1 /*
2  * This copyright notice applies to this header file only:
3  *
4  * Copyright (c) 2010-2016 NVIDIA Corporation
5  *
6  * Permission is hereby granted, free of charge, to any person
7  * obtaining a copy of this software and associated documentation
8  * files (the "Software"), to deal in the Software without
9  * restriction, including without limitation the rights to use,
10  * copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the software, and to permit persons to whom the
12  * software is furnished to do so, subject to the following
13  * conditions:
14  *
15  * The above copyright notice and this permission notice shall be
16  * included in all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
20  * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
21  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
22  * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
23  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
24  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
25  * OTHER DEALINGS IN THE SOFTWARE.
26  */
27 
28 /**
29  * \file cuviddec.h
30  * NvCuvid API provides Video Decoding interface to NVIDIA GPU devices.
31  * \date 2015-2016
32  * This file contains constants, structure definitions and function prototypes used for decoding.
33  */
34 
35 #if !defined(__CUDA_VIDEO_H__)
36 #define __CUDA_VIDEO_H__
37 
38 #if defined(__x86_64) || defined(AMD64) || defined(_M_AMD64)
39 #if (CUDA_VERSION >= 3020) && (!defined(CUDA_FORCE_API_VERSION) || (CUDA_FORCE_API_VERSION >= 3020))
40 #define __CUVID_DEVPTR64
41 #endif
42 #endif
43 
44 #if defined(__cplusplus)
45 extern "C" {
46 #endif /* __cplusplus */
47 
48 typedef void *CUvideodecoder;
49 typedef struct _CUcontextlock_st *CUvideoctxlock;
50 
51 /**
52  * \addtogroup VIDEO_DECODER Video Decoder
53  * @{
54  */
55 
56 /*!
57  * \enum cudaVideoCodec
58  * Video Codec Enums
59  */
60 typedef enum cudaVideoCodec_enum {
61     cudaVideoCodec_MPEG1=0,                 /**<  MPEG1   */
62     cudaVideoCodec_MPEG2,                   /**<  MPEG2  */
63     cudaVideoCodec_MPEG4,                   /**<  MPEG4   */
64     cudaVideoCodec_VC1,                     /**<  VC1   */
65     cudaVideoCodec_H264,                    /**<  H264   */
66     cudaVideoCodec_JPEG,                    /**<  JPEG   */
67     cudaVideoCodec_H264_SVC,                /**<  H264-SVC   */
68     cudaVideoCodec_H264_MVC,                /**<  H264-MVC   */
69     cudaVideoCodec_HEVC,                    /**<  HEVC   */
70     cudaVideoCodec_VP8,                     /**<  VP8   */
71     cudaVideoCodec_VP9,                     /**<  VP9   */
72     cudaVideoCodec_NumCodecs,               /**<  Max COdecs   */
73     // Uncompressed YUV
74     cudaVideoCodec_YUV420 = (('I'<<24)|('Y'<<16)|('U'<<8)|('V')),   /**< Y,U,V (4:2:0)  */
75     cudaVideoCodec_YV12   = (('Y'<<24)|('V'<<16)|('1'<<8)|('2')),   /**< Y,V,U (4:2:0)  */
76     cudaVideoCodec_NV12   = (('N'<<24)|('V'<<16)|('1'<<8)|('2')),   /**< Y,UV  (4:2:0)  */
77     cudaVideoCodec_YUYV   = (('Y'<<24)|('U'<<16)|('Y'<<8)|('V')),   /**< YUYV/YUY2 (4:2:2)  */
78     cudaVideoCodec_UYVY   = (('U'<<24)|('Y'<<16)|('V'<<8)|('Y'))    /**< UYVY (4:2:2)  */
79 } cudaVideoCodec;
80 
81 /*!
82  * \enum cudaVideoSurfaceFormat
83  * Video Surface Formats Enums
84  */
85 typedef enum cudaVideoSurfaceFormat_enum {
86     cudaVideoSurfaceFormat_NV12=0,      /**< NV12  */
87     cudaVideoSurfaceFormat_P016=1       /**< P016  */
88 } cudaVideoSurfaceFormat;
89 
90 /*!
