| 1 |
/* |
| 2 |
* SPDX-FileCopyrightText: Copyright (c) 1993-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved. |
| 3 |
* SPDX-License-Identifier: MIT |
| 4 |
* |
| 5 |
* Permission is hereby granted, free of charge, to any person obtaining a |
| 6 |
* copy of this software and associated documentation files (the "Software"), |
| 7 |
* to deal in the Software without restriction, including without limitation |
| 8 |
* the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 9 |
* and/or sell copies of the Software, and to permit persons to whom the |
| 10 |
* Software is furnished to do so, subject to the following conditions: |
| 11 |
* |
| 12 |
* The above copyright notice and this permission notice shall be included in |
| 13 |
* all copies or substantial portions of the Software. |
| 14 |
* |
| 15 |
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 16 |
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 17 |
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 18 |
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 19 |
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
| 20 |
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER |
| 21 |
* DEALINGS IN THE SOFTWARE. |
| 22 |
*/ |
| 23 |
|
| 24 |
#ifndef NVENC_DRV_H |
| 25 |
#define NVENC_DRV_H |
| 26 |
// |
| 27 |
// CLASS NV_C9B7_NVENC |
| 28 |
// =================== |
| 29 |
// |
| 30 |
// |
| 31 |
// 1 - INTRODUCTION |
| 32 |
// |
| 33 |
// The NVENC is a video encoder engine. Currently only H.264 baseline, main |
| 34 |
// and high profile are supported. For baseline, only 1 slicegroup is allowed. |
| 35 |
// NVENC has built-in motion estimation that enables it to do encode on its own |
| 36 |
// but also has the ability to work with external hints that could be produced |
| 37 |
// by another motion search implementation such as CEA.<br> |
| 38 |
// NVENC motion search is divided into a simple integer search module and a high |
| 39 |
// perf subpel refinement module. The integer search capability of NVENC is |
| 40 |
// limited (perf wise) so for higher perf the CEA approach should be used. |
| 41 |
// Motion search/estimation (ME) results in a set of vectors with scores |
| 42 |
// which it will evaluate for encoding cost and eligibility. It also searches |
| 43 |
// intra encoding modes for the lowest cost encoding, picks a final winner |
| 44 |
// from all inter and intra modes that have been evaluated and encodes the |
| 45 |
// macroblock according to that mode. Skip and direct modes are also checked. |
| 46 |
// Search and encoding involve forward dct and quantization, inverse dct and |
| 47 |
// quantization, reconstruction, and optionally deblocking. Entropy encode can |
| 48 |
// do either CAVLC or CABAC. |
| 49 |
// |
| 50 |
// |
| 51 |
// 1.4. CLASS OPERATION |
| 52 |
// ------------------------ |
| 53 |
// |
| 54 |
// *** TBD *** |
| 55 |
// |
| 56 |
// If MVC is enabled, HW need to write the prefix NAL unit(14) when current view is a base view. |
| 57 |
// As SVC is not supported in nvenc, so svc_extension_flag should be always false. Then, in the prefix NAL unit, |
| 58 |
// only the syntax elements(H.7.3.1.1 NAL unit header MVC extension syntax) in the prefix NAL unit should be written: |
| 59 |
// nal_unit_header_mvc_extension( ) { |
| 60 |
// non_idr_flag |
| 61 |
// priority_id |
| 62 |
// view_id |
| 63 |
// temporal_id |
| 64 |
// anchor_pic_flag |
| 65 |
// inter_view_flag |
| 66 |
// reserved_one_bit |
| 67 |
// } |
| 68 |
// |
| 69 |
// 1.5. RESTRICTIONS |
| 70 |
// ------------------------ |
| 71 |
// |
| 72 |
// (1) Reference pictures to be accessed by NVENC must be in tiled 16x16 format |
| 73 |
// (2) Input pictures are either all in BL format or all in tiled 16x16 format |
| 74 |
// (3) Input pictures and reference pictures must be of the kind specified by |
| 75 |
// 'pic_struct' (either fields or frames). NVENC cannot encode an interlaced |
| 76 |
// input picture (2 fields) as a frame or vice versa, and cannot use frame |
| 77 |
// reference pictures when encoding a field picture, or field reference |
| 78 |
// pictures when encoding a frame. |
| 79 |
// |
| 80 |
// |
| 81 |
// |
| 82 |
// 2 - APPLICATION MEMORY STRUCTURES |
| 83 |
// |
| 84 |
// 2.1. DRV PICTURE SETUP BUFFER |
| 85 |
// ----------------------------- |
| 86 |
// |
| 87 |
// DrvPictureSetupBuffer contains sequence, picture, and slice level control |
| 88 |
// data required for encoding. |
| 89 |
// |
| 90 |
// Input pictures (to be encoded) can be any size (HxV) but the encoded output |
| 91 |
// is always an integer number of macroblocks high and wide, which means that |
| 92 |
// H and V are rounded up to the next multiple of 16 (H'xV') and all pixels |
| 93 |
// within this range are used to encode the macroblocks. NVENC does not clear, |
| 94 |
// pad, duplicate or otherwise modify any pixels values within the H'xV' sized |
| 95 |
// input picture contained in the input surface. |
| 96 |
// The values for frame/pic_width_minus1 and frame/pic_height_minus1 contained |
| 97 |
// in the config structures are the original width and height (display size) |
| 98 |
// of the pictures, and from this the nvenc app will derive the cropping |
| 99 |
// information. Cropping will only be done on the right and bottom side of the |
| 100 |
// picture. |
| 101 |
|
| 102 |
typedef struct |
| 103 |
{ |
| 104 |
unsigned int luma_log2_weight_denom : 3; // the base 2 logarithm of the denominator for all luma weighting factors. The value of luma_log2_weight_denom shall be in the range of 0 to 7, inclusive |
| 105 |
unsigned int chroma_log2_weight_denom : 3; // the base 2 logarithm of the denominator for all chroma weighting factors. The value of chroma_log2_weight_denom shall be in the range of 0 to 7, inclusive. |
| 106 |
unsigned int reversed : 26; |
| 107 |
|
| 108 |
unsigned char luma_weight_l0_flag[16]; // equal to 1 specifies that weighting factors for the luma component of list 0 prediction are present, luma_weight_l0_flag[8~15] not used now |
| 109 |
unsigned char chroma_weight_l0_flag[16]; // equal to 1 specifies that weighting factors for the chroma prediction values of list 0 prediction are present, chroma_weight_l0_flag[8~15] not used now |
| 110 |
|
| 111 |
unsigned char luma_weight_l1_flag[16]; // have the same semantics as luma_weight_l0_flag, chroma_weight_l0_flag, not used now |
| 112 |
unsigned char chroma_weight_l1_flag[16]; // not used now |
| 113 |
|
| 114 |
short luma_weight_l0[16]; // is the weighting factor applied to the luma prediction value for list 0 prediction using RefPicList0[ i ]. When luma_weight_l0_flag is equal to 1, the value of luma_weight_l0[ i ] shall be in the range of -128 to 127, inclusive. -128 to 255 for H265 |
| 115 |
short luma_offset_l0[16]; // is the additive offset applied to the luma prediction value for list 0 prediction using RefPicList0[ i ]. The value of luma_offset_l0[ i ] shall be in the range of -128 to 127, inclusive. -2^9 to 2^9-1 for h265 |
| 116 |
short luma_weight_l1[16]; // is the weighting factor applied to the luma prediction value for list 0 prediction using RefPicList0[ i ]. When luma_weight_l0_flag is equal to 1, the value of luma_weight_l0[ i ] shall be in the range of -128 to 127, inclusive. -128 to 255 for H265 |
| 117 |
short luma_offset_l1[16]; // is the additive offset applied to the luma prediction value for list 0 prediction using RefPicList0[ i ]. The value of luma_offset_l0[ i ] shall be in the range of -128 to 127, inclusive. -2^9 to 2^9-1 for h265 |
| 118 |
short chroma_weight_l0[2][16]; // has the similar meaning with luma_weight/offset |
| 119 |
short chroma_offset_l0[2][16]; // has the similar meaning with luma_weight/offset |
| 120 |
short chroma_weight_l1[2][16]; // has the similar meaning with luma_weight/offset |
| 121 |
short chroma_offset_l1[2][16]; // has the similar meaning with luma_weight/offset |
| 122 |
|
| 123 |
} nvenc_pred_weight_table_s; // 452 bytes |
| 124 |
|
| 125 |
typedef struct |
| 126 |
{ |
| 127 |
unsigned short frame_width_minus1; // frame width in pixels minus 1, range 0-4095 |
| 128 |
unsigned short frame_height_minus1; // frame height in pixels minus 1, range 0-4095 |
| 129 |
unsigned short sfc_pitch; // pitch of luma plane |
| 130 |
unsigned short sfc_pitch_chroma; // pitch of chroma plane |
| 131 |
unsigned short sfc_trans_mode; // least 3 significient bits are used to stand for 8 modes, normal, xflip, yflip etc. |
| 132 |
unsigned short reserved2; // pad to int |
| 133 |
|
| 134 |
// do not use offset from cmod trace as golden, cmod behavior is different. |
| 135 |
unsigned int luma_top_frm_offset; // offset of luma top field or frame in units of 256 |
| 136 |
unsigned int luma_bot_offset; // offset of luma bottom field in units of 256. Not used if frame format. |
| 137 |
unsigned int chroma_top_frm_offset; // offset of chroma top field or frame, or offset of chroma U(Cb) plane in planar mode, both in units of 256 |
| 138 |
unsigned int chroma_bot_offset; // offset of chroma bottom field, not used if frame format. Or offset of chroma V(Cr) plane in planar mode. Both in units of 256. |
| 139 |
unsigned int block_height : 7; // BL mapping block height setting |
| 140 |
unsigned int tiled_16x16 : 1; // Surface is 16x16 tiled instead of BL mapped (must be 1 for refpics) |
| 141 |
unsigned int memory_mode : 2; // 0: semi-planar, 1: planar |
| 142 |
unsigned int nv21_enable : 1; // the surface format is yuv or yvu |
| 143 |
unsigned int input_bl_mode : 2; // the input block linear mode: 0~gpu bl; 1~tegra bl; 2~naive bl this is only for cmod and cmod/plugin |
| 144 |
unsigned int reserved : 19; // pad to NvU32 |
| 145 |
} nvenc_h264_surface_cfg_s; // 32 bytes |
| 146 |
|
| 147 |
typedef struct |
| 148 |
{ |
| 149 |
unsigned int profile_idc : 8; // seq_parameters(_ext) regs |
| 150 |
unsigned int level_idc : 8; |
| 151 |
unsigned int chroma_format_idc : 2; // 0=monochrome, 1=yuv4:2:0, 2-3=not supported |
| 152 |
unsigned int pic_order_cnt_type : 2; // only support values 0 and 2 |
| 153 |
unsigned int log2_max_frame_num_minus4 : 4; |
| 154 |
unsigned int log2_max_pic_order_cnt_lsb_minus4 : 4; |
| 155 |
unsigned int frame_mbs_only : 1; |
| 156 |
//mvc |
| 157 |
unsigned int stereo_mvc_enable : 1; //if this bit is enabled, the profile_idc should be 128 |
| 158 |
unsigned int separate_colour_plane_flag : 1; //if this bit is enabled, the chroma_format_idc should be 3 |
| 159 |
unsigned int lossless_qpprime_flag : 1; //if this bit is enabled and QP=0, bypass trans/q and encoded as lossless MB (hori or vert) |
| 160 |
} nvenc_h264_sps_data_s; // 4 bytes |
| 161 |
|
| 162 |
typedef struct |
| 163 |
{ |
| 164 |
unsigned int pic_param_set_id : 8; // picture parameter set identification |
| 165 |
unsigned int entropy_coding_mode_flag : 1; // select entropy copding mode: cabac, cavlc |
| 166 |
unsigned int num_ref_idx_l0_active_minus1 : 5; // number of currently active reference pictures in list 0 minus 1, range 0..30 |
| 167 |
unsigned int num_ref_idx_l1_active_minus1 : 5; // number of currently active reference pictures in list 1 minus 1, range 0..30 |
| 168 |
unsigned int weighted_bipred_idc : 2; // weighted prediction mode for B: only 0 (default) and 2 (implicit) are supported |
| 169 |
int pic_init_qp_minus26 : 6; // initial QP value, range -26..+25 |
| 170 |
int chroma_qp_index_offset : 5; // offset to add to chroma QPy for QPc table indexing for Cb, range -12..+12 |
| 171 |
|
| 172 |
int second_chroma_qp_index_offset : 5; // offset to add to chroma QPy for QPc table indexing for Cr, range -12..+12 |
| 173 |
unsigned int constrained_intra_pred_flag : 1; // if set, intra prediction can only use pixels from macroblocks that are also intra coded |
| 174 |
unsigned int deblocking_filter_control_present_flag : 1;// if set, deblock filter controls syntax elements will be present/encoded in the stream |
| 175 |
unsigned int transform_8x8_mode_flag : 1; // if set, enables the use of transform size 8x8 (flags will be encoded in stream per macroblock) |
| 176 |
unsigned int pic_order_present_flag : 1; // if set, pic_order control info is encoded in the stream |
| 177 |
unsigned int weighted_pred_flag : 1; |
| 178 |
unsigned int reserved0 : 22; // pad to full int |
| 179 |
} nvenc_h264_pps_data_s; // 8 bytes |
| 180 |
|
| 181 |
typedef struct |
| 182 |
{ |
| 183 |
unsigned int self_temporal_stamp_l0 : 3; // stamp to use for L0 integer search in stamp based mode |
| 184 |
unsigned int self_temporal_stamp_l1 : 3; // stamp to use for L1 integer search in stamp based mode |
| 185 |
unsigned int self_temporal_explicit : 1; // explicitly evaluate these vectors |
| 186 |
unsigned int self_temporal_search : 1; // integer search enable for this hint |
| 187 |
unsigned int self_temporal_refine : 1; // subpel search enable for this hint |
| 188 |
unsigned int self_temporal_enable : 1; // hint enable (enables processing + fetching of data) (only changeable at first slice in picture) |
| 189 |
|
| 190 |
unsigned int coloc_stamp_l0 : 3; // stamp to use for L0 integer search in stamp based mode |
| 191 |
unsigned int coloc_stamp_l1 : 3; // stamp to use for L1 integer search in stamp based mode |
| 192 |
unsigned int coloc_explicit : 1; // explicitly evaluate these vectors |
| 193 |
unsigned int coloc_search : 1; // integer search enable for this hint |
| 194 |
unsigned int coloc_refine : 1; // subpel search enable for this hint |
| 195 |
unsigned int coloc_enable : 1; // hint enable (enables processing + fetching of data) (only changeable at first slice in picture) |
| 196 |
|
| 197 |
unsigned int self_spatial_stamp_l0 : 3; // stamp to use for L0 integer search in stamp based mode |
| 198 |
unsigned int self_spatial_stamp_l1 : 3; // stamp to use for L1 integer search in stamp based mode |
| 199 |
unsigned int self_spatial_explicit : 1; // explicitly evaluate these vectors |
| 200 |
unsigned int self_spatial_search : 1; // integer search enable for this hint |
| 201 |
unsigned int self_spatial_refine : 1; // subpel search enable for this hint |
| 202 |
unsigned int self_spatial_enable : 1; // hint enable (enables processing + fetching of data) (only changeable at first slice in picture) |
| 203 |
unsigned int reserved0 : 2; // pad to full int |
| 204 |
|
| 205 |
unsigned int external_stamp_l0_refidx0_stamp : 3; // Stamp to use for external L0 hint with refidx = 0 |
| 206 |
unsigned int external_stamp_l0_refidx1_stamp : 3; // Stamp to use for external L0 hint with refidx = 1 |
| 207 |
unsigned int external_stamp_l0_refidx2_stamp : 3; // Stamp to use for external L0 hint with refidx = 2 |
| 208 |
unsigned int external_stamp_l0_refidx3_stamp : 3; // Stamp to use for external L0 hint with refidx = 3 |
| 209 |
unsigned int external_stamp_l0_refidx4_stamp : 3; // Stamp to use for external L0 hint with refidx = 4 |
| 210 |
unsigned int external_stamp_l0_refidx5_stamp : 3; // Stamp to use for external L0 hint with refidx = 5 |
| 211 |
unsigned int external_stamp_l0_refidx6_stamp : 3; // Stamp to use for external L0 hint with refidx = 6 |
| 212 |
unsigned int external_stamp_l0_refidx7_stamp : 3; // Stamp to use for external L0 hint with refidx = 7 |
| 213 |
unsigned int external_stamp_l1_refidx0_stamp : 3; // Stamp to use for external L1 hint with refidx = 0 |
| 214 |
unsigned int external_stamp_l1_refidx1_stamp : 3; // Stamp to use for external L1 hint with refidx = 1 |
| 215 |
unsigned int reserved1 : 2; // pad to full int |
| 216 |
|
| 217 |
|
| 218 |
unsigned int external_stamp_l1_refidx2_stamp : 3; // Stamp to use for external L1 hint with refidx = 2 |
| 219 |
unsigned int external_stamp_l1_refidx3_stamp : 3; // Stamp to use for external L1 hint with refidx = 3 |
| 220 |
unsigned int external_stamp_l1_refidx4_stamp : 3; // Stamp to use for external L1 hint with refidx = 4 |
| 221 |
unsigned int external_stamp_l1_refidx5_stamp : 3; // Stamp to use for external L1 hint with refidx = 5 |
| 222 |
unsigned int external_stamp_l1_refidx6_stamp : 3; // Stamp to use for external L1 hint with refidx = 6 |
| 223 |
unsigned int external_stamp_l1_refidx7_stamp : 3; // Stamp to use for external L1 hint with refidx = 7 |
| 224 |
|
| 225 |
|
| 226 |
unsigned int external_explicit : 1; // explicitly evaluate these vectors |
| 227 |
unsigned int external_search : 1; // integer search enable for this hint |
| 228 |
unsigned int external_refine : 1; // subpel search enable for this hint |
| 229 |
unsigned int external_enable : 1; // hint enable (enables processing + fetching of data) (only changeable at first slice in picture) |
| 230 |
|
| 231 |
unsigned int const_mv_stamp_l0 : 3; // stamp to use for L0 integer search in stamp based mode |
| 232 |
unsigned int const_mv_stamp_l1 : 3; // stamp to use for L1 integer search in stamp based mode |
| 233 |
unsigned int const_mv_explicit : 1; // explicitly evaluate these vectors |
| 234 |
unsigned int const_mv_search : 1; // integer search enable for this hint |
| 235 |
unsigned int const_mv_refine : 1; // subpel search enable for this hint |
| 236 |
unsigned int const_mv_enable : 1; // hint enable (enables processing + fetching of data) |
| 237 |
|
| 238 |
unsigned int stamp_refidx1_stamp : 3; // Stamp to use for multi-ref hint with refidx = 1 |
| 239 |
unsigned int stamp_refidx2_stamp : 3; // Stamp to use for multi-ref hint with refidx = 2 |
| 240 |
unsigned int stamp_refidx3_stamp : 3; // Stamp to use for multi-ref hint with refidx = 3 |
| 241 |
unsigned int stamp_refidx4_stamp : 3; // Stamp to use for multi-ref hint with refidx = 4 |
| 242 |
unsigned int stamp_refidx5_stamp : 3; // Stamp to use for multi-ref hint with refidx = 5 |
| 243 |
unsigned int stamp_refidx6_stamp : 3; // Stamp to use for multi-ref hint with refidx = 6 |
| 244 |
unsigned int stamp_refidx7_stamp : 3; // Stamp to use for multi-ref hint with refidx = 7 |
| 245 |
unsigned int reserved2 : 11; // pad to full int |
| 246 |
|
| 247 |
} nvenc_h264_me_hint_cfg_s; //16 bytes |
| 248 |
|
| 249 |
typedef struct |
| 250 |
{ |
| 251 |
unsigned int mvx_frac : 2; // vector X component fraction |
| 252 |
int mvx_int : 12; // vector X component integer portion |
| 253 |
unsigned int reserved_x : 2; // padding X to 16 bit |
| 254 |
unsigned int mvy_frac : 2; // vector Y component fraction |
| 255 |
int mvy_int : 10; // vector Y component integer portion |
| 256 |
unsigned int reserved_y : 4; // padding Y to 16 bit |
| 257 |
} nvenc_h264_me_const_vec_s; // 4 bytes |
| 258 |
|
| 259 |
typedef struct |
| 260 |
{ |
| 261 |
unsigned int bitmask[2] ; // 8x8 search point pattern bitmask |
| 262 |
unsigned int hor_adjust : 1; // shift hor based on lsb of pred |
| 263 |
unsigned int ver_adjust : 1; // shift ver based on lsb of pred |
| 264 |
unsigned int reserved : 30; |
| 265 |
} nvenc_h264_me_stamp_s; // 12 bytes |
| 266 |
|
| 267 |
typedef struct |
| 268 |
{ |
| 269 |
unsigned int mv_limit_enable : 1; // 0=disable, 1=enable set the absolute MV range (Cisco flip camera request) |
| 270 |
unsigned int left_mvx_frac : 2; // vector X component fraction |
| 271 |
int left_mvx_int : 12; // vector X component integer portion |
| 272 |
unsigned int reserved1 : 1; // padding X to 16 bit |
| 273 |
unsigned int top_mvy_frac : 2; // vector Y component fraction |
| 274 |
int top_mvy_int : 10; // vector Y component integer portion |
| 275 |
unsigned int reserved2 : 4; // padding Y to 16 bit |
| 276 |
|
| 277 |
unsigned int right_mvx_frac : 2; // vector X component fraction |
| 278 |
int right_mvx_int : 12; // vector X component integer portion |
| 279 |
unsigned int reserved3 : 2; // padding X to 16 bit |
| 280 |
unsigned int bottom_mvy_frac : 2; // vector Y component fraction |
| 281 |
int bottom_mvy_int : 10; // vector Y component integer portion |
| 282 |
unsigned int reserved4 : 4; // padding Y to 16 bit |
| 283 |
} nvenc_h264_me_limit_mv_s; // 8 bytes |
| 284 |
|
| 285 |
typedef struct |
| 286 |
{ |
| 287 |
unsigned char teb_K ; // TEB K |
| 288 |
unsigned char teb_N ; // TEB N |
| 289 |
unsigned char teb_P ; // TEB P |
| 290 |
unsigned char teb_S : 4; // TEB S |
| 291 |
unsigned char teb_mode : 4; // TEB mode |
| 292 |
} nvenc_me_tebif_s; |
| 293 |
|
| 294 |
typedef struct |
| 295 |
{ |
| 296 |
unsigned int me_predictor_mode : 1; // 0=MDP, 1=const |
| 297 |
unsigned int refinement_mode : 1; // 0=hpel, 1=qpel |
| 298 |
unsigned int lambda_mode : 1; // 0=MDP, 1=const |
| 299 |
unsigned int const_lambda : 11; // U8.3 format lambda, used if lambda_mode=1 |
| 300 |
unsigned int refine_on_search_enable : 1; // 0=disable, 1=enable subpel refine for search candidates |
| 301 |
unsigned int me_only_mode : 1; |
| 302 |
unsigned int fps_mvcost : 1; |
| 303 |
unsigned int sps_mvcost : 1; |
| 304 |
unsigned int sps_cost_func : 1; |
| 305 |
unsigned int me_icc_mode_mad : 1; |
| 306 |
|
| 307 |
unsigned int sps_filter : 3; |
| 308 |
unsigned int mc_filter : 3; |
| 309 |
unsigned int vc1_fastuv_mc : 1; |
| 310 |
unsigned int vc1_mc_rnd : 1; |
| 311 |
unsigned int mbc_ctrl_arbitor : 2; |
| 312 |
unsigned int mv_only_enable : 1; |
| 313 |
unsigned int average_mvhint_enable : 1; |
| 314 |
|
| 315 |
nvenc_h264_me_limit_mv_s limit_mv; |
| 316 |
nvenc_h264_me_hint_cfg_s predsrc; // predictor sources |
| 317 |
nvenc_h264_me_const_vec_s l0_hint; // constant MV used as L0 hint |
| 318 |
nvenc_h264_me_const_vec_s l1_hint; // constant MV used as L1 hint |
| 319 |
nvenc_h264_me_const_vec_s l0_pred; // constant MV used as L0 predictor |
| 320 |
nvenc_h264_me_const_vec_s l1_pred; // constant MV used as L1 predictor |
| 321 |
nvenc_h264_me_stamp_s shape0; // priority 0 stamping pattern |
| 322 |
nvenc_h264_me_stamp_s shape1; // priority 1 stamping pattern |
| 323 |
nvenc_h264_me_stamp_s shape2; // priority 2 stamping pattern |
| 324 |
nvenc_h264_me_stamp_s shape3; // priority 3 stamping pattern |
| 325 |
nvenc_h264_me_stamp_s shape4; // priority 4 stamping pattern |
| 326 |
nvenc_h264_me_stamp_s shape5; // priority 5 stamping pattern |
| 327 |
nvenc_h264_me_stamp_s shape6; // priority 6 stamping pattern |
| 328 |
nvenc_h264_me_stamp_s shape7; // priority 7 stamping pattern |
| 329 |
nvenc_me_tebif_s teb_para; // TEB parameter |
| 330 |
unsigned int mbc_mb_size : 9; // MBC size in MB |
| 331 |
unsigned int rmvp_source_half_scaled : 1; // only support picture size half scale down in horz&vert |
| 332 |
unsigned int partDecisionMadeByFPP : 1; // enable CU32 FPP; |
| 333 |
unsigned int penalty_factor_FPP : 6; // MV with cost no larger than min_cost + ((min_cost * factor) >> 8) will go to SPS, not used from nvenc7 |
| 334 |
unsigned int spatial_hint_pattern : 4; |
| 335 |
unsigned int temporal_hint_pattern : 6; |
| 336 |
unsigned int Cu16partDecisionMadeByFPP : 1; // enable CU16 making part decision |
| 337 |
unsigned int sps_mvp_from_fbm : 1; //1:sps use fbm'MVP and lambda;0:based on me_predictor_mode and lambda_mode to set the mvp and lambda for sps |
| 338 |
unsigned int fbm_select_best_cu16_parttype_num : 3; // #winner partitions selected for cu16x16. valid range is 0~7; |
| 339 |
unsigned int fbm_op_winner_num_p_frame : 3; // FBM output number of integer winners for each PU. when fbm_op_winner_num_p_frame_cu8 has non-zero value, this field means the PU winners for CU64~CU16, i.e. w/o CU8 |
| 340 |
unsigned int fbm_op_winner_num_b_frame_l0 : 3; // FBM output number of integer winners for each PU |
| 341 |
unsigned int fbm_op_winner_num_b_frame_l1 : 3; // FBM output number of integer winners for each PU |
| 342 |
unsigned int fbm_select_best_cu32_parttype_num : 3; // cu32 select best parttype num |
| 343 |
unsigned int sps_evaluate_merge_cand_num : 3; // number of merg candidate for sps to explicit evaluate |
| 344 |
unsigned int fps_quad_thresh_hold : 10; |
| 345 |
unsigned int external_hint_order : 1; //0:Insert external hint at first ref; 1:insert external hint at last ref; |
| 346 |
unsigned int coloc_hint_order : 1; //0:Insert coloc hint at first ref; 1: insert coloc hint at last ref. |
| 347 |
unsigned int reserved1 : 5; |
| 348 |
short ct_threshold ; // for OFS only, threshold in Census Transform calculation |
| 349 |
//(1<<hint_type0)|(1<<hint_type1)|(1<<hint_type2)|(1<<hint_type3)|(1<<hint_type4) should be equal to 0x1f |
| 350 |
unsigned short hint_type0 : 3; //0:const hint; 1:spatial hint; 2:temporal hint; 3:coloc hint; 4:external hint |
| 351 |
unsigned short hint_type1 : 3; //the same as hint_type0 |
| 352 |
unsigned short hint_type2 : 3; //the same as hint_type0 |
| 353 |
unsigned short hint_type3 : 3; //the same as hint_type0 |
| 354 |
unsigned short hint_type4 : 3; //the same as hint_type0 |
| 355 |
unsigned short pyramidal_hint_order : 1; // for OFS only, used when do pyramidal ME, 1: enable, the hint order will be: if one MB has external hint, use exhint, otherwise use spatial/temproal/const hints. 0: disable. |
| 356 |
unsigned char left_hint_delay_N : 4; // for OFS only, specify left hint delay MB number(1~8), used for high-perf mode (left_hint_delay_N need to set to 8 to achieve 4K@120FPS). set 0 for using default value, ie 3 for ofs/H265, 8 for H264 |
| 357 |
unsigned char reserved3 : 4; |
| 358 |
|
| 359 |
unsigned char cu8partDecisionMadeByFPP : 1; //enable CU8x8 partition selection in FBM |
| 360 |
unsigned char fbm_select_best_cu8_parttype_num : 3; //#winner partitions selected for cu8x8. valid range is 0~7; |
| 361 |
unsigned char fbm_op_winner_num_p_frame_cu8 : 3; // FBM output number of integer winners for each PU of CU size 8. |
| 362 |
unsigned char reserved4 : 1; |
| 363 |
|
| 364 |
unsigned char cu64partDecisionMadeByFPP : 1; //enable CU64x64 partition selection in FBM |
| 365 |
unsigned char fbm_select_best_cu64_parttype_num : 3; //#winner partitions selected for cu64x64. valid range is 0~7; |
| 366 |
unsigned char reserved5 : 4; |
| 367 |
|
| 368 |
unsigned char reserved[33]; // padding to make me_control=192 to allow 3 reads of 64 in ucode |
| 369 |
// this is optimal size as 160=5 reads of 32 will exceed dma queue |
| 370 |
} nvenc_h264_me_control_s; // 192 bytes |
| 371 |
|
| 372 |
typedef struct |
| 373 |
{ |
| 374 |
unsigned short intra_refresh_cnt; // distance between forced-intra MBs in slice; 0 means OFF |
| 375 |
unsigned short intra_refresh_offs; // first forced-intra MB in slice |
| 376 |
|
| 377 |
unsigned int intra_luma4x4_mode_enable : 9; // bitmask indicating which intra luma 4x4 modes to enable |
| 378 |
unsigned int intra_luma8x8_mode_enable : 9; // bitmask indicating which intra luma 8x8 modes to enable |
| 379 |
unsigned int intra_luma16x16_mode_enable : 4; // bitmask indicating which intra luma 16x16 modes to enable |
| 380 |
unsigned int intra_chroma_mode_enable : 4; // bitmask indicating which intra chroma modes to enable |
| 381 |
unsigned int inter_penalty_factor_for_ip1 : 6; // early teminate ip1 intra search if intra cost is larger than inter_cost + ((inter_cost * factor) >> 8) |
| 382 |
|
| 383 |
unsigned int l0_part_16x16_enable : 1; // enable L0 prediction for 16x16 |
| 384 |
unsigned int l0_part_16x8_enable : 1; // enable L0 prediction for 16x8 |
| 385 |
unsigned int l0_part_8x16_enable : 1; // enable L0 prediction for 8x16 |
| 386 |
unsigned int l0_part_8x8_enable : 1; // enable L0 prediction for 8x8 |
| 387 |
unsigned int l0_part_8x4_enable : 1; // enable L0 prediction for 8x4 |
| 388 |
unsigned int l0_part_4x8_enable : 1; // enable L0 prediction for 4x8 |
| 389 |
unsigned int l0_part_4x4_enable : 1; // enable L0 prediction for 4x4 |
| 390 |
unsigned int l0_part_reserved : 1; // reserved for future L0 prediction extensions |
| 391 |
unsigned int l1_part_16x16_enable : 1; // enable L1 prediction for 16x16 |
| 392 |
unsigned int l1_part_16x8_enable : 1; // enable L1 prediction for 16x8 |
| 393 |
unsigned int l1_part_8x16_enable : 1; // enable L1 prediction for 8x16 |
| 394 |
unsigned int l1_part_8x8_enable : 1; // enable L1 prediction for 8x8 |
| 395 |
unsigned int l1_part_reserved : 4; // reserved for future L1 prediction extensions |
| 396 |
unsigned int bi_part_16x16_enable : 1; // enable Bi prediction for 16x16 |
| 397 |
unsigned int bi_part_16x8_enable : 1; // enable Bi prediction for 16x8 |
| 398 |
unsigned int bi_part_8x16_enable : 1; // enable Bi prediction for 8x16 |
| 399 |
unsigned int bi_part_8x8_enable : 1; // enable Bi prediction for 8x8 |
| 400 |
unsigned int bi_part_reserved : 4; // reserved for future Bi prediction extensions |
| 401 |
unsigned int bdirect_mode : 2; // 0: disable, 1: spatial, 2: temporal |
| 402 |
unsigned int bskip_enable : 1; // enable b_skip encoding (requires b_direct encoding) |
| 403 |
unsigned int pskip_enable : 1; // enable p_skip encoding |
| 404 |
unsigned int special_reserved : 4; // reserved for future special prediction extensions |
| 405 |
|
| 406 |
short bias_inter_16x16; // bias for inter 16x16 (subtracts from inter 16x16 cost) |
| 407 |
short bias_inter_16x8; // bias for inter 16x8 (subtracts from inter 16x8 cost) |
| 408 |
|
| 409 |
short bias_inter_8x16; // bias for inter 8x16 (subtracts from inter 8x16 cost) |
| 410 |
short bias_inter_8x8; // bias for inter 8x8 (subtracts from inter 8x8 cost) |
| 411 |
|
| 412 |
short bias_pskip; // bias for skip (subtracts from skip cost) |
| 413 |
short bias_bdir; // bias for bdirect (subtracts from bdirect cost) |
| 414 |
|
| 415 |
short bias_intra_over_inter; // bias for any intra (subtracts from intra cost) |
| 416 |
short bias_intra_16x16; // bias for intra 16x16 (subtracts from intra 16x16 cost) |
| 417 |
|
| 418 |
short bias_intra_8x8; // bias for intra 8x8 (subtracts from intra 8x8 cost) |
| 419 |
short bias_intra_4x4; // bias for intra 4x4 (subtracts from intra 4x4 cost) |
| 420 |
|
| 421 |
short bias_intra_most_prob; // bias for intra most probable mode |
| 422 |
unsigned short mv_cost_bias; // bias applied to internal motion vector cost calculation |
| 423 |
|
| 424 |
unsigned short intra_nxn_bias_multiplier; // 0..255 range; used for internal bias calculation |
| 425 |
unsigned short intra_most_prob_bias_multiplier; // 0..31 range; used for internal bias calculation |
| 426 |
|
| 427 |
short pskip_bias_multiplier; // S16 range; used for internal bias calculation |
| 428 |
short bdirect_bias_multiplier; // S16 range; used for internal bias calculation |
| 429 |
|
| 430 |
unsigned short pskip_esc_threshold; // if pskip cost is below this, disable full search & attempt pskip encode |
| 431 |
unsigned short bdirect_esc_threshold; // if bdirect cost is below this, disable full search & attempt bdirect encode |
| 432 |
|
| 433 |
unsigned short early_intra_disable_mpeb_threshold; // if early intra cost is below this, disable full search & select intra |
| 434 |
unsigned short tempmv_wt_spread_threshold; // The threshold for the sum of differences between 8x8s and 16x16 mvs. Used for temporal mv weight |
| 435 |
unsigned int tempmv_wt_distort_threshold : 16; // The threshold for the cost of the 16x16 mv. Used for temporal mv weight |
| 436 |
unsigned int mv_cost_predictor_control : 1; // 0=use 16x16 predictor only; 1=use accurate predictor (slower but better) |
| 437 |
unsigned int mv_cost_enable : 1; // enable mv cost calculations in mode decision |
| 438 |
unsigned int intra_most_prob_force_on : 1; // always evaluate most probably 4x4 or 8x8 intra mode during full search |
