, 0) | |||
| 1877 | * @param y_off vertical position of block from origin (0, 0) | |||
| 1878 | * @param block_w width of block (16, 8 or 4) | |||
| 1879 | * @param block_h height of block (always same as block_w) | |||
| 1880 | * @param width width of src/dst plane data | |||
| 1881 | * @param height height of src/dst plane data | |||
| 1882 | * @param linesize size of a single line of plane data, including padding | |||
| 1883 | * @param mc_func motion compensation function pointers (bilinear or sixtap MC) | |||
| 1884 | */ | |||
| 1885 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1886 | void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1, | |||
| 1887 | uint8_t *dst2, const ProgressFrame *ref, const VP8mv *mv, | |||
| 1888 | int x_off, int y_off, int block_w, int block_h, | |||
| 1889 | int width, int height, ptrdiff_t linesize, | |||
| 1890 | vp8_mc_func mc_func[3][3]) | |||
| 1891 | { | |||
| 1892 | const uint8_t *src1 = ref->f->data[1], *src2 = ref->f->data[2]; | |||
| 1893 | ||||
| 1894 | if (AV_RN32A(mv)(((const av_alias32*)(mv))->u32)) { | |||
| ||||
| 1895 | int mx = mv->x & 7, mx_idx = subpel_idx[0][mx]; | |||
| 1896 | int my = mv->y & 7, my_idx = subpel_idx[0][my]; | |||
| 1897 | ||||
| 1898 | x_off += mv->x >> 3; | |||
| 1899 | y_off += mv->y >> 3; | |||
| 1900 | ||||
| 1901 | // edge emulation | |||
| 1902 | src1 += y_off * linesize + x_off; | |||
| 1903 | src2 += y_off * linesize + x_off; | |||
| 1904 | ff_progress_frame_await(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 3); | |||
| 1905 | if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] || | |||
| 1906 | y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) { | |||
| 1907 | s->vdsp.emulated_edge_mc(td->edge_emu_buffer, | |||
| 1908 | src1 - my_idx * linesize - mx_idx, | |||
| 1909 | EDGE_EMU_LINESIZE32, linesize, | |||
| 1910 | block_w + subpel_idx[1][mx], | |||
| 1911 | block_h + subpel_idx[1][my], | |||
| 1912 | x_off - mx_idx, y_off - my_idx, width, height); | |||
| 1913 | src1 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE32 * my_idx; | |||
| 1914 | mc_func[my_idx][mx_idx](dst1, linesize, src1, EDGE_EMU_LINESIZE32, block_h, mx, my); | |||
| 1915 | ||||
| 1916 | s->vdsp.emulated_edge_mc(td->edge_emu_buffer, | |||
| 1917 | src2 - my_idx * linesize - mx_idx, | |||
| 1918 | EDGE_EMU_LINESIZE32, linesize, | |||
| 1919 | block_w + subpel_idx[1][mx], | |||
| 1920 | block_h + subpel_idx[1][my], | |||
| 1921 | x_off - mx_idx, y_off - my_idx, width, height); | |||
| 1922 | src2 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE32 * my_idx; | |||
| 1923 | mc_func[my_idx][mx_idx](dst2, linesize, src2, EDGE_EMU_LINESIZE32, block_h, mx, my); | |||
| 1924 | } else { | |||
| 1925 | mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my); | |||
| 1926 | mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my); | |||
| 1927 | } | |||
| 1928 | } else { | |||
| 1929 | ff_progress_frame_await(ref, (3 + y_off + block_h) >> 3); | |||
| 1930 | mc_func[0][0](dst1, linesize, src1 + y_off * linesize + x_off, linesize, block_h, 0, 0); | |||
| 1931 | mc_func[0][0](dst2, linesize, src2 + y_off * linesize + x_off, linesize, block_h, 0, 0); | |||
| 1932 | } | |||
| 1933 | } | |||
| 1934 | ||||
| 1935 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1936 | void vp8_mc_part(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], | |||
| 1937 | const ProgressFrame *ref_frame, int x_off, int y_off, | |||
| 1938 | int bx_off, int by_off, int block_w, int block_h, | |||
| 1939 | int width, int height, const VP8mv *mv) | |||
| 1940 | { | |||
| 1941 | VP8mv uvmv = *mv; | |||
| 1942 | ||||
| 1943 | /* Y */ | |||
| 1944 | vp8_mc_luma(s, td, dst[0] + by_off * s->linesize + bx_off, | |||
| 1945 | ref_frame, mv, x_off + bx_off, y_off + by_off, | |||
| 1946 | block_w, block_h, width, height, s->linesize, | |||
| 1947 | s->put_pixels_tab[block_w == 8]); | |||
| 1948 | ||||
| 1949 | /* U/V */ | |||
| 1950 | if (s->profile == 3) { | |||
| 1951 | /* this block only applies VP8; it is safe to check | |||
| 1952 | * only the profile, as VP7 profile <= 1 */ | |||
| 1953 | uvmv.x &= ~7; | |||
| 1954 | uvmv.y &= ~7; | |||
| 1955 | } | |||
| 1956 | x_off >>= 1; | |||
| 1957 | y_off >>= 1; | |||
| 1958 | bx_off >>= 1; | |||
| 1959 | by_off >>= 1; | |||
| 1960 | width >>= 1; | |||
| 1961 | height >>= 1; | |||
| 1962 | block_w >>= 1; | |||
| 1963 | block_h >>= 1; | |||
| 1964 | vp8_mc_chroma(s, td, dst[1] + by_off * s->uvlinesize + bx_off, | |||
| 1965 | dst[2] + by_off * s->uvlinesize + bx_off, ref_frame, | |||
| 1966 | &uvmv, x_off + bx_off, y_off + by_off, | |||
| 1967 | block_w, block_h, width, height, s->uvlinesize, | |||
| 1968 | s->put_pixels_tab[1 + (block_w == 4)]); | |||
| 1969 | } | |||
| 1970 | ||||
| 1971 | /* Fetch pixels for estimated mv 4 macroblocks ahead. | |||
| 1972 | * Optimized for 64-byte cache lines. Inspired by ffh264 prefetch_motion. */ | |||
| 1973 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1974 | void prefetch_motion(const VP8Context *s, const VP8Macroblock *mb, | |||
| 1975 | int mb_x, int mb_y, int mb_xy, int ref) | |||
| 1976 | { | |||
| 1977 | /* Don't prefetch refs that haven't been used very often this frame. */ | |||
| 1978 | if (s->ref_count[ref - 1] > (mb_xy >> 5)) { | |||
| 1979 | int x_off = mb_x << 4, y_off = mb_y << 4; | |||
| 1980 | int mx = (mb->mv.x >> 2) + x_off + 8; | |||
| 1981 | int my = (mb->mv.y >> 2) + y_off; | |||
| 1982 | uint8_t **src = s->framep[ref]->tf.f->data; | |||
| 1983 | int off = mx + (my + (mb_x & 3) * 4) * s->linesize + 64; | |||
| 1984 | /* For threading, a ff_thread_await_progress here might be useful, but | |||
| 1985 | * it actually slows down the decoder. Since a bad prefetch doesn't | |||
| 1986 | * generate bad decoder output, we don't run it here. */ | |||
| 1987 | s->vdsp.prefetch(src[0] + off, s->linesize, 4); | |||
| 1988 | off = (mx >> 1) + ((my >> 1) + (mb_x & 7)) * s->uvlinesize + 64; | |||
| 1989 | s->vdsp.prefetch(src[1] + off, src[2] - src[1], 2); | |||
| 1990 | } | |||
| 1991 | } | |||
| 1992 | ||||
| 1993 | /** | |||
| 1994 | * Apply motion vectors to prediction buffer, chapter 18. | |||
| 1995 | */ | |||
| 1996 | static av_always_inline__attribute__((always_inline)) inline | |||
| 1997 | void inter_predict(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], | |||
| 1998 | VP8Macroblock *mb, int mb_x, int mb_y) | |||
| 1999 | { | |||
| 2000 | int x_off = mb_x << 4, y_off = mb_y << 4; | |||
| 2001 | int width = 16 * s->mb_width, height = 16 * s->mb_height; | |||
| 2002 | const ProgressFrame *ref = &s->framep[mb->ref_frame]->tf; | |||
| 2003 | const VP8mv *bmv = mb->bmv; | |||
| 2004 | ||||
| 2005 | switch (mb->partitioning) { | |||
| 2006 | case VP8_SPLITMVMODE_NONE: | |||
| 2007 | vp8_mc_part(s, td, dst, ref, x_off, y_off, | |||
| 2008 | 0, 0, 16, 16, width, height, &mb->mv); | |||
| 2009 | break; | |||
| 2010 | case VP8_SPLITMVMODE_4x4: { | |||
| 2011 | int x, y; | |||
| 2012 | VP8mv uvmv; | |||
| 2013 | ||||
| 2014 | /* Y */ | |||
| 2015 | for (y = 0; y < 4; y++) { | |||
| 2016 | for (x = 0; x < 4; x++) { | |||
| 2017 | vp8_mc_luma(s, td, dst[0] + 4 * y * s->linesize + x * 4, | |||
| 2018 | ref, &bmv[4 * y + x], | |||
| 2019 | 4 * x + x_off, 4 * y + y_off, 4, 4, | |||
| 2020 | width, height, s->linesize, | |||
| 2021 | s->put_pixels_tab[2]); | |||
| 2022 | } | |||
| 2023 | } | |||
| 2024 | ||||
| 2025 | /* U/V */ | |||
| 2026 | x_off >>= 1; | |||
| 2027 | y_off >>= 1; | |||
| 2028 | width >>= 1; | |||
| 2029 | height >>= 1; | |||
| 2030 | for (y = 0; y < 2; y++) { | |||
| 2031 | for (x = 0; x < 2; x++) { | |||
| 2032 | uvmv.x = mb->bmv[2 * y * 4 + 2 * x ].x + | |||
| 2033 | mb->bmv[2 * y * 4 + 2 * x + 1].x + | |||
| 2034 | mb->bmv[(2 * y + 1) * 4 + 2 * x ].x + | |||