91  * \enum cudaVideoDeinterlaceMode
92  * Deinterlacing Modes Enums
93  */
94 typedef enum cudaVideoDeinterlaceMode_enum {
95     cudaVideoDeinterlaceMode_Weave=0,   /**< Weave both fields (no deinterlacing) */
96     cudaVideoDeinterlaceMode_Bob,       /**< Drop one field  */
97     cudaVideoDeinterlaceMode_Adaptive   /**< Adaptive deinterlacing  */
98 } cudaVideoDeinterlaceMode;
99 
100 /*!
101  * \enum cudaVideoChromaFormat
102  * Chroma Formats Enums
103  */
104 typedef enum cudaVideoChromaFormat_enum {
105     cudaVideoChromaFormat_Monochrome=0,  /**< MonoChrome */
106     cudaVideoChromaFormat_420,           /**< 4:2:0 */
107     cudaVideoChromaFormat_422,           /**< 4:2:2 */
108     cudaVideoChromaFormat_444            /**< 4:4:4 */
109 } cudaVideoChromaFormat;
110 
111 /*!
112  * \enum cudaVideoCreateFlags
113  * Decoder Flags Enums
114  */
115 typedef enum cudaVideoCreateFlags_enum {
116     cudaVideoCreate_Default = 0x00,     /**< Default operation mode: use dedicated video engines */
117     cudaVideoCreate_PreferCUDA = 0x01,  /**< Use a CUDA-based decoder if faster than dedicated engines (requires a valid vidLock object for multi-threading) */
118     cudaVideoCreate_PreferDXVA = 0x02,  /**< Go through DXVA internally if possible (requires D3D9 interop) */
119     cudaVideoCreate_PreferCUVID = 0x04  /**< Use dedicated video engines directly */
120 } cudaVideoCreateFlags;
121 
122 /*!
123  * \struct CUVIDDECODECREATEINFO
124  * Struct used in create decoder
125  */
126 typedef struct _CUVIDDECODECREATEINFO
127 {
128     unsigned long ulWidth;              /**< Coded Sequence Width */
129     unsigned long ulHeight;             /**< Coded Sequence Height */
130     unsigned long ulNumDecodeSurfaces;  /**< Maximum number of internal decode surfaces */
131     cudaVideoCodec CodecType;           /**< cudaVideoCodec_XXX */
132     cudaVideoChromaFormat ChromaFormat; /**< cudaVideoChromaFormat_XXX (only 4:2:0 is currently supported) */
133     unsigned long ulCreationFlags;      /**< Decoder creation flags (cudaVideoCreateFlags_XXX) */
134     unsigned long bitDepthMinus8;
135     unsigned long Reserved1[4];         /**< Reserved for future use - set to zero */
136     /**
137     * area of the frame that should be displayed
138     */
139     struct {
140         short left;
141         short top;
142         short right;
143         short bottom;
144     } display_area;
145 
146     cudaVideoSurfaceFormat OutputFormat;       /**< cudaVideoSurfaceFormat_XXX */
147     cudaVideoDeinterlaceMode DeinterlaceMode;  /**< cudaVideoDeinterlaceMode_XXX */
148     unsigned long ulTargetWidth;               /**< Post-processed Output Width (Should be aligned to 2) */
149     unsigned long ulTargetHeight;              /**< Post-processed Output Height (Should be aligbed to 2) */
150     unsigned long ulNumOutputSurfaces;         /**< Maximum number of output surfaces simultaneously mapped */
151     CUvideoctxlock vidLock;                    /**< If non-NULL, context lock used for synchronizing ownership of the cuda context */
152     /**
153     * target rectangle in the output frame (for aspect ratio conversion)
154     * if a null rectangle is specified, {0,0,ulTargetWidth,ulTargetHeight} will be used
155     */
156     struct {
157         short left;
158         short top;
159         short right;
160         short bottom;
161     } target_rect;
162     unsigned long Reserved2[5];                /**< Reserved for future use - set to zero */
163 } CUVIDDECODECREATEINFO;
164 
165 /*!
