Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 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_process.c |
| 25 | * |
| 26 | * @brief |
| 27 | * Contains functions for codec thread |
| 28 | * |
| 29 | * @author |
| 30 | * Harish |
| 31 | * |
| 32 | * @par List of Functions: |
| 33 | * - ih264e_generate_sps_pps() |
| 34 | * - ih264e_init_entropy_ctxt() |
| 35 | * - ih264e_entropy() |
| 36 | * - ih264e_pack_header_data() |
| 37 | * - ih264e_update_proc_ctxt() |
| 38 | * - ih264e_init_proc_ctxt() |
| 39 | * - ih264e_pad_recon_buffer() |
| 40 | * - ih264e_dblk_pad_hpel_processing_n_mbs() |
| 41 | * - ih264e_process() |
| 42 | * - ih264e_set_rc_pic_params() |
| 43 | * - ih264e_update_rc_post_enc() |
| 44 | * - ih264e_process_thread() |
| 45 | * |
| 46 | * @remarks |
| 47 | * None |
| 48 | * |
| 49 | ******************************************************************************* |
| 50 | */ |
| 51 | |
| 52 | /*****************************************************************************/ |
| 53 | /* File Includes */ |
| 54 | /*****************************************************************************/ |
| 55 | |
| 56 | /* System include files */ |
| 57 | #include <stdio.h> |
| 58 | #include <stddef.h> |
| 59 | #include <stdlib.h> |
| 60 | #include <string.h> |
| 61 | #include <limits.h> |
| 62 | #include <assert.h> |
| 63 | |
| 64 | /* User include files */ |
| 65 | #include "ih264_typedefs.h" |
| 66 | #include "iv2.h" |
| 67 | #include "ive2.h" |
| 68 | #include "ih264_defs.h" |
| 69 | #include "ih264_debug.h" |
| 70 | #include "ime_distortion_metrics.h" |
| 71 | #include "ime_structs.h" |
| 72 | #include "ih264_defs.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_platform_macros.h" |
| 82 | #include "ih264_macros.h" |
| 83 | #include "ih264_error.h" |
| 84 | #include "ih264_buf_mgr.h" |
| 85 | #include "ih264e_error.h" |
| 86 | #include "ih264e_bitstream.h" |
| 87 | #include "ih264_structs.h" |
| 88 | #include "ih264_common_tables.h" |
| 89 | #include "ih264_list.h" |
| 90 | #include "ih264e_defs.h" |
| 91 | #include "irc_cntrl_param.h" |
| 92 | #include "irc_frame_info_collector.h" |
| 93 | #include "ih264e_rate_control.h" |
| 94 | #include "ih264e_structs.h" |
| 95 | #include "ih264e_process.h" |
| 96 | #include "ithread.h" |
| 97 | #include "ih264e_intra_modes_eval.h" |
| 98 | #include "ih264e_encode_header.h" |
| 99 | #include "ih264e_globals.h" |
| 100 | #include "ih264e_config.h" |
| 101 | #include "ih264e_trace.h" |
| 102 | #include "ih264e_statistics.h" |
| 103 | #include "ih264_cavlc_tables.h" |
| 104 | #include "ih264e_cavlc.h" |
| 105 | #include "ih264e_deblk.h" |
| 106 | #include "ih264e_me.h" |
| 107 | #include "ih264e_debug.h" |
| 108 | #include "ih264e_process.h" |
| 109 | #include "ih264e_master.h" |
| 110 | #include "ih264e_utils.h" |
| 111 | #include "irc_mem_req_and_acq.h" |
| 112 | #include "irc_cntrl_param.h" |
| 113 | #include "irc_frame_info_collector.h" |
| 114 | #include "irc_rate_control_api.h" |
| 115 | #include "ih264e_platform_macros.h" |
| 116 | #include "ih264_padding.h" |
| 117 | #include "ime_statistics.h" |
| 118 | |
| 119 | |
| 120 | /*****************************************************************************/ |
| 121 | /* Function Definitions */ |
| 122 | /*****************************************************************************/ |
| 123 | |
| 124 | /** |
| 125 | ****************************************************************************** |
| 126 | * |
| 127 | * @brief This function generates sps, pps set on request |
| 128 | * |
| 129 | * @par Description |
| 130 | * When the encoder is set in header generation mode, the following function |
| 131 | * is called. This generates sps and pps headers and returns the control back |
| 132 | * to caller. |
| 133 | * |
| 134 | * @param[in] ps_codec |
| 135 | * pointer to codec context |
| 136 | * |
| 137 | * @return success or failure error code |
| 138 | * |
| 139 | ****************************************************************************** |
| 140 | */ |
| 141 | IH264E_ERROR_T ih264e_generate_sps_pps(codec_t *ps_codec) |
| 142 | { |
| 143 | /* choose between ping-pong process buffer set */ |
| 144 | WORD32 ctxt_sel = ps_codec->i4_encode_api_call_cnt & 1; |
| 145 | |
| 146 | /* entropy ctxt */ |
| 147 | entropy_ctxt_t *ps_entropy = &ps_codec->as_process[ctxt_sel * MAX_PROCESS_THREADS].s_entropy; |
| 148 | |
| 149 | /* Bitstream structure */ |
| 150 | bitstrm_t *ps_bitstrm = ps_entropy->ps_bitstrm; |
| 151 | |
| 152 | /* sps */ |
| 153 | sps_t *ps_sps = NULL; |
| 154 | |
| 155 | /* pps */ |
| 156 | pps_t *ps_pps = NULL; |
| 157 | |
| 158 | /* output buff */ |
| 159 | out_buf_t *ps_out_buf = &ps_codec->as_out_buf[ctxt_sel]; |
| 160 | |
| 161 | |
| 162 | /********************************************************************/ |
| 163 | /* initialize the bit stream buffer */ |
| 164 | /********************************************************************/ |
| 165 | ih264e_bitstrm_init(ps_bitstrm, ps_out_buf->s_bits_buf.pv_buf, ps_out_buf->s_bits_buf.u4_bufsize); |
| 166 | |
| 167 | /********************************************************************/ |
| 168 | /* BEGIN HEADER GENERATION */ |
| 169 | /********************************************************************/ |
| 170 | /*ps_codec->i4_pps_id ++;*/ |
| 171 | ps_codec->i4_pps_id %= MAX_PPS_CNT; |
| 172 | |
| 173 | /*ps_codec->i4_sps_id ++;*/ |
| 174 | ps_codec->i4_sps_id %= MAX_SPS_CNT; |
| 175 | |
| 176 | /* populate sps header */ |
| 177 | ps_sps = ps_codec->ps_sps_base + ps_codec->i4_sps_id; |
| 178 | ih264e_populate_sps(ps_codec, ps_sps); |
| 179 | |
| 180 | /* populate pps header */ |
| 181 | ps_pps = ps_codec->ps_pps_base + ps_codec->i4_pps_id; |
| 182 | ih264e_populate_pps(ps_codec, ps_pps); |
| 183 | |
| 184 | ps_entropy->i4_error_code = IH264E_SUCCESS; |
| 185 | |
| 186 | /* generate sps */ |
| 187 | ps_entropy->i4_error_code |= ih264e_generate_sps(ps_bitstrm, ps_sps); |
| 188 | |
| 189 | /* generate pps */ |
| 190 | ps_entropy->i4_error_code |= ih264e_generate_pps(ps_bitstrm, ps_pps, ps_sps); |
| 191 | |
| 192 | /* queue output buffer */ |
| 193 | ps_out_buf->s_bits_buf.u4_bytes = ps_bitstrm->u4_strm_buf_offset; |
| 194 | |
| 195 | return ps_entropy->i4_error_code; |
| 196 | } |
| 197 | |
| 198 | /** |
| 199 | ******************************************************************************* |
| 200 | * |
| 201 | * @brief initialize entropy context. |
| 202 | * |
| 203 | * @par Description: |
| 204 | * Before invoking the call to perform to entropy coding the entropy context |
| 205 | * associated with the job needs to be initialized. This involves the start |
| 206 | * mb address, end mb address, slice index and the pointer to location at |
| 207 | * which the mb residue info and mb header info are packed. |
| 208 | * |
| 209 | * @param[in] ps_proc |
| 210 | * Pointer to the current process context |
| 211 | * |
| 212 | * @returns error status |
| 213 | * |
| 214 | * @remarks none |
| 215 | * |
| 216 | ******************************************************************************* |
| 217 | */ |
| 218 | IH264E_ERROR_T ih264e_init_entropy_ctxt(process_ctxt_t *ps_proc) |
| 219 | { |
| 220 | /* codec context */ |
| 221 | codec_t *ps_codec = ps_proc->ps_codec; |
| 222 | |
| 223 | /* entropy ctxt */ |
| 224 | entropy_ctxt_t *ps_entropy = &ps_proc->s_entropy; |
| 225 | |
| 226 | /* start address */ |
| 227 | ps_entropy->i4_mb_start_add = ps_entropy->i4_mb_y * ps_entropy->i4_wd_mbs + ps_entropy->i4_mb_x; |
| 228 | |
| 229 | /* end address */ |
| 230 | ps_entropy->i4_mb_end_add = ps_entropy->i4_mb_start_add + ps_entropy->i4_mb_cnt; |
