9ee9933cdc298f03205eaed396988b43c4d0d547
[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/imgutils.h"
39 #include "libavutil/internal.h"
40 #include "libavutil/pixdesc.h"
41
42 #include "avcodec.h"
43 #include "dv.h"
44 #include "dvdata.h"
45 #include "get_bits.h"
46 #include "idctdsp.h"
47 #include "internal.h"
48 #include "put_bits.h"
49 #include "simple_idct.h"
50
51 typedef struct BlockInfo {
52 const uint32_t *factor_table;
53 const uint8_t *scan_table;
54 uint8_t pos; /* position in block */
55 void (*idct_put)(uint8_t *dest, int line_size, int16_t *block);
56 uint8_t partial_bit_count;
57 uint32_t partial_bit_buffer;
58 int shift_offset;
59 } BlockInfo;
60
61 static const int dv_iweight_bits = 14;
62
63 static av_cold int dvvideo_decode_init(AVCodecContext *avctx)
64 {
65 DVVideoContext *s = avctx->priv_data;
66 IDCTDSPContext idsp;
67 int i;
68
69 ff_idctdsp_init(&idsp, avctx);
70
71 for (i = 0; i < 64; i++)
72 s->dv_zigzag[0][i] = idsp.idct_permutation[ff_zigzag_direct[i]];
73
74 memcpy(s->dv_zigzag[1], ff_dv_zigzag248_direct, sizeof(s->dv_zigzag[1]));
75
76 s->idct_put[0] = idsp.idct_put;
77 s->idct_put[1] = ff_simple_idct248_put;
78
79 return ff_dvvideo_init(avctx);
80 }
81
82 /* decode AC coefficients */
83 static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, int16_t *block)
84 {
85 int last_index = gb->size_in_bits;
86 const uint8_t *scan_table = mb->scan_table;
87 const uint32_t *factor_table = mb->factor_table;
88 int pos = mb->pos;
89 int partial_bit_count = mb->partial_bit_count;
90 int level, run, vlc_len, index;
91
92 OPEN_READER_NOSIZE(re, gb);
93 UPDATE_CACHE(re, gb);
94
95 /* if we must parse a partial VLC, we do it here */
96 if (partial_bit_count > 0) {
97 re_cache = re_cache >> partial_bit_count |
98 mb->partial_bit_buffer;
99 re_index -= partial_bit_count;
100 mb->partial_bit_count = 0;
101 }
102
103 /* get the AC coefficients until last_index is reached */
104 for (;;) {
105 ff_dlog(NULL, "%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16),
106 re_index);
107 /* our own optimized GET_RL_VLC */
108 index = NEG_USR32(re_cache, TEX_VLC_BITS);
109 vlc_len = ff_dv_rl_vlc[index].len;
110 if (vlc_len < 0) {
111 index = NEG_USR32((unsigned) re_cache << TEX_VLC_BITS, -vlc_len) +
112 ff_dv_rl_vlc[index].level;
113 vlc_len = TEX_VLC_BITS - vlc_len;
114 }
115 level = ff_dv_rl_vlc[index].level;
116 run = ff_dv_rl_vlc[index].run;
117
118 /* gotta check if we're still within gb boundaries */
119 if (re_index + vlc_len > last_index) {
120 /* should be < 16 bits otherwise a codeword could have been parsed */
121 mb->partial_bit_count = last_index - re_index;
122 mb->partial_bit_buffer = re_cache & ~(-1u >> mb->partial_bit_count);
123 re_index = last_index;
124 break;
125 }
126 re_index += vlc_len;
127
128 ff_dlog(NULL, "run=%d level=%d\n", run, level);
129 pos += run;
130 if (pos >= 64)
131 break;
132
133 level = (level * factor_table[pos] + (1 << (dv_iweight_bits - 1))) >>
134 dv_iweight_bits;
135 block[scan_table[pos]] = level;
136
137 UPDATE_CACHE(re, gb);
138 }
139 CLOSE_READER(re, gb);
140 mb->pos = pos;
141 }
142
143 static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
144 {
145 int bits_left = get_bits_left(gb);
146 while (bits_left >= MIN_CACHE_BITS) {
147 put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
148 bits_left -= MIN_CACHE_BITS;
149 }
150 if (bits_left > 0)
151 put_bits(pb, bits_left, get_bits(gb, bits_left));
152 }
153
154 /* mb_x and mb_y are in units of 8 pixels */
155 static int dv_decode_video_segment(AVCodecContext *avctx, void *arg)
156 {
157 DVVideoContext *s = avctx->priv_data;
158 DVwork_chunk *work_chunk = arg;
159 int quant, dc, dct_mode, class1, j;
160 int mb_index, mb_x, mb_y, last_index;
161 int y_stride, linesize;
162 int16_t *block, *block1;
163 int c_offset;
164 uint8_t *y_ptr;
165 const uint8_t *buf_ptr;
166 PutBitContext pb, vs_pb;
167 GetBitContext gb;
168 BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1;
169 LOCAL_ALIGNED_16(int16_t, sblock, [5 * DV_MAX_BPM], [64]);
