Merge commit '3c650efb81aaa3b395ba4606ee68a47ee4efb57b'
[ffmpeg.git] / libavcodec / vda_h264.c
1 /*
2  * VDA H264 HW acceleration.
3  *
4  * copyright (c) 2011 Sebastien Zwickert
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 #include <CoreFoundation/CFDictionary.h>
24 #include <CoreFoundation/CFNumber.h>
25 #include <CoreFoundation/CFData.h>
26
27 #include "vda.h"
28 #include "libavutil/avutil.h"
29 #include "h264.h"
30
31 struct vda_buffer {
32     CVPixelBufferRef cv_buffer;
33 };
34 #include "internal.h"
35 #include "vda_internal.h"
36
37 typedef struct VDAContext {
38     // The current bitstream buffer.
39     uint8_t             *bitstream;
40
41     // The current size of the bitstream.
42     int                  bitstream_size;
43
44     // The reference size used for fast reallocation.
45     int                  allocated_size;
46
47     CVImageBufferRef frame;
48 } VDAContext;
49
50 /* Decoder callback that adds the vda frame to the queue in display order. */
51 static void vda_decoder_callback(void *vda_hw_ctx,
52                                  CFDictionaryRef user_info,
53                                  OSStatus status,
54                                  uint32_t infoFlags,
55                                  CVImageBufferRef image_buffer)
56 {
57     struct vda_context *vda_ctx = vda_hw_ctx;
58
59     if (infoFlags & kVDADecodeInfo_FrameDropped)
60         vda_ctx->cv_buffer = NULL;
61
62     if (!image_buffer)
63         return;
64
65     if (vda_ctx->cv_pix_fmt_type != CVPixelBufferGetPixelFormatType(image_buffer))
66         return;
67
68     vda_ctx->cv_buffer = CVPixelBufferRetain(image_buffer);
69 }
70
71 static int vda_sync_decode(VDAContext *ctx, struct vda_context *vda_ctx)
72 {
73     OSStatus status;
74     CFDataRef coded_frame;
75     uint32_t flush_flags = 1 << 0; ///< kVDADecoderFlush_emitFrames
76
77     coded_frame = CFDataCreate(kCFAllocatorDefault,
78                                ctx->bitstream,
79                                ctx->bitstream_size);
80
81     status = VDADecoderDecode(vda_ctx->decoder, 0, coded_frame, NULL);
82
83     if (kVDADecoderNoErr == status)
84         status = VDADecoderFlush(vda_ctx->decoder, flush_flags);
85
86     CFRelease(coded_frame);
87
88     return status;
89 }
90
91
92 static int vda_old_h264_start_frame(AVCodecContext *avctx,
93                                 av_unused const uint8_t *buffer,
94                                 av_unused uint32_t size)
95 {
96     VDAContext *vda = avctx->internal->hwaccel_priv_data;
97     struct vda_context *vda_ctx = avctx->hwaccel_context;
98
99     if (!vda_ctx->decoder)
100         return -1;
101
102     vda->bitstream_size = 0;
103
104     return 0;
105 }
106
107 static int vda_old_h264_decode_slice(AVCodecContext *avctx,
108                                  const uint8_t *buffer,
109                                  uint32_t size)
110 {
111     VDAContext *vda             = avctx->internal->hwaccel_priv_data;
112     struct vda_context *vda_ctx = avctx->hwaccel_context;
113     void *tmp;
114
115     if (!vda_ctx->decoder)
116         return -1;
117
118     tmp = av_fast_realloc(vda->bitstream,
119                           &vda->allocated_size,
120                           vda->bitstream_size + size + 4);
121     if (!tmp)
122         return AVERROR(ENOMEM);
123
124     vda->bitstream = tmp;
125
126     AV_WB32(vda->bitstream + vda->bitstream_size, size);
127     memcpy(vda->bitstream + vda->bitstream_size + 4, buffer, size);
128
