3 * Copyright (c) 2002 Fabrice Bellard
4 * Copyright (c) 2004 Roman Shaposhnik
6 * 50 Mbps (DVCPRO50) support
7 * Copyright (c) 2006 Daniel Maas <dmaas@maasdigital.com>
9 * 100 Mbps (DVCPRO HD) support
10 * Initial code by Daniel Maas <dmaas@maasdigital.com> (funded by BBC R&D)
11 * Final code by Roman Shaposhnik
13 * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth
14 * of DV technical info.
16 * This file is part of Libav.
18 * Libav is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU Lesser General Public
20 * License as published by the Free Software Foundation; either
21 * version 2.1 of the License, or (at your option) any later version.
23 * Libav is distributed in the hope that it will be useful,
24 * but WITHOUT ANY WARRANTY; without even the implied warranty of
25 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
26 * Lesser General Public License for more details.
28 * You should have received a copy of the GNU Lesser General Public
29 * License along with Libav; if not, write to the Free Software
30 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
38 #include "libavutil/pixdesc.h"
43 #include "simple_idct.h"
47 typedef struct BlockInfo {
48 const uint32_t *factor_table;
49 const uint8_t *scan_table;
50 uint8_t pos; /* position in block */
51 void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block);
52 uint8_t partial_bit_count;
53 uint32_t partial_bit_buffer;
57 /* decode AC coefficients */
58 static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block)
60 int last_index = gb->size_in_bits;
61 const uint8_t *scan_table = mb->scan_table;
62 const uint32_t *factor_table = mb->factor_table;
64 int partial_bit_count = mb->partial_bit_count;
65 int level, run, vlc_len, index;
70 /* if we must parse a partial VLC, we do it here */
71 if (partial_bit_count > 0) {
72 re_cache = re_cache >> partial_bit_count | mb->partial_bit_buffer;
73 re_index -= partial_bit_count;
74 mb->partial_bit_count = 0;
77 /* get the AC coefficients until last_index is reached */
79 av_dlog(NULL, "%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16),
81 /* our own optimized GET_RL_VLC */
82 index = NEG_USR32(re_cache, TEX_VLC_BITS);
83 vlc_len = ff_dv_rl_vlc[index].len;
85 index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) +
86 ff_dv_rl_vlc[index].level;
87 vlc_len = TEX_VLC_BITS - vlc_len;
89 level = ff_dv_rl_vlc[index].level;
90 run = ff_dv_rl_vlc[index].run;
92 /* gotta check if we're still within gb boundaries */
93 if (re_index + vlc_len > last_index) {
94 /* should be < 16 bits otherwise a codeword could have been parsed */
95 mb->partial_bit_count = last_index - re_index;
96 mb->partial_bit_buffer = re_cache & ~(-1u >> mb->partial_bit_count);
97 re_index = last_index;
102 av_dlog(NULL, "run=%d level=%d\n", run, level);
107 level = (level * factor_table[pos] + (1 << (dv_iweight_bits - 1))) >> dv_iweight_bits;
108 block[scan_table[pos]] = level;
110 UPDATE_CACHE(re, gb);
112 CLOSE_READER(re, gb);
116 static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
118 int bits_left = get_bits_left(gb);
119 while (bits_left >= MIN_CACHE_BITS) {
120 put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
121 bits_left -= MIN_CACHE_BITS;
124 put_bits(pb, bits_left, get_bits(gb, bits_left));
128 /* mb_x and mb_y are in units of 8 pixels */
129 static int dv_decode_video_segment(AVCodecContext *avctx, void *arg)
131 DVVideoContext *s = avctx->priv_data;
132 DVwork_chunk *work_chunk = arg;
133 int quant, dc, dct_mode, class1, j;
134 int mb_index, mb_x, mb_y, last_index;
135 int y_stride, linesize;
136 DCTELEM *block, *block1;
139 const uint8_t *buf_ptr;
140 PutBitContext pb, vs_pb;