166  * \struct CUVIDH264DPBENTRY
167  * H.264 DPB Entry
168  */
169 typedef struct _CUVIDH264DPBENTRY
170 {
171     int PicIdx;                 /**< picture index of reference frame */
172     int FrameIdx;               /**< frame_num(short-term) or LongTermFrameIdx(long-term) */
173     int is_long_term;           /**< 0=short term reference, 1=long term reference */
174     int not_existing;           /**< non-existing reference frame (corresponding PicIdx should be set to -1) */
175     int used_for_reference;     /**< 0=unused, 1=top_field, 2=bottom_field, 3=both_fields */
176     int FieldOrderCnt[2];       /**< field order count of top and bottom fields */
177 } CUVIDH264DPBENTRY;
178 
179 /*!
180  * \struct CUVIDH264MVCEXT
181  * H.264 MVC Picture Parameters Ext
182  */
183 typedef struct _CUVIDH264MVCEXT
184 {
185     int num_views_minus1;
186     int view_id;
187     unsigned char inter_view_flag;
188     unsigned char num_inter_view_refs_l0;
189     unsigned char num_inter_view_refs_l1;
190     unsigned char MVCReserved8Bits;
191     int InterViewRefsL0[16];
192     int InterViewRefsL1[16];
193 } CUVIDH264MVCEXT;
194 
195 /*!
196  * \struct CUVIDH264SVCEXT
197  * H.264 SVC Picture Parameters Ext
198  */
199 typedef struct _CUVIDH264SVCEXT
200 {
201     unsigned char profile_idc;
202     unsigned char level_idc;
203     unsigned char DQId;
204     unsigned char DQIdMax;
205     unsigned char disable_inter_layer_deblocking_filter_idc;
206     unsigned char ref_layer_chroma_phase_y_plus1;
207     signed char   inter_layer_slice_alpha_c0_offset_div2;
208     signed char   inter_layer_slice_beta_offset_div2;
209 
210     unsigned short DPBEntryValidFlag;
211     unsigned char inter_layer_deblocking_filter_control_present_flag;
212     unsigned char extended_spatial_scalability_idc;
213     unsigned char adaptive_tcoeff_level_prediction_flag;
214     unsigned char slice_header_restriction_flag;
215     unsigned char chroma_phase_x_plus1_flag;
216     unsigned char chroma_phase_y_plus1;
217 
218     unsigned char tcoeff_level_prediction_flag;
219     unsigned char constrained_intra_resampling_flag;
220     unsigned char ref_layer_chroma_phase_x_plus1_flag;
221     unsigned char store_ref_base_pic_flag;
222     unsigned char Reserved8BitsA;
223     unsigned char Reserved8BitsB;
224     // For the 4 scaled_ref_layer_XX fields below,
225     // if (extended_spatial_scalability_idc == 1), SPS field, G.7.3.2.1.4, add prefix "seq_"
226     // if (extended_spatial_scalability_idc == 2), SLH field, G.7.3.3.4,
227     short scaled_ref_layer_left_offset;
228     short scaled_ref_layer_top_offset;
229     short scaled_ref_layer_right_offset;
230     short scaled_ref_layer_bottom_offset;
231     unsigned short Reserved16Bits;
232     struct _CUVIDPICPARAMS *pNextLayer; /**< Points to the picparams for the next layer to be decoded. Linked list ends at the target layer. */
233     int bRefBaseLayer;                  /**< whether to store ref base pic */
234 } CUVIDH264SVCEXT;
235 
236 /*!
