2 * - CrystalHD decoder module -
4 * Copyright(C) 2010,2011 Philip Langdale <ffmpeg.philipl@overt.org>
6 * This file is part of FFmpeg.
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.
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.
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
24 * - Principles of Operation -
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.
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.
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.
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.
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.
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.
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).
72 /*****************************************************************************
74 ****************************************************************************/
76 #define _XOPEN_SOURCE 600
82 #include <libcrystalhd/bc_dts_types.h>
83 #include <libcrystalhd/bc_dts_defs.h>
84 #include <libcrystalhd/libcrystalhd_if.h>
88 #include "libavutil/imgutils.h"
89 #include "libavutil/intreadwrite.h"
90 #include "libavutil/opt.h"
92 /** Timeout parameter passed to DtsProcOutput() in us */
93 #define OUTPUT_PROC_TIMEOUT 50
94 /** Step between fake timestamps passed to hardware in units of 100ns */
95 #define TIMESTAMP_UNIT 100000
96 /** Initial value in us of the wait in decode() */
97 #define BASE_WAIT 10000
98 /** Increment in us to adjust wait in decode() */
99 #define WAIT_UNIT 1000
102 /*****************************************************************************
103 * Module private data
104 ****************************************************************************/
110 RET_SKIP_NEXT_COPY = 2,
111 RET_COPY_NEXT_FIELD = 3,
114 typedef struct OpaqueList {
115 struct OpaqueList *next;
116 uint64_t fake_timestamp;
117 uint64_t reordered_opaque;
123 AVCodecContext *avctx;
127 uint8_t *orig_extradata;
128 uint32_t orig_extradata_size;
130 AVBitStreamFilterContext *bsfc;
131 AVCodecParserContext *parser;
134 uint8_t *sps_pps_buf;
135 uint32_t sps_pps_size;
137 uint8_t output_ready;
138 uint8_t need_second_field;
139 uint8_t skip_next_output;
140 uint64_t decode_wait;
142 uint64_t last_picture;
152 static const AVOption options[] = {
153 { "crystalhd_downscale_width",
154 "Turn on downscaling to the specified width",
155 offsetof(CHDContext, sWidth),
156 AV_OPT_TYPE_INT, 0, 0, UINT32_MAX,
157 AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM, },
162 /*****************************************************************************
164 ****************************************************************************/
166 static inline BC_MEDIA_SUBTYPE id2subtype(CHDContext *priv, enum CodecID id)
170 return BC_MSUBTYPE_DIVX;
171 case CODEC_ID_MSMPEG4V3:
172 return BC_MSUBTYPE_DIVX311;
173 case CODEC_ID_MPEG2VIDEO:
174 return BC_MSUBTYPE_MPEG2VIDEO;
176 return BC_MSUBTYPE_VC1;
178 return BC_MSUBTYPE_WMV3;
180 return priv->is_nal ? BC_MSUBTYPE_AVC1 : BC_MSUBTYPE_H264;
182 return BC_MSUBTYPE_INVALID;
186 static inline void print_frame_info(CHDContext *priv, BC_DTS_PROC_OUT *output)
188 av_log(priv->avctx, AV_LOG_VERBOSE, "\tYBuffSz: %u\n", output->YbuffSz);
189 av_log(priv->avctx, AV_LOG_VERBOSE, "\tYBuffDoneSz: %u\n",
190 output->YBuffDoneSz);
191 av_log(priv->avctx, AV_LOG_VERBOSE, "\tUVBuffDoneSz: %u\n",
192 output->UVBuffDoneSz);
193 av_log(priv->avctx, AV_LOG_VERBOSE, "\tTimestamp: %"PRIu64"\n",
194 output->PicInfo.timeStamp);
195 av_log(priv->avctx, AV_LOG_VERBOSE, "\tPicture Number: %u\n",
196 output->PicInfo.picture_number);
197 av_log(priv->avctx, AV_LOG_VERBOSE, "\tWidth: %u\n",
198 output->PicInfo.width);
199 av_log(priv->avctx, AV_LOG_VERBOSE, "\tHeight: %u\n",
200 output->PicInfo.height);
201 av_log(priv->avctx, AV_LOG_VERBOSE, "\tChroma: 0x%03x\n",
202 output->PicInfo.chroma_format);
203 av_log(priv->avctx, AV_LOG_VERBOSE, "\tPulldown: %u\n",
