Statistics
| Branch: | Revision:

ffmpeg / libavcodec / crystalhd.c @ ca0eed7e

History | View | Annotate | Download (33.8 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 "libavutil/imgutils.h"
88
#include "libavutil/intreadwrite.h"
89

    
90
/** Timeout parameter passed to DtsProcOutput() in us */
91
#define OUTPUT_PROC_TIMEOUT 50
92
/** Step between fake timestamps passed to hardware in units of 100ns */
93
#define TIMESTAMP_UNIT 100000
94
/** Initial value in us of the wait in decode() */
95
#define BASE_WAIT 10000
96
/** Increment in us to adjust wait in decode() */
97
#define WAIT_UNIT 1000
98

    
99

    
100
/*****************************************************************************
101
 * Module private data
102
 ****************************************************************************/
103

    
104
typedef enum {
105
    RET_ERROR           = -1,
106
    RET_OK              = 0,
107
    RET_COPY_AGAIN      = 1,
108
    RET_SKIP_NEXT_COPY  = 2,
109
    RET_COPY_NEXT_FIELD = 3,
110
} CopyRet;
111

    
112
typedef struct OpaqueList {
113
    struct OpaqueList *next;
114
    uint64_t fake_timestamp;
115
    uint64_t reordered_opaque;
116
} OpaqueList;
117

    
118
typedef struct {
119
    AVCodecContext *avctx;
120
    AVFrame pic;
121
    HANDLE dev;
122

    
123
    uint8_t is_70012;
124
    uint8_t *sps_pps_buf;
125
    uint32_t sps_pps_size;
126
    uint8_t is_nal;
127
    uint8_t output_ready;
128
    uint8_t need_second_field;
129
    uint8_t skip_next_output;
130
    uint64_t decode_wait;
131

    
132
    uint64_t last_picture;
133

    
134
    OpaqueList *head;
135
    OpaqueList *tail;
136
} CHDContext;
137

    
138

    
139
/*****************************************************************************
140
 * Helper functions
141
 ****************************************************************************/
142

    
143
static inline BC_MEDIA_SUBTYPE id2subtype(CHDContext *priv, enum CodecID id)
144
{
145
    switch (id) {
146
    case CODEC_ID_MPEG4:
147
        return BC_MSUBTYPE_DIVX;
148
    case CODEC_ID_MSMPEG4V3:
149
        return BC_MSUBTYPE_DIVX311;
150
    case CODEC_ID_MPEG2VIDEO:
151
        return BC_MSUBTYPE_MPEG2VIDEO;
152
    case CODEC_ID_VC1:
153
        return BC_MSUBTYPE_VC1;
154
    case CODEC_ID_WMV3:
155
        return BC_MSUBTYPE_WMV3;
156
    case CODEC_ID_H264:
157
        return priv->is_nal ? BC_MSUBTYPE_AVC1 : BC_MSUBTYPE_H264;
158
    default:
159
        return BC_MSUBTYPE_INVALID;
160
    }
161
}
162

    
163
static inline void print_frame_info(CHDContext *priv, BC_DTS_PROC_OUT *output)
164
{
165
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tYBuffSz: %u\n", output->YbuffSz);
166
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tYBuffDoneSz: %u\n",
167
           output->YBuffDoneSz);
168
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tUVBuffDoneSz: %u\n",
169
           output->UVBuffDoneSz);
170
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tTimestamp: %"PRIu64"\n",
171
           output->PicInfo.timeStamp);
172
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tPicture Number: %u\n",
173
           output->PicInfo.picture_number);
174
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tWidth: %u\n",
175
           output->PicInfo.width);
176
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tHeight: %u\n",
177
           output->PicInfo.height);
178
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tChroma: 0x%03x\n",
179
           output->PicInfo.chroma_format);
180
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tPulldown: %u\n",
181
           output->PicInfo.pulldown);
182
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tFlags: 0x%08x\n",
183
           output->PicInfo.flags);
184
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tFrame Rate/Res: %u\n",
185
           output->PicInfo.frame_rate);
186
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tAspect Ratio: %u\n",
187
           output->PicInfo.aspect_ratio);
188
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tColor Primaries: %u\n",
189
           output->PicInfo.colour_primaries);
190
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tMetaData: %u\n",
191
           output->PicInfo.picture_meta_payload);
192
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tSession Number: %u\n",
193
           output->PicInfo.sess_num);
194
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tycom: %u\n",
195
           output->PicInfo.ycom);
196
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tCustom Aspect: %u\n",
197
           output->PicInfo.custom_aspect_ratio_width_height);
198
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tFrames to Drop: %u\n",
199
           output->PicInfo.n_drop);
200
    av_log(priv->avctx, AV_LOG_VERBOSE, "\tH264 Valid Fields: 0x%08x\n",
201
           output->PicInfo.other.h264.valid);
202
}
203

