Statistics
| Branch: | Revision:

ffmpeg / libavcodec / crystalhd.c @ ae7a4a15

History | View | Annotate | Download (36.6 KB)

1
/*
2
 * - CrystalHD decoder module -
3
 *
4
 * Copyright(C) 2010,2011 Philip Langdale <ffmpeg.philipl@overt.org>
5
 *
6
 * This file is part of FFmpeg.
7
 *
8
 * FFmpeg is free software; you can redistribute it and/or
9
 * modify it under the terms of the GNU Lesser General Public
10
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
12
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16
 * Lesser General Public License for more details.
17
 *
18
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with FFmpeg; if not, write to the Free Software
20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21
 */
22

    
23
/*
24
 * - Principles of Operation -
25
 *
26
 * The CrystalHD decoder operates at the bitstream level - which is an even
27
 * higher level than the decoding hardware you typically see in modern GPUs.
28
 * This means it has a very simple interface, in principle. You feed demuxed
29
 * packets in one end and get decoded picture (fields/frames) out the other.
30
 *
31
 * Of course, nothing is ever that simple. Due, at the very least, to b-frame
32
 * dependencies in the supported formats, the hardware has a delay between
33
 * when a packet goes in, and when a picture comes out. Furthermore, this delay
34
 * is not just a function of time, but also one of the dependency on additional
35
 * frames being fed into the decoder to satisfy the b-frame dependencies.
36
 *
37
 * As such, a pipeline will build up that is roughly equivalent to the required
38
 * DPB for the file being played. If that was all it took, things would still
39
 * be simple - so, of course, it isn't.
40
 *
41
 * The hardware has a way of indicating that a picture is ready to be copied out,
42
 * but this is unreliable - and sometimes the attempt will still fail so, based
43
 * on testing, the code will wait until 3 pictures are ready before starting
44
 * to copy out - and this has the effect of extending the pipeline.
45
 *
46
 * Finally, while it is tempting to say that once the decoder starts outputing
47
 * frames, the software should never fail to return a frame from a decode(),
48
 * this is a hard assertion to make, because the stream may switch between
49
 * differently encoded content (number of b-frames, interlacing, etc) which
50
 * might require a longer pipeline than before. If that happened, you could
51
 * deadlock trying to retrieve a frame that can't be decoded without feeding
52
 * in additional packets.
53
 *
54
 * As such, the code will return in the event that a picture cannot be copied
55
 * out, leading to an increase in the length of the pipeline. This in turn,
56
 * means we have to be sensitive to the time it takes to decode a picture;
57
 * We do not want to give up just because the hardware needed a little more
58
 * time to prepare the picture! For this reason, there are delays included
59
 * in the decode() path that ensure that, under normal conditions, the hardware
60
 * will only fail to return a frame if it really needs additional packets to
61
 * complete the decoding.
62
 *
63
 * Finally, to be explicit, we do not want the pipeline to grow without bound
64
 * for two reasons: 1) The hardware can only buffer a finite number of packets,
65
 * and 2) The client application may not be able to cope with arbitrarily long
66
 * delays in the video path relative to the audio path. For example. MPlayer
67
 * can only handle a 20 picture delay (although this is arbitrary, and needs
68
 * to be extended to fully support the CrystalHD where the delay could be up
69
 * to 32 pictures - consider PAFF H.264 content with 16 b-frames).
70
 */
71

    
72
/*****************************************************************************
73
 * Includes
74
 ****************************************************************************/
75

    
76
#define _XOPEN_SOURCE 600
77
#include <inttypes.h>
78
#include <stdio.h>
79
#include <stdlib.h>
80
#include <unistd.h>
81

    
82
#include <libcrystalhd/bc_dts_types.h>
83
#include <libcrystalhd/bc_dts_defs.h>
84
#include <libcrystalhd/libcrystalhd_if.h>
85

    
86
#include "avcodec.h"
87
#include "h264.h"
88
#include "libavutil/imgutils.h"
89
#include "libavutil/intreadwrite.h"
90

    
91
/** Timeout parameter passed to DtsProcOutput() in us */
92
#define OUTPUT_PROC_TIMEOUT 50
93
/** Step between fake timestamps passed to hardware in units of 100ns */
94
#define TIMESTAMP_UNIT 100000
95
/** Initial value in us of the wait in decode() */
96
#define BASE_WAIT 10000
97
/** Increment in us to adjust wait in decode() */
98
#define WAIT_UNIT 1000
99

    
100

    
101
/*****************************************************************************
102
 * Module private data
103
 ****************************************************************************/
104

    
105
typedef enum {
106
    RET_ERROR           = -1,
107
    RET_OK              = 0,
108
    RET_COPY_AGAIN      = 1,
109
    RET_SKIP_NEXT_COPY  = 2,
110
    RET_COPY_NEXT_FIELD = 3,
111
} CopyRet;
112

    
113
typedef struct OpaqueList {
114
    struct OpaqueList *next;
115
    uint64_t fake_timestamp;
116
    uint64_t reordered_opaque;
117
    uint8_t pic_type;
118
} OpaqueList;
119

    
120
typedef struct {
121
    AVCodecContext *avctx;
122
    AVFrame pic;
123
    HANDLE dev;
124