204 output->PicInfo.pulldown);
205 av_log(priv->avctx, AV_LOG_VERBOSE, "\tFlags: 0x%08x\n",
206 output->PicInfo.flags);
207 av_log(priv->avctx, AV_LOG_VERBOSE, "\tFrame Rate/Res: %u\n",
208 output->PicInfo.frame_rate);
209 av_log(priv->avctx, AV_LOG_VERBOSE, "\tAspect Ratio: %u\n",
210 output->PicInfo.aspect_ratio);
211 av_log(priv->avctx, AV_LOG_VERBOSE, "\tColor Primaries: %u\n",
212 output->PicInfo.colour_primaries);
213 av_log(priv->avctx, AV_LOG_VERBOSE, "\tMetaData: %u\n",
214 output->PicInfo.picture_meta_payload);
215 av_log(priv->avctx, AV_LOG_VERBOSE, "\tSession Number: %u\n",
216 output->PicInfo.sess_num);
217 av_log(priv->avctx, AV_LOG_VERBOSE, "\tycom: %u\n",
218 output->PicInfo.ycom);
219 av_log(priv->avctx, AV_LOG_VERBOSE, "\tCustom Aspect: %u\n",
220 output->PicInfo.custom_aspect_ratio_width_height);
221 av_log(priv->avctx, AV_LOG_VERBOSE, "\tFrames to Drop: %u\n",
222 output->PicInfo.n_drop);
223 av_log(priv->avctx, AV_LOG_VERBOSE, "\tH264 Valid Fields: 0x%08x\n",
224 output->PicInfo.other.h264.valid);
228 /*****************************************************************************
229 * OpaqueList functions
230 ****************************************************************************/
232 static uint64_t opaque_list_push(CHDContext *priv, uint64_t reordered_opaque,
235 OpaqueList *newNode = av_mallocz(sizeof (OpaqueList));
237 av_log(priv->avctx, AV_LOG_ERROR,
238 "Unable to allocate new node in OpaqueList.\n");
242 newNode->fake_timestamp = TIMESTAMP_UNIT;
243 priv->head = newNode;
245 newNode->fake_timestamp = priv->tail->fake_timestamp + TIMESTAMP_UNIT;
246 priv->tail->next = newNode;
248 priv->tail = newNode;
249 newNode->reordered_opaque = reordered_opaque;
250 newNode->pic_type = pic_type;
252 return newNode->fake_timestamp;
256 * The OpaqueList is built in decode order, while elements will be removed
257 * in presentation order. If frames are reordered, this means we must be
258 * able to remove elements that are not the first element.
260 * Returned node must be freed by caller.
262 static OpaqueList *opaque_list_pop(CHDContext *priv, uint64_t fake_timestamp)
264 OpaqueList *node = priv->head;
267 av_log(priv->avctx, AV_LOG_ERROR,
268 "CrystalHD: Attempted to query non-existent timestamps.\n");
273 * The first element is special-cased because we have to manipulate
274 * the head pointer rather than the previous element in the list.
276 if (priv->head->fake_timestamp == fake_timestamp) {
277 priv->head = node->next;
279 if (!priv->head->next)
280 priv->tail = priv->head;
287 * The list is processed at arm's length so that we have the
288 * previous element available to rewrite its next pointer.
291 OpaqueList *current = node->next;
292 if (current->fake_timestamp == fake_timestamp) {
293 node->next = current->next;
298 current->next = NULL;
305 av_log(priv->avctx, AV_LOG_VERBOSE,
306 "CrystalHD: Couldn't match fake_timestamp.\n");
311 /*****************************************************************************
312 * Video decoder API function definitions
313 ****************************************************************************/
315 static void flush(AVCodecContext *avctx)
317 CHDContext *priv = avctx->priv_data;
319 avctx->has_b_frames = 0;
320 priv->last_picture = -1;
321 priv->output_ready = 0;
322 priv->need_second_field = 0;
323 priv->skip_next_output = 0;
324 priv->decode_wait = BASE_WAIT;
326 if (priv->pic.data[0])
327 avctx->release_buffer(avctx, &priv->pic);
329 /* Flush mode 4 flushes all software and hardware buffers. */
330 DtsFlushInput(priv->dev, 4);
334 static av_cold int uninit(AVCodecContext *avctx)
336 CHDContext *priv = avctx->priv_data;
340 DtsStopDecoder(device);
341 DtsCloseDecoder(device);
342 DtsDeviceClose(device);
345 * Restore original extradata, so that if the decoder is
346 * reinitialised, the bitstream detection and filtering
347 * will work as expected.