    
204

    
205
/*****************************************************************************
206
 * OpaqueList functions
207
 ****************************************************************************/
208

    
209
static uint64_t opaque_list_push(CHDContext *priv, uint64_t reordered_opaque)
210
{
211
    OpaqueList *newNode = av_mallocz(sizeof (OpaqueList));
212
    if (!newNode) {
213
        av_log(priv->avctx, AV_LOG_ERROR,
214
               "Unable to allocate new node in OpaqueList.\n");
215
        return 0;
216
    }
217
    if (!priv->head) {
218
        newNode->fake_timestamp = TIMESTAMP_UNIT;
219
        priv->head              = newNode;
220
    } else {
221
        newNode->fake_timestamp = priv->tail->fake_timestamp + TIMESTAMP_UNIT;
222
        priv->tail->next        = newNode;
223
    }
224
    priv->tail = newNode;
225
    newNode->reordered_opaque = reordered_opaque;
226

    
227
    return newNode->fake_timestamp;
228
}
229

    
230
/*
231
 * The OpaqueList is built in decode order, while elements will be removed
232
 * in presentation order. If frames are reordered, this means we must be
233
 * able to remove elements that are not the first element.
234
 */
235
static uint64_t opaque_list_pop(CHDContext *priv, uint64_t fake_timestamp)
236
{
237
    OpaqueList *node = priv->head;
238

    
239
    if (!priv->head) {
240
        av_log(priv->avctx, AV_LOG_ERROR,
241
               "CrystalHD: Attempted to query non-existent timestamps.\n");
242
        return AV_NOPTS_VALUE;
243
    }
244

    
245
    /*
246
     * The first element is special-cased because we have to manipulate
247
     * the head pointer rather than the previous element in the list.
248
     */
249
    if (priv->head->fake_timestamp == fake_timestamp) {
250
        uint64_t reordered_opaque = node->reordered_opaque;
251
        priv->head = node->next;
252
        av_free(node);
253

    
254
        if (!priv->head->next)
255
            priv->tail = priv->head;
256

    
257
        return reordered_opaque;
258
    }
259

    
260
    /*
261
     * The list is processed at arm's length so that we have the
262
     * previous element available to rewrite its next pointer.
263
     */
264
    while (node->next) {
265
        OpaqueList *next = node->next;
266
        if (next->fake_timestamp == fake_timestamp) {
267
            uint64_t reordered_opaque = next->reordered_opaque;
268
            node->next = next->next;
269
            av_free(next);
270

    
271
            if (!node->next)
272
               priv->tail = node;
273

    
274
            return reordered_opaque;
275
        } else {
276
            node = next;
277
        }
278
    }
279

    
280
    av_log(priv->avctx, AV_LOG_VERBOSE,
281
           "CrystalHD: Couldn't match fake_timestamp.\n");
282
    return AV_NOPTS_VALUE;
283
}
284

    
285

    
286
/*****************************************************************************
287
 * Video decoder API function definitions
288
 ****************************************************************************/
289

    
290
static void flush(AVCodecContext *avctx)
291
{
292
    CHDContext *priv = avctx->priv_data;
293

    
294
    avctx->has_b_frames     = 0;
295
    priv->last_picture      = -1;
296
    priv->output_ready      = 0;
297
    priv->need_second_field = 0;
298
    priv->skip_next_output  = 0;
299
    priv->decode_wait       = BASE_WAIT;
300

    
301
    if (priv->pic.data[0])
302
        avctx->release_buffer(avctx, &priv->pic);
303

    
304
    /* Flush mode 4 flushes all software and hardware buffers. */
305
    DtsFlushInput(priv->dev, 4);
306
}
307

    
308

    
309
static av_cold int uninit(AVCodecContext *avctx)
310
{
311
    CHDContext *priv = avctx->priv_data;
312
    HANDLE device;
313

    
314
    device = priv->dev;
315
    DtsStopDecoder(device);
316
    DtsCloseDecoder(device);
317
    DtsDeviceClose(device);
318

    
319
    av_free(priv->sps_pps_buf);
320

    
321
    if (priv->pic.data[0])
322
        avctx->release_buffer(avctx, &priv->pic);
323