| 439 |
unsigned int early_intra_mode_control : 2; // 0=inter; 1=eval_intra; 2=ext_hint; 3=none (low perf,best qual) |
| 440 |
unsigned int early_intra_mode_type_16x16dc : 1; // evaluate 16x16dc if early_intra_mode_control==1 |
| 441 |
unsigned int early_intra_mode_type_16x16h : 1; // evaluate 16x16h if early_intra_mode_control==1 |
| 442 |
unsigned int early_intra_mode_type_16x16v : 1; // evaluate 16x16v if early_intra_mode_control==1 |
| 443 |
unsigned int early_ip_is_final : 1; // if set, result of early intra is final (disable full intra search) |
| 444 |
unsigned int chroma_eval_mode : 1; // for intra chroma search 0 = U and V; 1 = U only |
| 445 |
unsigned int ip_search_mode : 3; // which intra sizes to search before making IP decision b0=4x4,b1=8x8,b2=16x16 |
| 446 |
unsigned int multiply_bias_with_lambda : 1; // setting this to 1 will result in bias being multiplied by lambda |
| 447 |
unsigned int force_ipcm : 1; // 0=normal encode; 1=force mpeb/mpec encode as IPCM |
| 448 |
unsigned int early_termination_ip1 : 1; // enable IP1 early termination |
| 449 |
short bias_favor_intra_16x16_early; // S16 range; used for internal bias calculation |
| 450 |
unsigned short priority_ipred_type_ip1 : 3; // which intra type search first for early termination, must be a subset of ip_search_mode |
| 451 |
unsigned short intra_ssd_cnt_4x4 : 4; // SSD cost evaluation between N top intra modes selected by satd |
| 452 |
unsigned short intra_ssd_cnt_8x8 : 4; // SSD cost evaluation between N top intra modes selected by satd |
| 453 |
unsigned short intra_ssd_cnt_16x16 : 4; // SSD cost evaluation between N top intra modes selected by satd |
| 454 |
unsigned short skip_evaluate_enable : 1; // 0: not check skip ssd cost; 1, evaluate skip ssd cost |
| 455 |
unsigned int rdo_level : 3; // 0:mode decision by satd 1:mode decision by ssd |
| 456 |
unsigned int tu_search_num : 3; // tu search part num per MB |
| 457 |
unsigned int luma_residual_zero_eval : 2; // luma residual zero cost evaluate, bit0:intra; bit1:inter |
| 458 |
unsigned int num_1div8_lambda_ssd : 7; // number of 1/8 lambda ssd for lambda_coef delta. lambda_coef = (1+num_1div8_lambda_ssd*1/8)*lambda_ssd |
| 459 |
unsigned int ofs_mvx_range : 2; // control mvx precision/range, 0: ouptput format: S10.5, range[-1024, 1023]; 1: output format: S11.4, range[-2048, 2047]; 2: output format: S12.3, range[-4096, 4095]; 3: output format: S12.2, range[-4096, 4095] |
| 460 |
unsigned int reserved4 : 15; // pad to 32 bytes |
| 461 |
short bias_tu_4x4; // bias for TU 4x4 |
| 462 |
short bias_tu_8x8; // bias for TU 8x8 |
| 463 |
unsigned int ssim_rdo : 2; // SSIM-RDO: 0 disable, 1: enable SW SSIM-RDO, 2: enable HW SSIM-RDO, |
| 464 |
unsigned int calc_ssim : 1; // Calculate SSIM distortion: 0 disable, 1: enable. ssim_rdo = 1 => calc_ssim = 1 |
| 465 |
unsigned int reserved5 : 29; // reserved |
| 466 |
unsigned int ssd_over_ssim_factor; // distortion ssd/ssim ratio, in U24.8 format |
| 467 |
unsigned int reserved6[13]; // reserved |
| 468 |
} nvenc_h264_md_control_s; // 128 bytes |
| 469 |
|
| 470 |
typedef struct |
| 471 |
{ |
| 472 |
unsigned short qpp_run_vector_4x4; // cost values for 4x4 transform (16 bit total vector) |
| 473 |
unsigned short qpp_run_vector_8x8[3]; // cost values for 8x8 transform (48 bit total vector) |
| 474 |
// 2 bits each for first 12 coefs, then 1 bit each for next 24 |
| 475 |
unsigned char qpp_luma8x8_cost; // luma 8x8 cost threshold, 0 = throw out all coefs, range 0-15 |
| 476 |
unsigned char qpp_luma16x16_cost; // luma 16x16 cost threshold, 0 = throw out all coefs, range 0-15 |
| 477 |
unsigned char qpp_chroma_cost; // chroma cost threshold, 0 = throw out all coefs, range 0-15 |
| 478 |
unsigned char qpp_mode : 2; // 0 = OFF, 1 = 8x8, 2 = 16x16_8x8 |
| 479 |
unsigned char reserved1 : 4; // padding |
| 480 |
//add by cl |
| 481 |
unsigned char quant_intra_sat_flag : 1; // 0: no saturation is applied; 1: saturation is applied |
| 482 |
unsigned char quant_inter_sat_flag : 1; // 0: no saturation is applied; 1: saturation is applied |
| 483 |
unsigned int quant_intra_sat_limit : 16; // quantization saturation limit for intra MB |
| 484 |
unsigned int quant_inter_sat_limit : 16; // quantization saturation limit for inter MB |
| 485 |
//~ |
| 486 |
unsigned short dz_4x4_YI[16]; // deadzone for 4x4 transform of Luma Intra |
| 487 |
unsigned short dz_4x4_YP[16]; // deadzone for 4x4 transform of Luma Inter |
| 488 |
unsigned short dz_4x4_CI; // deadzone for 4x4 transform of Chroma Intra |
| 489 |
unsigned short dz_4x4_CP; // deadzone for 4x4 transform of Chroma Inter |
| 490 |
unsigned short dz_8x8_YI[16]; // deadzone for 8x8 transform of Luma Intra, 16 values are mapped to 64 coefs |
| 491 |
unsigned short dz_8x8_YP[16]; // deadzone for 8x8 transform of Luma Inter, 16 values are mapped to 64 coefs |
| 492 |
unsigned int reserved2[11]; // pad to 192 bytes to allow 3 reads of 64 bytes in ucode |
| 493 |
// this is optimal size as 160=5 reads of 32 will exceed dma queue |
| 494 |
} nvenc_h264_quant_control_s; // 192 bytes |
| 495 |
|
| 496 |
typedef struct |
| 497 |
{ |
| 498 |
unsigned int reorder_l0_cmd_count : 4; |
| 499 |
unsigned int reorder_l1_cmd_count : 4; |
| 500 |
unsigned int mmco_cmd_count : 4; |
| 501 |
unsigned int no_output_of_prior_pic_flag : 1; |
| 502 |
unsigned int long_term_ref_flag : 1; |
| 503 |
unsigned int reserved :18; |
| 504 |
} nvenc_h264_refpic_cmd_s; // 4 bytes |
| 505 |
|
| 506 |
typedef struct |
| 507 |
{ |
| 508 |
// ref_pic_list_reorder_cmd includes both opcode and op arguments |
| 509 |
unsigned int reordering_of_pic_nums_idc : 3; // Bit 31-29 (0-5 (Table 7-7) |
| 510 |
unsigned int abs_diff_pic_num_minus1 :17; // (opcode=0/1) |
| 511 |
unsigned int long_term_pic_num : 4; // (opcode=2): |
| 512 |
unsigned int abs_diff_view_idx_minus1 : 4; // (opcode=4/5): |
| 513 |
unsigned int reserved : 4; |
| 514 |
} nvenc_h264_ref_pic_reorder_s; // 4 bytes |
| 515 |
|
| 516 |
typedef struct |
| 517 |
{ |
| 518 |
unsigned int mmco_cmd_id : 3; // totally 6 commands |
| 519 |
|
| 520 |
unsigned int difference_of_pic_nums_minus1 :17; // 1) mark a short_term ref as long (mmco=3); 2) mark a short_term as un-used (mmco=1) |
| 521 |
unsigned int long_term_pic_num : 3; // [0 to MaxLongTermFrameIdx ]; mark a long as "unused for ref" (mmco=2); |
| 522 |
unsigned int long_term_frame_idx : 3; // [0 to MaxLongTermFrameIdx] 1) mark a picture to long (mmco=3, mmco=6) |
| 523 |
unsigned int max_long_term_frame_idx_plus1 : 4; // [0, num_ref_frames] max long term frm indexes allowed for long term pictures. |
| 524 |
// max_long_term_frame_idx_plus1=0 -> MaxLongTermFrameIdx="no-long-term" |
| 525 |
// max_long_term_frame_idx_plus1>0 -> MaxLongTermFrameIdx= max_long_term_frame_idx_plus1-1 |
| 526 |
unsigned int reserved : 2; |
| 527 |
} nvenc_h264_mmco_s; // 4 bytes |
| 528 |
|
| 529 |
typedef struct |
| 530 |
{ |
| 531 |
// int0 |
| 532 |
unsigned int num_mb : 19; // number of macroblocks in this slice, support upto 8kx8k |
| 533 |
unsigned int qp_avr : 8; // 6bit value used by hw. Other bits can be used for rounding. |
| 534 |
unsigned int reserved : 5; |
| 535 |
// int1 |
| 536 |
unsigned int slice_tgt_rate; // target bit rate (size) (if RC mode == slice) |
| 537 |
|
| 538 |
// int2 |
| 539 |
char slice_alpha_c0_offset_div2 : 4; |
| 540 |
char slice_beta_offset_div2 : 4; |
| 541 |
unsigned char qp_slice_min; // min slice qp value |
| 542 |
unsigned short qp_slice_max : 8; // max slice qp value |
| 543 |
unsigned short force_intra : 1; // force entire slice to be intra |
| 544 |
unsigned short disable_deblocking_filter_idc : 2; |
| 545 |
unsigned short cabac_init_idc : 2; |
| 546 |
unsigned short disable_slice_header : 1; // if set to 1, disables HW slice header output |
| 547 |
unsigned short reserved0 : 2; |
| 548 |
|
| 549 |
// int3 |
| 550 |
unsigned int num_ref_idx_active_override_flag : 1; |
| 551 |
unsigned int num_ref_idx_l0_active_minus1 : 5; // [0.31] |
| 552 |
unsigned int num_ref_idx_l1_active_minus1 : 5; // [0,31] |
| 553 |
unsigned int reserved1 : 9; // padding |
| 554 |
unsigned int me_control_idx : 4; // index in array of nvenc_h264_me_control_s to use for this slice |
| 555 |
unsigned int md_control_idx : 4; // index in array of nvenc_h264_md_control_s to use for this slice |
| 556 |
unsigned int q_control_idx : 4; // index in array of nvenc_h264_quant_control_s to use for this slice |
| 557 |
|
| 558 |
// int 4 |
| 559 |
unsigned int limit_slice_top_boundary : 1; // limit ME to slice boundaries |
| 560 |
unsigned int limit_slice_bot_boundary : 1; // limit ME to slice boundaries |
| 561 |
unsigned int limit_slice_left_boundary : 1; // limit ME to slice boundaries |
| 562 |
unsigned int limit_slice_right_boundary : 1; // limit ME to slice boundaries |
| 563 |
unsigned int ROI_enable : 1; // Enable region of interest processing |
| 564 |
unsigned int ROI_me_control_idx : 4; |
| 565 |
unsigned int ROI_md_control_idx : 4; |
| 566 |
unsigned int ROI_q_control_idx : 4; |
| 567 |
unsigned int ROI_qp_delta : 8; // QP delta in ROI relative to picture outside ROI |
| 568 |
unsigned int wp_control_idx : 4; // index in array of nvenc_pred_weight_table_s to use for this slice |
| 569 |
unsigned int reserved2 : 3; // reserved for alignment |
| 570 |
|
| 571 |
// int 5 |
| 572 |
unsigned int ROI_top_mbx : 8; // top left mbx of the ROI |
| 573 |
unsigned int ROI_top_mby : 8; |
| 574 |
unsigned int ROI_bot_mbx : 8; // bottom right mbx of the ROI |
| 575 |
unsigned int ROI_bot_mby : 8; |
| 576 |
|
| 577 |
// int 6: ref_pic_cmd_cfg |
| 578 |
nvenc_h264_refpic_cmd_s ref_pic_cmd_cfg; |
| 579 |
|
| 580 |
// int 7-22: RefPicListReorder/Modification: Defined in Section 7.4.3.1 |
| 581 |
nvenc_h264_ref_pic_reorder_s ref_pic_list_reorder_cmd[2][8]; // size shall not exceed num_ref_idx_lx_active_minus1 + 1 (8+8=16) |
| 582 |
|
| 583 |
//int 23-30: MMCO: Defined in Section: 7.4.3.3 |
| 584 |
nvenc_h264_mmco_s mmco_cmd[8] ; |
| 585 |
|
| 586 |
// int 31 |
| 587 |
unsigned int reserved3; |
| 588 |
|
| 589 |
} nvenc_h264_slice_control_s; // 128 bytes |
| 590 |
|
| 591 |
|
| 592 |
// Quarter-resolution modifier flags for first_pass_source_half_scaled |
| 593 |
#define NVENC_QRES_FLAG_ENABLE 0x1 // Set if quarter-resolution is enabled |
| 594 |
#define NVENC_QRES_FLAG_DECIMATED 0x10 // Set to indicate that subsampling was performed with decimation rather than low-pass |
| 595 |
#define NVENC_QRES_FLAG_PASS2REFS 0x20 // Set to indicate that 1st-pass uses subsampled 2nd pass references instead of self-generated references |
| 596 |
|
| 597 |
// Lookahead macros for ab_beta(INFO1) and prev_act(INFO2) |
| 598 |
#define NVENC_LOOKAHEAD_INFO1(Wp2Wi) ((unsigned char)(Wp2Wi)) // clamp((int)(Wp*256.0/Wi), 0, 255) -> ratio of inter over intra complexity, 0=no lookahead |
| 599 |
// Lookahead INFO2: (depth<<8)|(nextI) |
| 600 |
// nextI: 1..255=distance to next scene change, 0=unknown or greater than lookahead depth; |
| 601 |
// depth: 1..255=max lookahead depth (0=no lookahead) |
| 602 |
#define NVENC_LOOKAHEAD_INFO2(nextI,depth) (((unsigned short)(nextI)) | (((unsigned short)(depth))<<8)) |
| 603 |
|
| 604 |
// HRD conformance and misc RC flags |
| 605 |
#define NVENC_RC_HRD_VCL 0x1 // vcl_cpb_size and vcl_bitrate |
| 606 |
#define NVENC_RC_HRD_NAL 0x2 // nal_cpb_size and nal_bitrate |
| 607 |
#define NVENC_RC_HRD_STRICTGOPTARGET 0x4 // Strict GOP target (minimizes GOP-to-GOP rate fluctuations) |
| 608 |
#define NVENC_RC_FIXED_PBRATIO 0x8U // Disable dynamic B-frame rhopbi adjustments |
| 609 |
|
| 610 |
// needed only if picture level RC is enabled; as part of picture level RC, it will also perform HRD verification |
| 611 |
// The contents of this file is produced by the driver during sequence level operations and should not be modified. |
| 612 |
typedef struct |
| 613 |
{ |
| 614 |
unsigned char hrd_type; // nal (=2) and vcl (=1) type |
| 615 |
unsigned char QP[3]; // QP for 0:P picture, ,1:B picture 2; I picture |
| 616 |
unsigned char minQP[3]; // min QP for 0:P picture, ,1:B picture 2; I picture |
| 617 |
unsigned char maxQP[3]; // max QP for 0:P picture, ,1:B picture 2; I picture |
| 618 |
unsigned char maxQPD; // max QP delta between two consecutive QP updates |
| 619 |
unsigned char baseQPD; // initial QP delta between two consecutive QP updates |
| 620 |
int rhopbi[3]; // 23.8 signed fixed point quant ratios P/I, B/I, I/I(=1) |
| 621 |
|
| 622 |
int framerate; // fps |
| 623 |
unsigned int buffersize; // total buffer size |
| 624 |
|
| 625 |
int nal_cpb_size; // size in bits |
| 626 |
int nal_bitrate; // rate in bps |
| 627 |
int vcl_cpb_size; // size in bits |
| 628 |
int vcl_bitrate; // rate in bps |
| 629 |
|
| 630 |
unsigned int gop_length; // I period, gop_length == 0xffffffff is used for infinite gop length |
| 631 |
int Np; // 27.4 signed fixed (gop_length + num_b_frames -1)/(num_b_frames + 1) |
| 632 |
int Bmin; // 23.8 signed fixed point min buffer level, updated by driver at sequence level only. |
| 633 |
int Ravg; // 23.8 signed fixed point average rate, updated by driver at sequence level only. |
| 634 |
int R; // 23.8 signed fixed point rate |
| 635 |
|
| 636 |
unsigned char ab_alpha; // VBR/CBR: min/target quality level (1..51); rcmode4: weight of prev frame activity compared to part of the current picture |
| 637 |
unsigned char ab_beta; // lookahead info1: ratio of inter over intra complexity (fix8); rcmode4: weight of current activity compared to neighbors |
| 638 |
unsigned short prev_act; // lookahead info2: nextI, depth and valid flag; rcmode4: average activity level of the first reference (L0) |
| 639 |
|
| 640 |
unsigned char aqMode : 3; // 0: disable adaptive quantization, 1: AQ mode fast, 2: AQ mode full |
| 641 |
unsigned char dump_aq_stats : 1; // 1: reads aq stats |
| 642 |
unsigned char is_emphasis_level : 1; // 1: enables emphasis level map, treats QP map as level map |
| 643 |
unsigned char reserve0 : 3; // Reserved for alignment |
| 644 |
unsigned char single_frame_VBV; // 1: VBV buffer size is set to average frame size; 0: otherwise |
| 645 |
unsigned char two_pass_rc; // 0: single pass rc , 1: first pass of 2 pass rc 2: second pass of 2 pass rc |
| 646 |
unsigned char rc_class; // reserved |
| 647 |
|
| 648 |
unsigned char first_pass_source_half_scaled; // 0 : first pass on half resolution 1: first pass on full resolution (along with NVENC_QRES_FLAG_XXX) |
| 649 |
unsigned char iSizeRatioX; // ratio between I picture target size over average picture size numerator |
| 650 |
unsigned char iSizeRatioY; // ratio between I picture target size over average picture size denominator |
| 651 |
unsigned char ext_scene_change_flag; // Scene change flag set as hint by external preprocessing unit. 0: No scene change, 1: Current picture is first in scene |
| 652 |
unsigned int ext_intra_satd; // If non zero this represents the intra SATD for current picture computed by external preprocessing unit |
| 653 |
unsigned char ext_picture_rc_hints; // picture rc hints are set by external source. 0: no hints, 1: hints are avaialble for current picture |
| 654 |
unsigned char session_max_qp; // Current frame qp will never exceed this value when ext_picture_rc_hints is set |
| 655 |
unsigned char reserve[2]; |
| 656 |
} nvenc_h264_rc_s; // 88 bytes |
| 657 |
|
| 658 |
typedef struct |
| 659 |
{ |
| 660 |
unsigned int var_min; |
| 661 |
unsigned int var_max; |
| 662 |
unsigned int var_avg; |
| 663 |
unsigned char reserved[4]; |
| 664 |
} nvenc_aq_stat_s; // 16 bytes |
| 665 |
|
| 666 |
typedef struct |
| 667 |
{ |
| 668 |
char l0[8]; // reference picture list 0 |
| 669 |
char l1[8]; // reference picture list 1 |
| 670 |
char temp_dist_l0[8]; // temporal distance of each ref pic in list 0 DiffPicOrderCnt( currPicOrField, pic0) ) |
| 671 |
char temp_dist_l1[8]; // temporal distance of each ref pic in list 1 DiffPicOrderCnt( currPicOrField, pic1) ) |
| 672 |
short dist_scale_factor[8][8]; // (h264 spec eq 8-203) array [refidx1][refidx0] signed 11 bit values |
| 673 |
unsigned int diff_pic_order_cnt_zero[2]; // This is a 2-dimensional array of booleans (1 bit flags) indicating that DiffPicOrderCnt(refidx1, refidx0) == 0 |
| 674 |
unsigned int max_slice_size; // for use in slice_mode 1 only: dynamic insertion at/before this boundary (#bytes) |
| 675 |
unsigned int max_byte_count_before_resid_zero; // Maximum byte count before micrcode forces the residuals to zero. |
| 676 |
int delta_pic_order_cnt_bottom; // direct value of syntax element to be put in every slice of the picture |
| 677 |
unsigned short frame_num; // value for slice header syntax element |
| 678 |
unsigned short pic_order_cnt_lsb; // value of the corresponding syntax element |
| 679 |
unsigned short idr_pic_id; |
| 680 |
unsigned short colour_plane_id; |
| 681 |
unsigned int longTermFlag; // Bits 7:0 for pictures in RefPicList0 and bits 23:16 for pictures in RefPicList1 |
| 682 |
unsigned char num_forced_slices_minus1; // number of forced slice boundary locations minus1(number of entries in h264_slice_control_s) |
| 683 |
unsigned char num_me_controls_minus1; // number of nvenc_h264_me_control_s entries in array minus1 |
| 684 |
unsigned char num_md_controls_minus1; // number of nvenc_h264_md_control_s entries in array minus1 |
| 685 |
unsigned char num_q_controls_minus1; // number of nvenc_h264_quant_control_s entries in array minus1 |
| 686 |
unsigned int slice_control_offset; // offset from start of top level "nvenc_h264_drv_pic_setup_s" structure |
| 687 |
// to start of array of "nvenc_h264_slice_control_s" structures |
| 688 |
unsigned int me_control_offset; // offset from start of top level "nvenc_h264_drv_pic_setup_s" structure |
| 689 |
// to start of array of "nvenc_h264_me_control_s" structures |
| 690 |
unsigned int md_control_offset; // offset from start of top level "nvenc_h264_drv_pic_setup_s" structure |
| 691 |
// to start of array of "nvenc_h264_md_control_s" structures |
| 692 |
unsigned int q_control_offset; // offset from start of top level "nvenc_h264_drv_pic_setup_s" structure |
| 693 |
// to start of array of "nvenc_h264_quant_control_s" structures |
| 694 |
unsigned int hist_buf_size; // size in bytes of the buffer allocated for history |
| 695 |
unsigned int bitstream_buf_size; // size in bytes of the buffer allocated for bitstream slice/mb data |
| 696 |
unsigned int bitstream_start_pos; // start position in bitstream buffer where data should be written (byte offset) |
| 697 |
unsigned int pic_struct : 2; // 0 = frame, 1 = top/first field, 2 = bot/second field |
| 698 |
unsigned int pic_type : 2; // 0 = P, 1 = B, 2 = I, 3 = IDR |
| 699 |
unsigned int ref_pic_flag : 1; // reference picture (0 = no, 1 = yes) |
| 700 |
unsigned int slice_mode : 1; // 0 = dynamic slice insertion based on slice size |
| 701 |
// 1 = static slice insertion based on slice control array |
| 702 |
// (insert slices as defined in array, if all array elements used |
| 703 |
// before end of picture is reached, start over with entry 0) |
| 704 |
unsigned int ipcm_rewind_enable : 1; |
| 705 |
unsigned int cabac_zero_word_enable : 1; |
| 706 |
|
| 707 |
//mvc |
| 708 |
unsigned int base_view : 1; // 1: current view component is a base view; 0: non-base view |
| 709 |
unsigned int priority_id : 6; // priority identifier for the NAL unit |
| 710 |
unsigned int view_id : 10; // view identifier for the NAL unit |
| 711 |
unsigned int temporal_id : 3; // temporal identifier for the NAL unit |
| 712 |
unsigned int anchor_pic_flag : 1; // equal to 1 specifies that the current access unit is an anchor access unit |
| 713 |
unsigned int inter_view_flag : 1; // equal to 1 specifies that the current view component may be used for inter-view prediction by other view components in the current access unit |
| 714 |
unsigned int opm_enable : 1; // check with OPM if encryption needed |
| 715 |
unsigned int temporal_hint8x8_enable : 1; // 8x8 hint enable |
| 716 |
char cur_interview_ref_pic; //current interview ref pic |
| 717 |
char prev_interview_ref_pic; //previous interview ref pic to replace current interview ref pic when write the regs ME_MBC_REFLIST0_DPBLUT and ME_MBC_REFLIST1_DPBLUT |
| 718 |
unsigned char codec : 3; |
| 719 |
unsigned char e4byteStartCode : 1; // if enable to 1, all the slices in a picture will use 4 byte start code 00000001(used in rtp mode), original is 000001 |
| 720 |
unsigned char qpfifo : 1; // 0 = Polling mode, 1 = Interrupt mode. |
| 721 |
unsigned int me_candidates_output_struct : 1; // 0 = me-only mode, 1 = MV & cost for all blocks (hybrid format) |
| 722 |
unsigned int me_candidates_output_enable : 1; // hybrid mode results configuration (see me_candidates_output_struct). 0 = results disabled. 1 = results enabled (for fullscale image). |
| 723 |
unsigned char medma_output_cost_mode : 1; // 0 = residual cost only. 1 = residual cost + mv cost |
| 724 |
unsigned char intraRefreshCount; // Set to number of pictures over which intra refresh will happen. Set to 0 for non intra refresh picture |
| 725 |
unsigned int mpec_threshold : 30; // collect mpec stats after threshold mbs |
| 726 |
// when mpec_stat_on is enabled and mpec_threshold is not equal to picWidthInMbs, |
| 727 |
// rc mode 0 will be used to collect stats |
| 728 |
unsigned int me_lambda_fifo_enable : 1; // lambda map fifo enable. 1: enable. 0: disable |
| 729 |
unsigned int me_candidates_rdo_mode : 1; // me candidates rdo mode enable. 1: enable. 0: disable. |
| 730 |
unsigned int slice_stat_offset; // offset from start of top level "nvenc_stat_data_s" structure |
| 731 |
// to start of array of "nvenc_slice_stat_s" structure |
| 732 |
unsigned int mpec_stat_offset; // offset from start of top level "nvenc_stat_data_s" structure |
| 733 |
// to start of array of "nvenc_mpec_stat_s" structure |
| 734 |
unsigned int wp_control_offset; // offset from start of top level "nvenc_h264_drv_pic_setup_s" structure |
| 735 |
// to start of array of "nvenc_pred_weight_table_s" structure |
| 736 |
// to start of array of "nvenc_mpec_stat_s" structure |
| 737 |
|
| 738 |
unsigned char num_wp_controls_minus1; // number of nvenc_pred_weight_table_s entries in array minus 1 |
| 739 |
unsigned char b_as_ref : 1; // use b frame as reference |
| 740 |
unsigned char ofs_mode : 2; // ofs mode. 0: OFF; 1: Optical Flow; 2: Stereo (output mvx only), 3: Stereo (output mvx/mvy) |
| 741 |
unsigned char new_subframe : 1; |
| 742 |
unsigned char bit_depth_minus_8 : 4; |
| 743 |
unsigned char stripID; |
| 744 |
unsigned char strips_in_frame; |
| 745 |
unsigned short slice_encoding_row_start; // start mb row for slice encoding |
| 746 |
unsigned short slice_encoding_row_num; // num of mb row for slice encoding |
| 747 |
|
| 748 |
unsigned char segment_enable : 1; // used by ofs mode only for segment map enable flag, 0: (default) disable/ 1: enable |
| 749 |
unsigned char background_seg : 5; // used by ofs mode only when segment map is enabled, to specify the segment id of background |
| 750 |
unsigned char input_sub_sampling : 1; // used by PDMA 2x2 sub-sampling |
| 751 |
unsigned char output_sub_sampling : 1; // used by MPEB 2x2 sub-sampling output |
| 752 |
unsigned char segment_P2 ; |
| 753 |
unsigned char segment_thr ; |
| 754 |
unsigned char accurate_skip_mv : 1; |
| 755 |
unsigned char intra_sel_4x4_enable : 1; // seletive intra4x4 , 0:disable(always do intra4x4 serach), 1:enable(enable or disable intra4x4 search based on pixel variance) |
| 756 |
unsigned int enable_source_image_padding : 1; // enable/disable HW source image padding |
| 757 |
unsigned int luma_padding_mode : 2; // HW souce padding mode for luma: 0: copy frame boundary, 1: zero padding, 2: half value, i.e 1<<(bitdepth-1) |
| 758 |
unsigned int chroma_padding_mode : 2; // HW souce padding mode for chroma: 0: copy frame boundary, 1: zero padding, 2: half value, i.e 1<<(bitdepth-1) |
| 759 |
unsigned int stats_fifo_enable : 1; // enable per-mb short stats fifo |
| 760 |
unsigned int chroma_skip_threshold_4x4 : 16; // low 3 bits is the factional, to check if chroma residual is big in skip mb, can't simply set chroma residual to 0 even though luma has been decided to skip |
| 761 |
unsigned int intra_sel_4x4_threshold : 16; // disable intra4x4 search when intra_sel_4x4_enable=true and max variance of 8x8 blocks in current MB is less than intra_sel_4x4_threshold |
| 762 |
unsigned int aq_stat_offset; // start of array of "nvenc_aq_stat_s" structure |
| 763 |
unsigned int act_stat_offset; // offset from start of "nvenc_rc_pic_stat_s" structure, |
| 764 |
// in the RC stats buffer. This field points to the start |
| 765 |
// of data for external (delta-)QP map. |
| 766 |
unsigned int stats_fifo_offset; // offset from start of top level "nvenc_stat_data_s" structure |
| 767 |
} nvenc_h264_pic_control_s; // 276 bytes |
| 768 |
|
| 769 |
|
| 770 |
// This is the top level config structure passed by driver to nvenc. It must be aligned |
| 771 |
// to a 256B boundary. The first integer in this struct is a 'magic' number which driver |
| 772 |
// must set to a specific value, and which the main Falcon app must check to see if it |
| 773 |
// matches the expected value. This prevents running wrong combinations of traces/drvr |
| 774 |
// code and falcon code. Encoding is as follows: |
| 775 |
// Bit 31..16 = 0xc9b7 |
| 776 |
// Bit 15..8 = Version number |
| 777 |
// Bit 7..0 = Revision number |
| 778 |
// This number changes whenever there is a change to the class. If the change is small and |
| 779 |
// backward compatible, only the revision number is incremented. If the change is major or |
| 780 |
// not backward compatible, the revision number is reset to zero and the version number |
| 781 |
// incremented. Falcon app should check if the upper 24 bits match the expected value, and |
| 782 |
// terminate with error code "NvC9B7EncErrorH264BadMagic" if there is a mismatch. The value |
| 783 |
// for the current class is defined as: |
| 784 |
|
| 785 |
#ifdef NV_NVENC_8_2 |
| 786 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xC9B70006 |
| 787 |
#elif NV_NVENC_8_0 |
| 788 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xC8B70006 |
| 789 |
#elif NV_NVENC_7_3 |
| 790 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xC7B70006 |
| 791 |
#elif NV_NVENC_7_2 |
| 792 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xC4B70006 |
| 793 |
#elif NV_NVENC_7_0 |
| 794 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xC5B70006 |
| 795 |
#elif NV_NVENC_6_8 |
| 796 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xB6B70006 |
| 797 |
#elif NV_NVENC_6_6 |
| 798 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xB4B70006 |
| 799 |
#elif NV_NVENC_6_4 |
| 800 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xC3B70006 |
| 801 |
#elif NV_NVENC_6_2 |
| 802 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xC2B70006 |
| 803 |
#elif NV_NVENC_6_0 |
| 804 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xc1b70006 |
| 805 |
#elif NV_NVENC_5_0 |
| 806 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xd0b70006 |
| 807 |
#elif NV_NVENC_1_0 |
| 808 |
#define NV_NVENC_DRV_MAGIC_VALUE 0xc0b70006 |
| 809 |
#elif NV_MSENC_2_0 |
| 810 |
#define NV_MSENC_DRV_MAGIC_VALUE 0xa0b70006 |
| 811 |
#else |
| 812 |
#define NV_MSENC_DRV_MAGIC_VALUE 0x90b70006 |
| 813 |
#endif |
| 814 |
|
| 815 |
#define HIST_BLOCK_SIZE 192 |
| 816 |
// Move all local define of these values to this file |
| 817 |
//align to current AV1HIST MAS |
| 818 |
const static int AV1_HIST_PIXEL_FRAME_SIZE = (8 * 8192); |
| 819 |
const static int AV1_HIST_MV_SB_SIZE = (256 * 4); |
| 820 |
|
| 821 |
typedef struct |
| 822 |
{ |
| 823 |
// 256 aligned |
| 824 |
unsigned int magic; // magic number, see text above 4 bytes |
| 825 |
nvenc_h264_surface_cfg_s refpic_cfg; // defines layout of reference pictures 32 bytes |
| 826 |
nvenc_h264_surface_cfg_s input_cfg; // defines layout of input pictures 32 bytes |
| 827 |
nvenc_h264_surface_cfg_s outputpic_cfg; // defines layout of reconstructed pictures 32 bytes |
| 828 |
nvenc_h264_sps_data_s sps_data; // 4 bytes |
| 829 |
nvenc_h264_pps_data_s pps_data; // 8 bytes |
| 830 |
nvenc_h264_rc_s rate_control; // Rate Control information 88 bytes |
| 831 |
nvenc_h264_pic_control_s pic_control; // 276 bytes |
| 832 |
nvenc_h264_surface_cfg_s half_scaled_outputpic_cfg; // defines layout of 2x2 subsampled reconstructed picture 32 bytes |
| 833 |
unsigned int gpTimer_timeout_val; // GPTimer cycle count set from driver 4 bytes |
| 834 |
|
| 835 |
} nvenc_h264_drv_pic_setup_s; // 512 bytes |
| 836 |
|
| 837 |
|
| 838 |
// **************************************************************************** |
| 839 |
// **************************************************************************** |
| 840 |
// H265 |
| 841 |
// **************************************************************************** |
| 842 |
// **************************************************************************** |
| 843 |
typedef struct |
| 844 |
{ |
| 845 |
unsigned short frame_width_minus1; // frame width in pixels minus 1, range 0-4095 |
| 846 |
unsigned short frame_height_minus1; // frame height in pixels minus 1, range 0-4095 |
| 847 |
unsigned short sfc_pitch; // pitch of luma plane |
| 848 |
unsigned short sfc_pitch_chroma; // pitch of chroma plane |
| 849 |
unsigned short sfc_trans_mode; // least 3 significient bits are used to stand for 8 modes, normal, xflip, yflip etc. |
| 850 |
unsigned short reserved2; // pad to int |
| 851 |
|
| 852 |
unsigned int luma_top_frm_offset; // offset of luma top field or frame in units of 256 |
| 853 |
unsigned int luma_bot_offset; // offset of luma bottom field in units of 256. Not used if frame format. |
| 854 |
unsigned int chroma_top_frm_offset; // offset of chroma top field or frame, or offset of chroma U(Cb) plane in planar mode, both in units of 256 |
| 855 |
unsigned int chroma_bot_offset; // offset of chroma bottom field, not used if frame format. Or offset of chroma V(Cr) plane in planar mode. Both in units of 256. |
| 856 |
unsigned int block_height : 7; // BL mapping block height setting |
| 857 |
unsigned int tiled_16x16 : 1; // Surface is 16x16 tiled instead of BL mapped (must be 1 for refpics) |
| 858 |
unsigned int memory_mode : 2; // 0: semi-planar, 1: planar |
| 859 |
unsigned int nv21_enable : 1; // the surface format is yuv or yvu |
| 860 |
unsigned int input_bl_mode : 2; // the input block linear mode: 0~gpu bl; 1~tegra bl; 2~naive bl this is only for cmod and cmod/plugin |
| 861 |
unsigned int reserved : 19; // pad to NvU32 |
| 862 |
} nvenc_h265_surface_cfg_s; // 32 bytes |
| 863 |
|
| 864 |
typedef struct |
| 865 |
{ |
| 866 |
unsigned int chroma_format_idc : 2; // 0=monochrome, 1=yuv4:2:0, 3=yuv4:4:4, 2=not supported |
| 867 |
unsigned int separate_colour_plane_flag : 1; //if this bit is enabled, the chroma_format_idc should be 3 |
| 868 |
unsigned int log2_max_pic_order_cnt_lsb_minus4 : 4; |
| 869 |
unsigned int min_cu_size : 2; //0-8x8, 1-16x16, 2-32x32, 3-64x64 |
| 870 |
unsigned int max_cu_size : 2; //0-8x8, 1-16x16, 2-32x32, 3-64x64 |
| 871 |
unsigned int min_tu_size : 2; //0-4x4, 1-8x8, 2-16x16, 3-32x32 |
| 872 |