| 2035 | mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].x; | |||
| 2036 | uvmv.y = mb->bmv[2 * y * 4 + 2 * x ].y + | |||
| 2037 | mb->bmv[2 * y * 4 + 2 * x + 1].y + | |||
| 2038 | mb->bmv[(2 * y + 1) * 4 + 2 * x ].y + | |||
| 2039 | mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].y; | |||
| 2040 | uvmv.x = (uvmv.x + 2 + FF_SIGNBIT(uvmv.x)((uvmv.x) >> 8 * sizeof(uvmv.x) - 1)) >> 2; | |||
| 2041 | uvmv.y = (uvmv.y + 2 + FF_SIGNBIT(uvmv.y)((uvmv.y) >> 8 * sizeof(uvmv.y) - 1)) >> 2; | |||
| 2042 | if (s->profile == 3) { | |||
| 2043 | uvmv.x &= ~7; | |||
| 2044 | uvmv.y &= ~7; | |||
| 2045 | } | |||
| 2046 | vp8_mc_chroma(s, td, dst[1] + 4 * y * s->uvlinesize + x * 4, | |||
| 2047 | dst[2] + 4 * y * s->uvlinesize + x * 4, ref, | |||
| 2048 | &uvmv, 4 * x + x_off, 4 * y + y_off, 4, 4, | |||
| 2049 | width, height, s->uvlinesize, | |||
| 2050 | s->put_pixels_tab[2]); | |||
| 2051 | } | |||
| 2052 | } | |||
| 2053 | break; | |||
| 2054 | } | |||
| 2055 | case VP8_SPLITMVMODE_16x8: | |||
| 2056 | vp8_mc_part(s, td, dst, ref, x_off, y_off, | |||
| 2057 | 0, 0, 16, 8, width, height, &bmv[0]); | |||
| 2058 | vp8_mc_part(s, td, dst, ref, x_off, y_off, | |||
| 2059 | 0, 8, 16, 8, width, height, &bmv[1]); | |||
| 2060 | break; | |||
| 2061 | case VP8_SPLITMVMODE_8x16: | |||
| 2062 | vp8_mc_part(s, td, dst, ref, x_off, y_off, | |||
| 2063 | 0, 0, 8, 16, width, height, &bmv[0]); | |||
| 2064 | vp8_mc_part(s, td, dst, ref, x_off, y_off, | |||
| 2065 | 8, 0, 8, 16, width, height, &bmv[1]); | |||
| 2066 | break; | |||
| 2067 | case VP8_SPLITMVMODE_8x8: | |||
| 2068 | vp8_mc_part(s, td, dst, ref, x_off, y_off, | |||
| 2069 | 0, 0, 8, 8, width, height, &bmv[0]); | |||
| 2070 | vp8_mc_part(s, td, dst, ref, x_off, y_off, | |||
| 2071 | 8, 0, 8, 8, width, height, &bmv[1]); | |||
| 2072 | vp8_mc_part(s, td, dst, ref, x_off, y_off, | |||
| 2073 | 0, 8, 8, 8, width, height, &bmv[2]); | |||
| 2074 | vp8_mc_part(s, td, dst, ref, x_off, y_off, | |||
| 2075 | 8, 8, 8, 8, width, height, &bmv[3]); | |||
| 2076 | break; | |||
| 2077 | } | |||
| 2078 | } | |||
| 2079 | ||||
| 2080 | static av_always_inline__attribute__((always_inline)) inline | |||
| 2081 | void idct_mb(VP8Context *s, VP8ThreadData *td, uint8_t *const dst[3], | |||
| 2082 | const VP8Macroblock *mb) | |||
| 2083 | { | |||
| 2084 | int x, y, ch; | |||
| 2085 | ||||
| 2086 | if (mb->mode != MODE_I4x44) { | |||
| 2087 | uint8_t *y_dst = dst[0]; | |||
| 2088 | for (y = 0; y < 4; y++) { | |||
| 2089 | uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[y])(((const union unaligned_32 *) (td->non_zero_count_cache[y ]))->l); | |||
| 2090 | if (nnz4) { | |||
| 2091 | if (nnz4 & ~0x01010101) { | |||
| 2092 | for (x = 0; x < 4; x++) { | |||
| 2093 | if ((uint8_t) nnz4 == 1) | |||
| 2094 | s->vp8dsp.vp8_idct_dc_add(y_dst + 4 * x, | |||
| 2095 | td->block[y][x], | |||
| 2096 | s->linesize); | |||
| 2097 | else if ((uint8_t) nnz4 > 1) | |||
| 2098 | s->vp8dsp.vp8_idct_add(y_dst + 4 * x, | |||
| 2099 | td->block[y][x], | |||
| 2100 | s->linesize); | |||
| 2101 | nnz4 >>= 8; | |||
| 2102 | if (!nnz4) | |||
| 2103 | break; | |||
| 2104 | } | |||
| 2105 | } else { | |||
| 2106 | s->vp8dsp.vp8_idct_dc_add4y(y_dst, td->block[y], s->linesize); | |||
| 2107 | } | |||
| 2108 | } | |||
| 2109 | y_dst += 4 * s->linesize; | |||
| 2110 | } | |||
| 2111 | } | |||
| 2112 | ||||
| 2113 | for (ch = 0; ch < 2; ch++) { | |||
| 2114 | uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[4 + ch])(((const union unaligned_32 *) (td->non_zero_count_cache[4 + ch]))->l); | |||
| 2115 | if (nnz4) { | |||
| 2116 | uint8_t *ch_dst = dst[1 + ch]; | |||
| 2117 | if (nnz4 & ~0x01010101) { | |||
| 2118 | for (y = 0; y < 2; y++) { | |||
| 2119 | for (x = 0; x < 2; x++) { | |||
| 2120 | if ((uint8_t) nnz4 == 1) | |||
| 2121 | s->vp8dsp.vp8_idct_dc_add(ch_dst + 4 * x, | |||
| 2122 | td->block[4 + ch][(y << 1) + x], | |||
| 2123 | s->uvlinesize); | |||
| 2124 | else if ((uint8_t) nnz4 > 1) | |||
| 2125 | s->vp8dsp.vp8_idct_add(ch_dst + 4 * x, | |||
| 2126 | td->block[4 + ch][(y << 1) + x], | |||
| 2127 | s->uvlinesize); | |||
| 2128 | nnz4 >>= 8; | |||
| 2129 | if (!nnz4) | |||
| 2130 | goto chroma_idct_end; | |||
| 2131 | } | |||
| 2132 | ch_dst += 4 * s->uvlinesize; | |||
| 2133 | } | |||
| 2134 | } else { | |||
| 2135 | s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, td->block[4 + ch], s->uvlinesize); | |||
| 2136 | } | |||
| 2137 | } | |||
| 2138 | chroma_idct_end: | |||
| 2139 | ; | |||
| 2140 | } | |||
| 2141 | } | |||
| 2142 | ||||
| 2143 | static av_always_inline__attribute__((always_inline)) inline | |||
| 2144 | void filter_level_for_mb(const VP8Context *s, const VP8Macroblock *mb, | |||
| 2145 | VP8FilterStrength *f, int is_vp7) | |||
| 2146 | { | |||
| 2147 | int interior_limit, filter_level; | |||
| 2148 | ||||
| 2149 | if (s->segmentation.enabled) { | |||
| 2150 | filter_level = s->segmentation.filter_level[mb->segment]; | |||
| 2151 | if (!s->segmentation.absolute_vals) | |||
| 2152 | filter_level += s->filter.level; | |||
| 2153 | } else | |||
| 2154 | filter_level = s->filter.level; | |||
| 2155 | ||||
| 2156 | if (s->lf_delta.enabled) { | |||
| 2157 | filter_level += s->lf_delta.ref[mb->ref_frame]; | |||
| 2158 | filter_level += s->lf_delta.mode[mb->mode]; | |||
| 2159 | } | |||
| 2160 | ||||
| 2161 | filter_level = av_clip_uintp2av_clip_uintp2_c(filter_level, 6); | |||
| 2162 | ||||
| 2163 | interior_limit = filter_level; | |||
| 2164 | if (s->filter.sharpness) { | |||
| 2165 | interior_limit >>= (s->filter.sharpness + 3) >> 2; | |||
| 2166 | interior_limit = FFMIN(interior_limit, 9 - s->filter.sharpness)((interior_limit) > (9 - s->filter.sharpness) ? (9 - s-> filter.sharpness) : (interior_limit)); | |||
| 2167 | } | |||
| 2168 | interior_limit = FFMAX(interior_limit, 1)((interior_limit) > (1) ? (interior_limit) : (1)); | |||
| 2169 | ||||
| 2170 | f->filter_level = filter_level; | |||
| 2171 | f->inner_limit = interior_limit; | |||
| 2172 | f->inner_filter = is_vp7 || !mb->skip || mb->mode == MODE_I4x44 || | |||
| 2173 | mb->mode == VP8_MVMODE_SPLIT; | |||
| 2174 | } | |||
| 2175 | ||||
| 2176 | static av_always_inline__attribute__((always_inline)) inline | |||
| 2177 | void filter_mb(const VP8Context *s, uint8_t *const dst[3], const VP8FilterStrength *f, | |||
| 2178 | int mb_x, int mb_y, int is_vp7) | |||
| 2179 | { | |||
| 2180 | int mbedge_lim, bedge_lim_y, bedge_lim_uv, hev_thresh; | |||
| 2181 | int filter_level = f->filter_level; | |||
| 2182 | int inner_limit = f->inner_limit; | |||
| 2183 | int inner_filter = f->inner_filter; | |||
| 2184 | ptrdiff_t linesize = s->linesize; | |||
| 2185 | ptrdiff_t uvlinesize = s->uvlinesize; | |||
| 2186 | static const uint8_t hev_thresh_lut[2][64] = { | |||
| 2187 | { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, | |||
| 2188 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |||
| 2189 | 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |||
| 2190 | 3, 3, 3, 3 }, | |||
| 2191 | { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, | |||
| 2192 | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |||
| 2193 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |||
| 2194 | 2, 2, 2, 2 } | |||
| 2195 | }; | |||
| 2196 | ||||
| 2197 | if (!filter_level) | |||
| 2198 | return; | |||
| 2199 | ||||
| 2200 | if (is_vp7) { | |||
| 2201 | bedge_lim_y = filter_level; | |||
| 2202 | bedge_lim_uv = filter_level * 2; | |||
| 2203 | mbedge_lim = filter_level + 2; | |||
| 2204 | } else { | |||
| 2205 | bedge_lim_y = | |||
| 2206 | bedge_lim_uv = filter_level * 2 + inner_limit; | |||
| 2207 | mbedge_lim = bedge_lim_y + 4; | |||
| 2208 | } | |||
| 2209 | ||||
| 2210 | hev_thresh = hev_thresh_lut[s->keyframe][filter_level]; | |||
| 2211 | ||||
| 2212 | if (mb_x) { | |||
| 2213 | s->vp8dsp.vp8_h_loop_filter16y(dst[0], linesize, | |||
