1 /* 2 * Copyright (c) 2007-2017 Intel Corporation. All Rights Reserved. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the 6 * "Software"), to deal in the Software without restriction, including 7 * without limitation the rights to use, copy, modify, merge, publish, 8 * distribute, sub license, and/or sell copies of the Software, and to 9 * permit persons to whom the Software is furnished to do so, subject to 10 * the following conditions: 11 * 12 * The above copyright notice and this permission notice (including the 13 * next paragraph) shall be included in all copies or substantial portions 14 * of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 17 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 18 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 19 * IN NO EVENT SHALL INTEL AND/OR ITS SUPPLIERS BE LIABLE FOR 20 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 21 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 22 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 23 */ 24 25 /** 26 * \file va_fei_h264.h 27 * \brief The FEI encoding H264 special API 28 */ 29 30 #ifndef VA_FEI_H264_H 31 #define VA_FEI_H264_H 32 33 #ifdef __cplusplus 34 extern "C" { 35 #endif 36 37 #include <stdint.h> 38 #include "va_fei.h" 39 40 /** 41 * \defgroup api_fei_h264 H.264 FEI encoding API 42 * 43 * @{ 44 */ 45 46 /** \brief FEI frame level control buffer for H.264 */ 47 typedef struct _VAEncMiscParameterFEIFrameControlH264 { 48 uint32_t function; /* one of the VAConfigAttribFEIFunctionType values */ 49 /** \brief MB (16x16) control input buffer. It is valid only when (mb_input | mb_size_ctrl) 50 * is set to 1. The data in this buffer correspond to the input source. 16x16 MB is in raster scan order, 51 * each MB control data structure is defined by VAEncFEIMBControlH264. 52 * Buffer size shall not be less than the number of 16x16 blocks multiplied by 53 * sizeof(VAEncFEIMBControlH264). 54 * Note: if mb_qp is set, VAEncQPBufferH264 is expected. 55 */ 56 VABufferID mb_ctrl; 57 /** \brief distortion output of MB ENC or ENC_PAK. 58 * Each 16x16 block has one distortion data with VAEncFEIDistortionH264 layout 59 * Buffer size shall not be less than the number of 16x16 blocks multiplied by 60 * sizeof(VAEncFEIDistortionH264). 61 */ 62 VABufferID distortion; 63 /** \brief MVs data output of MB ENC. 64 * Each 16x16 block has one MVs data with layout VAMotionVector 65 * Buffer size shall not be less than the number of 16x16 blocks multiplied by 66 * sizeof(VAMotionVector) * 16. 67 */ 68 VABufferID mv_data; 69 /** \brief MBCode data output of MB ENC. 70 * Each 16x16 block has one MB Code data with layout VAEncFEIMBCodeH264 71 * Buffer size shall not be less than the number of 16x16 blocks multiplied by 72 * sizeof(VAEncFEIMBCodeH264). 73 */ 74 VABufferID mb_code_data; 75 /** \brief QP input buffer with layout VAEncQPBufferH264. It is valid only when mb_qp is set to 1. 76 * The data in this buffer correspond to the input source. 77 * One QP per 16x16 block in raster scan order, each QP is a signed char (8-bit) value. 78 **/ 79 VABufferID qp; 80 /** \brief MV predictor. It is valid only when mv_predictor_enable is set to 1. 81 * Each 16x16 block has one or more pair of motion vectors and the corresponding 82 * reference indexes as defined by VAEncFEIMVPredictorH264. 16x16 block is in raster scan order. 