349 if (priv->orig_extradata) {
350 av_free(avctx->extradata);
351 avctx->extradata = priv->orig_extradata;
352 avctx->extradata_size = priv->orig_extradata_size;
353 priv->orig_extradata = NULL;
354 priv->orig_extradata_size = 0;
357 av_parser_close(priv->parser);
359 av_bitstream_filter_close(priv->bsfc);
362 av_free(priv->sps_pps_buf);
364 if (priv->pic.data[0])
365 avctx->release_buffer(avctx, &priv->pic);
368 OpaqueList *node = priv->head;
370 OpaqueList *next = node->next;
380 static av_cold int init(AVCodecContext *avctx)
384 BC_INFO_CRYSTAL version;
385 BC_INPUT_FORMAT format = {
388 .OptFlags = 0x80000000 | vdecFrameRate59_94 | 0x40,
389 .width = avctx->width,
390 .height = avctx->height,
393 BC_MEDIA_SUBTYPE subtype;
395 uint32_t mode = DTS_PLAYBACK_MODE |
396 DTS_LOAD_FILE_PLAY_FW |
397 DTS_SKIP_TX_CHK_CPB |
398 DTS_PLAYBACK_DROP_RPT_MODE |
399 DTS_SINGLE_THREADED_MODE |
400 DTS_DFLT_RESOLUTION(vdecRESOLUTION_1080p23_976);
402 av_log(avctx, AV_LOG_VERBOSE, "CrystalHD Init for %s\n",
405 avctx->pix_fmt = PIX_FMT_YUYV422;
407 /* Initialize the library */
408 priv = avctx->priv_data;
410 priv->is_nal = avctx->extradata_size > 0 && *(avctx->extradata) == 1;
411 priv->last_picture = -1;
412 priv->decode_wait = BASE_WAIT;
414 subtype = id2subtype(priv, avctx->codec->id);
416 case BC_MSUBTYPE_AVC1:
421 /* Back up the extradata so it can be restored at close time. */
422 priv->orig_extradata = av_malloc(avctx->extradata_size);
423 if (!priv->orig_extradata) {
424 av_log(avctx, AV_LOG_ERROR,
425 "Failed to allocate copy of extradata\n");
426 return AVERROR(ENOMEM);
428 priv->orig_extradata_size = avctx->extradata_size;
429 memcpy(priv->orig_extradata, avctx->extradata, avctx->extradata_size);
431 priv->bsfc = av_bitstream_filter_init("h264_mp4toannexb");
433 av_log(avctx, AV_LOG_ERROR,
434 "Cannot open the h264_mp4toannexb BSF!\n");
435 return AVERROR_BSF_NOT_FOUND;
437 av_bitstream_filter_filter(priv->bsfc, avctx, NULL, &dummy_p,
438 &dummy_int, NULL, 0, 0);
440 subtype = BC_MSUBTYPE_H264;
442 case BC_MSUBTYPE_H264:
443 format.startCodeSz = 4;
445 case BC_MSUBTYPE_VC1:
446 case BC_MSUBTYPE_WVC1:
447 case BC_MSUBTYPE_WMV3:
448 case BC_MSUBTYPE_WMVA:
449 case BC_MSUBTYPE_MPEG2VIDEO:
450 case BC_MSUBTYPE_DIVX:
451 case BC_MSUBTYPE_DIVX311:
452 format.pMetaData = avctx->extradata;
453 format.metaDataSz = avctx->extradata_size;
456 av_log(avctx, AV_LOG_ERROR, "CrystalHD: Unknown codec name\n");
457 return AVERROR(EINVAL);
459 format.mSubtype = subtype;
462 format.bEnableScaling = 1;
463 format.ScalingParams.sWidth = priv->sWidth;
466 /* Get a decoder instance */
467 av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: starting up\n");
468 // Initialize the Link and Decoder devices
469 ret = DtsDeviceOpen(&priv->dev, mode);
470 if (ret != BC_STS_SUCCESS) {
471 av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: DtsDeviceOpen failed\n");
475 ret = DtsCrystalHDVersion(priv->dev, &version);
476 if (ret != BC_STS_SUCCESS) {
477 av_log(avctx, AV_LOG_VERBOSE,
478 "CrystalHD: DtsCrystalHDVersion failed\n");
481 priv->is_70012 = version.device == 0;
483 if (priv->is_70012 &&
484 (subtype == BC_MSUBTYPE_DIVX || subtype == BC_MSUBTYPE_DIVX311)) {
485 av_log(avctx, AV_LOG_VERBOSE,
486 "CrystalHD: BCM70012 doesn't support MPEG4-ASP/DivX/Xvid\n");
490 ret = DtsSetInputFormat(priv->dev, &format);
491 if (ret != BC_STS_SUCCESS) {
492 av_log(avctx, AV_LOG_ERROR, "CrystalHD: SetInputFormat failed\n");
496 ret = DtsOpenDecoder(priv->dev, BC_STREAM_TYPE_ES);
497 if (ret != BC_STS_SUCCESS) {