    
324
    if (priv->head) {
325
       OpaqueList *node = priv->head;
326
       while (node) {
327
          OpaqueList *next = node->next;
328
          av_free(node);
329
          node = next;
330
       }
331
    }
332

    
333
    return 0;
334
}
335

    
336

    
337
static av_cold int init(AVCodecContext *avctx)
338
{
339
    CHDContext* priv;
340
    BC_STATUS ret;
341
    BC_INFO_CRYSTAL version;
342
    BC_INPUT_FORMAT format = {
343
        .FGTEnable   = FALSE,
344
        .Progressive = TRUE,
345
        .OptFlags    = 0x80000000 | vdecFrameRate59_94 | 0x40,
346
        .width       = avctx->width,
347
        .height      = avctx->height,
348
    };
349

    
350
    BC_MEDIA_SUBTYPE subtype;
351

    
352
    uint32_t mode = DTS_PLAYBACK_MODE |
353
                    DTS_LOAD_FILE_PLAY_FW |
354
                    DTS_SKIP_TX_CHK_CPB |
355
                    DTS_PLAYBACK_DROP_RPT_MODE |
356
                    DTS_SINGLE_THREADED_MODE |
357
                    DTS_DFLT_RESOLUTION(vdecRESOLUTION_1080p23_976);
358

    
359
    av_log(avctx, AV_LOG_VERBOSE, "CrystalHD Init for %s\n",
360
           avctx->codec->name);
361

    
362
    avctx->pix_fmt = PIX_FMT_YUYV422;
363

    
364
    /* Initialize the library */
365
    priv               = avctx->priv_data;
366
    priv->avctx        = avctx;
367
    priv->is_nal       = avctx->extradata_size > 0 && *(avctx->extradata) == 1;
368
    priv->last_picture = -1;
369
    priv->decode_wait  = BASE_WAIT;
370

    
371
    subtype = id2subtype(priv, avctx->codec->id);
372
    switch (subtype) {
373
    case BC_MSUBTYPE_AVC1:
374
        {
375
            uint8_t *dummy_p;
376
            int dummy_int;
377
            AVBitStreamFilterContext *bsfc;
378

    
379
            uint32_t orig_data_size = avctx->extradata_size;
380
            uint8_t *orig_data = av_malloc(orig_data_size);
381
            if (!orig_data) {
382
                av_log(avctx, AV_LOG_ERROR,
383
                       "Failed to allocate copy of extradata\n");
384
                return AVERROR(ENOMEM);
385
            }
386
            memcpy(orig_data, avctx->extradata, orig_data_size);
387

    
388

    
389
            bsfc = av_bitstream_filter_init("h264_mp4toannexb");
390
            if (!bsfc) {
391
                av_log(avctx, AV_LOG_ERROR,
392
                       "Cannot open the h264_mp4toannexb BSF!\n");
393
                av_free(orig_data);
394
                return AVERROR_BSF_NOT_FOUND;
395
            }
396
            av_bitstream_filter_filter(bsfc, avctx, NULL, &dummy_p,
397
                                       &dummy_int, NULL, 0, 0);
398
            av_bitstream_filter_close(bsfc);
399

    
400
            priv->sps_pps_buf     = avctx->extradata;
401
            priv->sps_pps_size    = avctx->extradata_size;
402
            avctx->extradata      = orig_data;
403
            avctx->extradata_size = orig_data_size;
404

    
405
            format.pMetaData   = priv->sps_pps_buf;
406
            format.metaDataSz  = priv->sps_pps_size;
407
            format.startCodeSz = (avctx->extradata[4] & 0x03) + 1;
408
        }
409
        break;
410
    case BC_MSUBTYPE_H264:
411
        format.startCodeSz = 4;
412
        // Fall-through
413
    case BC_MSUBTYPE_VC1:
414
    case BC_MSUBTYPE_WVC1:
415
    case BC_MSUBTYPE_WMV3:
416
    case BC_MSUBTYPE_WMVA:
417
    case BC_MSUBTYPE_MPEG2VIDEO:
418
    case BC_MSUBTYPE_DIVX:
419
    case BC_MSUBTYPE_DIVX311:
420
        format.pMetaData  = avctx->extradata;
421
        format.metaDataSz = avctx->extradata_size;
422
        break;
423
    default:
424
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: Unknown codec name\n");
425
        return AVERROR(EINVAL);
426
    }
427
    format.mSubtype = subtype;
428