| 2214 | mbedge_lim, inner_limit, hev_thresh); | |||
| 2215 | s->vp8dsp.vp8_h_loop_filter8uv(dst[1], dst[2], uvlinesize, | |||
| 2216 | mbedge_lim, inner_limit, hev_thresh); | |||
| 2217 | } | |||
| 2218 | ||||
| 2219 | #define H_LOOP_FILTER_16Y_INNER(cond)if (cond && inner_filter) { s->vp8dsp.vp8_h_loop_filter16y_inner (dst[0] + 4, linesize, bedge_lim_y, inner_limit, hev_thresh); s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 8, linesize , bedge_lim_y, inner_limit, hev_thresh); s->vp8dsp.vp8_h_loop_filter16y_inner (dst[0] + 12, linesize, bedge_lim_y, inner_limit, hev_thresh) ; s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4, uvlinesize, bedge_lim_uv, inner_limit, hev_thresh); } \ | |||
| 2220 | if (cond && inner_filter) { \ | |||
| 2221 | s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 4, linesize, \ | |||
| 2222 | bedge_lim_y, inner_limit, \ | |||
| 2223 | hev_thresh); \ | |||
| 2224 | s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 8, linesize, \ | |||
| 2225 | bedge_lim_y, inner_limit, \ | |||
| 2226 | hev_thresh); \ | |||
| 2227 | s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 12, linesize, \ | |||
| 2228 | bedge_lim_y, inner_limit, \ | |||
| 2229 | hev_thresh); \ | |||
| 2230 | s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4, \ | |||
| 2231 | uvlinesize, bedge_lim_uv, \ | |||
| 2232 | inner_limit, hev_thresh); \ | |||
| 2233 | } | |||
| 2234 | ||||
| 2235 | H_LOOP_FILTER_16Y_INNER(!is_vp7)if (!is_vp7 && inner_filter) { s->vp8dsp.vp8_h_loop_filter16y_inner (dst[0] + 4, linesize, bedge_lim_y, inner_limit, hev_thresh); s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 8, linesize , bedge_lim_y, inner_limit, hev_thresh); s->vp8dsp.vp8_h_loop_filter16y_inner (dst[0] + 12, linesize, bedge_lim_y, inner_limit, hev_thresh) ; s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4, uvlinesize, bedge_lim_uv, inner_limit, hev_thresh); } | |||
| 2236 | ||||
| 2237 | if (mb_y) { | |||
| 2238 | s->vp8dsp.vp8_v_loop_filter16y(dst[0], linesize, | |||
| 2239 | mbedge_lim, inner_limit, hev_thresh); | |||
| 2240 | s->vp8dsp.vp8_v_loop_filter8uv(dst[1], dst[2], uvlinesize, | |||
| 2241 | mbedge_lim, inner_limit, hev_thresh); | |||
| 2242 | } | |||
| 2243 | ||||
| 2244 | if (inner_filter) { | |||
| 2245 | s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 4 * linesize, | |||
| 2246 | linesize, bedge_lim_y, | |||
| 2247 | inner_limit, hev_thresh); | |||
| 2248 | s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 8 * linesize, | |||
| 2249 | linesize, bedge_lim_y, | |||
| 2250 | inner_limit, hev_thresh); | |||
| 2251 | s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 12 * linesize, | |||
| 2252 | linesize, bedge_lim_y, | |||
| 2253 | inner_limit, hev_thresh); | |||
| 2254 | s->vp8dsp.vp8_v_loop_filter8uv_inner(dst[1] + 4 * uvlinesize, | |||
| 2255 | dst[2] + 4 * uvlinesize, | |||
| 2256 | uvlinesize, bedge_lim_uv, | |||
| 2257 | inner_limit, hev_thresh); | |||
| 2258 | } | |||
| 2259 | ||||
| 2260 | H_LOOP_FILTER_16Y_INNER(is_vp7)if (is_vp7 && inner_filter) { s->vp8dsp.vp8_h_loop_filter16y_inner (dst[0] + 4, linesize, bedge_lim_y, inner_limit, hev_thresh); s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 8, linesize , bedge_lim_y, inner_limit, hev_thresh); s->vp8dsp.vp8_h_loop_filter16y_inner (dst[0] + 12, linesize, bedge_lim_y, inner_limit, hev_thresh) ; s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4, uvlinesize, bedge_lim_uv, inner_limit, hev_thresh); } | |||
| 2261 | } | |||
| 2262 | ||||
| 2263 | static av_always_inline__attribute__((always_inline)) inline | |||
| 2264 | void filter_mb_simple(const VP8Context *s, uint8_t *dst, const VP8FilterStrength *f, | |||
| 2265 | int mb_x, int mb_y) | |||
| 2266 | { | |||
| 2267 | int mbedge_lim, bedge_lim; | |||
| 2268 | int filter_level = f->filter_level; | |||
| 2269 | int inner_limit = f->inner_limit; | |||
| 2270 | int inner_filter = f->inner_filter; | |||
| 2271 | ptrdiff_t linesize = s->linesize; | |||
| 2272 | ||||
| 2273 | if (!filter_level) | |||
| 2274 | return; | |||
| 2275 | ||||
| 2276 | bedge_lim = 2 * filter_level + inner_limit; | |||
| 2277 | mbedge_lim = bedge_lim + 4; | |||
| 2278 | ||||
| 2279 | if (mb_x) | |||
| 2280 | s->vp8dsp.vp8_h_loop_filter_simple(dst, linesize, mbedge_lim); | |||
| 2281 | if (inner_filter) { | |||
| 2282 | s->vp8dsp.vp8_h_loop_filter_simple(dst + 4, linesize, bedge_lim); | |||
| 2283 | s->vp8dsp.vp8_h_loop_filter_simple(dst + 8, linesize, bedge_lim); | |||
| 2284 | s->vp8dsp.vp8_h_loop_filter_simple(dst + 12, linesize, bedge_lim); | |||
| 2285 | } | |||
| 2286 | ||||
| 2287 | if (mb_y) | |||
| 2288 | s->vp8dsp.vp8_v_loop_filter_simple(dst, linesize, mbedge_lim); | |||
| 2289 | if (inner_filter) { | |||
| 2290 | s->vp8dsp.vp8_v_loop_filter_simple(dst + 4 * linesize, linesize, bedge_lim); | |||
| 2291 | s->vp8dsp.vp8_v_loop_filter_simple(dst + 8 * linesize, linesize, bedge_lim); | |||
| 2292 | s->vp8dsp.vp8_v_loop_filter_simple(dst + 12 * linesize, linesize, bedge_lim); | |||
| 2293 | } | |||
| 2294 | } | |||
| 2295 | ||||
| 2296 | #define MARGIN(16 << 2) (16 << 2) | |||
| 2297 | static av_always_inline__attribute__((always_inline)) inline | |||
| 2298 | int vp78_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *curframe, | |||
| 2299 | const VP8Frame *prev_frame, int is_vp7) | |||
| 2300 | { | |||
| 2301 | VP8Context *s = avctx->priv_data; | |||
| 2302 | int mb_x, mb_y; | |||
| 2303 | ||||
| 2304 | s->mv_bounds.mv_min.y = -MARGIN(16 << 2); | |||
| 2305 | s->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN(16 << 2); | |||
| 2306 | for (mb_y = 0; mb_y < s->mb_height; mb_y++) { | |||
| 2307 | VP8Macroblock *mb = s->macroblocks_base + | |||
| 2308 | ((s->mb_width + 1) * (mb_y + 1) + 1); | |||
| 2309 | int mb_xy = mb_y * s->mb_width; | |||
| 2310 | ||||
| 2311 | AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101)(((av_alias32*)(s->intra4x4_pred_mode_left))->u32 = (2 * 0x01010101)); | |||
| 2312 | ||||
| 2313 | s->mv_bounds.mv_min.x = -MARGIN(16 << 2); | |||
| 2314 | s->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN(16 << 2); | |||
| 2315 | ||||
| 2316 | for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) { | |||
| 2317 | if (vpx_rac_is_end(&s->c)) { | |||
| 2318 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 2319 | } | |||
| 2320 | if (mb_y == 0) | |||
| 2321 | AV_WN32A((mb - s->mb_width - 1)->intra4x4_pred_mode_top,(((av_alias32*)((mb - s->mb_width - 1)->intra4x4_pred_mode_top ))->u32 = (2 * 0x01010101)) | |||
| 2322 | DC_PRED * 0x01010101)(((av_alias32*)((mb - s->mb_width - 1)->intra4x4_pred_mode_top ))->u32 = (2 * 0x01010101)); | |||
| 2323 | decode_mb_mode(s, &s->mv_bounds, mb, mb_x, mb_y, curframe->seg_map + mb_xy, | |||
| 2324 | prev_frame && prev_frame->seg_map ? | |||
| 2325 | prev_frame->seg_map + mb_xy : NULL((void*)0), 1, is_vp7); | |||
| 2326 | s->mv_bounds.mv_min.x -= 64; | |||
| 2327 | s->mv_bounds.mv_max.x -= 64; | |||
| 2328 | } | |||
| 2329 | s->mv_bounds.mv_min.y -= 64; | |||
| 2330 | s->mv_bounds.mv_max.y -= 64; | |||
| 2331 | } | |||
| 2332 | return 0; | |||
| 2333 | } | |||
| 2334 | ||||
| 2335 | static int vp7_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame, | |||
| 2336 | const VP8Frame *prev_frame) | |||
| 2337 | { | |||
| 2338 | return vp78_decode_mv_mb_modes(avctx, cur_frame, prev_frame, IS_VP71); | |||
| 2339 | } | |||
| 2340 | ||||
| 2341 | static int vp8_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame, | |||
| 2342 | const VP8Frame *prev_frame) | |||
| 2343 | { | |||
| 2344 | return vp78_decode_mv_mb_modes(avctx, cur_frame, prev_frame, IS_VP80); | |||
| 2345 | } | |||
| 2346 | ||||
| 2347 | #if HAVE_THREADS1 | |||