129     vda->bitstream_size += size + 4;
130
131     return 0;
132 }
133
134 static void vda_h264_release_buffer(void *opaque, uint8_t *data)
135 {
136     struct vda_buffer *context = opaque;
137     CVPixelBufferRelease(context->cv_buffer);
138     av_free(context);
139 }
140
141 static int vda_old_h264_end_frame(AVCodecContext *avctx)
142 {
143     H264Context *h                      = avctx->priv_data;
144     VDAContext *vda                     = avctx->internal->hwaccel_priv_data;
145     struct vda_context *vda_ctx         = avctx->hwaccel_context;
146     AVFrame *frame                      = &h->cur_pic_ptr->f;
147     struct vda_buffer *context;
148     AVBufferRef *buffer;
149     int status;
150
151     if (!vda_ctx->decoder || !vda->bitstream)
152         return -1;
153
154     status = vda_sync_decode(vda, vda_ctx);
155     frame->data[3] = (void*)vda_ctx->cv_buffer;
156
157     if (status)
158         av_log(avctx, AV_LOG_ERROR, "Failed to decode frame (%d)\n", status);
159
160     if (!vda_ctx->use_ref_buffer || status)
161         return status;
162
163     context = av_mallocz(sizeof(*context));
164     buffer = av_buffer_create(NULL, 0, vda_h264_release_buffer, context, 0);
165     if (!context || !buffer) {
166         CVPixelBufferRelease(vda_ctx->cv_buffer);
167         av_free(context);
168         return -1;
169     }
170
171     context->cv_buffer = vda_ctx->cv_buffer;
172     frame->buf[3] = buffer;
173
174     return status;
175 }
176
177 int ff_vda_create_decoder(struct vda_context *vda_ctx,
178                           uint8_t *extradata,
179                           int extradata_size)
180 {
181     OSStatus status;
182     CFNumberRef height;
183     CFNumberRef width;
184     CFNumberRef format;
185     CFDataRef avc_data;
186     CFMutableDictionaryRef config_info;
187     CFMutableDictionaryRef buffer_attributes;
188     CFMutableDictionaryRef io_surface_properties;
189     CFNumberRef cv_pix_fmt;
190
191     vda_ctx->priv_bitstream = NULL;
192     vda_ctx->priv_allocated_size = 0;
193
194     /* Each VCL NAL in the bitstream sent to the decoder
195      * is preceded by a 4 bytes length header.
196      * Change the avcC atom header if needed, to signal headers of 4 bytes. */
197     if (extradata_size >= 4 && (extradata[4] & 0x03) != 0x03) {
198         uint8_t *rw_extradata;
199
200         if (!(rw_extradata = av_malloc(extradata_size)))
201             return AVERROR(ENOMEM);
202
203         memcpy(rw_extradata, extradata, extradata_size);
204
205         rw_extradata[4] |= 0x03;
206
207         avc_data = CFDataCreate(kCFAllocatorDefault, rw_extradata, extradata_size);
208
209         av_freep(&rw_extradata);
210     } else {
211         avc_data = CFDataCreate(kCFAllocatorDefault, extradata, extradata_size);
212     }
213
214     config_info = CFDictionaryCreateMutable(kCFAllocatorDefault,
215                                             4,
216                                             &kCFTypeDictionaryKeyCallBacks,
217                                             &kCFTypeDictionaryValueCallBacks);
218
219     height   = CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt32Type, &vda_ctx->height);
220     width    = CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt32Type, &vda_ctx->width);
221     format   = CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt32Type, &vda_ctx->format);
222
223     CFDictionarySetValue(config_info, kVDADecoderConfiguration_Height, height);
224     CFDictionarySetValue(config_info, kVDADecoderConfiguration_Width, width);