    
125
    AVCodecParserContext *parser;
126

    
127
    uint8_t is_70012;
128
    uint8_t *sps_pps_buf;
129
    uint32_t sps_pps_size;
130
    uint8_t is_nal;
131
    uint8_t output_ready;
132
    uint8_t need_second_field;
133
    uint8_t skip_next_output;
134
    uint64_t decode_wait;
135

    
136
    uint64_t last_picture;
137

    
138
    OpaqueList *head;
139
    OpaqueList *tail;
140
} CHDContext;
141

    
142

    
143
/*****************************************************************************
144
 * Helper functions
145
 ****************************************************************************/
146

    
147
static inline BC_MEDIA_SUBTYPE id2subtype(CHDContext *priv, enum CodecID id)
148
{
149
    switch (id) {
150
    case CODEC_ID_MPEG4:
151
        return BC_MSUBTYPE_DIVX;
152
    case CODEC_ID_MSMPEG4V3:
153
        return BC_MSUBTYPE_DIVX311;
154
    case CODEC_ID_MPEG2VIDEO:
155
        return BC_MSUBTYPE_MPEG2VIDEO;
156
    case CODEC_ID_VC1:
157
        return BC_MSUBTYPE_VC1;
158
    case CODEC_ID_WMV3:
159
        return BC_MSUBTYPE_WMV3;
160
    case CODEC_ID_H264:
161
        return priv->is_nal ? BC_MSUBTYPE_AVC1 : BC_MSUBTYPE_H264;
162
    default:
163
        return BC_MSUBTYPE_INVALID;
164
    }
165
}
166

    
167
static inline void print_frame_info(CHDContext *priv, BC_DTS_PROC_OUT *output)
168
{
169
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tYBuffSz: %u\n", output->YbuffSz);
170
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tYBuffDoneSz: %u\n",
171
           output->YBuffDoneSz);
172
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tUVBuffDoneSz: %u\n",
173
           output->UVBuffDoneSz);
174
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tTimestamp: %"PRIu64"\n",
175
           output->PicInfo.timeStamp);
176
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tPicture Number: %u\n",
177
           output->PicInfo.picture_number);
178
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tWidth: %u\n",
179
           output->PicInfo.width);
180
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tHeight: %u\n",
181
           output->PicInfo.height);
182
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tChroma: 0x%03x\n",
183
           output->PicInfo.chroma_format);
184
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tPulldown: %u\n",
185
           output->PicInfo.pulldown);
186
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tFlags: 0x%08x\n",
187
           output->PicInfo.flags);
188
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tFrame Rate/Res: %u\n",
189
           output->PicInfo.frame_rate);
190
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tAspect Ratio: %u\n",
191
           output->PicInfo.aspect_ratio);
192
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tColor Primaries: %u\n",
193
           output->PicInfo.colour_primaries);
194
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tMetaData: %u\n",
195
           output->PicInfo.picture_meta_payload);
196
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tSession Number: %u\n",
197
           output->PicInfo.sess_num);
198
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tycom: %u\n",
199
           output->PicInfo.ycom);
200
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tCustom Aspect: %u\n",
201
           output->PicInfo.custom_aspect_ratio_width_height);
202
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tFrames to Drop: %u\n",
203
           output->PicInfo.n_drop);
204
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tH264 Valid Fields: 0x%08x\n",
205
           output->PicInfo.other.h264.valid);
206
}
207

    
208

    
209
/*****************************************************************************
210
 * OpaqueList functions
211
 ****************************************************************************/
212

    
213
static uint64_t opaque_list_push(CHDContext *priv, uint64_t reordered_opaque,
214
                                 uint8_t pic_type)
215
{
216
    OpaqueList *newNode = av_mallocz(sizeof (OpaqueList));
217
    if (!newNode) {
218
        av_log(priv->avctx, AV_LOG_ERROR,
219
               "Unable to allocate new node in OpaqueList.\n");
220
        return 0;
221
    }
222
    if (!priv->head) {
223
        newNode->fake_timestamp = TIMESTAMP_UNIT;
224
        priv->head              = newNode;
225
    } else {
226
        newNode->fake_timestamp = priv->tail->fake_timestamp + TIMESTAMP_UNIT;
227
        priv->tail->next        = newNode;
228
    }
229
    priv->tail = newNode;
230
    newNode->reordered_opaque = reordered_opaque;
231
    newNode->pic_type = pic_type;
232

    
233
    return newNode->fake_timestamp;
234
}
235

    
236
/*
237
 * The OpaqueList is built in decode order, while elements will be removed
238
 * in presentation order. If frames are reordered, this means we must be
239
 * able to remove elements that are not the first element.
240
 *
241
 * Returned node must be freed by caller.
242
 */
243
static OpaqueList *opaque_list_pop(CHDContext *priv, uint64_t fake_timestamp)
244
{
245
    OpaqueList *node = priv->head;
246

    
247
    if (!priv->head) {
248
        av_log(priv->avctx, AV_LOG_ERROR,
249
               "CrystalHD: Attempted to query non-existent timestamps.\n");
250
        return NULL;
251
    }
252

    
253
    /*
254
     * The first element is special-cased because we have to manipulate
255
     * the head pointer rather than the previous element in the list.
256
     */
257
    if (priv->head->fake_timestamp == fake_timestamp) {
258
        priv->head = node->next;
259