    
429
    /* Get a decoder instance */
430
    av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: starting up\n");
431
    // Initialize the Link and Decoder devices
432
    ret = DtsDeviceOpen(&priv->dev, mode);
433
    if (ret != BC_STS_SUCCESS) {
434
        av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: DtsDeviceOpen failed\n");
435
        goto fail;
436
    }
437

    
438
    ret = DtsCrystalHDVersion(priv->dev, &version);
439
    if (ret != BC_STS_SUCCESS) {
440
        av_log(avctx, AV_LOG_VERBOSE,
441
               "CrystalHD: DtsCrystalHDVersion failed\n");
442
        goto fail;
443
    }
444
    priv->is_70012 = version.device == 0;
445

    
446
    if (priv->is_70012 &&
447
        (subtype == BC_MSUBTYPE_DIVX || subtype == BC_MSUBTYPE_DIVX311)) {
448
        av_log(avctx, AV_LOG_VERBOSE,
449
               "CrystalHD: BCM70012 doesn't support MPEG4-ASP/DivX/Xvid\n");
450
        goto fail;
451
    }
452

    
453
    ret = DtsSetInputFormat(priv->dev, &format);
454
    if (ret != BC_STS_SUCCESS) {
455
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: SetInputFormat failed\n");
456
        goto fail;
457
    }
458

    
459
    ret = DtsOpenDecoder(priv->dev, BC_STREAM_TYPE_ES);
460
    if (ret != BC_STS_SUCCESS) {
461
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsOpenDecoder failed\n");
462
        goto fail;
463
    }
464

    
465
    ret = DtsSetColorSpace(priv->dev, OUTPUT_MODE422_YUY2);
466
    if (ret != BC_STS_SUCCESS) {
467
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsSetColorSpace failed\n");
468
        goto fail;
469
    }
470
    ret = DtsStartDecoder(priv->dev);
471
    if (ret != BC_STS_SUCCESS) {
472
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsStartDecoder failed\n");
473
        goto fail;
474
    }
475
    ret = DtsStartCapture(priv->dev);
476
    if (ret != BC_STS_SUCCESS) {
477
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsStartCapture failed\n");
478
        goto fail;
479
    }
480

    
481
    av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Init complete.\n");
482

    
483
    return 0;
484

    
485
 fail:
486
    uninit(avctx);
487
    return -1;
488
}
489

    
490

    
491
/*
492
 * The CrystalHD doesn't report interlaced H.264 content in a way that allows
493
 * us to distinguish between specific cases that require different handling.
494
 * So, for now, we have to hard-code the behaviour we want.
495
 *
496
 * The default behaviour is to assume MBAFF with input and output fieldpairs.
497
 *
498
 * Define ASSUME_PAFF_OVER_MBAFF to treat input as PAFF with separate input
499
 * and output fields.
500
 *
501
 * Define ASSUME_TWO_INPUTS_ONE_OUTPUT to treat input as separate fields but
502
 * output as a single fieldpair.
503
 *
504
 * Define both to mess up your playback.
505
 */
506
#define ASSUME_PAFF_OVER_MBAFF 0
507
#define ASSUME_TWO_INPUTS_ONE_OUTPUT 0
508
static inline CopyRet copy_frame(AVCodecContext *avctx,
509
                                 BC_DTS_PROC_OUT *output,
510
                                 void *data, int *data_size,
511
                                 uint8_t second_field)
512
{
513
    BC_STATUS ret;
514
    BC_DTS_STATUS decoder_status;
515
    uint8_t is_paff;
516
    uint8_t next_frame_same;
517
    uint8_t interlaced;
518

    
519
    CHDContext *priv = avctx->priv_data;
520

    
521
    uint8_t bottom_field = (output->PicInfo.flags & VDEC_FLAG_BOTTOMFIELD) ==
522
                           VDEC_FLAG_BOTTOMFIELD;
523
    uint8_t bottom_first = !!(output->PicInfo.flags & VDEC_FLAG_BOTTOM_FIRST);
524

    
525
    int width    = output->PicInfo.width;
526
    int height   = output->PicInfo.height;
527
    int bwidth;
528
    uint8_t *src = output->Ybuff;
529
    int sStride;
530
    uint8_t *dst;
531
    int dStride;
532

    
533
    ret = DtsGetDriverStatus(priv->dev, &decoder_status);
534
    if (ret != BC_STS_SUCCESS) {
535
        av_log(avctx, AV_LOG_ERROR,
536
               "CrystalHD: GetDriverStatus failed: %u\n", ret);
537
       return RET_ERROR;
538
    }
539