237  * \struct CUVIDH264PICPARAMS
238  * H.264 Picture Parameters
239  */
240 typedef struct _CUVIDH264PICPARAMS
241 {
242     // SPS
243     int log2_max_frame_num_minus4;
244     int pic_order_cnt_type;
245     int log2_max_pic_order_cnt_lsb_minus4;
246     int delta_pic_order_always_zero_flag;
247     int frame_mbs_only_flag;
248     int direct_8x8_inference_flag;
249     int num_ref_frames;             // NOTE: shall meet level 4.1 restrictions
250     unsigned char residual_colour_transform_flag;
251     unsigned char bit_depth_luma_minus8;    // Must be 0 (only 8-bit supported)
252     unsigned char bit_depth_chroma_minus8;  // Must be 0 (only 8-bit supported)
253     unsigned char qpprime_y_zero_transform_bypass_flag;
254     // PPS
255     int entropy_coding_mode_flag;
256     int pic_order_present_flag;
257     int num_ref_idx_l0_active_minus1;
258     int num_ref_idx_l1_active_minus1;
259     int weighted_pred_flag;
260     int weighted_bipred_idc;
261     int pic_init_qp_minus26;
262     int deblocking_filter_control_present_flag;
263     int redundant_pic_cnt_present_flag;
264     int transform_8x8_mode_flag;
265     int MbaffFrameFlag;
266     int constrained_intra_pred_flag;
267     int chroma_qp_index_offset;
268     int second_chroma_qp_index_offset;
269     int ref_pic_flag;
270     int frame_num;
271     int CurrFieldOrderCnt[2];
272     // DPB
273     CUVIDH264DPBENTRY dpb[16];          // List of reference frames within the DPB
274     // Quantization Matrices (raster-order)
275     unsigned char WeightScale4x4[6][16];
276     unsigned char WeightScale8x8[2][64];
277     // FMO/ASO
278     unsigned char fmo_aso_enable;
279     unsigned char num_slice_groups_minus1;
280     unsigned char slice_group_map_type;
281     signed char pic_init_qs_minus26;
282     unsigned int slice_group_change_rate_minus1;
283     union
284     {
285         unsigned long long slice_group_map_addr;
286         const unsigned char *pMb2SliceGroupMap;
287     } fmo;
288     unsigned int  Reserved[12];
289     // SVC/MVC
290     union
291     {
292         CUVIDH264MVCEXT mvcext;
293         CUVIDH264SVCEXT svcext;
294     } svcmvc;
295 } CUVIDH264PICPARAMS;
296 
297 
298 /*!
299  * \struct CUVIDMPEG2PICPARAMS
300  * MPEG-2 Picture Parameters
301  */
302 typedef struct _CUVIDMPEG2PICPARAMS
303 {
304     int ForwardRefIdx;          // Picture index of forward reference (P/B-frames)
305     int BackwardRefIdx;         // Picture index of backward reference (B-frames)
306     int picture_coding_type;
307     int full_pel_forward_vector;
308     int full_pel_backward_vector;
309     int f_code[2][2];
310     int intra_dc_precision;
311     int frame_pred_frame_dct;
312     int concealment_motion_vectors;
313     int q_scale_type;
314     int intra_vlc_format;
315     int alternate_scan;
316     int top_field_first;
317     // Quantization matrices (raster order)
318     unsigned char QuantMatrixIntra[64];
319     unsigned char QuantMatrixInter[64];
320 } CUVIDMPEG2PICPARAMS;
321 
322 ////////////////////////////////////////////////////////////////////////////////////////////////
323 //
324 // MPEG-4 Picture Parameters
325 //
326 
327 // MPEG-4 has VOP types instead of Picture types
328 #define I_VOP 0
329 #define P_VOP 1
330 #define B_VOP 2
331 #define S_VOP 3
332 
333 /*!
334  * \struct CUVIDMPEG4PICPARAMS
335  * MPEG-4 Picture Parameters
336  */
337 typedef struct _CUVIDMPEG4PICPARAMS
338 {
339     int ForwardRefIdx;          // Picture index of forward reference (P/B-frames)
340     int BackwardRefIdx;         // Picture index of backward reference (B-frames)
341     // VOL
342     int video_object_layer_width;
343     int video_object_layer_height;
344     int vop_time_increment_bitcount;
345     int top_field_first;
346     int resync_marker_disable;
347     int quant_type;
348     int quarter_sample;
349     int short_video_header;
350     int divx_flags;
351     // VOP
352     int vop_coding_type;
353     int vop_coded;
354     int vop_rounding_type;
355     int alternate_vertical_scan_flag;
356     int interlaced;
357     int vop_fcode_forward;
358     int vop_fcode_backward;
359     int trd[2];
360     int trb[2];
361     // Quantization matrices (raster order)
362     unsigned char QuantMatrixIntra[64];
363     unsigned char QuantMatrixInter[64];
364     int gmc_enabled;
365 } CUVIDMPEG4PICPARAMS;
366 
367 /*!