    
260
        if (!priv->head->next)
261
            priv->tail = priv->head;
262

    
263
        node->next = NULL;
264
        return node;
265
    }
266

    
267
    /*
268
     * The list is processed at arm's length so that we have the
269
     * previous element available to rewrite its next pointer.
270
     */
271
    while (node->next) {
272
        OpaqueList *current = node->next;
273
        if (current->fake_timestamp == fake_timestamp) {
274
            node->next = current->next;
275

    
276
            if (!node->next)
277
               priv->tail = node;
278

    
279
            current->next = NULL;
280
            return current;
281
        } else {
282
            node = current;
283
        }
284
    }
285

    
286
    av_log(priv->avctx, AV_LOG_VERBOSE,
287
           "CrystalHD: Couldn't match fake_timestamp.\n");
288
    return NULL;
289
}
290

    
291

    
292
/*****************************************************************************
293
 * Video decoder API function definitions
294
 ****************************************************************************/
295

    
296
static void flush(AVCodecContext *avctx)
297
{
298
    CHDContext *priv = avctx->priv_data;
299

    
300
    avctx->has_b_frames     = 0;
301
    priv->last_picture      = -1;
302
    priv->output_ready      = 0;
303
    priv->need_second_field = 0;
304
    priv->skip_next_output  = 0;
305
    priv->decode_wait       = BASE_WAIT;
306

    
307
    if (priv->pic.data[0])
308
        avctx->release_buffer(avctx, &priv->pic);
309

    
310
    /* Flush mode 4 flushes all software and hardware buffers. */
311
    DtsFlushInput(priv->dev, 4);
312
}
313

    
314

    
315
static av_cold int uninit(AVCodecContext *avctx)
316
{
317
    CHDContext *priv = avctx->priv_data;
318
    HANDLE device;
319

    
320
    device = priv->dev;
321
    DtsStopDecoder(device);
322
    DtsCloseDecoder(device);
323
    DtsDeviceClose(device);
324

    
325
    av_parser_close(priv->parser);
326

    
327
    av_free(priv->sps_pps_buf);
328

    
329
    if (priv->pic.data[0])
330
        avctx->release_buffer(avctx, &priv->pic);
331

    
332
    if (priv->head) {
333
       OpaqueList *node = priv->head;
334
       while (node) {
335
          OpaqueList *next = node->next;
336
          av_free(node);
337
          node = next;
338
       }
339
    }
340

    
341
    return 0;
342
}
343

    
344

    
345
static av_cold int init(AVCodecContext *avctx)
346
{
347
    CHDContext* priv;
348
    BC_STATUS ret;
349
    BC_INFO_CRYSTAL version;
350
    BC_INPUT_FORMAT format = {
351
        .FGTEnable   = FALSE,
352
        .Progressive = TRUE,
353
        .OptFlags    = 0x80000000 | vdecFrameRate59_94 | 0x40,
354
        .width       = avctx->width,
355
        .height      = avctx->height,
356
    };
357

    
358
    BC_MEDIA_SUBTYPE subtype;
359

    
360
    uint32_t mode = DTS_PLAYBACK_MODE |
361
                    DTS_LOAD_FILE_PLAY_FW |
362
                    DTS_SKIP_TX_CHK_CPB |
363
                    DTS_PLAYBACK_DROP_RPT_MODE |
364
                    DTS_SINGLE_THREADED_MODE |
365
                    DTS_DFLT_RESOLUTION(vdecRESOLUTION_1080p23_976);
366

    
367
    av_log(avctx, AV_LOG_VERBOSE, "CrystalHD Init for %s\n",
368
           avctx->codec->name);
369

    
370
    avctx->pix_fmt = PIX_FMT_YUYV422;
371

    
372
    /* Initialize the library */
373
    priv               = avctx->priv_data;
374
    priv->avctx        = avctx;
375
    priv->is_nal       = avctx->extradata_size > 0 && *(avctx->extradata) == 1;
376
    priv->last_picture = -1;
377
    priv->decode_wait  = BASE_WAIT;
378

    
379
    subtype = id2subtype(priv, avctx->codec->id);
380
    switch (subtype) {
381
    case BC_MSUBTYPE_AVC1:
382
        {
383
            uint8_t *dummy_p;
384
            int dummy_int;
385
            AVBitStreamFilterContext *bsfc;
386

    
387
            uint32_t orig_data_size = avctx->extradata_size;
388
            uint8_t *orig_data = av_malloc(orig_data_size);
389
            if (!orig_data) {
390
                av_log(avctx, AV_LOG_ERROR,
391
                       "Failed to allocate copy of extradata\n");
392
                return AVERROR(ENOMEM);
393
            }
394
            memcpy(orig_data, avctx->extradata, orig_data_size);
395

    
396

    
397
            bsfc = av_bitstream_filter_init("h264_mp4toannexb");
398
            if (!bsfc) {
399
                av_log(avctx, AV_LOG_ERROR,
400
                       "Cannot open the h264_mp4toannexb BSF!\n");
401
                av_free(orig_data);
402
                return AVERROR_BSF_NOT_FOUND;
403
            }
404
            av_bitstream_filter_filter(bsfc, avctx, NULL, &dummy_p,
405
                                       &dummy_int, NULL, 0, 0);
406
            av_bitstream_filter_close(bsfc);
407