unsigned int max_tu_size : 2; //0-4x4, 1-8x8, 2-16x16, 3-32x32 |
| 873 |
unsigned int reserved0 : 1; // pad to full short |
| 874 |
|
| 875 |
unsigned int max_transform_depth_inter : 2; // max_transform_hierarchy_depth_inter |
| 876 |
unsigned int max_transform_depth_intra : 2; // max_transform_hierarchy_depth_intra |
| 877 |
unsigned int pcm_enabled_flag : 1; // if this bit is 1 , PCM data is present is the sequence |
| 878 |
unsigned int pcm_loop_filter_disable_flag : 1; // Disable loop filter and SAO for CU with pcm_flag to 1 |
| 879 |
unsigned int long_term_ref_pics_present_flag : 1; // if set to 1 , it indicates long term pictures are used for inter prediction. |
| 880 |
unsigned int amp_enabled_flag : 1; // if set asymmetric motion partitions can be used for inter predictions. |
| 881 |
unsigned int strong_intra_smoothing_enabled_flag : 1; // if set , bilinear interpolation is conditionally used filtering process for intra prediction. |
| 882 |
unsigned int num_short_term_ref_pic_sets : 7; // max value is 64 |
| 883 |
unsigned int num_long_term_ref_pics_sps : 5; |
| 884 |
unsigned int bit_depth_minus_8 : 4; // luma/chroma should have the same value |
| 885 |
unsigned int sample_adaptive_offset_enabled_flag : 1; // set 1 specifies that the sample adaptive offset process is applied to the reconstructed picture after the deblocking filter process |
| 886 |
unsigned int sps_temporal_mvp_enabled_flag : 1; // if set, slice_temporal_mvp_enabled_flag is present in slice header |
| 887 |
unsigned int reserved1 : 21; |
| 888 |
} nvenc_h265_sps_data_s; // 8 bytes |
| 889 |
|
| 890 |
typedef struct |
| 891 |
{ |
| 892 |
unsigned int pic_param_set_id : 8; // picture parameter set identification |
| 893 |
unsigned int num_ref_idx_l0_active_minus1 : 5; // number of currently active reference pictures in list 0 minus 1, range 0..30 |
| 894 |
unsigned int num_ref_idx_l1_active_minus1 : 5; // number of currently active reference pictures in list 1 minus 1, range 0..30 |
| 895 |
unsigned int constrained_intra_pred_flag : 1; // if set, intra prediction can only use pixels from macroblocks that are also intra coded |
| 896 |
unsigned int transquant_bypass_enabled_flag : 1; // if set, transform ,scaling and inloop filtering is bypassed. |
| 897 |
unsigned int transform_skip_enabled_flag : 1; // if set , it enables transform bypass during residual coding. |
| 898 |
unsigned int deblocking_filter_override_enabled_flag : 1; // it set, the deblocking_filter_override_flag is encoded in slice segment header |
| 899 |
unsigned int pps_loop_filter_across_slices_enabled_flag : 1; // if set, it enables loop filter across slice boundary. |
| 900 |
unsigned int cu_qp_delta_enabled_flag : 1; // if set , it enables cu_qp_delta_abs is encoded at coding unit layer. |
| 901 |
unsigned int lists_modification_present_flag : 1; // if set , it indicates reference picture list modification command might be present is slice segment header. |
| 902 |
unsigned int output_flag_present_flag : 1; // if set, pic_output syntax element is added to slice syntax header |
| 903 |
unsigned int cabac_init_present_flag : 1; // if set, cabac_init_flag is present in slice segment header. |
| 904 |
unsigned int pps_slice_chroma_qp_offsets_present : 1; // if set to 1, cb_qp_offset and cr_qp_offset is signalled in the slice segment header. |
| 905 |
unsigned int weighted_pred_flag : 1; // weighted_pred_flag equal to 1 specifies that weighted prediction is applied to P slices |
| 906 |
unsigned int weighted_bipred_flag : 1; // weighted_bipred_flag equal to 1 specifies that weighted prediction is applied to B slices |
| 907 |
unsigned int reserved : 2; |
| 908 |
|
| 909 |
int pic_init_qp_minus26 : 6; // initial QP value, range -26..+25. -38...+25 for 10bit, -74...+25 for 16bits. TODO: eable this syntex for pps |
| 910 |
int pps_cb_qp_offset : 5; |
| 911 |
int pps_cr_qp_offset : 5; |
| 912 |
unsigned int dependent_slice_segments_enabled_flag : 1; |
| 913 |
unsigned int sign_data_hiding_enabled_flag : 1; |
| 914 |
unsigned int diff_cu_qp_delta_depth : 2; |
| 915 |
unsigned int log2_parallel_merge_level_minus2 : 3; |
| 916 |
unsigned int num_extra_slice_header_bits : 3; |
| 917 |
|
| 918 |
int reserved1 : 6; |
| 919 |
} nvenc_h265_pps_data_s; // 8 bytes |
| 920 |
|
| 921 |
// needed only if picture level RC is enabled; as part of picture level RC, it will also perform HRD verification |
| 922 |
// The contents of this file is produced by the driver during sequence level operations and should not be modified. |
| 923 |
typedef struct |
| 924 |
{ |
| 925 |
unsigned char hrd_type; // nal (=2) and vcl (=1) type |
| 926 |
unsigned char QP[3]; // QP for 0:P picture, ,1:B picture 2; I picture |
| 927 |
unsigned char minQP[3]; // min QP for 0:P picture, ,1:B picture 2; I picture |
| 928 |
unsigned char maxQP[3]; // max QP for 0:P picture, ,1:B picture 2; I picture |
| 929 |
unsigned char maxQPD; // max QP delta between two consecutive QP updates |
| 930 |
unsigned char baseQPD; // initial QP delta between two consecutive QP updates |
| 931 |
int rhopbi[3]; // 23.8 signed fixed point quant ratios P/I, B/I, I/I(=1) |
| 932 |
|
| 933 |
int framerate; // fps |
| 934 |
unsigned int buffersize; // total buffer size |
| 935 |
|
| 936 |
int nal_cpb_size; // size in bits |
| 937 |
int nal_bitrate; // rate in bps |
| 938 |
int vcl_cpb_size; // size in bits |
| 939 |
int vcl_bitrate; // rate in bps |
| 940 |
|
| 941 |
unsigned int gop_length; // I period, gop_length == 0xffffffff is used for infinite gop length |
| 942 |
int Np; // 27.4 signed fixed (gop_length + num_b_frames -1)/(num_b_frames + 1) |
| 943 |
int Bmin; // 23.8 signed fixed point min buffer level, updated by driver at sequence level only. |
| 944 |
int Ravg; // 23.8 signed fixed point average rate, updated by driver at sequence level only. |
| 945 |
int R; // 23.8 signed fixed point rate |
| 946 |
|
| 947 |
unsigned char ab_alpha; // VBR/CBR: min/target quality level (1..51); rcmode4: weight of prev frame activity compared to part of the current picture |
| 948 |
unsigned char ab_beta; // lookahead info1: ratio of inter over intra complexity (fix8); rcmode4: weight of current activity compared to neighbors |
| 949 |
unsigned short prev_act; // lookahead info2: NVENC_LOOKAHEAD_INFO2; rcmode4: average activity level of the first reference (L0) |
| 950 |
|
| 951 |
unsigned char reserved0; |
| 952 |
unsigned char single_frame_vbv; // 1: VBV buffer size is set to average frame size; 0: otherwise |
| 953 |
unsigned char two_pass_rc; // 0: single pass rc , 1: first pass of 2 pass rc 2: second pass of 2 pass rc |
| 954 |
unsigned char rc_class; // reserved |
| 955 |
unsigned char first_pass_source_half_scaled; // 0 : first pass on half resolution 1: first pass on full resolution (along with NVENC_QRES_FLAG_XXX) |
| 956 |
unsigned char i_size_ratio_x; // ratio between I picture target size over average picture size numerator |
| 957 |
unsigned char i_size_ratio_y; // ratio between I picture target size over average picture size denominator |
| 958 |
unsigned char aqMode; // 0: disable adaptive quantization 1: AQ mode fast (unimplemented) 2: AQ mode full (unimplemented) 3: External qp delta map |
| 959 |
unsigned char dump_aq_stats; // 1: dump aq stats |
| 960 |
unsigned char ext_scene_change_flag; // Scene change flag set as hint by external preprocessing unit. 0: No scene change, 1: Current picture is first in scene |
| 961 |
unsigned char ext_picture_rc_hints; // picture rc hints are set by external source. 0: no hints, 1: hints are avaialble for current picture |
| 962 |
unsigned char session_max_qp; // Current frame qp will never exceed this value when ext_picture_rc_hints is set |
| 963 |
unsigned int ext_intra_satd; // If non zero this represents the intra SATD for current picture computed by external preprocessing unit |
| 964 |
unsigned char reserved2[8]; |
| 965 |
} nvenc_h265_rc_s; // 96 bytes |
| 966 |
|
| 967 |
typedef struct |
| 968 |
{ |
| 969 |
int inter_ref_pic_set_prediction_flag : 1; |
| 970 |
int delta_idx_minus1 : 6; |
| 971 |
int delta_rps_sign : 1; |
| 972 |
int abs_delta_rps_minus1 : 16; |
| 973 |
int reserved : 8; |
| 974 |
unsigned short used_by_curr_pic_flag; |
| 975 |
unsigned short use_delta_flag; |
| 976 |
short num_negative_pics; |
| 977 |
short num_positive_pics; |
| 978 |
short delta_poc_s0_minus1[16]; |
| 979 |
unsigned short used_by_curr_pic_s0_flag; |
| 980 |
short delta_poc_s1_minus1[16]; |
| 981 |
unsigned short used_by_curr_pic_s1_flag; |
| 982 |
}nvenc_h265_strps_s; // 80bytes multiple of 4 |
| 983 |
|
| 984 |
typedef struct |
| 985 |
{ |
| 986 |
int num_long_term_sps : 5; |
| 987 |
int num_long_term_pics : 5; |
| 988 |
int reserved0 : 22; |
| 989 |
unsigned char lt_idx_sps[16]; |
| 990 |
unsigned short poc_lsb_lt[16]; |
| 991 |
unsigned short used_by_curr_pic_lt_flag; |
| 992 |
unsigned short delta_poc_msb_present_flag; |
| 993 |
int delta_poc_msb_cycle_lt[16]; |
| 994 |
}nvenc_h265_ltrps_s; // 120bytes multiple of 4 |
| 995 |
|
| 996 |
typedef struct |
| 997 |
{ |
| 998 |
int NumPocTotalCurr : 5; |
| 999 |
int ref_pic_list_mod_flag_l0 : 1; |
| 1000 |
int ref_pic_list_mod_flag_l1 : 1; |
| 1001 |
int reserved0 : 25; |
| 1002 |
unsigned char list_entry_l0[16]; |
| 1003 |
unsigned char list_entry_l1[16]; |
| 1004 |
}nvenc_h265_ref_pic_list_modification_s; //36bytes multiple of 4 |
| 1005 |
|
| 1006 |
typedef struct |
| 1007 |
{ |
| 1008 |
unsigned int vps_poc_lsb_not_present_flag : 1; |
| 1009 |
unsigned int vps_default_ref_layers_active_flag : 1; |
| 1010 |
unsigned int vps_max_one_active_ref_layer_flag : 1; |
| 1011 |
unsigned int vps_num_direct_ref_layers : 6; |
| 1012 |
unsigned int slh_discardable_flag : 1; // mv_hevc, equal to 1 specifies that the coded picture is not used as a reference picture |
| 1013 |
unsigned int slh_cross_layer_bla_flag : 1; // mv_hevc |
| 1014 |
unsigned int slh_num_active_ref_layer_pics : 6; // active ref layer pics of current picture |
| 1015 |
unsigned int slh_inter_layer_pred_enabled_flag : 1; |
| 1016 |
unsigned int slh_num_inter_layer_ref_pics_minus1 : 6; |
| 1017 |
unsigned int reserved0 : 8; |
| 1018 |
unsigned char slh_inter_layer_pred_layer_idc[8]; //NumPicTotalCurr <= 8 |
| 1019 |
}nvenc_mv_hevc_cfg_s; //12bytes multiple of 4 |
| 1020 |
|
| 1021 |
typedef struct |
| 1022 |
{ |
| 1023 |
char l0[8]; // reference picture list 0 |
| 1024 |
char l1[8]; // reference picture list 1 |
| 1025 |
char temp_dist_l0[8]; // temporal distance of each ref pic in list 0 DiffPicOrderCnt( currPicOrField, pic0) ) |
| 1026 |
char temp_dist_l1[8]; // temporal distance of each ref pic in list 1 DiffPicOrderCnt( currPicOrField, pic1) ) |
| 1027 |
short dist_scale_factor[8][8]; // (h264 spec eq 8-203) array [refidx1][refidx0] signed 11 bit values |
| 1028 |
unsigned int diff_pic_order_cnt_zero[2]; // This is a 2-dimensional array of booleans (1 bit flags) indicating that DiffPicOrderCnt(refidx1, refidx0) == 0 |
| 1029 |
unsigned int max_slice_size; // for use in slice_mode 1 only: dynamic insertion at/before this boundary (#bytes) |
| 1030 |
unsigned int max_byte_count_before_resid_zero; // Maximum byte count before micrcode forces the residuals to zero. |
| 1031 |
|
| 1032 |
unsigned short no_output_of_prior_pics_flag : 1; // used to modify the DPB after decoding a IDR or BLA picture |
| 1033 |
unsigned short pic_output_flag : 1; // affects the decoded picture output and removal processes as specified in Annex C |
| 1034 |
unsigned short mode_decision_only_flag : 1; // only output h265 mode and mocomp result, no recon and bitstream output |
| 1035 |
unsigned short cu_early_terminatiom_enable : 1; // 0: disable CU early termination, 1: enable CU early termination |
| 1036 |
unsigned short input_sub_sampling : 1; // used by PDMA 2x2 sub-sampling |
| 1037 |
unsigned short output_sub_sampling : 1; // used by MPEB 2x2 sub-sampling output |
| 1038 |
unsigned short reserved0 : 10; |
| 1039 |
unsigned short pic_order_cnt_lsb; // value of the corresponding syntax element |
| 1040 |
|
| 1041 |
unsigned char num_forced_slices_minus1; // number of forced slice boundary locations minus1(number of entries in h264_slice_control_s) |
| 1042 |
unsigned char num_me_controls_minus1; // number of nvenc_h264_me_control_s entries in array minus1 |
| 1043 |
unsigned char num_md_controls_minus1; // number of nvenc_h264_md_control_s entries in array minus1 |
| 1044 |
unsigned char num_q_controls_minus1; // number of nvenc_h264_quant_control_s entries in array minus1 |
| 1045 |
unsigned int slice_control_offset; // offset from start of top level "nvenc_h264_drv_pic_setup_s" structure |
| 1046 |
// to start of array of "nvenc_h264_slice_control_s" structures |
| 1047 |
unsigned int me_control_offset; // offset from start of top level "nvenc_h264_drv_pic_setup_s" structure |
| 1048 |
// to start of array of "nvenc_h264_me_control_s" structures |
| 1049 |
unsigned int md_control_offset; // offset from start of top level "nvenc_h264_drv_pic_setup_s" structure |
| 1050 |
// to start of array of "nvenc_h264_md_control_s" structures |
| 1051 |
unsigned int q_control_offset; // offset from start of top level "nvenc_h264_drv_pic_setup_s" structure |
| 1052 |
// to start of array of "nvenc_h264_quant_control_s" structures |
| 1053 |
unsigned int hist_buf_size; // size in bytes of the buffer allocated for history |
| 1054 |
unsigned int bitstream_buf_size; // size in bytes of the buffer allocated for bitstream slice/mb data |
| 1055 |
unsigned int bitstream_start_pos; // start position in bitstream buffer where data should be written (byte offset) |
| 1056 |
unsigned int pic_type : 2; // 0 = P, 1 = B, 2 = I, 3 = IDR |
| 1057 |
unsigned int ref_pic_flag : 1; // reference picture (0 = no, 1 = yes) |
| 1058 |
unsigned int slice_mode : 1; // 0 = dynamic slice insertion based on slice size |
| 1059 |
// 1 = static slice insertion based on slice control array |
| 1060 |
// (insert slices as defined in array, if all array elements used |
| 1061 |
// before end of picture is reached, start over with entry 0) |
| 1062 |
unsigned int cabac_zero_word_enable : 1; |
| 1063 |
unsigned int codec : 3; |
| 1064 |
unsigned int ceil_log2_pic_size_in_ctbs : 6; // ceillog2(picSizeInCtbs) for HW needs too much area/time for 16kx16k case |
| 1065 |
unsigned int sao_luma_mode : 5; |
| 1066 |
unsigned int sao_chroma_mode : 5; |
| 1067 |
unsigned int reserved1 : 8; |
| 1068 |
|
| 1069 |
unsigned int mpec_threshold; // collect mpec stats after threshold mbs |
| 1070 |
// when mpec_stat_on is enabled and mpec_threshold is not equal to picWidthInMbs, |
| 1071 |
// rc mode 0 will be used to collect stats |
| 1072 |
unsigned int slice_stat_offset; // offset from start of top level "nvenc_h265_stat_data_s" structure |
| 1073 |
// to start of array of "nvenc_h265_slice_stat_s" structure |
| 1074 |
unsigned int mpec_stat_offset; // offset from start of top level "nvenc_h265_stat_data_s" structure |
| 1075 |
// to start of array of "nvenc_h265_mpec_stat_s" structure |
| 1076 |
unsigned int aq_stat_offset; // to start of array of "nvenc_h265_aq_stat_s" structure |
| 1077 |
unsigned int short_term_ref_pic_set_sps_flag : 1; // if set to 1 ,the short term RPS is signalled using SPS, if set to 0 the short term RPS is encoded in the current slice header |
| 1078 |
unsigned int short_term_ref_pic_set_idx : 6; // used to signal index into the list of the short term RPS in the SPS. |
| 1079 |
unsigned int e4byteStartCode : 1; // if enable to 1, all the slices in a picture will use 4 byte start code 00000001(used in rtp mode), original is 000001 |
| 1080 |
unsigned int nal_unit_type : 6; // specifies the type of RBSP data structure contained in the NAL unit |
| 1081 |
unsigned int nuh_layer_id : 6; // reserved and shall be equal to 0. |
| 1082 |
unsigned int nuh_temporal_id_plus1 : 3; // specifies the temporal_ip_plus1 of RBSP data structure contained in the NAL unit |
| 1083 |
unsigned int qpfifo : 1; // 0 = Polling mode, 1 = Interrupt mode. |
| 1084 |
unsigned int enable_source_image_padding : 1; // enable/disable HW source image padding |
| 1085 |
unsigned int luma_padding_mode : 2; // HW souce padding mode for luma: 0: copy frame boundary, 1: zero padding, 2: half value, i.e 1<<(bitdepth-1) |
| 1086 |
unsigned int chroma_padding_mode : 2; // HW souce padding mode for chroma: 0: copy frame boundary, 1: zero padding, 2: half value, i.e 1<<(bitdepth-1) |
| 1087 |
unsigned int stats_fifo_enable : 1; // enable per-ctb short stats fifo |
| 1088 |
unsigned int reserved2 : 2; |
| 1089 |
|
| 1090 |
nvenc_h265_strps_s short_term_rps; // if short_term_ref_pic_set_sps_flag is set to 0, the nvenc_h265_strps_s structure is signalled in the slice segment segment header. |
| 1091 |
nvenc_h265_ltrps_s long_term_rps; // if long_term_ref_pics_present_flag is set to 1 in SPS , nvenc_h265_ltrps_s structure is signalled in the slice segment segment header |
| 1092 |
nvenc_h265_ref_pic_list_modification_s ref_pic_list_modification; // if lists_modification_present_flag is set to 1, then nvenc_h265_ref_pic_list_modification_s structure is encoded in slice segment header. |
| 1093 |
unsigned int rc_pass2_offset; |
| 1094 |
|
| 1095 |
unsigned int wp_control_offset; // offset from start of top level "nvenc_h265_drv_pic_setup_s" structure |
| 1096 |
// to start of array of "nvenc_pred_weight_table_s" structure |
| 1097 |
|
| 1098 |
unsigned char num_wp_controls_minus1; // number of nvenc_pred_weight_table_s entries in array minus 1 |
| 1099 |
unsigned char stripID; |
| 1100 |
unsigned char strips_in_frame; |
| 1101 |
unsigned char slice_idx_start; |
| 1102 |
unsigned short slice_encoding_row_start; // start ctb row for slice encoding |
| 1103 |
unsigned short slice_encoding_row_num; // num of ctb row for slice encoding |
| 1104 |
|
| 1105 |
nvenc_mv_hevc_cfg_s mv_hevc_cfg; |
| 1106 |
unsigned int chroma_skip_threshold_4x4 : 16; // low 3 bits is the factional, to check if chroma residual is big in skip mb, can't simply set chroma residual to 0 even though luma has been decided to skip |
| 1107 |
unsigned int longTermFlag : 16; // Bits 7:0 for pictures in RefPicList0 and bits 15:8 for pictures in RefPicList1 |
| 1108 |
|
| 1109 |
char poc_entry_l0[8]; // ref pic list to ref pic entry map |
| 1110 |
char poc_entry_l1[8]; // ref pic list to ref pic entry map |
| 1111 |
unsigned int same_poc[2]; // set to 1 when the poc of cur ref pic and neighbor ref pic are same, otherwise set it to 0. |
| 1112 |
unsigned int act_stat_offset; // offset from start of "nvenc_rc_pic_stat_s" structure, points to data for external QP map |
| 1113 |
unsigned int stats_fifo_offset; // offset from start of top level "nvenc_stat_data_s" structure |
| 1114 |
} nvenc_h265_pic_control_s; // 536 bytes |
| 1115 |
|
| 1116 |
typedef struct |
| 1117 |
{ |
| 1118 |
// 256 aligned |
| 1119 |
unsigned int magic; // magic number, see text above // 4 bytes |
| 1120 |
nvenc_h265_surface_cfg_s refpic_cfg; // defines layout of reference pictures // 32 bytes |
| 1121 |
nvenc_h265_surface_cfg_s input_cfg; // defines layout of input pictures // 32 bytes |
| 1122 |
nvenc_h265_surface_cfg_s outputpic_cfg; // defines layout of reconstructed pictures // 32 bytes |
| 1123 |
nvenc_h265_sps_data_s sps_data; // 8 bytes |
| 1124 |
nvenc_h265_pps_data_s pps_data; // 8 bytes |
| 1125 |
nvenc_h265_rc_s rate_control; // Rate Control information // 96 bytes |
| 1126 |
nvenc_h265_pic_control_s pic_control; //536 bytes |
| 1127 |
nvenc_h265_surface_cfg_s half_scaled_outputpic_cfg; // defines layout of 2x2 subsampled reconstructed picture // 32 bytes |
| 1128 |
unsigned int gpTimer_timeout_val; // GPTimer cycle count set from driver 4 bytes |
| 1129 |
unsigned char reserved[240]; // Pad to 1024 bytes |
| 1130 |
|
| 1131 |
} nvenc_h265_drv_pic_setup_s; |
| 1132 |
|
| 1133 |
typedef struct |
| 1134 |
{ |
| 1135 |
unsigned short intra_refresh_cnt; // distance between forced-intra MBs in slice; 0 means OFF |
| 1136 |
unsigned short intra_refresh_offs; // first forced-intra MB in slice |
| 1137 |
|
| 1138 |
unsigned int h265_intra_luma4x4_mode_enable; //bitmask indicate which intra luma 4x4 modes to enable in ip1 intra search |
| 1139 |
unsigned int h265_intra_luma8x8_mode_enable; //bitmask indicate which intra luma 8x8 modes to enable in ip1 intra search |
| 1140 |
unsigned int h265_intra_luma16x16_mode_enable; //bitmask indicate which intra luma 16x16 modes to enable in ip1 intra search |
| 1141 |
unsigned int h265_intra_luma32x32_mode_enable; //bitmask indicate which intra luma 32x32 modes to enable in ip1 intra search |
| 1142 |
unsigned int h265_intra_luma64x64_mode_enable; //bitmask indicate which intra luma 64x64 modes to enable in ip1 intra search |
| 1143 |
unsigned int h265_intra_luma_mode_enable_leftbits : 30; //bitmask indicate intra luma left modes(32~34) from ip1 luma4x4 to ip2 luma64x64. |
| 1144 |
unsigned int reserved0 : 2; |
| 1145 |
|
| 1146 |
|
| 1147 |
unsigned char cu16x16_part_2Nx2N_enable : 1; // enable prediction for 2Nx2N |
| 1148 |
unsigned char cu16x16_part_2NxN_enable : 1; // enable prediction for 2NxN |
| 1149 |
unsigned char cu16x16_part_Nx2N_enable : 1; // enable prediction for Nx2N |
| 1150 |
unsigned char cu16x16_part_NxN_enable : 1; // enable prediction for NxN |
| 1151 |
unsigned char cu16x16_part_2NxnU_enable : 1; // enable prediction for 2NxnU |
| 1152 |
unsigned char cu16x16_part_2NxnD_enable : 1; // enable prediction for 2NxnD |
| 1153 |
unsigned char cu16x16_part_nLx2N_enable : 1; // enable prediction for nLx2N |
| 1154 |
unsigned char cu16x16_part_nRx2N_enable : 1; // enable prediction for nRx2N |
| 1155 |
|
| 1156 |
unsigned char cu32x32_part_2Nx2N_enable : 1; // enable prediction for 2Nx2N |
| 1157 |
unsigned char cu32x32_part_2NxN_enable : 1; // enable prediction for 2NxN |
| 1158 |
unsigned char cu32x32_part_Nx2N_enable : 1; // enable prediction for Nx2N |
| 1159 |
unsigned char cu32x32_part_NxN_enable : 1; // enable prediction for NxN |
| 1160 |
unsigned char cu32x32_part_2NxnU_enable : 1; // enable prediction for 2NxnU |
| 1161 |
unsigned char cu32x32_part_2NxnD_enable : 1; // enable prediction for 2NxnD |
| 1162 |
unsigned char cu32x32_part_nLx2N_enable : 1; // enable prediction for nLx2N |
| 1163 |
unsigned char cu32x32_part_nRx2N_enable : 1; // enable prediction for nRx2N |
| 1164 |
|
| 1165 |
unsigned char cu64x64_part_2Nx2N_enable : 1; // enable prediction for 2Nx2N |
| 1166 |
unsigned char cu64x64_part_2NxN_enable : 1; // enable prediction for 2NxN |
| 1167 |
unsigned char cu64x64_part_Nx2N_enable : 1; // enable prediction for Nx2N |
| 1168 |
unsigned char cu64x64_part_NxN_enable : 1; // enable prediction for NxN |
| 1169 |
unsigned char cu64x64_part_2NxnU_enable : 1; // enable prediction for 2NxnU |
| 1170 |
unsigned char cu64x64_part_2NxnD_enable : 1; // enable prediction for 2NxnD |
| 1171 |
unsigned char cu64x64_part_nLx2N_enable : 1; // enable prediction for nLx2N |
| 1172 |
unsigned char cu64x64_part_nRx2N_enable : 1; // enable prediction for nRx2N |
| 1173 |
|
| 1174 |
unsigned char intra_mode_4x4_enable : 1; // enable intra prediction in mdp for 4x4 |
| 1175 |
unsigned char intra_mode_8x8_enable : 1; // enable intra prediction in mdp for 8x8 |
| 1176 |
unsigned char intra_mode_16x16_enable : 1; // enable intra prediction in mdp for 16x16 |
| 1177 |
unsigned char intra_mode_32x32_enable : 1; // enable intra prediction in mdp for 32x32 |
| 1178 |
unsigned char intra_mode_64x64_enable : 1; // enable intra prediction in mdp for 32x32 |
| 1179 |
unsigned char priority_ipred_type_ip1 : 3; // which intra type search first for early termination, must be a subset of ip_search_mode. reserved, not implemented in HW |
| 1180 |
|
| 1181 |
unsigned int h265_intra_chroma_mode_enable : 5; //bitmask indicate which intra_chroma_pred_mode to enable |
| 1182 |
unsigned int inter_penalty_factor_for_ip1 : 6; // early teminate ip1 intra search if intra cost is larger than inter_cost + ((inter_cost * factor) >> 8) |
| 1183 |
unsigned int tusize_4x4_enable : 1; //enable 4x4 transform |
| 1184 |
unsigned int tusize_8x8_enable : 1; //enable 8x8 transform |
| 1185 |
unsigned int tusize_16x16_enable : 1; //enable 16x16 transform |
| 1186 |
unsigned int tusize_32x32_enable : 1; //enable 32x32 transform |
| 1187 |
unsigned int intra_tusize_decision : 1; //enable intra tu size decision, reserved, not implemented in HW |
| 1188 |
unsigned int pskip_enable : 1; //enable p_skip encoding |
| 1189 |
unsigned int tu_search_basedon_pusize : 1; //enable tu search based on pu size |
| 1190 |
unsigned int reserved1 : 14; // reserved for future special prediction extensions |
| 1191 |
|
| 1192 |
short bias_part_2Nx2N; //bias for 2Nx2N PU |
| 1193 |
short bias_part_2NxN; //bias for 2NxN PU |
| 1194 |
short bias_part_Nx2N; //bias for Nx2N PU |
| 1195 |
short bias_part_NxN; //bias for NxN PU |
| 1196 |
short bias_part_2NxnU; //bias for 2NxnU PU |
| 1197 |
short bias_part_2NxnD; //bias for 2NxnD PU |
| 1198 |
short bias_part_nLx2N; //bias for nLx2N PU |
| 1199 |
short bias_part_nRxN; //bias for nRxN PU |
| 1200 |
|
| 1201 |
short bias_cusize_8x8; //bias for 8x8 CU |
| 1202 |
short bias_cusize_16x16; //bias for 16x16 CU |
| 1203 |
short bias_cusize_32x32; //bias for 32x32 CU |
| 1204 |
short bias_cusize_64x64; //bias for 64x64 CU |
| 1205 |
short bias_pskip; // bias for pskip (subtracts from pskip cost) |
| 1206 |
|
| 1207 |
short bias_intra_over_inter; // bias for any intra (subtracts from intra cost) |
| 1208 |
short bias_intra_most_prob; // bias for intra most probable mode |
| 1209 |
short bias_qpp; |
| 1210 |
short bias_transform_skip; // bias for transform skip mode |
| 1211 |
|
| 1212 |
unsigned short mv_cost_bias; // bias applied to internal motion vector cost calculation |
| 1213 |
unsigned short pskip_esc_threshold; // if pskip cost is below this, disable full search & attempt pskip encode |
| 1214 |
|
| 1215 |
unsigned short tempmv_wt_spread_threshold; // The threshold for the sum of differences between 8x8s and 16x16 mvs. Used for temporal mv weight |
| 1216 |
unsigned int tempmv_wt_distort_threshold : 16; // The threshold for the cost of the 16x16 mv. Used for temporal mv weight |
| 1217 |
unsigned int mv_cost_predictor_control : 1; // 0=use 16x16 predictor only; 1=use accurate predictor (slower but better) |
| 1218 |
unsigned int mv_cost_enable : 1; // enable mv cost calculations in mode decision |
| 1219 |
unsigned int chroma_eval_mode : 1; // for intra chroma search 0 = U and V; 1 = U only |
| 1220 |
unsigned int ip_search_mode : 4; // which intra sizes to search before making IP decision b0=4x4,b1=8x8,b2=16x16,b3=32x32 |
| 1221 |
unsigned int multiply_bias_with_lambda : 1; // setting this to 1 will result in bias being multiplied by lambda |
| 1222 |
unsigned int force_ipcm : 1; // 0=normal encode; 1=force mpeb/mpec encode as IPCM |
| 1223 |
unsigned int early_termination_ip1 : 1; // enable IP1 early termination |
| 1224 |
unsigned int tu_search_num_cu16 : 3; // tu search part num in cu16 |
| 1225 |
unsigned int tu_search_num_cu32 : 3; // tu search part num in cu32 |
| 1226 |
unsigned int early_termination_mode_ip1; //enable intra mode search for early termination, 32 bits for the 35 modes other than H, V and DC, these 3 modes enabled as default |
| 1227 |
short bias_tu_4x4; // bias for TU 4x4 |
| 1228 |
short bias_tu_8x8; // bias for TU 8x8 |
| 1229 |
short bias_tu_16x16; // bias for TU 16x16 |
| 1230 |
short bias_tu_32x32; // bias for TU 32x32 |
| 1231 |
short num_1div8_lambda_ssd : 7; // number of 1/8 lambda ssd for lambda_coef delta. lambda_coef = (1+num_1div8_lambda_ssd*1/8)*lambda_ssd |
| 1232 |
short skip_evaluate_enable_cu8 : 1; // 0: not check skip ssd cost; 1, evaluate skip ssd cost |
| 1233 |
short skip_evaluate_enable_cu16 : 1; // 0: not check skip ssd cost; 1, evaluate skip ssd cost |
| 1234 |
short skip_evaluate_enable_cu32 : 1; // 0: not check skip ssd cost; 1, evaluate skip ssd cost |
| 1235 |
short skip_evaluate_enable_cu64 : 1; // 0: not check skip ssd cost; 1, evaluate skip ssd cost |
| 1236 |
short reserved2 : 1; |
| 1237 |
short intra_ssd_cnt_4x4 : 4; // SSD cost evaluation between N top intra modes selected by satd |
| 1238 |
short intra_ssd_cnt_8x8 : 4; // SSD cost evaluation between N top intra modes selected by satd |
| 1239 |
short intra_ssd_cnt_16x16 : 4; // SSD cost evaluation between N top intra modes selected by satd |
| 1240 |
short intra_ssd_cnt_32x32 : 4; // SSD cost evaluation between N top intra modes selected by satd |
| 1241 |
short rdo_level : 3; // 0:cu size decided by satd in P frame 1:cu size decided by ssd |
| 1242 |
short reserved3 : 1; // pad to short |
| 1243 |
|
| 1244 |
short bias_skip; // bias for skip mv's ssd (subtracts from skip mv's ssd cost) |
| 1245 |
short bias_intra_32x32; // bias for intra 32x32 (subtracts from intra 32x32 cost) |
| 1246 |
short bias_intra_16x16; // bias for intra 16x16 (subtracts from intra 16x16 cost) |
| 1247 |
short bias_intra_8x8; // bias for intra 8x8 (subtracts from intra 8x8 cost) |
| 1248 |
short bias_intra_4x4; // bias for intra 4x4 (subtracts from intra 4x4 cost) |
| 1249 |
unsigned char reserved[22]; |
| 1250 |
} nvenc_h265_md_control_s; // 128 bytes |
| 1251 |
|
| 1252 |
typedef struct |
| 1253 |
{ |
| 1254 |
unsigned int num_ctus : 19; // number of ctus in this slice, support upto 8kx8k |
| 1255 |
unsigned int qp_avr : 8; // 6bit value used by hw. Other bits can be used for rounding. |
| 1256 |
unsigned int reserved0 : 5; |
| 1257 |
|
| 1258 |
unsigned int dependent_slice_segment_flag : 1; // slice segment header syntax element present or not |
| 1259 |
unsigned int slice_type : 2; // 0 = B, 1 = P, 2 = I |
| 1260 |
unsigned int five_minus_max_num_merge_cand : 3; // specifies the maximum number of merging MVP candidates supported |
| 1261 |
unsigned int cabac_init_flag : 1; |
| 1262 |
unsigned int deblocking_filter_override_flag : 1; // specifies the deblocking control params is preset in slice header. |
| 1263 |
unsigned int slice_deblocking_filter_disabled_flag : 1; // if set to 1 , it disables deblcoking for the current slice. |
| 1264 |
// If deblocking_filter_override_flag is 0, this is set to pps_deblocking_filter_disabled_flag |
| 1265 |
unsigned int slice_loop_filter_across_slices_enabled_flag : 1; // if set to 1, it enables loop filter accross slice boundary. |
| 1266 |
unsigned int qp_slice_min : 8; // min slice qp value |
| 1267 |
unsigned int qp_slice_max : 8; // max slice qp value |
| 1268 |
unsigned int force_intra : 1; |
| 1269 |
unsigned int slice_sao_luma_flag : 1; // equal to 1 specifies that SAO is enabled for the luma component in the current slice |
| 1270 |
unsigned int slice_sao_chroma_flag : 1; // equal to 1 specifies that SAO is enabled for the chroma component in the current slice |
| 1271 |
unsigned int reserved1 : 3; |
| 1272 |
|
| 1273 |
int slice_beta_offset_div2 : 4; // if deblocking_filter_override_flag = 1 , then slice_beta_offset_div2 is signalled in slice segment header |
| 1274 |
int slice_tc_offset_div2 : 4; // if deblocking_filter_override_flag = 1 , then slice_tc_offset_div2 is signalled in slice segment header. |
| 1275 |
int slice_cb_qp_offset : 5; // offset added to pps_cb_qp_offset. It shall be in the range -12 to +12. |
