2917de93fa31541a1673d9bf6996762c5e3a0f64
[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 "get_bits.h"
42 #include "put_bits.h"
43 #include "simple_idct.h"
44 #include "dvdata.h"
45
46 typedef struct BlockInfo {
47 const uint32_t *factor_table;
48 const uint8_t *scan_table;
49 uint8_t pos; /* position in block */
50 void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block);
51 uint8_t partial_bit_count;
52 uint32_t partial_bit_buffer;
53 int shift_offset;
54 } BlockInfo;
55
56 static const int dv_iweight_bits = 14;
57
58 /* decode AC coefficients */
59 static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block)
60 {
61 int last_index = gb->size_in_bits;
62 const uint8_t *scan_table = mb->scan_table;
63 const uint32_t *factor_table = mb->factor_table;
64 int pos = mb->pos;
65 int partial_bit_count = mb->partial_bit_count;
66 int level, run, vlc_len, index;
67
68 OPEN_READER(re, gb);
69 UPDATE_CACHE(re, gb);
70
71 /* if we must parse a partial VLC, we do it here */
72 if (partial_bit_count > 0) {
73 re_cache = re_cache >> partial_bit_count | mb->partial_bit_buffer;
74 re_index -= partial_bit_count;
75 mb->partial_bit_count = 0;
76 }
77
78 /* get the AC coefficients until last_index is reached */
79 for (;;) {
80 av_dlog(NULL, "%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16),
81 re_index);
82 /* our own optimized GET_RL_VLC */
83 index = NEG_USR32(re_cache, TEX_VLC_BITS);
84 vlc_len = ff_dv_rl_vlc[index].len;
85 if (vlc_len < 0) {
86 index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) +
87 ff_dv_rl_vlc[index].level;
88 vlc_len = TEX_VLC_BITS - vlc_len;
89 }
90 level = ff_dv_rl_vlc[index].level;
91 run = ff_dv_rl_vlc[index].run;
92
93 /* gotta check if we're still within gb boundaries */
94 if (re_index + vlc_len > last_index) {
95 /* should be < 16 bits otherwise a codeword could have been parsed */
96 mb->partial_bit_count = last_index - re_index;
97 mb->partial_bit_buffer = re_cache & ~(-1u >> mb->partial_bit_count);
98 re_index = last_index;
99 break;
100 }
101 re_index += vlc_len;
102
103 av_dlog(NULL, "run=%d level=%d\n", run, level);
104 pos += run;
105 if (pos >= 64)
106 break;
107
108 level = (level * factor_table[pos] + (1 << (dv_iweight_bits - 1))) >> dv_iweight_bits;
109 block[scan_table[pos]] = level;
110
111 UPDATE_CACHE(re, gb);
112 }
113 CLOSE_READER(re, gb);
114 mb->pos = pos;
115 }
116
117 static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
118 {
119 int bits_left = get_bits_left(gb);
120 while (bits_left >= MIN_CACHE_BITS) {
121 put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
122 bits_left -= MIN_CACHE_BITS;
123 }
124 if (bits_left > 0) {
125 put_bits(pb, bits_left, get_bits(gb, bits_left));
126 }
127 }
128
129 /* mb_x and mb_y are in units of 8 pixels */
130 static int dv_decode_video_segment(AVCodecContext *avctx, void *arg)
131 {
132 DVVideoContext *s = avctx->priv_data;
133 DVwork_chunk *work_chunk = arg;
134 int quant, dc, dct_mode, class1, j;
135 int mb_index, mb_x, mb_y, last_index;
136 int y_stride, linesize;
137 DCTELEM *block, *block1;
138 int c_offset;
139 uint8_t *y_ptr;
140 const uint8_t *buf_ptr;
141 PutBitContext pb, vs_pb;
142 GetBitContext gb;
143 BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1;
144 LOCAL_ALIGNED_16(DCTELEM, sblock, [5*DV_MAX_BPM], [64]);
145 LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [ 80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
146 LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [5*80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
147 const int log2_blocksize = 3;
148 int is_field_mode[5];
149
150 assert((((int)mb_bit_buffer) & 7) == 0);
151 assert((((int)vs_bit_buffer) & 7) == 0);
152
153 memset(sblock, 0, 5*DV_MAX_BPM*sizeof(*sblock));
154
155 /* pass 1: read DC and AC coefficients in blocks */
156 buf_ptr = &s->buf[work_chunk->buf_offset*80];
157 block1 = &sblock[0][0];