    
408
            priv->sps_pps_buf     = avctx->extradata;
409
            priv->sps_pps_size    = avctx->extradata_size;
410
            avctx->extradata      = orig_data;
411
            avctx->extradata_size = orig_data_size;
412

    
413
            format.pMetaData   = priv->sps_pps_buf;
414
            format.metaDataSz  = priv->sps_pps_size;
415
            format.startCodeSz = (avctx->extradata[4] & 0x03) + 1;
416
        }
417
        break;
418
    case BC_MSUBTYPE_H264:
419
        format.startCodeSz = 4;
420
        // Fall-through
421
    case BC_MSUBTYPE_VC1:
422
    case BC_MSUBTYPE_WVC1:
423
    case BC_MSUBTYPE_WMV3:
424
    case BC_MSUBTYPE_WMVA:
425
    case BC_MSUBTYPE_MPEG2VIDEO:
426
    case BC_MSUBTYPE_DIVX:
427
    case BC_MSUBTYPE_DIVX311:
428
        format.pMetaData  = avctx->extradata;
429
        format.metaDataSz = avctx->extradata_size;
430
        break;
431
    default:
432
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: Unknown codec name\n");
433
        return AVERROR(EINVAL);
434
    }
435
    format.mSubtype = subtype;
436

    
437
    /* Get a decoder instance */
438
    av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: starting up\n");
439
    // Initialize the Link and Decoder devices
440
    ret = DtsDeviceOpen(&priv->dev, mode);
441
    if (ret != BC_STS_SUCCESS) {
442
        av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: DtsDeviceOpen failed\n");
443
        goto fail;
444
    }
445

    
446
    ret = DtsCrystalHDVersion(priv->dev, &version);
447
    if (ret != BC_STS_SUCCESS) {
448
        av_log(avctx, AV_LOG_VERBOSE,
449
               "CrystalHD: DtsCrystalHDVersion failed\n");
450
        goto fail;
451
    }
452
    priv->is_70012 = version.device == 0;
453

    
454
    if (priv->is_70012 &&
455
        (subtype == BC_MSUBTYPE_DIVX || subtype == BC_MSUBTYPE_DIVX311)) {
456
        av_log(avctx, AV_LOG_VERBOSE,
457
               "CrystalHD: BCM70012 doesn't support MPEG4-ASP/DivX/Xvid\n");
458
        goto fail;
459
    }
460

    
461
    ret = DtsSetInputFormat(priv->dev, &format);
462
    if (ret != BC_STS_SUCCESS) {
463
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: SetInputFormat failed\n");
464
        goto fail;
465
    }
466

    
467
    ret = DtsOpenDecoder(priv->dev, BC_STREAM_TYPE_ES);
468
    if (ret != BC_STS_SUCCESS) {
469
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsOpenDecoder failed\n");
470
        goto fail;
471
    }
472

    
473
    ret = DtsSetColorSpace(priv->dev, OUTPUT_MODE422_YUY2);
474
    if (ret != BC_STS_SUCCESS) {
475
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsSetColorSpace failed\n");
476
        goto fail;
477
    }
478
    ret = DtsStartDecoder(priv->dev);
479
    if (ret != BC_STS_SUCCESS) {
480
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsStartDecoder failed\n");
481
        goto fail;
482
    }
483
    ret = DtsStartCapture(priv->dev);
484
    if (ret != BC_STS_SUCCESS) {
485
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsStartCapture failed\n");
486
        goto fail;
487
    }
488

    
489
    if (avctx->codec->id == CODEC_ID_H264) {
490
        priv->parser = av_parser_init(avctx->codec->id);
491
        if (!priv->parser)
492
            av_log(avctx, AV_LOG_WARNING,
493
                   "Cannot open the h.264 parser! Interlaced h.264 content "
494
                   "will not be detected reliably.\n");
495
    }
496
    av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Init complete.\n");
497

    
498
    return 0;
499

    
500
 fail:
501
    uninit(avctx);
502
    return -1;
503
}
504

    
505

    
506
static inline CopyRet copy_frame(AVCodecContext *avctx,
507
                                 BC_DTS_PROC_OUT *output,
508
                                 void *data, int *data_size)
509
{
510
    BC_STATUS ret;
511
    BC_DTS_STATUS decoder_status;
512
    uint8_t trust_interlaced;
513
    uint8_t interlaced;
514

    
515
    CHDContext *priv = avctx->priv_data;
516
    int64_t pkt_pts  = AV_NOPTS_VALUE;
517
    uint8_t pic_type = 0;
518

    
519
    uint8_t bottom_field = (output->PicInfo.flags & VDEC_FLAG_BOTTOMFIELD) ==
520
                           VDEC_FLAG_BOTTOMFIELD;
521
    uint8_t bottom_first = !!(output->PicInfo.flags & VDEC_FLAG_BOTTOM_FIRST);
522

    
523
    int width    = output->PicInfo.width;
524
    int height   = output->PicInfo.height;
525
    int bwidth;
526
    uint8_t *src = output->Ybuff;
527
    int sStride;
528
    uint8_t *dst;
529
    int dStride;
530