498 av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsOpenDecoder failed\n");
502 ret = DtsSetColorSpace(priv->dev, OUTPUT_MODE422_YUY2);
503 if (ret != BC_STS_SUCCESS) {
504 av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsSetColorSpace failed\n");
507 ret = DtsStartDecoder(priv->dev);
508 if (ret != BC_STS_SUCCESS) {
509 av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsStartDecoder failed\n");
512 ret = DtsStartCapture(priv->dev);
513 if (ret != BC_STS_SUCCESS) {
514 av_log(avctx, AV_LOG_ERROR, "CrystalHD: DtsStartCapture failed\n");
518 if (avctx->codec->id == CODEC_ID_H264) {
519 priv->parser = av_parser_init(avctx->codec->id);
521 av_log(avctx, AV_LOG_WARNING,
522 "Cannot open the h.264 parser! Interlaced h.264 content "
523 "will not be detected reliably.\n");
524 priv->parser->flags = PARSER_FLAG_COMPLETE_FRAMES;
526 av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Init complete.\n");
536 static inline CopyRet copy_frame(AVCodecContext *avctx,
537 BC_DTS_PROC_OUT *output,
538 void *data, int *data_size)
541 BC_DTS_STATUS decoder_status = { 0, };
542 uint8_t trust_interlaced;
545 CHDContext *priv = avctx->priv_data;
546 int64_t pkt_pts = AV_NOPTS_VALUE;
547 uint8_t pic_type = 0;
549 uint8_t bottom_field = (output->PicInfo.flags & VDEC_FLAG_BOTTOMFIELD) ==
550 VDEC_FLAG_BOTTOMFIELD;
551 uint8_t bottom_first = !!(output->PicInfo.flags & VDEC_FLAG_BOTTOM_FIRST);
553 int width = output->PicInfo.width;
554 int height = output->PicInfo.height;
556 uint8_t *src = output->Ybuff;
561 if (output->PicInfo.timeStamp != 0) {
562 OpaqueList *node = opaque_list_pop(priv, output->PicInfo.timeStamp);
564 pkt_pts = node->reordered_opaque;
565 pic_type = node->pic_type;
569 * We will encounter a situation where a timestamp cannot be
570 * popped if a second field is being returned. In this case,
571 * each field has the same timestamp and the first one will
572 * cause it to be popped. To keep subsequent calculations
573 * simple, pic_type should be set a FIELD value - doesn't
574 * matter which, but I chose BOTTOM.
576 pic_type = PICT_BOTTOM_FIELD;
578 av_log(avctx, AV_LOG_VERBOSE, "output \"pts\": %"PRIu64"\n",
579 output->PicInfo.timeStamp);
580 av_log(avctx, AV_LOG_VERBOSE, "output picture type %d\n",
584 ret = DtsGetDriverStatus(priv->dev, &decoder_status);
585 if (ret != BC_STS_SUCCESS) {
586 av_log(avctx, AV_LOG_ERROR,
587 "CrystalHD: GetDriverStatus failed: %u\n", ret);
592 * For most content, we can trust the interlaced flag returned
593 * by the hardware, but sometimes we can't. These are the
594 * conditions under which we can trust the flag:
596 * 1) It's not h.264 content
597 * 2) The UNKNOWN_SRC flag is not set
598 * 3) We know we're expecting a second field
599 * 4) The hardware reports this picture and the next picture
600 * have the same picture number.
602 * Note that there can still be interlaced content that will
603 * fail this check, if the hardware hasn't decoded the next
604 * picture or if there is a corruption in the stream. (In either
605 * case a 0 will be returned for the next picture number)
607 trust_interlaced = avctx->codec->id != CODEC_ID_H264 ||
608 !(output->PicInfo.flags & VDEC_FLAG_UNKNOWN_SRC) ||
609 priv->need_second_field ||
610 (decoder_status.picNumFlags & ~0x40000000) ==
611 output->PicInfo.picture_number;
614 * If we got a false negative for trust_interlaced on the first field,
615 * we will realise our mistake here when we see that the picture number is that
616 * of the previous picture. We cannot recover the frame and should discard the
617 * second field to keep the correct number of output frames.