| 231 | |
| 232 | /* slice index */ |
| 233 | ps_entropy->i4_cur_slice_idx = ps_proc->pu1_slice_idx[ps_entropy->i4_mb_start_add]; |
| 234 | |
| 235 | /* sof */ |
| 236 | /* @ start of frame or start of a new slice, set sof flag */ |
| 237 | if (ps_entropy->i4_mb_start_add == 0) |
| 238 | { |
| 239 | ps_entropy->i4_sof = 1; |
| 240 | } |
| 241 | |
| 242 | if (ps_entropy->i4_mb_x == 0) |
| 243 | { |
| 244 | /* packed mb coeff data */ |
| 245 | ps_entropy->pv_mb_coeff_data = ((UWORD8 *)ps_entropy->pv_pic_mb_coeff_data) + |
| 246 | ps_entropy->i4_mb_y * ps_codec->u4_size_coeff_data; |
| 247 | |
| 248 | /* packed mb header data */ |
| 249 | ps_entropy->pv_mb_header_data = ((UWORD8 *)ps_entropy->pv_pic_mb_header_data) + |
| 250 | ps_entropy->i4_mb_y * ps_codec->u4_size_header_data; |
| 251 | } |
| 252 | |
| 253 | return IH264E_SUCCESS; |
| 254 | } |
| 255 | |
| 256 | /** |
| 257 | ******************************************************************************* |
| 258 | * |
| 259 | * @brief entry point for entropy coding |
| 260 | * |
| 261 | * @par Description |
| 262 | * This function calls lower level functions to perform entropy coding for a |
| 263 | * group (n rows) of mb's. After encoding 1 row of mb's, the function takes |
| 264 | * back the control, updates the ctxt and calls lower level functions again. |
| 265 | * This process is repeated till all the rows or group of mb's (which ever is |
| 266 | * minimum) are coded |
| 267 | * |
| 268 | * @param[in] ps_proc |
| 269 | * process context |
| 270 | * |
| 271 | * @returns error status |
| 272 | * |
| 273 | * @remarks |
| 274 | * |
| 275 | ******************************************************************************* |
| 276 | */ |
| 277 | #define GET_NUM_BITS(ps_bitstream) ((ps_bitstream->u4_strm_buf_offset << 3) + WORD_SIZE - ps_bitstream->i4_bits_left_in_cw) |
| 278 | |
| 279 | IH264E_ERROR_T ih264e_entropy(process_ctxt_t *ps_proc) |
| 280 | { |
| 281 | /* codec context */ |
| 282 | codec_t *ps_codec = ps_proc->ps_codec; |
| 283 | |
| 284 | /* entropy context */ |
| 285 | entropy_ctxt_t *ps_entropy = &ps_proc->s_entropy; |
| 286 | |
| 287 | /* sps */ |
| 288 | sps_t *ps_sps = ps_entropy->ps_sps_base + (ps_entropy->u4_sps_id % MAX_SPS_CNT); |
| 289 | |
| 290 | /* pps */ |
| 291 | pps_t *ps_pps = ps_entropy->ps_pps_base + (ps_entropy->u4_pps_id % MAX_PPS_CNT); |
| 292 | |
| 293 | /* slice header */ |
| 294 | slice_header_t *ps_slice_hdr = ps_entropy->ps_slice_hdr_base + (ps_entropy->i4_cur_slice_idx % MAX_SLICE_HDR_CNT); |
| 295 | |
| 296 | /* slice type */ |
| 297 | WORD32 i4_slice_type = ps_proc->i4_slice_type; |
| 298 | |
| 299 | /* Bitstream structure */ |
| 300 | bitstrm_t *ps_bitstrm = ps_entropy->ps_bitstrm; |
| 301 | |
| 302 | /* output buff */ |
| 303 | out_buf_t s_out_buf; |
| 304 | |
| 305 | /* proc map */ |
| 306 | UWORD8 *pu1_proc_map; |
| 307 | |
| 308 | /* entropy map */ |
| 309 | UWORD8 *pu1_entropy_map_curr; |
| 310 | |
| 311 | /* proc base idx */ |
| 312 | WORD32 ctxt_sel = ps_proc->i4_encode_api_call_cnt & 1; |
| 313 | |
| 314 | /* temp var */ |
| 315 | WORD32 i4_wd_mbs, i4_ht_mbs; |
| 316 | UWORD32 u4_mb_cnt, u4_mb_idx, u4_mb_end_idx; |
| 317 | |
| 318 | /********************************************************************/ |
| 319 | /* BEGIN INIT */ |
| 320 | /********************************************************************/ |
| 321 | |
| 322 | /* entropy encode start address */ |
| 323 | u4_mb_idx = ps_entropy->i4_mb_start_add; |
| 324 | |
| 325 | /* entropy encode end address */ |
| 326 | u4_mb_end_idx = ps_entropy->i4_mb_end_add; |
| 327 | |
| 328 | /* width in mbs */ |
| 329 | i4_wd_mbs = ps_entropy->i4_wd_mbs; |
| 330 | |
| 331 | /* height in mbs */ |
| 332 | i4_ht_mbs = ps_entropy->i4_ht_mbs; |
| 333 | |
| 334 | /* total mb cnt */ |
| 335 | u4_mb_cnt = i4_wd_mbs * i4_ht_mbs; |
| 336 | |
| 337 | /* proc map */ |
| 338 | pu1_proc_map = ps_proc->pu1_proc_map + ps_entropy->i4_mb_y * i4_wd_mbs; |
| 339 | |
| 340 | /* entropy map */ |
| 341 | pu1_entropy_map_curr = ps_entropy->pu1_entropy_map + ps_entropy->i4_mb_y * i4_wd_mbs; |
| 342 | |
| 343 | /********************************************************************/ |
| 344 | /* @ start of frame / slice, */ |
| 345 | /* initialize the output buffer, */ |
| 346 | /* initialize the bit stream buffer, */ |
| 347 | /* check if sps and pps headers have to be generated, */ |
| 348 | /* populate and generate slice header */ |
| 349 | /********************************************************************/ |
| 350 | if (ps_entropy->i4_sof) |
| 351 | { |
| 352 | /********************************************************************/ |
| 353 | /* initialize the output buffer */ |
| 354 | /********************************************************************/ |
| 355 | s_out_buf = ps_codec->as_out_buf[ctxt_sel]; |
| 356 | |
| 357 | /* is last frame to encode */ |
| 358 | s_out_buf.u4_is_last = ps_entropy->u4_is_last; |
| 359 | |
| 360 | /* frame idx */ |
| 361 | s_out_buf.u4_timestamp_high = ps_entropy->u4_timestamp_high; |
| 362 | s_out_buf.u4_timestamp_low = ps_entropy->u4_timestamp_low; |
| 363 | |
| 364 | /********************************************************************/ |
| 365 | /* initialize the bit stream buffer */ |
| 366 | /********************************************************************/ |
| 367 | ih264e_bitstrm_init(ps_bitstrm, s_out_buf.s_bits_buf.pv_buf, s_out_buf.s_bits_buf.u4_bufsize); |
| 368 | |
| 369 | /********************************************************************/ |
| 370 | /* BEGIN HEADER GENERATION */ |
| 371 | /********************************************************************/ |
| 372 | if (1 == ps_entropy->i4_gen_header) |
| 373 | { |
| 374 | /* generate sps */ |
| 375 | ps_entropy->i4_error_code |= ih264e_generate_sps(ps_bitstrm, ps_sps); |
| 376 | |
| 377 | /* generate pps */ |
| 378 | ps_entropy->i4_error_code |= ih264e_generate_pps(ps_bitstrm, ps_pps, ps_sps); |
| 379 | |
| 380 | /* reset i4_gen_header */ |
| 381 | ps_entropy->i4_gen_header = 0; |
| 382 | } |
| 383 | |
| 384 | /* populate slice header */ |
| 385 | ih264e_populate_slice_header(ps_proc, ps_slice_hdr, ps_pps, ps_sps); |
| 386 | |
| 387 | /* generate slice header */ |
| 388 | ps_entropy->i4_error_code |= ih264e_generate_slice_header(ps_bitstrm, ps_slice_hdr, |
| 389 | ps_pps, ps_sps); |
| 390 | |
| 391 | /* once start of frame / slice is done, you can reset it */ |
| 392 | /* it is the responsibility of the caller to set this flag */ |
| 393 | ps_entropy->i4_sof = 0; |
| 394 | } |
| 395 | |
| 396 | /* begin entropy coding for the mb set */ |
| 397 | while (u4_mb_idx < u4_mb_end_idx) |
| 398 | { |
| 399 | /* init ptrs/indices */ |
| 400 | if (ps_entropy->i4_mb_x == i4_wd_mbs) |
| 401 | { |
| 402 | ps_entropy->i4_mb_y ++; |
| 403 | ps_entropy->i4_mb_x = 0; |
| 404 | |
| 405 | /* packed mb coeff data */ |
| 406 | ps_entropy->pv_mb_coeff_data = ((UWORD8 *)ps_entropy->pv_pic_mb_coeff_data) + |
| 407 | ps_entropy->i4_mb_y * ps_codec->u4_size_coeff_data; |
| 408 | |
| 409 | /* packed mb header data */ |
| 410 | ps_entropy->pv_mb_header_data = ((UWORD8 *)ps_entropy->pv_pic_mb_header_data) + |
| 411 | ps_entropy->i4_mb_y * ps_codec->u4_size_header_data; |
| 412 | |
| 413 | /* proc map */ |
| 414 | pu1_proc_map = ps_proc->pu1_proc_map + ps_entropy->i4_mb_y * i4_wd_mbs; |
| 415 | |
| 416 | /* entropy map */ |
| 417 | pu1_entropy_map_curr = ps_entropy->pu1_entropy_map + ps_entropy->i4_mb_y * i4_wd_mbs; |
| 418 | } |
| 419 | |
| 420 | DEBUG("\nmb indices x, y %d, %d", ps_entropy->i4_mb_x, ps_entropy->i4_mb_y); |
| 421 | ENTROPY_TRACE("mb index x %d", ps_entropy->i4_mb_x); |