142 BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1;
143 LOCAL_ALIGNED_16(DCTELEM, sblock, [5*DV_MAX_BPM], [64]);
144 LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [ 80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
145 LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [5*80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
146 const int log2_blocksize = 3-s->avctx->lowres;
147 int is_field_mode[5];
149 assert((((int)mb_bit_buffer) & 7) == 0);
150 assert((((int)vs_bit_buffer) & 7) == 0);
152 memset(sblock, 0, 5*DV_MAX_BPM*sizeof(*sblock));
154 /* pass 1: read DC and AC coefficients in blocks */
155 buf_ptr = &s->buf[work_chunk->buf_offset*80];
156 block1 = &sblock[0][0];
158 init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
159 for (mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) {
161 quant = buf_ptr[3] & 0x0f;
163 init_put_bits(&pb, mb_bit_buffer, 80);
166 is_field_mode[mb_index] = 0;
167 for (j = 0; j < s->sys->bpm; j++) {
168 last_index = s->sys->block_sizes[j];
169 init_get_bits(&gb, buf_ptr, last_index);
172 dc = get_sbits(&gb, 9);
173 dct_mode = get_bits1(&gb);
174 class1 = get_bits(&gb, 2);
175 if (DV_PROFILE_IS_HD(s->sys)) {
176 mb->idct_put = s->idct_put[0];
177 mb->scan_table = s->dv_zigzag[0];
178 mb->factor_table = &s->sys->idct_factor[(j >= 4)*4*16*64 + class1*16*64 + quant*64];
179 is_field_mode[mb_index] |= !j && dct_mode;
181 mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3];
182 mb->scan_table = s->dv_zigzag[dct_mode];
183 mb->factor_table = &s->sys->idct_factor[(class1 == 3)*2*22*64 + dct_mode*22*64 +
184 (quant + dv_quant_offset[class1])*64];
187 /* convert to unsigned because 128 is not added in the
191 buf_ptr += last_index >> 3;
193 mb->partial_bit_count = 0;
195 av_dlog(avctx, "MB block: %d, %d ", mb_index, j);
196 dv_decode_ac(&gb, mb, block);
198 /* write the remaining bits in a new buffer only if the
207 /* pass 2: we can do it just after */
208 av_dlog(avctx, "***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
211 init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
212 put_bits32(&pb, 0); // padding must be zeroed
214 for (j = 0; j < s->sys->bpm; j++, block += 64, mb++) {
215 if (mb->pos < 64 && get_bits_left(&gb) > 0) {
216 dv_decode_ac(&gb, mb, block);
217 /* if still not finished, no need to parse other blocks */
222 /* all blocks are finished, so the extra bytes can be used at
223 the video segment level */
224 if (j >= s->sys->bpm)
225 bit_copy(&vs_pb, &gb);
228 /* we need a pass over the whole video segment */
229 av_dlog(avctx, "***pass 3 size=%d\n", put_bits_count(&vs_pb));
230 block = &sblock[0][0];
232 init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
233 put_bits32(&vs_pb, 0); // padding must be zeroed
234 flush_put_bits(&vs_pb);
235 for (mb_index = 0; mb_index < 5; mb_index++) {
236 for (j = 0; j < s->sys->bpm; j++) {
238 av_dlog(avctx, "start %d:%d\n", mb_index, j);
239 dv_decode_ac(&gb, mb, block);
241 if (mb->pos >= 64 && mb->pos < 127)
242 av_log(avctx, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);
248 /* compute idct and place blocks */
249 block = &sblock[0][0];
251 for (mb_index = 0; mb_index < 5; mb_index++) {
252 dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y);
254 /* idct_put'ting luminance */
255 if ((s->sys->pix_fmt == PIX_FMT_YUV420P) ||
256 (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) ||
257 (s->sys->height >= 720 && mb_y != 134)) {
258 y_stride = (s->picture.linesize[0] << ((!is_field_mode[mb_index]) * log2_blocksize));
260 y_stride = (2 << log2_blocksize);
262 y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x) << log2_blocksize);
263 linesize = s->picture.linesize[0] << is_field_mode[mb_index];