158 mb1 = mb_data;
159 init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
160 for (mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) {
161 /* skip header */
162 quant = buf_ptr[3] & 0x0f;
163 buf_ptr += 4;
164 init_put_bits(&pb, mb_bit_buffer, 80);
165 mb = mb1;
166 block = block1;
167 is_field_mode[mb_index] = 0;
168 for (j = 0; j < s->sys->bpm; j++) {
169 last_index = s->sys->block_sizes[j];
170 init_get_bits(&gb, buf_ptr, last_index);
171
172 /* get the DC */
173 dc = get_sbits(&gb, 9);
174 dct_mode = get_bits1(&gb);
175 class1 = get_bits(&gb, 2);
176 if (DV_PROFILE_IS_HD(s->sys)) {
177 mb->idct_put = s->idct_put[0];
178 mb->scan_table = s->dv_zigzag[0];
179 mb->factor_table = &s->sys->idct_factor[(j >= 4)*4*16*64 + class1*16*64 + quant*64];
180 is_field_mode[mb_index] |= !j && dct_mode;
181 } else {
182 mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3];
183 mb->scan_table = s->dv_zigzag[dct_mode];
184 mb->factor_table = &s->sys->idct_factor[(class1 == 3)*2*22*64 + dct_mode*22*64 +
185 (quant + ff_dv_quant_offset[class1])*64];
186 }
187 dc = dc << 2;
188 /* convert to unsigned because 128 is not added in the
189 standard IDCT */
190 dc += 1024;
191 block[0] = dc;
192 buf_ptr += last_index >> 3;
193 mb->pos = 0;
194 mb->partial_bit_count = 0;
195
196 av_dlog(avctx, "MB block: %d, %d ", mb_index, j);
197 dv_decode_ac(&gb, mb, block);
198
199 /* write the remaining bits in a new buffer only if the
200 block is finished */
201 if (mb->pos >= 64)
202 bit_copy(&pb, &gb);
203
204 block += 64;
205 mb++;
206 }
207
208 /* pass 2: we can do it just after */
209 av_dlog(avctx, "***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
210 block = block1;
211 mb = mb1;
212 init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
213 put_bits32(&pb, 0); // padding must be zeroed
214 flush_put_bits(&pb);
215 for (j = 0; j < s->sys->bpm; j++, block += 64, mb++) {
216 if (mb->pos < 64 && get_bits_left(&gb) > 0) {
217 dv_decode_ac(&gb, mb, block);
218 /* if still not finished, no need to parse other blocks */
219 if (mb->pos < 64)
220 break;
221 }
222 }
223 /* all blocks are finished, so the extra bytes can be used at
224 the video segment level */
225 if (j >= s->sys->bpm)
226 bit_copy(&vs_pb, &gb);
227 }
228
229 /* we need a pass over the whole video segment */
230 av_dlog(avctx, "***pass 3 size=%d\n", put_bits_count(&vs_pb));
231 block = &sblock[0][0];
232 mb = mb_data;
233 init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
234 put_bits32(&vs_pb, 0); // padding must be zeroed
235 flush_put_bits(&vs_pb);
236 for (mb_index = 0; mb_index < 5; mb_index++) {
237 for (j = 0; j < s->sys->bpm; j++) {
238 if (mb->pos < 64) {
239 av_dlog(avctx, "start %d:%d\n", mb_index, j);
240 dv_decode_ac(&gb, mb, block);
241 }
242 if (mb->pos >= 64 && mb->pos < 127)
243 av_log(avctx, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);
244 block += 64;
245 mb++;
246 }
247 }
248
249 /* compute idct and place blocks */
250 block = &sblock[0][0];
251 mb = mb_data;
252 for (mb_index = 0; mb_index < 5; mb_index++) {
253 dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y);
254
255 /* idct_put'ting luminance */
256 if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) ||
257 (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) ||
258 (s->sys->height >= 720 && mb_y != 134)) {
259 y_stride = (s->picture.linesize[0] << ((!is_field_mode[mb_index]) * log2_blocksize));
260 } else {
261 y_stride = (2 << log2_blocksize);
262 }
263 y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x) << log2_blocksize);
264 linesize = s->picture.linesize[0] << is_field_mode[mb_index];
265 mb[0] .idct_put(y_ptr , linesize, block + 0*64);
266 if (s->sys->video_stype == 4) { /* SD 422 */
267 mb[2].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 2*64);
268 } else {
269 mb[1].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 1*64);