368  * \struct CUVIDVC1PICPARAMS
369  * VC1 Picture Parameters
370  */
371 typedef struct _CUVIDVC1PICPARAMS
372 {
373     int ForwardRefIdx;      /**< Picture index of forward reference (P/B-frames) */
374     int BackwardRefIdx;     /**< Picture index of backward reference (B-frames) */
375     int FrameWidth;         /**< Actual frame width */
376     int FrameHeight;        /**< Actual frame height */
377     // PICTURE
378     int intra_pic_flag;     /**< Set to 1 for I,BI frames */
379     int ref_pic_flag;       /**< Set to 1 for I,P frames */
380     int progressive_fcm;    /**< Progressive frame */
381     // SEQUENCE
382     int profile;
383     int postprocflag;
384     int pulldown;
385     int interlace;
386     int tfcntrflag;
387     int finterpflag;
388     int psf;
389     int multires;
390     int syncmarker;
391     int rangered;
392     int maxbframes;
393     // ENTRYPOINT
394     int panscan_flag;
395     int refdist_flag;
396     int extended_mv;
397     int dquant;
398     int vstransform;
399     int loopfilter;
400     int fastuvmc;
401     int overlap;
402     int quantizer;
403     int extended_dmv;
404     int range_mapy_flag;
405     int range_mapy;
406     int range_mapuv_flag;
407     int range_mapuv;
408     int rangeredfrm;    // range reduction state
409 } CUVIDVC1PICPARAMS;
410 
411 /*!
412  * \struct CUVIDJPEGPICPARAMS
413  * JPEG Picture Parameters
414  */
415 typedef struct _CUVIDJPEGPICPARAMS
416 {
417     int Reserved;
418 } CUVIDJPEGPICPARAMS;
419 
420 
421  /*!
422  * \struct CUVIDHEVCPICPARAMS
423  * HEVC Picture Parameters
424  */
425 typedef struct _CUVIDHEVCPICPARAMS
426 {
427     // sps
428     int pic_width_in_luma_samples;
429     int pic_height_in_luma_samples;
430     unsigned char log2_min_luma_coding_block_size_minus3;
431     unsigned char log2_diff_max_min_luma_coding_block_size;
432     unsigned char log2_min_transform_block_size_minus2;
433     unsigned char log2_diff_max_min_transform_block_size;
434     unsigned char pcm_enabled_flag;
435     unsigned char log2_min_pcm_luma_coding_block_size_minus3;
436     unsigned char log2_diff_max_min_pcm_luma_coding_block_size;
437     unsigned char pcm_sample_bit_depth_luma_minus1;
438 
439     unsigned char pcm_sample_bit_depth_chroma_minus1;
440     unsigned char pcm_loop_filter_disabled_flag;
441     unsigned char strong_intra_smoothing_enabled_flag;
442     unsigned char max_transform_hierarchy_depth_intra;
443     unsigned char max_transform_hierarchy_depth_inter;
444     unsigned char amp_enabled_flag;
445     unsigned char separate_colour_plane_flag;
446     unsigned char log2_max_pic_order_cnt_lsb_minus4;
447 
448     unsigned char num_short_term_ref_pic_sets;
449     unsigned char long_term_ref_pics_present_flag;
450     unsigned char num_long_term_ref_pics_sps;
451     unsigned char sps_temporal_mvp_enabled_flag;
452     unsigned char sample_adaptive_offset_enabled_flag;
453     unsigned char scaling_list_enable_flag;
454     unsigned char IrapPicFlag;
455     unsigned char IdrPicFlag;
456 
457     unsigned char bit_depth_luma_minus8;
458     unsigned char bit_depth_chroma_minus8;
459     unsigned char reserved1[14];
460 
461     // pps
462     unsigned char dependent_slice_segments_enabled_flag;
463     unsigned char slice_segment_header_extension_present_flag;
464     unsigned char sign_data_hiding_enabled_flag;
465     unsigned char cu_qp_delta_enabled_flag;
466     unsigned char diff_cu_qp_delta_depth;
467     signed char init_qp_minus26;
468     signed char pps_cb_qp_offset;
469     signed char pps_cr_qp_offset;
470 
471     unsigned char constrained_intra_pred_flag;
