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 FFmpeg.
18 * FFmpeg 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 * FFmpeg 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 FFmpeg; if not, write to the Free Software
30 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
38 #include "libavutil/avassert.h"
39 #include "libavutil/internal.h"
40 #include "libavutil/imgutils.h"
41 #include "libavutil/pixdesc.h"
46 #include "simple_idct.h"
50 typedef struct BlockInfo {
51 const uint32_t *factor_table;
52 const uint8_t *scan_table;
53 uint8_t pos; /* position in block */
54 void (*idct_put)(uint8_t *dest, int line_size, int16_t *block);
55 uint8_t partial_bit_count;
56 uint32_t partial_bit_buffer;
60 static const int dv_iweight_bits = 14;
62 static av_cold int dvvideo_decode_init(AVCodecContext *avctx)
64 DVVideoContext *s = avctx->priv_data;
68 memset(&dsp,0, sizeof(dsp));
69 ff_dsputil_init(&dsp, avctx);
71 for (i = 0; i < 64; i++)
72 s->dv_zigzag[0][i] = dsp.idct_permutation[ff_zigzag_direct[i]];
75 for (i = 0; i < 64; i++){
76 int j = ff_dv_zigzag248_direct[i];
77 s->dv_zigzag[1][i] = dsp.idct_permutation[(j & 7) + (j & 8) * 4 + (j & 48) / 2];
80 memcpy(s->dv_zigzag[1], ff_dv_zigzag248_direct, sizeof(s->dv_zigzag[1]));
82 s->idct_put[0] = dsp.idct_put;
83 s->idct_put[1] = ff_simple_idct248_put;
85 return ff_dvvideo_init(avctx);
88 /* decode AC coefficients */
89 static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, int16_t *block)
91 int last_index = gb->size_in_bits;
92 const uint8_t *scan_table = mb->scan_table;
93 const uint32_t *factor_table = mb->factor_table;
95 int partial_bit_count = mb->partial_bit_count;
96 int level, run, vlc_len, index;
101 /* if we must parse a partial VLC, we do it here */
102 if (partial_bit_count > 0) {
103 re_cache = re_cache >> partial_bit_count | mb->partial_bit_buffer;
104 re_index -= partial_bit_count;
105 mb->partial_bit_count = 0;
108 /* get the AC coefficients until last_index is reached */
110 av_dlog(NULL, "%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16),
112 /* our own optimized GET_RL_VLC */
113 index = NEG_USR32(re_cache, TEX_VLC_BITS);
114 vlc_len = ff_dv_rl_vlc[index].len;
116 index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) +
117 ff_dv_rl_vlc[index].level;
118 vlc_len = TEX_VLC_BITS - vlc_len;
120 level = ff_dv_rl_vlc[index].level;
121 run = ff_dv_rl_vlc[index].run;
123 /* gotta check if we're still within gb boundaries */
124 if (re_index + vlc_len > last_index) {
125 /* should be < 16 bits otherwise a codeword could have been parsed */
126 mb->partial_bit_count = last_index - re_index;
127 mb->partial_bit_buffer = re_cache & ~(-1u >> mb->partial_bit_count);
128 re_index = last_index;
133 av_dlog(NULL, "run=%d level=%d\n", run, level);
138 level = (level * factor_table[pos] + (1 << (dv_iweight_bits - 1))) >> dv_iweight_bits;
139 block[scan_table[pos]] = level;
141 UPDATE_CACHE(re, gb);
143 CLOSE_READER(re, gb);
147 static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
149 int bits_left = get_bits_left(gb);
150 while (bits_left >= MIN_CACHE_BITS) {
151 put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
152 bits_left -= MIN_CACHE_BITS;
155 put_bits(pb, bits_left, get_bits(gb, bits_left));
159 /* mb_x and mb_y are in units of 8 pixels */
160 static int dv_decode_video_segment(AVCodecContext *avctx, void *arg)
162 DVVideoContext *s = avctx->priv_data;
163 DVwork_chunk *work_chunk = arg;
164 int quant, dc, dct_mode, class1, j;
165 int mb_index, mb_x, mb_y, last_index;
166 int y_stride, linesize;
167 int16_t *block, *block1;
170 const uint8_t *buf_ptr;
171 PutBitContext pb, vs_pb;