619 if (output->PicInfo.picture_number == priv->last_picture && !priv->need_second_field) {
620 av_log(avctx, AV_LOG_WARNING,
621 "Incorrectly guessed progressive frame. Discarding second field\n");
622 /* Returning without providing a picture. */
626 interlaced = (output->PicInfo.flags & VDEC_FLAG_INTERLACED_SRC) &&
629 if (!trust_interlaced && (decoder_status.picNumFlags & ~0x40000000) == 0) {
630 av_log(avctx, AV_LOG_VERBOSE,
631 "Next picture number unknown. Assuming progressive frame.\n");
634 av_log(avctx, AV_LOG_VERBOSE, "Interlaced state: %d | trust_interlaced %d\n",
635 interlaced, trust_interlaced);
637 if (priv->pic.data[0] && !priv->need_second_field)
638 avctx->release_buffer(avctx, &priv->pic);
640 priv->need_second_field = interlaced && !priv->need_second_field;
642 priv->pic.buffer_hints = FF_BUFFER_HINTS_VALID | FF_BUFFER_HINTS_PRESERVE |
643 FF_BUFFER_HINTS_REUSABLE;
644 if (!priv->pic.data[0]) {
645 if (avctx->get_buffer(avctx, &priv->pic) < 0) {
646 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
651 bwidth = av_image_get_linesize(avctx->pix_fmt, width, 0);
652 if (priv->is_70012) {
657 else if (width <= 1280)
660 sStride = av_image_get_linesize(avctx->pix_fmt, pStride, 0);
665 dStride = priv->pic.linesize[0];
666 dst = priv->pic.data[0];
668 av_log(priv->avctx, AV_LOG_VERBOSE, "CrystalHD: Copying out frame\n");
676 av_log(priv->avctx, AV_LOG_VERBOSE, "Interlaced: bottom field\n");
679 av_log(priv->avctx, AV_LOG_VERBOSE, "Interlaced: top field\n");
683 for (sY = 0; sY < height; dY++, sY++) {
684 memcpy(&(dst[dY * dStride]), &(src[sY * sStride]), bwidth);
688 av_image_copy_plane(dst, dStride, src, sStride, bwidth, height);
691 priv->pic.interlaced_frame = interlaced;
693 priv->pic.top_field_first = !bottom_first;
695 priv->pic.pkt_pts = pkt_pts;
697 if (!priv->need_second_field) {
698 *data_size = sizeof(AVFrame);
699 *(AVFrame *)data = priv->pic;
703 * Two types of PAFF content have been observed. One form causes the
704 * hardware to return a field pair and the other individual fields,
705 * even though the input is always individual fields. We must skip
706 * copying on the next decode() call to maintain pipeline length in
709 if (!interlaced && (output->PicInfo.flags & VDEC_FLAG_UNKNOWN_SRC) &&
710 (pic_type == PICT_TOP_FIELD || pic_type == PICT_BOTTOM_FIELD)) {
711 av_log(priv->avctx, AV_LOG_VERBOSE, "Fieldpair from two packets.\n");
712 return RET_SKIP_NEXT_COPY;
716 * The logic here is purely based on empirical testing with samples.
717 * If we need a second field, it could come from a second input packet,
718 * or it could come from the same field-pair input packet at the current
719 * field. In the first case, we should return and wait for the next time
720 * round to get the second field, while in the second case, we should
721 * ask the decoder for it immediately.
723 * Testing has shown that we are dealing with the fieldpair -> two fields
724 * case if the VDEC_FLAG_UNKNOWN_SRC is not set or if the input picture
725 * type was PICT_FRAME (in this second case, the flag might still be set)
727 return priv->need_second_field &&
728 (!(output->PicInfo.flags & VDEC_FLAG_UNKNOWN_SRC) ||
729 pic_type == PICT_FRAME) ?
730 RET_COPY_NEXT_FIELD : RET_OK;
734 static inline CopyRet receive_frame(AVCodecContext *avctx,
735 void *data, int *data_size)
738 BC_DTS_PROC_OUT output = {
739 .PicInfo.width = avctx->width,
740 .PicInfo.height = avctx->height,
742 CHDContext *priv = avctx->priv_data;
743 HANDLE dev = priv->dev;
747 // Request decoded data from the driver
748 ret = DtsProcOutputNoCopy(dev, OUTPUT_PROC_TIMEOUT, &output);
749 if (ret == BC_STS_FMT_CHANGE) {
750 av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Initial format change\n");
751 avctx->width = output.PicInfo.width;
752 avctx->height = output.PicInfo.height;
753 switch ( output.PicInfo.aspect_ratio ) {
754 case vdecAspectRatioSquare:
755 avctx->sample_aspect_ratio = (AVRational) { 1, 1};
757 case vdecAspectRatio12_11:
758 avctx->sample_aspect_ratio = (AVRational) { 12, 11};
760 case vdecAspectRatio10_11:
761 avctx->sample_aspect_ratio = (AVRational) { 10, 11};
763 case vdecAspectRatio16_11:
764 avctx->sample_aspect_ratio = (AVRational) { 16, 11};
766 case vdecAspectRatio40_33:
767 avctx->sample_aspect_ratio = (AVRational) { 40, 33};
769 case vdecAspectRatio24_11:
770 avctx->sample_aspect_ratio = (AVRational) { 24, 11};
772 case vdecAspectRatio20_11:
773 avctx->sample_aspect_ratio = (AVRational) { 20, 11};
775 case vdecAspectRatio32_11:
776 avctx->sample_aspect_ratio = (AVRational) { 32, 11};
778 case vdecAspectRatio80_33:
779 avctx->sample_aspect_ratio = (AVRational) { 80, 33};
781 case vdecAspectRatio18_11:
782 avctx->sample_aspect_ratio = (AVRational) { 18, 11};
784 case vdecAspectRatio15_11:
785 avctx->sample_aspect_ratio = (AVRational) { 15, 11};
787 case vdecAspectRatio64_33:
788 avctx->sample_aspect_ratio = (AVRational) { 64, 33};
790 case vdecAspectRatio160_99:
791 avctx->sample_aspect_ratio = (AVRational) {160, 99};
793 case vdecAspectRatio4_3:
794 avctx->sample_aspect_ratio = (AVRational) { 4, 3};
796 case vdecAspectRatio16_9:
797 avctx->sample_aspect_ratio = (AVRational) { 16, 9};
799 case vdecAspectRatio221_1:
800 avctx->sample_aspect_ratio = (AVRational) {221, 1};
803 return RET_COPY_AGAIN;
804 } else if (ret == BC_STS_SUCCESS) {
806 if (output.PoutFlags & BC_POUT_FLAGS_PIB_VALID) {
807 if (priv->last_picture == -1) {
809 * Init to one less, so that the incrementing code doesn't
810 * need to be special-cased.