| 2348 | #define check_thread_pos(td, otd, mb_x_check, mb_y_check)do { int tmp = (mb_y_check << 16) | (mb_x_check & 0xFFFF ); if (__c11_atomic_load(&otd->thread_mb_pos, 5) < tmp ) { strict_pthread_mutex_lock(&otd->lock); __c11_atomic_store (&td->wait_mb_pos, tmp, 5); do { if (__c11_atomic_load (&otd->thread_mb_pos, 5) >= tmp) break; strict_pthread_cond_wait (&otd->cond, &otd->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&otd->lock); } } while (0) \ | |||
| 2349 | do { \ | |||
| 2350 | int tmp = (mb_y_check << 16) | (mb_x_check & 0xFFFF); \ | |||
| 2351 | if (atomic_load(&otd->thread_mb_pos)__c11_atomic_load(&otd->thread_mb_pos, 5) < tmp) { \ | |||
| 2352 | pthread_mutex_lockstrict_pthread_mutex_lock(&otd->lock); \ | |||
| 2353 | atomic_store(&td->wait_mb_pos, tmp)__c11_atomic_store(&td->wait_mb_pos, tmp, 5); \ | |||
| 2354 | do { \ | |||
| 2355 | if (atomic_load(&otd->thread_mb_pos)__c11_atomic_load(&otd->thread_mb_pos, 5) >= tmp) \ | |||
| 2356 | break; \ | |||
| 2357 | pthread_cond_waitstrict_pthread_cond_wait(&otd->cond, &otd->lock); \ | |||
| 2358 | } while (1); \ | |||
| 2359 | atomic_store(&td->wait_mb_pos, INT_MAX)__c11_atomic_store(&td->wait_mb_pos, 2147483647, 5); \ | |||
| 2360 | pthread_mutex_unlockstrict_pthread_mutex_unlock(&otd->lock); \ | |||
| 2361 | } \ | |||
| 2362 | } while (0) | |||
| 2363 | ||||
| 2364 | #define update_pos(td, mb_y, mb_x)do { int pos = (mb_y << 16) | (mb_x & 0xFFFF); int sliced_threading = (avctx->active_thread_type == 2) && (num_jobs > 1); int is_null = !next_td || !prev_td; int pos_check = (is_null ) ? 1 : (next_td != td && pos >= __c11_atomic_load (&next_td->wait_mb_pos, 5)) || (prev_td != td && pos >= __c11_atomic_load(&prev_td->wait_mb_pos, 5) ); __c11_atomic_store(&td->thread_mb_pos, pos, 5); if ( sliced_threading && pos_check) { strict_pthread_mutex_lock (&td->lock); strict_pthread_cond_broadcast(&td-> cond); strict_pthread_mutex_unlock(&td->lock); } } while (0) \ | |||
| 2365 | do { \ | |||
| 2366 | int pos = (mb_y << 16) | (mb_x & 0xFFFF); \ | |||
| 2367 | int sliced_threading = (avctx->active_thread_type == FF_THREAD_SLICE2) && \ | |||
| 2368 | (num_jobs > 1); \ | |||
| 2369 | int is_null = !next_td || !prev_td; \ | |||
| 2370 | int pos_check = (is_null) ? 1 : \ | |||
| 2371 | (next_td != td && pos >= atomic_load(&next_td->wait_mb_pos)__c11_atomic_load(&next_td->wait_mb_pos, 5)) || \ | |||
| 2372 | (prev_td != td && pos >= atomic_load(&prev_td->wait_mb_pos)__c11_atomic_load(&prev_td->wait_mb_pos, 5)); \ | |||
| 2373 | atomic_store(&td->thread_mb_pos, pos)__c11_atomic_store(&td->thread_mb_pos, pos, 5); \ | |||
| 2374 | if (sliced_threading && pos_check) { \ | |||
| 2375 | pthread_mutex_lockstrict_pthread_mutex_lock(&td->lock); \ | |||
| 2376 | pthread_cond_broadcaststrict_pthread_cond_broadcast(&td->cond); \ | |||
| 2377 | pthread_mutex_unlockstrict_pthread_mutex_unlock(&td->lock); \ | |||
| 2378 | } \ | |||
| 2379 | } while (0) | |||
| 2380 | #else | |||
| 2381 | #define check_thread_pos(td, otd, mb_x_check, mb_y_check)do { int tmp = (mb_y_check << 16) | (mb_x_check & 0xFFFF ); if (__c11_atomic_load(&otd->thread_mb_pos, 5) < tmp ) { strict_pthread_mutex_lock(&otd->lock); __c11_atomic_store (&td->wait_mb_pos, tmp, 5); do { if (__c11_atomic_load (&otd->thread_mb_pos, 5) >= tmp) break; strict_pthread_cond_wait (&otd->cond, &otd->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&otd->lock); } } while (0) while(0) | |||
| 2382 | #define update_pos(td, mb_y, mb_x)do { int pos = (mb_y << 16) | (mb_x & 0xFFFF); int sliced_threading = (avctx->active_thread_type == 2) && (num_jobs > 1); int is_null = !next_td || !prev_td; int pos_check = (is_null ) ? 1 : (next_td != td && pos >= __c11_atomic_load (&next_td->wait_mb_pos, 5)) || (prev_td != td && pos >= __c11_atomic_load(&prev_td->wait_mb_pos, 5) ); __c11_atomic_store(&td->thread_mb_pos, pos, 5); if ( sliced_threading && pos_check) { strict_pthread_mutex_lock (&td->lock); strict_pthread_cond_broadcast(&td-> cond); strict_pthread_mutex_unlock(&td->lock); } } while (0) while(0) | |||
| 2383 | #endif | |||
| 2384 | ||||
| 2385 | static av_always_inline__attribute__((always_inline)) inline int decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata, | |||
| 2386 | int jobnr, int threadnr, int is_vp7) | |||
| 2387 | { | |||
| 2388 | VP8Context *s = avctx->priv_data; | |||
| 2389 | VP8ThreadData *prev_td, *next_td, *td = &s->thread_data[threadnr]; | |||
| 2390 | int mb_y = atomic_load(&td->thread_mb_pos)__c11_atomic_load(&td->thread_mb_pos, 5) >> 16; | |||
| 2391 | int mb_x, mb_xy = mb_y * s->mb_width; | |||
| 2392 | int num_jobs = s->num_jobs; | |||
| 2393 | const VP8Frame *prev_frame = s->prev_frame; | |||
| 2394 | VP8Frame *curframe = s->curframe; | |||
| 2395 | VPXRangeCoder *coeff_c = &s->coeff_partition[mb_y & (s->num_coeff_partitions - 1)]; | |||
| 2396 | ||||
| 2397 | VP8Macroblock *mb; | |||
| 2398 | uint8_t *dst[3] = { | |||
| 2399 | curframe->tf.f->data[0] + 16 * mb_y * s->linesize, | |||
| 2400 | curframe->tf.f->data[1] + 8 * mb_y * s->uvlinesize, | |||
| 2401 | curframe->tf.f->data[2] + 8 * mb_y * s->uvlinesize | |||
| 2402 | }; | |||
| 2403 | ||||
| 2404 | if (vpx_rac_is_end(&s->c)) | |||
| ||||
| 2405 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 2406 | ||||
| 2407 | if (mb_y == 0) | |||
| 2408 | prev_td = td; | |||
| 2409 | else | |||
| 2410 | prev_td = &s->thread_data[(jobnr + num_jobs - 1) % num_jobs]; | |||
| 2411 | if (mb_y == s->mb_height - 1) | |||
| 2412 | next_td = td; | |||
| 2413 | else | |||
| 2414 | next_td = &s->thread_data[(jobnr + 1) % num_jobs]; | |||
| 2415 | if (s->mb_layout == 1) | |||
| 2416 | mb = s->macroblocks_base + ((s->mb_width + 1) * (mb_y + 1) + 1); | |||
| 2417 | else { | |||
| 2418 | // Make sure the previous frame has read its segmentation map, | |||
| 2419 | // if we reuse the same map. | |||
| 2420 | if (prev_frame && s->segmentation.enabled && | |||
| 2421 | !s->segmentation.update_map) | |||
| 2422 | ff_progress_frame_await(&prev_frame->tf, mb_y); | |||
| 2423 | mb = s->macroblocks + (s->mb_height - mb_y - 1) * 2; | |||
| 2424 | memset(mb - 1, 0, sizeof(*mb)); // zero left macroblock | |||
| 2425 | AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101)(((av_alias32*)(s->intra4x4_pred_mode_left))->u32 = (2 * 0x01010101)); | |||
| 2426 | } | |||
| 2427 | ||||
| 2428 | if (!is_vp7 || mb_y
| |||
| 2429 | memset(td->left_nnz, 0, sizeof(td->left_nnz)); | |||
| 2430 | ||||
| 2431 | td->mv_bounds.mv_min.x = -MARGIN(16 << 2); | |||
| 2432 | td->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN(16 << 2); | |||
| 2433 | ||||
| 2434 | for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) { | |||
| 2435 | if (vpx_rac_is_end(&s->c)) | |||
| 2436 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 2437 | // Wait for previous thread to read mb_x+2, and reach mb_y-1. | |||
| 2438 | if (prev_td != td) { | |||
| 2439 | if (threadnr != 0) { | |||
| 2440 | check_thread_pos(td, prev_td,do { int tmp = (mb_y - (is_vp7 ? 2 : 1) << 16) | (mb_x + (is_vp7 ? 2 : 1) & 0xFFFF); if (__c11_atomic_load(&prev_td ->thread_mb_pos, 5) < tmp) { strict_pthread_mutex_lock( &prev_td->lock); __c11_atomic_store(&td->wait_mb_pos , tmp, 5); do { if (__c11_atomic_load(&prev_td->thread_mb_pos , 5) >= tmp) break; strict_pthread_cond_wait(&prev_td-> cond, &prev_td->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&prev_td->lock); } } while (0) | |||
| 2441 | mb_x + (is_vp7 ? 2 : 1),do { int tmp = (mb_y - (is_vp7 ? 2 : 1) << 16) | (mb_x + (is_vp7 ? 2 : 1) & 0xFFFF); if (__c11_atomic_load(&prev_td ->thread_mb_pos, 5) < tmp) { strict_pthread_mutex_lock( &prev_td->lock); __c11_atomic_store(&td->wait_mb_pos , tmp, 5); do { if (__c11_atomic_load(&prev_td->thread_mb_pos , 5) >= tmp) break; strict_pthread_cond_wait(&prev_td-> cond, &prev_td->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&prev_td->lock); } } while (0) | |||