    
531
    if (output->PicInfo.timeStamp != 0) {
532
        OpaqueList *node = opaque_list_pop(priv, output->PicInfo.timeStamp);
533
        if (node) {
534
            pkt_pts = node->reordered_opaque;
535
            pic_type = node->pic_type;
536
            av_free(node);
537
        } else {
538
            /*
539
             * We will encounter a situation where a timestamp cannot be
540
             * popped if a second field is being returned. In this case,
541
             * each field has the same timestamp and the first one will
542
             * cause it to be popped. To keep subsequent calculations
543
             * simple, pic_type should be set a FIELD value - doesn't
544
             * matter which, but I chose BOTTOM.
545
             */
546
            pic_type = PICT_BOTTOM_FIELD;
547
        }
548
        av_log(avctx, AV_LOG_VERBOSE, "output \"pts\": %"PRIu64"\n",
549
               output->PicInfo.timeStamp);
550
        av_log(avctx, AV_LOG_VERBOSE, "output picture type %d\n",
551
               pic_type);
552
    }
553

    
554
    ret = DtsGetDriverStatus(priv->dev, &decoder_status);
555
    if (ret != BC_STS_SUCCESS) {
556
        av_log(avctx, AV_LOG_ERROR,
557
               "CrystalHD: GetDriverStatus failed: %u\n", ret);
558
       return RET_ERROR;
559
    }
560

    
561
    /*
562
     * For most content, we can trust the interlaced flag returned
563
     * by the hardware, but sometimes we can't. These are the
564
     * conditions under which we can trust the flag:
565
     *
566
     * 1) It's not h.264 content
567
     * 2) The UNKNOWN_SRC flag is not set
568
     * 3) We know we're expecting a second field
569
     * 4) The hardware reports this picture and the next picture
570
     *    have the same picture number.
571
     *
572
     * Note that there can still be interlaced content that will
573
     * fail this check, if the hardware hasn't decoded the next
574
     * picture or if there is a corruption in the stream. (In either
575
     * case a 0 will be returned for the next picture number)
576
     */
577
    trust_interlaced = avctx->codec->id != CODEC_ID_H264 ||
578
                       !(output->PicInfo.flags & VDEC_FLAG_UNKNOWN_SRC) ||
579
                       priv->need_second_field ||
580
                       (decoder_status.picNumFlags & ~0x40000000) ==
581
                       output->PicInfo.picture_number;
582

    
583
    /*
584
     * If we got a false negative for trust_interlaced on the first field,
585
     * we will realise our mistake here when we see that the picture number is that
586
     * of the previous picture. We cannot recover the frame and should discard the
587
     * second field to keep the correct number of output frames.
588
     */
589
    if (output->PicInfo.picture_number == priv->last_picture && !priv->need_second_field) {
590
        av_log(avctx, AV_LOG_WARNING,
591
               "Incorrectly guessed progressive frame. Discarding second field\n");
592
        /* Returning without providing a picture. */
593
        return RET_OK;
594
    }
595

    
596
    interlaced = (output->PicInfo.flags & VDEC_FLAG_INTERLACED_SRC) &&
597
                 trust_interlaced;
598

    
599
    if (!trust_interlaced && (decoder_status.picNumFlags & ~0x40000000) == 0) {
600
        av_log(avctx, AV_LOG_VERBOSE,
601
               "Next picture number unknown. Assuming progressive frame.\n");
602
    }
603

    
604
    av_log(avctx, AV_LOG_VERBOSE, "Interlaced state: %d | trust_interlaced %d\n",
605
           interlaced, trust_interlaced);
606

    
607
    if (priv->pic.data[0] && !priv->need_second_field)
608
        avctx->release_buffer(avctx, &priv->pic);
609

    
610
    priv->need_second_field = interlaced && !priv->need_second_field;
611

    
612
    priv->pic.buffer_hints = FF_BUFFER_HINTS_VALID | FF_BUFFER_HINTS_PRESERVE |
613
                             FF_BUFFER_HINTS_REUSABLE;
614
    if (!priv->pic.data[0]) {
615
        if (avctx->get_buffer(avctx, &priv->pic) < 0) {
616
            av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
617
            return RET_ERROR;
618
        }
619
    }
620

    
621
    bwidth = av_image_get_linesize(avctx->pix_fmt, width, 0);
622
    if (priv->is_70012) {
623
        int pStride;
624

    
625
        if (width <= 720)
626
            pStride = 720;
627
        else if (width <= 1280)
628
            pStride = 1280;
629
        else if (width <= 1080)
630
            pStride = 1080;
631
        sStride = av_image_get_linesize(avctx->pix_fmt, pStride, 0);
632
    } else {
633
        sStride = bwidth;
634
    }
635

    
636
    dStride = priv->pic.linesize[0];
637
    dst     = priv->pic.data[0];
638

    
639
    av_log(priv->avctx, AV_LOG_VERBOSE, "CrystalHD: Copying out frame\n");
640

    
641
    if (interlaced) {
642
        int dY = 0;
643
        int sY = 0;
644

    
645
        height /= 2;
646
        if (bottom_field) {
647
            av_log(priv->avctx, AV_LOG_VERBOSE, "Interlaced: bottom field\n");
648
            dY = 1;
649
        } else {
650
            av_log(priv->avctx, AV_LOG_VERBOSE, "Interlaced: top field\n");
651
            dY = 0;
652
        }
653

    
654
        for (sY = 0; sY < height; dY++, sY++) {
655
            memcpy(&(dst[dY * dStride]), &(src[sY * sStride]), bwidth);
656
            dY++;
657
        }
658
    } else {
659
        av_image_copy_plane(dst, dStride, src, sStride, bwidth, height);
660
    }
661