225     CFDictionarySetValue(config_info, kVDADecoderConfiguration_SourceFormat, format);
226     CFDictionarySetValue(config_info, kVDADecoderConfiguration_avcCData, avc_data);
227
228     buffer_attributes = CFDictionaryCreateMutable(kCFAllocatorDefault,
229                                                   2,
230                                                   &kCFTypeDictionaryKeyCallBacks,
231                                                   &kCFTypeDictionaryValueCallBacks);
232     io_surface_properties = CFDictionaryCreateMutable(kCFAllocatorDefault,
233                                                       0,
234                                                       &kCFTypeDictionaryKeyCallBacks,
235                                                       &kCFTypeDictionaryValueCallBacks);
236     cv_pix_fmt  = CFNumberCreate(kCFAllocatorDefault,
237                                  kCFNumberSInt32Type,
238                                  &vda_ctx->cv_pix_fmt_type);
239     CFDictionarySetValue(buffer_attributes,
240                          kCVPixelBufferPixelFormatTypeKey,
241                          cv_pix_fmt);
242     CFDictionarySetValue(buffer_attributes,
243                          kCVPixelBufferIOSurfacePropertiesKey,
244                          io_surface_properties);
245
246     status = VDADecoderCreate(config_info,
247                               buffer_attributes,
248                               (VDADecoderOutputCallback *)vda_decoder_callback,
249                               vda_ctx,
250                               &vda_ctx->decoder);
251
252     CFRelease(height);
253     CFRelease(width);
254     CFRelease(format);
255     CFRelease(avc_data);
256     CFRelease(config_info);
257     CFRelease(io_surface_properties);
258     CFRelease(cv_pix_fmt);
259     CFRelease(buffer_attributes);
260
261     return status;
262 }
263
264 int ff_vda_destroy_decoder(struct vda_context *vda_ctx)
265 {
266     OSStatus status = kVDADecoderNoErr;
267
268     if (vda_ctx->decoder)
269         status = VDADecoderDestroy(vda_ctx->decoder);
270
271     return status;
272 }
273
274 static int vda_h264_uninit(AVCodecContext *avctx)
275 {
276     VDAContext *vda = avctx->internal->hwaccel_priv_data;
277     if (vda) {
278         av_freep(&vda->bitstream);
279         if (vda->frame)
280             CVPixelBufferRelease(vda->frame);
281     }
282     return 0;
283 }
284
285 AVHWAccel ff_h264_vda_old_hwaccel = {
286     .name           = "h264_vda",
287     .type           = AVMEDIA_TYPE_VIDEO,
288     .id             = AV_CODEC_ID_H264,
289     .pix_fmt        = AV_PIX_FMT_VDA_VLD,
290     .start_frame    = vda_old_h264_start_frame,
291     .decode_slice   = vda_old_h264_decode_slice,
292     .end_frame      = vda_old_h264_end_frame,
293     .uninit         = vda_h264_uninit,
294     .priv_data_size = sizeof(VDAContext),
295 };
296
297 void ff_vda_output_callback(void *opaque,
298                             CFDictionaryRef user_info,
299                             OSStatus status,
300                             uint32_t infoFlags,
301                             CVImageBufferRef image_buffer)
302 {
303     AVCodecContext *ctx = opaque;
304     VDAContext *vda = ctx->internal->hwaccel_priv_data;
305
306
307     if (vda->frame) {
308         CVPixelBufferRelease(vda->frame);
309         vda->frame = NULL;
310     }
311
312     if (!image_buffer)
313         return;
314
315     vda->frame = CVPixelBufferRetain(image_buffer);