| 2442 | mb_y - (is_vp7 ? 2 : 1))do { int tmp = (mb_y - (is_vp7 ? 2 : 1) << 16) | (mb_x + (is_vp7 ? 2 : 1) & 0xFFFF); if (__c11_atomic_load(&prev_td ->thread_mb_pos, 5) < tmp) { strict_pthread_mutex_lock( &prev_td->lock); __c11_atomic_store(&td->wait_mb_pos , tmp, 5); do { if (__c11_atomic_load(&prev_td->thread_mb_pos , 5) >= tmp) break; strict_pthread_cond_wait(&prev_td-> cond, &prev_td->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&prev_td->lock); } } while (0); | |||
| 2443 | } else { | |||
| 2444 | check_thread_pos(td, prev_td,do { int tmp = (mb_y - (is_vp7 ? 2 : 1) << 16) | (mb_x + (is_vp7 ? 2 : 1) + s->mb_width + 3 & 0xFFFF); if (__c11_atomic_load (&prev_td->thread_mb_pos, 5) < tmp) { strict_pthread_mutex_lock (&prev_td->lock); __c11_atomic_store(&td->wait_mb_pos , tmp, 5); do { if (__c11_atomic_load(&prev_td->thread_mb_pos , 5) >= tmp) break; strict_pthread_cond_wait(&prev_td-> cond, &prev_td->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&prev_td->lock); } } while (0) | |||
| 2445 | mb_x + (is_vp7 ? 2 : 1) + s->mb_width + 3,do { int tmp = (mb_y - (is_vp7 ? 2 : 1) << 16) | (mb_x + (is_vp7 ? 2 : 1) + s->mb_width + 3 & 0xFFFF); if (__c11_atomic_load (&prev_td->thread_mb_pos, 5) < tmp) { strict_pthread_mutex_lock (&prev_td->lock); __c11_atomic_store(&td->wait_mb_pos , tmp, 5); do { if (__c11_atomic_load(&prev_td->thread_mb_pos , 5) >= tmp) break; strict_pthread_cond_wait(&prev_td-> cond, &prev_td->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&prev_td->lock); } } while (0) | |||
| 2446 | mb_y - (is_vp7 ? 2 : 1))do { int tmp = (mb_y - (is_vp7 ? 2 : 1) << 16) | (mb_x + (is_vp7 ? 2 : 1) + s->mb_width + 3 & 0xFFFF); if (__c11_atomic_load (&prev_td->thread_mb_pos, 5) < tmp) { strict_pthread_mutex_lock (&prev_td->lock); __c11_atomic_store(&td->wait_mb_pos , tmp, 5); do { if (__c11_atomic_load(&prev_td->thread_mb_pos , 5) >= tmp) break; strict_pthread_cond_wait(&prev_td-> cond, &prev_td->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&prev_td->lock); } } while (0); | |||
| 2447 | } | |||
| 2448 | } | |||
| 2449 | ||||
| 2450 | s->vdsp.prefetch(dst[0] + (mb_x & 3) * 4 * s->linesize + 64, | |||
| 2451 | s->linesize, 4); | |||
| 2452 | s->vdsp.prefetch(dst[1] + (mb_x & 7) * s->uvlinesize + 64, | |||
| 2453 | dst[2] - dst[1], 2); | |||
| 2454 | ||||
| 2455 | if (!s->mb_layout) | |||
| 2456 | decode_mb_mode(s, &td->mv_bounds, mb, mb_x, mb_y, curframe->seg_map + mb_xy, | |||
| 2457 | prev_frame && prev_frame->seg_map ? | |||
| 2458 | prev_frame->seg_map + mb_xy : NULL((void*)0), 0, is_vp7); | |||
| 2459 | ||||
| 2460 | prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP8_FRAME_PREVIOUS); | |||
| 2461 | ||||
| 2462 | if (!mb->skip) { | |||
| 2463 | if (vpx_rac_is_end(coeff_c)) | |||
| 2464 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 2465 | decode_mb_coeffs(s, td, coeff_c, mb, s->top_nnz[mb_x], td->left_nnz, is_vp7); | |||
| 2466 | } | |||
| 2467 | ||||
| 2468 | if (mb->mode <= MODE_I4x44) | |||
| 2469 | intra_predict(s, td, dst, mb, mb_x, mb_y, is_vp7); | |||
| 2470 | else | |||
| 2471 | inter_predict(s, td, dst, mb, mb_x, mb_y); | |||
| 2472 | ||||
| 2473 | prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP8_FRAME_GOLDEN); | |||
| 2474 | ||||
| 2475 | if (!mb->skip) { | |||
| 2476 | idct_mb(s, td, dst, mb); | |||
| 2477 | } else { | |||
| 2478 | AV_ZERO64(td->left_nnz)(((av_alias64*)(td->left_nnz))->u64 = 0); | |||
| 2479 | AV_WN64(s->top_nnz[mb_x], 0)((((union unaligned_64 *) (s->top_nnz[mb_x]))->l) = (0) ); // array of 9, so unaligned | |||
| 2480 | ||||
| 2481 | /* Reset DC block predictors if they would exist | |||
| 2482 | * if the mb had coefficients */ | |||
| 2483 | if (mb->mode != MODE_I4x44 && mb->mode != VP8_MVMODE_SPLIT) { | |||
| 2484 | td->left_nnz[8] = 0; | |||
| 2485 | s->top_nnz[mb_x][8] = 0; | |||
| 2486 | } | |||
| 2487 | } | |||
| 2488 | ||||
| 2489 | if (s->deblock_filter) | |||
| 2490 | filter_level_for_mb(s, mb, &td->filter_strength[mb_x], is_vp7); | |||
| 2491 | ||||
| 2492 | if (s->deblock_filter && num_jobs != 1 && threadnr == num_jobs - 1) { | |||
| 2493 | if (s->filter.simple) | |||
| 2494 | backup_mb_border(s->top_border[mb_x + 1], dst[0], | |||
| 2495 | NULL((void*)0), NULL((void*)0), s->linesize, 0, 1); | |||
| 2496 | else | |||
| 2497 | backup_mb_border(s->top_border[mb_x + 1], dst[0], | |||
| 2498 | dst[1], dst[2], s->linesize, s->uvlinesize, 0); | |||
| 2499 | } | |||
| 2500 | ||||
| 2501 | prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP8_FRAME_ALTREF); | |||
| 2502 | ||||
| 2503 | dst[0] += 16; | |||
| 2504 | dst[1] += 8; | |||
| 2505 | dst[2] += 8; | |||
| 2506 | td->mv_bounds.mv_min.x -= 64; | |||
| 2507 | td->mv_bounds.mv_max.x -= 64; | |||
| 2508 | ||||
| 2509 | if (mb_x == s->mb_width + 1) { | |||
| 2510 | update_pos(td, mb_y, s->mb_width + 3)do { int pos = (mb_y << 16) | (s->mb_width + 3 & 0xFFFF); int sliced_threading = (avctx->active_thread_type == 2) && (num_jobs > 1); int is_null = !next_td || !prev_td; int pos_check = (is_null) ? 1 : (next_td != td && pos >= __c11_atomic_load(&next_td->wait_mb_pos, 5) ) || (prev_td != td && pos >= __c11_atomic_load(& prev_td->wait_mb_pos, 5)); __c11_atomic_store(&td-> thread_mb_pos, pos, 5); if (sliced_threading && pos_check ) { strict_pthread_mutex_lock(&td->lock); strict_pthread_cond_broadcast (&td->cond); strict_pthread_mutex_unlock(&td->lock ); } } while (0); | |||
| 2511 | } else { | |||
| 2512 | update_pos(td, mb_y, mb_x)do { int pos = (mb_y << 16) | (mb_x & 0xFFFF); int sliced_threading = (avctx->active_thread_type == 2) && (num_jobs > 1); int is_null = !next_td || !prev_td; int pos_check = (is_null ) ? 1 : (next_td != td && pos >= __c11_atomic_load (&next_td->wait_mb_pos, 5)) || (prev_td != td && pos >= __c11_atomic_load(&prev_td->wait_mb_pos, 5) ); __c11_atomic_store(&td->thread_mb_pos, pos, 5); if ( sliced_threading && pos_check) { strict_pthread_mutex_lock (&td->lock); strict_pthread_cond_broadcast(&td-> cond); strict_pthread_mutex_unlock(&td->lock); } } while (0); | |||
| 2513 | } | |||
| 2514 | } | |||
| 2515 | return 0; | |||
| 2516 | } | |||
| 2517 | ||||
| 2518 | static av_always_inline__attribute__((always_inline)) inline void filter_mb_row(AVCodecContext *avctx, void *tdata, | |||
| 2519 | int jobnr, int threadnr, int is_vp7) | |||
| 2520 | { | |||
| 2521 | VP8Context *s = avctx->priv_data; | |||
| 2522 | VP8ThreadData *td = &s->thread_data[threadnr]; | |||
| 2523 | int mb_x, mb_y = atomic_load(&td->thread_mb_pos)__c11_atomic_load(&td->thread_mb_pos, 5) >> 16, num_jobs = s->num_jobs; | |||
| 2524 | AVFrame *curframe = s->curframe->tf.f; | |||
| 2525 | VP8ThreadData *prev_td, *next_td; | |||
| 2526 | uint8_t *dst[3] = { | |||
| 2527 | curframe->data[0] + 16 * mb_y * s->linesize, | |||
| 2528 | curframe->data[1] + 8 * mb_y * s->uvlinesize, | |||
| 2529 | curframe->data[2] + 8 * mb_y * s->uvlinesize | |||
| 2530 | }; | |||
| 2531 | ||||
| 2532 | if (mb_y == 0) | |||
| 2533 | prev_td = td; | |||
| 2534 | else | |||
| 2535 | prev_td = &s->thread_data[(jobnr + num_jobs - 1) % num_jobs]; | |||
| 2536 | if (mb_y == s->mb_height - 1) | |||
| 2537 | next_td = td; | |||
| 2538 | else | |||
| 2539 | next_td = &s->thread_data[(jobnr + 1) % num_jobs]; | |||
| 2540 | ||||
| 2541 | for (mb_x = 0; mb_x < s->mb_width; mb_x++) { | |||
| 2542 | const VP8FilterStrength *f = &td->filter_strength[mb_x]; | |||
| 2543 | if (prev_td != td) | |||
| 2544 | check_thread_pos(td, prev_td,do { int tmp = (mb_y - 1 << 16) | ((mb_x + 1) + (s-> mb_width + 3) & 0xFFFF); if (__c11_atomic_load(&prev_td ->thread_mb_pos, 5) < tmp) { strict_pthread_mutex_lock( &prev_td->lock); __c11_atomic_store(&td->wait_mb_pos , tmp, 5); do { if (__c11_atomic_load(&prev_td->thread_mb_pos , 5) >= tmp) break; strict_pthread_cond_wait(&prev_td-> cond, &prev_td->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&prev_td->lock); } } while (0) | |||