83 * Buffer size shall not be less than the number of 16x16 blocks multiplied by 84 * sizeof(VAEncFEIMVPredictorH264). */ 85 VABufferID mv_predictor; 86 87 /** \brief number of MV predictors. It must not be greater than maximum supported MV predictor. */ 88 uint32_t num_mv_predictors_l0 : 16; 89 uint32_t num_mv_predictors_l1 : 16; 90 91 /** \brief motion search method definition 92 * 0: default value, diamond search 93 * 1: full search 94 * 2: diamond search 95 **/ 96 uint32_t search_path : 8; 97 /** \brief maximum number of Search Units, valid range is [1, 63] 98 * 0 is treated as 1. reference search locations are grouped in a predefined pattern, 99 * and all locations within the same group must be either all are chosen or all are skipped. 100 * These predefined groups are called search unit (SU).*/ 101 uint32_t len_sp : 8; 102 uint32_t reserved0 : 16; 103 /** \brief defines the bit-mask for disabling sub-partition 104 * The lower 4 bits are for the major partitions (sub-macroblock) and the higher 3 bits for minor partitions (with sub-partition for 4x(8x8) sub-macroblocks. 105 * xxxxxx1 : 16x16 sub-macroblock disabled 106 * xxxxx1x : 2x(16x8) sub-macroblock within 16x16 disabled 107 * xxxx1xx : 2x(8x16) sub-macroblock within 16x16 disabled 108 * xxx1xxx : 1x(8x8) sub-partition for 4x(8x8) within 16x16 disabled 109 * xx1xxxx : 2x(8x4) sub-partition for 4x(8x8) within 16x16 disabled 110 * x1xxxxx : 2x(4x8) sub-partition for 4x(8x8) within 16x16 disabled 111 * 1xxxxxx : 4x(4x4) sub-partition for 4x(8x8) within 16x16 disabled 112 * 1111111 : Invalid 113 * 0000000 : default value */ 114 uint32_t sub_mb_part_mask : 7; 115 /** specifies which Luma Intra partition is enabled/disabled for intra mode decision. 116 * xxxx1: luma_intra_16x16 disabled 117 * xxx1x: luma_intra_8x8 disabled 118 * xx1xx: luma_intra_4x4 disabled 119 * xx111: intra prediction is disabled */ 120 uint32_t intra_part_mask : 5; 121 /** when set to 1, neighbor MV will be used as predictor; when set to 0, no neighbor MV will be used as predictor.*/ 122 uint32_t multi_pred_l0 : 1; 123 /** when set to 1, neighbor MV will be used as predictor; when set to 0, no neighbor MV will be used as predictor.*/ 124 uint32_t multi_pred_l1 : 1; 125 /**defines the half/quarter pel modes. The mode is inclusive, ie., higher precision mode samples lower precision locations. 126 * 00b: integer mode searching 127 * 01b: half-pel mode searching 128 * 10b: reserved 129 * 11b: quarter-pel mode searching */ 130 uint32_t sub_pel_mode : 2; 131 /** specifies distortion measure adjustments used for the inter motion search SAD comparison. 132 * 00b: none 133 * 10b: Haar transform adjusted*/ 134 uint32_t inter_sad : 2; 135 /** specifies distortion measure adjustments used for the intra motion search SAD comparison. 136 * 00b: none 137 * 10b: Haar transform adjusted*/ 138 uint32_t intra_sad : 2; 139 /** specifies if the output distortion is the raw distortion or cost adjusted distortion. 140 * 0: Raw Distortion without Cost 141 * 1: Distortion with added Cost */ 142 uint32_t distortion_type : 1; 143 /** when set to 1, enables the additional calls on Fraction & Bidirectional Refinement*/ 144 uint32_t repartition_check_enable : 1; 145 /** defines whether adaptive searching is enabled for IME(Integer Motion Estimation). 146 * 0: disable 147 * 1: enable */ 148 uint32_t adaptive_search : 1; 149 /** enables using the motion vector as an extra predictor provided by the host. If it is set, 150 * host needs to provide a buffer with motion vectors and the associated reference index for 151 * each 16x16 block as defined . The host can call processing function to get motion vectors and use as predictor. 