812 priv->last_picture = output.PicInfo.picture_number - 1;
815 if (avctx->codec->id == CODEC_ID_MPEG4 &&
816 output.PicInfo.timeStamp == 0 && priv->bframe_bug) {
817 av_log(avctx, AV_LOG_VERBOSE,
818 "CrystalHD: Not returning packed frame twice.\n");
819 priv->last_picture++;
820 DtsReleaseOutputBuffs(dev, NULL, FALSE);
821 return RET_COPY_AGAIN;
824 print_frame_info(priv, &output);
826 if (priv->last_picture + 1 < output.PicInfo.picture_number) {
827 av_log(avctx, AV_LOG_WARNING,
828 "CrystalHD: Picture Number discontinuity\n");
830 * Have we lost frames? If so, we need to shrink the
831 * pipeline length appropriately.
833 * XXX: I have no idea what the semantics of this situation
834 * are so I don't even know if we've lost frames or which
837 * In any case, only warn the first time.
839 priv->last_picture = output.PicInfo.picture_number - 1;
842 copy_ret = copy_frame(avctx, &output, data, data_size);
843 if (*data_size > 0) {
844 avctx->has_b_frames--;
845 priv->last_picture++;
846 av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Pipeline length: %u\n",
847 avctx->has_b_frames);
851 * An invalid frame has been consumed.
853 av_log(avctx, AV_LOG_ERROR, "CrystalHD: ProcOutput succeeded with "
855 avctx->has_b_frames--;
858 DtsReleaseOutputBuffs(dev, NULL, FALSE);
861 } else if (ret == BC_STS_BUSY) {
862 return RET_COPY_AGAIN;
864 av_log(avctx, AV_LOG_ERROR, "CrystalHD: ProcOutput failed %d\n", ret);
870 static int decode(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt)
873 BC_DTS_STATUS decoder_status = { 0, };
875 CHDContext *priv = avctx->priv_data;
876 HANDLE dev = priv->dev;
877 uint8_t *in_data = avpkt->data;
878 int len = avpkt->size;
880 uint8_t pic_type = 0;
882 av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: decode_frame\n");
884 if (avpkt->size == 7 && !priv->bframe_bug) {
886 * The use of a drop frame triggers the bug
888 av_log(avctx, AV_LOG_INFO,
889 "CrystalHD: Enabling work-around for packed b-frame bug\n");
890 priv->bframe_bug = 1;
891 } else if (avpkt->size == 8 && priv->bframe_bug) {
893 * Delay frames don't trigger the bug
895 av_log(avctx, AV_LOG_INFO,
896 "CrystalHD: Disabling work-around for packed b-frame bug\n");
897 priv->bframe_bug = 0;
901 int32_t tx_free = (int32_t)DtsTxFreeSize(dev);
907 ret = av_bitstream_filter_filter(priv->bsfc, avctx, NULL,
909 avpkt->data, len, 0);
917 H264Context *h = priv->parser->priv_data;
919 index = av_parser_parse2(priv->parser, avctx, &pout, &psize,
920 in_data, len, avctx->pkt->pts,
923 av_log(avctx, AV_LOG_WARNING,
924 "CrystalHD: Failed to parse h.264 packet to "
925 "detect interlacing.\n");
926 } else if (index != len) {
927 av_log(avctx, AV_LOG_WARNING,
928 "CrystalHD: Failed to parse h.264 packet "
929 "completely. Interlaced frames may be "
930 "incorrectly detected\n.");
932 av_log(avctx, AV_LOG_VERBOSE,
933 "CrystalHD: parser picture type %d\n",
934 h->s.picture_structure);
935 pic_type = h->s.picture_structure;
938 av_log(avctx, AV_LOG_WARNING,
939 "CrystalHD: mp4toannexb filter failed to filter "
940 "packet. Interlaced frames may be incorrectly "
945 if (len < tx_free - 1024) {
947 * Despite being notionally opaque, either libcrystalhd or
948 * the hardware itself will mangle pts values that are too
949 * small or too large. The docs claim it should be in units
950 * of 100ns. Given that we're nominally dealing with a black
951 * box on both sides, any transform we do has no guarantee of
952 * avoiding mangling so we need to build a mapping to values
953 * we know will not be mangled.