| 422 | ENTROPY_TRACE("mb index y %d", ps_entropy->i4_mb_y); |
| 423 | |
| 424 | /* wait until the curr mb is core coded */ |
| 425 | /* The wait for curr mb to be core coded is essential when entropy is launched |
| 426 | * as a separate job |
| 427 | */ |
| 428 | while (1) |
| 429 | { |
| 430 | volatile UWORD8 *pu1_buf1; |
| 431 | WORD32 idx = ps_entropy->i4_mb_x; |
| 432 | |
| 433 | pu1_buf1 = pu1_proc_map + idx; |
| 434 | if(*pu1_buf1) |
| 435 | break; |
| 436 | ithread_yield(); |
| 437 | } |
| 438 | |
| 439 | /* write mb layer */ |
| 440 | ps_codec->pf_write_mb_syntax_layer[i4_slice_type](ps_entropy); |
| 441 | |
| 442 | /* set entropy map */ |
| 443 | pu1_entropy_map_curr[ps_entropy->i4_mb_x] = 1; |
| 444 | |
| 445 | u4_mb_idx ++; |
| 446 | ps_entropy->i4_mb_x ++; |
| 447 | |
| 448 | if (ps_entropy->i4_mb_x == i4_wd_mbs) |
| 449 | { |
| 450 | /* if slices are enabled */ |
| 451 | if (ps_codec->s_cfg.e_slice_mode == IVE_SLICE_MODE_BLOCKS) |
| 452 | { |
| 453 | /* current slice index */ |
| 454 | WORD32 i4_curr_slice_idx = ps_entropy->i4_cur_slice_idx; |
| 455 | |
| 456 | /* slice map */ |
| 457 | UWORD8 *pu1_slice_idx = ps_entropy->pu1_slice_idx; |
| 458 | |
| 459 | /* No need to open a slice at end of frame. The current slice can be closed at the time |
| 460 | * of signaling eof flag. |
| 461 | */ |
| 462 | if ( (u4_mb_idx != u4_mb_cnt) && (i4_curr_slice_idx != pu1_slice_idx[u4_mb_idx])) |
| 463 | { |
| 464 | /* mb skip run */ |
| 465 | if ((i4_slice_type != ISLICE) && *ps_entropy->pi4_mb_skip_run) |
| 466 | { |
| 467 | if (*ps_entropy->pi4_mb_skip_run) |
| 468 | { |
| 469 | PUT_BITS_UEV(ps_bitstrm, *ps_entropy->pi4_mb_skip_run, ps_entropy->i4_error_code, "mb skip run"); |
| 470 | *ps_entropy->pi4_mb_skip_run = 0; |
| 471 | } |
| 472 | } |
| 473 | |
| 474 | /* put rbsp trailing bits for the previous slice */ |
| 475 | ps_entropy->i4_error_code |= ih264e_put_rbsp_trailing_bits(ps_bitstrm); |
| 476 | |
| 477 | /* update slice header pointer */ |
| 478 | i4_curr_slice_idx = pu1_slice_idx[u4_mb_idx]; |
| 479 | ps_entropy->i4_cur_slice_idx = i4_curr_slice_idx; |
| 480 | ps_slice_hdr = ps_entropy->ps_slice_hdr_base + (i4_curr_slice_idx % MAX_SLICE_HDR_CNT); |
| 481 | |
| 482 | /* populate slice header */ |
| 483 | ps_entropy->i4_mb_start_add = u4_mb_idx; |
| 484 | ih264e_populate_slice_header(ps_proc, ps_slice_hdr, ps_pps, ps_sps); |
| 485 | |
| 486 | /* generate slice header */ |
| 487 | ps_entropy->i4_error_code |= ih264e_generate_slice_header(ps_bitstrm, ps_slice_hdr, |
| 488 | ps_pps, ps_sps); |
| 489 | } |
| 490 | } |
| 491 | |
| 492 | /* Dont execute any further instructions until store synchronization took place */ |
| 493 | DATA_SYNC(); |
| 494 | } |
| 495 | } |
| 496 | |
| 497 | /* check for eof */ |
| 498 | if (u4_mb_idx == u4_mb_cnt) |
| 499 | { |
| 500 | /* set end of frame flag */ |
| 501 | ps_entropy->i4_eof = 1; |
| 502 | } |
| 503 | |
| 504 | if (ps_entropy->i4_eof) |
| 505 | { |
| 506 | /* mb skip run */ |
| 507 | if ((i4_slice_type != ISLICE) && *ps_entropy->pi4_mb_skip_run) |
| 508 | { |
| 509 | if (*ps_entropy->pi4_mb_skip_run) |
| 510 | { |
| 511 | PUT_BITS_UEV(ps_bitstrm, *ps_entropy->pi4_mb_skip_run, ps_entropy->i4_error_code, "mb skip run"); |
| 512 | *ps_entropy->pi4_mb_skip_run = 0; |
| 513 | } |
| 514 | } |
| 515 | |
| 516 | /* put rbsp trailing bits */ |
| 517 | ps_entropy->i4_error_code |= ih264e_put_rbsp_trailing_bits(ps_bitstrm); |
| 518 | |
| 519 | /* update current frame stats to rc library */ |
| 520 | if (IVE_RC_NONE != ps_codec->s_cfg.e_rc_mode) |
| 521 | { |
| 522 | /* number of bytes to stuff */ |
| 523 | WORD32 i4_stuff_bytes; |
| 524 | |
| 525 | /* update */ |
| 526 | i4_stuff_bytes = ih264e_update_rc_post_enc(ps_codec, ctxt_sel, ps_proc->i4_pic_cnt); |
| 527 | |
| 528 | /* cbr rc - house keeping */ |
| 529 | if (ps_codec->s_rate_control.post_encode_skip[ctxt_sel]) |
| 530 | { |
| 531 | ps_entropy->ps_bitstrm->u4_strm_buf_offset = 0; |
| 532 | } |
| 533 | else if (i4_stuff_bytes) |
| 534 | { |
| 535 | /* add filler nal units */ |
| 536 | ps_entropy->i4_error_code |= ih264e_add_filler_nal_unit(ps_bitstrm, i4_stuff_bytes); |
| 537 | } |
| 538 | } |
| 539 | |
| 540 | /********************************************************************/ |
| 541 | /* signal the output */ |
| 542 | /********************************************************************/ |
| 543 | ps_codec->as_out_buf[ctxt_sel].s_bits_buf.u4_bytes = ps_entropy->ps_bitstrm->u4_strm_buf_offset; |
| 544 | |
| 545 | DEBUG("entropy status %x", ps_entropy->i4_error_code); |
| 546 | } |
| 547 | |
| 548 | /* allow threads to dequeue entropy jobs */ |
| 549 | ps_codec->au4_entropy_thread_active[ctxt_sel] = 0; |
| 550 | |
| 551 | return ps_entropy->i4_error_code; |
| 552 | } |
| 553 | |
| 554 | /** |
| 555 | ******************************************************************************* |
| 556 | * |
| 557 | * @brief Packs header information of a mb in to a buffer |
| 558 | * |
| 559 | * @par Description: |
| 560 | * After the deciding the mode info of a macroblock, the syntax elements |
| 561 | * associated with the mb are packed and stored. The entropy thread unpacks |
| 562 | * this buffer and generates the end bit stream. |
| 563 | * |
| 564 | * @param[in] ps_proc |
| 565 | * Pointer to the current process context |
| 566 | * |
| 567 | * @returns error status |
| 568 | * |
| 569 | * @remarks none |
| 570 | * |
| 571 | ******************************************************************************* |
| 572 | */ |
| 573 | IH264E_ERROR_T ih264e_pack_header_data(process_ctxt_t *ps_proc) |
| 574 | { |
| 575 | /* curr mb type */ |
| 576 | UWORD32 u4_mb_type = ps_proc->u4_mb_type; |
| 577 | |
| 578 | /* pack mb syntax layer of curr mb (used for entropy coding) */ |
| 579 | if (u4_mb_type == I4x4) |
| 580 | { |
| 581 | /* pointer to mb header storage space */ |
| 582 | UWORD8 *pu1_ptr = ps_proc->pv_mb_header_data; |
| 583 | |
| 584 | /* temp var */ |
| 585 | WORD32 i4, byte; |
| 586 | |
| 587 | /* mb type plus mode */ |
| 588 | *pu1_ptr++ = (ps_proc->u1_c_i8_mode << 6) + u4_mb_type; |
| 589 | |
| 590 | /* cbp */ |
| 591 | *pu1_ptr++ = ps_proc->u4_cbp; |
| 592 | |
| 593 | /* mb qp delta */ |
| 594 | *pu1_ptr++ = ps_proc->u4_mb_qp - ps_proc->u4_mb_qp_prev; |
| 595 | |
| 596 | /* sub mb modes */ |
| 597 | for (i4 = 0; i4 < 16; i4 ++) |
| 598 | { |
| 599 | byte = 0; |
| 600 | |
| 601 | if (ps_proc->au1_predicted_intra_luma_mb_4x4_modes[i4] == |
| 602 | ps_proc->au1_intra_luma_mb_4x4_modes[i4]) |
| 603 | { |
| 604 | byte |= 1; |
| 605 | } |
| 606 | else |
| 607 | { |
| 608 | |
| 609 | if (ps_proc->au1_intra_luma_mb_4x4_modes[i4] < |
| 610 | ps_proc->au1_predicted_intra_luma_mb_4x4_modes[i4]) |
| 611 | { |
| 612 | byte |= (ps_proc->au1_intra_luma_mb_4x4_modes[i4] << 1); |
| 613 | } |
| 614 | else |
| 615 | { |
| 616 | byte |= (ps_proc->au1_intra_luma_mb_4x4_modes[i4] - 1) << 1; |
| 617 | } |
| 618 | } |
| 619 | |
| 620 | i4++; |
| 621 | |
| 622 | if (ps_proc->au1_predicted_intra_luma_mb_4x4_modes[i4] == |
| 623 | ps_proc->au1_intra_luma_mb_4x4_modes[i4]) |
| 624 | { |
| 625 | byte |= 16; |
| 626 | } |
| 627 | else |
| 628 | { |
| 629 | |
| 630 | if (ps_proc->au1_intra_luma_mb_4x4_modes[i4] < |
| 631 | ps_proc->au1_predicted_intra_luma_mb_4x4_modes[i4]) |
| 632 | { |
| 633 | byte |= (ps_proc->au1_intra_luma_mb_4x4_modes[i4] << 5); |
| 634 | } |
| 635 | else |
| 636 | { |
| 637 | byte |= (ps_proc->au1_intra_luma_mb_4x4_modes[i4] - 1) << 5; |
| 638 | } |
| 639 | } |
| 640 | |
| 641 | *pu1_ptr++ = byte; |
| 642 | } |
| 643 | |
| 644 | /* end of mb layer */ |
| 645 | ps_proc->pv_mb_header_data = pu1_ptr; |
| 646 | } |
