b4cb23f60489bfb69b3d2576e05ad29a992d0d74
[libav.git] / libavcodec / dvdec.c
1 /*
2 * DV decoder
3 * Copyright (c) 2002 Fabrice Bellard
4 * Copyright (c) 2004 Roman Shaposhnik
5 *
6 * 50 Mbps (DVCPRO50) support
7 * Copyright (c) 2006 Daniel Maas <dmaas@maasdigital.com>
8 *
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
12 *
13 * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth
14 * of DV technical info.
15 *
16 * This file is part of Libav.
17 *
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.
22 *
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.
27 *
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
31 */
32
33 /**
34 * @file
35 * DV decoder
36 */
37
38 #include "libavutil/pixdesc.h"
39 #include "avcodec.h"
40 #include "dsputil.h"
41 #include "internal.h"
42 #include "get_bits.h"
43 #include "put_bits.h"
44 #include "simple_idct.h"
45 #include "dvdata.h"
46
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;
54 int shift_offset;
55 } BlockInfo;
56
57 static const int dv_iweight_bits = 14;
58
59 /* decode AC coefficients */
60 static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block)
61 {
62 int last_index = gb->size_in_bits;
63 const uint8_t *scan_table = mb->scan_table;
64 const uint32_t *factor_table = mb->factor_table;
65 int pos = mb->pos;
66 int partial_bit_count = mb->partial_bit_count;
67 int level, run, vlc_len, index;
68
69 OPEN_READER(re, gb);
70 UPDATE_CACHE(re, gb);
71
72 /* if we must parse a partial VLC, we do it here */
73 if (partial_bit_count > 0) {
74 re_cache = re_cache >> partial_bit_count | mb->partial_bit_buffer;
75 re_index -= partial_bit_count;
76 mb->partial_bit_count = 0;
77 }
78
79 /* get the AC coefficients until last_index is reached */
80 for (;;) {
81 av_dlog(NULL, "%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16),
82 re_index);
83 /* our own optimized GET_RL_VLC */
84 index = NEG_USR32(re_cache, TEX_VLC_BITS);
85 vlc_len = ff_dv_rl_vlc[index].len;
86 if (vlc_len < 0) {
87 index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) +
88 ff_dv_rl_vlc[index].level;
89 vlc_len = TEX_VLC_BITS - vlc_len;
90 }
91 level = ff_dv_rl_vlc[index].level;
92 run = ff_dv_rl_vlc[index].run;
93
94 /* gotta check if we're still within gb boundaries */
95 if (re_index + vlc_len > last_index) {
96 /* should be < 16 bits otherwise a codeword could have been parsed */
97 mb->partial_bit_count = last_index - re_index;
98 mb->partial_bit_buffer = re_cache & ~(-1u >> mb->partial_bit_count);
99 re_index = last_index;
100 break;
101 }
102 re_index += vlc_len;
103
104 av_dlog(NULL, "run=%d level=%d\n", run, level);
105 pos += run;
106 if (pos >= 64)
107 break;
108
109 level = (level * factor_table[pos] + (1 << (dv_iweight_bits - 1))) >> dv_iweight_bits;
110 block[scan_table[pos]] = level;
111
112 UPDATE_CACHE(re, gb);
113 }
114 CLOSE_READER(re, gb);
115 mb->pos = pos;
116 }
117
118 static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
119 {
120 int bits_left = get_bits_left(gb);
121 while (bits_left >= MIN_CACHE_BITS) {
122 put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
123 bits_left -= MIN_CACHE_BITS;
124 }
125 if (bits_left > 0) {
126 put_bits(pb, bits_left, get_bits(gb, bits_left));
127 }
128 }
129
130 /* mb_x and mb_y are in units of 8 pixels */
131 static int dv_decode_video_segment(AVCodecContext *avctx, void *arg)
132 {
133 DVVideoContext *s = avctx->priv_data;
134 DVwork_chunk *work_chunk = arg;
135 int quant, dc, dct_mode, class1, j;
136 int mb_index, mb_x, mb_y, last_index;
137 int y_stride, linesize;
138 DCTELEM *block, *block1;
139 int c_offset;
140 uint8_t *y_ptr;
141 const uint8_t *buf_ptr;
142 PutBitContext pb, vs_pb;
143 GetBitContext gb;
144 BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1;
145 LOCAL_ALIGNED_16(DCTELEM, sblock, [5*DV_MAX_BPM], [64]);
146 LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [ 80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
147 LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [5*80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