472     unsigned char weighted_pred_flag;
473     unsigned char weighted_bipred_flag;
474     unsigned char transform_skip_enabled_flag;
475     unsigned char transquant_bypass_enabled_flag;
476     unsigned char entropy_coding_sync_enabled_flag;
477     unsigned char log2_parallel_merge_level_minus2;
478     unsigned char num_extra_slice_header_bits;
479 
480     unsigned char loop_filter_across_tiles_enabled_flag;
481     unsigned char loop_filter_across_slices_enabled_flag;
482     unsigned char output_flag_present_flag;
483     unsigned char num_ref_idx_l0_default_active_minus1;
484     unsigned char num_ref_idx_l1_default_active_minus1;
485     unsigned char lists_modification_present_flag;
486     unsigned char cabac_init_present_flag;
487     unsigned char pps_slice_chroma_qp_offsets_present_flag;
488 
489     unsigned char deblocking_filter_override_enabled_flag;
490     unsigned char pps_deblocking_filter_disabled_flag;
491     signed char pps_beta_offset_div2;
492     signed char pps_tc_offset_div2;
493     unsigned char tiles_enabled_flag;
494     unsigned char uniform_spacing_flag;
495     unsigned char num_tile_columns_minus1;
496     unsigned char num_tile_rows_minus1;
497 
498     unsigned short column_width_minus1[21];
499     unsigned short row_height_minus1[21];
500     unsigned int reserved3[15];
501 
502     // RefPicSets
503     int NumBitsForShortTermRPSInSlice;
504     int NumDeltaPocsOfRefRpsIdx;
505     int NumPocTotalCurr;
506     int NumPocStCurrBefore;
507     int NumPocStCurrAfter;
508     int NumPocLtCurr;
509     int CurrPicOrderCntVal;
510     int RefPicIdx[16];                  // [refpic] Indices of valid reference pictures (-1 if unused for reference)
511     int PicOrderCntVal[16];             // [refpic]
512     unsigned char IsLongTerm[16];       // [refpic] 0=not a long-term reference, 1=long-term reference
513     unsigned char RefPicSetStCurrBefore[8]; // [0..NumPocStCurrBefore-1] -> refpic (0..15)
514     unsigned char RefPicSetStCurrAfter[8];  // [0..NumPocStCurrAfter-1] -> refpic (0..15)
515     unsigned char RefPicSetLtCurr[8];       // [0..NumPocLtCurr-1] -> refpic (0..15)
516     unsigned char RefPicSetInterLayer0[8];
517     unsigned char RefPicSetInterLayer1[8];
518     unsigned int reserved4[12];
519 
520     // scaling lists (diag order)
521     unsigned char ScalingList4x4[6][16];       // [matrixId][i]
522     unsigned char ScalingList8x8[6][64];       // [matrixId][i]
523     unsigned char ScalingList16x16[6][64];     // [matrixId][i]
524     unsigned char ScalingList32x32[2][64];     // [matrixId][i]
525     unsigned char ScalingListDCCoeff16x16[6];  // [matrixId]
526     unsigned char ScalingListDCCoeff32x32[2];  // [matrixId]
527 } CUVIDHEVCPICPARAMS;
528 
529 
530 /*!
531  * \struct CUVIDVP8PICPARAMS
532  * VP8 Picture Parameters
533  */
534 typedef struct _CUVIDVP8PICPARAMS
535 {
536     int width;
537     int height;
538     unsigned int first_partition_size;
539     //Frame Indexes
540     unsigned char LastRefIdx;
541     unsigned char GoldenRefIdx;
542     unsigned char AltRefIdx;
543     union {
544         struct {
545             unsigned char frame_type : 1;    /**< 0 = KEYFRAME, 1 = INTERFRAME  */
546             unsigned char version : 3;
547             unsigned char show_frame : 1;
548             unsigned char update_mb_segmentation_data : 1;    /**< Must be 0 if segmentation is not enabled */
549             unsigned char Reserved2Bits : 2;
550         };
551         unsigned char wFrameTagFlags;
552     } tagflags;
553     unsigned char Reserved1[4];
554     unsigned int  Reserved2[3];
555 } CUVIDVP8PICPARAMS;
556 
557 /*!