    
662
    priv->pic.interlaced_frame = interlaced;
663
    if (interlaced)
664
        priv->pic.top_field_first = !bottom_first;
665

    
666
    priv->pic.pkt_pts = pkt_pts;
667

    
668
    if (!priv->need_second_field) {
669
        *data_size       = sizeof(AVFrame);
670
        *(AVFrame *)data = priv->pic;
671
    }
672

    
673
    /*
674
     * Two types of PAFF content have been observed. One form causes the
675
     * hardware to return a field pair and the other individual fields,
676
     * even though the input is always individual fields. We must skip
677
     * copying on the next decode() call to maintain pipeline length in
678
     * the first case.
679
     */
680
    if (!interlaced && (output->PicInfo.flags & VDEC_FLAG_UNKNOWN_SRC) &&
681
        (pic_type == PICT_TOP_FIELD || pic_type == PICT_BOTTOM_FIELD)) {
682
        av_log(priv->avctx, AV_LOG_VERBOSE, "Fieldpair from two packets.\n");
683
        return RET_SKIP_NEXT_COPY;
684
    }
685

    
686
    /*
687
     * Testing has shown that in all cases where we don't want to return the
688
     * full frame immediately, VDEC_FLAG_UNKNOWN_SRC is set.
689
     */
690
    return priv->need_second_field &&
691
           !(output->PicInfo.flags & VDEC_FLAG_UNKNOWN_SRC) ?
692
           RET_COPY_NEXT_FIELD : RET_OK;
693
}
694

    
695

    
696
static inline CopyRet receive_frame(AVCodecContext *avctx,
697
                                    void *data, int *data_size)
698
{
699
    BC_STATUS ret;
700
    BC_DTS_PROC_OUT output = {
701
        .PicInfo.width  = avctx->width,
702
        .PicInfo.height = avctx->height,
703
    };
704
    CHDContext *priv = avctx->priv_data;
705
    HANDLE dev       = priv->dev;
706

    
707
    *data_size = 0;
708

    
709
    // Request decoded data from the driver
710
    ret = DtsProcOutputNoCopy(dev, OUTPUT_PROC_TIMEOUT, &output);
711
    if (ret == BC_STS_FMT_CHANGE) {
712
        av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Initial format change\n");
713
        avctx->width  = output.PicInfo.width;
714
        avctx->height = output.PicInfo.height;
715
        return RET_COPY_AGAIN;
716
    } else if (ret == BC_STS_SUCCESS) {
717
        int copy_ret = -1;
718
        if (output.PoutFlags & BC_POUT_FLAGS_PIB_VALID) {
719
            if (priv->last_picture == -1) {
720
                /*
721
                 * Init to one less, so that the incrementing code doesn't
722
                 * need to be special-cased.
723
                 */
724
                priv->last_picture = output.PicInfo.picture_number - 1;
725
            }
726

    
727
            if (avctx->codec->id == CODEC_ID_MPEG4 &&
728
                output.PicInfo.timeStamp == 0) {
729
                av_log(avctx, AV_LOG_VERBOSE,
730
                       "CrystalHD: Not returning packed frame twice.\n");
731
                priv->last_picture++;
732
                DtsReleaseOutputBuffs(dev, NULL, FALSE);
733
                return RET_COPY_AGAIN;
734
            }
735

    
736
            print_frame_info(priv, &output);
737

    
738
            if (priv->last_picture + 1 < output.PicInfo.picture_number) {
739
                av_log(avctx, AV_LOG_WARNING,
740
                       "CrystalHD: Picture Number discontinuity\n");
741
                /*
742
                 * Have we lost frames? If so, we need to shrink the
743
                 * pipeline length appropriately.
744
                 *
745
                 * XXX: I have no idea what the semantics of this situation
746
                 * are so I don't even know if we've lost frames or which
747
                 * ones.
748
                 *
749
                 * In any case, only warn the first time.
750
                 */
751
               priv->last_picture = output.PicInfo.picture_number - 1;
752
            }
753

    
754
            copy_ret = copy_frame(avctx, &output, data, data_size);
755
            if (*data_size > 0) {
756
                avctx->has_b_frames--;
757
                priv->last_picture++;
758
                av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Pipeline length: %u\n",
759
                       avctx->has_b_frames);
760
            }
761
        } else {
762
            /*
763
             * An invalid frame has been consumed.
764
             */
765
            av_log(avctx, AV_LOG_ERROR, "CrystalHD: ProcOutput succeeded with "
766
                                        "invalid PIB\n");
767
            avctx->has_b_frames--;
768
            copy_ret = RET_OK;
769
        }
770
        DtsReleaseOutputBuffs(dev, NULL, FALSE);
771

    
772
        return copy_ret;
773
    } else if (ret == BC_STS_BUSY) {
774
        return RET_COPY_AGAIN;
775
    } else {
776
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: ProcOutput failed %d\n", ret);
777
        return RET_ERROR;
778
    }
779
}
780

    
781

    
782
static int decode(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt)
783
{
784
    BC_STATUS ret;
785
    BC_DTS_STATUS decoder_status;
786
    CopyRet rec_ret;
787
    CHDContext *priv   = avctx->priv_data;
788
    HANDLE dev         = priv->dev;
789
    int len            = avpkt->size;
790
    uint8_t pic_type   = 0;
791