264 mb[0] .idct_put(y_ptr , linesize, block + 0*64);
265 if (s->sys->video_stype == 4) { /* SD 422 */
266 mb[2].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 2*64);
268 mb[1].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 1*64);
269 mb[2].idct_put(y_ptr + y_stride, linesize, block + 2*64);
270 mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3*64);
275 /* idct_put'ting chrominance */
276 c_offset = (((mb_y >> (s->sys->pix_fmt == PIX_FMT_YUV420P)) * s->picture.linesize[1] +
277 (mb_x >> ((s->sys->pix_fmt == PIX_FMT_YUV411P) ? 2 : 1))) << log2_blocksize);
278 for (j = 2; j; j--) {
279 uint8_t *c_ptr = s->picture.data[j] + c_offset;
280 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
281 uint64_t aligned_pixels[64/8];
282 uint8_t *pixels = (uint8_t*)aligned_pixels;
283 uint8_t *c_ptr1, *ptr1;
285 mb->idct_put(pixels, 8, block);
286 for (y = 0; y < (1 << log2_blocksize); y++, c_ptr += s->picture.linesize[j], pixels += 8) {
287 ptr1 = pixels + (1 << (log2_blocksize - 1));
288 c_ptr1 = c_ptr + (s->picture.linesize[j] << log2_blocksize);
289 for (x = 0; x < (1 << (log2_blocksize - 1)); x++) {
290 c_ptr[x] = pixels[x];
296 y_stride = (mb_y == 134) ? (1 << log2_blocksize) :
297 s->picture.linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize);
298 linesize = s->picture.linesize[j] << is_field_mode[mb_index];
299 (mb++)-> idct_put(c_ptr , linesize, block); block += 64;
300 if (s->sys->bpm == 8) {
301 (mb++)->idct_put(c_ptr + y_stride, linesize, block); block += 64;
309 /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
310 144000 bytes for PAL - or twice those for 50Mbps) */
311 static int dvvideo_decode_frame(AVCodecContext *avctx,
312 void *data, int *data_size,
315 const uint8_t *buf = avpkt->data;
316 int buf_size = avpkt->size;
317 DVVideoContext *s = avctx->priv_data;
318 const uint8_t* vsc_pack;
321 s->sys = avpriv_dv_frame_profile(s->sys, buf, buf_size);
322 if (!s->sys || buf_size < s->sys->frame_size || ff_dv_init_dynamic_tables(s->sys)) {
323 av_log(avctx, AV_LOG_ERROR, "could not find dv frame profile\n");
324 return -1; /* NOTE: we only accept several full frames */
327 if (s->picture.data[0])
328 avctx->release_buffer(avctx, &s->picture);
330 s->picture.reference = 0;
331 s->picture.key_frame = 1;
332 s->picture.pict_type = AV_PICTURE_TYPE_I;
333 avctx->pix_fmt = s->sys->pix_fmt;
334 avctx->time_base = s->sys->time_base;
335 avcodec_set_dimensions(avctx, s->sys->width, s->sys->height);
336 if (avctx->get_buffer(avctx, &s->picture) < 0) {
337 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
340 s->picture.interlaced_frame = 1;
341 s->picture.top_field_first = 0;
344 avctx->execute(avctx, dv_decode_video_segment, s->sys->work_chunks, NULL,
345 dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
350 *data_size = sizeof(AVFrame);
351 *(AVFrame*)data = s->picture;
353 /* Determine the codec's sample_aspect ratio from the packet */
354 vsc_pack = buf + 80*5 + 48 + 5;
355 if ( *vsc_pack == dv_video_control ) {
357 is16_9 = (vsc_pack && ((vsc_pack[2] & 0x07) == 0x02 || (!apt && (vsc_pack[2] & 0x07) == 0x07)));
358 avctx->sample_aspect_ratio = s->sys->sar[is16_9];
361 return s->sys->frame_size;
364 static int dvvideo_close(AVCodecContext *c)
366 DVVideoContext *s = c->priv_data;
368 if (s->picture.data[0])
369 c->release_buffer(c, &s->picture);
374 AVCodec ff_dvvideo_decoder = {
376 .type = AVMEDIA_TYPE_VIDEO,
377 .id = CODEC_ID_DVVIDEO,
378 .priv_data_size = sizeof(DVVideoContext),
379 .init = ff_dvvideo_init,
380 .close = dvvideo_close,
381 .decode = dvvideo_decode_frame,
382 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS,
384 .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),