558  * \struct CUVIDVP9PICPARAMS
559  * VP9 Picture Parameters
560  */
561 typedef struct _CUVIDVP9PICPARAMS
562 {
563     unsigned int width;
564     unsigned int height;
565 
566     //Frame Indices
567     unsigned char LastRefIdx;
568     unsigned char GoldenRefIdx;
569     unsigned char AltRefIdx;
570     unsigned char colorSpace;
571 
572     unsigned short profile : 3;
573     unsigned short frameContextIdx : 2;
574     unsigned short frameType : 1;
575     unsigned short showFrame : 1;
576     unsigned short errorResilient : 1;
577     unsigned short frameParallelDecoding : 1;
578     unsigned short subSamplingX : 1;
579     unsigned short subSamplingY : 1;
580     unsigned short intraOnly : 1;
581     unsigned short allow_high_precision_mv : 1;
582     unsigned short refreshEntropyProbs : 1;
583     unsigned short reserved2Bits : 2;
584 
585     unsigned short reserved16Bits;
586 
587     unsigned char  refFrameSignBias[4];
588 
589     unsigned char bitDepthMinus8Luma;
590     unsigned char bitDepthMinus8Chroma;
591     unsigned char loopFilterLevel;
592     unsigned char loopFilterSharpness;
593 
594     unsigned char modeRefLfEnabled;
595     unsigned char log2_tile_columns;
596     unsigned char log2_tile_rows;
597 
598     unsigned char segmentEnabled : 1;
599     unsigned char segmentMapUpdate : 1;
600     unsigned char segmentMapTemporalUpdate : 1;
601     unsigned char segmentFeatureMode : 1;
602     unsigned char reserved4Bits : 4;
603 
604 
605     unsigned char segmentFeatureEnable[8][4];
606     short segmentFeatureData[8][4];
607     unsigned char mb_segment_tree_probs[7];
608     unsigned char segment_pred_probs[3];
609     unsigned char reservedSegment16Bits[2];
610 
611     int qpYAc;
612     int qpYDc;
613     int qpChDc;
614     int qpChAc;
615 
616     unsigned int activeRefIdx[3];
617     unsigned int resetFrameContext;
618     unsigned int mcomp_filter_type;
619     unsigned int mbRefLfDelta[4];
620     unsigned int mbModeLfDelta[2];
621     unsigned int frameTagSize;
622     unsigned int offsetToDctParts;
623     unsigned int reserved128Bits[4];
624 
625 } CUVIDVP9PICPARAMS;
626 
627 
628 /*!
629  * \struct CUVIDPICPARAMS
630  * Picture Parameters for Decoding
631  */
632 typedef struct _CUVIDPICPARAMS
633 {
634     int PicWidthInMbs;                    /**< Coded Frame Size */
635     int FrameHeightInMbs;                 /**< Coded Frame Height */
636     int CurrPicIdx;                       /**< Output index of the current picture */
637     int field_pic_flag;                   /**< 0=frame picture, 1=field picture */
638     int bottom_field_flag;                /**< 0=top field, 1=bottom field (ignored if field_pic_flag=0) */
639     int second_field;                     /**< Second field of a complementary field pair */
640     // Bitstream data
641     unsigned int nBitstreamDataLen;        /**< Number of bytes in bitstream data buffer */
642     const unsigned char *pBitstreamData;   /**< Ptr to bitstream data for this picture (slice-layer) */
643     unsigned int nNumSlices;               /**< Number of slices in this picture */
644     const unsigned int *pSliceDataOffsets; /**< nNumSlices entries, contains offset of each slice within the bitstream data buffer */
645     int ref_pic_flag;                      /**< This picture is a reference picture */
646     int intra_pic_flag;                    /**< This picture is entirely intra coded */
647     unsigned int Reserved[30];             /**< Reserved for future use */
648     // Codec-specific data
649     union {
650         CUVIDMPEG2PICPARAMS mpeg2;         /**< Also used for MPEG-1 */
651         CUVIDH264PICPARAMS h264;
652         CUVIDVC1PICPARAMS vc1;
653         CUVIDMPEG4PICPARAMS mpeg4;
654         CUVIDJPEGPICPARAMS jpeg;
655         CUVIDHEVCPICPARAMS hevc;
656         CUVIDVP8PICPARAMS vp8;
657         CUVIDVP9PICPARAMS vp9;
658         unsigned int CodecReserved[1024];
659     } CodecSpecific;
660 } CUVIDPICPARAMS;
661 
662 
663 /*!