    
792
    av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: decode_frame\n");
793

    
794
    if (len) {
795
        int32_t tx_free = (int32_t)DtsTxFreeSize(dev);
796

    
797
        if (priv->parser) {
798
            uint8_t *pout;
799
            int psize = len;
800
            H264Context *h = priv->parser->priv_data;
801

    
802
            while (psize)
803
                ret = av_parser_parse2(priv->parser, avctx, &pout, &psize,
804
                                       avpkt->data, len, avctx->pkt->pts,
805
                                       avctx->pkt->dts, len - psize);
806
            av_log(avctx, AV_LOG_VERBOSE,
807
                   "CrystalHD: parser picture type %d\n",
808
                   h->s.picture_structure);
809
            pic_type = h->s.picture_structure;
810
        }
811

    
812
        if (len < tx_free - 1024) {
813
            /*
814
             * Despite being notionally opaque, either libcrystalhd or
815
             * the hardware itself will mangle pts values that are too
816
             * small or too large. The docs claim it should be in units
817
             * of 100ns. Given that we're nominally dealing with a black
818
             * box on both sides, any transform we do has no guarantee of
819
             * avoiding mangling so we need to build a mapping to values
820
             * we know will not be mangled.
821
             */
822
            uint64_t pts = opaque_list_push(priv, avctx->pkt->pts, pic_type);
823
            if (!pts) {
824
                return AVERROR(ENOMEM);
825
            }
826
            av_log(priv->avctx, AV_LOG_VERBOSE,
827
                   "input \"pts\": %"PRIu64"\n", pts);
828
            ret = DtsProcInput(dev, avpkt->data, len, pts, 0);
829
            if (ret == BC_STS_BUSY) {
830
                av_log(avctx, AV_LOG_WARNING,
831
                       "CrystalHD: ProcInput returned busy\n");
832
                usleep(BASE_WAIT);
833
                return AVERROR(EBUSY);
834
            } else if (ret != BC_STS_SUCCESS) {
835
                av_log(avctx, AV_LOG_ERROR,
836
                       "CrystalHD: ProcInput failed: %u\n", ret);
837
                return -1;
838
            }
839
            avctx->has_b_frames++;
840
        } else {
841
            av_log(avctx, AV_LOG_WARNING, "CrystalHD: Input buffer full\n");
842
            len = 0; // We didn't consume any bytes.
843
        }
844
    } else {
845
        av_log(avctx, AV_LOG_INFO, "CrystalHD: No more input data\n");
846
    }
847

    
848
    if (priv->skip_next_output) {
849
        av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Skipping next output.\n");
850
        priv->skip_next_output = 0;
851
        avctx->has_b_frames--;
852
        return len;
853
    }
854

    
855
    ret = DtsGetDriverStatus(dev, &decoder_status);
856
    if (ret != BC_STS_SUCCESS) {
857
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: GetDriverStatus failed\n");
858
        return -1;
859
    }
860

    
861
    /*
862
     * No frames ready. Don't try to extract.
863
     *
864
     * Empirical testing shows that ReadyListCount can be a damn lie,
865
     * and ProcOut still fails when count > 0. The same testing showed
866
     * that two more iterations were needed before ProcOutput would
867
     * succeed.
868
     */
869
    if (priv->output_ready < 2) {
870
        if (decoder_status.ReadyListCount != 0)
871
            priv->output_ready++;
872
        usleep(BASE_WAIT);
873
        av_log(avctx, AV_LOG_INFO, "CrystalHD: Filling pipeline.\n");
874
        return len;
875
    } else if (decoder_status.ReadyListCount == 0) {
876
        /*
877
         * After the pipeline is established, if we encounter a lack of frames
878
         * that probably means we're not giving the hardware enough time to
879
         * decode them, so start increasing the wait time at the end of a
880
         * decode call.
881
         */
882
        usleep(BASE_WAIT);
883
        priv->decode_wait += WAIT_UNIT;
884
        av_log(avctx, AV_LOG_INFO, "CrystalHD: No frames ready. Returning\n");
885
        return len;
886
    }
887