| 1276 |
int slice_cr_qp_offset : 5; // offset added to pps_cr_qp_offset. It shall be in the range -12 to +12. |
| 1277 |
unsigned int slice_temporal_mvp_enabled_flag : 1; // set 1, temporal motion vector predictors may be used |
| 1278 |
unsigned int collocated_from_l0_flag : 1; // set 1, the collocated picture for tmvp is derived from ref. pic. list 0 |
| 1279 |
unsigned int collocated_ref_idx : 5; // specifies ref. index of the collocated picture for tmvp |
| 1280 |
int reserved2 : 7; |
| 1281 |
|
| 1282 |
unsigned int num_ref_idx_active_override_flag : 1; |
| 1283 |
unsigned int num_ref_idx_l0_active_minus1 : 5; // [0.31] |
| 1284 |
unsigned int num_ref_idx_l1_active_minus1 : 5; // [0,31] |
| 1285 |
unsigned int me_control_idx : 4; // index in array of nvenc_h264_me_control_s to use for this slice |
| 1286 |
unsigned int md_control_idx : 4; // index in array of nvenc_h264_md_control_s to use for this slice |
| 1287 |
unsigned int q_control_idx : 4; // index in array of nvenc_h264_quant_control_s to use for this slice |
| 1288 |
unsigned int wp_control_idx : 4; // index in array of nvenc_pred_weight_table_s to use for this slice |
| 1289 |
unsigned int reserved3 : 5; |
| 1290 |
|
| 1291 |
unsigned int limit_slice_top_boundary : 1; // limit ME to slice boundaries |
| 1292 |
unsigned int limit_slice_bot_boundary : 1; // limit ME to slice boundaries |
| 1293 |
unsigned int limit_slice_left_boundary : 1; // limit ME to slice boundaries |
| 1294 |
unsigned int limit_slice_right_boundary : 1; // limit ME to slice boundaries |
| 1295 |
unsigned int ROI_enable : 1; // Enable region of interest processing |
| 1296 |
unsigned int ROI_me_control_idx : 4; |
| 1297 |
unsigned int ROI_md_control_idx : 4; |
| 1298 |
unsigned int ROI_q_control_idx : 4; |
| 1299 |
unsigned int ROI_qp_delta : 8; // QP delta in ROI relative to picture outside ROI |
| 1300 |
unsigned int reserved4 : 7; // reserved for alignment |
| 1301 |
|
| 1302 |
unsigned int ROI_top_mbx : 16; // top left mbx of the ROI |
| 1303 |
unsigned int ROI_top_mby : 16; // top left mby of the ROI |
| 1304 |
unsigned int ROI_bot_mbx : 16; // bottom right mbx of the ROI |
| 1305 |
unsigned int ROI_bot_mby : 16; // bottom right mby of the ROI |
| 1306 |
|
| 1307 |
unsigned char reserved5[36]; // padding to 64 bytes |
| 1308 |
} nvenc_h265_slice_control_s; // 64 bytes |
| 1309 |
|
| 1310 |
// h265 not support QPP, reserved it but not used |
| 1311 |
typedef struct |
| 1312 |
{ |
| 1313 |
// QPP has not been defined for H265 |
| 1314 |
unsigned short qpp_run_vector_4x4; // cost values for 4x4 transform (16 bit total vector) |
| 1315 |
unsigned short qpp_run_vector_8x8[3]; // cost values for 8x8 transform (48 bit total vector) |
| 1316 |
// 2 bits each for first 12 coefs, then 1 bit each for next 24 |
| 1317 |
unsigned char qpp_luma8x8_cost; // luma 8x8 cost threshold, 0 = throw out all coefs, range 0-15 |
| 1318 |
unsigned char qpp_luma16x16_cost; // luma 16x16 cost threshold, 0 = throw out all coefs, range 0-15 |
| 1319 |
unsigned char qpp_chroma_cost; // chroma cost threshold, 0 = throw out all coefs, range 0-15 |
| 1320 |
unsigned char qpp_mode : 2; // 0 = OFF, 1 = 8x8, 2 = 16x16_8x8 |
| 1321 |
unsigned char reserved0 : 4; // padding |
| 1322 |
|
| 1323 |
unsigned int reserved1; // padding to 16 bytes |
| 1324 |
} nvenc_h265_quant_control_s; // 16 bytes |
| 1325 |
|
| 1326 |
//***************************************************************************** |
| 1327 |
// VP9 |
| 1328 |
//***************************************************************************** |
| 1329 |
|
| 1330 |
//vp9 filter config |
| 1331 |
typedef struct |
| 1332 |
{ |
| 1333 |
unsigned char filter_level[8]; // loop filter strength for each segment. |
| 1334 |
char ref_delta[4]; // Loop filter strength adjustments based on frame type (intra, inter) |
| 1335 |
char mode_delta[2]; // Loop filter strength adjustments based on mode (zero, new mv) |
| 1336 |
unsigned char disable : 1; // disable loop-filter |
| 1337 |
unsigned char sharpness : 3; // sharpness level |
| 1338 |
unsigned char baselevel_gt32 : 1; // filter base level greater than 32 flag |
| 1339 |
unsigned char reserved : 3; |
| 1340 |
unsigned char reserved2; |
| 1341 |
}nvenc_vp9_filter_cfg_s; //16 bytes |
| 1342 |
|
| 1343 |
typedef struct |
| 1344 |
{ |
| 1345 |
unsigned int round_chroma_ac : 3; // round for chroma AC |
| 1346 |
unsigned int round_chroma_dc : 3; // round for chroma DC |
| 1347 |
unsigned int round_luma_ac : 3; // round for luma AC |
| 1348 |
unsigned int round_luma_dc : 3; // round for luma DC |
| 1349 |
unsigned int round_delta_inter : 3; // round delta for inter |
| 1350 |
unsigned int reserved : 17; |
| 1351 |
}vp9_round_cfg_s; |
| 1352 |
|
| 1353 |
//vp9 quant config |
| 1354 |
typedef struct |
| 1355 |
{ |
| 1356 |
unsigned char qindex_seg[8]; // Base frame qindex (index in quant lookup table) for each segment 0...255 |
| 1357 |
int delta_y_dc : 5; // explicit qindex adjustment for y dccoefficient, -15...15 |
| 1358 |
int delta_uv_dc : 5; // qindex adjustment for uv accoefficient, -15...15 |
| 1359 |
int delta_uv_ac : 5; // qindex adjustment for uv dccoefficient, -15...15 |
| 1360 |
int reserved : 17; |
| 1361 |
//round parameters |
| 1362 |
vp9_round_cfg_s round_seg[8]; // round config for each segment |
| 1363 |
}nvenc_vp9_quant_cfg_s; //44 bytes |
| 1364 |
|
| 1365 |
// vp9 dma config for each tile |
| 1366 |
typedef struct |
| 1367 |
{ |
| 1368 |
unsigned int tile_width : 6; // tile width in super blocks(64x64) |
| 1369 |
unsigned int tile_height : 6; // tile height in super blocks(64x64) |
| 1370 |
unsigned int tile_pos_x : 6; // tile x-position in super blocks unit |
| 1371 |
unsigned int reserved : 14; |
| 1372 |
}nvenc_vp9_modedma_cfg_s; |
| 1373 |
|
| 1374 |
//vp9 smalle config |
| 1375 |
typedef struct |
| 1376 |
{ |
| 1377 |
unsigned int high_prec_mv_enable : 1; // high precision mv enable |
| 1378 |
unsigned int temporal_mv_enable : 1; // temporal mv enable |
| 1379 |
unsigned int transform_mode : 3; // transform mode: 0: only4x4, 1: allow_8x8, 2: allow_16x16, 3: allow_32x32, 4:transform_mode_select |
| 1380 |
unsigned int lossless_enable : 1; // lossless encoding enable |
| 1381 |
unsigned int mocomp_filter_type : 3; // mocomp filter type:0 eight-tap smooth, 1: eight-tap, 2:eight-tap sharp, 3: bilinear, 4:switchable |
| 1382 |
unsigned int tile_enable : 1; // enable tile encoding |
| 1383 |
unsigned int segment_enable : 1; // enable segment |
| 1384 |
unsigned int segment_upd_enable : 1; // enable explicit map update for current frame. If true, then segment_temp_enable says how to update |
| 1385 |
unsigned int segment_temp_enable : 1; // Type of map update |
| 1386 |
unsigned int reserved : 19; |
| 1387 |
}nvenc_vp9_smalle_config_s; |
| 1388 |
|
| 1389 |
//vp9 frame-level parameters |
| 1390 |
typedef struct |
| 1391 |
{ |
| 1392 |
unsigned int pic_width : 13; // picture width in pixels |
| 1393 |
unsigned int pic_height : 13; // picture height in pixels |
| 1394 |
unsigned int rec_disable : 1; // disable recon output |
| 1395 |
unsigned int type : 1; // picture type: 0 Intra/1 Inter |
| 1396 |
unsigned int refframe : 2; // reference frame type used 1 = last 2= golden 3= alt-ref |
| 1397 |
unsigned int reserved : 2; |
| 1398 |
|
| 1399 |
//ref-frame param |
| 1400 |
unsigned char ref_index[8]; // ref-frame index for each segment |
| 1401 |
//skip param |
| 1402 |
unsigned char skip_mode[8]; // skip-mode enable for each segment: skip mode implies both the use of a (0,0) motion vector and that no residual will be coded |
| 1403 |
//filter param |
| 1404 |
nvenc_vp9_filter_cfg_s filter_cfg; // filter config |
| 1405 |
//quant param |
| 1406 |
nvenc_vp9_quant_cfg_s quant_cfg; // quant & round config |
| 1407 |
|
| 1408 |
nvenc_vp9_smalle_config_s smalle_cfg; // small-e config |
| 1409 |
unsigned int num_tile_controls_minus1; // number of nvenc_vp9_tile_control_s entries in array minus1 |
| 1410 |
unsigned int tile_controls_offset; // offset from start of top level "nvenc_vp9_pic_control_s" structure |
| 1411 |
// to start of array of "nvenc_vp9_tile_control_s" structures |
| 1412 |
|
| 1413 |
unsigned int bitstream_buf_start_pos; // bitstream buffer offset start position byte offset |
| 1414 |
unsigned int bitstream_buf_size; // bitstream buffer size in bytes |
| 1415 |
|
| 1416 |
unsigned int combined_linebuf_size; // combined line buffer size in bytes |
| 1417 |
|
| 1418 |
unsigned int filterline_linebuf_size; // filter line line-buffer size in bytes |
| 1419 |
|
| 1420 |
unsigned int filtercol_linebuf_size; // filter column line-buffer size in bytes |
| 1421 |
}nvenc_vp9_pic_control_s; //112 bytes |
| 1422 |
|
| 1423 |
typedef struct |
| 1424 |
{ |
| 1425 |
unsigned int col_pos : 6; // tile position in Super Block of Y axis |
| 1426 |
unsigned int bs_size : 24; // tile size in Super Block |
| 1427 |
unsigned int reserved : 2; |
| 1428 |
nvenc_vp9_modedma_cfg_s mode_dma_cfg; // mode-dma config for this tile |
| 1429 |
unsigned char tile_idx; // tile index in the picture |
| 1430 |
unsigned char reserved2[55]; |
| 1431 |
}nvenc_vp9_tile_control_s; //64bytes |
| 1432 |
|
| 1433 |
typedef struct |
| 1434 |
{ |
| 1435 |
unsigned char QP[3]; // QP for 0:P picture, ,1:B picture 2; I picture |
| 1436 |
unsigned char minQP[3]; // min QP for 0:P picture, ,1:B picture 2; I picture |
| 1437 |
unsigned char maxQP[3]; // max QP for 0:P picture, ,1:B picture 2; I picture |
| 1438 |
unsigned char reserved[3]; |
| 1439 |
int rhopbi[3]; // 23.8 signed fixed point quant ratios P/I, B/I, I/I(=1) |
| 1440 |
|
| 1441 |
int framerate; // fps |
| 1442 |
unsigned int buffersize; // total buffer size |
| 1443 |
|
| 1444 |
int nal_cpb_size; // size in bits |
| 1445 |
int nal_bitrate; // rate in bps |
| 1446 |
|
| 1447 |
int gop_length; // I period |
| 1448 |
int Bmin; // 23.8 signed fixed point min buffer level, updated by driver at sequence level only. |
| 1449 |
int Ravg; // 23.8 signed fixed point average rate, updated by driver at sequence level only. |
| 1450 |
int R; // 23.8 signed fixed point rate |
| 1451 |
|
| 1452 |
|
| 1453 |
} nvenc_vp9_rc_s; // 56 bytes, currently keep the same as vp8 |
| 1454 |
|
| 1455 |
|
| 1456 |
typedef struct |
| 1457 |
{ |
| 1458 |
unsigned int picture_index; // value received from SetPictureIndex method |
| 1459 |
unsigned int error_status : 2; // report error if any |
| 1460 |
unsigned int ucode_error_status : 30; // report error status from ucode to driver |
| 1461 |
unsigned int total_byte_count; // total bitstream size in bytes |
| 1462 |
unsigned int bitstream_start_pos; // byte_offset where VP9 BS DMA start writing bitstream |
| 1463 |
unsigned short pic_type; // inter/key frame type |
| 1464 |
unsigned short avgQP; // average QP |
| 1465 |
unsigned int cycle_count; // total cycles taked for execute. Written out only when DumpCycleCount in SetControlParams is set |
| 1466 |
unsigned int intra_cu32x32_count; // number of intra 32x32 CUs. |
| 1467 |
unsigned int inter_cu32x32_count; // number of inter 32x32 CUs. |
| 1468 |
unsigned int intra_cu16x16_count; // number of intra 16x16 CUs. |
| 1469 |
unsigned int inter_cu16x16_count; // number of inter 16x16 CUs. |
| 1470 |
unsigned int intra_cu8x8_count; // number of intra 8x8 CUs. |
| 1471 |
unsigned int inter_cu8x8_count; // number of inter 8x8 CUs. |
| 1472 |
unsigned int cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 1473 |
unsigned int cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 1474 |
unsigned int total_intra_cost; // sum of all intra best mode's SATDs, each mb has one intra SATD. |
| 1475 |
unsigned int total_inter_cost; // sum of all inter best mode's SATDs, each mb has one inter SATD. |
| 1476 |
|
| 1477 |
} nvenc_vp9_pic_stat_s; // 64 bytes |
| 1478 |
|
| 1479 |
// **************************************************************************** |
| 1480 |
// **************************************************************************** |
| 1481 |
// VP8 |
| 1482 |
// **************************************************************************** |
| 1483 |
// **************************************************************************** |
| 1484 |
|
| 1485 |
// TO DO: Find out if we should define # maxSegments as 4 |
| 1486 |
#define MAXSEGMENTS 4 |
| 1487 |
|
| 1488 |
typedef struct |
| 1489 |
{ |
| 1490 |
unsigned short frame_width_minus1; // frame width in pixels minus 1, range 0-4095 |
| 1491 |
unsigned short frame_height_minus1; // frame height in pixels minus 1, range 0-4095 |
| 1492 |
|
| 1493 |
unsigned int sfc_pitch : 10; // pitch of luma plane in macorblock units( i.e stride >>4) |
| 1494 |
unsigned int sfc_pitch_chroma : 10; // pitch of chroma plane in macroblock units |
| 1495 |
unsigned int reserved2 : 12; |
| 1496 |
|
| 1497 |
unsigned int sfc_trans_mode : 4; // least 3 significient bits are used to stand for 8 modes, normal, xflip, yflip etc. |
| 1498 |
unsigned int block_height : 7; // BL mapping block height setting |
| 1499 |
unsigned int tiled_16x16 : 1; // Surface is 16x16 tiled instead of BL mapped (must be 1 for refpics) |
| 1500 |
unsigned int memory_mode : 2; // 0: semi-planar, 1: planar |
| 1501 |
unsigned int input_bl_mode : 2; // the input block linear mode: 0~gpu bl; 1~tegra bl; 2~naive bl this is only for cmod and cmod/plugin |
| 1502 |
unsigned int reserved : 16; // pad to int |
| 1503 |
|
| 1504 |
unsigned int luma_top_frm_offset; // offset of luma top field or frame in units of 256 |
| 1505 |
unsigned int luma_bot_offset; // offset of luma bottom field in units of 256. Not used if frame format. |
| 1506 |
unsigned int chroma_top_frm_offset; // offset of chroma top field or frame, or offset of chroma U(Cb) plane in planar mode, both in units of 256 |
| 1507 |
unsigned int chroma_bot_offset; // offset of chroma bottom field, not used if frame format. Or offset of chroma V(Cr) plane in planar mode. Both in units of 256. |
| 1508 |
|
| 1509 |
} nvenc_vp8_surface_cfg_s; // 28 bytes |
| 1510 |
|
| 1511 |
typedef struct |
| 1512 |
{ |
| 1513 |
//unsigned int coeff_partitioning_enabled : 1; // 0: coefficient partitioning disabled 1: coeffient partitioning enabled (SE: redundant, same as multi_token_partition!=0) |
| 1514 |
//unsigned int num_segments_minus_one : 2; // number of segments |
| 1515 |
unsigned int horiz_scale : 2; // 0=1, 1=5/4, 2=5/3, 3=2 |
| 1516 |
unsigned int vert_scale : 2; // 0=1, 1=5/4, 2=5/3, 3=2 |
| 1517 |
//unsigned int golden_frame_enable : 1; // 0: golden frame disabled, 1: golden frame enabled |
| 1518 |
//unsigned int alt_ref_frame_enable : 1; // 0: alt_ref_frame_disabled, 1: alt_ref_frame_enabled |
| 1519 |
//unsigned int adaptive_golden_frame_selection : 1; // 0: fixed distance between golden frames 1: adaptive golden frame selection |
| 1520 |
//unsigned int adaptive_altref_frame_selection : 1; // 0: fixed distance between alt ref frames 1: adaptive alt ref frame selection |
| 1521 |
unsigned int reserved : 28; // Align to word boundary |
| 1522 |
|
| 1523 |
//unsigned int golden_frame_interval : 16; // golden frame interval if adaptive_golden_frame_selection is fixed |
| 1524 |
//unsigned int altref_frame_interval : 16; // altref frame interval if adaptive_altref_frame_selection is fixed |
| 1525 |
|
| 1526 |
} nvenc_vp8_sequence_data_s; // 4 bytes |
| 1527 |
|
| 1528 |
typedef struct |
| 1529 |
{ |
| 1530 |
unsigned int frame_type : 1; // 0=key frame, 1=interframe |
| 1531 |
unsigned int version : 3; // VP8 version, also affects luma/chroma mocomp: 0=bicubic/bicubic, 1,2=bilinear/bilinear, 3..7=bilinear/fullpel (SE: perhaps separate flags for bilinear/fullpel?) |
| 1532 |
unsigned int show_frame : 1; // 0=don't display, 1=display |
| 1533 |
unsigned int clr_type : 1; // color space: 0=ITU-R BT.601, 1=reserved (multiplier free) |
| 1534 |
unsigned int clamp_type : 1; // 0=clamping of reconstructed samples required, 1=no clamping necessary (SE: not supported by NVENC?) |
| 1535 |
// <loop filter parameters> |
| 1536 |
unsigned int filter_type : 2; // 0=normal, 1=simple 2: disable (SE: filter_type==2 is redundant, same as filter_level==0) |
| 1537 |
unsigned int filter_level : 6; // loop filter strength (1..63), 0=disabled |
| 1538 |
unsigned int sharpness_level : 3; // loop filter sharpness (0..7) |
| 1539 |
unsigned int mode_ref_lf_delta_enabled : 1; // apply mb param dependent loop filter level deltas: 0=disabled, 1=enabled; if disabled the following parameters are not used |
| 1540 |
unsigned int mode_ref_lf_delta_update : 1; // update ref_lf_deltas and mode_lf_deltas: 0=keep previous, 1=update |
| 1541 |
unsigned int ref_lf_deltas_update_0 : 1; // update ref_lf_deltas[0]: 0=keep previous, 1=update |
| 1542 |
unsigned int ref_lf_deltas_update_1 : 1; // update ref_lf_deltas[1]: 0=keep previous, 1=update |
| 1543 |
unsigned int ref_lf_deltas_update_2 : 1; // update ref_lf_deltas[2]: 0=keep previous, 1=update |
| 1544 |
unsigned int ref_lf_deltas_update_3 : 1; // update ref_lf_deltas[3]: 0=keep previous, 1=update |
| 1545 |
unsigned int mode_lf_deltas_update_0 : 1; // update mode_lf_deltas[0]: 0=keep previous, 1=update |
| 1546 |
unsigned int mode_lf_deltas_update_1 : 1; // update mode_lf_deltas[1]: 0=keep previous, 1=update |
| 1547 |
unsigned int mode_lf_deltas_update_2 : 1; // update mode_lf_deltas[2]: 0=keep previous, 1=update |
| 1548 |
unsigned int mode_lf_deltas_update_3 : 1; // update mode_lf_deltas[3]: 0=keep previous, 1=update |
| 1549 |
unsigned int reserved0 : 4; // Align to word boundary |
| 1550 |
signed char ref_lf_deltas[4]; // reference frame dependent loop filter level delta (-63..+63) |
| 1551 |
// Index 0: macroblock is INTRA |
| 1552 |
// 1: macroblock is INTER and reference frame is LAST |
| 1553 |
// 2: macroblock is INTER and reference frame is GOLDEN |
| 1554 |
// 3: macroblock is INTER and reference frame is ALTREF |
| 1555 |
// lower 7 bits of each entry used as a signed value -63..+63, MS Bit is ignored by HW (SE: invalid values should be rejected?) |
| 1556 |
signed char mode_lf_deltas[4]; // macroblock mode dependent loop filter level delta (-63..+63) |
| 1557 |
// Index 0: B_PRED (macroblock is INTRA 4x4) |
| 1558 |
// 1: ZEROMV (macroblock is INTER and mode is ZEROMV) |
| 1559 |
// 2: MV (macroblock is INTER and mode is neither ZEROMV nor SPLITMV) |
| 1560 |
// 3: SPLITMV (macroblock is INTER and mode is SPLITMV (not 16x16)) |
| 1561 |
// lower 7 bits of each entry used as a signed value -63..+63, MS Bit is ignored by HW (SE: invalid values should be rejected?) |
| 1562 |
// Other frame header parameters |
| 1563 |
unsigned int multi_token_partition : 2; // number of tcoeff token partitions (0,1,2,3 -> 1,2,4,8): num_part = 1 << multi_token_partition |
| 1564 |
unsigned int refresh_entropy_probs : 1; // 0=restore previous probs at end of frame, 1=keep updated probs for next frame |
| 1565 |
unsigned int mb_no_coeff_skip : 1; // per-mb skip flag, 0=disable 1:enable (misnomer?) |
| 1566 |
unsigned int update_ymode_prob : 1; // update ymode probs, 0=no, 1=yes |
| 1567 |
unsigned int update_uvmode_prob : 1; // update uvmode probs, 0=no, 1=yes |
| 1568 |
//unsigned int coef_update_probs : 1; // update tcoeff token probs, 0=no, 1=yes (SE: isnt't this redundant with coef_probs_update array?) |
| 1569 |
|
| 1570 |
unsigned int base_qindex : 7; // quantizer index (0..127) |
| 1571 |
|
| 1572 |
unsigned int refresh_last_frame : 1; // use decoded frame as last reference frame |
| 1573 |
unsigned int refresh_golden_frame : 1; // use decoded frame as golden reference frame |
| 1574 |
unsigned int refresh_alt_ref_frame : 1; // use decoded frame as alt_ref reference frame |
| 1575 |
unsigned int copy_buffer_to_gf : 2; // update golden frame from: 0=none (keep), 1=last_frame, 2=alt_ref_frame |
| 1576 |
unsigned int copy_buffer_to_arf : 2; // update alt_ref frame from: 0=none (keep), 1=last_frame, 2=golden_frame |
| 1577 |
unsigned int ref_frame_sign_bias_gf : 1; // golden frame is future frame (invert mv sign) |
| 1578 |
unsigned int ref_frame_sign_bias_arf : 1; // alt_ref frame is future frame (invert mv sign) |
| 1579 |
unsigned int fixed_ref_frame_probs : 1; //use fixed ref frame probability |
| 1580 |
unsigned int reserved1 : 9; // Align to word boundary |
| 1581 |
|
| 1582 |
|
| 1583 |
signed int y1dc_delta_q : 5; // quantizer delta for luma DC (-15..+15) |
| 1584 |
signed int y2dc_delta_q : 5; // quantizer delta for WHT DC (-15..+15) |
| 1585 |
signed int y2ac_delta_q : 5; // quantizer delta for WHT AC (-15..+15) |
| 1586 |
signed int uvdc_delta_q : 5; // quantizer delta for chroma DC (-15..+15) |
| 1587 |
signed int uvac_delta_q : 5; // quantizer delta for chroma AC (-15..+15) |
| 1588 |
|
| 1589 |
// <segmentation parameters> |
| 1590 |
unsigned int segmentation_enabled : 1; // 0=disabled, 1=enabled; if disabled the following parameters are not used |
| 1591 |
unsigned int update_mb_segmentation_map : 1; // per-mb segment ID map update: 0=keep previous, 1=update (SE: have to provide the actual map) |
| 1592 |
unsigned int update_mb_segmentation_data : 1; // mb_segment_abs_delta, seg_filter_level, and seg_qindex update: 0=keep previous, 1=update |
| 1593 |
unsigned int mb_segment_abs_delta : 1; // seg_filter_level and seg_qindex are: 0=delta values, 1=absolute values |
| 1594 |
unsigned int reserved2 : 3; // Align to word boundary |
| 1595 |
signed char seg_filter_level[MAXSEGMENTS]; // per-segment loop filter levels |
| 1596 |
// if mb_segment_abs_delta==0: relative to filter_level (-63..+63) |
| 1597 |
// if mb_segment_abs_delta==1: absolute value (0..63) |
| 1598 |
signed char seg_qindex[MAXSEGMENTS]; // per-segment qindex values |
| 1599 |
// if mb_segment_abs_delta==0: relative to base_qindex (-127..+127) |
| 1600 |
// if mb_segment_abs_delta==1: absolute value (0..127) |
| 1601 |
|
| 1602 |
} nvenc_vp8_picture_data_s; // 32 bytes |
| 1603 |
|
| 1604 |
typedef struct |
| 1605 |
{ |
| 1606 |
unsigned int early_intra_mode_control : 2; // 0=inter; 1=eval_intra; 2=ext_hint; 3=none (low perf,best qual) |
| 1607 |
unsigned int early_intra_mode_type_16x16dc : 1; // evaluate 16x16dc if early_intra_mode_control==1 |
| 1608 |
unsigned int early_intra_mode_type_16x16h : 1; // evaluate 16x16h if early_intra_mode_control==1 |
| 1609 |
unsigned int early_intra_mode_type_16x16v : 1; // evaluate 16x16v if early_intra_mode_control==1 |
| 1610 |
unsigned int early_ip_is_final : 1; // if set, result of early intra is final (disable full intra search) |
| 1611 |
unsigned int chroma_eval_mode : 1; // for intra chroma search 0 = U and V; 1 = U only |
| 1612 |
unsigned int nearest_mv_enable : 1; // enable nearest mv mode for 16x16 |
| 1613 |
unsigned int near_mv_enable : 1; // enable near mv mode for 16x16 |
| 1614 |
unsigned int zero_mv_enable : 1; // enable zero mv mode 16x16 |
| 1615 |
unsigned int new_mv_enable : 1; // enable new mv mode 16x16 |
| 1616 |
unsigned int above4x4_mv_enable : 1; // enable above4x4 mv mode for VP8 SPLITMV modes |
| 1617 |
unsigned int left4x4_mv_enable : 1; // enable left4x4 mv mode for VP8 SPLITMV modes |
| 1618 |
unsigned int zero4x4_mv_enable : 1; // enable zero4x4 mv mode for VP8 SPLITMV modes |
| 1619 |
unsigned int new4x4_mv_enable : 1; // enable new4x4 mv mode for VP8 SPLITMV modes |
| 1620 |
unsigned int l0_part_16x16_enable : 1; // enable L0 prediction for 16x16 |
| 1621 |
unsigned int l0_part_16x8_enable : 1; // enable L0 prediction for 16x8 |
| 1622 |
unsigned int l0_part_8x16_enable : 1; // enable L0 prediction for 8x16 |
| 1623 |
unsigned int l0_part_8x8_enable : 1; // enable L0 prediction for 8x8 |
| 1624 |
unsigned int pskip_enable : 1; // enable p_skip encoding (still used by ME for special modes, can this be removed?) |
| 1625 |
//unsigned int use_variance : 1; // to decide between inter and intra (Most likely wont be used) |
| 1626 |
unsigned int explicit_eval_nearestmv : 1; // enable explicit search vp8 nearest mv |
| 1627 |
unsigned int explicit_eval_nearmv : 1; // enable explicit search vp8 near mv |
| 1628 |
unsigned int explicit_eval_left_topmv : 1; // enable explicit search vp8 8x8 left_top mv |
| 1629 |
unsigned int explicit_eval_left_bottommv : 1; // enable explicit search vp8 8x8 left_bottom mv |
| 1630 |
unsigned int explicit_eval_top_leftmv : 1; // enable explicit search vp8 8x8 top_left mv |
| 1631 |
unsigned int explicit_eval_top_rightmv : 1; // enable explicit search vp8 8x8 top_right mv |
| 1632 |
unsigned int explicit_eval_zeromv : 1; // enable explicit search vp8 zero mv |
| 1633 |
unsigned int refframe_pic : 2; // reference frame type used 1 = last 2= golden 3= alt-ref |
| 1634 |
unsigned int reserved0 : 3; // Pad to int |
| 1635 |
|
| 1636 |
unsigned int intra_luma4x4_mode_enable : 9; // bitmask indicating which intra luma 4x4 modes to enable |
| 1637 |
unsigned int intra_luma16x16_mode_enable : 4; // bitmask indicating which intra luma 16x16 modes to enable |
| 1638 |
unsigned int intra_chroma_mode_enable : 4; // bitmask indicating which intra chroma modes to enable |
| 1639 |
unsigned int intra_tm_mode_enable : 1; // bitmask indicating intra 4x4 true motion mode enable |
| 1640 |
unsigned int ip_search_mode : 3; // which intra sizes to search before making IP decision b0=4x4,b2=16x16 // We dont need a bit for 8x8 for VP8 but it might be simpler for reg field to keep it? |
| 1641 |
unsigned int mv_cost_enable : 1; // enable mv cost calculations in mode decision |
| 1642 |
unsigned int reserved1 : 10; // pad to full int |
| 1643 |
|
| 1644 |
unsigned short early_intra_disable_mpeb_threshold; // if early intra cost is below this, disable full search & select intra |
| 1645 |
//unsigned short lambda_multiplier_mvcost; // used for mv cost calculation |
| 1646 |
short bias_favor_intra_16x16_early; // bias for early intra (subtracts from early intra 16x16 cost) |
| 1647 |
|
| 1648 |
unsigned short lambda_multiplier_intra[4]; // used for intra mode cost calculation [segment_id] |
| 1649 |
|
| 1650 |
unsigned int mdpLambdaCoefMd :16; |
| 1651 |
unsigned int mdpLambdaCoefMv :16; |
| 1652 |
|
| 1653 |
short bias_intra_16x16; // bias for intra 16x16 (subtracts from intra 16x16 cost) |
| 1654 |
short bias_intra_4x4; // bias for intra 4x4 (subtracts from intra 4x4 cost) |
| 1655 |
|
| 1656 |
short bias_inter_16x16; // bias for inter 16x16 (subtracts from inter 16x16 cost) |
| 1657 |
short bias_inter_16x8; // bias for inter 16x8 (subtracts from inter 16x8 cost) |
| 1658 |
|
| 1659 |
short bias_inter_8x16; // bias for inter 8x16 (subtracts from inter 8x16 cost) |
| 1660 |
short bias_inter_8x8; // bias for inter 8x8 (subtracts from inter 8x8 cost) |
| 1661 |
|
| 1662 |
short bias_nearmv; // bias for nearmv (subtracts from nearmv cost) for 16x16 |
| 1663 |
short bias_nearestmv; // bias for nearestmv (subtracts from nearestmv cost) for 16x16 |
| 1664 |
|
| 1665 |
short bias_zeromv; // bias for zeromv (subtracts from zeromv cost) for 16x16 |
| 1666 |
short bias_newmv; // bias for newmv (subtracts from newmv cost) for 16x16 |
| 1667 |
|
| 1668 |
short bias_above4x4mv; // bias for above4x4mv (subtracts from above4x4 cost) for VP8 SPLITMV modes |
| 1669 |
short bias_left4x4mv; // bias for left4x4mv (subtracts from left4x4 cost) for VP8 SPLITMV modes |
| 1670 |
|
| 1671 |
short bias_zero4x4mv; // bias for zero4x4mv (subtracts from zero4x4 cost) for VP8 SPLITMV modes |
| 1672 |
short bias_new4x4mv; // bias for new4x4mv (subtracts from new4x4 cost) for VP8 SPLITMV modes |
| 1673 |
|
| 1674 |
short mv_cost_bias; // bias applied to internal motion vector cost calculation |
| 1675 |
short bias_intra_over_inter; // bias for any intra (subtracts from intra cost) |
| 1676 |
unsigned short intra_refresh_cnt; // distance between forced-intra MBs; 0 means OFF |
| 1677 |
unsigned short intra_refresh_offs; // first forced-intra MB |
| 1678 |
|
| 1679 |
unsigned short tempmv_wt_spread_threshold; // The threshold for the sum of difference between 8x8 and 16x16 mvs. Used for temporal mv weight. |
| 1680 |
unsigned short tempmv_wt_distort_threshold; // The threshold for cost of the 16x16 mv. Used for temporal mv weight. |
| 1681 |
} nvenc_vp8_md_control_s; // 64 bytes |
| 1682 |
|
| 1683 |
typedef struct |
| 1684 |
{ |
| 1685 |
unsigned int self_temporal_stamp_l0 : 3; // stamp to use for L0 integer search in stamp based mode |
| 1686 |
//unsigned int self_temporal_stamp_l1 : 3; // stamp to use for L1 integer search in stamp based mode |
| 1687 |
unsigned int self_temporal_explicit : 1; // explicitly evaluate these vectors |
| 1688 |
unsigned int self_temporal_search : 1; // integer search enable for this hint |
| 1689 |
unsigned int self_temporal_refine : 1; // subpel search enable for this hint |
| 1690 |
unsigned int self_temporal_enable : 1; // hint enable (enables processing + fetching of data) (only changeable at first slice in picture) |
| 1691 |
|
| 1692 |
unsigned int coloc_stamp_l0 : 3; // stamp to use for L0 integer search in stamp based mode |
| 1693 |
//unsigned int coloc_stamp_l1 : 3; // stamp to use for L1 integer search in stamp based mode |
| 1694 |
unsigned int coloc_explicit : 1; // explicitly evaluate these vectors |
| 1695 |
unsigned int coloc_search : 1; // integer search enable for this hint |
| 1696 |
unsigned int coloc_refine : 1; // subpel search enable for this hint |
| 1697 |
unsigned int coloc_enable : 1; // hint enable (enables processing + fetching of data) (only changeable at first slice in picture) |
| 1698 |
|
| 1699 |
unsigned int self_spatial_stamp_l0 : 3; // stamp to use for L0 integer search in stamp based mode |
| 1700 |
//unsigned int self_spatial_stamp_l1 : 3; // stamp to use for L1 integer search in stamp based mode |
| 1701 |
unsigned int self_spatial_explicit : 1; // explicitly evaluate these vectors |
| 1702 |
unsigned int self_spatial_search : 1; // integer search enable for this hint |
| 1703 |
unsigned int self_spatial_refine : 1; // subpel search enable for this hint |
| 1704 |
unsigned int self_spatial_enable : 1; // hint enable (enables processing + fetching of data) (only changeable at first slice in picture) |
| 1705 |
|
| 1706 |
unsigned int reserved0 : 11; // Pad to int |
| 1707 |
|
| 1708 |
unsigned int external_stamp_lastref_stamp : 3; // Stamp to use for external L0 hint with refidx = 0 |
| 1709 |
unsigned int external_stamp_goldref_stamp : 3; // Stamp to use for external L0 hint with refidx = 1 |
| 1710 |
unsigned int external_stamp_altref_stamp : 3; // Stamp to use for external L0 hint with refidx = 2 |
| 1711 |
unsigned int external_explicit : 1; // explicitly evaluate these vectors |
| 1712 |
unsigned int external_search : 1; // integer search enable for this hint |
| 1713 |
unsigned int external_refine : 1; // subpel search enable for this hint |
| 1714 |
unsigned int external_enable : 1; // hint enable (enables processing + fetching of data) (only changeable at first slice in picture) |
| 1715 |
|
| 1716 |
unsigned int const_mv_stamp_l0 : 3; // stamp to use for L0 integer search in stamp based mode |
| 1717 |
//unsigned int const_mv_stamp_l1 : 3; // stamp to use for L1 integer search in stamp based mode |
| 1718 |
unsigned int const_mv_explicit : 1; // explicitly evaluate these vectors |
| 1719 |