173 BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1;
174 LOCAL_ALIGNED_16(int16_t, sblock, [5*DV_MAX_BPM], [64]);
175 LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [ 80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
176 LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [5*80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
177 const int log2_blocksize = 3-s->avctx->lowres;
178 int is_field_mode[5];
180 av_assert1((((int)mb_bit_buffer) & 7) == 0);
181 av_assert1((((int)vs_bit_buffer) & 7) == 0);
183 memset(sblock, 0, 5*DV_MAX_BPM*sizeof(*sblock));
185 /* pass 1: read DC and AC coefficients in blocks */
186 buf_ptr = &s->buf[work_chunk->buf_offset*80];
187 block1 = &sblock[0][0];
189 init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
190 for (mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) {
192 quant = buf_ptr[3] & 0x0f;
194 init_put_bits(&pb, mb_bit_buffer, 80);
197 is_field_mode[mb_index] = 0;
198 for (j = 0; j < s->sys->bpm; j++) {
199 last_index = s->sys->block_sizes[j];
200 init_get_bits(&gb, buf_ptr, last_index);
203 dc = get_sbits(&gb, 9);
204 dct_mode = get_bits1(&gb);
205 class1 = get_bits(&gb, 2);
206 if (DV_PROFILE_IS_HD(s->sys)) {
207 mb->idct_put = s->idct_put[0];
208 mb->scan_table = s->dv_zigzag[0];
209 mb->factor_table = &s->idct_factor[(j >= 4)*4*16*64 + class1*16*64 + quant*64];
210 is_field_mode[mb_index] |= !j && dct_mode;
212 mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3];
213 mb->scan_table = s->dv_zigzag[dct_mode];
214 mb->factor_table = &s->idct_factor[(class1 == 3)*2*22*64 + dct_mode*22*64 +
215 (quant + ff_dv_quant_offset[class1])*64];
218 /* convert to unsigned because 128 is not added in the
222 buf_ptr += last_index >> 3;
224 mb->partial_bit_count = 0;
226 av_dlog(avctx, "MB block: %d, %d ", mb_index, j);
227 dv_decode_ac(&gb, mb, block);
229 /* write the remaining bits in a new buffer only if the
238 /* pass 2: we can do it just after */
239 av_dlog(avctx, "***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
242 init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
243 put_bits32(&pb, 0); // padding must be zeroed
245 for (j = 0; j < s->sys->bpm; j++, block += 64, mb++) {
246 if (mb->pos < 64 && get_bits_left(&gb) > 0) {
247 dv_decode_ac(&gb, mb, block);
248 /* if still not finished, no need to parse other blocks */
253 /* all blocks are finished, so the extra bytes can be used at
254 the video segment level */
255 if (j >= s->sys->bpm)
256 bit_copy(&vs_pb, &gb);
259 /* we need a pass over the whole video segment */
260 av_dlog(avctx, "***pass 3 size=%d\n", put_bits_count(&vs_pb));
261 block = &sblock[0][0];
263 init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
264 put_bits32(&vs_pb, 0); // padding must be zeroed
265 flush_put_bits(&vs_pb);
266 for (mb_index = 0; mb_index < 5; mb_index++) {
267 for (j = 0; j < s->sys->bpm; j++) {
268 if (mb->pos < 64 && get_bits_left(&gb) > 0) {
269 av_dlog(avctx, "start %d:%d\n", mb_index, j);
270 dv_decode_ac(&gb, mb, block);
272 if (mb->pos >= 64 && mb->pos < 127)
273 av_log(avctx, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);
279 /* compute idct and place blocks */
280 block = &sblock[0][0];
282 for (mb_index = 0; mb_index < 5; mb_index++) {
283 dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y);
285 /* idct_put'ting luminance */
286 if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) ||
287 (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) ||
288 (s->sys->height >= 720 && mb_y != 134)) {
289 y_stride = (s->frame->linesize[0] << ((!is_field_mode[mb_index]) * log2_blocksize));
291 y_stride = (2 << log2_blocksize);