270 mb[2].idct_put(y_ptr + y_stride, linesize, block + 2*64);
271 mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3*64);
272 }
273 mb += 4;
274 block += 4*64;
275
276 /* idct_put'ting chrominance */
277 c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->picture.linesize[1] +
278 (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << log2_blocksize);
279 for (j = 2; j; j--) {
280 uint8_t *c_ptr = s->picture.data[j] + c_offset;
281 if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
282 uint64_t aligned_pixels[64/8];
283 uint8_t *pixels = (uint8_t*)aligned_pixels;
284 uint8_t *c_ptr1, *ptr1;
285 int x, y;
286 mb->idct_put(pixels, 8, block);
287 for (y = 0; y < (1 << log2_blocksize); y++, c_ptr += s->picture.linesize[j], pixels += 8) {
288 ptr1 = pixels + (1 << (log2_blocksize - 1));
289 c_ptr1 = c_ptr + (s->picture.linesize[j] << log2_blocksize);
290 for (x = 0; x < (1 << (log2_blocksize - 1)); x++) {
291 c_ptr[x] = pixels[x];
292 c_ptr1[x] = ptr1[x];
293 }
294 }
295 block += 64; mb++;
296 } else {
297 y_stride = (mb_y == 134) ? (1 << log2_blocksize) :
298 s->picture.linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize);
299 linesize = s->picture.linesize[j] << is_field_mode[mb_index];
300 (mb++)-> idct_put(c_ptr , linesize, block); block += 64;
301 if (s->sys->bpm == 8) {
302 (mb++)->idct_put(c_ptr + y_stride, linesize, block); block += 64;
303 }
304 }
305 }
306 }
307 return 0;
308 }
309
310 /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
311 144000 bytes for PAL - or twice those for 50Mbps) */
312 static int dvvideo_decode_frame(AVCodecContext *avctx,
313 void *data, int *data_size,
314 AVPacket *avpkt)
315 {
316 uint8_t *buf = avpkt->data;
317 int buf_size = avpkt->size;
318 DVVideoContext *s = avctx->priv_data;
319 const uint8_t* vsc_pack;
320 int apt, is16_9;
321
322 s->sys = avpriv_dv_frame_profile(s->sys, buf, buf_size);
323 if (!s->sys || buf_size < s->sys->frame_size || ff_dv_init_dynamic_tables(s->sys)) {
324 av_log(avctx, AV_LOG_ERROR, "could not find dv frame profile\n");
325 return -1; /* NOTE: we only accept several full frames */
326 }
327
328 if (s->picture.data[0])
329 avctx->release_buffer(avctx, &s->picture);
330
331 s->picture.reference = 0;
332 s->picture.key_frame = 1;
333 s->picture.pict_type = AV_PICTURE_TYPE_I;
334 avctx->pix_fmt = s->sys->pix_fmt;
335 avctx->time_base = s->sys->time_base;
336 avcodec_set_dimensions(avctx, s->sys->width, s->sys->height);
337 if (avctx->get_buffer(avctx, &s->picture) < 0) {
338 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
339 return -1;
340 }
341 s->picture.interlaced_frame = 1;
342 s->picture.top_field_first = 0;
343
344 s->buf = buf;
345 avctx->execute(avctx, dv_decode_video_segment, s->sys->work_chunks, NULL,
346 dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
347
348 emms_c();
349
350 /* return image */
351 *data_size = sizeof(AVFrame);
352 *(AVFrame*)data = s->picture;
353
354 /* Determine the codec's sample_aspect ratio from the packet */
355 vsc_pack = buf + 80*5 + 48 + 5;
356 if ( *vsc_pack == dv_video_control ) {
357 apt = buf[4] & 0x07;
358 is16_9 = (vsc_pack && ((vsc_pack[2] & 0x07) == 0x02 || (!apt && (vsc_pack[2] & 0x07) == 0x07)));
359 avctx->sample_aspect_ratio = s->sys->sar[is16_9];
360 }
361
362 return s->sys->frame_size;
363 }
364
365 static int dvvideo_close(AVCodecContext *c)
366 {
367 DVVideoContext *s = c->priv_data;
368
369 if (s->picture.data[0])
370 c->release_buffer(c, &s->picture);
371
372 return 0;
373 }
374
375 AVCodec ff_dvvideo_decoder = {
376 .name = "dvvideo",
377 .type = AVMEDIA_TYPE_VIDEO,
378 .id = AV_CODEC_ID_DVVIDEO,
379 .priv_data_size = sizeof(DVVideoContext),
380 .init = ff_dvvideo_init,
381 .close = dvvideo_close,
382 .decode = dvvideo_decode_frame,
383 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS,
384 .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),
385 };