unsigned int const_mv_search : 1; // integer search enable for this hint |
| 1720 |
unsigned int const_mv_refine : 1; // subpel search enable for this hint |
| 1721 |
unsigned int const_mv_enable : 1; // hint enable (enables processing + fetching of data) |
| 1722 |
|
| 1723 |
unsigned int reserved1 : 12; // pad to int |
| 1724 |
|
| 1725 |
} nvenc_vp8_me_hint_cfg_s; // 8 bytes |
| 1726 |
|
| 1727 |
// Same as H264 me_limit_mv struct |
| 1728 |
typedef struct |
| 1729 |
{ |
| 1730 |
unsigned int mv_limit_enable : 1; // 0=disable, 1=enable set the absolute MV range (Cisco flip camera request) |
| 1731 |
unsigned int left_mvx_frac : 2; // vector X component fraction |
| 1732 |
int left_mvx_int : 12; // vector X component integer portion |
| 1733 |
unsigned int reserved1 : 1; // padding X to 16 bit |
| 1734 |
unsigned int top_mvy_frac : 2; // vector Y component fraction |
| 1735 |
int top_mvy_int : 10; // vector Y component integer portion |
| 1736 |
unsigned int reserved2 : 4; // padding Y to 16 bit |
| 1737 |
|
| 1738 |
unsigned int right_mvx_frac : 2; // vector X component fraction |
| 1739 |
int right_mvx_int : 12; // vector X component integer portion |
| 1740 |
unsigned int reserved3 : 2; // padding X to 16 bit |
| 1741 |
unsigned int bottom_mvy_frac : 2; // vector Y component fraction |
| 1742 |
int bottom_mvy_int : 10; // vector Y component integer portion |
| 1743 |
unsigned int reserved4 : 4; // padding Y to 16 bit |
| 1744 |
|
| 1745 |
} nvenc_vp8_me_limit_mv_s; // 8 bytes |
| 1746 |
|
| 1747 |
typedef struct |
| 1748 |
{ |
| 1749 |
unsigned int mvx_frac : 2; // vector X component fraction |
| 1750 |
int mvx_int : 12; // vector X component integer portion |
| 1751 |
unsigned int reserved_x : 2; // padding X to 16 bit |
| 1752 |
unsigned int mvy_frac : 2; // vector Y component fraction |
| 1753 |
int mvy_int : 10; // vector Y component integer portion |
| 1754 |
unsigned int reserved_y : 4; // padding Y to 16 bit |
| 1755 |
|
| 1756 |
} nvenc_vp8_me_const_vec_s; // 4 bytes |
| 1757 |
|
| 1758 |
// TO DO: We are wasting ~7 ints here we can combine this to pack hor_adjust and ver_adjust in diff int |
| 1759 |
// but dont do it if we are going to throw it away in padding anyways |
| 1760 |
typedef struct |
| 1761 |
{ |
| 1762 |
unsigned int bitmask[2] ; // 8x8 search point pattern bitmask |
| 1763 |
unsigned int hor_adjust : 1; // shift hor based on lsb of pred |
| 1764 |
unsigned int ver_adjust : 1; // shift ver based on lsb of pred |
| 1765 |
unsigned int reserved : 30; |
| 1766 |
|
| 1767 |
} nvenc_vp8_me_stamp_s; // 12 bytes |
| 1768 |
|
| 1769 |
typedef struct |
| 1770 |
{ |
| 1771 |
unsigned int me_predictor_mode : 1; // 0=MDP, 1=const |
| 1772 |
unsigned int refinement_mode : 1; // 0=hpel, 1=qpel |
| 1773 |
unsigned int lambda_mode : 1; // 0=MDP, 1=const |
| 1774 |
unsigned int const_lambda : 11; // U8.3 format lambda, used if lambda_mode=1 |
| 1775 |
unsigned int refine_on_search_enable : 1; // 0=disable, 1=enable subpel refine for search candidates |
| 1776 |
unsigned int me_only_mode : 1; |
| 1777 |
unsigned int fps_mvcost : 1; |
| 1778 |
unsigned int sps_mvcost : 1; |
| 1779 |
unsigned int sps_cost_func : 1; |
| 1780 |
unsigned int me_icc_mode_mad : 1; |
| 1781 |
unsigned int sps_filter : 3; |
| 1782 |
unsigned int mc_filter : 3; |
| 1783 |
unsigned int mbc_ctrl_arbitor : 2; |
| 1784 |
unsigned int average_mvhint_enable : 1; |
| 1785 |
unsigned int rmvp_source_half_scaled : 1; // only support picture size half scale down in horz&vert |
| 1786 |
unsigned int reserved : 2; |
| 1787 |
|
| 1788 |
nvenc_vp8_me_limit_mv_s limit_mv; |
| 1789 |
nvenc_vp8_me_hint_cfg_s predsrc; // predictor sources |
| 1790 |
nvenc_vp8_me_const_vec_s l0_hint; // constant MV used as L0 hint |
| 1791 |
nvenc_vp8_me_const_vec_s l1_hint; // constant MV used as L1 hint |
| 1792 |
nvenc_vp8_me_const_vec_s l0_pred; // constant MV used as L0 predictor |
| 1793 |
// should this be made an array? |
| 1794 |
nvenc_vp8_me_stamp_s shape0; // priority 0 stamping pattern |
| 1795 |
nvenc_vp8_me_stamp_s shape1; // priority 1 stamping pattern |
| 1796 |
nvenc_vp8_me_stamp_s shape2; // priority 2 stamping pattern |
| 1797 |
nvenc_vp8_me_stamp_s shape3; // priority 3 stamping pattern |
| 1798 |
nvenc_vp8_me_stamp_s shape4; // priority 4 stamping pattern |
| 1799 |
nvenc_vp8_me_stamp_s shape5; // priority 5 stamping pattern |
| 1800 |
nvenc_vp8_me_stamp_s shape6; // priority 6 stamping pattern |
| 1801 |
nvenc_vp8_me_stamp_s shape7; // priority 7 stamping pattern |
| 1802 |
|
| 1803 |
nvenc_me_tebif_s teb_para; // teb parameters |
| 1804 |
unsigned char mbc_mb_size; // MBC size in MB |
| 1805 |
unsigned char tempmv_scale; // temporal mv scale |
| 1806 |
unsigned char reserved1[2]; |
| 1807 |
|
| 1808 |
} nvenc_vp8_me_control_s; // 136 bytes |
| 1809 |
|
| 1810 |
typedef struct |
| 1811 |
{ |
| 1812 |
unsigned int qpp_run_vector_4x4 : 16; // cost values for 4x4 transform (16 bit total vector) |
| 1813 |
unsigned int qpp_luma8x8_cost : 4; // luma 8x8 cost threshold, 0 = throw out all coefs, range 0-15 |
| 1814 |
unsigned int qpp_luma16x16_cost : 4; // luma 16x16 cost threshold, 0 = throw out all coefs, range 0-15 |
| 1815 |
unsigned int qpp_chroma_cost : 4; // chroma cost threshold, 0 = throw out all coefs, range 0-15 |
| 1816 |
unsigned int qpp_mode : 2; // 0 = OFF, 1 = 8x8, 2 = 16x16_8x8 |
| 1817 |
//unsigned int quant_intra_sat_flag : 1; // 0: no saturation is applied; 1: saturation is applied |
| 1818 |
//unsigned int quant_inter_sat_flag : 1; // 0: no saturation is applied; 1: saturation is applied |
| 1819 |
unsigned int reserved : 2; |
| 1820 |
|
| 1821 |
//unsigned int quant_intra_sat_limit : 16; // quantization saturation limit for intra MB |
| 1822 |
//unsigned int quant_inter_sat_limit : 16; // quantization saturation limit for inter MB |
| 1823 |
|
| 1824 |
unsigned short dz_4x4_YI[16]; // deadzone for 4x4 transform of Luma Intra |
| 1825 |
unsigned short dz_4x4_YP[16]; // deadzone for 4x4 transform of Luma Inter |
| 1826 |
unsigned short dz_4x4_CI; // deadzone for 4x4 transform of Chroma Intra |
| 1827 |
unsigned short dz_4x4_CP; // deadzone for 4x4 transform of Chroma Inter |
| 1828 |
|
| 1829 |
} nvenc_vp8_quant_control_s; // 72 bytes |
| 1830 |
|
| 1831 |
typedef struct |
| 1832 |
{ |
| 1833 |
//char temp_dist_l0[3]; // temporal distance of each ref pic (last, gold, altref) |
| 1834 |
//char reserved; |
| 1835 |
unsigned int hist_buf_size; // size in bytes of the buffer allocated for history |
| 1836 |
unsigned int bitstream_buf_size; // size in bytes of the buffer allocated for bitstream slice/mb data |
| 1837 |
unsigned int bitstream_start_pos; // start position in bitstream buffer where data should be written (byte offset) |
| 1838 |
unsigned int bitstream_residual_buf_size; // size in bytes of the buffer allocated for residual partition bitstream slice/mb data |
| 1839 |
unsigned int bitstream_residual_start_pos[8]; // start position in residual partition bitstream buffer where data for residual partitions should be written (byte offset) |
| 1840 |
|
| 1841 |
} nvenc_vp8_pic_control_s; // 48 bytes |
| 1842 |
|
| 1843 |
typedef struct |
| 1844 |
{ |
| 1845 |
//unsigned char hrd_type; // nal (=2) and vcl (=1) type |
| 1846 |
unsigned char QP[3]; // QP for 0:P picture, ,1:B picture 2; I picture |
| 1847 |
unsigned char minQP[3]; // min QP for 0:P picture, ,1:B picture 2; I picture |
| 1848 |
unsigned char maxQP[3]; // max QP for 0:P picture, ,1:B picture 2; I picture |
| 1849 |
//unsigned char maxQPD; // max QP delta between two consecutive QP updates |
| 1850 |
//unsigned char baseQPD; // initial QP delta between two consecutive QP updates |
| 1851 |
unsigned char reserved[3]; |
| 1852 |
int rhopbi[3]; // 23.8 signed fixed point quant ratios P/I, B/I, I/I(=1) |
| 1853 |
|
| 1854 |
int framerate; // fps |
| 1855 |
unsigned int buffersize; // total buffer size |
| 1856 |
|
| 1857 |
int nal_cpb_size; // size in bits |
| 1858 |
int nal_bitrate; // rate in bps |
| 1859 |
//int vcl_cpb_size; // size in bits |
| 1860 |
//int vcl_bitrate; // rate in bps |
| 1861 |
|
| 1862 |
int gop_length; // I period |
| 1863 |
//int Np; // 27.4 signed fixed (gop_length + num_b_frames -1)/(num_b_frames + 1) |
| 1864 |
int Bmin; // 23.8 signed fixed point min buffer level, updated by driver at sequence level only. |
| 1865 |
int Ravg; // 23.8 signed fixed point average rate, updated by driver at sequence level only. |
| 1866 |
int R; // 23.8 signed fixed point rate |
| 1867 |
|
| 1868 |
//unsigned char ab_alpha; // weight of prev frame activity compared to part of the current picture |
| 1869 |
//unsigned char ab_beta; // weight of current activity compared to neighbors |
| 1870 |
//unsigned short prev_act; // average activity level of the first reference (L0) |
| 1871 |
} nvenc_vp8_rc_s; // 56 bytes |
| 1872 |
|
| 1873 |
typedef struct |
| 1874 |
{ |
| 1875 |
// 256 aligned |
| 1876 |
unsigned int magic; // magic number |
| 1877 |
nvenc_vp8_surface_cfg_s refpic_cfg; // 28 bytes |
| 1878 |
nvenc_vp8_surface_cfg_s input_cfg; // 28 bytes |
| 1879 |
nvenc_vp8_surface_cfg_s outputpic_cfg; // 28 bytes |
| 1880 |
nvenc_vp8_sequence_data_s sequence_data; // 4 bytes |
| 1881 |
nvenc_vp8_picture_data_s picture_data; // 32 bytes |
| 1882 |
nvenc_vp8_pic_control_s pic_control; // 48 bytes |
| 1883 |
nvenc_vp8_me_control_s me_control; // 136 bytes |
| 1884 |
nvenc_vp8_md_control_s md_control; // 64 bytes |
| 1885 |
nvenc_vp8_quant_control_s quant_control; // 72 bytes |
| 1886 |
nvenc_vp8_rc_s rate_control; // 56 bytes |
| 1887 |
unsigned char reserved[16]; // pad to 512 bytes |
| 1888 |
|
| 1889 |
} nvenc_vp8_drv_pic_setup_s; // 512 bytes |
| 1890 |
|
| 1891 |
|
| 1892 |
// 2.2. Colocated Motion Data Buffer |
| 1893 |
// --------------------------------- |
| 1894 |
// |
| 1895 |
// This buffer will be used to store co-located motion data. This data is produced for |
| 1896 |
// pictures that are marked as reference pictures. The buffer is not needed for non- |
| 1897 |
// reference pictures and in such case the corresponding method may be skipped. The |
| 1898 |
// data in this buffer will be needed for future pictures encoded by both CEA and NVENC. |
| 1899 |
// There are two of these buffers: one output, written only if the current picture is a |
| 1900 |
// reference picture, and one input, read if the current picture is being encoded |
| 1901 |
// as a B picture, and possible also during P picture to serve as hints for motion search. |
| 1902 |
// Each buffer holds coloc data for all macroblocks in the picture. This |
| 1903 |
// data is packed into an array of 16 entries of 4 bytes. Each entry holds one motion |
| 1904 |
// vector and 6 bits of additional data, which is formatted as follows: |
| 1905 |
// |
| 1906 |
// coldata array entry N: contents of 'e': |
| 1907 |
// 0, 4, 8, 12 0 = bottom field flag for ref picture |
| 1908 |
// 4..1 = reference picture number (index in DPB) |
| 1909 |
// 5 = zero |
| 1910 |
// 1, 5, 9, 13 3..0 = ColZero flag for blocks N+2 .. N-1 |
| 1911 |
// 5..4 = subpartition size for 8x8 block (N-1)/4 |
| 1912 |
// (0=8x8, 1=8x4, 2=4x8, 3=4x4) |
| 1913 |
// 15 0 = field mode flag |
| 1914 |
// 1 = intra coded |
| 1915 |
// 3..2 = partition size for macroblock |
| 1916 |
// (0=16x16, 1=16x8, 2=8x16, 3=8x8) |
| 1917 |
// 5..4 = zero |
| 1918 |
// all others 5..0 = zero |
| 1919 |
// |
| 1920 |
// Note that motion vectors are duplicated for each 4x4 index from the top-left 4x4 block |
| 1921 |
// of each (sub)partition. Note also that since NVENC does not currently support sub-8x8 |
| 1922 |
// inter partitioning, the 4 vectors within an 8x8 are replicated values. |
| 1923 |
|
| 1924 |
typedef struct |
| 1925 |
{ |
| 1926 |
unsigned int mvx : 14; // bit 0..13: x component of MV |
| 1927 |
unsigned int mvy : 12; // bit 14..25: y component of MV |
| 1928 |
unsigned int e : 6; // bit 26..31: extra data, different per entry |
| 1929 |
} nvenc_h264_coloc_entry_s; |
| 1930 |
|
| 1931 |
typedef struct // 64 byte aligned block |
| 1932 |
{ |
| 1933 |
nvenc_h264_coloc_entry_s coldata[16]; // col data for 1 macroblock = array of 16 entries |
| 1934 |
} nvenc_h264_coloc_mb_s; |
| 1935 |
|
| 1936 |
|
| 1937 |
// 2.3. History Buffer |
| 1938 |
// ------------------- |
| 1939 |
// |
| 1940 |
// This buffer will be used to keep history data (data produced for one macroblock that |
| 1941 |
// must be kept until the macroblock in the row directly below is processed). This buffer |
| 1942 |
// should be at least as large as (((frame_width_minus1+16) / 16) + 16)*128 bytes. |
| 1943 |
// |
| 1944 |
// After the execute completes, the data in this buffer is no longer needed and the |
| 1945 |
// buffer can be discarded/deallocated. |
| 1946 |
// |
| 1947 |
// The internal structure of the history record is implementation specific and therefore |
| 1948 |
// subject to change. The data is produced and consumed by NVENC only and should never be |
| 1949 |
// interpreted by any other engine or by software. The current structure is defined in |
| 1950 |
// the following visio document in the HW development tree: |
| 1951 |
// |
| 1952 |
// doc/nvenc/arch/hist_layout.vsd |
| 1953 |
// |
| 1954 |
// 2.4. Reference and Input Picture Buffer |
| 1955 |
// --------------------------------------- |
| 1956 |
// |
| 1957 |
// Reference Picture Buffer contains reference pictures produced by NVENC during |
| 1958 |
// encode opeations of previous pictures. The Input Picture Buffer contains the |
| 1959 |
// current picture that needs to be encoded into a bitstream. |
| 1960 |
// |
| 1961 |
// Each picture is actually a separate buffer with its own surface offset method. |
| 1962 |
// Relative position of luma, chroma fields are defined in the picture setup buffer |
| 1963 |
// and shown in the figure below. |
| 1964 |
// |
| 1965 |
// Driver needs to send SetRefPicOffset#() method for only the used reference pictures |
| 1966 |
// for the current execute. |
| 1967 |
// |
| 1968 |
// Pictures are by default stored in NV24 field or NV12 frame format. NVENC will |
| 1969 |
// always output in the same format as the current picture that is being encoded |
| 1970 |
// (frame or field). True picaff (using frame refs to encode a field or using field |
| 1971 |
// refs to encode a frame) is NOT supported. |
| 1972 |
// |
| 1973 |
// The input picture must contain fields if the picture is to be encoded as fields, |
| 1974 |
// and must contain frames if the picture is to be encoded as a frame. |
| 1975 |
// |
| 1976 |
// |
| 1977 |
// +---------------+ <-- RefPicOffset (start of buffer) |
| 1978 |
// |XXXXXXXXXXXXXXX| |
| 1979 |
// +---------------+ <-- luma_top_frm_offset (in picture setup buffer, in units of 256 bytes) |
| 1980 |
// |TOP/FRM luma | |
| 1981 |
// | | |
| 1982 |
// +XXXXXXXXXXXXXXX+ |
| 1983 |
// +---------------+ <-- luma_bot_offset |
| 1984 |
// |BOTTOM luma | |
| 1985 |
// | | |
| 1986 |
// +XXXXXXXXXXXXXXX+ |
| 1987 |
// +---------------+ <-- chroma_top_frm_offset |
| 1988 |
// |TOP/FRM chroma | |
| 1989 |
// +XXXXXXXXXXXXXXX+ |
| 1990 |
// +---------------+ <-- chroma_bot_offset |
| 1991 |
// |BOTTOM chroma | |
| 1992 |
// +XXXXXXXXXXXXXXX+ |
| 1993 |
// +---------------+ <-- end of buffer |
| 1994 |
// |
| 1995 |
// |
| 1996 |
// 2.5. CEA Hints Data Buffer |
| 1997 |
// --------------------------- |
| 1998 |
// |
| 1999 |
// CEA (a CUDA based integer motion search kernel) can provide hints to NVENC. |
| 2000 |
// These hints can either be taken as explicit vectors to evaluate or as guides |
| 2001 |
// for NVENC's built-in integer search. |
| 2002 |
// In this mode of operation, CEA and NVENC typically work on the same frame, |
| 2003 |
// with CEA slightly leading (by a few macroblock rows). NVENC would then read |
| 2004 |
// in the data already produced by CEA for the current picture and use it. CEA |
| 2005 |
// has to provide information on how many macroblocks worth of data it has |
| 2006 |
// already completed and written out into this buffer, so that NVENC can read |
| 2007 |
// up to that point. The buffer must be big enough to contain all data for the |
| 2008 |
// entire frame. |
| 2009 |
// |
| 2010 |
// This buffer consists of 2 sections: a header and a data section. One such |
| 2011 |
// buffer must be provided/allocated by driver for each picture that needs to |
| 2012 |
// be encoded. The header section must be prefilled with zeroes by driver. The |
| 2013 |
// data section can remain uninitialized. |
| 2014 |
// |
| 2015 |
// The data section contains the motion vector candidates for each macroblock. |
| 2016 |
// It is written by CEA and read by NVENC. |
| 2017 |
// |
| 2018 |
// The header section contains flow control and status information produced by |
| 2019 |
// CEA for this picture, and contains an offset to the start of the data |
| 2020 |
// section. The header section begins at offset 0 in this buffer. |
| 2021 |
|
| 2022 |
|
| 2023 |
typedef struct |
| 2024 |
{ |
| 2025 |
unsigned int cand_cnt_blk_16x16 : 4; // 3:0 Number of candidates per block types |
| 2026 |
unsigned int cand_cnt_blk_16x8 : 4; // 7:4 |
| 2027 |
unsigned int cand_cnt_blk_8x16 : 4; // 11:8 |
| 2028 |
unsigned int cand_cnt_blk_8x8 : 4; // 15:12 |
| 2029 |
unsigned int cand_cnt_blk_8x4 : 4; // 19:16 |
| 2030 |
unsigned int cand_cnt_blk_4x8 : 4; // 23:20 |
| 2031 |
unsigned int cand_cnt_blk_4x4 : 4; // 27:24 |
| 2032 |
unsigned int cand_bdirect_spatial_zmv : 1; // 28 enable Spatial direct enable ZMV 4x4 scores |
| 2033 |
unsigned int cand_bdirect_spatial_nzmv : 1; // 29 enable Spatial direct enable non-zero MV 4x4 scores |
| 2034 |
unsigned int cand_bdirect_temporal : 1; // 30 enable Temporal direct mode. Only valid for Bi-type. |
| 2035 |
unsigned int cand_pskip : 1; // 31 enable p-skip mode. Only valid for L0 controls. |
| 2036 |
} cea_cand_count_blktype_s; |
| 2037 |
|
| 2038 |
typedef struct |
| 2039 |
{ |
| 2040 |
// 64 aligned |
| 2041 |
unsigned int picture_index; // picture index given by driver (via method) |
| 2042 |
unsigned int data_start; // start offset in buffer of data section which must be 256B aligned |
| 2043 |
// the following fields are only there right now to make it easier to implement the cmodel plugin mode |
| 2044 |
// they might be removed at a later stage |
| 2045 |
unsigned int cea_lambda; |
| 2046 |
cea_cand_count_blktype_s blk_counts[3]; |
| 2047 |
int reserved[10]; //padded by 8 bytes to align the end of the structure to 32 bytes for dmread operations |
| 2048 |
} cea_hints_header_s; |
| 2049 |
|
| 2050 |
|
| 2051 |
// Please refer to the document named "nvenc_cea_mvinfo_format.doc" for more information. |
| 2052 |
// This document can be found in the nvenc HW development tree under doc/nvenc/arch |
| 2053 |
|
| 2054 |
typedef struct |
| 2055 |
{ |
| 2056 |
int mvx : 12; // 11:00 x component of integer pixel MV (relative to current MB) S12.0 |
| 2057 |
int mvy : 10; // 21:12 y component of integer pixel MV (relative to current MB) S10.0 |
| 2058 |
int refidx : 5; // 26:22 reference index (31=invalid) |
| 2059 |
int dir : 1; // 27:27 0=L0 1=L1 |
| 2060 |
int part_type : 2; // 29:28 0=16x16 1=16x8 2=8x16 3=8x8 (blocks in partition must be consecutive) |
| 2061 |
int last_of_part: 1; // 30:30 last MV of (sub) partition |
| 2062 |
int last_of_mb : 1; // 31:31 last MV of macroblock |
| 2063 |
} cea_hint_s; |
| 2064 |
|
| 2065 |
// External hint data structure for Orin AV1 |
| 2066 |
typedef struct |
| 2067 |
{ |
| 2068 |
short refidx : 5; // 4:0 reference index (31=invalid) |
| 2069 |
short direction : 1; // 5:5 0=L0 1=L1 |
| 2070 |
short bi : 1; // 6:6 0=single mv, 1=compound mv |
| 2071 |
short partition_type : 3; // 9:7 parition type: 0: 2NX2N, 1:2NxN, 2:Nx2N. reserved 3bits for modes in later projects |
| 2072 |
short x8 : 3; // 12:10 x8: current parition's top left x position in 8 pixel unit |
| 2073 |
short last_of_cu : 1; // 13:13 last mv of current CU |
| 2074 |
short last_of_sb : 1; // 14:14 last mv of current SB |
| 2075 |
short reserved0 : 1; |
| 2076 |
|
| 2077 |
short mvx : 14; // 13:0 mvx, in S12.2 format |
| 2078 |
short cu_size : 2; // 15:14 CU size: 0: 8x8, 1: 16x16, 2:32x32, 3:64x64 |
| 2079 |
|
| 2080 |
short mvy : 12; // 11:0 mvy, in S10.2 format |
| 2081 |
short y8 : 3; // 14:12 y8: current parition's top left y position in 8 pixel unit |
| 2082 |
short reserved1 : 1; |
| 2083 |
} av1_hint_data; //6 bytes |
| 2084 |
|
| 2085 |
|
| 2086 |
typedef struct |
| 2087 |
{ |
| 2088 |
unsigned int segID[4]; //segment ID for one MB. each 4x4 block has a segID which is 8 bits, total is 8bitx16 = 128 bits. maximun segments supported is 32 |
| 2089 |
} cea_segment_s; |
| 2090 |
|
| 2091 |
|
| 2092 |
|
| 2093 |
// 2.6. Encoder Status Buffer |
| 2094 |
// --------------------------- |
| 2095 |
// |
| 2096 |
// This buffer provides some additional data about the encoded bit stream that can |
| 2097 |
// be used for rate control and multi-pass encoding as well as status information |
| 2098 |
// that can be used for debug. |
| 2099 |
|
| 2100 |
typedef struct |
| 2101 |
{ |
| 2102 |
unsigned int picture_index; // value received from SetPictureIndex method |
| 2103 |
unsigned int error_status : 2; // report error if any |
| 2104 |
unsigned int ucode_error_status : 30; // report error status from ucode to driver |
| 2105 |
unsigned int total_bit_count; // picture size in bits |
| 2106 |
unsigned int type1_bit_count; // type1 bit count for the entire picture |
| 2107 |
unsigned short pic_type; // copied from pic_type in nvenc_h264_pic_control_s |
| 2108 |
unsigned short num_slices; // number of slices produced |
| 2109 |
unsigned short ave_activity; // report average activity if activity based RC enabled |
| 2110 |
unsigned short avgQP; // average QP |
| 2111 |
unsigned int cycle_count; // total cycles taked for execute. Written out only when DumpCycleCount in SetControlParams is set |
| 2112 |
int hrdFullness; |
| 2113 |
unsigned int bitstream_start_pos; // byte_offset where mpec will start writing slice header |
| 2114 |
unsigned int last_valid_byte_offset; // greatest offset in output buffer which is assured to have valid compressed data |
| 2115 |
unsigned short intra_mb_count; // number of intra MBs. |
| 2116 |
unsigned short inter_mb_count; // number of inter MBs. |
| 2117 |
unsigned int cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 2118 |
unsigned int cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 2119 |
unsigned int total_intra_cost; // sum of all intra best mode's SATDs, each mb has one intra SATD. |
| 2120 |
unsigned int total_inter_cost; // sum of all inter best mode's SATDs, each mb has one inter SATD. |
| 2121 |
unsigned int complexity; // sum of mbRowBitcount * current QS |
| 2122 |
short average_mvx; // average mvx info, can be used for ME hint |
| 2123 |
short average_mvy; // average mvy info, can be used for ME hint |
| 2124 |
unsigned short actual_min_qp_used; // actual minimum qp used in the frame |
| 2125 |
unsigned short actual_max_qp_used; // actual maximum qp used in the frame |
| 2126 |
unsigned int hw_perf_me_sps; // me slice per second |
| 2127 |
unsigned int hw_perf_me_fps; // me frame per second |
| 2128 |
unsigned int hw_perf_mpec_active_cnt; // mpec active count |
| 2129 |
unsigned int hw_perf_mpeb_active_cnt; // mpeb active count |
| 2130 |
unsigned long long int total_dist_ssd; // sum of all ssd distortion of the MB best mode |
| 2131 |
unsigned int total_dist_ssim; // sum of all ssim distortion of the MB best mode |
| 2132 |
unsigned int best_satd_l0; // sum of all best satd cost of L0 |
| 2133 |
unsigned int best_satd_l1; // sum of all best satd cost of L1 |
| 2134 |
unsigned int reserved[5]; // align to 128 bytes |
| 2135 |
|
| 2136 |
} nvenc_pic_stat_s; // 128 bytes |
| 2137 |
|
| 2138 |
|
| 2139 |
typedef struct |
| 2140 |
{ |
| 2141 |
unsigned int slice_offset; // slice offset in bytes |
| 2142 |
unsigned int slice_size; // slice size in bytes |
| 2143 |
unsigned int slh_bit_count; // slice header size in bits |
| 2144 |
unsigned char reserved[4]; |
| 2145 |
} nvenc_slice_stat_s; // 16 bytes per status |
| 2146 |
|
| 2147 |
|
| 2148 |
typedef struct |
| 2149 |
{ |
| 2150 |
unsigned int intra_res_bit_count; // #bits spent on intra residuals |
| 2151 |
unsigned int inter_res_bit_count; // #bits spent on inter residuals |
| 2152 |
unsigned int intra_mode_bit_count; // #bits spent on intra modes (mbtype, pred mode) |
| 2153 |
unsigned int inter_mode_bit_count; // #bits spent on inter modes (mbtype, submbtype, refidx, mv) |
| 2154 |
unsigned int ipcm_bit_count; // # bits spent on MB mode that caused IPCM rewind. |
| 2155 |
unsigned int type1_bit_count; // count of bits in type I bitstream (excludes leading_zero_8bits, |
| 2156 |
// start_codeprefix_one_3bytes and trailing_zero_8bits) |
| 2157 |
unsigned int total_bit_count; // total number of bits produced |
| 2158 |
|
| 2159 |
unsigned int total_bin_count; // total number of bins produced by cabac |
| 2160 |
unsigned int cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 2161 |
unsigned int cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 2162 |
unsigned int cumulative_ipcm_cost; // sum of all inter SA(T)Ds for the MBs that caused IPCM rewind |
| 2163 |
unsigned int cumulative_skip_cost; // sum of all P and B skip MB SA(T)Ds |
| 2164 |
|
| 2165 |
unsigned short intra_mb_count; // number of intra MBs. |
| 2166 |
unsigned short inter_mb_count; // number of inter MBs. |
| 2167 |
unsigned short skip_mb_count; // number of skip MBs. |
| 2168 |
unsigned short ipcm_mb_count; // number of ipcm MBs. |
| 2169 |
|
| 2170 |
unsigned int total_intra_cost; // sum of all intra best mode's SATDs, each ctb has one intra SATD. |
| 2171 |
unsigned int total_inter_cost; // sum of all inter best mode's SATDs, each ctb has one inter SATD. |
| 2172 |
} nvenc_mpec_stat_s; // 64 bytes per status |
| 2173 |
|
| 2174 |
typedef struct |
| 2175 |
{ |
| 2176 |
unsigned int qp_val : 8 ; // qp used for encoding this block |
| 2177 |
unsigned int total_bit_count : 24 ; // total number of bits produced |
| 2178 |
unsigned int res_cost : 24 ; // satd cost of residual error |
| 2179 |
unsigned int flags : 8 ; // flags |
| 2180 |
} nvenc_stats_fifo_s; // 8 bytes per stat |
| 2181 |
|
| 2182 |
typedef struct |
| 2183 |
{ |
| 2184 |
nvenc_pic_stat_s pic_stat; // 64Bytes |
| 2185 |
nvenc_slice_stat_s slice_stat[((4096/16)*(4096/16)) + 3]; // To account for 1mb/slice and 4kx4k, plus 3 for alignment |
| 2186 |
nvenc_mpec_stat_s mpec_stat[((4096/16)*(4096/16)) + 3]; // To account for stat/1mb and 4kx4k, plus 3 for alignment |
| 2187 |
nvenc_aq_stat_s aq_stats[((4096/16)*(4096/16)) + 3]; // AQ stats from PDMA |
| 2188 |
nvenc_stats_fifo_s stats_fifo[((4096/16)*(4096/16)) + 3]; |
| 2189 |
} nvenc_stat_data_s; |
| 2190 |
|
| 2191 |
typedef struct |
| 2192 |
{ |
| 2193 |
unsigned int picture_index; // value received from SetPictureIndex method |
| 2194 |
unsigned int error_status : 2; // report error if any |
| 2195 |
unsigned int ucode_error_status : 30; // report error status from ucode to driver |
| 2196 |
unsigned int total_bit_count; // picture size in bits |
| 2197 |
unsigned int type1_bit_count; // type1 bit count for the entire picture |
| 2198 |
unsigned short pic_type; // copied from pic_type in nvenc_h264_pic_control_s |
| 2199 |
unsigned short num_slices; // number of slices produced, maxinum slice number is 65535 |
| 2200 |
unsigned short ave_activity; // report average activity if activity based RC enabled |
| 2201 |
unsigned short avgQP; // average QP |
| 2202 |
unsigned int cycle_count; // total cycles taked for execute. Written out only when DumpCycleCount in SetControlParams is set |
| 2203 |
int hrdFullness; |
| 2204 |
unsigned int bitstream_start_pos; // byte_offset where mpec will start writing slice header |
| 2205 |
unsigned int last_valid_byte_offset; // greatest offset in output buffer which is assured to have valid compressed data |
| 2206 |
unsigned int intra_cu32x32_count; // number of intra 32x32 CUs. |
| 2207 |
unsigned int inter_cu32x32_count; // number of inter 32x32 CUs. |
| 2208 |
unsigned int intra_cu16x16_count; // number of intra 16x16 CUs. |
| 2209 |
unsigned int inter_cu16x16_count; // number of inter 16x16 CUs. |
| 2210 |
unsigned int intra_cu8x8_count; // number of intra 8x8 CUs. |
| 2211 |
unsigned int inter_cu8x8_count; // number of inter 8x8 CUs. |
| 2212 |
unsigned int cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 2213 |
unsigned int cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 2214 |
unsigned int total_intra_cost; // sum of all intra best mode's SATDs, each mb has one intra SATD. |
| 2215 |
unsigned int total_inter_cost; // sum of all inter best mode's SATDs, each mb has one inter SATD. |
| 2216 |
short average_mvx; // average mvx info, can be used for ME hint |
| 2217 |
short average_mvy; // average mvy info, can be used for ME hint |
| 2218 |
unsigned short actual_min_qp_used; // actual minimum qp used in the frame |
| 2219 |
unsigned short actual_max_qp_used; // actual maximum qp used in the frame |
| 2220 |
unsigned int hw_perf_me_sps; // me slice per second |
| 2221 |
unsigned int hw_perf_me_fps; // me frame per second |
| 2222 |
unsigned int hw_perf_mpec_active_cnt; // mpec active count |
| 2223 |
unsigned int hw_perf_mpeb_active_cnt; // mpeb active count |
| 2224 |
unsigned char reserved[24]; |
| 2225 |
} nvenc_h265_pic_stat_s; // 128 bytes |
| 2226 |
|
| 2227 |
typedef struct |
| 2228 |
{ |
| 2229 |
unsigned int slice_offset; // slice offset in bytes |
| 2230 |
unsigned int slice_size; // slice size in bytes |
| 2231 |
unsigned int slh_bit_count; // slice header size in bits |
| 2232 |
unsigned int slice_size_ctbs; // slice size in ctbs |
| 2233 |
}nvenc_h265_slice_stat_s; // 16 bytes per status |
| 2234 |
|
| 2235 |
typedef struct |
| 2236 |
{ |
| 2237 |
unsigned int intra_res_bit_count; // #bits spent on intra residuals |
| 2238 |
unsigned int inter_res_bit_count; // #bits spent on inter residuals |
| 2239 |
unsigned int intra_mode_bit_count; // #bits spent on intra modes (mbtype, pred mode) |
| 2240 |
unsigned int inter_mode_bit_count; // #bits spent on inter modes (mbtype, submbtype, refidx, mv) |
| 2241 |
unsigned int ipcm_bit_count; // # bits spent on MB mode that caused IPCM rewind. |
| 2242 |
unsigned int type1_bit_count; // count of bits in type I bitstream (excludes leading_zero_8bits, |
| 2243 |
// start_codeprefix_one_3bytes and trailing_zero_8bits) |
| 2244 |
unsigned int total_bit_count; // total number of bits produced |
| 2245 |
|
| 2246 |
unsigned int total_bin_count; // total number of bins produced by cabac |
| 2247 |
unsigned int cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 2248 |
unsigned int cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 2249 |
unsigned int cumulative_ipcm_cost; // sum of all inter SA(T)Ds for the MBs that caused IPCM rewind |
| 2250 |
unsigned int cumulative_skip_cost; // sum of all P and B skip MB SA(T)Ds |
| 2251 |
unsigned int total_intra_cost; // sum of all intra best mode's SATDs, each ctb has one intra SATD. |
| 2252 |
unsigned int total_inter_cost; // sum of all inter best mode's SATDs, each ctb has one inter SATD. |