    
540
    is_paff           = ASSUME_PAFF_OVER_MBAFF ||
541
                        !(output->PicInfo.flags & VDEC_FLAG_UNKNOWN_SRC);
542
    next_frame_same   = output->PicInfo.picture_number ==
543
                        (decoder_status.picNumFlags & ~0x40000000);
544
    interlaced        = ((output->PicInfo.flags &
545
                          VDEC_FLAG_INTERLACED_SRC) && is_paff) ||
546
                         next_frame_same || bottom_field || second_field;
547

    
548
    av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: next_frame_same: %u | %u | %u\n",
549
           next_frame_same, output->PicInfo.picture_number,
550
           decoder_status.picNumFlags & ~0x40000000);
551

    
552
    if (priv->pic.data[0] && !priv->need_second_field)
553
        avctx->release_buffer(avctx, &priv->pic);
554

    
555
    priv->need_second_field = interlaced && !priv->need_second_field;
556

    
557
    priv->pic.buffer_hints = FF_BUFFER_HINTS_VALID | FF_BUFFER_HINTS_PRESERVE |
558
                             FF_BUFFER_HINTS_REUSABLE;
559
    if (!priv->pic.data[0]) {
560
        if (avctx->get_buffer(avctx, &priv->pic) < 0) {
561
            av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
562
            return RET_ERROR;
563
        }
564
    }
565

    
566
    bwidth = av_image_get_linesize(avctx->pix_fmt, width, 0);
567
    if (priv->is_70012) {
568
        int pStride;
569

    
570
        if (width <= 720)
571
            pStride = 720;
572
        else if (width <= 1280)
573
            pStride = 1280;
574
        else if (width <= 1080)
575
            pStride = 1080;
576
        sStride = av_image_get_linesize(avctx->pix_fmt, pStride, 0);
577
    } else {
578
        sStride = bwidth;
579
    }
580

    
581
    dStride = priv->pic.linesize[0];
582
    dst     = priv->pic.data[0];
583

    
584
    av_log(priv->avctx, AV_LOG_VERBOSE, "CrystalHD: Copying out frame\n");
585

    
586
    if (interlaced) {
587
        int dY = 0;
588
        int sY = 0;
589

    
590
        height /= 2;
591
        if (bottom_field) {
592
            av_log(priv->avctx, AV_LOG_VERBOSE, "Interlaced: bottom field\n");
593
            dY = 1;
594
        } else {
595
            av_log(priv->avctx, AV_LOG_VERBOSE, "Interlaced: top field\n");
596
            dY = 0;
597
        }
598

    
599
        for (sY = 0; sY < height; dY++, sY++) {
600
            memcpy(&(dst[dY * dStride]), &(src[sY * sStride]), bwidth);
601
            dY++;
602
        }
603
    } else {
604
        av_image_copy_plane(dst, dStride, src, sStride, bwidth, height);
605
    }
606

    
607
    priv->pic.interlaced_frame = interlaced;
608
    if (interlaced)
609
        priv->pic.top_field_first = !bottom_first;
610

    
611
    if (output->PicInfo.timeStamp != 0) {
612
        priv->pic.pkt_pts = opaque_list_pop(priv, output->PicInfo.timeStamp);
613
        av_log(avctx, AV_LOG_VERBOSE, "output \"pts\": %"PRIu64"\n",
614
               priv->pic.pkt_pts);
615
    }
616

    
617
    if (!priv->need_second_field) {
618
        *data_size       = sizeof(AVFrame);
619
        *(AVFrame *)data = priv->pic;
620
    }
621

    
622
    if (ASSUME_TWO_INPUTS_ONE_OUTPUT &&
623
        output->PicInfo.flags & VDEC_FLAG_UNKNOWN_SRC) {
624
        av_log(priv->avctx, AV_LOG_VERBOSE, "Fieldpair from two packets.\n");
625
        return RET_SKIP_NEXT_COPY;
626
    }
627

    
628
    /*
629
     * Testing has shown that in all cases where we don't want to return the
630
     * full frame immediately, VDEC_FLAG_UNKNOWN_SRC is set.
631
     */
632
    return priv->need_second_field &&
633
           !(output->PicInfo.flags & VDEC_FLAG_UNKNOWN_SRC) ?
634
           RET_COPY_NEXT_FIELD : RET_OK;
635
}
636