152 * 0: MV predictor disabled 153 * 1: MV predictor enabled */ 154 uint32_t mv_predictor_enable : 1; 155 /** enables using the QP buffer to set the QP for each block*/ 156 uint32_t mb_qp : 1; 157 /** enable mb_ctrl buffer to handle MB*/ 158 uint32_t mb_input : 1; 159 /** when this flag is set, mb_ctrl must be set too and a buffer with per MB input 160 * needs to be provided and MaxSizeInWord and */ 161 uint32_t mb_size_ctrl : 1; 162 /** when this flag is set, extra distortion between the current MB and co-located MB is provided. 163 * Extra distortion output has performance impact, set it only when it is needed.*/ 164 uint32_t colocated_mb_distortion : 1; 165 uint32_t reserved1 : 4; 166 167 /** \brief motion search window(ref_width * ref_height) */ 168 uint32_t ref_width : 8; 169 uint32_t ref_height : 8; 170 /** \brief predefined motion search windows. If selected, len_sp, window(ref_width * ref_eight) 171 * and search_path setting are ignored. 172 * 0: not use predefined search window 173 * 1: Tiny, len_sp=4, 24x24 window and diamond search 174 * 2: Small, len_sp=9, 28x28 window and diamond search 175 * 3: Diamond, len_sp=16, 48x40 window and diamond search 176 * 4: Large Diamond, len_sp=32, 48x40 window and diamond search 177 * 5: Exhaustive, len_sp=48, 48x40 window and full search 178 * 6: Extend Diamond, len_sp=16, 64x40 window and diamond search 179 * 7: Extend Large Diamond, len_sp=32, 64x40 window and diamond search 180 * 8: Extend Exhaustive, len_sp=48, 64x40 window and full search 181 **/ 182 uint32_t search_window : 4; 183 uint32_t reserved2 : 12; 184 185 /** \brief max frame size control with multi passes QP setting */ 186 uint32_t max_frame_size; 187 /** \brief number of passes, every pass has different QP */ 188 uint32_t num_passes; 189 /** \brief delta QP list for every pass */ 190 uint8_t *delta_qp; 191 uint32_t reserved3[VA_PADDING_LOW]; 192 } VAEncMiscParameterFEIFrameControlH264; 193 194 /** \brief FEI MB level control data structure */ 195 typedef struct _VAEncFEIMBControlH264 { 196 /** \brief when set, correposndent MB is coded as intra */ 197 uint32_t force_to_intra : 1; 198 /** \brief when set, correposndent MB is coded as skip */ 199 uint32_t force_to_skip : 1; 200 /** \brief specifies whether this macroblock should be coded as a non-skipped macroblock. */ 201 uint32_t force_to_nonskip : 1; 202 uint32_t enable_direct_bias_adjustment : 1; 203 uint32_t enable_motion_bias_adjustment : 1; 204 uint32_t ext_mv_cost_scaling_factor : 3; 205 uint32_t reserved0 : 24; 206 207 uint32_t reserved1; 208 209 uint32_t reserved2; 210 211 uint32_t reserved3 : 16; 212 /** \brief when mb_size_ctrl is set, size here is used to budget accumulatively. Set to 0xFF if don't care. */ 213 uint32_t target_size_in_word : 8; 214 /** \brief specifies the max size of each MB */ 215 uint32_t max_size_in_word : 8; 216 } VAEncFEIMBControlH264; 217 218 219 /** \brief Application can use this definition as reference to allocate the buffer 220 * based on MaxNumPredictor returned from attribute VAConfigAttribFEIMVPredictors query. 221 **/ 222 typedef struct _VAEncFEIMVPredictorH264 { 223 /** \brief Reference index corresponding to the entry of RefPicList0 & RefPicList1 in VAEncSliceParameterBufferH264. 224 * Note that RefPicList0 & RefPicList1 needs to be the same for all slices. 