| 2545 | (mb_x + 1) + (s->mb_width + 3), mb_y - 1)do { int tmp = (mb_y - 1 << 16) | ((mb_x + 1) + (s-> mb_width + 3) & 0xFFFF); if (__c11_atomic_load(&prev_td ->thread_mb_pos, 5) < tmp) { strict_pthread_mutex_lock( &prev_td->lock); __c11_atomic_store(&td->wait_mb_pos , tmp, 5); do { if (__c11_atomic_load(&prev_td->thread_mb_pos , 5) >= tmp) break; strict_pthread_cond_wait(&prev_td-> cond, &prev_td->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&prev_td->lock); } } while (0); | |||
| 2546 | if (next_td != td) | |||
| 2547 | if (next_td != &s->thread_data[0]) | |||
| 2548 | check_thread_pos(td, next_td, mb_x + 1, mb_y + 1)do { int tmp = (mb_y + 1 << 16) | (mb_x + 1 & 0xFFFF ); if (__c11_atomic_load(&next_td->thread_mb_pos, 5) < tmp) { strict_pthread_mutex_lock(&next_td->lock); __c11_atomic_store (&td->wait_mb_pos, tmp, 5); do { if (__c11_atomic_load (&next_td->thread_mb_pos, 5) >= tmp) break; strict_pthread_cond_wait (&next_td->cond, &next_td->lock); } while (1); __c11_atomic_store (&td->wait_mb_pos, 2147483647, 5); strict_pthread_mutex_unlock (&next_td->lock); } } while (0); | |||
| 2549 | ||||
| 2550 | if (num_jobs == 1) { | |||
| 2551 | if (s->filter.simple) | |||
| 2552 | backup_mb_border(s->top_border[mb_x + 1], dst[0], | |||
| 2553 | NULL((void*)0), NULL((void*)0), s->linesize, 0, 1); | |||
| 2554 | else | |||
| 2555 | backup_mb_border(s->top_border[mb_x + 1], dst[0], | |||
| 2556 | dst[1], dst[2], s->linesize, s->uvlinesize, 0); | |||
| 2557 | } | |||
| 2558 | ||||
| 2559 | if (s->filter.simple) | |||
| 2560 | filter_mb_simple(s, dst[0], f, mb_x, mb_y); | |||
| 2561 | else | |||
| 2562 | filter_mb(s, dst, f, mb_x, mb_y, is_vp7); | |||
| 2563 | dst[0] += 16; | |||
| 2564 | dst[1] += 8; | |||
| 2565 | dst[2] += 8; | |||
| 2566 | ||||
| 2567 | update_pos(td, mb_y, (s->mb_width + 3) + mb_x)do { int pos = (mb_y << 16) | ((s->mb_width + 3) + mb_x & 0xFFFF); int sliced_threading = (avctx->active_thread_type == 2) && (num_jobs > 1); int is_null = !next_td || !prev_td; int pos_check = (is_null) ? 1 : (next_td != td && pos >= __c11_atomic_load(&next_td->wait_mb_pos, 5) ) || (prev_td != td && pos >= __c11_atomic_load(& prev_td->wait_mb_pos, 5)); __c11_atomic_store(&td-> thread_mb_pos, pos, 5); if (sliced_threading && pos_check ) { strict_pthread_mutex_lock(&td->lock); strict_pthread_cond_broadcast (&td->cond); strict_pthread_mutex_unlock(&td->lock ); } } while (0); | |||
| 2568 | } | |||
| 2569 | } | |||
| 2570 | ||||
| 2571 | static av_always_inline__attribute__((always_inline)) inline | |||
| 2572 | int vp78_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr, | |||
| 2573 | int threadnr, int is_vp7) | |||
| 2574 | { | |||
| 2575 | const VP8Context *s = avctx->priv_data; | |||
| 2576 | VP8ThreadData *td = &s->thread_data[jobnr]; | |||
| 2577 | VP8ThreadData *next_td = NULL((void*)0), *prev_td = NULL((void*)0); | |||
| 2578 | VP8Frame *curframe = s->curframe; | |||
| 2579 | int mb_y, num_jobs = s->num_jobs; | |||
| 2580 | int ret; | |||
| 2581 | ||||
| 2582 | td->thread_nr = threadnr; | |||
| 2583 | td->mv_bounds.mv_min.y = -MARGIN(16 << 2) - 64 * threadnr; | |||
| 2584 | td->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN(16 << 2) - 64 * threadnr; | |||
| 2585 | for (mb_y = jobnr; mb_y < s->mb_height; mb_y += num_jobs) { | |||
| 2586 | atomic_store(&td->thread_mb_pos, mb_y << 16)__c11_atomic_store(&td->thread_mb_pos, mb_y << 16 , 5); | |||
| 2587 | ret = s->decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr); | |||
| 2588 | if (ret < 0) { | |||
| 2589 | update_pos(td, s->mb_height, INT_MAX & 0xFFFF)do { int pos = (s->mb_height << 16) | (2147483647 & 0xFFFF & 0xFFFF); int sliced_threading = (avctx->active_thread_type == 2) && (num_jobs > 1); int is_null = !next_td || !prev_td; int pos_check = (is_null) ? 1 : (next_td != td && pos >= __c11_atomic_load(&next_td->wait_mb_pos, 5) ) || (prev_td != td && pos >= __c11_atomic_load(& prev_td->wait_mb_pos, 5)); __c11_atomic_store(&td-> thread_mb_pos, pos, 5); if (sliced_threading && pos_check ) { strict_pthread_mutex_lock(&td->lock); strict_pthread_cond_broadcast (&td->cond); strict_pthread_mutex_unlock(&td->lock ); } } while (0); | |||
| 2590 | return ret; | |||
| 2591 | } | |||
| 2592 | if (s->deblock_filter) | |||
| 2593 | s->filter_mb_row(avctx, tdata, jobnr, threadnr); | |||
| 2594 | update_pos(td, mb_y, INT_MAX & 0xFFFF)do { int pos = (mb_y << 16) | (2147483647 & 0xFFFF & 0xFFFF); int sliced_threading = (avctx->active_thread_type == 2) && (num_jobs > 1); int is_null = !next_td || !prev_td; int pos_check = (is_null) ? 1 : (next_td != td && pos >= __c11_atomic_load(&next_td->wait_mb_pos, 5) ) || (prev_td != td && pos >= __c11_atomic_load(& prev_td->wait_mb_pos, 5)); __c11_atomic_store(&td-> thread_mb_pos, pos, 5); if (sliced_threading && pos_check ) { strict_pthread_mutex_lock(&td->lock); strict_pthread_cond_broadcast (&td->cond); strict_pthread_mutex_unlock(&td->lock ); } } while (0); | |||
| 2595 | ||||
| 2596 | td->mv_bounds.mv_min.y -= 64 * num_jobs; | |||
| 2597 | td->mv_bounds.mv_max.y -= 64 * num_jobs; | |||
| 2598 | ||||
| 2599 | if (avctx->active_thread_type == FF_THREAD_FRAME1) | |||
| 2600 | ff_progress_frame_report(&curframe->tf, mb_y); | |||
| 2601 | } | |||
| 2602 | ||||
| 2603 | return 0; | |||
| 2604 | } | |||
| 2605 | ||||
| 2606 | static int vp7_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, | |||
| 2607 | int jobnr, int threadnr) | |||
| 2608 | { | |||
| 2609 | return vp78_decode_mb_row_sliced(avctx, tdata, jobnr, threadnr, IS_VP71); | |||
| 2610 | } | |||
| 2611 | ||||
| 2612 | static int vp8_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, | |||
| 2613 | int jobnr, int threadnr) | |||
| 2614 | { | |||
| 2615 | return vp78_decode_mb_row_sliced(avctx, tdata, jobnr, threadnr, IS_VP80); | |||
| 2616 | } | |||
| 2617 | ||||
| 2618 | static av_always_inline__attribute__((always_inline)) inline | |||
| 2619 | int vp78_decode_frame(AVCodecContext *avctx, AVFrame *rframe, int *got_frame, | |||
| 2620 | const AVPacket *avpkt, int is_vp7) | |||
| 2621 | { | |||
| 2622 | VP8Context *s = avctx->priv_data; | |||
| 2623 | int ret, i, referenced, num_jobs; | |||
| 2624 | enum AVDiscard skip_thresh; | |||
| 2625 | VP8Frame *av_uninit(curframe)curframe=curframe, *prev_frame; | |||
| 2626 | ||||
| 2627 | if (is_vp7) | |||
| 2628 | ret = vp7_decode_frame_header(s, avpkt->data, avpkt->size); | |||
| 2629 | else | |||
| 2630 | ret = vp8_decode_frame_header(s, avpkt->data, avpkt->size); | |||
| 2631 | ||||
| 2632 | if (ret < 0) | |||
| 2633 | goto err; | |||
| 2634 | ||||
| 2635 | if (!is_vp7 && s->actually_webp) { | |||
| 2636 | // VP8 in WebP is supposed to be intra-only. Enforce this here | |||
| 2637 | // to ensure that output is reproducible with frame-threading. | |||
| 2638 | if (!s->keyframe) | |||
| 2639 | return AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 2640 | // avctx->pix_fmt already set in caller. | |||
| 2641 | } else if (!is_vp7 && s->pix_fmt == AV_PIX_FMT_NONE) { | |||
| 2642 | s->pix_fmt = get_pixel_format(s); | |||
| 2643 | if (s->pix_fmt < 0) { | |||
| 2644 | ret = AVERROR(EINVAL)(-(22)); | |||
| 2645 | goto err; | |||
| 2646 | } | |||
| 2647 | avctx->pix_fmt = s->pix_fmt; | |||
| 2648 | } | |||
| 2649 | ||||
| 2650 | prev_frame = s->framep[VP8_FRAME_CURRENT]; | |||
| 2651 | ||||
| 2652 | referenced = s->update_last || s->update_golden == VP8_FRAME_CURRENT || | |||
| 2653 | s->update_altref == VP8_FRAME_CURRENT; | |||
| 2654 | ||||
| 2655 | skip_thresh = !referenced ? AVDISCARD_NONREF | |||
| 2656 | : !s->keyframe ? AVDISCARD_NONKEY | |||
| 2657 | : AVDISCARD_ALL; | |||
| 2658 | ||||
| 2659 | if (avctx->skip_frame >= skip_thresh) { | |||
| 2660 | s->invisible = 1; | |||