    
888
    do {
889
        rec_ret = receive_frame(avctx, data, data_size);
890
        if (rec_ret == RET_OK && *data_size == 0) {
891
            /*
892
             * This case is for when the encoded fields are stored
893
             * separately and we get a separate avpkt for each one. To keep
894
             * the pipeline stable, we should return nothing and wait for
895
             * the next time round to grab the second field.
896
             * H.264 PAFF is an example of this.
897
             */
898
            av_log(avctx, AV_LOG_VERBOSE, "Returning after first field.\n");
899
            avctx->has_b_frames--;
900
        } else if (rec_ret == RET_COPY_NEXT_FIELD) {
901
            /*
902
             * This case is for when the encoded fields are stored in a
903
             * single avpkt but the hardware returns then separately. Unless
904
             * we grab the second field before returning, we'll slip another
905
             * frame in the pipeline and if that happens a lot, we're sunk.
906
             * So we have to get that second field now.
907
             * Interlaced mpeg2 and vc1 are examples of this.
908
             */
909
            av_log(avctx, AV_LOG_VERBOSE, "Trying to get second field.\n");
910
            while (1) {
911
                usleep(priv->decode_wait);
912
                ret = DtsGetDriverStatus(dev, &decoder_status);
913
                if (ret == BC_STS_SUCCESS &&
914
                    decoder_status.ReadyListCount > 0) {
915
                    rec_ret = receive_frame(avctx, data, data_size);
916
                    if ((rec_ret == RET_OK && *data_size > 0) ||
917
                        rec_ret == RET_ERROR)
918
                        break;
919
                }
920
            }
921
            av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Got second field.\n");
922
        } else if (rec_ret == RET_SKIP_NEXT_COPY) {
923
            /*
924
             * Two input packets got turned into a field pair. Gawd.
925
             */
926
            av_log(avctx, AV_LOG_VERBOSE,
927
                   "Don't output on next decode call.\n");
928
            priv->skip_next_output = 1;
929
        }
930
        /*
931
         * If rec_ret == RET_COPY_AGAIN, that means that either we just handled
932
         * a FMT_CHANGE event and need to go around again for the actual frame,
933
         * we got a busy status and need to try again, or we're dealing with
934
         * packed b-frames, where the hardware strangely returns the packed
935
         * p-frame twice. We choose to keep the second copy as it carries the
936
         * valid pts.
937
         */
938
    } while (rec_ret == RET_COPY_AGAIN);
939
    usleep(priv->decode_wait);
940
    return len;
941
}
942

    
943

    
944
#if CONFIG_H264_CRYSTALHD_DECODER
945
AVCodec ff_h264_crystalhd_decoder = {
946
    .name           = "h264_crystalhd",
947
    .type           = AVMEDIA_TYPE_VIDEO,
948
    .id             = CODEC_ID_H264,
949
    .priv_data_size = sizeof(CHDContext),
950
    .init           = init,
951
    .close          = uninit,
952
    .decode         = decode,
953
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
954
    .flush          = flush,
955
    .long_name      = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (CrystalHD acceleration)"),
956
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
957
};
958
#endif
959

    
960
#if CONFIG_MPEG2_CRYSTALHD_DECODER
961
AVCodec ff_mpeg2_crystalhd_decoder = {
962
    .name           = "mpeg2_crystalhd",
963
    .type           = AVMEDIA_TYPE_VIDEO,
964
    .id             = CODEC_ID_MPEG2VIDEO,
965
    .priv_data_size = sizeof(CHDContext),
966
    .init           = init,
967
    .close          = uninit,
968
    .decode         = decode,
969
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
970
    .flush          = flush,
971
    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-2 Video (CrystalHD acceleration)"),
972
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
973
};
974
#endif
975

    
976
#if CONFIG_MPEG4_CRYSTALHD_DECODER
977
AVCodec ff_mpeg4_crystalhd_decoder = {
978
    .name           = "mpeg4_crystalhd",
979
    .type           = AVMEDIA_TYPE_VIDEO,
980
    .id             = CODEC_ID_MPEG4,
981
    .priv_data_size = sizeof(CHDContext),
982
    .init           = init,
983
    .close          = uninit,
984
    .decode         = decode,
985
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
986
    .flush          = flush,
987
    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 Part 2 (CrystalHD acceleration)"),
988
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
989
};
990
#endif
991

    
992
#if CONFIG_MSMPEG4_CRYSTALHD_DECODER
993
AVCodec ff_msmpeg4_crystalhd_decoder = {
994
    .name           = "msmpeg4_crystalhd",
995
    .type           = AVMEDIA_TYPE_VIDEO,
996
    .id             = CODEC_ID_MSMPEG4V3,
997
    .priv_data_size = sizeof(CHDContext),
998
    .init           = init,
999
    .close          = uninit,
1000
    .decode         = decode,
1001
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
1002
    .flush          = flush,
1003
    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 Part 2 Microsoft variant version 3 (CrystalHD acceleration)"),
1004
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
1005
};
1006
#endif
1007

    
1008
#if CONFIG_VC1_CRYSTALHD_DECODER
1009
AVCodec ff_vc1_crystalhd_decoder = {
1010
    .name           = "vc1_crystalhd",
1011
    .type           = AVMEDIA_TYPE_VIDEO,
1012
    .id             = CODEC_ID_VC1,
1013
    .priv_data_size = sizeof(CHDContext),
1014
    .init           = init,
1015
    .close          = uninit,
1016
    .decode         = decode,
1017
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
1018
    .flush          = flush,
1019
    .long_name      = NULL_IF_CONFIG_SMALL("SMPTE VC-1 (CrystalHD acceleration)"),
1020
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
1021
};
1022
#endif
1023

    
1024
#if CONFIG_WMV3_CRYSTALHD_DECODER
1025
AVCodec ff_wmv3_crystalhd_decoder = {
1026
    .name           = "wmv3_crystalhd",
1027
    .type           = AVMEDIA_TYPE_VIDEO,
1028
    .id             = CODEC_ID_WMV3,
1029
    .priv_data_size = sizeof(CHDContext),
1030
    .init           = init,
1031
    .close          = uninit,
1032
    .decode         = decode,
1033
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
1034
    .flush          = flush,
1035
    .long_name      = NULL_IF_CONFIG_SMALL("Windows Media Video 9 (CrystalHD acceleration)"),
1036
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
1037
};
1038
#endif