225 * ref_idx_l0_x : index to RefPicList0; ref_idx_l1_x : index to RefPicList1; x : 0 - MaxNumPredictor. 226 **/ 227 struct { 228 uint8_t ref_idx_l0 : 4; 229 uint8_t ref_idx_l1 : 4; 230 } ref_idx[4]; /* index is predictor number */ 231 uint32_t reserved; 232 /** \brief MV. MaxNumPredictor must be the returned value from attribute VAConfigAttribFEIMVPredictors query. 233 * Even application doesn't use the maximum predictors, the VAFEIMVPredictorH264 structure size 234 * has to be defined as maximum so each MB can be at a fixed location. 235 * Note that 0x8000 must be used for correspondent intra block. 236 **/ 237 VAMotionVector mv[4]; /* MaxNumPredictor is 4 */ 238 } VAEncFEIMVPredictorH264; 239 240 /** \brief FEI output */ 241 /** 242 * Motion vector output is per 4x4 block. For each 4x4 block there is a pair of MVs 243 * for RefPicList0 and RefPicList1 and each MV is 4 bytes including horizontal and vertical directions. 244 * Depending on Subblock partition, for the shape that is not 4x4, the MV is replicated 245 * so each 4x4 block has a pair of MVs. The 16x16 block has 32 MVs (128 bytes). 246 * 0x8000 is used for correspondent intra block. The 16x16 block is in raster scan order, 247 * within the 16x16 block, each 4x4 block MV is ordered as below in memory. 248 * The buffer size shall be greater than or equal to the number of 16x16 blocks multiplied by 128 bytes. 249 * Note that, when separate ENC and PAK is enabled, the exact layout of this buffer is needed for PAK input. 250 * App can reuse this buffer, or copy to a different buffer as PAK input. 251 * Layout is defined as Generic motion vector data structure VAMotionVector 252 * 16x16 Block 253 * ----------------------------------------- 254 * | 1 | 2 | 5 | 6 | 255 * ----------------------------------------- 256 * | 3 | 4 | 7 | 8 | 257 * ----------------------------------------- 258 * | 9 | 10 | 13 | 14 | 259 * ----------------------------------------- 260 * | 11 | 12 | 15 | 16 | 261 * ----------------------------------------- 262 **/ 263 264 /** \brief VAEncFEIMBCodeH264 defines the data structure for VAEncFEIMBCodeBufferType per 16x16 MB block. 265 * it is output buffer of ENC and ENC_PAK modes, it's also input buffer of PAK mode. 266 * The 16x16 block is in raster scan order. Buffer size shall not be less than the number of 16x16 blocks 267 * multiplied by sizeof(VAEncFEIMBCodeH264). Note that, when separate ENC and PAK is enabled, 268 * the exact layout of this buffer is needed for PAK input. App can reuse this buffer, 269 * or copy to a different buffer as PAK input, reserved elements must not be modified when used as PAK input. 270 **/ 271 typedef struct _VAEncFEIMBCodeH264 { 272 //DWORD 0~2 273 uint32_t reserved0[3]; 274 275 //DWORD 3 276 uint32_t inter_mb_mode : 2; 277 uint32_t mb_skip_flag : 1; 278 uint32_t reserved1 : 1; 279 uint32_t intra_mb_mode : 2; 280 uint32_t reserved2 : 1; 281 uint32_t field_mb_polarity_flag : 1; 282 uint32_t mb_type : 5; 283 uint32_t intra_mb_flag : 1; 284 uint32_t field_mb_flag : 1; 285 uint32_t transform8x8_flag : 1; 286 uint32_t reserved3 : 1; 287 uint32_t dc_block_coded_cr_flag : 1; 288 uint32_t dc_block_coded_cb_flag : 1; 289 uint32_t dc_block_coded_y_flag : 1; 290 uint32_t reserved4 : 12; 291 292 //DWORD 4 293 uint32_t horz_origin : 8; 294 uint32_t vert_origin : 8; 295 uint32_t cbp_y : 16; 296 297 //DWORD 5 298 uint32_t cbp_cb : 16; 299 uint32_t cbp_cr : 16; 300 301 //DWORD 6 302 uint32_t qp_prime_y : 