664  * \struct CUVIDPROCPARAMS
665  * Picture Parameters for Postprocessing
666  */
667 typedef struct _CUVIDPROCPARAMS
668 {
669     int progressive_frame;  /**< Input is progressive (deinterlace_mode will be ignored)  */
670     int second_field;       /**< Output the second field (ignored if deinterlace mode is Weave) */
671     int top_field_first;    /**< Input frame is top field first (1st field is top, 2nd field is bottom) */
672     int unpaired_field;     /**< Input only contains one field (2nd field is invalid) */
673     // The fields below are used for raw YUV input
674     unsigned int reserved_flags;        /**< Reserved for future use (set to zero) */
675     unsigned int reserved_zero;         /**< Reserved (set to zero) */
676     unsigned long long raw_input_dptr;  /**< Input CUdeviceptr for raw YUV extensions */
677     unsigned int raw_input_pitch;       /**< pitch in bytes of raw YUV input (should be aligned appropriately) */
678     unsigned int raw_input_format;      /**< Reserved for future use (set to zero) */
679     unsigned long long raw_output_dptr; /**< Reserved for future use (set to zero) */
680     unsigned int raw_output_pitch;      /**< Reserved for future use (set to zero) */
681     unsigned int Reserved[48];
682     void *Reserved3[3];
683 } CUVIDPROCPARAMS;
684 
685 
686 /**
687  *
688  * In order to minimize decode latencies, there should be always at least 2 pictures in the decode
689  * queue at any time, in order to make sure that all decode engines are always busy.
690  *
691  * Overall data flow:
692  *  - cuvidCreateDecoder(...)
693  *  For each picture:
694  *  - cuvidDecodePicture(N)
695  *  - cuvidMapVideoFrame(N-4)
696  *  - do some processing in cuda
697  *  - cuvidUnmapVideoFrame(N-4)
698  *  - cuvidDecodePicture(N+1)
699  *  - cuvidMapVideoFrame(N-3)
700  *    ...
701  *  - cuvidDestroyDecoder(...)
702  *
703  * NOTE:
704  * - When the cuda context is created from a D3D device, the D3D device must also be created
705  *   with the D3DCREATE_MULTITHREADED flag.
706  * - There is a limit to how many pictures can be mapped simultaneously (ulNumOutputSurfaces)
707  * - cuVidDecodePicture may block the calling thread if there are too many pictures pending
708  *   in the decode queue
709  */
710 
711 /**
712  *
713  * Context-locking: to facilitate multi-threaded implementations, the following 4 functions
714  * provide a simple mutex-style host synchronization. If a non-NULL context is specified
715  * in CUVIDDECODECREATEINFO, the codec library will acquire the mutex associated with the given
716  * context before making any cuda calls.
717  * A multi-threaded application could create a lock associated with a context handle so that
718  * multiple threads can safely share the same cuda context:
719  *  - use cuCtxPopCurrent immediately after context creation in order to create a 'floating' context
720  *    that can be passed to cuvidCtxLockCreate.
721  *  - When using a floating context, all cuda calls should only be made within a cuvidCtxLock/cuvidCtxUnlock section.
722  *
723  * NOTE: This is a safer alternative to cuCtxPushCurrent and cuCtxPopCurrent, and is not related to video
724  * decoder in any way (implemented as a critical section associated with cuCtx{Push|Pop}Current calls).
725 */
726 /** @} */  /* End VIDEO_DECODER */
727 
728 #if defined(__cplusplus)
729 }
730 #endif /* __cplusplus */
731 
732 #endif // __CUDA_VIDEO_H__
733