316 }
317
318 static int vda_h264_start_frame(AVCodecContext *avctx,
319                                 const uint8_t *buffer,
320                                 uint32_t size)
321 {
322     VDAContext *vda = avctx->internal->hwaccel_priv_data;
323
324     vda->bitstream_size = 0;
325
326     return 0;
327 }
328
329 static int vda_h264_decode_slice(AVCodecContext *avctx,
330                                  const uint8_t *buffer,
331                                  uint32_t size)
332 {
333     VDAContext *vda       = avctx->internal->hwaccel_priv_data;
334     void *tmp;
335
336     tmp = av_fast_realloc(vda->bitstream,
337                           &vda->allocated_size,
338                           vda->bitstream_size + size + 4);
339     if (!tmp)
340         return AVERROR(ENOMEM);
341
342     vda->bitstream = tmp;
343
344     AV_WB32(vda->bitstream + vda->bitstream_size, size);
345     memcpy(vda->bitstream + vda->bitstream_size + 4, buffer, size);
346
347     vda->bitstream_size += size + 4;
348
349     return 0;
350 }
351
352 static void release_buffer(void *opaque, uint8_t *data)
353 {
354     CVImageBufferRef frame = (CVImageBufferRef)data;
355     CVPixelBufferRelease(frame);
356 }
357
358 static int vda_h264_end_frame(AVCodecContext *avctx)
359 {
360     H264Context *h        = avctx->priv_data;
361     VDAContext *vda       = avctx->internal->hwaccel_priv_data;
362     AVVDAContext *vda_ctx = avctx->hwaccel_context;
363     AVFrame *frame        = &h->cur_pic_ptr->f;
364     uint32_t flush_flags  = 1 << 0; ///< kVDADecoderFlush_emitFrames
365     CFDataRef coded_frame;
366     OSStatus status;
367
368     if (!vda->bitstream_size)
369         return AVERROR_INVALIDDATA;
370
371
372     coded_frame = CFDataCreate(kCFAllocatorDefault,
373                                vda->bitstream,
374                                vda->bitstream_size);
375
376     status = VDADecoderDecode(vda_ctx->decoder, 0, coded_frame, NULL);
377
378     if (status == kVDADecoderNoErr)
379         status = VDADecoderFlush(vda_ctx->decoder, flush_flags);
380
381     CFRelease(coded_frame);
382
383     if (status != kVDADecoderNoErr) {
384         av_log(avctx, AV_LOG_ERROR, "Failed to decode frame (%d)\n", status);
385         return AVERROR_UNKNOWN;
386     }
387
388     if (vda->frame) {
389         av_buffer_unref(&frame->buf[0]);
390
391         frame->buf[0] = av_buffer_create((uint8_t*)vda->frame,
392                                          sizeof(vda->frame),
393                                          release_buffer, NULL,
394                                          AV_BUFFER_FLAG_READONLY);
395         if (!frame->buf)
396             return AVERROR(ENOMEM);
397
398         frame->data[3] = (uint8_t*)vda->frame;
399         vda->frame = NULL;
400     }
401
402     return 0;
403 }
404
405 int ff_vda_default_init(AVCodecContext *avctx)
406 {
407     AVVDAContext *vda_ctx = avctx->hwaccel_context;
408     OSStatus status = kVDADecoderNoErr;
409     CFNumberRef height;
410     CFNumberRef width;
411     CFNumberRef format;
412     CFDataRef avc_data;
413     CFMutableDictionaryRef config_info;
414     CFMutableDictionaryRef buffer_attributes;
415     CFMutableDictionaryRef io_surface_properties;
416     CFNumberRef cv_pix_fmt;
417     int32_t fmt = 'avc1', pix_fmt = kCVPixelFormatType_422YpCbCr8;
418
419     // kCVPixelFormatType_420YpCbCr8Planar;
420
421     /* Each VCL NAL in the bitstream sent to the decoder
422      * is preceded by a 4 bytes length header.