8; 303 uint32_t reserved5 : 17; 304 uint32_t mb_skip_conv_disable : 1; 305 uint32_t is_last_mb : 1; 306 uint32_t enable_coefficient_clamp : 1; 307 uint32_t direct8x8_pattern : 4; 308 309 //DWORD 7 8 and 9 310 union { 311 /* Intra MBs */ 312 struct { 313 uint32_t luma_intra_pred_modes0 : 16; 314 uint32_t luma_intra_pred_modes1 : 16; 315 316 uint32_t luma_intra_pred_modes2 : 16; 317 uint32_t luma_intra_pred_modes3 : 16; 318 319 uint32_t chroma_intra_pred_mode : 2; 320 uint32_t intra_pred_avail_flag : 5; 321 uint32_t intra_pred_avail_flagF : 1; 322 uint32_t reserved6 : 24; 323 } intra_mb; 324 325 /* Inter MBs */ 326 struct { 327 uint32_t sub_mb_shapes : 8; 328 uint32_t sub_mb_pred_modes : 8; 329 uint32_t reserved7 : 16; 330 331 uint32_t ref_idx_l0_0 : 8; 332 uint32_t ref_idx_l0_1 : 8; 333 uint32_t ref_idx_l0_2 : 8; 334 uint32_t ref_idx_l0_3 : 8; 335 336 uint32_t ref_idx_l1_0 : 8; 337 uint32_t ref_idx_l1_1 : 8; 338 uint32_t ref_idx_l1_2 : 8; 339 uint32_t ref_idx_l1_3 : 8; 340 } inter_mb; 341 } mb_mode; 342 343 //DWORD 10 344 uint32_t reserved8 : 16; 345 uint32_t target_size_in_word : 8; 346 uint32_t max_size_in_word : 8; 347 348 //DWORD 11~14 349 uint32_t reserved9[4]; 350 351 //DWORD 15 352 uint32_t reserved10; 353 } VAEncFEIMBCodeH264; // 64 bytes 354 355 /** \brief VAEncFEIDistortionH264 defines the data structure for VAEncFEIDistortionBufferType per 16x16 MB block. 356 * It is output buffer of ENC and ENC_PAK modes, The 16x16 block is in raster scan order. 357 * Buffer size shall not be less than the number of 16x16 blocks multiple by sizeof(VAEncFEIDistortionH264). 358 **/ 359 typedef struct _VAEncFEIDistortionH264 { 360 /** \brief Inter-prediction-distortion associated with motion vector i (co-located with subblock_4x4_i). 361 * Its meaning is determined by sub-shape. It must be zero if the corresponding sub-shape is not chosen. 362 **/ 363 uint16_t inter_distortion[16]; 364 uint32_t best_inter_distortion : 16; 365 uint32_t best_intra_distortion : 16; 366 uint32_t colocated_mb_distortion : 16; 367 uint32_t reserved0 : 16; 368 uint32_t reserved1[2]; 369 } VAEncFEIDistortionH264; // 48 bytes 370 371 /** \brief Motion Vector and Statistics frame level controls. 372 * VAStatsStatisticsParameterBufferType for H264 16x16 block 373 **/ 374 typedef struct _VAStatsStatisticsParameterH264 { 375 VAStatsStatisticsParameter stats_params; 376 377 uint32_t frame_qp : 8; 378 /** \brief length of search path */ 379 uint32_t len_sp : 8; 380 /** \brief motion search method definition 381 * 0: default value, diamond search 382 * 1: full search 383 * 2: diamond search 384 **/ 385 uint32_t search_path : 8; 386 uint32_t reserved0 : 8; 387 388 uint32_t sub_mb_part_mask : 7; 389 /** \brief sub pixel mode definition 390 * 00b: integer mode searching 391 * 01b: half-pel mode searching 392 * 10b: reserved 393 * 11b: quarter-pel mode searching 394 **/ 395 uint32_t sub_pel_mode : 2; 396 /** \brief distortion measure adjustment for inter search SAD comparison 397 * 00b: none 398 * 01b: reserved 399 * 10b: Haar transform adjusted 400 * 11b: reserved 401 **/ 402 uint32_t inter_sad : 2; 403 /** \brief distortion measure adjustment for intra search SAD comparison 404 * 00b: none 405 * 01b: reserved 406 * 10b: Haar transform adjusted 407 * 11b: reserved 408 **/ 409 uint32_t intra_sad : 2; 410 uint32_t adaptive_search : 1; 411 /** \brief indicate if future or/and past MV in mv_predictor buffer is valid. 