148 const int log2_blocksize = 3;
149 int is_field_mode[5];
150
151 assert((((int)mb_bit_buffer) & 7) == 0);
152 assert((((int)vs_bit_buffer) & 7) == 0);
153
154 memset(sblock, 0, 5*DV_MAX_BPM*sizeof(*sblock));
155
156 /* pass 1: read DC and AC coefficients in blocks */
157 buf_ptr = &s->buf[work_chunk->buf_offset*80];
158 block1 = &sblock[0][0];
159 mb1 = mb_data;
160 init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
161 for (mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) {
162 /* skip header */
163 quant = buf_ptr[3] & 0x0f;
164 buf_ptr += 4;
165 init_put_bits(&pb, mb_bit_buffer, 80);
166 mb = mb1;
167 block = block1;
168 is_field_mode[mb_index] = 0;
169 for (j = 0; j < s->sys->bpm; j++) {
170 last_index = s->sys->block_sizes[j];
171 init_get_bits(&gb, buf_ptr, last_index);
172
173 /* get the DC */
174 dc = get_sbits(&gb, 9);
175 dct_mode = get_bits1(&gb);
176 class1 = get_bits(&gb, 2);
177 if (DV_PROFILE_IS_HD(s->sys)) {
178 mb->idct_put = s->idct_put[0];
179 mb->scan_table = s->dv_zigzag[0];
180 mb->factor_table = &s->sys->idct_factor[(j >= 4)*4*16*64 + class1*16*64 + quant*64];
181 is_field_mode[mb_index] |= !j && dct_mode;
182 } else {
183 mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3];
184 mb->scan_table = s->dv_zigzag[dct_mode];
185 mb->factor_table = &s->sys->idct_factor[(class1 == 3)*2*22*64 + dct_mode*22*64 +
186 (quant + ff_dv_quant_offset[class1])*64];
187 }
188 dc = dc << 2;
189 /* convert to unsigned because 128 is not added in the
190 standard IDCT */
191 dc += 1024;
192 block[0] = dc;
193 buf_ptr += last_index >> 3;
194 mb->pos = 0;
195 mb->partial_bit_count = 0;
196
197 av_dlog(avctx, "MB block: %d, %d ", mb_index, j);
198 dv_decode_ac(&gb, mb, block);
199
200 /* write the remaining bits in a new buffer only if the
201 block is finished */
202 if (mb->pos >= 64)
203 bit_copy(&pb, &gb);
204
205 block += 64;
206 mb++;
207 }
208
209 /* pass 2: we can do it just after */
210 av_dlog(avctx, "***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
211 block = block1;
212 mb = mb1;
213 init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
214 put_bits32(&pb, 0); // padding must be zeroed
215 flush_put_bits(&pb);
216 for (j = 0; j < s->sys->bpm; j++, block += 64, mb++) {
217 if (mb->pos < 64 && get_bits_left(&gb) > 0) {
218 dv_decode_ac(&gb, mb, block);
219 /* if still not finished, no need to parse other blocks */
220 if (mb->pos < 64)
221 break;
222 }
223 }
224 /* all blocks are finished, so the extra bytes can be used at
225 the video segment level */
226 if (j >= s->sys->bpm)
227 bit_copy(&vs_pb, &gb);
228 }
229
230 /* we need a pass over the whole video segment */
231 av_dlog(avctx, "***pass 3 size=%d\n", put_bits_count(&vs_pb));
232 block = &sblock[0][0];
233 mb = mb_data;
234 init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
235 put_bits32(&vs_pb, 0); // padding must be zeroed
236 flush_put_bits(&vs_pb);
237 for (mb_index = 0; mb_index < 5; mb_index++) {
238 for (j = 0; j < s->sys->bpm; j++) {
239 if (mb->pos < 64) {
240 av_dlog(avctx, "start %d:%d\n", mb_index, j);
241 dv_decode_ac(&gb, mb, block);
242 }
243 if (mb->pos >= 64 && mb->pos < 127)
244 av_log(avctx, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);
245 block += 64;
246 mb++;
247 }
248 }
249
250 /* compute idct and place blocks */
251 block = &sblock[0][0];
252 mb = mb_data;
253 for (mb_index = 0; mb_index < 5; mb_index++) {
254 dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y);
255
256 /* idct_put'ting luminance */
257 if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) ||
258 (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) ||
259 (s->sys->height >= 720 && mb_y != 134)) {
260 y_stride = (s->picture.linesize[0] << ((!is_field_mode[mb_index]) * log2_blocksize));
261 } else {
262 y_stride = (2 << log2_blocksize);
263 }
264 y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x) << log2_blocksize);
265 linesize = s->picture.linesize[0] << is_field_mode[mb_index];
266 mb[0] .idct_put(y_ptr , linesize, block + 0*64);
267 if (s->sys->video_stype == 4) { /* SD 422 */