170 LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [80 + AV_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
171 LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [80 * 5 + AV_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
172 const int log2_blocksize = 3;
173 int is_field_mode[5];
174
175 assert((((int) mb_bit_buffer) & 7) == 0);
176 assert((((int) vs_bit_buffer) & 7) == 0);
177
178 memset(sblock, 0, 5 * DV_MAX_BPM * sizeof(*sblock));
179
180 /* pass 1: read DC and AC coefficients in blocks */
181 buf_ptr = &s->buf[work_chunk->buf_offset * 80];
182 block1 = &sblock[0][0];
183 mb1 = mb_data;
184 init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
185 for (mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) {
186 /* skip header */
187 quant = buf_ptr[3] & 0x0f;
188 buf_ptr += 4;
189 init_put_bits(&pb, mb_bit_buffer, 80);
190 mb = mb1;
191 block = block1;
192 is_field_mode[mb_index] = 0;
193 for (j = 0; j < s->sys->bpm; j++) {
194 last_index = s->sys->block_sizes[j];
195 init_get_bits(&gb, buf_ptr, last_index);
196
197 /* get the DC */
198 dc = get_sbits(&gb, 9);
199 dct_mode = get_bits1(&gb);
200 class1 = get_bits(&gb, 2);
201 if (DV_PROFILE_IS_HD(s->sys)) {
202 mb->idct_put = s->idct_put[0];
203 mb->scan_table = s->dv_zigzag[0];
204 mb->factor_table = &s->idct_factor[(j >= 4) * 4 * 16 * 64 +
205 class1 * 16 * 64 +
206 quant * 64];
207 is_field_mode[mb_index] |= !j && dct_mode;
208 } else {
209 mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3];
210 mb->scan_table = s->dv_zigzag[dct_mode];
211 mb->factor_table =
212 &s->idct_factor[(class1 == 3) * 2 * 22 * 64 +
213 dct_mode * 22 * 64 +
214 (quant + ff_dv_quant_offset[class1]) * 64];
215 }
216 dc = dc << 2;
217 /* convert to unsigned because 128 is not added in the
218 * standard IDCT */
219 dc += 1024;
220 block[0] = dc;
221 buf_ptr += last_index >> 3;
222 mb->pos = 0;
223 mb->partial_bit_count = 0;
224
225 ff_dlog(avctx, "MB block: %d, %d ", mb_index, j);
226 dv_decode_ac(&gb, mb, block);
227
228 /* write the remaining bits in a new buffer only if the
229 * block is finished */
230 if (mb->pos >= 64)
231 bit_copy(&pb, &gb);
232
233 block += 64;
234 mb++;
235 }
236
237 /* pass 2: we can do it just after */
238 ff_dlog(avctx, "***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
239 block = block1;
240 mb = mb1;
241 init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
242 put_bits32(&pb, 0); // padding must be zeroed
243 flush_put_bits(&pb);
244 for (j = 0; j < s->sys->bpm; j++, block += 64, mb++) {
245 if (mb->pos < 64 && get_bits_left(&gb) > 0) {
246 dv_decode_ac(&gb, mb, block);
247 /* if still not finished, no need to parse other blocks */
248 if (mb->pos < 64)
249 break;
250 }
251 }
252 /* all blocks are finished, so the extra bytes can be used at
253 * the video segment level */
254 if (j >= s->sys->bpm)
255 bit_copy(&vs_pb, &gb);
256 }
257
258 /* we need a pass over the whole video segment */
259 ff_dlog(avctx, "***pass 3 size=%d\n", put_bits_count(&vs_pb));
260 block = &sblock[0][0];
261 mb = mb_data;
262 init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
263 put_bits32(&vs_pb, 0); // padding must be zeroed
264 flush_put_bits(&vs_pb);
265 for (mb_index = 0; mb_index < 5; mb_index++) {
266 for (j = 0; j < s->sys->bpm; j++) {
267 if (mb->pos < 64) {
268 ff_dlog(avctx, "start %d:%d\n", mb_index, j);
269 dv_decode_ac(&gb, mb, block);
270 }
271 if (mb->pos >= 64 && mb->pos < 127)
272 av_log(avctx, AV_LOG_ERROR,
273 "AC EOB marker is absent pos=%d\n", mb->pos);
274 block += 64;
275 mb++;
276 }
277 }
278
279 /* compute idct and place blocks */
280 block = &sblock[0][0];
281 mb = mb_data;
282 for (mb_index = 0; mb_index < 5; mb_index++) {
283 dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y);
284
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] <<
290 ((!is_field_mode[mb_index]) * log2_blocksize));
291 } else {
292 y_stride = (2 << log2_blocksize);
293 }
294 y_ptr = s->frame->data[0] +
295 ((mb_y * s->frame->linesize[0] + mb_x) << log2_blocksize);
296 linesize = s->frame->linesize[0] << is_field_mode[mb_index];