    
637

    
638
static inline CopyRet receive_frame(AVCodecContext *avctx,
639
                                    void *data, int *data_size,
640
                                    uint8_t second_field)
641
{
642
    BC_STATUS ret;
643
    BC_DTS_PROC_OUT output = {
644
        .PicInfo.width  = avctx->width,
645
        .PicInfo.height = avctx->height,
646
    };
647
    CHDContext *priv = avctx->priv_data;
648
    HANDLE dev       = priv->dev;
649

    
650
    *data_size = 0;
651

    
652
    // Request decoded data from the driver
653
    ret = DtsProcOutputNoCopy(dev, OUTPUT_PROC_TIMEOUT, &output);
654
    if (ret == BC_STS_FMT_CHANGE) {
655
        av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Initial format change\n");
656
        avctx->width  = output.PicInfo.width;
657
        avctx->height = output.PicInfo.height;
658
        return RET_COPY_AGAIN;
659
    } else if (ret == BC_STS_SUCCESS) {
660
        int copy_ret = -1;
661
        if (output.PoutFlags & BC_POUT_FLAGS_PIB_VALID) {
662
            if (priv->last_picture == -1) {
663
                /*
664
                 * Init to one less, so that the incrementing code doesn't
665
                 * need to be special-cased.
666
                 */
667
                priv->last_picture = output.PicInfo.picture_number - 1;
668
            }
669

    
670
            if (avctx->codec->id == CODEC_ID_MPEG4 &&
671
                output.PicInfo.timeStamp == 0) {
672
                av_log(avctx, AV_LOG_VERBOSE,
673
                       "CrystalHD: Not returning packed frame twice.\n");
674
                priv->last_picture++;
675
                DtsReleaseOutputBuffs(dev, NULL, FALSE);
676
                return RET_COPY_AGAIN;
677
            }
678

    
679
            print_frame_info(priv, &output);
680

    
681
            if (priv->last_picture + 1 < output.PicInfo.picture_number) {
682
                av_log(avctx, AV_LOG_WARNING,
683
                       "CrystalHD: Picture Number discontinuity\n");
684
                /*
685
                 * Have we lost frames? If so, we need to shrink the
686
                 * pipeline length appropriately.
687
                 *
688
                 * XXX: I have no idea what the semantics of this situation
689
                 * are so I don't even know if we've lost frames or which
690
                 * ones.
691
                 *
692
                 * In any case, only warn the first time.
693
                 */
694
               priv->last_picture = output.PicInfo.picture_number - 1;
695
            }
696

    
697
            copy_ret = copy_frame(avctx, &output, data, data_size, second_field);
698
            if (*data_size > 0) {
699
                avctx->has_b_frames--;
700
                priv->last_picture++;
701
                av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Pipeline length: %u\n",
702
                       avctx->has_b_frames);
703
            }
704
        } else {
705
            /*
706
             * An invalid frame has been consumed.
707
             */
708
            av_log(avctx, AV_LOG_ERROR, "CrystalHD: ProcOutput succeeded with "
709
                                        "invalid PIB\n");
710
            avctx->has_b_frames--;
711
            copy_ret = RET_OK;
712
        }
713
        DtsReleaseOutputBuffs(dev, NULL, FALSE);
714

    
715
        return copy_ret;
716
    } else if (ret == BC_STS_BUSY) {
717
        return RET_COPY_AGAIN;
718
    } else {
719
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: ProcOutput failed %d\n", ret);
720
        return RET_ERROR;
721
    }
722
}
723

    
724

    
725
static int decode(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt)
726
{
727
    BC_STATUS ret;
728
    BC_DTS_STATUS decoder_status;
729
    CopyRet rec_ret;
730
    CHDContext *priv   = avctx->priv_data;
731
    HANDLE dev         = priv->dev;
732
    int len            = avpkt->size;
733

    
734
    av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: decode_frame\n");
735