| 647 | else if (u4_mb_type == I16x16) |
| 648 | { |
| 649 | /* pointer to mb header storage space */ |
| 650 | UWORD8 *pu1_ptr = ps_proc->pv_mb_header_data; |
| 651 | |
| 652 | /* mb type plus mode */ |
| 653 | *pu1_ptr++ = (ps_proc->u1_c_i8_mode << 6) + (ps_proc->u1_l_i16_mode << 4) + u4_mb_type; |
| 654 | |
| 655 | /* cbp */ |
| 656 | *pu1_ptr++ = ps_proc->u4_cbp; |
| 657 | |
| 658 | /* mb qp delta */ |
| 659 | *pu1_ptr++ = ps_proc->u4_mb_qp - ps_proc->u4_mb_qp_prev; |
| 660 | |
| 661 | /* end of mb layer */ |
| 662 | ps_proc->pv_mb_header_data = pu1_ptr; |
| 663 | } |
| 664 | else if (u4_mb_type == P16x16) |
| 665 | { |
| 666 | /* pointer to mb header storage space */ |
| 667 | UWORD8 *pu1_ptr = ps_proc->pv_mb_header_data; |
| 668 | |
| 669 | WORD16 *i2_mv_ptr; |
| 670 | |
| 671 | /* mb type plus mode */ |
| 672 | *pu1_ptr++ = u4_mb_type; |
| 673 | |
| 674 | /* cbp */ |
| 675 | *pu1_ptr++ = ps_proc->u4_cbp; |
| 676 | |
| 677 | /* mb qp delta */ |
| 678 | *pu1_ptr++ = ps_proc->u4_mb_qp - ps_proc->u4_mb_qp_prev; |
| 679 | |
| 680 | i2_mv_ptr = (WORD16 *)pu1_ptr; |
| 681 | |
| 682 | *i2_mv_ptr++ = ps_proc->ps_pu->s_l0_mv.i2_mvx - ps_proc->ps_pred_mv->i2_mvx; |
| 683 | |
| 684 | *i2_mv_ptr++ = ps_proc->ps_pu->s_l0_mv.i2_mvy - ps_proc->ps_pred_mv->i2_mvy; |
| 685 | |
| 686 | /* end of mb layer */ |
| 687 | ps_proc->pv_mb_header_data = i2_mv_ptr; |
| 688 | } |
| 689 | else if (u4_mb_type == PSKIP) |
| 690 | { |
| 691 | /* pointer to mb header storage space */ |
| 692 | UWORD8 *pu1_ptr = ps_proc->pv_mb_header_data; |
| 693 | |
| 694 | /* mb type plus mode */ |
| 695 | *pu1_ptr++ = u4_mb_type; |
| 696 | |
| 697 | /* end of mb layer */ |
| 698 | ps_proc->pv_mb_header_data = pu1_ptr; |
| 699 | } |
| 700 | |
| 701 | return IH264E_SUCCESS; |
| 702 | } |
| 703 | |
| 704 | /** |
| 705 | ******************************************************************************* |
| 706 | * |
| 707 | * @brief update process context after encoding an mb. This involves preserving |
| 708 | * the current mb information for later use, initialize the proc ctxt elements to |
| 709 | * encode next mb. |
| 710 | * |
| 711 | * @par Description: |
| 712 | * This function performs house keeping tasks after encoding an mb. |
| 713 | * After encoding an mb, various elements of the process context needs to be |
| 714 | * updated to encode the next mb. For instance, the source, recon and reference |
| 715 | * pointers, mb indices have to be adjusted to the next mb. The slice index of |
| 716 | * the current mb needs to be updated. If mb qp modulation is enabled, then if |
| 717 | * the qp changes the quant param structure needs to be updated. Also to encoding |
| 718 | * the next mb, the current mb info is used as part of mode prediction or mv |
| 719 | * prediction. Hence the current mb info has to preserved at top/top left/left |
| 720 | * locations. |
| 721 | * |
| 722 | * @param[in] ps_proc |
| 723 | * Pointer to the current process context |
| 724 | * |
| 725 | * @returns none |
| 726 | * |
| 727 | * @remarks none |
| 728 | * |
| 729 | ******************************************************************************* |
| 730 | */ |
| 731 | WORD32 ih264e_update_proc_ctxt(process_ctxt_t *ps_proc) |
| 732 | { |
| 733 | /* error status */ |
| 734 | WORD32 error_status = IH264_SUCCESS; |
| 735 | |
| 736 | /* codec context */ |
| 737 | codec_t *ps_codec = ps_proc->ps_codec; |
| 738 | |
| 739 | /* curr mb indices */ |
| 740 | WORD32 i4_mb_x = ps_proc->i4_mb_x; |
| 741 | WORD32 i4_mb_y = ps_proc->i4_mb_y; |
| 742 | |
| 743 | /* mb syntax elements of neighbors */ |
| 744 | mb_info_t *ps_left_syn = &ps_proc->s_left_mb_syntax_ele; |
| 745 | mb_info_t *ps_top_syn = ps_proc->ps_top_row_mb_syntax_ele + i4_mb_x; |
| 746 | mb_info_t *ps_top_left_syn = &ps_proc->s_top_left_mb_syntax_ele; |
| 747 | |
| 748 | /* curr mb type */ |
| 749 | UWORD32 u4_mb_type = ps_proc->u4_mb_type; |
| 750 | |
| 751 | /* curr mb type */ |
| 752 | UWORD32 u4_is_intra = ps_proc->u4_is_intra; |
| 753 | |
| 754 | /* width in mbs */ |
| 755 | WORD32 i4_wd_mbs = ps_proc->i4_wd_mbs; |
| 756 | |
| 757 | /*height in mbs*/ |
| 758 | WORD32 i4_ht_mbs = ps_proc->i4_ht_mbs; |
| 759 | |
| 760 | /* proc map */ |
| 761 | UWORD8 *pu1_proc_map = ps_proc->pu1_proc_map + (i4_mb_y * i4_wd_mbs); |
| 762 | |
| 763 | /* deblk context */ |
| 764 | deblk_ctxt_t *ps_deblk = &ps_proc->s_deblk_ctxt; |
| 765 | |
| 766 | /* deblk bs context */ |
| 767 | bs_ctxt_t *ps_bs = &(ps_deblk->s_bs_ctxt); |
| 768 | |
| 769 | /* top row motion vector info */ |
| 770 | enc_pu_t *ps_top_row_pu = ps_proc->ps_top_row_pu + i4_mb_x; |
| 771 | |
| 772 | /* top left mb motion vector */ |
| 773 | enc_pu_t *ps_top_left_mb_pu = &ps_proc->s_top_left_mb_pu; |
| 774 | |
| 775 | /* left mb motion vector */ |
| 776 | enc_pu_t *ps_left_mb_pu = &ps_proc->s_left_mb_pu; |
| 777 | |
| 778 | /* sub mb modes */ |
| 779 | UWORD8 *pu1_top_mb_intra_modes = ps_proc->pu1_top_mb_intra_modes + (i4_mb_x << 4); |
| 780 | |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 781 | /*************************************************************/ |
| 782 | /* During MV prediction, when top right mb is not available, */ |
| 783 | /* top left mb info. is used for prediction. Hence the curr */ |
| 784 | /* top, which will be top left for the next mb needs to be */ |
| 785 | /* preserved before updating it with curr mb info. */ |
| 786 | /*************************************************************/ |
| 787 | |
| 788 | /* mb type, mb class, csbp */ |
| 789 | *ps_top_left_syn = *ps_top_syn; |
| 790 | |
| 791 | if (ps_proc->i4_slice_type == PSLICE) |
| 792 | { |
| 793 | /*****************************************/ |
| 794 | /* update top left with top info results */ |
| 795 | /*****************************************/ |
| 796 | |
| 797 | /* mv */ |
| 798 | *ps_top_left_mb_pu = *ps_top_row_pu; |
| 799 | } |
| 800 | |
| 801 | /*************************************************/ |
| 802 | /* update top and left with curr mb info results */ |
| 803 | /*************************************************/ |
| 804 | |
| 805 | /* mb type */ |
| 806 | ps_left_syn->u2_mb_type = ps_top_syn->u2_mb_type = u4_mb_type; |
| 807 | |
| 808 | /* mb class */ |
| 809 | ps_left_syn->u2_is_intra = ps_top_syn->u2_is_intra = u4_is_intra; |
| 810 | |
| 811 | /* csbp */ |
| 812 | ps_left_syn->u4_csbp = ps_top_syn->u4_csbp = ps_proc->u4_csbp; |
| 813 | |
| 814 | /* distortion */ |
| 815 | ps_left_syn->i4_mb_distortion = ps_top_syn->i4_mb_distortion = ps_proc->i4_mb_distortion; |
| 816 | |
| 817 | if (u4_is_intra) |
| 818 | { |
| 819 | /* mb / sub mb modes */ |
| 820 | if (I16x16 == u4_mb_type) |
| 821 | { |
| 822 | pu1_top_mb_intra_modes[0] = ps_proc->au1_left_mb_intra_modes[0] = ps_proc->u1_l_i16_mode; |
| 823 | } |
| 824 | else if (I4x4 == u4_mb_type) |
| 825 | { |
| 826 | ps_codec->pf_mem_cpy_mul8(ps_proc->au1_left_mb_intra_modes, ps_proc->au1_intra_luma_mb_4x4_modes, 16); |
| 827 | ps_codec->pf_mem_cpy_mul8(pu1_top_mb_intra_modes, ps_proc->au1_intra_luma_mb_4x4_modes, 16); |
| 828 | } |
| 829 | else if (I8x8 == u4_mb_type) |
| 830 | { |
| 831 | memcpy(ps_proc->au1_left_mb_intra_modes, ps_proc->au1_intra_luma_mb_8x8_modes, 4); |
| 832 | memcpy(pu1_top_mb_intra_modes, ps_proc->au1_intra_luma_mb_8x8_modes, 4); |
| 833 | } |
| 834 | |
| 835 | if (ps_proc->i4_slice_type == PSLICE) |
| 836 | { |
| 837 | /* mv */ |
| 838 | *ps_left_mb_pu = *ps_top_row_pu = *(ps_proc->ps_pu); |
| 839 | |
| 840 | // /* reset ngbr mv's */ |
| 841 | // ps_top_row_pu->i1_l0_ref_idx = -1; |
| 842 | // ps_top_row_pu->s_l0_mv = zero_mv; |
| 843 | // |
| 844 | // *ps_left_mb_pu = *ps_top_row_pu; |
| 845 | } |
| 846 | } |
| 847 | else |
| 848 | { |
| 849 | /* mv */ |
| 850 | *ps_left_mb_pu = *ps_top_row_pu = *(ps_proc->ps_pu); |
| 851 | } |
| 852 | |
| 853 | /* |
| 854 | * Mark that the MB has been coded intra |
| 855 | * So that future AIRs can skip it |
| 856 | */ |
| 857 | ps_proc->pu1_is_intra_coded[i4_mb_x + (i4_mb_y * i4_wd_mbs)] = u4_is_intra; |
| 858 | |
| 859 | /**************************************************/ |
| 860 | /* pack mb header info. for entropy coding */ |
| 861 | /**************************************************/ |
| 862 | ih264e_pack_header_data(ps_proc); |
| 863 | |
| 864 | /* update previous mb qp */ |
| 865 | ps_proc->u4_mb_qp_prev = ps_proc->u4_mb_qp; |
| 866 | |
| 867 | /* store qp */ |
| 868 | ps_proc->s_deblk_ctxt.s_bs_ctxt.pu1_pic_qp[(i4_mb_y * i4_wd_mbs) + i4_mb_x] = ps_proc->u4_mb_qp; |
| 869 | |
| 870 | /* |
| 871 | * We need to sync the cache to make sure that the nmv content of proc |
| 872 | * is updated to cache properly |
| 873 | */ |
| 874 | DATA_SYNC(); |
| 875 | |
| 876 | /* Just before finishing the row, enqueue the job in to entropy queue. |
| 877 | * The master thread depending on its convenience shall dequeue it and |
| 878 | * performs entropy. |
| 879 | * |
| 880 | * WARN !! Placing this block post proc map update can cause queuing of |
| 881 | * entropy jobs in out of order. |
| 882 | */ |
| 883 | if (i4_mb_x == i4_wd_mbs - 1) |
| 884 | { |
| 885 | /* job structures */ |
| 886 | job_t s_job; |
| 887 | |
| 888 | /* job class */ |
| 889 | s_job.i4_cmd = CMD_ENTROPY; |
| 890 | |
| 891 | /* number of mbs to be processed in the current job */ |
| 892 | s_job.i2_mb_cnt = ps_codec->s_cfg.i4_wd_mbs; |
| 893 | |
| 894 | /* job start index x */ |
| 895 | s_job.i2_mb_x = 0; |
| 896 | |
| 897 | /* job start index y */ |
| 898 | s_job.i2_mb_y = ps_proc->i4_mb_y; |
| 899 | |
| 900 | /* proc base idx */ |
| 901 | s_job.i2_proc_base_idx = (ps_codec->i4_encode_api_call_cnt & 1) ? (MAX_PROCESS_CTXT / 2): 0 ; |
| 902 | |
| 903 | /* queue the job */ |
| 904 | error_status |= ih264_list_queue(ps_proc->pv_entropy_jobq, &s_job, 1); |
| 905 | |
| 906 | if(ps_proc->i4_mb_y == (i4_ht_mbs - 1)) |
| 907 | ih264_list_terminate(ps_codec->pv_entropy_jobq); |
| 908 | } |
| 909 | |
| 910 | /* update proc map */ |
| 911 | pu1_proc_map[i4_mb_x] = 1; |
| 912 | |
| 913 | /**************************************************/ |
| 914 | /* update proc ctxt elements for encoding next mb */ |
| 915 | /**************************************************/ |
| 916 | /* update indices */ |
| 917 | i4_mb_x ++; |
| 918 | ps_proc->i4_mb_x = i4_mb_x; |
| 919 | |
| 920 | if (ps_proc->i4_mb_x == i4_wd_mbs) |
| 921 | { |
| 922 | ps_proc->i4_mb_y++; |
| 923 | ps_proc->i4_mb_x = 0; |
| 924 | } |
| 925 | |
| 926 | /* update slice index */ |
| 927 | ps_proc->i4_cur_slice_idx = ps_proc->pu1_slice_idx[ps_proc->i4_mb_y * i4_wd_mbs + ps_proc->i4_mb_x]; |
| 928 | |
| 929 | /* update buffers pointers */ |
| 930 | ps_proc->pu1_src_buf_luma += MB_SIZE; |
| 931 | ps_proc->pu1_rec_buf_luma += MB_SIZE; |
| 932 | ps_proc->pu1_ref_buf_luma += MB_SIZE; |
| 933 | |
| 934 | /* |
| 935 | * Note: Although chroma mb size is 8, as the chroma buffers are interleaved, |
| 936 | * the stride per MB is MB_SIZE |
| 937 | */ |
| 938 | ps_proc->pu1_src_buf_chroma += MB_SIZE; |
| 939 | ps_proc->pu1_rec_buf_chroma += MB_SIZE; |
| 940 | ps_proc->pu1_ref_buf_chroma += MB_SIZE; |
| 941 | |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 942 | |
| 943 | /* Reset cost, distortion params */ |
| 944 | ps_proc->i4_mb_cost = INT_MAX; |
| 945 | ps_proc->i4_mb_distortion = SHRT_MAX; |
| 946 | |
| 947 | ps_proc->ps_pu += *ps_proc->pu4_mb_pu_cnt; |
| 948 | |
| 949 | ps_proc->pu4_mb_pu_cnt += 1; |
| 950 | |
| 951 | /* deblk ctxts */ |
| 952 | if (ps_proc->u4_disable_deblock_level != 1) |
| 953 | { |
| 954 | /* indices */ |
| 955 | ps_bs->i4_mb_x = ps_proc->i4_mb_x; |
| 956 | ps_bs->i4_mb_y = ps_proc->i4_mb_y; |
| 957 | |
| 958 | #ifndef N_MB_ENABLE /* For N MB processing update take place inside deblocking function */ |
| 959 | ps_deblk->i4_mb_x ++; |
| 960 | |
| 961 | ps_deblk->pu1_cur_pic_luma += MB_SIZE; |
| 962 | /* |
| 963 | * Note: Although chroma mb size is 8, as the chroma buffers are interleaved, |
| 964 | * the stride per MB is MB_SIZE |
| 965 | */ |
| 966 | ps_deblk->pu1_cur_pic_chroma += MB_SIZE; |
| 967 | #endif |
| 968 | } |
| 969 | |
| 970 | return error_status; |
| 971 | } |
| 972 | |
| 973 | /** |
| 974 | ******************************************************************************* |
| 975 | * |
| 976 | * @brief initialize process context. |
| 977 | * |
| 978 | * @par Description: |
| 979 | * Before dispatching the current job to process thread, the process context |
| 980 | * associated with the job is initialized. Usually every job aims to encode one |
| 981 | * row of mb's. Basing on the row indices provided by the job, the process |
| 982 | * context's buffer ptrs, slice indices and other elements that are necessary |
| 983 | * during core-coding are initialized. |
| 984 | * |
| 985 | * @param[in] ps_proc |
| 986 | * Pointer to the current process context |
| 987 | * |
| 988 | * @returns error status |
| 989 | * |
| 990 | * @remarks none |
| 991 | * |
| 992 | ******************************************************************************* |
| 993 | */ |
| 994 | IH264E_ERROR_T ih264e_init_proc_ctxt(process_ctxt_t *ps_proc) |
| 995 | { |
| 996 | /* codec context */ |
| 997 | codec_t *ps_codec = ps_proc->ps_codec; |
| 998 | |
| 999 | /* nmb processing context*/ |
| 1000 | n_mb_process_ctxt_t *ps_n_mb_ctxt = &ps_proc->s_n_mb_ctxt; |
| 1001 | |
| 1002 | /* indices */ |
| 1003 | WORD32 i4_mb_x, i4_mb_y; |
| 1004 | |
| 1005 | /* strides */ |
| 1006 | WORD32 i4_src_strd = ps_proc->i4_src_strd; |
| 1007 | WORD32 i4_rec_strd = ps_proc->i4_rec_strd; |
| 1008 | |
| 1009 | /* quant params */ |
| 1010 | quant_params_t *ps_qp_params = ps_proc->ps_qp_params[0]; |
| 1011 | |
| 1012 | /* deblk ctxt */ |
| 1013 | deblk_ctxt_t *ps_deblk = &ps_proc->s_deblk_ctxt; |
| 1014 | |
| 1015 | /* deblk bs context */ |
| 1016 | bs_ctxt_t *ps_bs = &(ps_deblk->s_bs_ctxt); |
| 1017 | |
| 1018 | /* Pointer to mv_buffer of current frame */ |
| 1019 | mv_buf_t *ps_cur_mv_buf = ps_proc->ps_cur_mv_buf; |
| 1020 | |
| 1021 | /* Pointers for color space conversion */ |
| 1022 | UWORD8 *pu1_y_buf_base, *pu1_u_buf_base, *pu1_v_buf_base; |
| 1023 | |
| 1024 | /* Pad the MB to support non standard sizes */ |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1025 | UWORD32 u4_pad_right_sz = ps_codec->s_cfg.u4_wd - ps_codec->s_cfg.u4_disp_wd; |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1026 | UWORD32 u4_pad_bottom_sz = ps_codec->s_cfg.u4_ht - ps_codec->s_cfg.u4_disp_ht; |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1027 | UWORD16 u2_num_rows = MB_SIZE; |
| 1028 | WORD32 convert_uv_only; |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1029 | |
| 1030 | /********************************************************************/ |
| 1031 | /* BEGIN INIT */ |
| 1032 | /********************************************************************/ |
| 1033 | |
| 1034 | i4_mb_x = ps_proc->i4_mb_x; |
| 1035 | i4_mb_y = ps_proc->i4_mb_y; |
| 1036 | |
| 1037 | /* Number of mbs processed in one loop of process function */ |
| 1038 | ps_proc->i4_nmb_ntrpy = (ps_proc->i4_wd_mbs > MAX_NMB) ? MAX_NMB : ps_proc->i4_wd_mbs; |
| 1039 | ps_proc->u4_nmb_me = (ps_proc->i4_wd_mbs > MAX_NMB)? MAX_NMB : ps_proc->i4_wd_mbs; |
| 1040 | |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1041 | convert_uv_only = 1; |
| 1042 | if (u4_pad_bottom_sz && (ps_proc->i4_mb_y == ps_proc->i4_ht_mbs - 1)) |
| 1043 | { |
| 1044 | u2_num_rows = (UWORD16) MB_SIZE - u4_pad_bottom_sz; |
| 1045 | ps_proc->pu1_src_buf_luma_base = ps_codec->pu1_y_csc_buf_base; |
| 1046 | ps_proc->pu1_src_buf_luma = ps_proc->pu1_src_buf_luma_base + (i4_mb_x * MB_SIZE) + ps_codec->s_cfg.u4_max_wd * (i4_mb_y * MB_SIZE); |
| 1047 | convert_uv_only = 0; |
| 1048 | |
| 1049 | } |
| 1050 | else |
| 1051 | ps_proc->pu1_src_buf_luma = ps_proc->pu1_src_buf_luma_base + (i4_mb_x * MB_SIZE) + i4_src_strd * (i4_mb_y * MB_SIZE); |
| 1052 | |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1053 | /* init buffer pointers */ |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1054 | |
| 1055 | ps_proc->pu1_src_buf_chroma = ps_proc->pu1_src_buf_chroma_base + (i4_mb_x * MB_SIZE) + ps_codec->s_cfg.u4_max_wd * (i4_mb_y * BLK8x8SIZE); |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1056 | ps_proc->pu1_rec_buf_luma = ps_proc->pu1_rec_buf_luma_base + (i4_mb_x * MB_SIZE) + i4_rec_strd * (i4_mb_y * MB_SIZE); |
| 1057 | ps_proc->pu1_rec_buf_chroma = ps_proc->pu1_rec_buf_chroma_base + (i4_mb_x * MB_SIZE) + i4_rec_strd * (i4_mb_y * BLK8x8SIZE); |
| 1058 | ps_proc->pu1_ref_buf_luma = ps_proc->pu1_ref_buf_luma_base + (i4_mb_x * MB_SIZE) + i4_rec_strd * (i4_mb_y * MB_SIZE); |
| 1059 | ps_proc->pu1_ref_buf_chroma = ps_proc->pu1_ref_buf_chroma_base + (i4_mb_x * MB_SIZE) + i4_rec_strd * (i4_mb_y * BLK8x8SIZE); |
| 1060 | |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1061 | |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1062 | /* |
| 1063 | * Do color space conversion |
| 1064 | * NOTE : We assume there that the number of MB's to process will not span multiple rows |
| 1065 | */ |
| 1066 | switch (ps_codec->s_cfg.e_inp_color_fmt) |
| 1067 | { |
| 1068 | case IV_YUV_420SP_UV: |
| 1069 | case IV_YUV_420SP_VU: |
| 1070 | break; |
| 1071 | |
| 1072 | case IV_YUV_420P : |
| 1073 | pu1_y_buf_base = (UWORD8 *)ps_proc->s_inp_buf.s_raw_buf.apv_bufs[0] + (i4_mb_x * MB_SIZE) + |
| 1074 | ps_proc->s_inp_buf.s_raw_buf.au4_strd[0] * (i4_mb_y * MB_SIZE); |
| 1075 | |
| 1076 | pu1_u_buf_base = (UWORD8 *)ps_proc->s_inp_buf.s_raw_buf.apv_bufs[1] + (i4_mb_x * BLK8x8SIZE) + |
| 1077 | ps_proc->s_inp_buf.s_raw_buf.au4_strd[1] * (i4_mb_y * BLK8x8SIZE); |
| 1078 | |
| 1079 | pu1_v_buf_base = (UWORD8 *)ps_proc->s_inp_buf.s_raw_buf.apv_bufs[2] + (i4_mb_x * BLK8x8SIZE) + |
| 1080 | ps_proc->s_inp_buf.s_raw_buf.au4_strd[2] * (i4_mb_y * BLK8x8SIZE); |
| 1081 | |
| 1082 | ps_codec->pf_ih264e_conv_420p_to_420sp( |
| 1083 | pu1_y_buf_base, pu1_u_buf_base, pu1_v_buf_base, |
| 1084 | ps_proc->pu1_src_buf_luma, |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1085 | ps_proc->pu1_src_buf_chroma, u2_num_rows, |
| 1086 | ps_codec->s_cfg.u4_disp_wd, |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1087 | ps_proc->s_inp_buf.s_raw_buf.au4_strd[0], |
| 1088 | ps_proc->s_inp_buf.s_raw_buf.au4_strd[1], |
| 1089 | ps_proc->s_inp_buf.s_raw_buf.au4_strd[2], |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1090 | ps_proc->i4_src_strd, ps_proc->i4_src_strd, |
| 1091 | convert_uv_only); |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1092 | break; |
| 1093 | |
| 1094 | case IV_YUV_422ILE : |
| 1095 | pu1_y_buf_base = (UWORD8 *)ps_proc->s_inp_buf.s_raw_buf.apv_bufs[0] + (i4_mb_x * MB_SIZE * 2) |
| 1096 | + ps_proc->s_inp_buf.s_raw_buf.au4_strd[0] * (i4_mb_y * MB_SIZE); |
| 1097 | |
| 1098 | ps_codec->pf_ih264e_fmt_conv_422i_to_420sp( |
| 1099 | ps_proc->pu1_src_buf_luma, |
| 1100 | ps_proc->pu1_src_buf_chroma, |
| 1101 | ps_proc->pu1_src_buf_chroma + 1, pu1_y_buf_base, |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1102 | ps_codec->s_cfg.u4_disp_wd, u2_num_rows, |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1103 | ps_proc->i4_src_strd, ps_proc->i4_src_strd, |
| 1104 | ps_proc->i4_src_strd, |
| 1105 | ps_proc->s_inp_buf.s_raw_buf.au4_strd[0] >> 1); |
| 1106 | break; |
| 1107 | |
| 1108 | default: |
| 1109 | break; |
| 1110 | } |
| 1111 | |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1112 | if (u4_pad_right_sz && (ps_proc->i4_mb_x == 0) && |
| 1113 | (ps_proc->i4_src_strd > (WORD32)ps_codec->s_cfg.u4_disp_wd) ) |
| 1114 | { |
| 1115 | UWORD32 u4_pad_wd, u4_pad_ht; |
| 1116 | u4_pad_wd = (UWORD32)(ps_proc->i4_src_strd - ps_codec->s_cfg.u4_disp_wd); |
| 1117 | u4_pad_wd = MIN(u4_pad_right_sz, u4_pad_wd); |
| 1118 | u4_pad_ht = MB_SIZE; |
| 1119 | if(ps_proc->i4_mb_y == ps_proc->i4_ht_mbs - 1) |
| 1120 | u4_pad_ht = MIN(MB_SIZE, (MB_SIZE - u4_pad_bottom_sz)); |
| 1121 | |
| 1122 | ih264_pad_right_luma( |
| 1123 | ps_proc->pu1_src_buf_luma + ps_codec->s_cfg.u4_disp_wd, |
| 1124 | ps_proc->i4_src_strd, u4_pad_ht, u4_pad_wd); |
| 1125 | |
| 1126 | ih264_pad_right_chroma( |
| 1127 | ps_proc->pu1_src_buf_chroma + ps_codec->s_cfg.u4_disp_wd, |
| 1128 | ps_proc->i4_src_strd, u4_pad_ht / 2, u4_pad_wd); |
| 1129 | } |
| 1130 | |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1131 | /* pad bottom edge */ |
| 1132 | if (u4_pad_bottom_sz && (ps_proc->i4_mb_y == ps_proc->i4_ht_mbs - 1) && ps_proc->i4_mb_x == 0) |
| 1133 | { |
| 1134 | ih264_pad_bottom(ps_proc->pu1_src_buf_luma + (MB_SIZE - u4_pad_bottom_sz) * ps_proc->i4_src_strd, |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1135 | ps_proc->i4_src_strd, ps_proc->i4_src_strd, u4_pad_bottom_sz); |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1136 | |
| 1137 | ih264_pad_bottom(ps_proc->pu1_src_buf_chroma + (MB_SIZE - u4_pad_bottom_sz) * ps_proc->i4_src_strd / 2, |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1138 | ps_proc->i4_src_strd, ps_proc->i4_src_strd, (u4_pad_bottom_sz / 2)); |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1139 | } |
| 1140 | |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1141 | |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1142 | /* packed mb coeff data */ |
| 1143 | ps_proc->pv_mb_coeff_data = ((UWORD8 *)ps_proc->pv_pic_mb_coeff_data) + i4_mb_y * ps_codec->u4_size_coeff_data; |
| 1144 | |
| 1145 | /* packed mb header data */ |
| 1146 | ps_proc->pv_mb_header_data = ((UWORD8 *)ps_proc->pv_pic_mb_header_data) + i4_mb_y * ps_codec->u4_size_header_data; |
| 1147 | |
| 1148 | /* slice index */ |
| 1149 | ps_proc->i4_cur_slice_idx = ps_proc->pu1_slice_idx[i4_mb_y * ps_proc->i4_wd_mbs + i4_mb_x]; |
| 1150 | |
| 1151 | /*********************************************************************/ |
| 1152 | /* ih264e_init_quant_params() routine is called at the pic init level*/ |
| 1153 | /* this would have initialized the qp. */ |
| 1154 | /* TODO_LATER: currently it is assumed that quant params donot change*/ |
| 1155 | /* across mb's. When they do calculate update ps_qp_params accordingly*/ |
| 1156 | /*********************************************************************/ |
| 1157 | |
| 1158 | /* init mv buffer ptr */ |
| 1159 | ps_proc->ps_pu = ps_cur_mv_buf->ps_pic_pu + (i4_mb_y * ps_proc->i4_wd_mbs * (MIN_PU_SIZE * MIN_PU_SIZE)); |
| 1160 | |
| 1161 | if (i4_mb_y == 0) |
| 1162 | { |
| 1163 | ps_proc->ps_top_row_pu_ME = ps_cur_mv_buf->ps_pic_pu; |
| 1164 | } |
| 1165 | else |
| 1166 | { |
| 1167 | ps_proc->ps_top_row_pu_ME = ps_cur_mv_buf->ps_pic_pu + ((i4_mb_y - 1) * ps_proc->i4_wd_mbs * (MIN_PU_SIZE * MIN_PU_SIZE)); |
| 1168 | } |
| 1169 | |
| 1170 | ps_proc->pu4_mb_pu_cnt = ps_cur_mv_buf->pu4_mb_pu_cnt + (i4_mb_y * ps_proc->i4_wd_mbs); |
| 1171 | |
| 1172 | /* mb type */ |
| 1173 | ps_proc->u4_mb_type = I16x16; |
| 1174 | |
| 1175 | /* lambda */ |
| 1176 | ps_proc->u4_lambda = gu1_qp0[ps_qp_params->u1_mb_qp]; |
| 1177 | |
| 1178 | /* mb distortion */ |
| 1179 | ps_proc->i4_mb_distortion = SHRT_MAX; |
| 1180 | |
| 1181 | if (i4_mb_x == 0) |
| 1182 | { |
| 1183 | ps_proc->s_left_mb_syntax_ele.i4_mb_distortion = 0; |
| 1184 | |
| 1185 | ps_proc->s_top_left_mb_syntax_ele.i4_mb_distortion = 0; |
| 1186 | |
| 1187 | ps_proc->s_top_left_mb_syntax_ME.i4_mb_distortion = 0; |
| 1188 | |
| 1189 | if (i4_mb_y == 0) |
| 1190 | { |
| 1191 | memset(ps_proc->ps_top_row_mb_syntax_ele, 0, (ps_proc->i4_wd_mbs + 1)*sizeof(mb_info_t)); |
| 1192 | } |
| 1193 | } |
| 1194 | |
| 1195 | /* mb cost */ |
| 1196 | ps_proc->i4_mb_cost = INT_MAX; |
| 1197 | |
| 1198 | /**********************/ |
| 1199 | /* init deblk context */ |
| 1200 | /**********************/ |
| 1201 | ps_deblk->i4_mb_x = ps_proc->i4_mb_x; |
| 1202 | /* deblk lags the current mb proc by 1 row */ |
| 1203 | /* NOTE: Intra prediction has to happen with non deblocked samples used as reference */ |
| 1204 | /* Hence to deblk MB 0 of row 0, you have wait till MB 0 of row 1 is encoded. */ |
| 1205 | /* For simplicity, we chose to lag deblking by 1 Row wrt to proc */ |
| 1206 | ps_deblk->i4_mb_y = ps_proc->i4_mb_y - 1; |
| 1207 | |
| 1208 | /* buffer ptrs */ |
| 1209 | ps_deblk->pu1_cur_pic_luma = ps_proc->pu1_rec_buf_luma_base + i4_rec_strd * (ps_deblk->i4_mb_y * MB_SIZE); |
| 1210 | ps_deblk->pu1_cur_pic_chroma = ps_proc->pu1_rec_buf_chroma_base + i4_rec_strd * (ps_deblk->i4_mb_y * BLK8x8SIZE); |
| 1211 | |
| 1212 | /* init deblk bs context */ |
| 1213 | /* mb indices */ |
| 1214 | ps_bs->i4_mb_x = ps_proc->i4_mb_x; |
| 1215 | ps_bs->i4_mb_y = ps_proc->i4_mb_y; |
| 1216 | |
| 1217 | /* init n_mb_process context */ |
| 1218 | ps_n_mb_ctxt->i4_mb_x = 0; |
| 1219 | ps_n_mb_ctxt->i4_mb_y = ps_deblk->i4_mb_y; |
| 1220 | ps_n_mb_ctxt->i4_n_mbs = ps_proc->i4_nmb_ntrpy; |
| 1221 | |
| 1222 | return IH264E_SUCCESS; |
| 1223 | } |
| 1224 | |
| 1225 | /** |
| 1226 | ******************************************************************************* |
| 1227 | * |
| 1228 | * @brief This function performs luma & chroma padding |
| 1229 | * |
| 1230 | * @par Description: |
| 1231 | * |
| 1232 | * @param[in] ps_proc |
| 1233 | * Process context corresponding to the job |
| 1234 | * |
| 1235 | * @param[in] pu1_curr_pic_luma |
| 1236 | * Pointer to luma buffer |
| 1237 | * |
| 1238 | * @param[in] pu1_curr_pic_chroma |
| 1239 | * Pointer to chroma buffer |
| 1240 | * |
| 1241 | * @param[in] i4_mb_x |
| 1242 | * mb index x |
| 1243 | * |
| 1244 | * @param[in] i4_mb_y |
| 1245 | * mb index y |
| 1246 | * |
| 1247 | * @param[in] i4_pad_ht |
| 1248 | * number of rows to be padded |
| 1249 | * |
| 1250 | * @returns error status |
| 1251 | * |
| 1252 | * @remarks none |
| 1253 | * |
| 1254 | ******************************************************************************* |
| 1255 | */ |
| 1256 | IH264E_ERROR_T ih264e_pad_recon_buffer(process_ctxt_t *ps_proc, |
| 1257 | UWORD8 *pu1_curr_pic_luma, |
| 1258 | UWORD8 *pu1_curr_pic_chroma, |
| 1259 | WORD32 i4_mb_x, |
| 1260 | WORD32 i4_mb_y, |
| 1261 | WORD32 i4_pad_ht) |
| 1262 | { |
| 1263 | /* codec context */ |
| 1264 | codec_t *ps_codec = ps_proc->ps_codec; |
| 1265 | |
| 1266 | /* strides */ |
| 1267 | WORD32 i4_rec_strd = ps_proc->i4_rec_strd; |
| 1268 | |
| 1269 | if (i4_mb_x == 0) |
| 1270 | { |
| 1271 | /* padding left luma */ |
| 1272 | ps_codec->pf_pad_left_luma(pu1_curr_pic_luma, i4_rec_strd, i4_pad_ht, PAD_LEFT); |
| 1273 | |
| 1274 | /* padding left chroma */ |
| 1275 | ps_codec->pf_pad_left_chroma(pu1_curr_pic_chroma, i4_rec_strd, i4_pad_ht >> 1, PAD_LEFT); |
| 1276 | } |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1277 | if (i4_mb_x == ps_proc->i4_wd_mbs - 1) |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1278 | { |
| 1279 | /* padding right luma */ |
| 1280 | ps_codec->pf_pad_right_luma(pu1_curr_pic_luma + MB_SIZE, i4_rec_strd, i4_pad_ht, PAD_RIGHT); |
| 1281 | |
| 1282 | /* padding right chroma */ |
| 1283 | ps_codec->pf_pad_right_chroma(pu1_curr_pic_chroma + MB_SIZE, i4_rec_strd, i4_pad_ht >> 1, PAD_RIGHT); |
| 1284 | |
| 1285 | if (i4_mb_y == ps_proc->i4_ht_mbs - 1) |
| 1286 | { |
| 1287 | UWORD8 *pu1_rec_luma = pu1_curr_pic_luma + MB_SIZE + PAD_RIGHT + ((i4_pad_ht - 1) * i4_rec_strd); |
| 1288 | UWORD8 *pu1_rec_chroma = pu1_curr_pic_chroma + MB_SIZE + PAD_RIGHT + (((i4_pad_ht >> 1) - 1) * i4_rec_strd); |
| 1289 | |
| 1290 | /* padding bottom luma */ |
| 1291 | ps_codec->pf_pad_bottom(pu1_rec_luma, i4_rec_strd, i4_rec_strd, PAD_BOT); |
| 1292 | |
| 1293 | /* padding bottom chroma */ |
| 1294 | ps_codec->pf_pad_bottom(pu1_rec_chroma, i4_rec_strd, i4_rec_strd, (PAD_BOT >> 1)); |
| 1295 | } |
| 1296 | } |
| 1297 | |
| 1298 | if (i4_mb_y == 0) |
| 1299 | { |
| 1300 | UWORD8 *pu1_rec_luma = pu1_curr_pic_luma; |
| 1301 | UWORD8 *pu1_rec_chroma = pu1_curr_pic_chroma; |
| 1302 | WORD32 wd = MB_SIZE; |
| 1303 | |
| 1304 | if (i4_mb_x == 0) |
| 1305 | { |
| 1306 | pu1_rec_luma -= PAD_LEFT; |
| 1307 | pu1_rec_chroma -= PAD_LEFT; |
| 1308 | |
| 1309 | wd += PAD_LEFT; |
| 1310 | } |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1311 | if (i4_mb_x == ps_proc->i4_wd_mbs - 1) |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1312 | { |
| 1313 | wd += PAD_RIGHT; |
| 1314 | } |
| 1315 | |
| 1316 | /* padding top luma */ |
| 1317 | ps_codec->pf_pad_top(pu1_rec_luma, i4_rec_strd, wd, PAD_TOP); |
| 1318 | |
| 1319 | /* padding top chroma */ |
| 1320 | ps_codec->pf_pad_top(pu1_rec_chroma, i4_rec_strd, wd, (PAD_TOP >> 1)); |
| 1321 | } |
| 1322 | |
| 1323 | return IH264E_SUCCESS; |
| 1324 | } |
| 1325 | |
| 1326 | |
| 1327 | |
| 1328 | |
| 1329 | /** |
| 1330 | ******************************************************************************* |
| 1331 | * |
| 1332 | * @brief This function performs deblocking, padding and halfpel generation for |
| 1333 | * 'n' MBs |
| 1334 | * |
| 1335 | * @par Description: |
| 1336 | * |
| 1337 | * @param[in] ps_proc |
| 1338 | * Process context corresponding to the job |
| 1339 | * |
| 1340 | * @param[in] pu1_curr_pic_luma |
| 1341 | * Current MB being processed(Luma) |
| 1342 | * |
| 1343 | * @param[in] pu1_curr_pic_chroma |
| 1344 | * Current MB being processed(Chroma) |
| 1345 | * |
| 1346 | * @param[in] i4_mb_x |
| 1347 | * Column value of current MB processed |
| 1348 | * |
| 1349 | * @param[in] i4_mb_y |
| 1350 | * Curent row processed |
| 1351 | * |
| 1352 | * @returns error status |
| 1353 | * |
| 1354 | * @remarks none |
| 1355 | * |
| 1356 | ******************************************************************************* |
| 1357 | */ |
| 1358 | IH264E_ERROR_T ih264e_dblk_pad_hpel_processing_n_mbs(process_ctxt_t *ps_proc, |
| 1359 | UWORD8 *pu1_curr_pic_luma, |
| 1360 | UWORD8 *pu1_curr_pic_chroma, |
| 1361 | WORD32 i4_mb_x, |
| 1362 | WORD32 i4_mb_y) |
| 1363 | { |
| 1364 | /* codec context */ |
| 1365 | codec_t *ps_codec = ps_proc->ps_codec; |
| 1366 | |
| 1367 | /* n_mb processing context */ |
| 1368 | n_mb_process_ctxt_t *ps_n_mb_ctxt = &ps_proc->s_n_mb_ctxt; |
| 1369 | |
| 1370 | /* deblk context */ |
| 1371 | deblk_ctxt_t *ps_deblk = &ps_proc->s_deblk_ctxt; |
| 1372 | |
| 1373 | /* strides */ |
| 1374 | WORD32 i4_rec_strd = ps_proc->i4_rec_strd; |
| 1375 | |
| 1376 | /* loop variables */ |
| 1377 | WORD32 row, i, j, col; |
| 1378 | |
| 1379 | /* Padding Width */ |
| 1380 | UWORD32 u4_pad_wd; |
| 1381 | |
| 1382 | /* deblk_map of the row being deblocked */ |
| 1383 | UWORD8 *pu1_deblk_map = ps_proc->pu1_deblk_map + ps_deblk->i4_mb_y * ps_proc->i4_wd_mbs; |
| 1384 | |
| 1385 | /* deblk_map_previous row */ |
| 1386 | UWORD8 *pu1_deblk_map_prev_row = pu1_deblk_map - ps_proc->i4_wd_mbs; |
| 1387 | |
| 1388 | WORD32 u4_pad_top = 0; |
| 1389 | |
| 1390 | WORD32 u4_deblk_prev_row = 0; |
| 1391 | |