| 2661 | memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4); | |||
| 2662 | goto skip_decode; | |||
| 2663 | } | |||
| 2664 | s->deblock_filter = s->filter.level && avctx->skip_loop_filter < skip_thresh; | |||
| 2665 | ||||
| 2666 | // release no longer referenced frames | |||
| 2667 | for (i = 0; i < 5; i++) | |||
| 2668 | if (s->frames[i].tf.f && | |||
| 2669 | &s->frames[i] != prev_frame && | |||
| 2670 | &s->frames[i] != s->framep[VP8_FRAME_PREVIOUS] && | |||
| 2671 | &s->frames[i] != s->framep[VP8_FRAME_GOLDEN] && | |||
| 2672 | &s->frames[i] != s->framep[VP8_FRAME_ALTREF]) | |||
| 2673 | vp8_release_frame(&s->frames[i]); | |||
| 2674 | ||||
| 2675 | if (!s->colorspace) | |||
| 2676 | avctx->colorspace = AVCOL_SPC_BT470BG; | |||
| 2677 | if (s->fullrange) | |||
| 2678 | avctx->color_range = AVCOL_RANGE_JPEG; | |||
| 2679 | else | |||
| 2680 | avctx->color_range = AVCOL_RANGE_MPEG; | |||
| 2681 | ||||
| 2682 | /* Given that arithmetic probabilities are updated every frame, it's quite | |||
| 2683 | * likely that the values we have on a random interframe are complete | |||
| 2684 | * junk if we didn't start decode on a keyframe. So just don't display | |||
| 2685 | * anything rather than junk. */ | |||
| 2686 | if (!s->keyframe && (!s->framep[VP8_FRAME_PREVIOUS] || | |||
| 2687 | !s->framep[VP8_FRAME_GOLDEN] || | |||
| 2688 | !s->framep[VP8_FRAME_ALTREF])) { | |||
| 2689 | av_log(avctx, AV_LOG_WARNING24, | |||
| 2690 | "Discarding interframe without a prior keyframe!\n"); | |||
| 2691 | ret = AVERROR_INVALIDDATA(-(int)(('I') | (('N') << 8) | (('D') << 16) | (( unsigned)('A') << 24))); | |||
| 2692 | goto err; | |||
| 2693 | } | |||
| 2694 | ||||
| 2695 | curframe = vp8_find_free_buffer(s); | |||
| 2696 | if ((ret = vp8_alloc_frame(s, curframe, referenced)) < 0) | |||
| 2697 | goto err; | |||
| 2698 | s->framep[VP8_FRAME_CURRENT] = curframe; | |||
| 2699 | if (s->keyframe) | |||
| 2700 | curframe->tf.f->flags |= AV_FRAME_FLAG_KEY(1 << 1); | |||
| 2701 | else | |||
| 2702 | curframe->tf.f->flags &= ~AV_FRAME_FLAG_KEY(1 << 1); | |||
| 2703 | curframe->tf.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I | |||
| 2704 | : AV_PICTURE_TYPE_P; | |||
| 2705 | ||||
| 2706 | // check if golden and altref are swapped | |||
| 2707 | if (s->update_altref != VP8_FRAME_NONE) | |||
| 2708 | s->next_framep[VP8_FRAME_ALTREF] = s->framep[s->update_altref]; | |||
| 2709 | else | |||
| 2710 | s->next_framep[VP8_FRAME_ALTREF] = s->framep[VP8_FRAME_ALTREF]; | |||
| 2711 | ||||
| 2712 | if (s->update_golden != VP8_FRAME_NONE) | |||
| 2713 | s->next_framep[VP8_FRAME_GOLDEN] = s->framep[s->update_golden]; | |||
| 2714 | else | |||
| 2715 | s->next_framep[VP8_FRAME_GOLDEN] = s->framep[VP8_FRAME_GOLDEN]; | |||
| 2716 | ||||
| 2717 | if (s->update_last) | |||
| 2718 | s->next_framep[VP8_FRAME_PREVIOUS] = curframe; | |||
| 2719 | else | |||
| 2720 | s->next_framep[VP8_FRAME_PREVIOUS] = s->framep[VP8_FRAME_PREVIOUS]; | |||
| 2721 | ||||
| 2722 | s->next_framep[VP8_FRAME_CURRENT] = curframe; | |||
| 2723 | ||||
| 2724 | if (!is_vp7 && !s->actually_webp) | |||
| 2725 | ff_thread_finish_setup(avctx); | |||
| 2726 | ||||
| 2727 | if (!is_vp7 && avctx->hwaccel) { | |||
| 2728 | const FFHWAccel *hwaccel = ffhwaccel(avctx->hwaccel); | |||
| 2729 | ret = hwaccel->start_frame(avctx, avpkt->buf, avpkt->data, avpkt->size); | |||
| 2730 | if (ret < 0) | |||
| 2731 | goto err; | |||
| 2732 | ||||
| 2733 | ret = hwaccel->decode_slice(avctx, avpkt->data, avpkt->size); | |||
| 2734 | if (ret < 0) | |||
| 2735 | goto err; | |||
| 2736 | ||||
| 2737 | ret = hwaccel->end_frame(avctx); | |||
| 2738 | if (ret < 0) | |||
| 2739 | goto err; | |||
| 2740 | ||||
| 2741 | } else { | |||
| 2742 | s->linesize = curframe->tf.f->linesize[0]; | |||
| 2743 | s->uvlinesize = curframe->tf.f->linesize[1]; | |||
| 2744 | ||||
| 2745 | memset(s->top_nnz, 0, s->mb_width * sizeof(*s->top_nnz)); | |||
| 2746 | /* Zero macroblock structures for top/top-left prediction | |||
| 2747 | * from outside the frame. */ | |||
| 2748 | if (!s->mb_layout) | |||
| 2749 | memset(s->macroblocks + s->mb_height * 2 - 1, 0, | |||
| 2750 | (s->mb_width + 1) * sizeof(*s->macroblocks)); | |||
| 2751 | if (!s->mb_layout && s->keyframe) | |||
| 2752 | memset(s->intra4x4_pred_mode_top, DC_PRED2, s->mb_width * 4); | |||
| 2753 | ||||
| 2754 | memset(s->ref_count, 0, sizeof(s->ref_count)); | |||
| 2755 | ||||
| 2756 | if (s->mb_layout == 1) { | |||
| 2757 | // Make sure the previous frame has read its segmentation map, | |||
| 2758 | // if we reuse the same map. | |||
| 2759 | if (prev_frame && s->segmentation.enabled && | |||
| 2760 | !s->segmentation.update_map) | |||
| 2761 | ff_progress_frame_await(&prev_frame->tf, 1); | |||
| 2762 | if (is_vp7) | |||
| 2763 | ret = vp7_decode_mv_mb_modes(avctx, curframe, prev_frame); | |||
| 2764 | else | |||
| 2765 | ret = vp8_decode_mv_mb_modes(avctx, curframe, prev_frame); | |||
| 2766 | if (ret < 0) | |||
| 2767 | goto err; | |||
| 2768 | } | |||
| 2769 | ||||
| 2770 | if (avctx->active_thread_type == FF_THREAD_FRAME1) | |||
| 2771 | num_jobs = 1; | |||
| 2772 | else | |||
| 2773 | num_jobs = FFMIN(s->num_coeff_partitions, avctx->thread_count)((s->num_coeff_partitions) > (avctx->thread_count) ? (avctx->thread_count) : (s->num_coeff_partitions)); | |||
| 2774 | s->num_jobs = num_jobs; | |||
| 2775 | s->curframe = curframe; | |||
| 2776 | s->prev_frame = prev_frame; | |||
| 2777 | s->mv_bounds.mv_min.y = -MARGIN(16 << 2); | |||
| 2778 | s->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN(16 << 2); | |||
| 2779 | for (i = 0; i < MAX_THREADS8; i++) { | |||
| 2780 | VP8ThreadData *td = &s->thread_data[i]; | |||
| 2781 | atomic_init__c11_atomic_init(&td->thread_mb_pos, 0); | |||
| 2782 | atomic_init__c11_atomic_init(&td->wait_mb_pos, INT_MAX2147483647); | |||
| 2783 | } | |||
| 2784 | if (is_vp7) | |||
| 2785 | avctx->execute2(avctx, vp7_decode_mb_row_sliced, s->thread_data, NULL((void*)0), | |||
| 2786 | num_jobs); | |||
| 2787 | else | |||
| 2788 | avctx->execute2(avctx, vp8_decode_mb_row_sliced, s->thread_data, NULL((void*)0), | |||
| 2789 | num_jobs); | |||
| 2790 | } | |||
| 2791 | ||||
| 2792 | ff_progress_frame_report(&curframe->tf, INT_MAX2147483647); | |||
| 2793 | memcpy(&s->framep[0], &s->next_framep[0], sizeof(s->framep[0]) * 4); | |||
| 2794 | ||||
| 2795 | skip_decode: | |||
| 2796 | // if future frames don't use the updated probabilities, | |||
| 2797 | // reset them to the values we saved | |||
| 2798 | if (!s->update_probabilities) | |||
| 2799 | s->prob[0] = s->prob[1]; | |||
| 2800 | ||||
| 2801 | if (!s->invisible) { | |||
| 2802 | if ((ret = av_frame_ref(rframe, curframe->tf.f)) < 0) | |||
| 2803 | return ret; | |||
| 2804 | *got_frame = 1; | |||
| 2805 | } | |||
| 2806 | ||||
| 2807 | return avpkt->size; | |||
| 2808 | err: | |||
| 2809 | memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4); | |||
| 2810 | return ret; | |||
| 2811 | } | |||
| 2812 | ||||
| 2813 | av_cold__attribute__((cold)) int ff_vp8_decode_free(AVCodecContext *avctx) | |||
| 2814 | { | |||
| 2815 | vp8_decode_flush_impl(avctx, 1); | |||
| 2816 | ||||
| 2817 | return 0; | |||
| 2818 | } | |||
| 2819 | ||||
| 2820 | static av_cold__attribute__((cold)) void vp78_decode_init(AVCodecContext *avctx) | |||
| 2821 | { | |||
| 2822 | VP8Context *s = avctx->priv_data; | |||
| 2823 | ||||
| 2824 | s->avctx = avctx; | |||
| 2825 | s->pix_fmt = AV_PIX_FMT_NONE; | |||
| 2826 | avctx->pix_fmt = AV_PIX_FMT_YUV420P; | |||
| 2827 | ||||
| 2828 | ff_videodsp_init(&s->vdsp, 8); | |||
| 2829 | ||||
| 2830 | ff_vp78dsp_init(&s->vp8dsp); | |||
| 2831 | ||||
| 2832 | /* does not change for VP8 */ | |||
| 2833 | memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan)); | |||
| 2834 | } | |||
| 2835 | ||||