955 uint64_t pts = opaque_list_push(priv, avctx->pkt->pts, pic_type);
960 return AVERROR(ENOMEM);
962 av_log(priv->avctx, AV_LOG_VERBOSE,
963 "input \"pts\": %"PRIu64"\n", pts);
964 ret = DtsProcInput(dev, in_data, len, pts, 0);
968 if (ret == BC_STS_BUSY) {
969 av_log(avctx, AV_LOG_WARNING,
970 "CrystalHD: ProcInput returned busy\n");
972 return AVERROR(EBUSY);
973 } else if (ret != BC_STS_SUCCESS) {
974 av_log(avctx, AV_LOG_ERROR,
975 "CrystalHD: ProcInput failed: %u\n", ret);
978 avctx->has_b_frames++;
980 av_log(avctx, AV_LOG_WARNING, "CrystalHD: Input buffer full\n");
981 len = 0; // We didn't consume any bytes.
984 av_log(avctx, AV_LOG_INFO, "CrystalHD: No more input data\n");
987 if (priv->skip_next_output) {
988 av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Skipping next output.\n");
989 priv->skip_next_output = 0;
990 avctx->has_b_frames--;
994 ret = DtsGetDriverStatus(dev, &decoder_status);
995 if (ret != BC_STS_SUCCESS) {
996 av_log(avctx, AV_LOG_ERROR, "CrystalHD: GetDriverStatus failed\n");
1001 * No frames ready. Don't try to extract.
1003 * Empirical testing shows that ReadyListCount can be a damn lie,
1004 * and ProcOut still fails when count > 0. The same testing showed
1005 * that two more iterations were needed before ProcOutput would
1008 if (priv->output_ready < 2) {
1009 if (decoder_status.ReadyListCount != 0)
1010 priv->output_ready++;
1012 av_log(avctx, AV_LOG_INFO, "CrystalHD: Filling pipeline.\n");
1014 } else if (decoder_status.ReadyListCount == 0) {
1016 * After the pipeline is established, if we encounter a lack of frames
1017 * that probably means we're not giving the hardware enough time to
1018 * decode them, so start increasing the wait time at the end of a
1022 priv->decode_wait += WAIT_UNIT;
1023 av_log(avctx, AV_LOG_INFO, "CrystalHD: No frames ready. Returning\n");
1028 rec_ret = receive_frame(avctx, data, data_size);
1029 if (rec_ret == RET_OK && *data_size == 0) {
1031 * This case is for when the encoded fields are stored
1032 * separately and we get a separate avpkt for each one. To keep
1033 * the pipeline stable, we should return nothing and wait for
1034 * the next time round to grab the second field.
1035 * H.264 PAFF is an example of this.
1037 av_log(avctx, AV_LOG_VERBOSE, "Returning after first field.\n");
1038 avctx->has_b_frames--;
1039 } else if (rec_ret == RET_COPY_NEXT_FIELD) {
1041 * This case is for when the encoded fields are stored in a
1042 * single avpkt but the hardware returns then separately. Unless
1043 * we grab the second field before returning, we'll slip another
1044 * frame in the pipeline and if that happens a lot, we're sunk.
1045 * So we have to get that second field now.
1046 * Interlaced mpeg2 and vc1 are examples of this.
1048 av_log(avctx, AV_LOG_VERBOSE, "Trying to get second field.\n");
1050 usleep(priv->decode_wait);
1051 ret = DtsGetDriverStatus(dev, &decoder_status);
1052 if (ret == BC_STS_SUCCESS &&
1053 decoder_status.ReadyListCount > 0) {
1054 rec_ret = receive_frame(avctx, data, data_size);
1055 if ((rec_ret == RET_OK && *data_size > 0) ||
1056 rec_ret == RET_ERROR)
1060 av_log(avctx, AV_LOG_VERBOSE, "CrystalHD: Got second field.\n");
1061 } else if (rec_ret == RET_SKIP_NEXT_COPY) {
1063 * Two input packets got turned into a field pair. Gawd.