| 2253 |
|
| 2254 |
unsigned int intra_cu32x32_count; // number of intra 32x32 CUs. |
| 2255 |
unsigned int inter_cu32x32_count; // number of inter 32x32 CUs. |
| 2256 |
unsigned int skip_cu32x32_count; // number of skip 32x32 CUs. |
| 2257 |
unsigned int ipcm_cu32x32_count; // number of ipcm 32x32 CUs. |
| 2258 |
unsigned int intra_cu16x16_count; // number of intra 16x16 CUs. |
| 2259 |
unsigned int inter_cu16x16_count; // number of inter 16x16 CUs. |
| 2260 |
unsigned int skip_cu16x16_count; // number of skip 16x16 CUs. |
| 2261 |
unsigned int ipcm_cu16x16_count; // number of ipcm 16x16 CUs. |
| 2262 |
unsigned int intra_cu8x8_count; // number of intra 8x8 CUs. |
| 2263 |
unsigned int inter_cu8x8_count; // number of inter 8x8 CUs. |
| 2264 |
unsigned int skip_cu8x8_count; // number of skip 8x8 CUs. |
| 2265 |
unsigned int ipcm_cu8x8_count; // number of ipcm 8x8 CUs. |
| 2266 |
|
| 2267 |
unsigned int total_emulated_bit_count; // total number of emulated bits in total_bit_count |
| 2268 |
unsigned char reserved[20]; |
| 2269 |
} nvenc_h265_mpec_stat_s; // 128 bytes per status |
| 2270 |
|
| 2271 |
typedef struct |
| 2272 |
{ |
| 2273 |
unsigned int var_min; |
| 2274 |
unsigned int var_max; |
| 2275 |
unsigned int var_avg; |
| 2276 |
unsigned char reserved[4]; |
| 2277 |
} nvenc_h265_aq_stat_s; // 16 bytes |
| 2278 |
|
| 2279 |
typedef struct |
| 2280 |
{ |
| 2281 |
unsigned int qp_val : 8 ; // qp used for encoding this block |
| 2282 |
unsigned int total_bit_count : 24 ; // total number of bits produced |
| 2283 |
unsigned int res_cost : 24 ; // satd cost of residual error |
| 2284 |
unsigned int flags : 8 ; // flags |
| 2285 |
} nvenc_h265_stats_fifo_s; // 8 bytes per stat |
| 2286 |
|
| 2287 |
typedef struct |
| 2288 |
{ |
| 2289 |
nvenc_h265_pic_stat_s pic_stat; // 64Bytes |
| 2290 |
nvenc_h265_slice_stat_s slice_stat[((4096/16)*(4096/16)) + 3]; // To account for 1mb/slice and 4kx4k, plus 3 for alignment. support upto 65535 slices |
| 2291 |
nvenc_h265_mpec_stat_s mpec_stat[((4096/16)*(4096/16)) + 3]; // To account for stat/1mb and 4kx4k, plus 3 for alignment. support upto 65535 slices |
| 2292 |
nvenc_h265_aq_stat_s aq_stats [((4096/16)*(4096/16)) + 3]; // AQ stats from pdma. support upto 65535 slices |
| 2293 |
nvenc_h265_stats_fifo_s stats_fifo[((4096/16)*(4096/16)) + 3]; |
| 2294 |
} nvenc_h265_stat_data_s; |
| 2295 |
|
| 2296 |
typedef struct |
| 2297 |
{ |
| 2298 |
unsigned int picture_index; // value received from SetPictureIndex method |
| 2299 |
unsigned int error_status : 2; // report error if any |
| 2300 |
unsigned int ucode_error_status : 30; // report error status from ucode to driver |
| 2301 |
unsigned int total_bit_count; // picture size in bits |
| 2302 |
unsigned int type1_bit_count; // type1 bit count for the entire picture |
| 2303 |
unsigned short num_tiles; // number of tiles produced |
| 2304 |
unsigned char av1_start_QP; // AV1 QP value at frame encoding start (to be inserted into frame header by driver) |
| 2305 |
unsigned char loop_filter_level; // loop filer value computed by ucode (to be inserted into frame header by driver) |
| 2306 |
unsigned int pic_type : 14; // copied from pic_type in nvenc_h264_pic_control_s |
| 2307 |
unsigned int cdef_damping_minus_3 : 2; // CDEF damping value computed by ucode (to be inserted into frame header by driver) |
| 2308 |
int avgQP : 16; // average QP |
| 2309 |
unsigned int cycle_count; // total cycles taked for execute. Written out only when DumpCycleCount in SetControlParams is set |
| 2310 |
unsigned int hrdFullness; |
| 2311 |
unsigned int bitstream_start_pos; // byte_offset where mpec will start writing slice header |
| 2312 |
unsigned int intra_cu64x64_count; // number of intra 64x64 CUs. |
| 2313 |
unsigned int inter_cu64x64_count; // number of inter 64x64 CUs. |
| 2314 |
unsigned int intra_cu32x32_count; // number of intra 32x32 CUs. |
| 2315 |
unsigned int inter_cu32x32_count; // number of inter 32x32 CUs. |
| 2316 |
unsigned int intra_cu16x16_count; // number of intra 16x16 CUs. |
| 2317 |
unsigned int inter_cu16x16_count; // number of inter 16x16 CUs. |
| 2318 |
unsigned int intra_cu8x8_count; // number of intra 8x8 CUs. |
| 2319 |
unsigned int inter_cu8x8_count; // number of inter 8x8 CUs. |
| 2320 |
unsigned int cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 2321 |
unsigned int cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 2322 |
unsigned int total_intra_cost; // sum of all intra best mode's SATDs, each mb has one intra SATD. |
| 2323 |
unsigned int total_inter_cost; // sum of all inter best mode's SATDs, each mb has one inter SATD. |
| 2324 |
short average_mvx; // average mvx info, can be used for ME hint |
| 2325 |
short average_mvy; // average mvy info, can be used for ME hint |
| 2326 |
short actual_min_qp_used; // actual minimum qp used in the frame |
| 2327 |
short actual_max_qp_used; // actual maximum qp used in the frame |
| 2328 |
unsigned int hw_perf_me_sps; // me slice per second |
| 2329 |
unsigned int hw_perf_me_fps; // me frame per second |
| 2330 |
unsigned int hw_perf_entropy_active_cnt; // Entropy active count |
| 2331 |
unsigned int hw_perf_rdo_active_cnt; // RDO active count |
| 2332 |
unsigned short ave_activity; // report average activity if activity based RC enabled |
| 2333 |
unsigned char reserved[18]; |
| 2334 |
} nvenc_av1_pic_stat_s; // 128 bytes |
| 2335 |
|
| 2336 |
typedef struct |
| 2337 |
{ |
| 2338 |
unsigned int tile_offset; // tile offset in bytes |
| 2339 |
unsigned int tile_size; // tile size in bytes |
| 2340 |
unsigned short tile_bitstream_idx; // Index of bistream base address the tile offset refers to (for multi entropy cores use - set to 0 otherwise) |
| 2341 |
unsigned short tile_idx; // Tile index at strip level |
| 2342 |
unsigned int tile_size_sbs; // tile size in sbs |
| 2343 |
}nvenc_av1_tile_stat_s; // 16 bytes per status |
| 2344 |
|
| 2345 |
typedef struct |
| 2346 |
{ |
| 2347 |
unsigned int res_bit_count; // #bits spent on residuals |
| 2348 |
unsigned int mode_bit_count; // #bits spent on modes (mbtype, pred mode) |
| 2349 |
unsigned int type1_bit_count; // count of bits in type I bitstream (excludes leading_zero_8bits, |
| 2350 |
// start_codeprefix_one_3bytes and trailing_zero_8bits) |
| 2351 |
unsigned int total_bit_count; // total number of bits produced |
| 2352 |
|
| 2353 |
unsigned int total_bin_count; // total number of bins produced by arithmetic coder |
| 2354 |
unsigned int cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 2355 |
unsigned int cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 2356 |
unsigned int total_intra_cost; // sum of all intra best mode's SATDs, each ctb has one intra SATD. |
| 2357 |
unsigned int total_inter_cost; // sum of all inter best mode's SATDs, each ctb has one inter SATD. |
| 2358 |
|
| 2359 |
unsigned int intra_cu64x64_count; // number of intra 64x64 CUs. |
| 2360 |
unsigned int inter_cu64x64_count; // number of inter 64x64 CUs. |
| 2361 |
unsigned int intra_cu32x32_count; // number of intra 32x32 CUs. |
| 2362 |
unsigned int inter_cu32x32_count; // number of inter 32x32 CUs. |
| 2363 |
unsigned int intra_cu16x16_count; // number of intra 16x16 CUs. |
| 2364 |
unsigned int inter_cu16x16_count; // number of inter 16x16 CUs. |
| 2365 |
unsigned int intra_cu8x8_count; // number of intra 8x8 CUs. |
| 2366 |
unsigned int inter_cu8x8_count; // number of inter 8x8 CUs. |
| 2367 |
|
| 2368 |
|
| 2369 |
unsigned int total_emulated_bit_count; // total number of emulated bits in total_bit_count |
| 2370 |
unsigned char reserved[56]; |
| 2371 |
} nvenc_av1_mpec_stat_s; // 128 bytes per status |
| 2372 |
|
| 2373 |
typedef struct |
| 2374 |
{ |
| 2375 |
unsigned int var_min; |
| 2376 |
unsigned int var_max; |
| 2377 |
unsigned int var_avg; |
| 2378 |
unsigned char reserved[4]; |
| 2379 |
} nvenc_av1_aq_stat_s; // 16 bytes |
| 2380 |
|
| 2381 |
typedef struct |
| 2382 |
{ |
| 2383 |
unsigned int qp_val : 8 ; // qp used for encoding this block |
| 2384 |
unsigned int total_bit_count : 24 ; // total number of bits produced |
| 2385 |
unsigned int res_cost : 24 ; // satd cost of residual error |
| 2386 |
unsigned int flags : 8 ; // flags |
| 2387 |
} nvenc_av1_stats_fifo_s; // 8 bytes per stat |
| 2388 |
|
| 2389 |
typedef struct |
| 2390 |
{ |
| 2391 |
nvenc_av1_pic_stat_s pic_stat; // 64Bytes |
| 2392 |
nvenc_av1_tile_stat_s tile_stat[128]; // maximum tile number is 128 per AV1 spec |
| 2393 |
nvenc_av1_mpec_stat_s mpec_stat[((8192 / 64)*(8192 / 64)) ]; // To account for stat/1SB and 8kx8k, plus 3 for alignment? |
| 2394 |
nvenc_av1_aq_stat_s aq_stats[((8192 / 64)*(8192 / 64)) ]; // AQ stats from pdma |
| 2395 |
nvenc_av1_stats_fifo_s stats_fifo[((8192 / 64)*(8192 / 64)) ]; |
| 2396 |
} nvenc_av1_stat_data_s; |
| 2397 |
|
| 2398 |
|
| 2399 |
typedef struct |
| 2400 |
{ |
| 2401 |
unsigned int intra_res_bit_count; // #bits spent on intra residuals |
| 2402 |
unsigned int inter_res_bit_count; // #bits spent on inter residuals |
| 2403 |
unsigned int intra_mode_bit_count; // #bits spent on intra modes |
| 2404 |
unsigned int inter_mode_bit_count; // #bits spent on inter modes |
| 2405 |
unsigned int total_bit_count; // total number of bits produced |
| 2406 |
unsigned int total_bin_count; // total number of bins produced by bac |
| 2407 |
unsigned int cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 2408 |
unsigned int cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 2409 |
unsigned int cumulative_skip_cost; // sum of all P skip MB SA(T)Ds |
| 2410 |
unsigned short intra_mb_count; // number of intra MBs. |
| 2411 |
unsigned short inter_mb_count; // number of inter MBs. |
| 2412 |
unsigned short skip_mb_count; // number of skip MBs. |
| 2413 |
unsigned short reserved0; // align to a word boundary |
| 2414 |
} nvenc_vp8_segment_stat_s; // 44 Bytes |
| 2415 |
|
| 2416 |
typedef struct |
| 2417 |
{ |
| 2418 |
unsigned int bitstream_byte_count; // output bytes in main partition bitstream buffer |
| 2419 |
unsigned int bitstream_res_byte_count[8]; // output bytes in residual partition bitstream buffers |
| 2420 |
unsigned int cycle_count; // total cycles taked for execute. Written out only when DumpCycleCount in SetControlParams is set |
| 2421 |
unsigned int picture_index; // value received from SetPictureIndex method |
| 2422 |
unsigned int error_status; // report error if any |
| 2423 |
unsigned int total_bit_count; // picture size in bits /*(sum of header's & all the partition's bits. actually redundant)*/ |
| 2424 |
unsigned short pic_type; // copied from pic_type in nvenc_h264_pic_control_s |
| 2425 |
unsigned short ave_activity; // report average activity if activity based RC enabled |
| 2426 |
unsigned short num_segments; // number of segments produced |
| 2427 |
unsigned short base_qindex; // base_qindex used for the frame |
| 2428 |
nvenc_vp8_segment_stat_s segment_stat[4]; |
| 2429 |
unsigned int total_intra_cost; // sum of all intra best mode's SATDs, each mb has one intra SATD. |
| 2430 |
unsigned int total_inter_cost; // sum of all inter best mode's SATDs, each mb has one inter SATD. |
| 2431 |
unsigned int reserved1[3]; |
| 2432 |
} nvenc_vp8_stat_data_s; // 256 Bytes |
| 2433 |
|
| 2434 |
// Stats stored in HW format. |
| 2435 |
typedef struct |
| 2436 |
{ |
| 2437 |
unsigned int coef_token_cnt[3][3][4][8][4]; // [value/4][ctx][value%4][band][type] 4608 bytes (4096 + 512) |
| 2438 |
unsigned int mv_stats[20][4]; |
| 2439 |
} nvenc_vp8_token_data_s; |
| 2440 |
|
| 2441 |
// 8 bytes |
| 2442 |
|
| 2443 |
|
| 2444 |
////////////////////////////////////////////////////////////////// |
| 2445 |
// **************************************************************************** |
| 2446 |
// **************************************************************************** |
| 2447 |
// AV1 |
| 2448 |
// **************************************************************************** |
| 2449 |
// **************************************************************************** |
| 2450 |
//AV1 loop filter config |
| 2451 |
typedef struct |
| 2452 |
{ |
| 2453 |
unsigned char filter_level[4]; // loop filter strength based on edge direction and component - vertical/horizontal edge luma (0,1) and chroma (2,3) |
| 2454 |
char ref_delta[8]; // Loop filter level adjustments based on reference frame type - 0 = Intra, Last, Last2+Last3, GF, BRF, ARF2, AR |
| 2455 |
int mode_delta_0 : 8; // Loop filter level adjustments based on mode type - 0 = ZERO_MV, MV |
| 2456 |
int mode_delta_1 : 8; // Loop filter level adjustments based on mode type - 0 = ZERO_MV, MV |
| 2457 |
unsigned int update_ref_delta_flag : 8; // update delta for reference i if bit i is set |
| 2458 |
unsigned int update_mode_delta0 : 1; |
| 2459 |
unsigned int update_mode_delta1 : 1; |
| 2460 |
unsigned int delta_enabled : 1; // disable loop-filter |
| 2461 |
unsigned int sharpness : 3; // sharpness level |
| 2462 |
unsigned int reserved : 2; |
| 2463 |
}nvenc_av1_loop_filter_cfg_s; //16 bytes |
| 2464 |
|
| 2465 |
//AV1 CDEF config |
| 2466 |
// NOTE: the following is mandated by the current HW implementation: |
| 2467 |
// y_strength_0 and uv_strength_0 must always be set to 0 |
| 2468 |
// Luma/chroma strength combination 0 must always be (y_strength_0, uv_strength_0) i.e. (0,0) |
| 2469 |
typedef struct |
| 2470 |
{ |
| 2471 |
unsigned int cdef_damping_minus_3 : 2; // controls the amount of damping in the deringing filter |
| 2472 |
unsigned int y_strength_0 : 6; // specify (strength primary filter 0 << 2) | strength secondary filter 0 (luma) |
| 2473 |
unsigned int y_strength_1 : 6; // specify (strength primary filter 1 << 2) | strength secondary filter 1 (luma) |
| 2474 |
unsigned int y_strength_2 : 6; // specify (strength primary filter 2 << 2) | strength secondary filter 2 (luma) |
| 2475 |
unsigned int y_strength_3 : 6; // specify (strength primary filter 3 << 2) | strength secondary filter 3 (luma) |
| 2476 |
unsigned int y_strength_4 : 6; // specify (strength primary filter 4 << 2) | strength secondary filter 4 (luma). Not supported. |
| 2477 |
unsigned int y_strength_5 : 6; // specify (strength primary filter 5 << 2) | strength secondary filter 5 (luma). Not supported. |
| 2478 |
unsigned int y_strength_6 : 6; // specify (strength primary filter 6 << 2) | strength secondary filter 6 (luma). Not supported. |
| 2479 |
unsigned int y_strength_7 : 6; // specify (strength primary filter 7 << 2) | strength secondary filter 7 (luma). Not supported. |
| 2480 |
unsigned int cdef_bits : 2; // support up to 3. the number of bits needed to specify w2ich CDEF filter to apply |
| 2481 |
unsigned int uv_strength_0 : 6; // specify (strength primary filter 0 << 2) | strength secondary filter 0 (chroma) |
| 2482 |
unsigned int uv_strength_1 : 6; // specify (strength primary filter 1 << 2) | strength secondary filter 1 (chroma) |
| 2483 |
unsigned int uv_strength_2 : 6; // specify (strength primary filter 2 << 2) | strength secondary filter 2 (chroma) |
| 2484 |
unsigned int uv_strength_3 : 6; // specify (strength primary filter 3 << 2) | strength secondary filter 3 (chroma) |
| 2485 |
unsigned int uv_strength_4 : 6; // specify (strength primary filter 4 << 2) | strength secondary filter 4 (chroma). Not supported. |
| 2486 |
unsigned int uv_strength_5 : 6; // specify (strength primary filter 5 << 2) | strength secondary filter 5 (chroma). Not supported. |
| 2487 |
unsigned int uv_strength_6 : 6; // specify (strength primary filter 6 << 2) | strength secondary filter 6 (chroma). Not supported. |
| 2488 |
unsigned int reserved0 : 2; // Reserved bits |
| 2489 |
unsigned int uv_strength_7 : 6; // specify (strength primary filter 7 << 2) | strength secondary filter 7 (chroma). Not supported. |
| 2490 |
unsigned int cdef_y_strength_num_minus_1 : 3; // HW will only considers y strength values from index 0 to cdef_y_strength_num_minus_1 (from 0 to 7) |
| 2491 |
unsigned int cdef_uv_strength_num_minus_1 : 3; // HW will only considers uv strength values from index 0 to cdef_uv_strength_num_minus_1 (from 0 to 7) |
| 2492 |
unsigned int cdef_combined_idx_minus_1 : 3; // Specify max number of unique (y_strength_x, uv_strength_y) combinations for HW to consider (from 0 to 7) |
| 2493 |
// bits[2:0] must be set to 0 |
| 2494 |
unsigned int reserved1 : 17; // Reserved bits |
| 2495 |
unsigned int cdef_combined_luma_idx : 24; // List of (cdef_combined_idx_minus_1 + 1) y strength indexes, each index using 3 bits and bits[3(n+1)-1:3n] corresponding to index of y_strength in combination n |
| 2496 |
unsigned int reserved2 : 8; // Reserved bits |
| 2497 |
unsigned int cdef_combined_chroma_idx : 24; // List of (cdef_combined_idx_minus_1 + 1) uv strength indexes, each index using 3 bits and bits[3(n+1)-1:3n] corresponding to index of uv_strength in combination n |
| 2498 |
// bits[2:0] must be set to 0 |
| 2499 |
unsigned int reserved3 : 8; // Reserved bits |
| 2500 |
}nvenc_av1_cdef_cfg_s; //24 bytes |
| 2501 |
|
| 2502 |
//AV1 Loop Restoration Filter config |
| 2503 |
typedef struct |
| 2504 |
{ |
| 2505 |
unsigned int y_lr_type : 2; // luma restoration type |
| 2506 |
unsigned int uv_lr_type : 2; // chroma restoration type |
| 2507 |
unsigned int y_lr_unit_size : 2; // luma restoration unit size 0:32x32, 1:64x64, 2:128x128,3:256x256; |
| 2508 |
unsigned int uv_lr_unit_size : 2; // chroma restoration unit size 0:32x32, 1:64x64, 2:128x128,3:256x256; |
| 2509 |
unsigned int lr_codebook_enable : 16; // bit enable for up to 16 codebooks |
| 2510 |
unsigned int lr_pe_on_edge_enable : 1; // enable bit for LR parameter estimation on picture boundary SBs |
| 2511 |
unsigned int reserved : 7; // Reserved bits |
| 2512 |
}nvenc_av1_lr_filter_cfg_s; // 4 bytes |
| 2513 |
|
| 2514 |
//AV1 quant config |
| 2515 |
typedef struct |
| 2516 |
{ |
| 2517 |
unsigned char base_q_idx; // Base frame qindex (index in quant lookup table) 0...255 |
| 2518 |
char delta_y_dc; // explicit qindex adjustment for y dccoefficient, -63...63 |
| 2519 |
char delta_uv_dc; // qindex adjustment for uv accoefficient, -63...63 |
| 2520 |
char delta_uv_ac; // qindex adjustment for uv dccoefficient, -63...63 |
| 2521 |
}nvenc_av1_quant_cfg_s; //4 bytes |
| 2522 |
|
| 2523 |
typedef struct |
| 2524 |
{ |
| 2525 |
unsigned short frame_width_minus1; // frame width in pixels minus 1, range 191-8191 |
| 2526 |
unsigned short frame_height_minus1; // frame height in pixels minus 1, range 127-8191 |
| 2527 |
unsigned short sfc_pitch; // pitch of luma plane |
| 2528 |
unsigned short sfc_pitch_chroma; // pitch of chroma plane |
| 2529 |
unsigned short sfc_trans_mode; // least 3 significient bits are used to stand for 8 modes, normal, xflip, yflip etc. |
| 2530 |
unsigned short reserved2; // pad to NvS32 |
| 2531 |
|
| 2532 |
unsigned int luma_top_frm_offset; // offset of luma top field or frame in units of 256 |
| 2533 |
unsigned int luma_bot_offset; // (deprecated) offset of luma bottom field in units of 256. Not used if frame format. |
| 2534 |
unsigned int chroma_top_frm_offset; // offset of chroma top field or frame, or offset of chroma U(Cb) plane in planar mode, both in units of 256 |
| 2535 |
unsigned int chroma_bot_offset; // (deprecated) offset of chroma bottom field, not used if frame format. Or offset of chroma V(Cr) plane in planar mode. Both in units of 256. |
| 2536 |
unsigned int block_height : 7; // BL mapping block height setting |
| 2537 |
unsigned int tiled_16x16 : 1; // (deprecated) Surface is 16x16 tiled instead of BL mapped (must be 1 for refpics) (deprecated) |
| 2538 |
unsigned int memory_mode : 2; // (deprecated) 0: semi-planar, 1: planar |
| 2539 |
unsigned int nv21_enable : 1; // (deprecated) the surface format is yuv or yvu |
| 2540 |
unsigned int input_bl_mode : 2; // the input block linear mode: 0~gpu bl; 1~tegra bl; 2~naive bl this is only for cmod and cmod/plugin. TBL only for Orin. |
| 2541 |
unsigned int src_in_lsb : 1; // pixel is LSB aligned in 16bit container, only valid for high-bitdepth |
| 2542 |
unsigned int reserved : 18; // pad to NvU32 |
| 2543 |
} nvenc_av1_surface_cfg_s; // 32 bytes |
| 2544 |
|
| 2545 |
typedef struct |
| 2546 |
{ |
| 2547 |
//sequence header |
| 2548 |
unsigned int min_cu_size : 2; //0-8x8, 1-16x16, 2-32x32, 3-64x64. Orin AV1 supports 1~3 |
| 2549 |
unsigned int max_cu_size : 2; //0-8x8, 1-16x16, 2-32x32, 3-64x64 Orin AV1 supports 1~3 |
| 2550 |
|
| 2551 |
unsigned int chroma_format_idc : 2; // 0=not supported, 1=yuv4:2:0, 2=yuv4:2:2, 3=yuv4:4:4. Only support YUV420 |
| 2552 |
unsigned int monochrome_flag : 1; // only luma component if flag is set. Not supported. |
| 2553 |
unsigned int bit_depth_minus_8 : 4; // luma/chroma should have the same value. support 0 (8bit), 2(10bit), 4(12bit) for Orin AV1 |
| 2554 |
unsigned int input_bit_depth_minus8 : 4; // Internal high-bitdepth support in HW |
| 2555 |
// the field is to specify the bit-depth of source yuv. the decoded yuv bitdepth is specified nvenc_av1_seq_hdr_data_s::bit_depth_minus_8; HW will do the bitdepth conversion from source yuv-> recon yuv |
| 2556 |
// |
| 2557 |
unsigned int enable_filter_intra : 1; // enables/disables filterintra. Not supported |
| 2558 |
unsigned int enable_intra_edge_filter : 1; // enables/disables corner/edge/upsampling. Not supported |
| 2559 |
unsigned int enable_interintra_compound : 1; // enables/disables interintra_compound. Not supported |
| 2560 |
unsigned int enable_masked_compound : 1; // enables/disables masked compound. Not supported |
| 2561 |
unsigned int enable_dual_filter : 1; // enable or disable vertical and horiz filter selection. Not supported |
| 2562 |
unsigned int force_screen_content_tools : 2; // 0 - force off. Not supported. |
| 2563 |
// 1 - force on |
| 2564 |
// 2 - adaptive |
| 2565 |
unsigned int force_integer_mv : 2; // Not supported.0 - Not to force. MV can be in 1/4 or 1/8 |
| 2566 |
// 1 - force to integer |
| 2567 |
// 2 - adaptive |
| 2568 |
unsigned int enable_jnt_comp : 1; // enables/disables joint compound modes. not supported. Not supported. |
| 2569 |
unsigned int enable_ref_frame_mvs : 1; // enables/disables ref frame mvs. not supported. Not supported. |
| 2570 |
unsigned int enable_warped_motion : 1; // enables/disables warped motion for sequence. Not supported. |
| 2571 |
unsigned int enable_superres : 1; // enables/disables superres for sequence. Not supported. |
| 2572 |
unsigned int enable_cdef : 1; // enables/disables CDEF |
| 2573 |
unsigned int enable_restoration : 1; // enables/disables restoration filter, only support self-guided filter |
| 2574 |
unsigned int reserved0 : 2; // reserved bits |
| 2575 |
} nvenc_av1_seq_hdr_data_s; // 4 bytes |
| 2576 |
|
| 2577 |
//frame header |
| 2578 |
typedef struct |
| 2579 |
{ |
| 2580 |
unsigned int frame_type : 2; // 0:Key frame, 1:Inter frame, 2:intra only, 3:s-frame |
| 2581 |
unsigned int disable_cdf_update : 1; // disable CDF update during symbol decoding |
| 2582 |
unsigned int allow_screen_content_tools : 1; // screen content tool enable. Not supported. |
| 2583 |
unsigned int force_integer_mv : 1; // AMVR enable. Not supported. |
| 2584 |
unsigned int allow_high_precision_mv : 1; // 1/8 precision mv enabl. Not supported. |
| 2585 |
unsigned int switchable_motion_mode : 1; // 0: simple motion mode, 1: SIMPLE, OBMC, LOCAL WARP. Not supported, fixed to 0. |
| 2586 |
unsigned int is_filter_switchable : 1; // 1: filter selection is signaled at block level |
| 2587 |
unsigned int use_ref_frame_mvs : 1; // Not supported. enable/disable use of motion vector info from previous frame |
| 2588 |
unsigned int frame_end_update_cdf : 1; // enable/disable CDF update at end of frame |
| 2589 |
unsigned int interp_filter : 3; // interpolation filter : EIGHTTAP,...., SWITCHABLE. Only support EIGHTTAP &EIGHTTAP_SMOOTH. |
| 2590 |
unsigned int delta_q_present_flag : 1; // quantizer index delta values are present in the block level |
| 2591 |
unsigned int delta_q_res : 2; // left shift will apply to decoded quantizer index delta values. Not supported. |
| 2592 |
unsigned int allow_intrabc : 1; // IBC enable. Not supported. |
| 2593 |
unsigned int delta_lf_present_flag : 1; // specify whether loop filter delta values are present in the block level. Not supported. |
| 2594 |
unsigned int delta_lf_res : 2; // specify the left shift will apply to decoded loop filter values. Not supported. |
| 2595 |
unsigned int delta_lf_multi : 1; // seperate loop filter deltas for Hy,Vy,U,V edges. Not supported. |
| 2596 |
unsigned int using_qmatrix : 1; // enable quant matrix. Not supported. |
| 2597 |
unsigned int frame_tx_mode : 2; // 0:LARGEST (32x32), 1:SELECT, 2: ONLY4x4 (lossless only) |
| 2598 |
unsigned int frame_reference_mode : 2; // 0: single, 1: select, 2: compound |
| 2599 |
unsigned int skip_mode_flag : 1; // skip mode present flag. Not supported. |
| 2600 |
unsigned int allow_warped_motion : 1; // frame level warped motion enable. Not supported. |
| 2601 |
unsigned int reduced_tx_set_flag : 1; // whether the frame is restricted to a reduced subset of the full set of transform types. Not supported. |
| 2602 |
unsigned int coded_lossless : 1; // whether lossless coding is used for frame or not |
| 2603 |
unsigned int reserved0 : 2; // reserved bits |
| 2604 |
|
| 2605 |
unsigned int ssim_rdo_enable : 1; //enable variance based SSIM-RDO |
| 2606 |
unsigned int ssim_rdo_alt_frame_scale : 8; //frame level scaling factor for altref frame |
| 2607 |
unsigned int bicompound_enable : 1; //enable bi-compound for B-frames. |
| 2608 |
unsigned int unicompound_enable : 1; //enable uni-compound for P/B-frames. (only support in Pframes) |
| 2609 |
unsigned int mc_hfilter : 4; //bitmask for horizontal filters enabled 0: EIGHTTAP, 1: EIGHTTAP_SMOOTH, |
| 2610 |
unsigned int mc_vfilter : 4; //bitmask for vertical filters enabled 0: EIGHTTAP, 1: EIGHTTAP_SMOOTH, |
| 2611 |
|
| 2612 |
unsigned int segmentation_enable : 1; //enable segmentation |
| 2613 |
unsigned int segmentation_update_map : 1; //enable segmentation update map ,force 1 |
| 2614 |
unsigned int segment_temporal_update : 1; //enable segmentation temporal update,force 0 |
| 2615 |
unsigned int segIdPreSkip : 1; //segment id pre skip encoding ,force 1 |
| 2616 |
unsigned int LastActiveSegId : 3; //highest number of active segment . force 2 |
| 2617 |
|
| 2618 |
unsigned int reserved1 : 6; |
| 2619 |
|
| 2620 |
|
| 2621 |
|
| 2622 |
unsigned char ref_frame_sign_bias[8]; //Used by AV1 MVP. A sign bias equal to 0 indicates that the reference frame is a forwards reference ( |
| 2623 |
nvenc_av1_loop_filter_cfg_s lf_cfg; // loop filter configuration |
| 2624 |
nvenc_av1_cdef_cfg_s cdef_cfg; // CDEF configuration |
| 2625 |
nvenc_av1_quant_cfg_s quant_cfg; // quantizer configuration |
| 2626 |
nvenc_av1_lr_filter_cfg_s lrf_cfg; // loop restoration filter configuration |
| 2627 |
} nvenc_av1_frame_hdr_data_s; // 64 bytes |
| 2628 |
|
| 2629 |
// needed only if picture level RC is enabled; as part of picture level RC, it will also perform HRD verification |
| 2630 |
// The contents of this file is produced by the driver during sequence level operations and should not be modified. |
| 2631 |
typedef struct |
| 2632 |
{ |
| 2633 |
unsigned char hrd_type; // nal (=2) and vcl (=1) type |
| 2634 |
unsigned char QP[3]; // QP for 0:P picture, ,1:B picture 2; I picture |
| 2635 |
unsigned char minQP[3]; // min QP for 0:P picture, ,1:B picture 2; I picture |
| 2636 |
unsigned char maxQP[3]; // max QP for 0:P picture, ,1:B picture 2; I picture |
| 2637 |
char maxQPD; // max QP delta between two consecutive QP updates |
| 2638 |
char baseQPD; // initial QP delta between two consecutive QP updates |
| 2639 |
unsigned int rhopbi[3]; // 23.8 unsigned fixed point quant ratios P/I, B/I, I/I(=1) |
| 2640 |
|
| 2641 |
unsigned int framerate; // fps |
| 2642 |
unsigned int buffersize; // total buffer size |
| 2643 |
|
| 2644 |
unsigned int nal_cpb_size; // size in bits |
| 2645 |
unsigned int nal_bitrate; // rate in bps |
| 2646 |
unsigned int vcl_cpb_size; // size in bits |
| 2647 |
unsigned int vcl_bitrate; // rate in bps |
| 2648 |
|
| 2649 |
unsigned int gop_length; // I period, gop_length == 0xffffffff is used for infinite gop length |
| 2650 |
unsigned int Np; // 27.4 unsigned fixed (gop_length + num_b_frames -1)/(num_b_frames + 1) |
| 2651 |
unsigned int Bmin; // 23.8 unsigned fixed point min buffer level, updated by driver at sequence level only. |
| 2652 |
unsigned int Ravg; // 23.8 unsigned fixed point average rate, updated by driver at sequence level only. |
| 2653 |
unsigned int R; // 23.8 unsigned fixed point rate |
| 2654 |
|
| 2655 |
unsigned char ab_alpha; // VBR/CBR: min/target quality level (1..51); rcmode4: weight of prev frame activity compared to part of the current picture |
| 2656 |
unsigned char ab_beta; // lookahead info1: ratio of inter over intra complexity (fix8); rcmode4: weight of current activity compared to neighbors |
| 2657 |
unsigned short prev_act; // lookahead info2: NVENC_LOOKAHEAD_INFO2; rcmode4: average activity level of the first reference (L0) |
| 2658 |
|
| 2659 |
unsigned char reserved0; |
| 2660 |
unsigned char single_frame_vbv; // 1: VBV buffer size is set to average frame size; 0: otherwise |
| 2661 |
unsigned char two_pass_rc; // 0: single pass rc , 1: first pass of 2 pass rc 2: second pass of 2 pass rc |
| 2662 |
unsigned char rc_class; // reserved |
| 2663 |
unsigned char first_pass_source_half_scaled; // 0 : first pass on half resolution 1: first pass on full resolution (along with NVENC_QRES_FLAG_XXX) |
| 2664 |
unsigned char i_size_ratio_x; // ratio between I picture target size over average picture size numerator |
| 2665 |
unsigned char i_size_ratio_y; // ratio between I picture target size over average picture size denominator |
| 2666 |
unsigned char aqMode; // 0: disable adaptive quantization 1: AQ mode fast (unimplemented) 2: AQ mode full (unimplemented) 3: External qp delta map |
| 2667 |
unsigned char dump_aq_stats; // 1: dump aq stats |
| 2668 |
unsigned char ext_scene_change_flag; // Scene change flag set as hint by external preprocessing unit. 0: No scene change, 1: Current picture is first in scene |
| 2669 |
unsigned char ext_picture_rc_hints; // picture rc hints are set by external source. 0: no hints, 1: hints are avaialble for current picture |
| 2670 |
unsigned char session_max_qp; // Current frame qp will never exceed this value when ext_picture_rc_hints is set |
| 2671 |
unsigned int ext_intra_satd; // If non zero this represents the intra SATD for current picture computed by external preprocessing unit |
| 2672 |
unsigned char reserved2[8]; |
| 2673 |
} nvenc_av1_rc_s; // 96 bytes |
| 2674 |
////////////////////////////////////////////////////////////////// |
| 2675 |
|
| 2676 |
|
| 2677 |
typedef struct |
| 2678 |
{ |
| 2679 |
char l0[4]; // reference picture list 0, maximum 4 forward reference frame |
| 2680 |
char l1[3]; // reference picture list 1, maximum 3 backward reference frame |
| 2681 |
char temp_dist_l0[4]; // temporal distance of each ref pic in list 0 DiffPicOrderCnt( currPicOrField, pic0) ) |
| 2682 |
char temp_dist_l1[3]; // temporal distance of each ref pic in list 1 DiffPicOrderCnt( currPicOrField, pic1) ) |