412 * 0: MV predictor disabled 413 * 1: MV predictor enabled for past reference 414 * 2: MV predictor enabled for future reference 415 * 3: MV predictor enabled for both past and future references 416 **/ 417 uint32_t mv_predictor_ctrl : 3; 418 uint32_t mb_qp : 1; 419 /** \brief forward transform enable 420 * 0: disable 421 * 1: enable, needs frame_qp or mb_qp input for transform 422 **/ 423 uint32_t ft_enable : 1; 424 /** \brief luma intra mode partition mask 425 * xxxx1: luma_intra_16x16 disabled 426 * xxx1x: luma_intra_8x8 disabled 427 * xx1xx: luma_intra_4x4 disabled 428 * xx111: intra prediction is disabled 429 **/ 430 uint32_t intra_part_mask : 5; 431 uint32_t reserved1 : 8; 432 433 /** \brief motion search window(ref_width * ref_height) */ 434 uint32_t ref_width : 8; 435 uint32_t ref_height : 8; 436 /** \brief predefined motion search windows. If selected, len_sp, window(ref_width * ref_eight) 437 * and search_path setting are ignored. 438 * 0: not use predefined search window 439 * 1: Tiny, len_sp=4, 24x24 window and diamond search 440 * 2: Small, len_sp=9, 28x28 window and diamond search 441 * 3: Diamond, len_sp=16, 48x40 window and diamond search 442 * 4: Large Diamond, len_sp=32, 48x40 window and diamond search 443 * 5: Exhaustive, len_sp=48, 48x40 window and full search 444 * 6: Extend Diamond, len_sp=16, 64x40 window and diamond search 445 * 7: Extend Large Diamond, len_sp=32, 64x40 window and diamond search 446 * 8: Extend Exhaustive, len_sp=48, 64x40 window and full search 447 **/ 448 uint32_t search_window : 4; 449 uint32_t reserved2 : 12; 450 451 /** \brief MVOutput. When set to 1, MV output is NOT provided */ 452 uint32_t disable_mv_output : 1; 453 /** \brief StatisticsOutput. When set to 1, Statistics output is NOT provided. */ 454 uint32_t disable_statistics_output : 1; 455 /** \brief block 8x8 data enabling in statistics output */ 456 uint32_t enable_8x8_statistics : 1; 457 uint32_t reserved3 : 29; 458 uint32_t reserved4[2]; 459 } VAStatsStatisticsParameterH264; 460 461 /** \brief VAStatsStatisticsH264. H264 Statistics buffer layout for VAStatsStatisticsBufferType 462 * and VAStatsStatisticsBottomFieldBufferType(for interlaced only). 463 * Statistics output is per 16x16 block. Data structure per 16x16 block is defined below. 464 * The 16x16 block is in raster scan order. The buffer size shall be greater than or equal to 465 * the number of 16x16 blocks multiplied by sizeof(VAStatsStatisticsH264). 466 **/ 467 typedef struct _VAStatsStatisticsH264 { 468 /** \brief past reference */ 469 uint32_t best_inter_distortion0 : 16; 470 uint32_t inter_mode0 : 16; 471 472 /** \brief future reference */ 473 uint32_t best_inter_distortion1 : 16; 474 uint32_t inter_mode1 : 16; 475 476 uint32_t best_intra_distortion : 16; 477 uint32_t intra_mode : 16; 478 479 uint32_t num_non_zero_coef : 16; 480 uint32_t reserved0 : 16; 481 482 uint32_t sum_coef; 483 484 /** \brief DWORD 5 flat info **/ 485 uint32_t mb_is_flat : 1; 486 uint32_t reserved1 : 31; 487 488 /** \brief DWORD 6 variance for block16x16**/ 489 uint32_t variance_16x16; 490 /** \brief DWORD 7 ~ 10, variance for block8x8 **/ 491 uint32_t variance_8x8[4]; 492 493 /** \brief DWORD 11 pixel_average for block16x16 **/ 494 uint32_t pixel_average_16x16; 495 /** \brief DWORD 12 ~ 15, pixel_average for block8x8 **/ 496 uint32_t pixel_average_8x8[4]; 497 } VAStatsStatisticsH264; // 64 bytes 498 499 500 #ifdef __cplusplus 501 } 502 #endif 503 504 #endif /* VA_FEI_H264_H */ 505