268 mb[2].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 2*64);
269 } else {
270 mb[1].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 1*64);
271 mb[2].idct_put(y_ptr + y_stride, linesize, block + 2*64);
272 mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3*64);
273 }
274 mb += 4;
275 block += 4*64;
276
277 /* idct_put'ting chrominance */
278 c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->picture.linesize[1] +
279 (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << log2_blocksize);
280 for (j = 2; j; j--) {
281 uint8_t *c_ptr = s->picture.data[j] + c_offset;
282 if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
283 uint64_t aligned_pixels[64/8];
284 uint8_t *pixels = (uint8_t*)aligned_pixels;
285 uint8_t *c_ptr1, *ptr1;
286 int x, y;
287 mb->idct_put(pixels, 8, block);
288 for (y = 0; y < (1 << log2_blocksize); y++, c_ptr += s->picture.linesize[j], pixels += 8) {
289 ptr1 = pixels + (1 << (log2_blocksize - 1));
290 c_ptr1 = c_ptr + (s->picture.linesize[j] << log2_blocksize);
291 for (x = 0; x < (1 << (log2_blocksize - 1)); x++) {
292 c_ptr[x] = pixels[x];
293 c_ptr1[x] = ptr1[x];
294 }
295 }
296 block += 64; mb++;
297 } else {
298 y_stride = (mb_y == 134) ? (1 << log2_blocksize) :
299 s->picture.linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize);
300 linesize = s->picture.linesize[j] << is_field_mode[mb_index];
301 (mb++)-> idct_put(c_ptr , linesize, block); block += 64;
302 if (s->sys->bpm == 8) {
303 (mb++)->idct_put(c_ptr + y_stride, linesize, block); block += 64;
304 }
305 }
306 }
307 }
308 return 0;
309 }
310
311 /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
312 144000 bytes for PAL - or twice those for 50Mbps) */
313 static int dvvideo_decode_frame(AVCodecContext *avctx,
314 void *data, int *data_size,
315 AVPacket *avpkt)
316 {
317 uint8_t *buf = avpkt->data;
318 int buf_size = avpkt->size;
319 DVVideoContext *s = avctx->priv_data;
320 const uint8_t* vsc_pack;
321 int apt, is16_9;
322
323 s->sys = avpriv_dv_frame_profile(s->sys, buf, buf_size);
324 if (!s->sys || buf_size < s->sys->frame_size || ff_dv_init_dynamic_tables(s->sys)) {
325 av_log(avctx, AV_LOG_ERROR, "could not find dv frame profile\n");
326 return -1; /* NOTE: we only accept several full frames */
327 }
328
329 if (s->picture.data[0])
330 avctx->release_buffer(avctx, &s->picture);
331
332 s->picture.reference = 0;
333 s->picture.key_frame = 1;
334 s->picture.pict_type = AV_PICTURE_TYPE_I;
335 avctx->pix_fmt = s->sys->pix_fmt;
336 avctx->time_base = s->sys->time_base;
337 avcodec_set_dimensions(avctx, s->sys->width, s->sys->height);
338 if (ff_get_buffer(avctx, &s->picture) < 0) {
339 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
340 return -1;
341 }
342 s->picture.interlaced_frame = 1;
343 s->picture.top_field_first = 0;
344
345 s->buf = buf;
346 avctx->execute(avctx, dv_decode_video_segment, s->sys->work_chunks, NULL,
347 dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
348
349 emms_c();
350
351 /* return image */
352 *data_size = sizeof(AVFrame);
353 *(AVFrame*)data = s->picture;
354
355 /* Determine the codec's sample_aspect ratio from the packet */
356 vsc_pack = buf + 80*5 + 48 + 5;
357 if ( *vsc_pack == dv_video_control ) {
358 apt = buf[4] & 0x07;
359 is16_9 = (vsc_pack && ((vsc_pack[2] & 0x07) == 0x02 || (!apt && (vsc_pack[2] & 0x07) == 0x07)));
360 avctx->sample_aspect_ratio = s->sys->sar[is16_9];
361 }
362
363 return s->sys->frame_size;
364 }
365
366 static int dvvideo_close(AVCodecContext *c)
367 {
368 DVVideoContext *s = c->priv_data;
369
370 if (s->picture.data[0])
371 c->release_buffer(c, &s->picture);
372
373 return 0;
374 }
375
376 AVCodec ff_dvvideo_decoder = {
377 .name = "dvvideo",
378 .type = AVMEDIA_TYPE_VIDEO,
379 .id = AV_CODEC_ID_DVVIDEO,
380 .priv_data_size = sizeof(DVVideoContext),
381 .init = ff_dvvideo_init,
382 .close = dvvideo_close,
383 .decode = dvvideo_decode_frame,
384 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS,
385 .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),
386 };