423      * Change the avcC atom header if needed, to signal headers of 4 bytes. */
424     if (avctx->extradata_size >= 4 && (avctx->extradata[4] & 0x03) != 0x03) {
425         uint8_t *rw_extradata;
426
427         if (!(rw_extradata = av_malloc(avctx->extradata_size)))
428             return AVERROR(ENOMEM);
429
430         memcpy(rw_extradata, avctx->extradata, avctx->extradata_size);
431
432         rw_extradata[4] |= 0x03;
433
434         avc_data = CFDataCreate(kCFAllocatorDefault, rw_extradata, avctx->extradata_size);
435
436         av_freep(&rw_extradata);
437     } else {
438         avc_data = CFDataCreate(kCFAllocatorDefault,
439                                 avctx->extradata, avctx->extradata_size);
440     }
441
442     config_info = CFDictionaryCreateMutable(kCFAllocatorDefault,
443                                             4,
444                                             &kCFTypeDictionaryKeyCallBacks,
445                                             &kCFTypeDictionaryValueCallBacks);
446
447     height = CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt32Type, &avctx->height);
448     width  = CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt32Type, &avctx->width);
449     format = CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt32Type, &fmt);
450     CFDictionarySetValue(config_info, kVDADecoderConfiguration_Height, height);
451     CFDictionarySetValue(config_info, kVDADecoderConfiguration_Width, width);
452     CFDictionarySetValue(config_info, kVDADecoderConfiguration_avcCData, avc_data);
453     CFDictionarySetValue(config_info, kVDADecoderConfiguration_SourceFormat, format);
454
455     buffer_attributes = CFDictionaryCreateMutable(kCFAllocatorDefault,
456                                                   2,
457                                                   &kCFTypeDictionaryKeyCallBacks,
458                                                   &kCFTypeDictionaryValueCallBacks);
459     io_surface_properties = CFDictionaryCreateMutable(kCFAllocatorDefault,
460                                                       0,
461                                                       &kCFTypeDictionaryKeyCallBacks,
462                                                       &kCFTypeDictionaryValueCallBacks);
463     cv_pix_fmt      = CFNumberCreate(kCFAllocatorDefault,
464                                      kCFNumberSInt32Type,
465                                      &pix_fmt);
466
467     CFDictionarySetValue(buffer_attributes,
468                          kCVPixelBufferPixelFormatTypeKey,
469                          cv_pix_fmt);
470     CFDictionarySetValue(buffer_attributes,
471                          kCVPixelBufferIOSurfacePropertiesKey,
472                          io_surface_properties);
473
474     status = VDADecoderCreate(config_info,
475                               buffer_attributes,
476                               (VDADecoderOutputCallback *)ff_vda_output_callback,
477                               avctx,
478                               &vda_ctx->decoder);
479
480     CFRelease(format);
481     CFRelease(height);
482     CFRelease(width);
483     CFRelease(avc_data);
484     CFRelease(config_info);
485     CFRelease(cv_pix_fmt);
486     CFRelease(io_surface_properties);
487     CFRelease(buffer_attributes);
488
489     if (status != kVDADecoderNoErr) {
490         av_log(avctx, AV_LOG_ERROR, "Cannot initialize VDA %d\n", status);
491     }
492
493     switch (status) {
494     case kVDADecoderHardwareNotSupportedErr:
495     case kVDADecoderFormatNotSupportedErr:
496         return AVERROR(ENOSYS);
497     case kVDADecoderConfigurationError:
498         return AVERROR(EINVAL);
499     case kVDADecoderDecoderFailedErr:
500         return AVERROR_INVALIDDATA;
501     case kVDADecoderNoErr:
502         return 0;
503     default:
504         return AVERROR_UNKNOWN;
505     }
506 }
507
508 static int vda_h264_alloc_frame(AVCodecContext *avctx, AVFrame *frame)
509 {
510     frame->width  = avctx->width;
511     frame->height = avctx->height;
512     frame->format = avctx->pix_fmt;
513     frame->buf[0] = av_buffer_alloc(1);
514
515     if (!frame->buf[0])
516         return AVERROR(ENOMEM);
517     return 0;
518 }
519
520 AVHWAccel ff_h264_vda_hwaccel = {
521     .name           = "h264_vda",
522     .type           = AVMEDIA_TYPE_VIDEO,
523     .id             = AV_CODEC_ID_H264,
524     .pix_fmt        = AV_PIX_FMT_VDA,
525     .alloc_frame    = vda_h264_alloc_frame,
526     .start_frame    = vda_h264_start_frame,
527     .decode_slice   = vda_h264_decode_slice,
528     .end_frame      = vda_h264_end_frame,
529     .uninit         = vda_h264_uninit,
530     .priv_data_size = sizeof(VDAContext),
531 };