293 y_ptr = s->frame->data[0] + ((mb_y * s->frame->linesize[0] + mb_x) << log2_blocksize);
294 linesize = s->frame->linesize[0] << is_field_mode[mb_index];
295 mb[0] .idct_put(y_ptr , linesize, block + 0*64);
296 if (s->sys->video_stype == 4) { /* SD 422 */
297 mb[2].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 2*64);
299 mb[1].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 1*64);
300 mb[2].idct_put(y_ptr + y_stride, linesize, block + 2*64);
301 mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3*64);
306 /* idct_put'ting chrominance */
307 c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->frame->linesize[1] +
308 (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << log2_blocksize);
309 for (j = 2; j; j--) {
310 uint8_t *c_ptr = s->frame->data[j] + c_offset;
311 if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
312 uint64_t aligned_pixels[64/8];
313 uint8_t *pixels = (uint8_t*)aligned_pixels;
314 uint8_t *c_ptr1, *ptr1;
316 mb->idct_put(pixels, 8, block);
317 for (y = 0; y < (1 << log2_blocksize); y++, c_ptr += s->frame->linesize[j], pixels += 8) {
318 ptr1 = pixels + ((1 << (log2_blocksize))>>1);
319 c_ptr1 = c_ptr + (s->frame->linesize[j] << log2_blocksize);
320 for (x = 0; x < (1 << FFMAX(log2_blocksize - 1, 0)); x++) {
321 c_ptr[x] = pixels[x];
327 y_stride = (mb_y == 134) ? (1 << log2_blocksize) :
328 s->frame->linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize);
329 linesize = s->frame->linesize[j] << is_field_mode[mb_index];
330 (mb++)-> idct_put(c_ptr , linesize, block); block += 64;
331 if (s->sys->bpm == 8) {
332 (mb++)->idct_put(c_ptr + y_stride, linesize, block); block += 64;
340 /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
341 144000 bytes for PAL - or twice those for 50Mbps) */
342 static int dvvideo_decode_frame(AVCodecContext *avctx,
343 void *data, int *got_frame,
346 uint8_t *buf = avpkt->data;
347 int buf_size = avpkt->size;
348 DVVideoContext *s = avctx->priv_data;
349 const uint8_t* vsc_pack;
350 int apt, is16_9, ret;
351 const DVprofile *sys;
353 sys = avpriv_dv_frame_profile2(avctx, s->sys, buf, buf_size);
354 if (!sys || buf_size < sys->frame_size) {
355 av_log(avctx, AV_LOG_ERROR, "could not find dv frame profile\n");
356 return -1; /* NOTE: we only accept several full frames */
360 ret = ff_dv_init_dynamic_tables(s, sys);
362 av_log(avctx, AV_LOG_ERROR, "Error initializing the work tables.\n");
369 s->frame->key_frame = 1;
370 s->frame->pict_type = AV_PICTURE_TYPE_I;
371 avctx->pix_fmt = s->sys->pix_fmt;
372 avctx->time_base = s->sys->time_base;
374 ret = ff_set_dimensions(avctx, s->sys->width, s->sys->height);
378 /* Determine the codec's sample_aspect ratio from the packet */
379 vsc_pack = buf + 80*5 + 48 + 5;
380 if ( *vsc_pack == dv_video_control ) {
382 is16_9 = (vsc_pack && ((vsc_pack[2] & 0x07) == 0x02 || (!apt && (vsc_pack[2] & 0x07) == 0x07)));
383 ff_set_sar(avctx, s->sys->sar[is16_9]);
386 if ((ret = ff_get_buffer(avctx, s->frame, 0)) < 0)
388 s->frame->interlaced_frame = 1;
389 s->frame->top_field_first = 0;
391 /* Determine the codec's field order from the packet */
392 if ( *vsc_pack == dv_video_control ) {
393 s->frame->top_field_first = !(vsc_pack[3] & 0x40);
397 avctx->execute(avctx, dv_decode_video_segment, s->work_chunks, NULL,
398 dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
405 return s->sys->frame_size;
408 AVCodec ff_dvvideo_decoder = {
410 .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),
411 .type = AVMEDIA_TYPE_VIDEO,
412 .id = AV_CODEC_ID_DVVIDEO,
413 .priv_data_size = sizeof(DVVideoContext),
414 .init = dvvideo_decode_init,
415 .decode = dvvideo_decode_frame,
416 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS,