1 /****************************************************************************** 2 * 3 * Copyright (C) 2015 The Android Open Source Project 4 * 5 * Licensed under the Apache License, Version 2.0 (the "License"); 6 * you may not use this file except in compliance with the License. 7 * You may obtain a copy of the License at: 8 * 9 * http://www.apache.org/licenses/LICENSE-2.0 10 * 11 * Unless required by applicable law or agreed to in writing, software 12 * distributed under the License is distributed on an "AS IS" BASIS, 13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 14 * See the License for the specific language governing permissions and 15 * limitations under the License. 16 * 17 ***************************************************************************** 18 * Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore 19 */ 20 21 /** 22 ******************************************************************************* 23 * @file 24 * ih264e_utils.c 25 * 26 * @brief 27 * Contains miscellaneous utility functions used by the encoder 28 * 29 * @author 30 * ittiam 31 * 32 * @par List of Functions: 33 * - ih264e_get_min_level() 34 * - ih264e_get_lvl_idx() 35 * - ih264e_get_dpb_size() 36 * - ih264e_get_total_pic_buf_size() 37 * - ih264e_get_pic_mv_bank_size() 38 * - ih264e_pic_buf_mgr_add_bufs() 39 * - ih264e_mv_buf_mgr_add_bufs() 40 * - ih264e_init_quant_params() 41 * - ih264e_init_air_map() 42 * - ih264e_codec_init() 43 * - ih264e_pic_init() 44 * 45 * @remarks 46 * None 47 * 48 ******************************************************************************* 49 */ 50 51 /*****************************************************************************/ 52 /* File Includes */ 53 /*****************************************************************************/ 54 55 /* system include files */ 56 #include <stdio.h> 57 #include <stddef.h> 58 #include <stdlib.h> 59 #include <string.h> 60 #include <assert.h> 61 62 /* user include files */ 63 #include "ih264_typedefs.h" 64 #include "iv2.h" 65 #include "ive2.h" 66 #include "ih264e.h" 67 #include "ithread.h" 68 #include "ih264_defs.h" 69 #include "ih264_size_defs.h" 70 #include "ime_distortion_metrics.h" 71 #include "ime_defs.h" 72 #include "ime_structs.h" 73 #include "ih264_error.h" 74 #include "ih264_structs.h" 75 #include "ih264_trans_quant_itrans_iquant.h" 76 #include "ih264_inter_pred_filters.h" 77 #include "ih264_mem_fns.h" 78 #include "ih264_padding.h" 79 #include "ih264_intra_pred_filters.h" 80 #include "ih264_deblk_edge_filters.h" 81 #include "ih264_cabac_tables.h" 82 #include "ih264_macros.h" 83 #include "ih264_common_tables.h" 84 #include "ih264_debug.h" 85 #include "ih264_trans_data.h" 86 #include "ih264e_defs.h" 87 #include "ih264e_globals.h" 88 #include "ih264_buf_mgr.h" 89 #include "ih264_dpb_mgr.h" 90 #include "ih264e_error.h" 91 #include "ih264e_bitstream.h" 92 #include "irc_cntrl_param.h" 93 #include "irc_frame_info_collector.h" 94 #include "ih264e_rate_control.h" 95 #include "ih264e_cabac_structs.h" 96 #include "ih264e_structs.h" 97 #include "ih264e_cabac.h" 98 #include "ih264e_utils.h" 99 #include "ih264e_config.h" 100 #include "ih264e_statistics.h" 101 #include "ih264e_trace.h" 102 #include "ih264_list.h" 103 #include "ih264e_encode_header.h" 104 #include "ih264e_me.h" 105 #include "ime.h" 106 #include "ih264e_core_coding.h" 107 #include "ih264e_rc_mem_interface.h" 108 #include "ih264e_time_stamp.h" 109 #include "ih264e_debug.h" 110 #include "ih264e_process.h" 111 #include "ih264e_master.h" 112 #include "irc_rate_control_api.h" 113 #include "ime_statistics.h" 114 115 /*****************************************************************************/ 116 /* Function Definitions */ 117 /*****************************************************************************/ 118 119 /** 120 ******************************************************************************* 121 * 122 * @brief 123 * Queues the current buffer, gets back a another buffer for encoding with corrent 124 * picture type 125 * 126 * @par Description: 127 * This function performs 3 distinct but related functions. 128 * 1) Maintains an input queue [Note the the term queue donot imply a 129 * first-in first-out logic here] that queues input and dequeues them so 130 * that input frames can be encoded at any predetermined encoding order 131 * 2) Uses RC library to decide which frame must be encoded in current pass 132 * and which picture type it must be encoded to. 133 * 3) Uses RC library to decide the QP at which current frame has to be 134 * encoded 135 * 4) Determines if the current picture must be encoded or not based on 136 * PRE-ENC skip 137 * 138 * Input queue is used for storing input buffers till they are used for 139 * encoding. This queue is maintained at ps_codec->as_inp_list. Whenever a 140 * valid input comes, it is added to the end of queue. This same input is 141 * added to RC queue using the identifier as ps_codec->i4_pic_cnt. Hence any 142 * pic from RC can be located in the input queue easily. 143 * 144 * The dequeue operation does not start till we have ps_codec->s_cfg.u4_max_num_bframes 145 * frames in the queue. THis is done in order to ensure that once output starts 146 * we will have a constant stream of output with no gaps. 147 * 148 * THe output frame order is governed by RC library. When ever we dequeue a 149 * buffer from RC library, it ensures that we will get them in encoding order 150 * With the output of RC library, we can use the picture id to dequeue the 151 * corresponding buffer from input queue and encode it. 152 * 153 * Condition at the end of stream. 154 * ------------------------------- 155 * At the last valid buffer from the app, we will get ps_ive_ip->u4_is_last 156 * to be set. This will the given to lib when appropriate input buffer is 157 * given to encoding. 158 * 159 * Since we have to output is not in sync with input, we will have frames to 160 * encode even after we recive the last vaild input buffer. Hence we have to 161 * make sure that we donot queue any new buffers once we get the flag [It may 162 * mess up GOP ?]. This is acheived by setting ps_codec->i4_last_inp_buff_received 163 * to act as a permenent marker for last frame recived [This may not be needed, 164 * because in our current app, all buffers after the last are marked as last. 165 * But can we rely on that?] . Hence after this flgag is set no new buffers are 166 * queued. 167 * 168 * @param[in] ps_codec 169 * Pointer to codec descriptor 170 * 171 * @param[in] ps_ive_ip 172 * Current input buffer to the encoder 173 * 174 * @param[out] ps_inp 175 * Buffer to be encoded in the current pass 176 * 177 * @returns 178 * Flag indicating if we have a pre-enc skip or not 179 * 180 * @remarks 181 * TODO (bpic) 182 * The check for null ans is last is redudent. 183 * Need to see if we can remove it 184 * 185 ******************************************************************************* 186 */ 187 WORD32 ih264e_input_queue_update(codec_t *ps_codec, 188 ive_video_encode_ip_t *ps_ive_ip, 189 inp_buf_t *ps_enc_buff) 190 { 191 192 inp_buf_t *ps_inp_buf; 193 picture_type_e e_pictype; 194 WORD32 i4_skip; 195 UWORD32 ctxt_sel, u4_pic_id, u4_pic_disp_id; 196 UWORD8 u1_frame_qp; 197 UWORD32 max_frame_bits = 0x7FFFFFFF; 198 199 /* Mark that the last input frame has been received */ 200 if (ps_ive_ip->u4_is_last == 1) 201 { 202 ps_codec->i4_last_inp_buff_received = 1; 203 } 204 205 if (ps_ive_ip->s_inp_buf.apv_bufs[0] == NULL 206 && !ps_codec->i4_last_inp_buff_received) 207 { 208 ps_enc_buff->s_raw_buf.apv_bufs[0] = NULL; 209 ps_enc_buff->u4_is_last = ps_ive_ip->u4_is_last; 210 return 0; 211 } 212 213 /*************************************************************************** 214 * Check for pre enc skip 215 * When src and target frame rates donot match, we skip some frames to 216 * maintain the relation ship between them 217 **************************************************************************/ 218 { 219 WORD32 skip_src; 220 221 skip_src = ih264e_update_rc_framerates( 222 ps_codec->s_rate_control.pps_rate_control_api, 223 ps_codec->s_rate_control.pps_pd_frm_rate, 224 ps_codec->s_rate_control.pps_time_stamp, 225 ps_codec->s_rate_control.pps_frame_time); 226 227 if (skip_src) 228 { 229 ps_enc_buff->u4_is_last = ps_ive_ip->u4_is_last; 230 return 1; 231 } 232 } 233 234 /*************************************************************************** 235 *Queue the input to the queue 236 **************************************************************************/ 237 ps_inp_buf = &(ps_codec->as_inp_list[ps_codec->i4_pic_cnt 238 % MAX_NUM_BFRAMES]); 239 240 /* copy input info. to internal structure */ 241 ps_inp_buf->s_raw_buf = ps_ive_ip->s_inp_buf; 242 ps_inp_buf->u4_timestamp_low = ps_ive_ip->u4_timestamp_low; 243 ps_inp_buf->u4_timestamp_high = ps_ive_ip->u4_timestamp_high; 244 ps_inp_buf->u4_is_last = ps_ive_ip->u4_is_last; 245 ps_inp_buf->pv_mb_info = ps_ive_ip->pv_mb_info; 246 ps_inp_buf->u4_mb_info_type = ps_ive_ip->u4_mb_info_type; 247 ps_inp_buf->pv_pic_info = ps_ive_ip->pv_pic_info; 248 ps_inp_buf->u4_pic_info_type = ps_ive_ip->u4_pic_info_type; 249 250 /*************************************************************************** 251 * Now we should add the picture to RC stack here 252 **************************************************************************/ 253 /* 254 * If an I frame has been requested, ask RC to force it 255 * For IDR requests, we have to ask RC to force I and set IDR by our selves 256 * since RC Donot know about IDR. For forcing an IDR at dequeue stage we 257 * should record that an IDR has been requested some where. Hence we will 258 * store it in the u4_idr_inp_list at a position same as that of input frame 259 */ 260 { 261 WORD32 i4_force_idr, i4_force_i; 262 263 i4_force_idr = (ps_codec->force_curr_frame_type == IV_IDR_FRAME); 264 i4_force_idr |= !(ps_codec->i4_pic_cnt % ps_codec->s_cfg.u4_idr_frm_interval); 265 266 i4_force_i = (ps_codec->force_curr_frame_type == IV_I_FRAME); 267 268 ps_codec->i4_pending_idr_flag |= i4_force_idr; 269 270 if ((ps_codec->i4_pic_cnt > 0) && (i4_force_idr || i4_force_i)) 271 { 272 irc_force_I_frame(ps_codec->s_rate_control.pps_rate_control_api); 273 } 274 ps_codec->force_curr_frame_type = IV_NA_FRAME; 275 } 276 277 irc_add_picture_to_stack(ps_codec->s_rate_control.pps_rate_control_api, 278 ps_codec->i4_pic_cnt); 279 280 281 /* Delay */ 282 if (ps_codec->i4_encode_api_call_cnt 283 < (WORD32)(ps_codec->s_cfg.u4_num_bframes)) 284 { 285 ps_enc_buff->s_raw_buf.apv_bufs[0] = NULL; 286 ps_enc_buff->u4_is_last = 0; 287 return 0; 288 } 289 290 /*************************************************************************** 291 * Get a new pic to encode 292 **************************************************************************/ 293 /* Query the picture_type */ 294 e_pictype = ih264e_rc_get_picture_details( 295 ps_codec->s_rate_control.pps_rate_control_api, (WORD32 *)(&u4_pic_id), 296 (WORD32 *)(&u4_pic_disp_id)); 297 298 switch (e_pictype) 299 { 300 case I_PIC: 301 ps_codec->pic_type = PIC_I; 302 break; 303 case P_PIC: 304 ps_codec->pic_type = PIC_P; 305 break; 306 case B_PIC: 307 ps_codec->pic_type = PIC_B; 308 break; 309 default: 310 ps_codec->pic_type = PIC_NA; 311 ps_enc_buff->s_raw_buf.apv_bufs[0] = NULL; 312 return 0; 313 } 314 315 /* Set IDR if it has been requested */ 316 if (ps_codec->pic_type == PIC_I) 317 { 318 ps_codec->pic_type = ps_codec->i4_pending_idr_flag ? 319 PIC_IDR : ps_codec->pic_type; 320 ps_codec->i4_pending_idr_flag = 0; 321 } 322 323 /* Get current frame Qp */ 324 u1_frame_qp = (UWORD8)irc_get_frame_level_qp( 325 ps_codec->s_rate_control.pps_rate_control_api, e_pictype, 326 max_frame_bits); 327 ps_codec->u4_frame_qp = gau1_mpeg2_to_h264_qmap[u1_frame_qp]; 328 329 /* 330 * copy the pic id to poc because the display order is assumed to be same 331 * as input order 332 */ 333 ps_codec->i4_poc = u4_pic_id; 334 335 /*************************************************************************** 336 * Now retrieve the correct picture from the queue 337 **************************************************************************/ 338 339 /* Mark the skip flag */ 340 i4_skip = 0; 341 ctxt_sel = ps_codec->i4_encode_api_call_cnt % MAX_CTXT_SETS; 342 ps_codec->s_rate_control.pre_encode_skip[ctxt_sel] = i4_skip; 343 344 /* Get a buffer to encode */ 345 ps_inp_buf = &(ps_codec->as_inp_list[u4_pic_id % MAX_NUM_BFRAMES]); 346 347 /* copy dequeued input to output */ 348 ps_enc_buff->s_raw_buf = ps_inp_buf->s_raw_buf; 349 ps_enc_buff->u4_timestamp_low = ps_inp_buf->u4_timestamp_low; 350 ps_enc_buff->u4_timestamp_high = ps_inp_buf->u4_timestamp_high; 351 ps_enc_buff->u4_is_last = ps_inp_buf->u4_is_last; 352 ps_enc_buff->pv_mb_info = ps_inp_buf->pv_mb_info; 353 ps_enc_buff->u4_mb_info_type = ps_inp_buf->u4_mb_info_type; 354 ps_enc_buff->pv_pic_info = ps_inp_buf->pv_pic_info; 355 ps_enc_buff->u4_pic_info_type = ps_inp_buf->u4_pic_info_type; 356 357 /* Special case for encoding trailing B frames 358 * 359 * In encoding streams with B frames it may happen that we have a B frame 360 * at the end without a P/I frame after it. Hence when we are dequeing from 361 * the RC, it will return the P frame [next in display order but before in 362 * encoding order] first. Since the dequeue happens for an invalid frame we 363 * will get a frame with null buff and set u4_is_last. Hence lib with return 364 * last frame flag at this point and will stop encoding. 365 * 366 * Since for the last B frame, we does not have the forward ref frame 367 * it makes sense to force it into P. 368 * 369 * To solve this, in case the current frame is P and if the last frame flag 370 * is set, we need to see if there is and pending B frames. If there are any, 371 * we should just encode that picture as the current P frame and set 372 * that B frame as the last frame. Hence the encoder will terminate naturally 373 * once that B-frame is encoded after all the in between frames. 374 * 375 * Since we cannot touch RC stack directly, the option of actually swapping 376 * frames in RC is ruled out. We have to modify the as_inp_list to simulate 377 * such a behavior by RC. We can do that by 378 * 1) Search through as_inp_list to locate the largest u4_timestamp_low less 379 * than current u4_timestamp_low. This will give us the last B frame before 380 * the current P frame. Note that this will handle pre encode skip too since 381 * queue happens after pre enc skip. 382 * 2) Swap the position in as_inp_list. Hence now the last B frame is 383 * encoded as P frame. And the new last B frame will have u4_is_last 384 * set so that encoder will end naturally once we reached that B frame 385 * or any subsequent frame. Also the current GOP will have 1 less B frame 386 * Since we are swapping, the poc will also be in-order. 387 * 3) In case we have an IPP stream, the result of our search will be an 388 * I/P frame which is already encoded. Thus swap and encode will result 389 * in encoding of duplicate frames. Hence to avoid this we will only 390 * have this work around in case of u4_num_bframes > 0. 391 * 392 * In case we have forced an I/IDR frame In between this P frame and 393 * the last B frame -> This cannot happen as the current P frame is 394 * supposed to have u4_is_last set. Thus forcing an I/ IDR after this 395 * is illogical. 396 * 397 * In cae if we have forced an I such that the frame just before last frame 398 * in is I/P -> This case will never arise. Since we have a closed GOP now, 399 * once we force an I, the gop gets reset, hence there will be a B between 400 * I/P and I/P. 401 */ 402 if (ps_enc_buff->u4_is_last && (ps_codec->pic_type == PIC_P) 403 && ps_codec->s_cfg.u4_num_bframes && (ps_codec->i4_poc > 1)) 404 { 405 UWORD32 u4_cntr, u4_lst_bframe; 406 inp_buf_t *ps_swap_buff, *ps_inp_list, *ps_cur_pic; 407 408 u4_cntr = (u4_pic_id + 1) % MAX_NUM_BFRAMES; 409 u4_lst_bframe = u4_pic_id ? ((u4_pic_id - 1) % MAX_NUM_BFRAMES) : (MAX_NUM_BFRAMES - 1); 410 411 ps_inp_list = &ps_codec->as_inp_list[0]; 412 ps_cur_pic = &ps_inp_list[u4_pic_id % MAX_NUM_BFRAMES]; 413 414 /* Now search the pic in most recent past to current frame */ 415 for(; u4_cntr != (u4_pic_id % MAX_NUM_BFRAMES); 416 u4_cntr = ((u4_cntr + 1) % MAX_NUM_BFRAMES)) 417 { 418 if ( (ps_inp_list[u4_cntr].u4_timestamp_low <= ps_cur_pic->u4_timestamp_low) && 419 (ps_inp_list[u4_cntr].u4_timestamp_high <= ps_cur_pic->u4_timestamp_high) && 420 (ps_inp_list[u4_cntr].u4_timestamp_low >= ps_inp_list[u4_lst_bframe].u4_timestamp_low) && 421 (ps_inp_list[u4_cntr].u4_timestamp_high >= ps_inp_list[u4_lst_bframe].u4_timestamp_high)) 422 { 423 u4_lst_bframe = u4_cntr; 424 } 425 } 426 427 ps_swap_buff = &(ps_codec->as_inp_list[u4_lst_bframe]); 428 429 /* copy the last B buffer to output */ 430 *ps_enc_buff = *ps_swap_buff; 431 432 /* Store the current buf into the queue in place of last B buf */ 433 *ps_swap_buff = *ps_inp_buf; 434 435 } 436 437 if (ps_enc_buff->u4_is_last) 438 { 439 ps_codec->pic_type = PIC_NA; 440 } 441 442 /* Return the buffer status */ 443 return (0); 444 } 445 446 /** 447 ******************************************************************************* 448 * 449 * @brief 450 * Used to get minimum level index for a given picture size 451 * 452 * @par Description: 453 * Gets the minimum level index and then gets corresponding level. 454 * Also used to ignore invalid levels like 2.3, 3.3 etc 455 * 456 * @param[in] level 457 * Level of the stream 458 * 459 * @returns Level index for a given level 460 * 461 * @remarks 462 * 463 ******************************************************************************* 464 */ 465 WORD32 ih264e_get_min_level(WORD32 wd, WORD32 ht) 466 { 467 WORD32 lvl_idx = MAX_LEVEL, i; 468 WORD32 pic_size = wd * ht; 469 WORD32 max = MAX(wd, ht); 470 for (i = 0; i < MAX_LEVEL; i++) 471 { 472 if ((pic_size <= gai4_ih264_max_luma_pic_size[i]) && 473 (max <= gai4_ih264_max_wd_ht[i])) 474 { 475 lvl_idx = i; 476 break; 477 } 478 } 479 480 return gai4_ih264_levels[lvl_idx]; 481 } 482 483 /** 484 ******************************************************************************* 485 * 486 * @brief 487 * Used to get level index for a given level 488 * 489 * @par Description: 490 * Converts from level_idc (which is multiplied by 30) to an index that can be 491 * used as a lookup. Also used to ignore invalid levels like 2.2 , 3.2 etc 492 * 493 * @param[in] level 494 * Level of the stream 495 * 496 * @returns Level index for a given level 497 * 498 * @remarks 499 * 500 ******************************************************************************* 501 */ 502 WORD32 ih264e_get_lvl_idx(WORD32 level) 503 { 504 WORD32 lvl_idx = 0; 505 506 if (level < IH264_LEVEL_11) 507 { 508 lvl_idx = 0; 509 } 510 else if (level < IH264_LEVEL_12) 511 { 512 lvl_idx = 1; 513 } 514 else if (level < IH264_LEVEL_13) 515 { 516 lvl_idx = 2; 517 } 518 else if (level < IH264_LEVEL_20) 519 { 520 lvl_idx = 3; 521 } 522 else if (level < IH264_LEVEL_21) 523 { 524 lvl_idx = 4; 525 } 526 else if (level < IH264_LEVEL_22) 527 { 528 lvl_idx = 5; 529 } 530 else if (level < IH264_LEVEL_30) 531 { 532 lvl_idx = 6; 533 } 534 else if (level < IH264_LEVEL_31) 535 { 536 lvl_idx = 7; 537 } 538 else if (level < IH264_LEVEL_32) 539 { 540 lvl_idx = 8; 541 } 542 else if (level < IH264_LEVEL_40) 543 { 544 lvl_idx = 9; 545 } 546 else if (level < IH264_LEVEL_41) 547 { 548 lvl_idx = 10; 549 } 550 else if (level < IH264_LEVEL_42) 551 { 552 lvl_idx = 11; 553 } 554 else if (level < IH264_LEVEL_50) 555 { 556 lvl_idx = 12; 557 } 558 else if (level < IH264_LEVEL_51) 559 { 560 lvl_idx = 13; 561 } 562 else 563 { 564 lvl_idx = 14; 565 } 566 567 return (lvl_idx); 568 } 569 570 /** 571 ******************************************************************************* 572 * 573 * @brief returns maximum number of pictures allowed in dpb for a given level 574 * 575 * @par Description: 576 * For given width, height and level, number of pictures allowed in decoder 577 * picture buffer is computed as per Annex A.3.1 578 * 579 * @param[in] level 580 * level of the bit-stream 581 * 582 * @param[in] pic_size 583 * width * height 584 * 585 * @returns Number of buffers in DPB 586 * 587 * @remarks 588 * From annexure A.3.1 of H264 specification, 589 * max_dec_frame_buffering <= MaxDpbSize, where MaxDpbSize is equal to 590 * Min( 1024 * MaxDPB / ( PicWidthInMbs * FrameHeightInMbs * 384 ), 16 ) and 591 * MaxDPB is given in Table A-1 in units of 1024 bytes. However the MaxDPB size 592 * presented in the look up table gas_ih264_lvl_tbl is in units of 512 593 * bytes. Hence the expression is modified accordingly. 594 * 595 ******************************************************************************* 596 */ 597 WORD32 ih264e_get_dpb_size(WORD32 level, WORD32 pic_size) 598 { 599 /* dpb size */ 600 WORD32 max_dpb_size_bytes = 0; 601 602 /* dec frame buffering */ 603 WORD32 max_dpb_size_frames = 0; 604 605 /* temp var */ 606 WORD32 i; 607 608 /* determine max luma samples */ 609 for (i = 0; i < 16; i++) 610 if (level == (WORD32)gas_ih264_lvl_tbl[i].u4_level_idc) 611 max_dpb_size_bytes = gas_ih264_lvl_tbl[i].u4_max_dpb_size; 612 613 /* from Annexure A.3.1 h264 specification */ 614 max_dpb_size_frames = 615 MIN( 1024 * max_dpb_size_bytes / ( pic_size * 3 ), MAX_DPB_SIZE ); 616 617 return max_dpb_size_frames; 618 } 619 620 /** 621 ******************************************************************************* 622 * 623 * @brief 624 * Used to get reference picture buffer size for a given level and 625 * and padding used 626 * 627 * @par Description: 628 * Used to get reference picture buffer size for a given level and padding used 629 * Each picture is padded on all four sides 630 * 631 * @param[in] pic_size 632 * Number of luma samples (Width * Height) 633 * 634 * @param[in] level 635 * Level 636 * 637 * @param[in] horz_pad 638 * Total padding used in horizontal direction 639 * 640 * @param[in] vert_pad 641 * Total padding used in vertical direction 642 * 643 * @returns Total picture buffer size 644 * 645 * @remarks 646 * 647 * 648 ******************************************************************************* 649 */ 650 WORD32 ih264e_get_total_pic_buf_size(WORD32 pic_size, 651 WORD32 level, 652 WORD32 horz_pad, 653 WORD32 vert_pad, 654 WORD32 num_ref_frames, 655 WORD32 num_reorder_frames) 656 { 657 WORD32 size; 658 WORD32 num_luma_samples; 659 WORD32 lvl_idx; 660 WORD32 max_wd, min_ht; 661 WORD32 num_samples; 662 WORD32 max_num_bufs; 663 WORD32 pad = MAX(horz_pad, vert_pad); 664 665 /* 666 * If num_ref_frames and num_reorder_frmaes is specified 667 * Use minimum value 668 */ 669 max_num_bufs = (num_ref_frames + num_reorder_frames + MAX_CTXT_SETS); 670 671 /* Get level index */ 672 lvl_idx = ih264e_get_lvl_idx(level); 673 674 /* Maximum number of luma samples in a picture at given level */ 675 num_luma_samples = gai4_ih264_max_luma_pic_size[lvl_idx]; 676 num_luma_samples = MAX(num_luma_samples, pic_size); 677 678 /* Account for chroma */ 679 num_samples = num_luma_samples * 3 / 2; 680 681 /* Maximum width of luma samples in a picture at given level */ 682 max_wd = gai4_ih264_max_wd_ht[lvl_idx]; 683 684 /* Minimum height of luma samples in a picture at given level */ 685 min_ht = gai4_ih264_min_wd_ht[lvl_idx]; 686 687 /* Allocation is required for 688 * (Wd + horz_pad) * (Ht + vert_pad) * (2 * max_dpb_size + 1) 689 * 690 * Above expanded as 691 * ((Wd * Ht) + (horz_pad * vert_pad) + Wd * vert_pad + Ht * horz_pad) * (2 * max_dpb_size + 1) 692 * (Wd * Ht) * (2 * max_dpb_size + 1) + ((horz_pad * vert_pad) + Wd * vert_pad + Ht * horz_pad) * (2 * max_dpb_size + 1) 693 * Now max_dpb_size increases with smaller Wd and Ht, but Wd * ht * max_dpb_size will still be lesser or equal to max_wd * max_ht * dpb_size 694 * 695 * In the above equation (Wd * Ht) * (2 * max_dpb_size + 1) is accounted by using num_samples * (2 * max_dpb_size + 1) below 696 * 697 * For the padded area use MAX(horz_pad, vert_pad) as pad 698 * ((pad * pad) + pad * (Wd + Ht)) * (2 * max_dpb_size + 1) has to accounted from the above for padding 699 * 700 * Since Width and Height can change worst Wd + Ht is when One of the dimensions is max and other is min 701 * So use max_wd and min_ht 702 */ 703 704 /* Number of bytes in reference pictures */ 705 size = num_samples * max_num_bufs; 706 707 /* Account for padding area */ 708 size += ((pad * pad) + pad * (max_wd + min_ht)) * 3 / 2 * max_num_bufs; 709 710 return size; 711 } 712 713 /** 714 ******************************************************************************* 715 * 716 * @brief Returns MV bank buffer size for a given number of luma samples 717 * 718 * @par Description: 719 * For given number of luma samples one MV bank size is computed. 720 * Each MV bank includes pu_map and enc_pu_t for all the min PUs(4x4) in a picture 721 * 722 * @param[in] num_luma_samples 723 * Max number of luma pixels in the frame 724 * 725 * @returns Total MV Bank size 726 * 727 * @remarks 728 * 729 ******************************************************************************* 730 */ 731 WORD32 ih264e_get_pic_mv_bank_size(WORD32 num_luma_samples) 732 { 733 /* mv bank buffer size */ 734 WORD32 mv_bank_size = 0; 735 736 /* number of sub mb partitions possible */ 737 WORD32 num_pu = num_luma_samples / (ENC_MIN_PU_SIZE * ENC_MIN_PU_SIZE); 738 739 /* number of mbs */ 740 WORD32 num_mb = num_luma_samples / (MB_SIZE * MB_SIZE); 741 742 /* Size for storing enc_pu_t start index each MB */ 743 /* One extra entry is needed to compute number of PUs in the last MB */ 744 mv_bank_size += num_mb * sizeof(WORD32); 745 746 /* Size for pu_map */ 747 mv_bank_size += ALIGN4(num_pu); 748 749 /* Size for storing enc_pu_t for each PU */ 750 mv_bank_size += ALIGN4(num_pu * sizeof(enc_pu_t)); 751 752 return mv_bank_size; 753 } 754 755 /** 756 ******************************************************************************* 757 * 758 * @brief 759 * Function to initialize ps_pic_buf structs add pic buffers to 760 * buffer manager in case of non-shared mode 761 * 762 * @par Description: 763 * Function to initialize ps_pic_buf structs add pic buffers to 764 * buffer manager in case of non-shared mode 765 * To be called once per stream or for every reset 766 * 767 * @param[in] ps_codec 768 * Pointer to codec context 769 * 770 * @returns error status 771 * 772 * @remarks 773 * 774 ******************************************************************************* 775 */ 776 IH264E_ERROR_T ih264e_pic_buf_mgr_add_bufs(codec_t *ps_codec) 777 { 778 /* error status */ 779 IH264E_ERROR_T ret = IH264E_SUCCESS; 780 781 /* max ref buffer cnt */ 782 WORD32 max_num_bufs = ps_codec->i4_ref_buf_cnt; 783 784 /* total size for pic buffers */ 785 WORD32 pic_buf_size_allocated = ps_codec->i4_total_pic_buf_size 786 - BUF_MGR_MAX_CNT * sizeof(pic_buf_t); 787 788 /* temp var */ 789 UWORD8 *pu1_buf = (UWORD8 *) ps_codec->ps_pic_buf; 790 pic_buf_t *ps_pic_buf = (pic_buf_t *) ps_codec->ps_pic_buf; 791 WORD32 i; 792 793 pu1_buf += BUF_MGR_MAX_CNT * sizeof(pic_buf_t); 794 795 /* In case of non-shared mode, add picture buffers to buffer manager 796 * In case of shared mode, buffers are added in the run-time 797 */ 798 { 799 WORD32 buf_ret; 800 801 WORD32 luma_samples = (ps_codec->i4_rec_strd) 802 * (ps_codec->s_cfg.u4_ht + PAD_HT); 803 804 WORD32 chroma_samples = luma_samples >> 1; 805 806 /* Try and add as many buffers as possible for the memory that is allocated */ 807 /* If the number of buffers that can be added is less than max_num_bufs 808 * return with an error */ 809 for (i = 0; i < max_num_bufs; i++) 810 { 811 pic_buf_size_allocated -= (luma_samples + chroma_samples); 812 813 if (pic_buf_size_allocated < 0) 814 { 815 ps_codec->i4_error_code = IH264E_INSUFFICIENT_MEM_PICBUF; 816 return IH264E_INSUFFICIENT_MEM_PICBUF; 817 } 818 819 ps_pic_buf->pu1_luma = pu1_buf + ps_codec->i4_rec_strd * PAD_TOP 820 + PAD_LEFT; 821 pu1_buf += luma_samples; 822 823 ps_pic_buf->pu1_chroma = pu1_buf 824 + ps_codec->i4_rec_strd * (PAD_TOP / 2)+ PAD_LEFT; 825 pu1_buf += chroma_samples; 826 827 buf_ret = ih264_buf_mgr_add((buf_mgr_t *) ps_codec->pv_ref_buf_mgr, 828 ps_pic_buf, i); 829 830 if (0 != buf_ret) 831 { 832 ps_codec->i4_error_code = IH264E_BUF_MGR_ERROR; 833 return IH264E_BUF_MGR_ERROR; 834 } 835 pu1_buf += (HPEL_PLANES_CNT - 1) * (chroma_samples + luma_samples); 836 ps_pic_buf++; 837 } 838 } 839 840 return ret; 841 } 842 843 /** 844 ******************************************************************************* 845 * 846 * @brief Function to add buffers to MV Bank buffer manager 847 * 848 * @par Description: 849 * Function to add buffers to MV Bank buffer manager. To be called once per 850 * stream or for every reset 851 * 852 * @param[in] ps_codec 853 * Pointer to codec context 854 * 855 * @returns error status 856 * 857 * @remarks 858 * 859 ******************************************************************************* 860 */ 861 IH264E_ERROR_T ih264e_mv_buf_mgr_add_bufs(codec_t *ps_codec) 862 { 863 /* error status */ 864 IH264E_ERROR_T error_status = IH264E_SUCCESS; 865 IH264_ERROR_T ret; 866 867 /* max dpb size in frames */ 868 WORD32 max_dpb_size = 0; 869 870 /* mv bank size for the entire dpb */ 871 WORD32 mv_bank_size_allocated = 0; 872 873 /* mv bank size per pic */ 874 WORD32 pic_mv_bank_size = 0; 875 876 /* mv buffer ptr */ 877 mv_buf_t *ps_mv_buf = NULL; 878 879 /* num of luma samples */ 880 WORD32 num_luma_samples = ALIGN16(ps_codec->s_cfg.u4_wd) 881 * ALIGN16(ps_codec->s_cfg.u4_ht); 882 883 /* number of mb's & frame partitions */ 884 WORD32 num_pu, num_mb; 885 886 /* temp var */ 887 UWORD8 *pu1_buf = NULL; 888 WORD32 i; 889 890 /* Compute the number of MB Bank buffers needed */ 891 max_dpb_size = ps_codec->i4_ref_buf_cnt; 892 893 /* allocate memory for mv buffer array */ 894 ps_codec->ps_mv_buf = ps_codec->pv_mv_bank_buf_base; 895 pu1_buf = ps_codec->pv_mv_bank_buf_base; 896 pu1_buf += BUF_MGR_MAX_CNT * sizeof(mv_buf_t); 897 898 /********************************************************************/ 899 /* allocate memory for individual elements of mv buffer ptr */ 900 /********************************************************************/ 901 mv_bank_size_allocated = ps_codec->i4_total_mv_bank_size 902 - (BUF_MGR_MAX_CNT * sizeof(mv_buf_t)); 903 904 /* compute MV bank size per picture */ 905 pic_mv_bank_size = ih264e_get_pic_mv_bank_size(num_luma_samples); 906 907 num_pu = num_luma_samples / (ENC_MIN_PU_SIZE * ENC_MIN_PU_SIZE); 908 num_mb = num_luma_samples / (MB_SIZE * MB_SIZE); 909 i = 0; 910 ps_mv_buf = ps_codec->pv_mv_bank_buf_base; 911 912 while (i < max_dpb_size) 913 { 914 mv_bank_size_allocated -= pic_mv_bank_size; 915 916 if (mv_bank_size_allocated < 0) 917 { 918 ps_codec->i4_error_code = IH264E_INSUFFICIENT_MEM_MVBANK; 919 920 error_status = IH264E_INSUFFICIENT_MEM_MVBANK; 921 922 return error_status; 923 } 924 925 ps_mv_buf->pu4_mb_pu_cnt = (UWORD32 *) pu1_buf; 926 pu1_buf += num_mb * sizeof(WORD32); 927 928 ps_mv_buf->pu1_pic_pu_map = pu1_buf; 929 pu1_buf += ALIGN4(num_pu); 930 931 ps_mv_buf->ps_pic_pu = (enc_pu_t *) (pu1_buf); 932 pu1_buf += ALIGN4(num_pu * sizeof(enc_pu_t)); 933 934 ret = ih264_buf_mgr_add((buf_mgr_t *) ps_codec->pv_mv_buf_mgr, 935 ps_mv_buf, i); 936 937 if (IH264_SUCCESS != ret) 938 { 939 ps_codec->i4_error_code = IH264E_BUF_MGR_ERROR; 940 error_status = IH264E_BUF_MGR_ERROR; 941 return error_status; 942 } 943 944 ps_mv_buf++; 945 i++; 946 } 947 948 return error_status; 949 } 950 951 /** 952 ******************************************************************************* 953 * 954 * @brief Function to initialize quant params structure 955 * 956 * @par Description: 957 * The forward quantization modules depends on qp/6, qp mod 6, forward scale 958 * matrix, forward threshold matrix, weight list. The inverse quantization 959 * modules depends on qp/6, qp mod 6, inverse scale matrix, weight list. 960 * These params are initialized in this function. 961 * 962 * @param[in] ps_proc 963 * pointer to process context 964 * 965 * @param[in] qp 966 * quantization parameter 967 * 968 * @returns none 969 * 970 * @remarks 971 * 972 ******************************************************************************* 973 */ 974 void ih264e_init_quant_params(process_ctxt_t *ps_proc, int qp) 975 { 976 /* quant params */ 977 quant_params_t *ps_qp_params; 978 979 /* ptr to forward quant threshold matrix */ 980 const UWORD16 *pu2_thres_mat = NULL; 981 982 /* ptr to forward scale matrix */ 983 const UWORD16 *pu2_scale_mat = gu2_quant_scale_matrix_4x4; 984 985 /* ptr to inverse scale matrix */ 986 const UWORD16 *pu2_iscale_mat = gau2_ih264_iquant_scale_matrix_4x4; 987 988 /* temp var */ 989 UWORD32 u4_qp[3], u4_qp_div6, u4_qp_mod6; 990 COMPONENT_TYPE plane; 991 WORD32 i; 992 UWORD32 u4_satdq_t; 993 const UWORD16 *pu2_smat; 994 995 /********************************************************************/ 996 /* init quant params for all planes Y, U and V */ 997 /********************************************************************/ 998 /* luma qp */ 999 u4_qp[Y] = qp; 1000 1001 /* chroma qp 1002 * TODO_LATER : just in case if the chroma planes use different qp's this 1003 * needs to be corrected accordingly. 1004 */ 1005 u4_qp[U] = gu1_qpc_fqpi[qp]; 1006 u4_qp[V] = gu1_qpc_fqpi[qp]; 1007 1008 plane = Y; 1009 while (plane <= V) 1010 { 1011 u4_qp_div6 = (u4_qp[plane] / 6); 1012 u4_qp_mod6 = (u4_qp[plane] % 6); 1013 1014 ps_qp_params = ps_proc->ps_qp_params[plane]; 1015 1016 /* mb qp */ 1017 ps_qp_params->u1_mb_qp = u4_qp[plane]; 1018 1019 /* mb qp / 6 */ 1020 ps_qp_params->u1_qp_div = u4_qp_div6; 1021 1022 /* mb qp % 6 */ 1023 ps_qp_params->u1_qp_rem = u4_qp_mod6; 1024 1025 /* QP bits */ 1026 ps_qp_params->u1_qbits = QP_BITS_h264_4x4 + u4_qp_div6; 1027 1028 /* forward scale matrix */ 1029 ps_qp_params->pu2_scale_mat = pu2_scale_mat + (u4_qp_mod6 * 16); 1030 1031 /* threshold matrix & weight for quantization */ 1032 pu2_thres_mat = gu2_forward_quant_threshold_4x4 + (u4_qp_mod6 * 16); 1033 for (i = 0; i < 16; i++) 1034 { 1035 ps_qp_params->pu2_thres_mat[i] = pu2_thres_mat[i] 1036 >> (8 - u4_qp_div6); 1037 ps_qp_params->pu2_weigh_mat[i] = 16; 1038 } 1039 1040 /* qp dependent rounding constant */ 1041 ps_qp_params->u4_dead_zone = 1042 gu4_forward_quant_round_factor_4x4[u4_qp_div6]; 1043 1044 /* slice dependent rounding constant */ 1045 if (ps_proc->i4_slice_type != ISLICE 1046 && ps_proc->i4_slice_type != SISLICE) 1047 { 1048 ps_qp_params->u4_dead_zone >>= 1; 1049 } 1050 1051 /* SATQD threshold for zero block prediction */ 1052 if (ps_proc->ps_codec->s_cfg.u4_enable_satqd) 1053 { 1054 pu2_smat = ps_qp_params->pu2_scale_mat; 1055 1056 u4_satdq_t = ((1 << (ps_qp_params->u1_qbits)) - ps_qp_params->u4_dead_zone); 1057 1058 ps_qp_params->pu2_sad_thrsh[0] = u4_satdq_t / MAX(pu2_smat[3], pu2_smat[11]); 1059 ps_qp_params->pu2_sad_thrsh[1] = u4_satdq_t / MAX(pu2_smat[1], pu2_smat[9]); 1060 ps_qp_params->pu2_sad_thrsh[2] = u4_satdq_t / pu2_smat[15]; 1061 ps_qp_params->pu2_sad_thrsh[3] = u4_satdq_t / pu2_smat[7]; 1062 ps_qp_params->pu2_sad_thrsh[4] = u4_satdq_t / MAX(pu2_smat[12], pu2_smat[14]); 1063 ps_qp_params->pu2_sad_thrsh[5] = u4_satdq_t / MAX(pu2_smat[4], pu2_smat[6]); 1064 ps_qp_params->pu2_sad_thrsh[6] = u4_satdq_t / pu2_smat[13]; 1065 ps_qp_params->pu2_sad_thrsh[7] = u4_satdq_t / pu2_smat[5]; 1066 ps_qp_params->pu2_sad_thrsh[8] = u4_satdq_t / MAX(MAX3(pu2_smat[0], pu2_smat[2], pu2_smat[8]), pu2_smat[10]); 1067 } 1068 1069 /* inverse scale matrix */ 1070 ps_qp_params->pu2_iscale_mat = pu2_iscale_mat + (u4_qp_mod6 * 16); 1071 1072 plane += 1; 1073 } 1074 return ; 1075 } 1076 1077 /** 1078 ******************************************************************************* 1079 * 1080 * @brief 1081 * Initialize AIR mb frame Map 1082 * 1083 * @par Description: 1084 * Initialize AIR mb frame map 1085 * MB frame map indicates which frame an Mb should be coded as intra according to AIR 1086 * 1087 * @param[in] ps_codec 1088 * Pointer to codec context 1089 * 1090 * @returns error_status 1091 * 1092 * @remarks 1093 * 1094 * 1095 ******************************************************************************* 1096 */ 1097 IH264E_ERROR_T ih264e_init_air_map(codec_t *ps_codec) 1098 { 1099 /* intra refresh map */ 1100 UWORD16 *pu2_intr_rfrsh_map = ps_codec->pu2_intr_rfrsh_map; 1101 1102 /* air mode */ 1103 IVE_AIR_MODE_T air_mode = ps_codec->s_cfg.e_air_mode; 1104 1105 /* refresh period */ 1106 UWORD32 air_period = ps_codec->s_cfg.u4_air_refresh_period; 1107 1108 /* mb cnt */ 1109 UWORD32 u4_mb_cnt = ps_codec->s_cfg.i4_wd_mbs * ps_codec->s_cfg.i4_ht_mbs; 1110 1111 /* temp var */ 1112 UWORD32 curr_mb, seed_rand = 1; 1113 1114 switch (air_mode) 1115 { 1116 case IVE_AIR_MODE_CYCLIC: 1117 1118 for (curr_mb = 0; curr_mb < u4_mb_cnt; curr_mb++) 1119 { 1120 pu2_intr_rfrsh_map[curr_mb] = curr_mb % air_period; 1121 } 1122 break; 1123 1124 case IVE_AIR_MODE_RANDOM: 1125 1126 for (curr_mb = 0; curr_mb < u4_mb_cnt; curr_mb++) 1127 { 1128 seed_rand = (seed_rand * 32719 + 3) % 32749; 1129 pu2_intr_rfrsh_map[curr_mb] = seed_rand % air_period; 1130 } 1131 break; 1132 1133 default: 1134 1135 break; 1136 } 1137 1138 return IH264E_SUCCESS; 1139 } 1140 1141 /** 1142 ******************************************************************************* 1143 * 1144 * @brief 1145 * Codec level initializations 1146 * 1147 * @par Description: 1148 * Initializes the codec with parameters that needs to be set before encoding 1149 * first frame 1150 * 1151 * @param[in] ps_codec 1152 * Pointer to codec context 1153 * 1154 * @param[in] ps_inp_buf 1155 * Pointer to input buffer context 1156 * 1157 * @returns error_status 1158 * 1159 * @remarks 1160 * 1161 * 1162 ******************************************************************************* 1163 */ 1164 IH264E_ERROR_T ih264e_codec_init(codec_t *ps_codec) 1165 { 1166 /******************************************************************** 1167 * INITIALIZE CODEC CONTEXT * 1168 ********************************************************************/ 1169 /* encoder presets */ 1170 if (ps_codec->s_cfg.u4_enc_speed_preset != IVE_CONFIG) 1171 { 1172 if (ps_codec->s_cfg.u4_enc_speed_preset == IVE_SLOWEST) 1173 {/* high quality */ 1174 /* enable diamond search */ 1175 ps_codec->s_cfg.u4_me_speed_preset = DMND_SRCH; 1176 ps_codec->s_cfg.u4_enable_fast_sad = 0; 1177 1178 /* disable intra 4x4 */ 1179 ps_codec->s_cfg.u4_enable_intra_4x4 = 1; 1180 ps_codec->luma_energy_compaction[1] = 1181 ih264e_code_luma_intra_macroblock_4x4_rdopt_on; 1182 1183 /* sub pel off */ 1184 ps_codec->s_cfg.u4_enable_hpel = 1; 1185 1186 /* deblocking off */ 1187 ps_codec->s_cfg.u4_disable_deblock_level = DISABLE_DEBLK_LEVEL_0; 1188 1189 /* disabled intra inter gating in Inter slices */ 1190 ps_codec->u4_inter_gate = 0; 1191 } 1192 else if (ps_codec->s_cfg.u4_enc_speed_preset == IVE_NORMAL) 1193 {/* normal */ 1194 /* enable diamond search */ 1195 ps_codec->s_cfg.u4_me_speed_preset = DMND_SRCH; 1196 ps_codec->s_cfg.u4_enable_fast_sad = 0; 1197 1198 /* disable intra 4x4 */ 1199 ps_codec->s_cfg.u4_enable_intra_4x4 = 1; 1200 1201 /* sub pel off */ 1202 ps_codec->s_cfg.u4_enable_hpel = 1; 1203 1204 /* deblocking off */ 1205 ps_codec->s_cfg.u4_disable_deblock_level = DISABLE_DEBLK_LEVEL_0; 1206 1207 /* disabled intra inter gating in Inter slices */ 1208 ps_codec->u4_inter_gate = 0; 1209 } 1210 else if (ps_codec->s_cfg.u4_enc_speed_preset == IVE_FAST) 1211 {/* normal */ 1212 /* enable diamond search */ 1213 ps_codec->s_cfg.u4_me_speed_preset = DMND_SRCH; 1214 ps_codec->s_cfg.u4_enable_fast_sad = 0; 1215 1216 /* disable intra 4x4 */ 1217 ps_codec->s_cfg.u4_enable_intra_4x4 = 0; 1218 1219 /* sub pel off */ 1220 ps_codec->s_cfg.u4_enable_hpel = 1; 1221 1222 /* deblocking off */ 1223 ps_codec->s_cfg.u4_disable_deblock_level = DISABLE_DEBLK_LEVEL_0; 1224 1225 /* disabled intra inter gating in Inter slices */ 1226 ps_codec->u4_inter_gate = 1; 1227 } 1228 else if (ps_codec->s_cfg.u4_enc_speed_preset == IVE_HIGH_SPEED) 1229 {/* fast */ 1230 /* enable diamond search */ 1231 ps_codec->s_cfg.u4_me_speed_preset = DMND_SRCH; 1232 ps_codec->s_cfg.u4_enable_fast_sad = 0; 1233 1234 /* disable intra 4x4 */ 1235 ps_codec->s_cfg.u4_enable_intra_4x4 = 0; 1236 1237 /* sub pel off */ 1238 ps_codec->s_cfg.u4_enable_hpel = 0; 1239 1240 /* deblocking off */ 1241 ps_codec->s_cfg.u4_disable_deblock_level = DISABLE_DEBLK_LEVEL_4; 1242 1243 /* disabled intra inter gating in Inter slices */ 1244 ps_codec->u4_inter_gate = 0; 1245 } 1246 else if (ps_codec->s_cfg.u4_enc_speed_preset == IVE_FASTEST) 1247 {/* fastest */ 1248 /* enable diamond search */ 1249 ps_codec->s_cfg.u4_me_speed_preset = DMND_SRCH; 1250 1251 /* disable intra 4x4 */ 1252 ps_codec->s_cfg.u4_enable_intra_4x4 = 0; 1253 1254 /* sub pel off */ 1255 ps_codec->s_cfg.u4_enable_hpel = 0; 1256 1257 /* deblocking off */ 1258 ps_codec->s_cfg.u4_disable_deblock_level = DISABLE_DEBLK_LEVEL_4; 1259 1260 /* disabled intra inter gating in Inter slices */ 1261 ps_codec->u4_inter_gate = 1; 1262 } 1263 } 1264 1265 /***************************************************************** 1266 * Initialize AIR inside codec 1267 *****************************************************************/ 1268 if (IVE_AIR_MODE_NONE != ps_codec->s_cfg.e_air_mode) 1269 { 1270 ih264e_init_air_map(ps_codec); 1271 1272 ps_codec->i4_air_pic_cnt = -1; 1273 } 1274 1275 /****************************************************/ 1276 /* INITIALIZE RATE CONTROL */ 1277 /****************************************************/ 1278 { 1279 /* init qp */ 1280 UWORD8 au1_init_qp[MAX_PIC_TYPE]; 1281 1282 /* min max qp */ 1283 UWORD8 au1_min_max_qp[2 * MAX_PIC_TYPE]; 1284 1285 /* init i,p,b qp */ 1286 au1_init_qp[0] = gau1_h264_to_mpeg2_qmap[ps_codec->s_cfg.u4_i_qp]; 1287 au1_init_qp[1] = gau1_h264_to_mpeg2_qmap[ps_codec->s_cfg.u4_p_qp]; 1288 au1_init_qp[2] = gau1_h264_to_mpeg2_qmap[ps_codec->s_cfg.u4_b_qp]; 1289 1290 /* init min max qp */ 1291 au1_min_max_qp[2 * I_PIC] = 1292 gau1_h264_to_mpeg2_qmap[ps_codec->s_cfg.u4_i_qp_min]; 1293 au1_min_max_qp[2 * I_PIC + 1] = 1294 gau1_h264_to_mpeg2_qmap[ps_codec->s_cfg.u4_i_qp_max]; 1295 1296 au1_min_max_qp[2 * P_PIC] = 1297 gau1_h264_to_mpeg2_qmap[ps_codec->s_cfg.u4_p_qp_min]; 1298 au1_min_max_qp[2 * P_PIC + 1] = 1299 gau1_h264_to_mpeg2_qmap[ps_codec->s_cfg.u4_p_qp_max]; 1300 1301 au1_min_max_qp[2 * B_PIC] = 1302 gau1_h264_to_mpeg2_qmap[ps_codec->s_cfg.u4_b_qp_min]; 1303 au1_min_max_qp[2 * B_PIC + 1] = 1304 gau1_h264_to_mpeg2_qmap[ps_codec->s_cfg.u4_b_qp_max]; 1305 1306 /* get rc mode */ 1307 switch (ps_codec->s_cfg.e_rc_mode) 1308 { 1309 case IVE_RC_STORAGE: 1310 ps_codec->s_rate_control.e_rc_type = VBR_STORAGE; 1311 break; 1312 case IVE_RC_CBR_NON_LOW_DELAY: 1313 ps_codec->s_rate_control.e_rc_type = CBR_NLDRC; 1314 break; 1315 case IVE_RC_CBR_LOW_DELAY: 1316 ps_codec->s_rate_control.e_rc_type = CBR_LDRC; 1317 break; 1318 case IVE_RC_NONE: 1319 ps_codec->s_rate_control.e_rc_type = CONST_QP; 1320 break; 1321 default: 1322 break; 1323 } 1324 1325 /* init rate control */ 1326 ih264e_rc_init(ps_codec->s_rate_control.pps_rate_control_api, 1327 ps_codec->s_rate_control.pps_frame_time, 1328 ps_codec->s_rate_control.pps_time_stamp, 1329 ps_codec->s_rate_control.pps_pd_frm_rate, 1330 ps_codec->s_cfg.u4_max_framerate, 1331 ps_codec->s_cfg.u4_src_frame_rate, 1332 ps_codec->s_cfg.u4_tgt_frame_rate, 1333 ps_codec->s_rate_control.e_rc_type, 1334 ps_codec->s_cfg.u4_target_bitrate, 1335 ps_codec->s_cfg.u4_max_bitrate, 1336 ps_codec->s_cfg.u4_vbv_buffer_delay, 1337 ps_codec->s_cfg.u4_i_frm_interval, 1338 ps_codec->s_cfg.u4_num_bframes + 1, au1_init_qp, 1339 ps_codec->s_cfg.u4_num_bframes + 2 , au1_min_max_qp, 1340 MAX(ps_codec->s_cfg.u4_max_level, 1341 (UWORD32)ih264e_get_min_level(ps_codec->s_cfg.u4_max_wd, ps_codec->s_cfg.u4_max_ht))); 1342 } 1343 1344 /* recon stride */ 1345 ps_codec->i4_rec_strd = ALIGN16(ps_codec->s_cfg.u4_max_wd) + PAD_WD; 1346 1347 /* max ref and reorder cnt */ 1348 ps_codec->i4_ref_buf_cnt = ps_codec->s_cfg.u4_max_ref_cnt 1349 + ps_codec->s_cfg.u4_max_reorder_cnt; 1350 ps_codec->i4_ref_buf_cnt += MAX_CTXT_SETS; 1351 1352 DEBUG_HISTOGRAM_INIT(); 1353 1354 1355 /* Init dependecy vars */ 1356 ps_codec->i4_last_inp_buff_received = 0; 1357 1358 /* At codec start no IDR is pending */ 1359 ps_codec->i4_pending_idr_flag = 0; 1360 1361 return IH264E_SUCCESS; 1362 } 1363 1364 /** 1365 ******************************************************************************* 1366 * 1367 * @brief 1368 * Picture level initializations 1369 * 1370 * @par Description: 1371 * Before beginning to encode the frame, the current function initializes all 1372 * the ctxts (proc, entropy, me, ...) basing on the input configured params. 1373 * It locates space for storing recon in the encoder picture buffer set, fetches 1374 * reference frame from encoder picture buffer set. Calls RC pre-enc to get 1375 * qp and pic type for the current frame. Queues proc jobs so that 1376 * the other threads can begin encoding. In brief, this function sets up the 1377 * tone for the entire encoder. 1378 * 1379 * @param[in] ps_codec 1380 * Pointer to codec context 1381 * 1382 * @param[in] ps_inp_buf 1383 * Pointer to input buffer context 1384 * 1385 * @returns error_status 1386 * 1387 * @remarks 1388 * 1389 * 1390 ******************************************************************************* 1391 */ 1392 IH264E_ERROR_T ih264e_pic_init(codec_t *ps_codec, inp_buf_t *ps_inp_buf) 1393 { 1394 /* error status */ 1395 IH264E_ERROR_T error_status = IH264E_SUCCESS; 1396 IH264_ERROR_T ret = IH264_SUCCESS; 1397 1398 /* mv buff bank */ 1399 mv_buf_t *ps_mv_buf = NULL; 1400 WORD32 cur_mv_bank_buf_id; 1401 1402 /* recon buffer set */ 1403 pic_buf_t *ps_cur_pic; 1404 WORD32 cur_pic_buf_id; 1405 UWORD8 *pu1_cur_pic_luma, *pu1_cur_pic_chroma; 1406 1407 /* ref buffer set */ 1408 pic_buf_t *aps_ref_pic[MAX_REF_PIC_CNT] = {NULL, NULL}; 1409 mv_buf_t *aps_mv_buf[MAX_REF_PIC_CNT] = {NULL, NULL}; 1410 WORD32 ref_set_id; 1411 1412 /* pic time stamp */ 1413 UWORD32 u4_timestamp_high = ps_inp_buf->u4_timestamp_high; 1414 UWORD32 u4_timestamp_low = ps_inp_buf->u4_timestamp_low; 1415 1416 /* indices to access curr/prev frame info */ 1417 WORD32 ctxt_sel = ps_codec->i4_encode_api_call_cnt % MAX_CTXT_SETS; 1418 1419 /* curr pic type */ 1420 PIC_TYPE_T *pic_type = &ps_codec->pic_type; 1421 1422 /* Diamond search Iteration Max Cnt */ 1423 UWORD32 u4_num_layers = 1424 (ps_codec->s_cfg.u4_enc_speed_preset == IVE_FASTEST) ? 1425 (NUM_LAYERS >> 2) : NUM_LAYERS; 1426 1427 /* enable fast sad */ 1428 UWORD32 u4_enable_fast_sad = ps_codec->s_cfg.u4_enable_fast_sad; 1429 1430 /********************************************************************/ 1431 /* INITIALIZE CODEC CONTEXT */ 1432 /********************************************************************/ 1433 /* slice_type */ 1434 if ((PIC_I == *pic_type) || (PIC_IDR == *pic_type)) 1435 { 1436 ps_codec->i4_slice_type = ISLICE; 1437 } 1438 else if (PIC_P == *pic_type) 1439 { 1440 ps_codec->i4_slice_type = PSLICE; 1441 } 1442 else if(PIC_B == *pic_type) 1443 { 1444 ps_codec->i4_slice_type = BSLICE; 1445 } 1446 1447 1448 /*************************************************************************** 1449 * Set up variables for sending frame number, poc and reference 1450 * a) Set up alt ref too 1451 **************************************************************************/ 1452 1453 /* Check and set if the current frame is reference or not */ 1454 ps_codec->u4_is_curr_frm_ref = 0; 1455 1456 /* This frame is reference if its not a B pic, pending approval from alt ref */ 1457 ps_codec->u4_is_curr_frm_ref = (*pic_type != PIC_B); 1458 1459 /* In case if its a P pic, we will decide according to alt ref also */ 1460 if (ps_codec->s_cfg.u4_enable_alt_ref && (*pic_type == PIC_P) 1461 && (ps_codec->i4_pic_cnt 1462 % (ps_codec->s_cfg.u4_enable_alt_ref + 1))) 1463 { 1464 ps_codec->u4_is_curr_frm_ref = 0; 1465 } 1466 1467 /* 1468 * Override everything in case of IDR 1469 * Note that in case of IDR, at this point ps_codec->u4_is_curr_frm_ref must 1470 * be 1 1471 */ 1472 1473 /* is this an IDR pic */ 1474 ps_codec->u4_is_idr = 0; 1475 1476 if (PIC_IDR == *pic_type) 1477 { 1478 /* set idr flag */ 1479 ps_codec->u4_is_idr = 1; 1480 1481 /* reset frame num */ 1482 ps_codec->i4_frame_num = 0; 1483 1484 /* idr_pic_id */ 1485 ps_codec->i4_idr_pic_id++; 1486 } 1487 1488 /*************************************************************************** 1489 * Set up Deblock 1490 **************************************************************************/ 1491 1492 /* set deblock disable flags based on disable deblock level */ 1493 ps_codec->i4_disable_deblk_pic = 1; 1494 1495 if (ps_codec->s_cfg.u4_disable_deblock_level == DISABLE_DEBLK_LEVEL_0) 1496 { 1497 /* enable deblocking */ 1498 ps_codec->i4_disable_deblk_pic = 0; 1499 } 1500 else if (ps_codec->s_cfg.u4_disable_deblock_level == DISABLE_DEBLK_LEVEL_2) 1501 { 1502 /* enable deblocking after a period of frames */ 1503 if (ps_codec->i4_disable_deblk_pic_cnt == DISABLE_DEBLOCK_INTERVAL 1504 || ps_codec->i4_slice_type == ISLICE) 1505 { 1506 ps_codec->i4_disable_deblk_pic = 0; 1507 } 1508 } 1509 else if (ps_codec->s_cfg.u4_disable_deblock_level == DISABLE_DEBLK_LEVEL_3) 1510 { 1511 if (ps_codec->i4_slice_type == ISLICE) 1512 { 1513 ps_codec->i4_disable_deblk_pic = 0; 1514 } 1515 } 1516 1517 if (ps_codec->i4_disable_deblk_pic) 1518 { 1519 ps_codec->i4_disable_deblk_pic_cnt++; 1520 } 1521 else 1522 { 1523 ps_codec->i4_disable_deblk_pic_cnt = 0; 1524 } 1525 1526 /* In slice mode - lets not deblk mb edges that lie along slice boundaries */ 1527 if (ps_codec->i4_disable_deblk_pic == 0) 1528 { 1529 if (ps_codec->s_cfg.e_slice_mode != IVE_SLICE_MODE_NONE) 1530 { 1531 ps_codec->i4_disable_deblk_pic = 2; 1532 } 1533 } 1534 1535 /* error status */ 1536 ps_codec->i4_error_code = IH264E_SUCCESS; 1537 1538 /* populate header */ 1539 if (ps_codec->i4_gen_header) 1540 { 1541 /* sps */ 1542 sps_t *ps_sps = NULL; 1543 1544 /* pps */ 1545 pps_t *ps_pps = NULL; 1546 1547 /*ps_codec->i4_pps_id ++;*/ 1548 ps_codec->i4_pps_id %= MAX_PPS_CNT; 1549 1550 /*ps_codec->i4_sps_id ++;*/ 1551 ps_codec->i4_sps_id %= MAX_SPS_CNT; 1552 1553 /* populate sps header */ 1554 ps_sps = ps_codec->ps_sps_base + ps_codec->i4_sps_id; 1555 ih264e_populate_sps(ps_codec, ps_sps); 1556 1557 /* populate pps header */ 1558 ps_pps = ps_codec->ps_pps_base + ps_codec->i4_pps_id; 1559 ih264e_populate_pps(ps_codec, ps_pps); 1560 } 1561 1562 /*************************************************************************** 1563 * Reference and MV bank Buffer Manager 1564 * Here we will 1565 * 1) Find the correct ref pics for the current frame 1566 * 2) Free the ref pic that is not going to be used anywhere 1567 * 3) Find a free buff from the list and assign it as the recon of 1568 * current frame 1569 * 1570 * 1) Finding correct ref pic 1571 * All pics needed for future are arranged in a picture list called 1572 * ps_codec->as_ref_set. Each picture in this will have a pic buffer and 1573 * MV buffer that is marked appropriately as BUF_MGR_REF, BUF_MGR_IO or 1574 * BUF_MGR_CODEC. Also the pic_cnt and poc will also be present. 1575 * Hence to find the ref pic we will loop through the list and find 1576 * 2 pictures with maximum i4_pic_cnt . 1577 * 1578 * note that i4_pic_cnt == -1 is used to filter uninit ref pics. 1579 * Now since we only have max two ref pics, we will always find max 2 1580 * ref pics. 1581 1582 * 1583 * 2) 3) Self explanatory 1584 ***************************************************************************/ 1585 { 1586 /* Search for buffs with maximum pic cnt */ 1587 1588 WORD32 max_pic_cnt[] = { -1, -1 }; 1589 1590 mv_buf_t *ps_mv_buf_to_free[] = { NULL, NULL }; 1591 1592 /* temp var */ 1593 WORD32 i, buf_status; 1594 1595 for (i = 0; i < ps_codec->i4_ref_buf_cnt; i++) 1596 { 1597 if (ps_codec->as_ref_set[i].i4_pic_cnt == -1) 1598 continue; 1599 1600 buf_status = ih264_buf_mgr_get_status( 1601 ps_codec->pv_ref_buf_mgr, 1602 ps_codec->as_ref_set[i].ps_pic_buf->i4_buf_id); 1603 1604 /* Ideally we should look for buffer status of MV BUFF also. But since 1605 * the correponding MV buffs also will be at the same state. It dosent 1606 * matter as of now. But the check will make the logic better */ 1607 if ((max_pic_cnt[0] < ps_codec->as_ref_set[i].i4_pic_cnt) 1608 && (buf_status & BUF_MGR_REF)) 1609 { 1610 if (max_pic_cnt[1] < ps_codec->as_ref_set[i].i4_pic_cnt) 1611 { 1612 max_pic_cnt[0] = max_pic_cnt[1]; 1613 aps_ref_pic[0] = aps_ref_pic[1]; 1614 aps_mv_buf[0] = aps_mv_buf[1]; 1615 1616 ps_mv_buf_to_free[0] = ps_mv_buf_to_free[1]; 1617 1618 max_pic_cnt[1] = ps_codec->as_ref_set[i].i4_pic_cnt; 1619 aps_ref_pic[1] = ps_codec->as_ref_set[i].ps_pic_buf; 1620 aps_mv_buf[1] = ps_codec->as_ref_set[i].ps_mv_buf; 1621 ps_mv_buf_to_free[1] = ps_codec->as_ref_set[i].ps_mv_buf; 1622 1623 } 1624 else 1625 { 1626 max_pic_cnt[0] = ps_codec->as_ref_set[i].i4_pic_cnt; 1627 aps_ref_pic[0] = ps_codec->as_ref_set[i].ps_pic_buf; 1628 aps_mv_buf[0] = ps_codec->as_ref_set[i].ps_mv_buf; 1629 ps_mv_buf_to_free[0] = ps_codec->as_ref_set[i].ps_mv_buf; 1630 } 1631 } 1632 } 1633 1634 /* 1635 * Now if the current picture is I or P, we discard the back ref pic and 1636 * assign forward ref as backward ref 1637 */ 1638 if (*pic_type != PIC_B) 1639 { 1640 if (ps_mv_buf_to_free[0]) 1641 { 1642 /* release this frame from reference list */ 1643 ih264_buf_mgr_release(ps_codec->pv_mv_buf_mgr, 1644 ps_mv_buf_to_free[0]->i4_buf_id, 1645 BUF_MGR_REF); 1646 1647 ih264_buf_mgr_release(ps_codec->pv_ref_buf_mgr, 1648 aps_ref_pic[0]->i4_buf_id, BUF_MGR_REF); 1649 } 1650 1651 max_pic_cnt[0] = max_pic_cnt[1]; 1652 aps_ref_pic[0] = aps_ref_pic[1]; 1653 aps_mv_buf[0] = aps_mv_buf[1]; 1654 1655 /* Dummy */ 1656 max_pic_cnt[1] = -1; 1657 } 1658 1659 /* 1660 * Mark all reference pic with unused buffers to be free 1661 * We need this step since each one, ie ref, recon io etc only unset their 1662 * respective flags. Hence we need to combine togather and mark the ref set 1663 * accordingly 1664 */ 1665 ref_set_id = -1; 1666 for (i = 0; i < ps_codec->i4_ref_buf_cnt; i++) 1667 { 1668 if (ps_codec->as_ref_set[i].i4_pic_cnt == -1) 1669 { 1670 ref_set_id = i; 1671 continue; 1672 } 1673 1674 buf_status = ih264_buf_mgr_get_status( 1675 ps_codec->pv_ref_buf_mgr, 1676 ps_codec->as_ref_set[i].ps_pic_buf->i4_buf_id); 1677 1678 if ((buf_status & (BUF_MGR_REF | BUF_MGR_CODEC | BUF_MGR_IO)) == 0) 1679 { 1680 ps_codec->as_ref_set[i].i4_pic_cnt = -1; 1681 ps_codec->as_ref_set[i].i4_poc = 32768; 1682 1683 ref_set_id = i; 1684 } 1685 } 1686 /* An asssert failure here means we donot have any free buffs */ 1687 ASSERT(ref_set_id >= 0); 1688 } 1689 1690 { 1691 /*****************************************************************/ 1692 /* Get free MV Bank to hold current picture's motion vector data */ 1693 /* If there are no free buffers then return with an error code. */ 1694 /* If the buffer is to be freed by another thread, change the */ 1695 /* following to call thread yield and wait for buffer to be freed*/ 1696 /*****************************************************************/ 1697 ps_mv_buf = (mv_buf_t *) ih264_buf_mgr_get_next_free( 1698 (buf_mgr_t *) ps_codec->pv_mv_buf_mgr, 1699 &cur_mv_bank_buf_id); 1700 1701 if (NULL == ps_mv_buf) 1702 { 1703 ps_codec->i4_error_code = IH264E_NO_FREE_MVBANK; 1704 return IH264E_NO_FREE_MVBANK; 1705 } 1706 1707 /* mark the buffer as needed for reference if the curr pic is available for ref */ 1708 if (ps_codec->u4_is_curr_frm_ref) 1709 { 1710 ih264_buf_mgr_set_status(ps_codec->pv_mv_buf_mgr, 1711 cur_mv_bank_buf_id, BUF_MGR_REF); 1712 } 1713 1714 /* Set current ABS poc to ps_mv_buf, so that while freeing a reference buffer 1715 * corresponding mv buffer can be found by looping through ps_codec->ps_mv_buf array 1716 * and getting a buffer id to free 1717 */ 1718 ps_mv_buf->i4_abs_poc = ps_codec->i4_abs_pic_order_cnt; 1719 ps_mv_buf->i4_buf_id = cur_mv_bank_buf_id; 1720 } 1721 1722 { 1723 /*****************************************************************/ 1724 /* Get free pic buf to hold current picture's recon data */ 1725 /* If there are no free buffers then return with an error code. */ 1726 /* If the buffer is to be freed by another thread, change the */ 1727 /* following to call thread yield and wait for buffer to be freed*/ 1728 /*****************************************************************/ 1729 ps_cur_pic = (pic_buf_t *) ih264_buf_mgr_get_next_free( 1730 (buf_mgr_t *) ps_codec->pv_ref_buf_mgr, 1731 &cur_pic_buf_id); 1732 1733 if (NULL == ps_cur_pic) 1734 { 1735 ps_codec->i4_error_code = IH264E_NO_FREE_PICBUF; 1736 return IH264E_NO_FREE_PICBUF; 1737 } 1738 1739 /* mark the buffer as needed for reference if the curr pic is available for ref */ 1740 if (ps_codec->u4_is_curr_frm_ref) 1741 { 1742 ih264_buf_mgr_set_status(ps_codec->pv_ref_buf_mgr, cur_pic_buf_id, 1743 BUF_MGR_REF); 1744 } 1745 1746 /* Mark the current buffer as needed for IO if recon is enabled */ 1747 if (1 == ps_codec->s_cfg.u4_enable_recon) 1748 { 1749 ih264_buf_mgr_set_status(ps_codec->pv_ref_buf_mgr, cur_pic_buf_id, 1750 BUF_MGR_IO); 1751 } 1752 1753 /* Associate input timestamp with current buffer */ 1754 ps_cur_pic->u4_timestamp_high = ps_inp_buf->u4_timestamp_high; 1755 ps_cur_pic->u4_timestamp_low = ps_inp_buf->u4_timestamp_low; 1756 1757 ps_cur_pic->i4_abs_poc = ps_codec->i4_poc; 1758 ps_cur_pic->i4_poc_lsb = ps_codec->i4_pic_order_cnt_lsb; 1759 1760 ps_cur_pic->i4_buf_id = cur_pic_buf_id; 1761 1762 pu1_cur_pic_luma = ps_cur_pic->pu1_luma; 1763 pu1_cur_pic_chroma = ps_cur_pic->pu1_chroma; 1764 } 1765 1766 /* 1767 * Add the current picture to ref list independent of the fact that it is used 1768 * as reference or not. This is because, now recon is not in sync with output 1769 * hence we may need the current recon after some delay. By adding it to ref list 1770 * we can retrieve the recon any time we want. The information that it is used 1771 * for ref can still be found by checking the buffer status of pic buf. 1772 */ 1773 { 1774 ps_codec->as_ref_set[ref_set_id].i4_pic_cnt = ps_codec->i4_pic_cnt; 1775 ps_codec->as_ref_set[ref_set_id].i4_poc = ps_codec->i4_poc; 1776 ps_codec->as_ref_set[ref_set_id].ps_mv_buf = ps_mv_buf; 1777 ps_codec->as_ref_set[ref_set_id].ps_pic_buf = ps_cur_pic; 1778 } 1779 1780 /********************************************************************/ 1781 /* INITIALIZE PROCESS CONTEXT */ 1782 /********************************************************************/ 1783 { 1784 /* temp var */ 1785 WORD32 i, j = 0; 1786 1787 /* curr proc ctxt */ 1788 process_ctxt_t *ps_proc = NULL; 1789 1790 j = ctxt_sel * MAX_PROCESS_THREADS; 1791 1792 /* begin init */ 1793 for (i = j; i < (j + MAX_PROCESS_THREADS); i++) 1794 { 1795 ps_proc = &ps_codec->as_process[i]; 1796 1797 /* luma src buffer */ 1798 if (ps_codec->s_cfg.e_inp_color_fmt == IV_YUV_422ILE) 1799 { 1800 ps_proc->pu1_src_buf_luma_base = ps_codec->pu1_y_csc_buf_base; 1801 } 1802 else 1803 { 1804 ps_proc->pu1_src_buf_luma_base = 1805 ps_inp_buf->s_raw_buf.apv_bufs[0]; 1806 } 1807 1808 /* chroma src buffer */ 1809 if (ps_codec->s_cfg.e_inp_color_fmt == IV_YUV_422ILE 1810 || ps_codec->s_cfg.e_inp_color_fmt == IV_YUV_420P) 1811 { 1812 ps_proc->pu1_src_buf_chroma_base = 1813 ps_codec->pu1_uv_csc_buf_base; 1814 } 1815 else 1816 { 1817 ps_proc->pu1_src_buf_chroma_base = 1818 ps_inp_buf->s_raw_buf.apv_bufs[1]; 1819 } 1820 1821 /* luma rec buffer */ 1822 ps_proc->pu1_rec_buf_luma_base = pu1_cur_pic_luma; 1823 1824 /* chroma rec buffer */ 1825 ps_proc->pu1_rec_buf_chroma_base = pu1_cur_pic_chroma; 1826 1827 /* rec stride */ 1828 ps_proc->i4_rec_strd = ps_codec->i4_rec_strd; 1829 1830 /* frame num */ 1831 ps_proc->i4_frame_num = ps_codec->i4_frame_num; 1832 1833 /* is idr */ 1834 ps_proc->u4_is_idr = ps_codec->u4_is_idr; 1835 1836 /* idr pic id */ 1837 ps_proc->u4_idr_pic_id = ps_codec->i4_idr_pic_id; 1838 1839 /* slice_type */ 1840 ps_proc->i4_slice_type = ps_codec->i4_slice_type; 1841 1842 /* Input width in mbs */ 1843 ps_proc->i4_wd_mbs = ps_codec->s_cfg.i4_wd_mbs; 1844 1845 /* Input height in mbs */ 1846 ps_proc->i4_ht_mbs = ps_codec->s_cfg.i4_ht_mbs; 1847 1848 /* Half x plane offset from pic buf */ 1849 ps_proc->u4_half_x_offset = 0; 1850 1851 /* Half y plane offset from half x plane */ 1852 ps_proc->u4_half_y_offset = 0; 1853 1854 /* Half x plane offset from half y plane */ 1855 ps_proc->u4_half_xy_offset = 0; 1856 1857 /* top row syntax elements */ 1858 ps_proc->ps_top_row_mb_syntax_ele = 1859 ps_proc->ps_top_row_mb_syntax_ele_base; 1860 1861 ps_proc->pu1_top_mb_intra_modes = 1862 ps_proc->pu1_top_mb_intra_modes_base; 1863 1864 ps_proc->ps_top_row_pu = ps_proc->ps_top_row_pu_base; 1865 1866 /* initialize quant params */ 1867 ps_proc->u4_frame_qp = ps_codec->u4_frame_qp; 1868 ps_proc->u4_mb_qp = ps_codec->u4_frame_qp; 1869 ih264e_init_quant_params(ps_proc, ps_proc->u4_frame_qp); 1870 1871 /* previous mb qp*/ 1872 ps_proc->u4_mb_qp_prev = ps_proc->u4_frame_qp; 1873 1874 /* Reset frame info */ 1875 memset(&ps_proc->s_frame_info, 0, sizeof(frame_info_t)); 1876 1877 /* initialize proc, deblk and ME map */ 1878 if (i == j) 1879 { 1880 /* row '-1' */ 1881 memset(ps_proc->pu1_proc_map - ps_proc->i4_wd_mbs, 1, ps_proc->i4_wd_mbs); 1882 /* row 0 to ht in mbs */ 1883 memset(ps_proc->pu1_proc_map, 0, ps_proc->i4_wd_mbs * ps_proc->i4_ht_mbs); 1884 1885 /* row '-1' */ 1886 memset(ps_proc->pu1_deblk_map - ps_proc->i4_wd_mbs, 1, ps_proc->i4_wd_mbs); 1887 /* row 0 to ht in mbs */ 1888 memset(ps_proc->pu1_deblk_map, 0, ps_proc->i4_wd_mbs * ps_proc->i4_ht_mbs); 1889 1890 /* row '-1' */ 1891 memset(ps_proc->pu1_me_map - ps_proc->i4_wd_mbs, 1, ps_proc->i4_wd_mbs); 1892 /* row 0 to ht in mbs */ 1893 memset(ps_proc->pu1_me_map, 0, ps_proc->i4_wd_mbs * ps_proc->i4_ht_mbs); 1894 1895 /* at the start of air refresh period, reset intra coded map */ 1896 if (IVE_AIR_MODE_NONE != ps_codec->s_cfg.e_air_mode) 1897 { 1898 ps_codec->i4_air_pic_cnt = (ps_codec->i4_air_pic_cnt + 1) 1899 % ps_codec->s_cfg.u4_air_refresh_period; 1900 1901 if (!ps_codec->i4_air_pic_cnt) 1902 { 1903 memset(ps_proc->pu1_is_intra_coded, 0, ps_proc->i4_wd_mbs * ps_proc->i4_ht_mbs); 1904 } 1905 } 1906 } 1907 1908 /* deblock level */ 1909 ps_proc->u4_disable_deblock_level = ps_codec->i4_disable_deblk_pic; 1910 1911 /* slice index map */ 1912 /* no slice */ 1913 if (ps_codec->s_cfg.e_slice_mode == IVE_SLICE_MODE_NONE) 1914 { 1915 memset(ps_proc->pu1_slice_idx, 0, ps_proc->i4_wd_mbs * ps_proc->i4_ht_mbs); 1916 } 1917 /* generate slices for every 'n' rows, 'n' is given through slice param */ 1918 else if (ps_codec->s_cfg.e_slice_mode == IVE_SLICE_MODE_BLOCKS) 1919 { 1920 /* slice idx map */ 1921 UWORD8 *pu1_slice_idx = ps_proc->pu1_slice_idx; 1922 1923 /* temp var */ 1924 WORD32 i4_mb_y = 0, slice_idx = 0, cnt; 1925 1926 while (i4_mb_y < ps_proc->i4_ht_mbs) 1927 { 1928 if (i4_mb_y +(WORD32)ps_codec->s_cfg.u4_slice_param < ps_proc->i4_ht_mbs) 1929 { 1930 cnt = ps_codec->s_cfg.u4_slice_param * ps_proc->i4_wd_mbs; 1931 i4_mb_y += ps_codec->s_cfg.u4_slice_param; 1932 } 1933 else 1934 { 1935 cnt = (ps_proc->i4_ht_mbs - i4_mb_y) * ps_proc->i4_wd_mbs; 1936 i4_mb_y += (ps_proc->i4_ht_mbs - i4_mb_y); 1937 } 1938 memset(pu1_slice_idx, slice_idx, cnt); 1939 slice_idx++; 1940 pu1_slice_idx += cnt; 1941 } 1942 } 1943 1944 /* Current MV Bank's buffer ID */ 1945 ps_proc->i4_cur_mv_bank_buf_id = cur_mv_bank_buf_id; 1946 1947 /* Pointer to current picture buffer structure */ 1948 ps_proc->ps_cur_pic = ps_cur_pic; 1949 1950 /* Pointer to current pictures mv buffers */ 1951 ps_proc->ps_cur_mv_buf = ps_mv_buf; 1952 1953 /* 1954 * pointer to ref picture 1955 * 0 : Temporal back reference 1956 * 1 : Temporal forward reference 1957 */ 1958 ps_proc->aps_ref_pic[PRED_L0] = aps_ref_pic[PRED_L0]; 1959 ps_proc->aps_ref_pic[PRED_L1] = aps_ref_pic[PRED_L1]; 1960 if (ps_codec->pic_type == PIC_B) 1961 { 1962 ps_proc->aps_mv_buf[PRED_L0] = aps_mv_buf[PRED_L0]; 1963 ps_proc->aps_mv_buf[PRED_L1] = aps_mv_buf[PRED_L1]; 1964 } 1965 else 1966 { 1967 /* 1968 * Else is dummy since for non B pic we does not need this 1969 * But an assignment here will help in not having a segfault 1970 * when we calcualte colpic in P slices 1971 */ 1972 ps_proc->aps_mv_buf[PRED_L0] = ps_mv_buf; 1973 ps_proc->aps_mv_buf[PRED_L1] = ps_mv_buf; 1974 } 1975 1976 if ((*pic_type != PIC_IDR) && (*pic_type != PIC_I)) 1977 { 1978 /* temporal back an forward ref pointer luma and chroma */ 1979 ps_proc->apu1_ref_buf_luma_base[PRED_L0] = aps_ref_pic[PRED_L0]->pu1_luma; 1980 ps_proc->apu1_ref_buf_chroma_base[PRED_L0] = aps_ref_pic[PRED_L0]->pu1_chroma; 1981 1982 ps_proc->apu1_ref_buf_luma_base[PRED_L1] = aps_ref_pic[PRED_L1]->pu1_luma; 1983 ps_proc->apu1_ref_buf_chroma_base[PRED_L1] = aps_ref_pic[PRED_L1]->pu1_chroma; 1984 } 1985 1986 /* Structure for current input buffer */ 1987 ps_proc->s_inp_buf = *ps_inp_buf; 1988 1989 /* Number of encode frame API calls made */ 1990 ps_proc->i4_encode_api_call_cnt = ps_codec->i4_encode_api_call_cnt; 1991 1992 /* Current Picture count */ 1993 ps_proc->i4_pic_cnt = ps_codec->i4_pic_cnt; 1994 1995 /* error status */ 1996 ps_proc->i4_error_code = 0; 1997 1998 /********************************************************************/ 1999 /* INITIALIZE ENTROPY CONTEXT */ 2000 /********************************************************************/ 2001 { 2002 entropy_ctxt_t *ps_entropy = &ps_proc->s_entropy; 2003 2004 /* start of frame */ 2005 ps_entropy->i4_sof = 0; 2006 2007 /* end of frame */ 2008 ps_entropy->i4_eof = 0; 2009 2010 /* generate header */ 2011 ps_entropy->i4_gen_header = ps_codec->i4_gen_header; 2012 2013 /* sps ref_set_id */ 2014 ps_entropy->u4_sps_id = ps_codec->i4_sps_id; 2015 2016 /* sps base */ 2017 ps_entropy->ps_sps_base = ps_codec->ps_sps_base; 2018 2019 /* sps id */ 2020 ps_entropy->u4_pps_id = ps_codec->i4_pps_id; 2021 2022 /* sps base */ 2023 ps_entropy->ps_pps_base = ps_codec->ps_pps_base; 2024 2025 /* slice map */ 2026 ps_entropy->pu1_slice_idx = ps_proc->pu1_slice_idx; 2027 2028 /* slice hdr base */ 2029 ps_entropy->ps_slice_hdr_base = ps_proc->ps_slice_hdr_base; 2030 2031 /* Abs poc */ 2032 ps_entropy->i4_abs_pic_order_cnt = ps_proc->ps_codec->i4_poc; 2033 2034 /* initialize entropy map */ 2035 if (i == j) 2036 { 2037 /* row '-1' */ 2038 memset(ps_entropy->pu1_entropy_map - ps_proc->i4_wd_mbs, 1, ps_proc->i4_wd_mbs); 2039 /* row 0 to ht in mbs */ 2040 memset(ps_entropy->pu1_entropy_map, 0, ps_proc->i4_wd_mbs * ps_proc->i4_ht_mbs); 2041 2042 /* intialize cabac tables */ 2043 ih264e_init_cabac_table(ps_entropy); 2044 } 2045 2046 /* wd in mbs */ 2047 ps_entropy->i4_wd_mbs = ps_proc->i4_wd_mbs; 2048 2049 /* ht in mbs */ 2050 ps_entropy->i4_ht_mbs = ps_proc->i4_ht_mbs; 2051 2052 /* transform_8x8_mode_flag */ 2053 ps_entropy->i1_transform_8x8_mode_flag = 0; 2054 2055 /* entropy_coding_mode_flag */ 2056 ps_entropy->u1_entropy_coding_mode_flag = 2057 ps_codec->s_cfg.u4_entropy_coding_mode; 2058 2059 /* error code */ 2060 ps_entropy->i4_error_code = IH264E_SUCCESS; 2061 2062 /* mb skip run */ 2063 *(ps_proc->s_entropy.pi4_mb_skip_run) = 0; 2064 2065 /* last frame to encode */ 2066 ps_proc->s_entropy.u4_is_last = ps_inp_buf->u4_is_last; 2067 2068 /* Current Picture count */ 2069 ps_proc->s_entropy.i4_pic_cnt = ps_codec->i4_pic_cnt; 2070 2071 /* time stamps */ 2072 ps_entropy->u4_timestamp_low = u4_timestamp_low; 2073 ps_entropy->u4_timestamp_high = u4_timestamp_high; 2074 2075 /* init frame statistics */ 2076 ps_entropy->u4_header_bits[MB_TYPE_INTRA] = 0; 2077 ps_entropy->u4_header_bits[MB_TYPE_INTER] = 0; 2078 ps_entropy->u4_residue_bits[MB_TYPE_INTRA] = 0; 2079 ps_entropy->u4_residue_bits[MB_TYPE_INTER] = 0; 2080 } 2081 2082 /********************************************************************/ 2083 /* INITIALIZE DEBLOCK CONTEXT */ 2084 /********************************************************************/ 2085 { 2086 /* deblk ctxt */ 2087 deblk_ctxt_t *ps_deblk = &ps_proc->s_deblk_ctxt; 2088 2089 /* slice idx map */ 2090 ps_deblk->pu1_slice_idx = ps_proc->pu1_slice_idx; 2091 } 2092 2093 /********************************************************************/ 2094 /* INITIALIZE ME CONTEXT */ 2095 /********************************************************************/ 2096 { 2097 /* me ctxt */ 2098 me_ctxt_t *ps_me_ctxt = &ps_proc->s_me_ctxt; 2099 2100 /* srch range x */ 2101 ps_me_ctxt->ai2_srch_boundaries[0] = 2102 ps_codec->s_cfg.u4_srch_rng_x; 2103 2104 /* srch range y */ 2105 ps_me_ctxt->ai2_srch_boundaries[1] = 2106 ps_codec->s_cfg.u4_srch_rng_y; 2107 2108 /* rec stride */ 2109 ps_me_ctxt->i4_rec_strd = ps_codec->i4_rec_strd; 2110 2111 /* Half x plane offset from pic buf */ 2112 ps_me_ctxt->u4_half_x_offset = ps_proc->u4_half_x_offset; 2113 2114 /* Half y plane offset from half x plane */ 2115 ps_me_ctxt->u4_half_y_offset = ps_proc->u4_half_y_offset; 2116 2117 /* Half x plane offset from half y plane */ 2118 ps_me_ctxt->u4_half_xy_offset = ps_proc->u4_half_xy_offset; 2119 2120 /* enable fast sad */ 2121 ps_me_ctxt->u4_enable_fast_sad = u4_enable_fast_sad; 2122 2123 /* half pel */ 2124 ps_me_ctxt->u4_enable_hpel = ps_codec->s_cfg.u4_enable_hpel; 2125 2126 /* Diamond search Iteration Max Cnt */ 2127 ps_me_ctxt->u4_num_layers = u4_num_layers; 2128 2129 /* me speed preset */ 2130 ps_me_ctxt->u4_me_speed_preset = 2131 ps_codec->s_cfg.u4_me_speed_preset; 2132 2133 /* qp */ 2134 ps_me_ctxt->u1_mb_qp = ps_codec->u4_frame_qp; 2135 2136 if ((i == j) && (0 == ps_codec->i4_poc)) 2137 { 2138 /* init mv bits tables */ 2139 ih264e_init_mv_bits(ps_me_ctxt); 2140 } 2141 } 2142 2143 ps_proc->ps_ngbr_avbl = &(ps_proc->s_ngbr_avbl); 2144 2145 } 2146 2147 /* reset encoder header */ 2148 ps_codec->i4_gen_header = 0; 2149 } 2150 2151 /********************************************************************/ 2152 /* ADD JOBS TO THE QUEUE */ 2153 /********************************************************************/ 2154 { 2155 /* job structures */ 2156 job_t s_job; 2157 2158 /* temp var */ 2159 WORD32 i; 2160 2161 /* job class */ 2162 s_job.i4_cmd = CMD_PROCESS; 2163 2164 /* number of mbs to be processed in the current job */ 2165 s_job.i2_mb_cnt = ps_codec->s_cfg.i4_wd_mbs; 2166 2167 /* job start index x */ 2168 s_job.i2_mb_x = 0; 2169 2170 /* proc base idx */ 2171 s_job.i2_proc_base_idx = ctxt_sel ? (MAX_PROCESS_CTXT / 2) : 0; 2172 2173 for (i = 0; i < (WORD32)ps_codec->s_cfg.i4_ht_mbs; i++) 2174 { 2175 /* job start index y */ 2176 s_job.i2_mb_y = i; 2177 2178 /* queue the job */ 2179 ret = ih264_list_queue(ps_codec->pv_proc_jobq, &s_job, 1); 2180 if (ret != IH264_SUCCESS) 2181 { 2182 ps_codec->i4_error_code = ret; 2183 return IH264E_FAIL; 2184 } 2185 } 2186 2187 /* Once all the jobs are queued, terminate the queue */ 2188 /* Since the threads are created and deleted in each call, terminating 2189 here is not an issue */ 2190 ih264_list_terminate(ps_codec->pv_proc_jobq); 2191 } 2192 2193 return error_status; 2194 } 2195