    
736
    if (len) {
737
        int32_t tx_free = (int32_t)DtsTxFreeSize(dev);
738
        if (len < tx_free - 1024) {
739
            /*
740
             * Despite being notionally opaque, either libcrystalhd or
741
             * the hardware itself will mangle pts values that are too
742
             * small or too large. The docs claim it should be in units
743
             * of 100ns. Given that we're nominally dealing with a black
744
             * box on both sides, any transform we do has no guarantee of
745
             * avoiding mangling so we need to build a mapping to values
746
             * we know will not be mangled.
747
             */
748
            uint64_t pts = opaque_list_push(priv, avctx->pkt->pts);
749
            if (!pts) {
750
                return AVERROR(ENOMEM);
751
            }
752
            av_log(priv->avctx, AV_LOG_VERBOSE,
753
                   "input \"pts\": %"PRIu64"\n", pts);
754
            ret = DtsProcInput(dev, avpkt->data, len, pts, 0);
755
            if (ret == BC_STS_BUSY) {
756
                av_log(avctx, AV_LOG_WARNING,
757
                       "CrystalHD: ProcInput returned busy\n");
758
                usleep(BASE_WAIT);
759
                return AVERROR(EBUSY);
760
            } else if (ret != BC_STS_SUCCESS) {
761
                av_log(avctx, AV_LOG_ERROR,
762
                       "CrystalHD: ProcInput failed: %u\n", ret);
763
                return -1;
764
            }
765
            avctx->has_b_frames++;
766
        } else {
767
            av_log(avctx, AV_LOG_WARNING, "CrystalHD: Input buffer full\n");
768
            len = 0; // We didn't consume any bytes.
769
        }
770
    } else {
771
        av_log(avctx, AV_LOG_INFO, "CrystalHD: No more input data\n");
772
    }
773

    
774
    if (priv->skip_next_output) {
775
        av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Skipping next output.\n");
776
        priv->skip_next_output = 0;
777
        avctx->has_b_frames--;
778
        return len;
779
    }
780

    
781
    ret = DtsGetDriverStatus(dev, &decoder_status);
782
    if (ret != BC_STS_SUCCESS) {
783
        av_log(avctx, AV_LOG_ERROR, "CrystalHD: GetDriverStatus failed\n");
784
        return -1;
785
    }
786

    
787
    /*
788
     * No frames ready. Don't try to extract.
789
     *
790
     * Empirical testing shows that ReadyListCount can be a damn lie,
791
     * and ProcOut still fails when count > 0. The same testing showed
792
     * that two more iterations were needed before ProcOutput would
793
     * succeed.
794
     */
795
    if (priv->output_ready < 2) {
796
        if (decoder_status.ReadyListCount != 0)
797
            priv->output_ready++;
798
        usleep(BASE_WAIT);
799
        av_log(avctx, AV_LOG_INFO, "CrystalHD: Filling pipeline.\n");
800
        return len;
801
    } else if (decoder_status.ReadyListCount == 0) {
802
        /*
803
         * After the pipeline is established, if we encounter a lack of frames
804
         * that probably means we're not giving the hardware enough time to
805
         * decode them, so start increasing the wait time at the end of a
806
         * decode call.
807
         */
808
        usleep(BASE_WAIT);
809
        priv->decode_wait += WAIT_UNIT;
810
        av_log(avctx, AV_LOG_INFO, "CrystalHD: No frames ready. Returning\n");
811
        return len;
812
    }
813

    
814
    do {
815
        rec_ret = receive_frame(avctx, data, data_size, 0);
816
        if (rec_ret == RET_OK && *data_size == 0) {
817
            /*
818
             * This case is for when the encoded fields are stored
819
             * separately and we get a separate avpkt for each one. To keep
820
             * the pipeline stable, we should return nothing and wait for
821
             * the next time round to grab the second field.
822
             * H.264 PAFF is an example of this.
823
             */
824
            av_log(avctx, AV_LOG_VERBOSE, "Returning after first field.\n");
825
            avctx->has_b_frames--;
826
        } else if (rec_ret == RET_COPY_NEXT_FIELD) {
827
            /*
828
             * This case is for when the encoded fields are stored in a
829
             * single avpkt but the hardware returns then separately. Unless
830
             * we grab the second field before returning, we'll slip another
831
             * frame in the pipeline and if that happens a lot, we're sunk.
832
             * So we have to get that second field now.
833
             * Interlaced mpeg2 and vc1 are examples of this.
834
             */
835
            av_log(avctx, AV_LOG_VERBOSE, "Trying to get second field.\n");
836
            while (1) {
837
                usleep(priv->decode_wait);
838
                ret = DtsGetDriverStatus(dev, &decoder_status);
839
                if (ret == BC_STS_SUCCESS &&
840
                    decoder_status.ReadyListCount > 0) {
841
                    rec_ret = receive_frame(avctx, data, data_size, 1);
842
                    if ((rec_ret == RET_OK && *data_size > 0) ||
843
                        rec_ret == RET_ERROR)
844
                        break;
845
                }
846
            }
847
            av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Got second field.\n");
848
        } else if (rec_ret == RET_SKIP_NEXT_COPY) {
849
            /*
850
             * Two input packets got turned into a field pair. Gawd.
851
             */
852
            av_log(avctx, AV_LOG_VERBOSE,
853
                   "Don't output on next decode call.\n");
854
            priv->skip_next_output = 1;
855
        }
856
        /*
857
         * If rec_ret == RET_COPY_AGAIN, that means that either we just handled
858
         * a FMT_CHANGE event and need to go around again for the actual frame,
859
         * we got a busy status and need to try again, or we're dealing with
860
         * packed b-frames, where the hardware strangely returns the packed
861
         * p-frame twice. We choose to keep the second copy as it carries the
862
         * valid pts.
863
         */
864
    } while (rec_ret == RET_COPY_AGAIN);
865
    usleep(priv->decode_wait);
866
    return len;
867
}
868

    
869

    
870
#if CONFIG_H264_CRYSTALHD_DECODER
871
AVCodec ff_h264_crystalhd_decoder = {
872
    .name           = "h264_crystalhd",
873
    .type           = AVMEDIA_TYPE_VIDEO,
874
    .id             = CODEC_ID_H264,
875
    .priv_data_size = sizeof(CHDContext),
876
    .init           = init,
877
    .close          = uninit,
878
    .decode         = decode,
879
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
880
    .flush          = flush,
881
    .long_name      = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (CrystalHD acceleration)"),
882
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
883
};
884
#endif
885

    
886
#if CONFIG_MPEG2_CRYSTALHD_DECODER
887
AVCodec ff_mpeg2_crystalhd_decoder = {
888
    .name           = "mpeg2_crystalhd",
889
    .type           = AVMEDIA_TYPE_VIDEO,
890
    .id             = CODEC_ID_MPEG2VIDEO,
891
    .priv_data_size = sizeof(CHDContext),
892
    .init           = init,
893
    .close          = uninit,
894
    .decode         = decode,
895
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
896
    .flush          = flush,
897
    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-2 Video (CrystalHD acceleration)"),
898
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
899
};
900
#endif
901

    
902
#if CONFIG_MPEG4_CRYSTALHD_DECODER
903
AVCodec ff_mpeg4_crystalhd_decoder = {
904
    .name           = "mpeg4_crystalhd",
905
    .type           = AVMEDIA_TYPE_VIDEO,
906
    .id             = CODEC_ID_MPEG4,
907
    .priv_data_size = sizeof(CHDContext),
908
    .init           = init,
909
    .close          = uninit,
910
    .decode         = decode,
911
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
912
    .flush          = flush,
913
    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 Part 2 (CrystalHD acceleration)"),
914
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
915
};
916
#endif
917

    
918
#if CONFIG_MSMPEG4_CRYSTALHD_DECODER
919
AVCodec ff_msmpeg4_crystalhd_decoder = {
920
    .name           = "msmpeg4_crystalhd",
921
    .type           = AVMEDIA_TYPE_VIDEO,
922
    .id             = CODEC_ID_MSMPEG4V3,
923
    .priv_data_size = sizeof(CHDContext),
924
    .init           = init,
925
    .close          = uninit,
926
    .decode         = decode,
927
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
928
    .flush          = flush,
929
    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 Part 2 Microsoft variant version 3 (CrystalHD acceleration)"),
930
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
931
};
932
#endif
933

    
934
#if CONFIG_VC1_CRYSTALHD_DECODER
935
AVCodec ff_vc1_crystalhd_decoder = {
936
    .name           = "vc1_crystalhd",
937
    .type           = AVMEDIA_TYPE_VIDEO,
938
    .id             = CODEC_ID_VC1,
939
    .priv_data_size = sizeof(CHDContext),
940
    .init           = init,
941
    .close          = uninit,
942
    .decode         = decode,
943
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
944
    .flush          = flush,
945
    .long_name      = NULL_IF_CONFIG_SMALL("SMPTE VC-1 (CrystalHD acceleration)"),
946
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
947
};
948
#endif
949

    
950
#if CONFIG_WMV3_CRYSTALHD_DECODER
951
AVCodec ff_wmv3_crystalhd_decoder = {
952
    .name           = "wmv3_crystalhd",
953
    .type           = AVMEDIA_TYPE_VIDEO,
954
    .id             = CODEC_ID_WMV3,
955
    .priv_data_size = sizeof(CHDContext),
956
    .init           = init,
957
    .close          = uninit,
958
    .decode         = decode,
959
    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
960
    .flush          = flush,
961
    .long_name      = NULL_IF_CONFIG_SMALL("Windows Media Video 9 (CrystalHD acceleration)"),
962
    .pix_fmts       = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
963
};
964
#endif