297 mb[0].idct_put(y_ptr, linesize, block + 0 * 64);
298 if (s->sys->video_stype == 4) { /* SD 422 */
299 mb[2].idct_put(y_ptr + (1 << log2_blocksize), linesize, block + 2 * 64);
300 } else {
301 mb[1].idct_put(y_ptr + (1 << log2_blocksize), linesize, block + 1 * 64);
302 mb[2].idct_put(y_ptr + y_stride, linesize, block + 2 * 64);
303 mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3 * 64);
304 }
305 mb += 4;
306 block += 4 * 64;
307
308 /* idct_put'ting chrominance */
309 c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->frame->linesize[1] +
310 (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << log2_blocksize);
311 for (j = 2; j; j--) {
312 uint8_t *c_ptr = s->frame->data[j] + c_offset;
313 if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
314 uint64_t aligned_pixels[64 / 8];
315 uint8_t *pixels = (uint8_t *) aligned_pixels;
316 uint8_t *c_ptr1, *ptr1;
317 int x, y;
318 mb->idct_put(pixels, 8, block);
319 for (y = 0; y < (1 << log2_blocksize); y++, c_ptr += s->frame->linesize[j], pixels += 8) {
320 ptr1 = pixels + (1 << (log2_blocksize - 1));
321 c_ptr1 = c_ptr + (s->frame->linesize[j] << log2_blocksize);
322 for (x = 0; x < (1 << (log2_blocksize - 1)); x++) {
323 c_ptr[x] = pixels[x];
324 c_ptr1[x] = ptr1[x];
325 }
326 }
327 block += 64;
328 mb++;
329 } else {
330 y_stride = (mb_y == 134) ? (1 << log2_blocksize) :
331 s->frame->linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize);
332 linesize = s->frame->linesize[j] << is_field_mode[mb_index];
333 (mb++)->idct_put(c_ptr, linesize, block);
334 block += 64;
335 if (s->sys->bpm == 8) {
336 (mb++)->idct_put(c_ptr + y_stride, linesize, block);
337 block += 64;
338 }
339 }
340 }
341 }
342 return 0;
343 }
344
345 /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
346 * 144000 bytes for PAL - or twice those for 50Mbps) */
347 static int dvvideo_decode_frame(AVCodecContext *avctx, void *data,
348 int *got_frame, AVPacket *avpkt)
349 {
350 uint8_t *buf = avpkt->data;
351 int buf_size = avpkt->size;
352 DVVideoContext *s = avctx->priv_data;
353 const uint8_t *vsc_pack;
354 int apt, is16_9, ret;
355 const AVDVProfile *sys;
356
357 sys = av_dv_frame_profile(s->sys, buf, buf_size);
358 if (!sys || buf_size < sys->frame_size) {
359 av_log(avctx, AV_LOG_ERROR, "could not find dv frame profile\n");
360 return -1; /* NOTE: we only accept several full frames */
361 }
362
363 if (sys != s->sys) {
364 ret = ff_dv_init_dynamic_tables(s, sys);
365 if (ret < 0) {
366 av_log(avctx, AV_LOG_ERROR, "Error initializing the work tables.\n");
367 return ret;
368 }
369 s->sys = sys;
370 }
371
372 s->frame = data;
373 s->frame->key_frame = 1;
374 s->frame->pict_type = AV_PICTURE_TYPE_I;
375 avctx->pix_fmt = s->sys->pix_fmt;
376 avctx->framerate = av_inv_q(s->sys->time_base);
377
378 ret = ff_set_dimensions(avctx, s->sys->width, s->sys->height);
379 if (ret < 0)
380 return ret;
381
382 /* Determine the codec's sample_aspect ratio from the packet */
383 vsc_pack = buf + 80 * 5 + 48 + 5;
384 if (*vsc_pack == dv_video_control) {
385 apt = buf[4] & 0x07;
386 is16_9 = (vsc_pack[2] & 0x07) == 0x02 ||
387 (!apt && (vsc_pack[2] & 0x07) == 0x07);
388 ff_set_sar(avctx, s->sys->sar[is16_9]);
389 }
390
391 if (ff_get_buffer(avctx, s->frame, 0) < 0) {
392 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
393 return -1;
394 }
395 s->frame->interlaced_frame = 1;
396 s->frame->top_field_first = 0;
397
398 s->buf = buf;
399 avctx->execute(avctx, dv_decode_video_segment, s->work_chunks, NULL,
400 dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
401
402 emms_c();
403
404 /* return image */
405 *got_frame = 1;
406
407 return s->sys->frame_size;
408 }
409
410 AVCodec ff_dvvideo_decoder = {
411 .name = "dvvideo",
412 .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),
413 .type = AVMEDIA_TYPE_VIDEO,
414 .id = AV_CODEC_ID_DVVIDEO,
415 .priv_data_size = sizeof(DVVideoContext),
416 .init = dvvideo_decode_init,
417 .decode = dvvideo_decode_frame,
418 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_SLICE_THREADS,
419 };