1065 av_log(avctx, AV_LOG_VERBOSE,
1066 "Don't output on next decode call.\n");
1067 priv->skip_next_output = 1;
1070 * If rec_ret == RET_COPY_AGAIN, that means that either we just handled
1071 * a FMT_CHANGE event and need to go around again for the actual frame,
1072 * we got a busy status and need to try again, or we're dealing with
1073 * packed b-frames, where the hardware strangely returns the packed
1074 * p-frame twice. We choose to keep the second copy as it carries the
1077 } while (rec_ret == RET_COPY_AGAIN);
1078 usleep(priv->decode_wait);
1083 #if CONFIG_H264_CRYSTALHD_DECODER
1084 static AVClass h264_class = {
1086 av_default_item_name,
1088 LIBAVUTIL_VERSION_INT,
1091 AVCodec ff_h264_crystalhd_decoder = {
1092 .name = "h264_crystalhd",
1093 .type = AVMEDIA_TYPE_VIDEO,
1094 .id = CODEC_ID_H264,
1095 .priv_data_size = sizeof(CHDContext),
1099 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1101 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (CrystalHD acceleration)"),
1102 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
1103 .priv_class = &h264_class,
1107 #if CONFIG_MPEG2_CRYSTALHD_DECODER
1108 static AVClass mpeg2_class = {
1110 av_default_item_name,
1112 LIBAVUTIL_VERSION_INT,
1115 AVCodec ff_mpeg2_crystalhd_decoder = {
1116 .name = "mpeg2_crystalhd",
1117 .type = AVMEDIA_TYPE_VIDEO,
1118 .id = CODEC_ID_MPEG2VIDEO,
1119 .priv_data_size = sizeof(CHDContext),
1123 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1125 .long_name = NULL_IF_CONFIG_SMALL("MPEG-2 Video (CrystalHD acceleration)"),
1126 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
1127 .priv_class = &mpeg2_class,
1131 #if CONFIG_MPEG4_CRYSTALHD_DECODER
1132 static AVClass mpeg4_class = {
1134 av_default_item_name,
1136 LIBAVUTIL_VERSION_INT,
1139 AVCodec ff_mpeg4_crystalhd_decoder = {
1140 .name = "mpeg4_crystalhd",
1141 .type = AVMEDIA_TYPE_VIDEO,
1142 .id = CODEC_ID_MPEG4,
1143 .priv_data_size = sizeof(CHDContext),
1147 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1149 .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 Part 2 (CrystalHD acceleration)"),
1150 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
1151 .priv_class = &mpeg4_class,
1155 #if CONFIG_MSMPEG4_CRYSTALHD_DECODER
1156 static AVClass msmpeg4_class = {
1157 "msmpeg4_crystalhd",
1158 av_default_item_name,
1160 LIBAVUTIL_VERSION_INT,
1163 AVCodec ff_msmpeg4_crystalhd_decoder = {
1164 .name = "msmpeg4_crystalhd",
1165 .type = AVMEDIA_TYPE_VIDEO,
1166 .id = CODEC_ID_MSMPEG4V3,
1167 .priv_data_size = sizeof(CHDContext),
1171 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
1173 .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 Part 2 Microsoft variant version 3 (CrystalHD acceleration)"),
1174 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
1175 .priv_class = &msmpeg4_class,
1179 #if CONFIG_VC1_CRYSTALHD_DECODER
1180 static AVClass vc1_class = {
1182 av_default_item_name,
1184 LIBAVUTIL_VERSION_INT,
1187 AVCodec ff_vc1_crystalhd_decoder = {
1188 .name = "vc1_crystalhd",
1189 .type = AVMEDIA_TYPE_VIDEO,
1191 .priv_data_size = sizeof(CHDContext),
1195 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1197 .long_name = NULL_IF_CONFIG_SMALL("SMPTE VC-1 (CrystalHD acceleration)"),
1198 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
1199 .priv_class = &vc1_class,
1203 #if CONFIG_WMV3_CRYSTALHD_DECODER
1204 static AVClass wmv3_class = {
1206 av_default_item_name,
1208 LIBAVUTIL_VERSION_INT,
1211 AVCodec ff_wmv3_crystalhd_decoder = {
1212 .name = "wmv3_crystalhd",
1213 .type = AVMEDIA_TYPE_VIDEO,
1214 .id = CODEC_ID_WMV3,
1215 .priv_data_size = sizeof(CHDContext),
1219 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1221 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Video 9 (CrystalHD acceleration)"),
1222 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_YUYV422, PIX_FMT_NONE},
1223 .priv_class = &wmv3_class,