| 2683 |
char poc_entry_l0[4]; // ref pic list to ref pic entry map |
| 2684 |
char poc_entry_l1[3]; // ref pic list to ref pic entry map |
| 2685 |
char reserved0[3]; |
| 2686 |
|
| 2687 |
unsigned int max_byte_count_before_resid_zero; // Maximum byte count before microcode forces the residuals to zero. |
| 2688 |
|
| 2689 |
unsigned int mode_decision_only_flag : 1; // only output av1 mode and mocomp result, no recon and bitstream output. Not supported. |
| 2690 |
unsigned int cu_early_terminatiom_enable : 1; // 0: disable CU early termination, 1: enable CU early termination. Not supported. |
| 2691 |
unsigned int pic_output_flag : 1; // enable recon picture output |
| 2692 |
unsigned int input_sub_sampling : 1; // used by PDMA 2x2 sub-sampling. Not supported. |
| 2693 |
unsigned int output_sub_sampling : 1; // used by MPEB 2x2 sub-sampling output. Not supported. |
| 2694 |
unsigned int num_ref_idx_l0_active : 3; // [0..4] |
| 2695 |
unsigned int num_ref_idx_l1_active : 3; // [0..3] |
| 2696 |
unsigned int tile_idx_start : 8; // index of entry in av1_tile_control_s corresponding to first tile in strip |
| 2697 |
unsigned int num_strip_tiles_minus1 : 8; // number of tiles in strip (= num_tiles_minus1 if single strip) |
| 2698 |
unsigned int force_lf_on_flag : 1; // minimum level of loop filter is set to 1 instead of 0 when flag is set |
| 2699 |
unsigned int force_lf_off_flag : 1; // maximum level of loop filter is set to 0 when flag is set (i.e. filtering off) |
| 2700 |
unsigned int vp9_skip_chroma : 1; // HW will only process luma and skip chroma. used for vp9 autoprofile TODO: temporary add here, need to move to VP9 dedicated structrue. |
| 2701 |
unsigned int reserved1 : 2; |
| 2702 |
|
| 2703 |
unsigned int max_tu_depth_inter : 2; // max_transform_hierarchy_depth_inter. Orin AV1 supports 0~2 |
| 2704 |
unsigned int max_tu_depth_intra : 2; // max_transform_hierarchy_depth_intra. Orin AV1 supports 0 |
| 2705 |
unsigned int num_tiles_minus1 : 8; // number of entries in av1_tile_control_s |
| 2706 |
unsigned int enable_source_image_padding : 1; // enable/disable HW source image padding |
| 2707 |
unsigned int luma_padding_mode : 2; // HW souce padding mode for luma: 0: copy frame boundary, 1: zero padding, 2: half value, i.e 1<<(bitdepth-1) |
| 2708 |
unsigned int chroma_padding_mode : 2; // HW souce padding mode for chroma: 0: copy frame boundary, 1: zero padding, 2: half value, i.e 1<<(bitdepth-1) |
| 2709 |
unsigned int reserved2 : 15; |
| 2710 |
unsigned int prob_tables_offset; // When non-zero, ucode will adjust the base address of the probability tables surfaces based on picture QP value as follows |
| 2711 |
unsigned int tile_control_offset; // offset from start of top level "nvenc_av1_drv_pic_setup_s" structure |
| 2712 |
unsigned int hist_buf_size; // size in bytes of the buffer allocated for history |
| 2713 |
unsigned int bitstream_buf_size; // size in bytes of the buffer allocated for bitstream slice/mb data |
| 2714 |
unsigned int bitstream_start_pos; //start position in bitstream buffer where data should be written (byte offset) |
| 2715 |
unsigned int pic_type : 2; // 0 = P, 1 = B, 2 = I, 3 = IDR |
| 2716 |
unsigned int ref_pic_flag : 1; // reference picture (0 = no, 1 = yes) |
| 2717 |
unsigned int codec : 3; |
| 2718 |
unsigned int qpfifo : 1; // 0 = Polling mode, 1 = Interrupt mode. |
| 2719 |
unsigned int num_tile_cols : 8; // Number of tile columns in picture |
| 2720 |
unsigned int num_tile_rows : 8; // Number of tile rows in picture |
| 2721 |
unsigned int ctx_update_tile_id : 8; // update cdf for the specified tile id. value 255 is reserved to enable HW auto update CDF from tile with maximun tile data length |
| 2722 |
unsigned int stats_fifo_enable : 1; // enable per-sb short stats fifo |
| 2723 |
|
| 2724 |
// 1 = static slice insertion based on slice control array |
| 2725 |
// (insert slices as defined in array, if all array elements used |
| 2726 |
// before end of picture is reached, start over with entry 0) |
| 2727 |
|
| 2728 |
unsigned int mpec_threshold; // collect mpec stats after threshold mbs |
| 2729 |
// when mpec_stat_on is enabled and mpec_threshold is not equal to picWidthInMbs, |
| 2730 |
unsigned int tile_stat_offset; // offset from start of top level "nvenc_av1_stat_data_s" structure |
| 2731 |
// rc mode 0 will be used to collect stats |
| 2732 |
unsigned int mpec_stat_offset; // offset from start of top level "nvenc_stat_data_s" structure |
| 2733 |
// to start of array of "nvenc_mpec_stat_s" structure |
| 2734 |
unsigned int aq_stat_offset; // to start of array of "nvenc_av1_aq_stat_s" structure |
| 2735 |
|
| 2736 |
unsigned int rc_pass2_offset; |
| 2737 |
|
| 2738 |
unsigned int act_stat_offset; // offset from start of "nvenc_rc_pic_stat_s" structure, points to data for external QP map |
| 2739 |
unsigned int external_hint_number; // total external hint number for current frame, maximun external hints per SB: 128. |
| 2740 |
|
| 2741 |
unsigned char stripID; // strip index (from 0 to strips_in_frame-1) |
| 2742 |
unsigned char strips_in_frame; // total number of strips in frame |
| 2743 |
unsigned short strip_width_in_sbs; // strip width in SB unit,must be frame_width |
| 2744 |
unsigned short strip_height_in_sbs; // strip height in SB unit |
| 2745 |
unsigned short strip_width_offset_in_sbs; // Column index of first SB in strip ,must be 0 |
| 2746 |
unsigned short strip_height_offset_in_sbs; // Row index of first SB in strip |
| 2747 |
unsigned short reserved5; |
| 2748 |
|
| 2749 |
unsigned int stats_fifo_offset; // offset from start of top level "nvenc_stat_data_s" structure |
| 2750 |
unsigned char reserved6[8]; |
| 2751 |
} nvenc_av1_pic_control_s; // 112 bytes |
| 2752 |
|
| 2753 |
typedef struct |
| 2754 |
{ |
| 2755 |
unsigned int cu8x8_part_2Nx2N_enable : 1; // enable prediction for Inter CU8x8 2Nx2N |
| 2756 |
unsigned int cu16x16_part_2Nx2N_enable : 1; // enable prediction for Inter CU16x16 2Nx2N |
| 2757 |
unsigned int cu16x16_part_2NxN_enable : 1; // enable prediction for Inter CU16x16 2NxN |
| 2758 |
unsigned int cu16x16_part_Nx2N_enable : 1; // enable prediction for Inter CU16x16 Nx2N |
| 2759 |
unsigned int cu32x32_part_2Nx2N_enable : 1; // enable prediction for Inter CU32x32 2Nx2N |
| 2760 |
unsigned int cu32x32_part_2NxN_enable : 1; // enable prediction for Inter CU32x32 2NxN |
| 2761 |
unsigned int cu32x32_part_Nx2N_enable : 1; // enable prediction for Inter CU32x32 Nx2N |
| 2762 |
unsigned int cu64x64_part_2Nx2N_enable : 1; // enable prediction for Inter CU64x64 2Nx2N |
| 2763 |
unsigned int cu64x64_part_2NxN_enable : 1; // enable prediction for Inter CU64x64 2NxN |
| 2764 |
unsigned int cu64x64_part_Nx2N_enable : 1; // enable prediction for Inter CU64x64 Nx2N |
| 2765 |
unsigned int tusize_4x4_enable : 1; //enable 4x4 transform |
| 2766 |
unsigned int tusize_8x8_enable : 1; //enable 8x8/8x4/4x8 transform |
| 2767 |
unsigned int tusize_16x16_enable : 1; //enable 16x16/16x8/8x16 transform |
| 2768 |
unsigned int tusize_32x32_enable : 1; //enable 32x32/32x16/16x32 transform |
| 2769 |
unsigned int tusize_64x64_enable : 1; //enable 64x64/64x32/32x16 transform |
| 2770 |
unsigned int intra_tusize_decision : 1; //enable intra tu size decision, reserved, not implemented in HW |
| 2771 |
unsigned int pskip_enable : 1; //enable p_skip encoding. Not supported |
| 2772 |
unsigned int tu_search_basedon_pusize : 1; //enable tu search based on pu size. Not supported |
| 2773 |
unsigned int multiply_bias_with_lambda : 1; // setting this to 1 will result in bias being multiplied by lambda |
| 2774 |
unsigned int skip_evaluate_enable_cu8 : 1; // 0: not check skip ssd cost; 1, evaluate skip ssd cost. Not supported |
| 2775 |
unsigned int skip_evaluate_enable_cu16 : 1; // 0: not check skip ssd cost; 1, evaluate skip ssd cost. Not supported |
| 2776 |
unsigned int skip_evaluate_enable_cu32 : 1; // 0: not check skip ssd cost; 1, evaluate skip ssd cost. Not supported |
| 2777 |
unsigned int skip_evaluate_enable_cu64 : 1; // 0: not check skip ssd cost; 1, evaluate skip ssd cost. Not supported |
| 2778 |
unsigned int inter_rdo_cnt_8x8 : 3; // #RD cost evaluation for inter top N SATD winner partitions of CU8x8. support 0~4 |
| 2779 |
unsigned int inter_rdo_cnt_16x16 : 3; // #RD cost evaluation for inter top N SATD winner partitions of CU16x16. support 0~4 |
| 2780 |
unsigned int inter_rdo_cnt_32x32 : 3; // #RD cost evaluation for inter top N SATD winner partitions of CU32x32. support 0~4 |
| 2781 |
|
| 2782 |
unsigned int inter_rdo_cnt_64x64 : 3; // #RD cost evaluation for inter top N SATD winner partitions of Cu64x64. support 0~4 |
| 2783 |
unsigned int intra_rdo_cnt_4x4 : 3; // RD cost evaluation for N top satd winner modes of intra 4x4. Not supported |
| 2784 |
unsigned int intra_rdo_cnt_8x8 : 3; // RD cost evaluation for N top satd winner modes of intra 8x8. support 0~4 |
| 2785 |
unsigned int intra_rdo_cnt_8x4 : 3; // RD cost evaluation for N top satd winner modes of intra 8x4. support 0~4 |
| 2786 |
unsigned int intra_rdo_cnt_4x8 : 3; // RD cost evaluation for N top satd winner modes of intra 4x8. support 0~4 |
| 2787 |
unsigned int intra_rdo_cnt_16x16 : 3; // RD cost evaluation for N top satd winner modes of intra 16x16. support 0~4 |
| 2788 |
unsigned int intra_rdo_cnt_16x8 : 3; // RD cost evaluation for N top satd winner modes of intra 16x8. support 0~4 |
| 2789 |
unsigned int intra_rdo_cnt_8x16 : 3; // RD cost evaluation for N top satd winner modes of intra 8x16. support 0~4 |
| 2790 |
unsigned int intra_rdo_cnt_32x32 : 3; // RD cost evaluation for N top satd winner modes of intra 32x32. support 0~4 |
| 2791 |
unsigned int intra_rdo_cnt_32x16 : 3; // RD cost evaluation for N top satd winner modes of intra 32x16. support 0~4 |
| 2792 |
unsigned int luma_all_zero_check : 1; // enable TU level luma coefficient all zero check |
| 2793 |
unsigned int chroma_all_zero_check : 1; // enable TU level chroma coefficient all zero check. Not supported |
| 2794 |
|
| 2795 |
unsigned int intra_rdo_cnt_16x32 : 3; // RD cost evaluation for N top satd winner modes of intra 16x32. support 0~4 |
| 2796 |
unsigned int intra_rdo_cnt_64x64 : 3; // RD cost evaluation for N top satd winner modes of intra 64x64. support 0~4 |
| 2797 |
unsigned int intra_rdo_cnt_64x32 : 3; // RD cost evaluation for N top satd winner modes of intra 64x32. support 0~4 |
| 2798 |
unsigned int intra_rdo_cnt_32x64 : 3; // RD cost evaluation for N top satd winner modes of intra 32x64. support 0~4 |
| 2799 |
unsigned int num_1div8_lambda_ssd : 7; // number of 1/8 lambda ssd for lambda_coef delta. lambda_coef = (1+num_1div8_lambda_ssd*1/8)*lambda_ssd |
| 2800 |
unsigned int mv_cost_enable : 1; // enable mv cost calculations in mode decision |
| 2801 |
unsigned int reserved0 : 12; |
| 2802 |
|
| 2803 |
//Intra Luma angular delta mode controls: begin |
| 2804 |
unsigned int intra_luma_dir_delta_64x64_enable_0; // bit mask for enabling angular delta modes: {D135, D45, H, V} * {-3,-2,-1,0,1,2,3} |
| 2805 |
unsigned int intra_luma_dir_delta_64x64_enable_1; // bit mask for enabling angular delta modes: {D67, D203, D157, D113} * {-3,-2,-1,0,1,2,3} |
| 2806 |
unsigned int intra_luma_dir_delta_64x32_enable_0; |
| 2807 |
unsigned int intra_luma_dir_delta_64x32_enable_1; // for 64x32 and 32x64 |
| 2808 |
|
| 2809 |
unsigned int intra_luma_dir_delta_32x32_enable_0; // bit mask for enabling angular delta modes: {D135, D45, H, V} * {-3,-2,-1,0,1,2,3} |
| 2810 |
unsigned int intra_luma_dir_delta_32x32_enable_1; // bit mask for enabling angular delta modes: {D67, D203, D157, D113} * {-3,-2,-1,0,1,2,3} |
| 2811 |
unsigned int intra_luma_dir_delta_32x16_enable_0; // for 32x16 and 16x32 |
| 2812 |
unsigned int intra_luma_dir_delta_32x16_enable_1; |
| 2813 |
|
| 2814 |
unsigned int intra_luma_dir_delta_16x16_enable_0; // bit mask for enabling angular delta modes: {D135, D45, H, V} * {-3,-2,-1,0,1,2,3} |
| 2815 |
unsigned int intra_luma_dir_delta_16x16_enable_1; // bit mask for enabling angular delta modes: {D67, D203, D157, D113} * {-3,-2,-1,0,1,2,3} |
| 2816 |
unsigned int intra_luma_dir_delta_16x8_enable_0; // for 8x16 and 16x8 |
| 2817 |
unsigned int intra_luma_dir_delta_16x8_enable_1; |
| 2818 |
|
| 2819 |
unsigned int intra_luma_dir_delta_8x8_enable_0; // bit mask for enabling angular delta modes: {D135, D45, H, V} * {-3,-2,-1,0,1,2,3} |
| 2820 |
unsigned int intra_luma_dir_delta_8x8_enable_1; // bit mask for enabling angular delta modes: {D67, D203, D157, D113} * {-3,-2,-1,0,1,2,3} |
| 2821 |
//Intra Luma angular delta mode controls: end |
| 2822 |
|
| 2823 |
// 0:DC_PRED, 1: V_PRED, 2: H_PRED, 3:D45_PRED, |
| 2824 |
// 4:D135_PRED, 5:D113_PRED, 6:D157_PRED, 7: D203_PRED, |
| 2825 |
// 8: D67_PRED, 9:SMOOTH_PRED, 10:SMOOTH_V_PRED, 11:SMOOTH_H_PRED |
| 2826 |
// 12: PAETH |
| 2827 |
unsigned short intra_luma_base_mode_4x4_enable; //bit mask of intra luma base 4x4 mode: |
| 2828 |
unsigned short intra_luma_base_mode_8x8_enable; //bit mask of intra luma base 8x8/8x4/4x8 mode |
| 2829 |
unsigned short intra_luma_base_mode_16x16_enable; //bit mask of intra luma base 16x16/16x8/8x16 modes |
| 2830 |
unsigned short intra_luma_base_mode_32x32_enable; //bit mask of intra luma base 32x32/32x16/16x32 modes |
| 2831 |
unsigned short intra_luma_base_mode_64x64_enable; //bit mask of intra luma base 64x64/32x64/64x32 modes |
| 2832 |
|
| 2833 |
short bias_intra_over_inter; // bias for any intra (subtracts from intra cost) |
| 2834 |
unsigned short mv_cost_bias; // bias applied to internal motion vector cost calculation |
| 2835 |
|
| 2836 |
short bias_tu_4x4; // bias for TU 4x4 |
| 2837 |
short bias_tu_8x8; // bias for TU 8x8 |
| 2838 |
short bias_tu_16x16; // bias for TU 16x16 |
| 2839 |
short bias_tu_32x32; // bias for TU 32x32 |
| 2840 |
short bias_tu_64x64; // bias for TU 64x64 |
| 2841 |
short bias_tu_8x4; // bias for TU 8x4 |
| 2842 |
short bias_tu_4x8; // bias for TU 4x8 |
| 2843 |
short bias_tu_16x8; // bias for TU 16x8 |
| 2844 |
short bias_tu_8x16; // bias for TU 8x16 |
| 2845 |
short bias_tu_32x16; // bias for TU 32x16 |
| 2846 |
short bias_tu_16x32; // bias for TU 16x32 |
| 2847 |
|
| 2848 |
short bias_skip; // bias for skip mv's ssd (subtracts from skip mv's ssd cost). Not supported. |
| 2849 |
short bias_inter_64x64; // bias for inter 64x64 (subtracts from inter 64x64 cost) |
| 2850 |
short bias_inter_64x32; // bias for inter 64x32 (subtracts from inter 64x32 cost) |
| 2851 |
short bias_inter_32x64; // bias for inter 32x64 (subtracts from inter 32x64 cost) |
| 2852 |
short bias_inter_32x32; // bias for inter 32x32 (subtracts from inter 32x32 cost) |
| 2853 |
short bias_inter_32x16; // bias for inter 32x16 (subtracts from inter 32x16 cost) |
| 2854 |
short bias_inter_16x32; // bias for inter 16x32 (subtracts from inter 16x32 cost) |
| 2855 |
short bias_inter_16x16; // bias for inter 16x16 (subtracts from inter 16x16 cost) |
| 2856 |
short bias_inter_16x8; // bias for inter 16x8 (subtracts from inter 16x8 cost) |
| 2857 |
short bias_inter_8x16; // bias for inter 8x16 (subtracts from inter 8x16 cost) |
| 2858 |
short bias_inter_8x8; // bias for inter 8x8 (subtracts from inter 8x8 cost) |
| 2859 |
short bias_intra_64x64; // bias for intra 64x64 (subtracts from intra 64x64 cost) |
| 2860 |
short bias_intra_64x32; // bias for intra 64x32 (subtracts from intra 64x32 cost) |
| 2861 |
short bias_intra_32x64; // bias for intra 32x64 (subtracts from intra 32x64 cost) |
| 2862 |
short bias_intra_32x32; // bias for intra 32x32 (subtracts from intra 32x32 cost) |
| 2863 |
short bias_intra_32x16; // bias for intra 32x16 (subtracts from intra 32x16 cost) |
| 2864 |
short bias_intra_16x32; // bias for intra 16x32 (subtracts from intra 16x32 cost) |
| 2865 |
short bias_intra_16x16; // bias for intra 16x16 (subtracts from intra 16x16 cost) |
| 2866 |
short bias_intra_16x8; // bias for intra 16x8 (subtracts from intra 16x8 cost) |
| 2867 |
short bias_intra_8x16; // bias for intra 8x16 (subtracts from intra 8x16 cost) |
| 2868 |
short bias_intra_8x8; // bias for intra 8x8 (subtracts from intra 8x8 cost) |
| 2869 |
short bias_intra_8x4; // bias for intra 8x4 (subtracts from intra 8x4 cost) |
| 2870 |
short bias_intra_4x8; // bias for intra 4x8 (subtracts from intra 4x8 cost) |
| 2871 |
short bias_intra_4x4; // bias for intra 4x4 (subtracts from intra 4x4 cost) |
| 2872 |
|
| 2873 |
unsigned short tx_type_search_mask[2]; // bit mask of tx type enabled. bit 0~15, corresponding to: |
| 2874 |
// tx type DCT_DCT, ADST_DCT, DCT_ADST, ADST_ADST, FLIPADST_DCT, |
| 2875 |
// DCT_FLIPADST, FLIPADST_FLIPADST, ADST_FLIPADST, FLIPADST_ADST, |
| 2876 |
// IDTX, V_DCT, H_DCT, V_ADST, H_ADST, V_FLIPADST, H_FLIPADST, |
| 2877 |
// 0: control for RDO stage. 1: control for RECON stage |
| 2878 |
|
| 2879 |
unsigned int zbin_factor_lt_148 : 8; //Aggresive DZ control for zbin factor when quant value < 148. in 1/128 unit |
| 2880 |
unsigned int zbin_factor_gt_148 : 8; //Aggresive DZ control for zbin factor when quant value >= 148. in 1/128 unit |
| 2881 |
unsigned int round_factor_eq_0 : 8; // TBD |
| 2882 |
unsigned int round_factor_ne_0 : 8; // TBD |
| 2883 |
} nvenc_av1_md_control_s; // 160 bytes |
| 2884 |
|
| 2885 |
typedef nvenc_h264_me_control_s nvenc_me_control_s; |
| 2886 |
|
| 2887 |
typedef struct |
| 2888 |
{ |
| 2889 |
// 256 aligned |
| 2890 |
unsigned int magic; // magic number, see text above // 4 bytes |
| 2891 |
nvenc_av1_surface_cfg_s refpic_cfg; // defines layout of reference pictures // 32 bytes |
| 2892 |
nvenc_av1_surface_cfg_s input_cfg; // defines layout of input pictures // 32 bytes |
| 2893 |
nvenc_av1_surface_cfg_s outputpic_cfg; // defines layout of reconstructed pictures // 32 bytes |
| 2894 |
|
| 2895 |
nvenc_av1_seq_hdr_data_s seqhdr_data; // 4 bytes |
| 2896 |
nvenc_av1_frame_hdr_data_s framehdr_data; // 64 bytes |
| 2897 |
nvenc_av1_rc_s rate_control; // Rate Control information // 96 bytes |
| 2898 |
nvenc_av1_pic_control_s pic_control; // 112 bytes |
| 2899 |
nvenc_av1_surface_cfg_s half_scaled_outputpic_cfg; // defines layout of 2x2 subsampled reconstructed picture // 32 bytes |
| 2900 |
nvenc_av1_md_control_s md_control; // 160 bytes |
| 2901 |
nvenc_me_control_s me_control; // 192 bytes |
| 2902 |
unsigned int gpTimer_timeout_val; // GPTimer cycle count set from driver |
| 2903 |
unsigned char reserved0[4]; |
| 2904 |
} nvenc_av1_drv_pic_setup_s; //768 bytes |
| 2905 |
|
| 2906 |
typedef nvenc_av1_drv_pic_setup_s nvenc_vp9_drv_pic_setup_s; //TODO: Orin VP9 will share the same class as AV1. Define new one for Hopper VP9 |
| 2907 |
|
| 2908 |
typedef struct |
| 2909 |
{ |
| 2910 |
unsigned int tile_top_sbx : 16; // top left sbx of the tile |
| 2911 |
unsigned int tile_top_sby : 16; // top left sby of the tile |
| 2912 |
unsigned int tile_bot_sbx : 16; // bottom right sbx of the tile |
| 2913 |
unsigned int tile_bot_sby : 16; // bottom right sby of the tile |
| 2914 |
|
| 2915 |
unsigned int qp_tile_min : 8; // min slice qp value |
| 2916 |
unsigned int qp_tile_max : 8; // max slice qp value |
| 2917 |
unsigned int qp_avr : 8; // 8bit value used by hw |
| 2918 |
unsigned int tile_type : 2; // 0 = B, 1 = P, 2 = I |
| 2919 |
unsigned int force_intra : 1; |
| 2920 |
unsigned int tile_end_update_cdf : 1; // enable/disable CDF update at end of tile |
| 2921 |
unsigned int limit_tile_top_boundary : 1; // limit ME to tile boundaries |
| 2922 |
unsigned int limit_tile_bot_boundary : 1; // limit ME to tile boundaries |
| 2923 |
unsigned int limit_tile_left_boundary : 1; // limit ME to tile boundaries |
| 2924 |
unsigned int limit_tile_right_boundary : 1; // limit ME to tile boundaries |
| 2925 |
|
| 2926 |
unsigned int tile_idx : 8; // tile index at picture level (potentially different from index in array of nvenc_av1_tile_control_s when doing strip encoding) |
| 2927 |
unsigned int disable_tile_top_filters : 1; //disable filters at tile top boundary |
| 2928 |
unsigned int disable_tile_bottom_filters : 1; //disable filters at tile bottom boundary |
| 2929 |
unsigned int disable_next_tile_top_filters : 1; //disable filters at next tile top boundary |
| 2930 |
unsigned int disable_next_tile_bottom_filters : 1; //disable filters at next tile bottom boundary |
| 2931 |
unsigned int reserved0 : 20; |
| 2932 |
|
| 2933 |
unsigned int next_tile_top_sbx : 16; // top left sbx of next tile |
| 2934 |
unsigned int next_tile_top_sby : 16; // top left sby of next tile |
| 2935 |
unsigned int next_tile_bot_sbx : 16; // bottom right sbx of next tile |
| 2936 |
unsigned int next_tile_bot_sby : 16; // bottom right sby of next tile |
| 2937 |
|
| 2938 |
unsigned short intra_refresh_cnt; // distance between forced-intra SBs in slice; 0 means OFF |
| 2939 |
unsigned short intra_refresh_offs; // first forced-intra SB in slice |
| 2940 |
|
| 2941 |
unsigned char reserved2[4]; // padding to 32 bytes |
| 2942 |
} nvenc_av1_tile_control_s; // 32 bytes |
| 2943 |
|
| 2944 |
// AV1 not support QPP, reserved it but not used |
| 2945 |
typedef struct |
| 2946 |
{ |
| 2947 |
// QPP has not been defined for AV1 |
| 2948 |
unsigned short qpp_run_vector_4x4; // cost values for 4x4 transform (16 bit total vector) |
| 2949 |
unsigned short qpp_run_vector_8x8[3]; // cost values for 8x8 transform (48 bit total vector) |
| 2950 |
// 2 bits each for first 12 coefs, then 1 bit each for next 24 |
| 2951 |
unsigned int qpp_luma8x8_cost : 8; // luma 8x8 cost threshold, 0 = throw out all coefs, range 0-15 |
| 2952 |
unsigned int qpp_luma16x16_cost : 8; // luma 16x16 cost threshold, 0 = throw out all coefs, range 0-15 |
| 2953 |
unsigned int qpp_chroma_cost : 8; // chroma cost threshold, 0 = throw out all coefs, range 0-15 |
| 2954 |
unsigned int qpp_mode : 2; // 0 = OFF, 1 = 8x8, 2 = 16x16_8x8 |
| 2955 |
unsigned int reserved0 : 6; // padding |
| 2956 |
|
| 2957 |
unsigned int reserved1; // padding to 16 bytes |
| 2958 |
} nvenc_av1_quant_control_s; // 16 bytes |
| 2959 |
|
| 2960 |
// Status buffer holds for debug related information |
| 2961 |
// |
| 2962 |
// General guideline for using CRC interface debug feature: |
| 2963 |
// 1. Generate the golden CRC (e.g. using cmodel) for for all interfaces |
| 2964 |
// 2. program the golden crc to NV_CNVENC_SAFETY_CRC_GOLDEN_PART_[X]_INTERFACE_NAME, and config the SAFETY_CRC_CFG register (with values from dbg_crc_init_enable/comp_enable/addr_comp/mem_wr_dis), set SAFETY_CRC_CTRL_EN = 1 (TZ) |
| 2965 |
// 3. kick off HW and trigger HW to compare width golden by writing SAFETY_CRC_CFG_COMP_EN when finish processing the frame; compare result could read from NV_CNVENC_SAFETY_CRC_STATUS_PART_[X] |
| 2966 |
// 4. raw HW CRC values could be read from NV_CNVENC_SAFETY_CRC_HW_PART_[X]_INTERFACE_NAME |
| 2967 |
typedef struct |
| 2968 |
{ |
| 2969 |
unsigned int dbg_crc_enable : 1; //Global enable flag for enable/disable interface crc calculation in NVENC HW |
| 2970 |
unsigned int dbg_crc_init_enable: 1; //Enable crc initialization |
| 2971 |
unsigned int dbg_crc_comp_enable: 1; //Enable crc golden comparision |
| 2972 |
unsigned int dbg_crc_addr_comp : 1; //Enable crc calculation for fbif address |
| 2973 |
unsigned int dbg_crc_mem_wr_dis : 1; //Disable fb memory write |
| 2974 |
unsigned int reserved0 : 27; |
| 2975 |
|
| 2976 |
//Golden CRC values for each partition |
| 2977 |
unsigned int dbg_crc_partb_golden[32]; |
| 2978 |
unsigned int dbg_crc_partf_golden[32]; |
| 2979 |
unsigned int dbg_crc_partg_golden[32]; |
| 2980 |
unsigned int dbg_crc_parth_golden[32]; |
| 2981 |
unsigned int dbg_crc_parti_golden[32]; |
| 2982 |
unsigned int dbg_crc_partj_golden[32]; |
| 2983 |
unsigned int dbg_crc_partk_golden[32]; |
| 2984 |
unsigned int dbg_crc_partl_golden[35]; |
| 2985 |
|
| 2986 |
//Compare status register for each partition. |
| 2987 |
//each bit corresponding to the compare status for one interface (0: pass; 1: fail) |
| 2988 |
//The number of valid bits for each partition shall equal to the number of golden crcs for that partition |
| 2989 |
unsigned int dbg_crc_comp_partb; |
| 2990 |
unsigned int dbg_crc_comp_partf; |
| 2991 |
unsigned int dbg_crc_comp_partg; |
| 2992 |
unsigned int dbg_crc_comp_parth; |
| 2993 |
unsigned int dbg_crc_comp_parti; |
| 2994 |
unsigned int dbg_crc_comp_partj; |
| 2995 |
unsigned int dbg_crc_comp_partk; |
| 2996 |
unsigned int dbg_crc_comp_partl[2]; //split into two compare status registers |
| 2997 |
|
| 2998 |
|
| 2999 |
//HW CRC values generated from each partition. the number shall match with golden CRC number |
| 3000 |
unsigned int dbg_crc_partb_crc[32]; |
| 3001 |
unsigned int dbg_crc_partf_crc[32]; |
| 3002 |
unsigned int dbg_crc_partg_crc[32]; |
| 3003 |
unsigned int dbg_crc_parth_crc[32]; |
| 3004 |
unsigned int dbg_crc_parti_crc[32]; |
| 3005 |
unsigned int dbg_crc_partj_crc[32]; |
| 3006 |
unsigned int dbg_crc_partk_crc[32]; |
| 3007 |
unsigned int dbg_crc_partl_crc[35]; |
| 3008 |
|
| 3009 |
|
| 3010 |
//HW verification purpose control fields |
| 3011 |
unsigned int lambda_ssd; //lambda for SSD cost |
| 3012 |
unsigned int lambda_satd; //lambda for SATD cost |
| 3013 |
unsigned int lambda_ld; //lambda for laplacian cost |
| 3014 |
unsigned char reserved1[180]; |
| 3015 |
} nvenc_debug_status_s; //2304 bytes |
| 3016 |
|
| 3017 |
// 2.7. Bit Stream Output Data Buffer |
| 3018 |
// ---------------------------------- |
| 3019 |
// |
| 3020 |
// This buffer is used to store the encoded bit stream data at slice header |
| 3021 |
// level and below. |
| 3022 |
// The driver must ensure that there is enough free space in this buffer to |
| 3023 |
// contain the entire output produced for the current execute, including any |
| 3024 |
// headers that will be produced. Output will be written incrementally |
| 3025 |
// from the starting position relative to the 256 byte aligned start location |
| 3026 |
// specified for this buffer. |
| 3027 |
// |
| 3028 |
// 2.8. Motion Estimation Predictor Data Buffer |
| 3029 |
// -------------------------------------------- |
| 3030 |
// |
| 3031 |
// This buffer is used to store by the motion estimation unit to store vector |
| 3032 |
// hints that can be used as explicit candidates or as guides for motion |
| 3033 |
// search in future pictures. |
| 3034 |
// During encode one such surface can be written and another can be read. |
| 3035 |
// |
| 3036 |
// 2.9. Rate Control Process Buffer |
| 3037 |
// -------------------------------- |
| 3038 |
// |
| 3039 |
// This buffer is used to store persistent rate control information. The same |
| 3040 |
// buffer will be shared in all executes of a particular stream. |
| 3041 |
// |
| 3042 |
// This buffer is used only if picture level rate control is enabled ((rc_mode & 3) != 0) |
| 3043 |
|
| 3044 |
// Macro to initialize Bavg in VBR mode |
| 3045 |
#define NVENC_VBR_INIT_BAVG(framerate, ravg) (0) // zero-centered vbr fullness |
| 3046 |
// Macro to initialize Whist and Wpbi |
| 3047 |
#define NVENC_RC_INIT_COMPLEXITY(bpp_at_qs1) ((int)((bpp_at_qs1)*16)) // use this macro to decouple from implementation-dependent precision |
| 3048 |
|
| 3049 |
typedef struct |
| 3050 |
{ |
| 3051 |
|
| 3052 |
int nal_cpb_fullness; // 28.4 fixed point |
| 3053 |
int vcl_cpb_fullness; // 28.4 fixed point |
| 3054 |
|
| 3055 |
int Wpbi[3]; // 27.4 signed fixed point |
| 3056 |
int Whist[3][3]; // 27.4 signed fixed point [pbi][t] |
| 3057 |
int np; // 27.4 signed fixed point |
| 3058 |
int nb; // 27.4 signed fixed point |
| 3059 |
int t; // 27.4 signed fixed point |
| 3060 |
|
| 3061 |
int Rmean; // 23.8 signed fixed point peak bitrate (bits/pixel) |
| 3062 |
int Bavg; // 23.8 signed fixed point |
| 3063 |
|
| 3064 |
short average_mvx; // average mvx info, can be used for ME hint |
| 3065 |
short average_mvy; // average mvy info, can be used for ME hint |
| 3066 |
|
| 3067 |
int prevQS; // previous second pass Qs derived from rate |
| 3068 |
short prevQP; // previous 2nd pass qp |
| 3069 |
short fwdQP; // QP of forward reference |
| 3070 |
|
| 3071 |
unsigned char refQPcol[48]; // co-located mbrow qp of the most recently coded reference |
| 3072 |
unsigned char Rcol[48]; // co-located rate (5.3 precision, relative to avg rate) |
| 3073 |
unsigned char ref1QP[48]; // first-pass reference qp (when 1st pass is using its own reference frames) |
| 3074 |
|
| 3075 |
unsigned int ext_intra_satd_Hist[2]; // external satd History |
| 3076 |
unsigned char ext_scd_Hist; // scene change flag History |
| 3077 |
unsigned char scene_change; // last coded frame was a scene change |
| 3078 |
unsigned char qpp_modifier; // current qpp adjustment (subtracted from qpp thresholds) |
| 3079 |
unsigned char Ksatd; // pass1 satd to pass2 complexity measure |
| 3080 |
signed char Werr; // model deviation: Wpass2 / Wpass1 - 1.0, .8 signed fixed-point [0.5x..1.5x] |
| 3081 |
signed char WerrI; // same as above for I-frames |
| 3082 |
unsigned char avgBf; // 3.5 fixed point, average number of consecutive B-frames |
| 3083 |
unsigned char reserved; |
| 3084 |
unsigned int printf_offset; // Using for ucode print |
| 3085 |
unsigned char reserved2[4]; |
| 3086 |
|
| 3087 |
} nvenc_persistent_state_s; // 256Bytes |
| 3088 |
|
| 3089 |
// 2.a. Ucode Probability State |
| 3090 |
// -------------------------------- |
| 3091 |
// |
| 3092 |
// This buffer is used to store probability state computed and used during the last picture. |
| 3093 |
// This is used to compute the probability updates for current picture |
| 3094 |
// |
| 3095 |
|
| 3096 |
// DO NOT rearrange this structure. The alignment is crucial for dmreads |
| 3097 |
typedef struct |
| 3098 |
{ |
| 3099 |
unsigned char coef_probs[4][8][3][11+5]; // 1536 bytes = 4*(3*128) |
| 3100 |
// Padding cause this has to be read in parts |
| 3101 |
unsigned char ymode_prob[4]; // 4 bytes |
| 3102 |
unsigned char uvmode_prob[3]; // 3 bytes |
| 3103 |
unsigned char reserved0; // Pad to word boundary |
| 3104 |
unsigned char mv_context[2][19]; // 38 bytes |
| 3105 |
unsigned char reserved1[18]; // Align to 32 byte boundary |
| 3106 |
|
| 3107 |
}nvenc_vp8_prob_state_s; |
| 3108 |
|
| 3109 |
|
| 3110 |
// 2.b. Rate Control stat buffer used for 2 pass encoding |
| 3111 |
// ------------------------------------------------------ |
| 3112 |
// |
| 3113 |
// This buffer is used to store rate control information for 2 pass encoding. |
| 3114 |
// The same buffer will be shared in both the passes of the encoding. |
| 3115 |
// The first pass will write MPEC stats to the buffer and 2nd pass will read |
| 3116 |
// the stats from the same buffer for QP calculation. The layout of buffer is |
| 3117 |
// msenc_rc_pic_stat_s followed by msenc_mbrow_stat_s for each mb row. |
| 3118 |
// |
| 3119 |
|
| 3120 |
typedef struct |
| 3121 |
{ |
| 3122 |
unsigned int intra_mb_count; // number of intra MBs. |
| 3123 |
unsigned int inter_mb_count; // number of inter MBs. |
| 3124 |
unsigned int cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 3125 |
unsigned int cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 3126 |
unsigned int total_bit_count; |
| 3127 |
unsigned int avgQP; |
| 3128 |
unsigned int complexity; |
| 3129 |
//split encoding |
| 3130 |
unsigned int strip_complexity; |
| 3131 |