| 2836 | #if CONFIG_VP8_DECODER1 | |||
| 2837 | static int vp8_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata, | |||
| 2838 | int jobnr, int threadnr) | |||
| 2839 | { | |||
| 2840 | return decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr, 0); | |||
| 2841 | } | |||
| 2842 | ||||
| 2843 | static void vp8_filter_mb_row(AVCodecContext *avctx, void *tdata, | |||
| 2844 | int jobnr, int threadnr) | |||
| 2845 | { | |||
| 2846 | filter_mb_row(avctx, tdata, jobnr, threadnr, 0); | |||
| 2847 | } | |||
| 2848 | ||||
| 2849 | int ff_vp8_decode_frame(AVCodecContext *avctx, AVFrame *frame, | |||
| 2850 | int *got_frame, AVPacket *avpkt) | |||
| 2851 | { | |||
| 2852 | return vp78_decode_frame(avctx, frame, got_frame, avpkt, IS_VP80); | |||
| 2853 | } | |||
| 2854 | ||||
| 2855 | av_cold__attribute__((cold)) int ff_vp8_decode_init(AVCodecContext *avctx) | |||
| 2856 | { | |||
| 2857 | VP8Context *s = avctx->priv_data; | |||
| 2858 | ||||
| 2859 | vp78_decode_init(avctx); | |||
| 2860 | ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP8, 8, 1); | |||
| 2861 | ff_vp8dsp_init(&s->vp8dsp); | |||
| 2862 | s->decode_mb_row_no_filter = vp8_decode_mb_row_no_filter; | |||
| 2863 | s->filter_mb_row = vp8_filter_mb_row; | |||
| 2864 | ||||
| 2865 | return 0; | |||
| 2866 | } | |||
| 2867 | ||||
| 2868 | #if HAVE_THREADS1 | |||
| 2869 | static void vp8_replace_frame(VP8Frame *dst, const VP8Frame *src) | |||
| 2870 | { | |||
| 2871 | ff_progress_frame_replace(&dst->tf, &src->tf); | |||
| 2872 | av_refstruct_replace(&dst->seg_map, src->seg_map); | |||
| 2873 | av_refstruct_replace(&dst->hwaccel_picture_private, | |||
| 2874 | src->hwaccel_picture_private); | |||
| 2875 | } | |||
| 2876 | ||||
| 2877 | #define REBASE(pic)((pic) ? (pic) - &s_src->frames[0] + &s->frames [0] : ((void*)0)) ((pic) ? (pic) - &s_src->frames[0] + &s->frames[0] : NULL((void*)0)) | |||
| 2878 | ||||
| 2879 | static int vp8_decode_update_thread_context(AVCodecContext *dst, | |||
| 2880 | const AVCodecContext *src) | |||
| 2881 | { | |||
| 2882 | VP8Context *s = dst->priv_data, *s_src = src->priv_data; | |||
| 2883 | ||||
| 2884 | if (s->macroblocks_base && | |||
| 2885 | (s_src->mb_width != s->mb_width || s_src->mb_height != s->mb_height)) { | |||
| 2886 | free_buffers(s); | |||
| 2887 | s->mb_width = s_src->mb_width; | |||
| 2888 | s->mb_height = s_src->mb_height; | |||
| 2889 | } | |||
| 2890 | ||||
| 2891 | s->pix_fmt = s_src->pix_fmt; | |||
| 2892 | s->prob[0] = s_src->prob[!s_src->update_probabilities]; | |||
| 2893 | s->segmentation = s_src->segmentation; | |||
| 2894 | s->lf_delta = s_src->lf_delta; | |||
| 2895 | memcpy(s->sign_bias, s_src->sign_bias, sizeof(s->sign_bias)); | |||
| 2896 | ||||
| 2897 | for (int i = 0; i < FF_ARRAY_ELEMS(s_src->frames)(sizeof(s_src->frames) / sizeof((s_src->frames)[0])); i++) | |||
| 2898 | vp8_replace_frame(&s->frames[i], &s_src->frames[i]); | |||
| 2899 | ||||
| 2900 | s->framep[0] = REBASE(s_src->next_framep[0])((s_src->next_framep[0]) ? (s_src->next_framep[0]) - & s_src->frames[0] + &s->frames[0] : ((void*)0)); | |||
| 2901 | s->framep[1] = REBASE(s_src->next_framep[1])((s_src->next_framep[1]) ? (s_src->next_framep[1]) - & s_src->frames[0] + &s->frames[0] : ((void*)0)); | |||
| 2902 | s->framep[2] = REBASE(s_src->next_framep[2])((s_src->next_framep[2]) ? (s_src->next_framep[2]) - & s_src->frames[0] + &s->frames[0] : ((void*)0)); | |||
| 2903 | s->framep[3] = REBASE(s_src->next_framep[3])((s_src->next_framep[3]) ? (s_src->next_framep[3]) - & s_src->frames[0] + &s->frames[0] : ((void*)0)); | |||
| 2904 | ||||
| 2905 | return 0; | |||
| 2906 | } | |||
| 2907 | #endif /* HAVE_THREADS */ | |||
| 2908 | #endif /* CONFIG_VP8_DECODER */ | |||
| 2909 | ||||
| 2910 | #if CONFIG_VP7_DECODER0 | |||
| 2911 | static int vp7_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata, | |||
| 2912 | int jobnr, int threadnr) | |||
| 2913 | { | |||
| 2914 | return decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr, 1); | |||
| 2915 | } | |||
| 2916 | ||||
| 2917 | static void vp7_filter_mb_row(AVCodecContext *avctx, void *tdata, | |||
| 2918 | int jobnr, int threadnr) | |||
| 2919 | { | |||
| 2920 | filter_mb_row(avctx, tdata, jobnr, threadnr, 1); | |||
| 2921 | } | |||
| 2922 | ||||
| 2923 | static int vp7_decode_frame(AVCodecContext *avctx, AVFrame *frame, | |||
| 2924 | int *got_frame, AVPacket *avpkt) | |||
| 2925 | { | |||
| 2926 | return vp78_decode_frame(avctx, frame, got_frame, avpkt, IS_VP71); | |||
| 2927 | } | |||
| 2928 | ||||
| 2929 | av_cold__attribute__((cold)) static int vp7_decode_init(AVCodecContext *avctx) | |||
| 2930 | { | |||
| 2931 | VP8Context *s = avctx->priv_data; | |||
| 2932 | ||||
| 2933 | vp78_decode_init(avctx); | |||
| 2934 | ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP7, 8, 1); | |||
| 2935 | ff_vp7dsp_init(&s->vp8dsp); | |||
| 2936 | s->decode_mb_row_no_filter = vp7_decode_mb_row_no_filter; | |||
| 2937 | s->filter_mb_row = vp7_filter_mb_row; | |||
| 2938 | ||||
| 2939 | return 0; | |||
| 2940 | } | |||
| 2941 | ||||
| 2942 | const FFCodec ff_vp7_decoder = { | |||
| 2943 | .p.name = "vp7", | |||
| 2944 | CODEC_LONG_NAME("On2 VP7").p.long_name = "On2 VP7", | |||
| 2945 | .p.type = AVMEDIA_TYPE_VIDEO, | |||
| 2946 | .p.id = AV_CODEC_ID_VP7, | |||
| 2947 | .priv_data_size = sizeof(VP8Context), | |||
| 2948 | .init = vp7_decode_init, | |||
| 2949 | .close = ff_vp8_decode_free, | |||
| 2950 | FF_CODEC_DECODE_CB(vp7_decode_frame).is_decoder = 1, .cb_type = FF_CODEC_CB_TYPE_DECODE, .cb.decode = (vp7_decode_frame), | |||
| 2951 | .p.capabilities = AV_CODEC_CAP_DR1(1 << 1), | |||
| 2952 | .flush = vp8_decode_flush, | |||
| 2953 | .caps_internal = FF_CODEC_CAP_USES_PROGRESSFRAMES(1 << 6), | |||
| 2954 | }; | |||
| 2955 | #endif /* CONFIG_VP7_DECODER */ | |||
| 2956 | ||||
| 2957 | #if CONFIG_VP8_DECODER1 | |||
| 2958 | const FFCodec ff_vp8_decoder = { | |||
| 2959 | .p.name = "vp8", | |||
| 2960 | CODEC_LONG_NAME("On2 VP8").p.long_name = "On2 VP8", | |||
| 2961 | .p.type = AVMEDIA_TYPE_VIDEO, | |||
| 2962 | .p.id = AV_CODEC_ID_VP8, | |||
| 2963 | .priv_data_size = sizeof(VP8Context), | |||
| 2964 | .init = ff_vp8_decode_init, | |||
| 2965 | .close = ff_vp8_decode_free, | |||
| 2966 | FF_CODEC_DECODE_CB(ff_vp8_decode_frame).is_decoder = 1, .cb_type = FF_CODEC_CB_TYPE_DECODE, .cb.decode = (ff_vp8_decode_frame), | |||
| 2967 | .p.capabilities = AV_CODEC_CAP_DR1(1 << 1) | AV_CODEC_CAP_FRAME_THREADS(1 << 12) | | |||
| 2968 | AV_CODEC_CAP_SLICE_THREADS(1 << 13), | |||
| 2969 | .caps_internal = FF_CODEC_CAP_USES_PROGRESSFRAMES(1 << 6), | |||
| 2970 | .flush = vp8_decode_flush, | |||
| 2971 | UPDATE_THREAD_CONTEXT(vp8_decode_update_thread_context).update_thread_context = (vp8_decode_update_thread_context), | |||
| 2972 | .hw_configs = (const AVCodecHWConfigInternal *const []) { | |||
| 2973 | #if CONFIG_VP8_VAAPI_HWACCEL1 | |||
| 2974 | HWACCEL_VAAPI(vp8)&(const AVCodecHWConfigInternal) { .public = { .pix_fmt = AV_PIX_FMT_VAAPI, .methods = (1 ? AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX : 0) | (1 ? AV_CODEC_HW_CONFIG_METHOD_HW_FRAMES_CTX : 0) | ( 1 ? AV_CODEC_HW_CONFIG_METHOD_AD_HOC : 0), .device_type = AV_HWDEVICE_TYPE_VAAPI , }, .hwaccel = &ff_vp8_vaapi_hwaccel, }, | |||
| 2975 | #endif | |||
| 2976 | #if CONFIG_VP8_NVDEC_HWACCEL0 | |||
| 2977 | HWACCEL_NVDEC(vp8)&(const AVCodecHWConfigInternal) { .public = { .pix_fmt = AV_PIX_FMT_CUDA, .methods = (1 ? AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX : 0) | (1 ? AV_CODEC_HW_CONFIG_METHOD_HW_FRAMES_CTX : 0) | ( 0 ? AV_CODEC_HW_CONFIG_METHOD_AD_HOC : 0), .device_type = AV_HWDEVICE_TYPE_CUDA , }, .hwaccel = &ff_vp8_nvdec_hwaccel, }, | |||
| 2978 | #endif | |||
| 2979 | NULL((void*)0) | |||
| 2980 | }, | |||
| 2981 | }; | |||
| 2982 | #endif /* CONFIG_VP8_DECODER */ |