unsigned int strip_intra_mb_count; |
| 3132 |
unsigned char reserved[220]; |
| 3133 |
}nvenc_rc_pic_stat_s; //256 bytes |
| 3134 |
|
| 3135 |
typedef struct |
| 3136 |
{ |
| 3137 |
unsigned int mbrow_intra_mb_count : 12; // number of intra MBs |
| 3138 |
unsigned int mbrow_inter_mb_count : 12; // number of inter MBs |
| 3139 |
unsigned int mbrow_avgQP : 8; // average QP |
| 3140 |
unsigned int mbrow_cumulative_intra_cost; // sum of all intra SA(T)Ds |
| 3141 |
unsigned int mbrow_cumulative_inter_cost; // sum of all inter SA(T)Ds |
| 3142 |
unsigned int mbrow_bits_per_row; |
| 3143 |
}nvenc_mbrow_stat_s; |
| 3144 |
|
| 3145 |
typedef struct |
| 3146 |
{ |
| 3147 |
signed char act_qp_buf[4096]; // Buffer to store qp modifier based on avg activity for a MB row |
| 3148 |
}nvenc_act_stat_s; |
| 3149 |
|
| 3150 |
// 2.b. Generic Map Buffer |
| 3151 |
// -------------------------------- |
| 3152 |
// |
| 3153 |
// This buffer is used to store qp map when rcmode==4 & lambda_map when enabled. the structure is in run/level, run==0 means end. |
| 3154 |
// |
| 3155 |
typedef struct |
| 3156 |
{ |
| 3157 |
unsigned short run; // number of blocks with the same value |
| 3158 |
unsigned short level; // value |
| 3159 |
} nvenc_generic_map_s; |
| 3160 |
|
| 3161 |
typedef nvenc_generic_map_s nvenc_qp_map_s; |
| 3162 |
typedef nvenc_generic_map_s nvenc_lambda_map_s; |
| 3163 |
|
| 3164 |
|
| 3165 |
// 2.d. Encrypt Params buffer |
| 3166 |
// -------------------------------- |
| 3167 |
// |
| 3168 |
// This buffer is used by SEC / TSEC engine which will encrypt the encoded stream |
| 3169 |
// MSENC will output the # of 16 byte blocks present in the encoded bitstream to noOfInputBlocks field. |
| 3170 |
// MSENC ucode does not use any other field besides noOfInputBlocks. |
| 3171 |
// These will be populated by driver / TSEC ucode |
| 3172 |
// Currently used only for MSENC 2.0 for T114 |
| 3173 |
typedef struct _hdcp_encrypt_param_s |
| 3174 |
{ |
| 3175 |
unsigned int sessionId; // <<in |
| 3176 |
unsigned int noOfInputBlocks; // <<in |
| 3177 |
unsigned int streamId; // <<in |
| 3178 |
unsigned int encOffset; // <<in |
| 3179 |
unsigned int streamCtr; // >>out |
| 3180 |
unsigned int crcOffset; // >>out |
| 3181 |
unsigned long long inputCtr; // >>out |
| 3182 |
unsigned long long pesPriv[2]; // >>out |
| 3183 |
unsigned int retCode; // >>out |
| 3184 |
unsigned char reserved2[4]; |
| 3185 |
} hdcp_encrypt_param_s; |
| 3186 |
|
| 3187 |
// Error codes used across all codecs |
| 3188 |
typedef enum |
| 3189 |
{ |
| 3190 |
E_NONE = 0U, |
| 3191 |
E_GPTMR_TIMEOUT, |
| 3192 |
E_MAGIC_NUMBER_MISMATCH, |
| 3193 |
E_PICSETUP_OFFSET, |
| 3194 |
E_ENC_STATUS_OUTPUT_OFFSET, |
| 3195 |
E_OUTPUT_BITSTREAM_OFFSET, |
| 3196 |
E_INSUFFICIENT_WRDMA_BUF, |
| 3197 |
E_MDP_ERROR, |
| 3198 |
E_RDO_ERROR, |
| 3199 |
E_ME_PROGRAMMING, |
| 3200 |
E_ME_HW, |
| 3201 |
E_TFBIF_TIMEOUT, |
| 3202 |
E_SKIP_ENCODING, |
| 3203 |
E_INPUT_BITSTREAM_OFFSET, |
| 3204 |
E_INPUT_HISTORY_OFFSET, |
| 3205 |
E_MEPRED_OFFSET, |
| 3206 |
E_EXTHINT_OFFSET, |
| 3207 |
E_REFPIC_OFFSET, |
| 3208 |
E_QPMAP_OFFSET, |
| 3209 |
E_COLLOC_OFFSET, |
| 3210 |
E_PROB_OFFSET, |
| 3211 |
E_ENTROPY_OFFSET, |
| 3212 |
E_FILTER_OFFSET, |
| 3213 |
E_SSIM_RDO_OFFSET, |
| 3214 |
E_RC_ERROR, |
| 3215 |
E_SMALLE_BS_OVERFLOW, |
| 3216 |
E_SMALLE_METHODSERROR, |
| 3217 |
E_ENTROPY_ERR, |
| 3218 |
E_RECON_OVERFLOW, |
| 3219 |
TOTAL_ERR_CODES |
| 3220 |
} ERROR_CODES; |
| 3221 |
|
| 3222 |
// 3 - APPLICATION SPECIFIC METHOD REGISTERS |
| 3223 |
// |
| 3224 |
// The NVENC application method registers are accessed directly by the driver. |
| 3225 |
// The range assignment is 0x400 - 0x7FC: |
| 3226 |
// |
| 3227 |
// |
| 3228 |
// 3.1 - Common Methods |
| 3229 |
// -------------------- |
| 3230 |
// |
| 3231 |
// SetControlParams() |
| 3232 |
// ------------------ |
| 3233 |
// This method sets the control parameters that are used in this execution |
| 3234 |
// of the encode subclass. |
| 3235 |
// |
| 3236 |
// CODEC_TYPE field should be set as |
| 3237 |
// 3: H.264 // same codec enum as for msdec |
| 3238 |
// |
| 3239 |
// FORCE_OUTPIC indicates that the encoded picture (after deblocking if that is enabled) |
| 3240 |
// must be written out to the buffer specified by SetOutRefPic(), even if the |
| 3241 |
// current picture is not marked as a reference picture. |
| 3242 |
// |
| 3243 |
// FORCE_OUTCOL indicates that the colocated vector data must be written out to the |
| 3244 |
// buffer specified by SetOutColocData(), even if the current picture is |
| 3245 |
// not marked as a reference picture. |
| 3246 |
// |
| 3247 |
// GPTIMER_ON, this flag is used in activating an app based timer with a timeout value |
| 3248 |
// less than watchdog timer. When this app level timer expires, various hardware |
| 3249 |
// submodules in NVENC will be reset, the status will be informed in flowControl |
| 3250 |
// buffer if DEBUG_MODE is OFF (0) and app will return. |
| 3251 |
// |
| 3252 |
// ME_ONLY, indicates whether to run in normal mode of operation or activate only |
| 3253 |
// motion estimation and write out search vector results |
| 3254 |
// 0 = encode bitstream |
| 3255 |
// 1 = only ME, do not run mode decision/recon/entropy encode |
| 3256 |
// |
| 3257 |
// SLICE_STAT_ON indicates if slice level statistics dumping is enabled. If enabled, |
| 3258 |
// following the persistent state information, slice level statistics will be |
| 3259 |
// output in the Encoder Status buffer. It is assumed that the number of slices |
| 3260 |
// will be limited to picture height in units of MBs. If there are more slices in the |
| 3261 |
// picture, driver should turn off slice stat output, by resetting this flag to 0. |
| 3262 |
// |
| 3263 |
// MPEC_STAT_ON indicates if nMB level mpec statistics dumping is enabled. If SLICE_STAT_ON |
| 3264 |
// is enabled, MPEC_STAT_ON will be automatically get disabled and all the slice level |
| 3265 |
// mpec statistics will be dumped to nvenc_mpec_stat_s. |
| 3266 |
// |
| 3267 |
// DEBUG_MODE indicates whether the app will return non-zero error codes to OS in case |
| 3268 |
// of an error detected by the app. |
| 3269 |
// |
| 3270 |
// If the value is set to 1, upon detecting an error the app will NOT do a |
| 3271 |
// soft reset of hw, the app will return appropriate error code to OS. This |
| 3272 |
// shall be used during debugging phase. |
| 3273 |
// |
| 3274 |
// If the value is set to 0, upon detecting an error the app will do a soft |
| 3275 |
// reset of hw and return with zero to OS. This shall be used for product. |
| 3276 |
// |
| 3277 |
// SUBFRAME_MODE used in low latency applications where client need to read the bitstream |
| 3278 |
// data at slice boundary before frame done. |
| 3279 |
// 0: disable |
| 3280 |
// 1: flush at slice end and report last_valid_byte_offset |
| 3281 |
// 2: report slice offsets without flushing |
| 3282 |
// 3: report slice offsets with flush at slice end |
| 3283 |
// |
| 3284 |
// RC_MODE contains the rate control mode. |
| 3285 |
// 0: Constant QP |
| 3286 |
// 1: picture level VBR |
| 3287 |
// 2: picture level CBR |
| 3288 |
// 3: picture level VBR with const QP |
| 3289 |
// 4: QP Map |
| 3290 |
// 16: (same as 0) |
| 3291 |
// 17: mbrow level VBR |
| 3292 |
// 18: mbrow level CBR (QP is updated every macroblock row) |
| 3293 |
// 19: mbrow level VBR with const QP |
| 3294 |
// 32: macroblock level activity based rate control algorithm |
| 3295 |
// |
| 3296 |
// |
| 3297 |
// 31 24 20 16 15 12 8 4 0 |
| 3298 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3299 |
// | |D| |G|S|M|C|P| | CODEC | SetControlParams |
| 3300 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3301 |
// |
| 3302 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM 0x00000700 /* -W-4R */ |
| 3303 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_CODEC_TYPE 3:0 /* -W-UF */ |
| 3304 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_CODEC_TYPE_H264 0x00000003 |
| 3305 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_FORCE_OUTPIC 8:8 /* -W-UF */ |
| 3306 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_FORCE_OUTCOL 9:9 /* -W-UF */ |
| 3307 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_ME_ONLY 10:10 /* -W-UF */ |
| 3308 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_SLICE_STAT_ON 11:11 /* -W-UF */ |
| 3309 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_GPTIMER_ON 12:12 /* -W-UF */ |
| 3310 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_MPEC_STAT_ON 15:15 /* -W-UF */ |
| 3311 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_DEBUG_MODE 16:16 /* -W-UF */ |
| 3312 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_SUBFRAME_MODE 18:17 /* -W-UF */ |
| 3313 |
// #define NV_C9B7_NVENC_SET_CTRL_PARAM_RC_MODE 31:24 /* -W-UF */ |
| 3314 |
// |
| 3315 |
// |
| 3316 |
// SetPictureIndex() |
| 3317 |
// ----------------- |
| 3318 |
// |
| 3319 |
// This method sets the picture index to NVENC. For the same picture, CEA and NVENC |
| 3320 |
// should receive the same picture index. Picture index should be incremented by 1 |
| 3321 |
// in each new execute. CEA will put this in the CEA Hints data buffer. |
| 3322 |
// |
| 3323 |
// 31 24 23 16 15 8 7 0 |
| 3324 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3325 |
// | INDEX | SetPictureIndex |
| 3326 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3327 |
// |
| 3328 |
// #define NV_C9B7_NVENC_SET_PICTURE_INDEX 0x00000704 /* -W-4R */ |
| 3329 |
// #define NV_C9B7_NVENC_SET_PICTURE_INDEX_INDEX 31:0 /* -W-UF */ |
| 3330 |
// |
| 3331 |
// |
| 3332 |
// SetInRcData() |
| 3333 |
// ------------------ |
| 3334 |
// This method sets the location of the RCDataBuffer. It sets the |
| 3335 |
// upper bits (32 most significant bits) of the 40-bit offset where the buffer |
| 3336 |
// is located. This buffer will have RC configuration data from CEA or an app. |
| 3337 |
// |
| 3338 |
// The values of OFFSET is defined and initialized to zero during the |
| 3339 |
// instance's creation. |
| 3340 |
// |
| 3341 |
// 31 16 0 |
| 3342 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3343 |
// | OFFSET | SetInDrvPicSetup |
| 3344 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3345 |
// |
| 3346 |
// #define NV_C9B7_NVENC_SETIN_RC_DATA 0x0000070c /* -W-4R */ |
| 3347 |
// #define NV_C9B7_NVENC_SETIN_RC_DATA_OFFSET 31:0 /* -W-UF */ |
| 3348 |
// |
| 3349 |
// SetInDrvPicSetup() |
| 3350 |
// ------------------ |
| 3351 |
// This method sets the location of the DrvPictureSetupBuffer. It sets the |
| 3352 |
// upper bits (32 most significant bits) of the 40-bit offset where the buffer |
| 3353 |
// is located. |
| 3354 |
// |
| 3355 |
// The values of OFFSET is defined and initialized to zero during the |
| 3356 |
// instance's creation. |
| 3357 |
// |
| 3358 |
// 31 16 0 |
| 3359 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3360 |
// | OFFSET | SetInDrvPicSetup |
| 3361 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3362 |
// |
| 3363 |
// #define NV_C9B7_NVENC_SETIN_DRV_PIC_SETUP 0x00000710 /* -W-4R */ |
| 3364 |
// #define NV_C9B7_NVENC_SETIN_DRV_PIC_SETUP_OFFSET 31:0 /* -W-UF */ |
| 3365 |
// |
| 3366 |
// |
| 3367 |
// SetInCEAHintsData() |
| 3368 |
// ------------------- |
| 3369 |
// This method sets the CEA hints data buffer offset. It sets the upper bits (32 |
| 3370 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3371 |
// |
| 3372 |
// |
| 3373 |
// 31 16 0 |
| 3374 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3375 |
// | OFFSET | SetInCEAHintsData |
| 3376 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3377 |
// |
| 3378 |
// #define NV_C9B7_NVENC_SETIN_CEA_HINTS_DATA 0x00000714 /* -W-4R */ |
| 3379 |
// #define NV_C9B7_NVENC_SETIN_CEA_HINTS_DATA_OFFSET 31:0 /* -W-UF */ |
| 3380 |
// |
| 3381 |
// |
| 3382 |
// SetOutEncStatus() |
| 3383 |
// ----------------- |
| 3384 |
// This method sets the encoder status buffer offset. It sets the upper bits |
| 3385 |
// (32 most significant bits) of the 40-bit offset where the buffer is located. |
| 3386 |
// |
| 3387 |
// 31 16 0 |
| 3388 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3389 |
// | OFFSET | SetOutEncStatus |
| 3390 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3391 |
// |
| 3392 |
// #define NV_C9B7_NVENC_SETOUT_ENC_STATUS 0x00000718 /* -W-4R */ |
| 3393 |
// #define NV_C9B7_NVENC_SETOUT_ENC_STATUS_OFFSET 31:0 /* -W-UF */ |
| 3394 |
// |
| 3395 |
// |
| 3396 |
// SetOutBitstream() |
| 3397 |
// ----------------- |
| 3398 |
// This method sets the bitstream output data buffer offset. It sets the upper |
| 3399 |
// bits (32 most significant bits) of the 40-bit offset where the buffer is located. |
| 3400 |
// |
| 3401 |
// 31 16 0 |
| 3402 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3403 |
// | OFFSET | SetOutBitstream |
| 3404 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3405 |
// |
| 3406 |
// #define NV_C9B7_NVENC_SETOUT_BITSTREAM 0x0000071c /* -W-4R */ |
| 3407 |
// #define NV_C9B7_NVENC_SETOUT_BITSTREAM_OFFSET 31:0 /* -W-UF */ |
| 3408 |
// |
| 3409 |
// |
| 3410 |
// SetIOHistory() |
| 3411 |
// -------------- |
| 3412 |
// This method sets the history buffer offset. It sets the upper bits (32 |
| 3413 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3414 |
// |
| 3415 |
// |
| 3416 |
// 31 16 0 |
| 3417 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3418 |
// | OFFSET | SetIOHistory |
| 3419 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3420 |
// |
| 3421 |
// #define NV_C9B7_NVENC_SETIO_HISTORY 0x00000720 /* -W-4R */ |
| 3422 |
// #define NV_C9B7_NVENC_SETIO_HISTORY_OFFSET 31:0 /* -W-UF */ |
| 3423 |
// |
| 3424 |
// |
| 3425 |
// SetIORCProcess() |
| 3426 |
// ---------------- |
| 3427 |
// This method sets the RC Process buffer offset. It sets the upper bits |
| 3428 |
// (32 most significant bits) of the 40-bit offset where the buffer is located. |
| 3429 |
// |
| 3430 |
// 31 16 0 |
| 3431 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3432 |
// | OFFSET | SetIORCProcess |
| 3433 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3434 |
// |
| 3435 |
// #define NV_C9B7_NVENC_SETIO_RC_PROCESS 0x00000724 /* -W-4R */ |
| 3436 |
// #define NV_C9B7_NVENC_SETIO_PROCESS_OFFSET 31:0 /* -W-UF */ |
| 3437 |
// |
| 3438 |
// |
| 3439 |
// SetInColocData() |
| 3440 |
// ---------------- |
| 3441 |
// This method sets the input colocated motion data buffer offset. It sets the upper |
| 3442 |
// bits (32 most significant bits) of the 40-bit offset where the buffer is located. |
| 3443 |
// This is used only for encoding of B pictures. If the current picture is an I or P |
| 3444 |
// picture, this method can be omitted. |
| 3445 |
// |
| 3446 |
// |
| 3447 |
// 31 16 0 |
| 3448 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3449 |
// | OFFSET | SetInColocData |
| 3450 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3451 |
// |
| 3452 |
// #define NV_C9B7_NVENC_SETIN_COLOC_DATA 0x00000728 /* -W-4R */ |
| 3453 |
// #define NV_C9B7_NVENC_SETIN_COLOC_DATA_OFFSET 31:0 /* -W-UF */ |
| 3454 |
// |
| 3455 |
// |
| 3456 |
// SetOutColocData() |
| 3457 |
// ----------------- |
| 3458 |
// This method sets the output colocated motion data buffer offset. It sets the upper |
| 3459 |
// bits (32 most significant bits) of the 40-bit offset where the buffer is located. |
| 3460 |
// This is used only for encoding of pictures that are marked as reference pictures. |
| 3461 |
// If the current picture is not marked as a reference picture, this method can be omitted. |
| 3462 |
// |
| 3463 |
// |
| 3464 |
// 31 16 0 |
| 3465 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3466 |
// | OFFSET | SetOutColocData |
| 3467 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3468 |
// |
| 3469 |
// #define NV_C9B7_NVENC_SETOUT_COLOC_DATA 0x0000072c /* -W-4R */ |
| 3470 |
// #define NV_C9B7_NVENC_SETOUT_COLOC_DATA_OFFSET 31:0 /* -W-UF */ |
| 3471 |
// |
| 3472 |
// |
| 3473 |
// SetOutRefPic() |
| 3474 |
// -------------- |
| 3475 |
// This method sets the offset for the tiled 16x16 output reference picture produced by |
| 3476 |
// the encode of the current picture. It sets the upper bits (32 most significant bits) |
| 3477 |
// of the 40-bit offset where the buffer is located. This method must be sent if the |
| 3478 |
// current picture is marked as a reference picture, or the FORCE_OUTPIC bit is set in |
| 3479 |
// the SetControlParams(). |
| 3480 |
// |
| 3481 |
// 31 16 0 |
| 3482 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3483 |
// | OFFSET | SetOutRefPic |
| 3484 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3485 |
// |
| 3486 |
// #define NV_C9B7_NVENC_SETOUT_REFPIC 0x00000730 /* -W-4R */ |
| 3487 |
// #define NV_C9B7_NVENC_SETOUT_REFPIC_OFFSET 31:0 /* -W-UF */ |
| 3488 |
// |
| 3489 |
// |
| 3490 |
// SetInCurPic() |
| 3491 |
// ------------- |
| 3492 |
// This method sets the offset for current input picture to be encoded. It sets the upper |
| 3493 |
// bits (32 most significant bits) of the 40-bit offset where the buffer is located. |
| 3494 |
// |
| 3495 |
// 31 16 0 |
| 3496 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3497 |
// | OFFSET | SetInCurPic |
| 3498 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3499 |
// |
| 3500 |
// #define NV_C9B7_NVENC_SETIN_CURPIC 0x00000734 /* -W-4R */ |
| 3501 |
// #define NV_C9B7_NVENC_SETIN_CURPIC_OFFSET 31:0 /* -W-UF */ |
| 3502 |
// |
| 3503 |
// |
| 3504 |
// SetInMEPredData() |
| 3505 |
// ----------------- |
| 3506 |
// This method sets the offset for motion estimation predictor input data buffer. |
| 3507 |
// This is the buffer containing vector hints produced by an earlier execute of NVENC. |
| 3508 |
// The method sets the upper bits (32 most significant bits) of the 40-bit offset where |
| 3509 |
// the buffer is located. |
| 3510 |
// |
| 3511 |
// 31 16 0 |
| 3512 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3513 |
// | OFFSET | SetInMEPredData |
| 3514 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3515 |
// |
| 3516 |
// #define NV_C9B7_NVENC_SETIN_MEPRED_DATA 0x00000738 /* -W-4R */ |
| 3517 |
// #define NV_C9B7_NVENC_SETIN_MEPRED_DATA_OFFSET 31:0 /* -W-UF */ |
| 3518 |
// |
| 3519 |
// |
| 3520 |
// SetOutMEPredData() |
| 3521 |
// ------------------ |
| 3522 |
// This method sets the offset for motion estimation predictor output data Buffer. |
| 3523 |
// It sets the upper bits (32 most significant bits) of the 40-bit offset where the |
| 3524 |
// buffer is located. |
| 3525 |
// |
| 3526 |
// 31 16 0 |
| 3527 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3528 |
// | OFFSET | SetOutMEPredData |
| 3529 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3530 |
// |
| 3531 |
// #define NV_C9B7_NVENC_SETOUT_MEPRED_DATA 0x0000073c /* -W-4R */ |
| 3532 |
// #define NV_C9B7_NVENC_SETOUT_MEPRED_DATA_OFFSET 31:0 /* -W-UF */ |
| 3533 |
// |
| 3534 |
// |
| 3535 |
// SetInRefPic0() |
| 3536 |
// -------------- |
| 3537 |
// This method sets the offset for reference picture 0. It sets the upper bits (32 |
| 3538 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3539 |
// |
| 3540 |
// 31 16 0 |
| 3541 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3542 |
// | OFFSET | SetInRefPic0 |
| 3543 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3544 |
// |
| 3545 |
// #define NV_C9B7_NVENC_SETIN_REFPIC0 0x00000400 /* -W-4R */ |
| 3546 |
// #define NV_C9B7_NVENC_SETIN_REFPIC0_OFFSET 31:0 /* -W-UF */ |
| 3547 |
// |
| 3548 |
// |
| 3549 |
// SetInRefPic1() |
| 3550 |
// -------------- |
| 3551 |
// This method sets the offset for reference picture 1. It sets the upper bits (32 |
| 3552 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3553 |
// |
| 3554 |
// 31 16 0 |
| 3555 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3556 |
// | OFFSET | SetInRefPic1 |
| 3557 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3558 |
// |
| 3559 |
// #define NV_C9B7_NVENC_SETIN_REFPIC1 0x00000404 /* -W-4R */ |
| 3560 |
// #define NV_C9B7_NVENC_SETIN_REFPIC1_OFFSET 31:0 /* -W-UF */ |
| 3561 |
// |
| 3562 |
// |
| 3563 |
// SetInRefPic2() |
| 3564 |
// -------------- |
| 3565 |
// This method sets the offset for reference picture 2. It sets the upper bits (32 |
| 3566 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3567 |
// |
| 3568 |
// 31 16 0 |
| 3569 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3570 |
// | OFFSET | SetInRefPic2 |
| 3571 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3572 |
// |
| 3573 |
// #define NV_C9B7_NVENC_SETIN_REFPIC2 0x00000408 /* -W-4R */ |
| 3574 |
// #define NV_C9B7_NVENC_SETIN_REFPIC2_OFFSET 31:0 /* -W-UF */ |
| 3575 |
// |
| 3576 |
// |
| 3577 |
// SetInRefPic3() |
| 3578 |
// -------------- |
| 3579 |
// This method sets the offset for reference picture 3. It sets the upper bits (32 |
| 3580 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3581 |
// |
| 3582 |
// 31 16 0 |
| 3583 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3584 |
// | OFFSET | SetInRefPic3 |
| 3585 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3586 |
// |
| 3587 |
// #define NV_C9B7_NVENC_SETIN_REFPIC3 0x0000040c /* -W-4R */ |
| 3588 |
// #define NV_C9B7_NVENC_SETIN_REFPIC3_OFFSET 31:0 /* -W-UF */ |
| 3589 |
// |
| 3590 |
// |
| 3591 |
// SetInRefPic4() |
| 3592 |
// -------------- |
| 3593 |
// This method sets the offset for reference picture 4. It sets the upper bits (32 |
| 3594 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3595 |
// |
| 3596 |
// 31 16 0 |
| 3597 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3598 |
// | OFFSET | SetInRefPic4 |
| 3599 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3600 |
// |
| 3601 |
// #define NV_C9B7_NVENC_SETIN_REFPIC4 0x00000410 /* -W-4R */ |
| 3602 |
// #define NV_C9B7_NVENC_SETIN_REFPIC4_OFFSET 31:0 /* -W-UF */ |
| 3603 |
// |
| 3604 |
// |
| 3605 |
// SetInRefPic5() |
| 3606 |
// -------------- |
| 3607 |
// This method sets the offset for reference picture 5. It sets the upper bits (32 |
| 3608 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3609 |
// |
| 3610 |
// 31 16 0 |
| 3611 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3612 |
// | OFFSET | SetInRefPic5 |
| 3613 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3614 |
// |
| 3615 |
// #define NV_C9B7_NVENC_SETIN_REFPIC5 0x00000414 /* -W-4R */ |
| 3616 |
// #define NV_C9B7_NVENC_SETIN_REFPIC5_OFFSET 31:0 /* -W-UF */ |
| 3617 |
// |
| 3618 |
// |
| 3619 |
// SetInRefPic6() |
| 3620 |
// -------------- |
| 3621 |
// This method sets the offset for reference picture 6. It sets the upper bits (32 |
| 3622 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3623 |
// |
| 3624 |
// 31 16 0 |
| 3625 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3626 |
// | OFFSET | SetInRefPic6 |
| 3627 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3628 |
// |
| 3629 |
// #define NV_C9B7_NVENC_SETIN_REFPIC6 0x00000418 /* -W-4R */ |
| 3630 |
// #define NV_C9B7_NVENC_SETIN_REFPIC6_OFFSET 31:0 /* -W-UF */ |
| 3631 |
// |
| 3632 |
// |
| 3633 |
// SetInRefPic7() |
| 3634 |
// -------------- |
| 3635 |
// This method sets the offset for reference picture 7. It sets the upper bits (32 |
| 3636 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3637 |
// |
| 3638 |
// 31 16 0 |
| 3639 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3640 |
// | OFFSET | SetInRefPic7 |
| 3641 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3642 |
// |
| 3643 |
// #define NV_C9B7_NVENC_SETIN_REFPIC7 0x0000041c /* -W-4R */ |
| 3644 |
// #define NV_C9B7_NVENC_SETIN_REFPIC7_OFFSET 31:0 /* -W-UF */ |
| 3645 |
// |
| 3646 |
// |
| 3647 |
// SetInRefPic8() |
| 3648 |
// -------------- |
| 3649 |
// This method sets the offset for reference picture 8. It sets the upper bits (32 |
| 3650 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3651 |
// |
| 3652 |
// 31 16 0 |
| 3653 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3654 |
// | OFFSET | SetInRefPic8 |
| 3655 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3656 |
// |
| 3657 |
// #define NV_C9B7_NVENC_SETIN_REFPIC8 0x00000420 /* -W-4R */ |
| 3658 |
// #define NV_C9B7_NVENC_SETIN_REFPIC8_OFFSET 31:0 /* -W-UF */ |
| 3659 |
// |
| 3660 |
// |
| 3661 |
// SetInRefPic9() |
| 3662 |
// -------------- |
| 3663 |
// This method sets the offset for reference picture 9. It sets the upper bits (32 |
| 3664 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3665 |
// |
| 3666 |
// 31 16 0 |
| 3667 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3668 |
// | OFFSET | SetInRefPic9 |
| 3669 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3670 |
// |
| 3671 |
// #define NV_C9B7_NVENC_SETIN_REFPIC9 0x00000424 /* -W-4R */ |
| 3672 |
// #define NV_C9B7_NVENC_SETIN_REFPIC9_OFFSET 31:0 /* -W-UF */ |
| 3673 |
// |
| 3674 |
// |
| 3675 |
// SetInRefPic10() |
| 3676 |
// --------------- |
| 3677 |
// This method sets the offset for reference picture 10. It sets the upper bits (32 |
| 3678 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3679 |
// |
| 3680 |
// 31 16 0 |
| 3681 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3682 |
// | OFFSET | SetInRefPic10 |
| 3683 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3684 |
// |
| 3685 |
// #define NV_C9B7_NVENC_SETIN_REFPIC10 0x00000428 /* -W-4R */ |
| 3686 |
// #define NV_C9B7_NVENC_SETIN_REFPIC10_OFFSET 31:0 /* -W-UF */ |
| 3687 |
// |
| 3688 |
// |
| 3689 |
// SetInRefPic11() |
| 3690 |
// --------------- |
| 3691 |
// This method sets the offset for reference picture 11. It sets the upper bits (32 |
| 3692 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3693 |
// |
| 3694 |
// 31 16 0 |
| 3695 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3696 |
// | OFFSET | SetInRefPic11 |
| 3697 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3698 |
// |
| 3699 |
// #define NV_C9B7_NVENC_SETIN_REFPIC11 0x0000042c /* -W-4R */ |
| 3700 |
// #define NV_C9B7_NVENC_SETIN_REFPIC11_OFFSET 31:0 /* -W-UF */ |
| 3701 |
// |
| 3702 |
// |
| 3703 |
// SetInRefPic12() |
| 3704 |
// --------------- |
| 3705 |
// This method sets the offset for reference picture 12. It sets the upper bits (32 |
| 3706 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3707 |
// |
| 3708 |
// 31 16 0 |
| 3709 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3710 |
// | OFFSET | SetInRefPic12 |
| 3711 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3712 |
// |
| 3713 |
// #define NV_C9B7_NVENC_SETIN_REFPIC12 0x00000430 /* -W-4R */ |
| 3714 |
// #define NV_C9B7_NVENC_SETIN_REFPIC12_OFFSET 31:0 /* -W-UF */ |
| 3715 |
// |
| 3716 |
// |
| 3717 |
// SetInRefPic13() |
| 3718 |
// --------------- |
| 3719 |
// This method sets the offset for reference picture 13. It sets the upper bits (32 |
| 3720 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3721 |
// |
| 3722 |
// 31 16 0 |
| 3723 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3724 |
// | OFFSET | SetInRefPic13 |
| 3725 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3726 |
// |
| 3727 |
// #define NV_C9B7_NVENC_SETIN_REFPIC13 0x00000434 /* -W-4R */ |
| 3728 |
// #define NV_C9B7_NVENC_SETIN_REFPIC13_OFFSET 31:0 /* -W-UF */ |
| 3729 |
// |
| 3730 |
// |
| 3731 |
// SetInRefPic14() |
| 3732 |
// --------------- |
| 3733 |
// This method sets the offset for reference picture 14. It sets the upper bits (32 |
| 3734 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3735 |
// |
| 3736 |
// 31 16 0 |
| 3737 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3738 |
// | OFFSET | SetInRefPic14 |
| 3739 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3740 |
// |
| 3741 |
// #define NV_C9B7_NVENC_SETIN_REFPIC14 0x00000438 /* -W-4R */ |
| 3742 |
// #define NV_C9B7_NVENC_SETIN_REFPIC14_OFFSET 31:0 /* -W-UF */ |
| 3743 |
// |
| 3744 |
// |
| 3745 |
// SetInRefPic15() |
| 3746 |
// --------------- |
| 3747 |
// This method sets the offset for reference picture 15. It sets the upper bits (32 |
| 3748 |
// most significant bits) of the 40-bit offset where the buffer is located. |
| 3749 |
// |
| 3750 |
// 31 16 0 |
| 3751 |
// .-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-. |
| 3752 |
// | OFFSET | SetInRefPic15 |
| 3753 |
// `-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-' |
| 3754 |
// |
| 3755 |
// #define NV_C9B7_NVENC_SETIN_REFPIC15 0x0000043c /* -W-4R */ |
| 3756 |
// #define NV_C9B7_NVENC_SETIN_REFPIC15_OFFSET 31:0 /* -W-UF */ |
| 3757 |
// |
| 3758 |
// |
| 3759 |
// |
| 3760 |
// |
| 3761 |
// 4 - APPLICATION SPECIFIC ERROR CODES |
| 3762 |
// |
| 3763 |
// The upper 4 bit indicate the codec type. The lower 28 bit are the error code. |
| 3764 |
// |
| 3765 |
// 5 - Some shared defines |
| 3766 |
#define TEB_BUF_SIZE (256 * 1024) |
| 3767 |
#define HIST_BLOCK_SIZE 192 |
| 3768 |
|
| 3769 |
#define H264_HIST_BLOCK_SIZE 224 |
| 3770 |
#define H265_HIST_BLOCK_SIZE H264_HIST_BLOCK_SIZE |
| 3771 |
#define VP9_HIST_BLOCK_SIZE H265_HIST_BLOCK_SIZE |
| 3772 |
|
| 3773 |
//HIST BLOCK structre has not been finalized, subject to change. |
| 3774 |
#define AV1_HIST_BLOCK_SIZE VP9_HIST_BLOCK_SIZE |
| 3775 |
|
| 3776 |
// Errors |
| 3777 |
#define NvC9B7EncErrorH264ApptimerExpired 0x30000001 |
| 3778 |
#define NvC9B7EncErrorH264InvalidInput 0x30000002 |
| 3779 |
#define NvC9B7EncErrorH264HwerrInterrupt 0x30000003 |
| 3780 |
#define NvC9B7EncErrorH264BadMagic 0x30000004 |
| 3781 |
|
| 3782 |
// Appendix A - KEY |
| 3783 |
// |
| 3784 |
// Read |
| 3785 |
// ' ' = Other Information |
| 3786 |
// '-' = Field is part of a write-only register |
| 3787 |
// 'C' = Value read is always the same, constant value line follows (C) |
| 3788 |
// 'R' = Value is read |
| 3789 |
// |
| 3790 |
// Write |
| 3791 |
// ' ' = Other Information |
| 3792 |
// '-' = Must not be written (D), value ignored when written (R,A,F) |
| 3793 |
// 'W' = Can be written |
| 3794 |
// |
| 3795 |
// Internal State |
| 3796 |
// ' ' = Other Information |
| 3797 |
// '-' = No internal state |
| 3798 |
// 'X' = Internal state, initial value is unknown |
| 3799 |
// 'I' = Internal state, initial value is known and follows (I) |
| 3800 |
// 'V' = Internal state, initialize at volatile reset |
| 3801 |
// 'N' = Internal state, initial value and initialize at volatile reset |
| 3802 |
// 'C' = Internal state, initial value at object creation |
| 3803 |
// |
| 3804 |
// Declaration/Size |
| 3805 |
// ' ' = Other Information |
| 3806 |
// '-' = Does Not Apply |
| 3807 |
// 'V' = Type is void |
| 3808 |
// 'U' = Type is unsigned integer |
| 3809 |
// 'S' = Type is signed integer |
| 3810 |
// 'F' = Type is IEEE floating point |
| 3811 |
// '1' = Byte size (008) |
| 3812 |
// '2' = Short size (016) |
| 3813 |
// '3' = Three byte size (024) |
| 3814 |
// '4' = Word size (032) |
| 3815 |
// '8' = Double size (064) |
| 3816 |
// |
| 3817 |
// Define Indicator |
| 3818 |
// ' ' = Other Information |
| 3819 |
// 'D' = Device |
| 3820 |
// 'M' = Memory |
| 3821 |
// 'R' = Register |
| 3822 |
// 'A' = Array of Registers |
| 3823 |
// 'F' = Field |
| 3824 |
// 'V' = Value |
| 3825 |
// 'C' = Clear value |
| 3826 |
// 'S' = Set value |
| 3827 |
// 'T' = Call task function |
| 3828 |
|
| 3829 |
|
| 3830 |
#endif |