| 1392 | /* Number of mbs to be processed */ |
| 1393 | WORD32 i4_n_mbs = ps_n_mb_ctxt->i4_n_mbs; |
| 1394 | |
| 1395 | /* Number of mbs actually processed |
| 1396 | * (at the end of a row, when remaining number of MBs are less than i4_n_mbs) */ |
| 1397 | WORD32 i4_n_mb_process_count = 0; |
| 1398 | |
| 1399 | UWORD8 *pu1_pad_bottom_src = NULL; |
| 1400 | |
| 1401 | UWORD8 *pu1_pad_src_luma = NULL; |
| 1402 | UWORD8 *pu1_pad_src_chroma = NULL; |
| 1403 | |
| 1404 | if (ps_proc->u4_disable_deblock_level == 1) |
| 1405 | { |
| 1406 | /* If left most MB is processed, then pad left */ |
| 1407 | if (i4_mb_x == 0) |
| 1408 | { |
| 1409 | /* padding left luma */ |
| 1410 | ps_codec->pf_pad_left_luma(pu1_curr_pic_luma, i4_rec_strd, MB_SIZE, PAD_LEFT); |
| 1411 | |
| 1412 | /* padding left chroma */ |
| 1413 | ps_codec->pf_pad_left_chroma(pu1_curr_pic_chroma, i4_rec_strd, MB_SIZE >> 1, PAD_LEFT); |
| 1414 | } |
| 1415 | /*last col*/ |
| 1416 | if (i4_mb_x == (ps_proc->i4_wd_mbs - 1)) |
| 1417 | { |
| 1418 | /* padding right luma */ |
| 1419 | ps_codec->pf_pad_right_luma(pu1_curr_pic_luma + MB_SIZE, i4_rec_strd, MB_SIZE, PAD_RIGHT); |
| 1420 | |
| 1421 | /* padding right chroma */ |
| 1422 | ps_codec->pf_pad_right_chroma(pu1_curr_pic_chroma + MB_SIZE, i4_rec_strd, MB_SIZE >> 1, PAD_RIGHT); |
| 1423 | } |
| 1424 | } |
| 1425 | |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1426 | if ((i4_mb_y > 0) || (i4_mb_y == (ps_proc->i4_ht_mbs - 1))) |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1427 | { |
| 1428 | /* if number of mb's to be processed are less than 'N', go back. |
| 1429 | * exception to the above clause is end of row */ |
| 1430 | if ( ((i4_mb_x - (ps_n_mb_ctxt->i4_mb_x - 1)) < i4_n_mbs) && (i4_mb_x < (ps_proc->i4_wd_mbs - 1)) ) |
| 1431 | { |
| 1432 | return IH264E_SUCCESS; |
| 1433 | } |
| 1434 | else |
| 1435 | { |
| 1436 | i4_n_mb_process_count = MIN(i4_mb_x - (ps_n_mb_ctxt->i4_mb_x - 1), i4_n_mbs); |
| 1437 | |
| 1438 | u4_deblk_prev_row = 1; |
| 1439 | |
| 1440 | /* checking whether the top rows are deblocked */ |
| 1441 | for (col = 0; col < i4_n_mb_process_count; col++) |
| 1442 | { |
| 1443 | u4_deblk_prev_row &= pu1_deblk_map_prev_row[ps_deblk->i4_mb_x + col]; |
| 1444 | } |
| 1445 | |
| 1446 | /* checking whether the top right MB is deblocked */ |
| 1447 | if ((ps_deblk->i4_mb_x + i4_n_mb_process_count) != ps_proc->i4_wd_mbs) |
| 1448 | { |
| 1449 | u4_deblk_prev_row &= pu1_deblk_map_prev_row[ps_deblk->i4_mb_x + i4_n_mb_process_count]; |
| 1450 | } |
| 1451 | |
| 1452 | /* performing deblocking for required number of MBs */ |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1453 | if ((i4_mb_y > 0) && (ps_proc->u4_disable_deblock_level != 1)) |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1454 | { |
| 1455 | /* Top or Top right MBs not deblocked */ |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1456 | if ((u4_deblk_prev_row != 1) && (i4_mb_y > 0)) |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1457 | { |
| 1458 | return IH264E_SUCCESS; |
| 1459 | } |
| 1460 | |
| 1461 | for (row = 0; row < i4_n_mb_process_count; row++) |
| 1462 | { |
| 1463 | ih264e_deblock_mb(ps_proc, ps_deblk); |
| 1464 | |
| 1465 | pu1_deblk_map[ps_deblk->i4_mb_x] = 1; |
| 1466 | |
| 1467 | if (ps_deblk->i4_mb_y > 0) |
| 1468 | { |
| 1469 | if (ps_deblk->i4_mb_x == 0)/* If left most MB is processed, then pad left*/ |
| 1470 | { |
| 1471 | /* padding left luma */ |
| 1472 | ps_codec->pf_pad_left_luma(ps_deblk->pu1_cur_pic_luma - i4_rec_strd * MB_SIZE, i4_rec_strd, MB_SIZE, PAD_LEFT); |
| 1473 | |
| 1474 | /* padding left chroma */ |
| 1475 | ps_codec->pf_pad_left_chroma(ps_deblk->pu1_cur_pic_chroma - i4_rec_strd * BLK8x8SIZE, i4_rec_strd, MB_SIZE >> 1, PAD_LEFT); |
| 1476 | } |
| 1477 | |
| 1478 | if (ps_deblk->i4_mb_x == (ps_proc->i4_wd_mbs - 1))/*last column*/ |
| 1479 | { |
| 1480 | /* padding right luma */ |
| 1481 | ps_codec->pf_pad_right_luma(ps_deblk->pu1_cur_pic_luma - i4_rec_strd * MB_SIZE + MB_SIZE, i4_rec_strd, MB_SIZE, PAD_RIGHT); |
| 1482 | |
| 1483 | /* padding right chroma */ |
| 1484 | ps_codec->pf_pad_right_chroma(ps_deblk->pu1_cur_pic_chroma - i4_rec_strd * BLK8x8SIZE + MB_SIZE, i4_rec_strd, MB_SIZE >> 1, PAD_RIGHT); |
| 1485 | } |
| 1486 | } |
| 1487 | ps_deblk->i4_mb_x++; |
| 1488 | |
| 1489 | ps_deblk->pu1_cur_pic_luma += MB_SIZE; |
| 1490 | ps_deblk->pu1_cur_pic_chroma += MB_SIZE; |
| 1491 | |
| 1492 | } |
| 1493 | } |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1494 | else if(i4_mb_y > 0) |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1495 | { |
| 1496 | ps_deblk->i4_mb_x += i4_n_mb_process_count; |
| 1497 | |
| 1498 | ps_deblk->pu1_cur_pic_luma += i4_n_mb_process_count * MB_SIZE; |
| 1499 | ps_deblk->pu1_cur_pic_chroma += i4_n_mb_process_count * MB_SIZE; |
| 1500 | } |
| 1501 | |
| 1502 | if (i4_mb_y == 2) |
| 1503 | { |
| 1504 | u4_pad_wd = i4_n_mb_process_count * MB_SIZE; |
| 1505 | u4_pad_top = ps_n_mb_ctxt->i4_mb_x * MB_SIZE; |
| 1506 | |
| 1507 | if (ps_n_mb_ctxt->i4_mb_x == 0) |
| 1508 | { |
| 1509 | u4_pad_wd += PAD_LEFT; |
| 1510 | u4_pad_top = -PAD_LEFT; |
| 1511 | } |
| 1512 | |
| 1513 | if (i4_mb_x == ps_proc->i4_wd_mbs - 1) |
| 1514 | { |
| 1515 | u4_pad_wd += PAD_RIGHT; |
| 1516 | } |
| 1517 | |
| 1518 | /* padding top luma */ |
| 1519 | ps_codec->pf_pad_top(ps_proc->pu1_rec_buf_luma_base + u4_pad_top, i4_rec_strd, u4_pad_wd, PAD_TOP); |
| 1520 | |
| 1521 | /* padding top chroma */ |
| 1522 | ps_codec->pf_pad_top(ps_proc->pu1_rec_buf_chroma_base + u4_pad_top, i4_rec_strd, u4_pad_wd, (PAD_TOP >> 1)); |
| 1523 | } |
| 1524 | |
| 1525 | ps_n_mb_ctxt->i4_mb_x += i4_n_mb_process_count; |
| 1526 | |
| 1527 | if (i4_mb_x == ps_proc->i4_wd_mbs - 1) |
| 1528 | { |
| 1529 | if (ps_proc->i4_mb_y == ps_proc->i4_ht_mbs - 1) |
| 1530 | { |
| 1531 | /* Bottom Padding is done in one stretch for the entire width */ |
| 1532 | if (ps_proc->u4_disable_deblock_level != 1) |
| 1533 | { |
| 1534 | ps_deblk->pu1_cur_pic_luma = ps_proc->pu1_rec_buf_luma_base + (ps_proc->i4_ht_mbs - 1) * i4_rec_strd * MB_SIZE; |
| 1535 | |
| 1536 | ps_deblk->pu1_cur_pic_chroma = ps_proc->pu1_rec_buf_chroma_base + (ps_proc->i4_ht_mbs - 1) * i4_rec_strd * BLK8x8SIZE; |
| 1537 | |
| 1538 | ps_n_mb_ctxt->i4_mb_x = 0; |
| 1539 | ps_n_mb_ctxt->i4_mb_y = ps_proc->i4_mb_y; |
| 1540 | ps_deblk->i4_mb_x = 0; |
| 1541 | ps_deblk->i4_mb_y = ps_proc->i4_mb_y; |
| 1542 | |
| 1543 | /* update pic qp map (as update_proc_ctxt is still not called for the last MB) */ |
| 1544 | ps_proc->s_deblk_ctxt.s_bs_ctxt.pu1_pic_qp[(i4_mb_y * ps_proc->i4_wd_mbs) + i4_mb_x] = ps_proc->u4_mb_qp; |
| 1545 | |
| 1546 | i4_n_mb_process_count = (ps_proc->i4_wd_mbs) % i4_n_mbs; |
| 1547 | |
| 1548 | j = (ps_proc->i4_wd_mbs) / i4_n_mbs; |
| 1549 | |
| 1550 | for (i = 0; i < j; i++) |
| 1551 | { |
| 1552 | for (col = 0; col < i4_n_mbs; col++) |
| 1553 | { |
| 1554 | ih264e_deblock_mb(ps_proc, ps_deblk); |
| 1555 | |
| 1556 | pu1_deblk_map[ps_deblk->i4_mb_x] = 1; |
| 1557 | |
| 1558 | ps_deblk->i4_mb_x++; |
| 1559 | ps_deblk->pu1_cur_pic_luma += MB_SIZE; |
| 1560 | ps_deblk->pu1_cur_pic_chroma += MB_SIZE; |
| 1561 | ps_n_mb_ctxt->i4_mb_x++; |
| 1562 | } |
| 1563 | } |
| 1564 | |
| 1565 | for (col = 0; col < i4_n_mb_process_count; col++) |
| 1566 | { |
| 1567 | ih264e_deblock_mb(ps_proc, ps_deblk); |
| 1568 | |
| 1569 | pu1_deblk_map[ps_deblk->i4_mb_x] = 1; |
| 1570 | |
| 1571 | ps_deblk->i4_mb_x++; |
| 1572 | ps_deblk->pu1_cur_pic_luma += MB_SIZE; |
| 1573 | ps_deblk->pu1_cur_pic_chroma += MB_SIZE; |
| 1574 | ps_n_mb_ctxt->i4_mb_x++; |
| 1575 | } |
| 1576 | |
| 1577 | pu1_pad_src_luma = ps_proc->pu1_rec_buf_luma_base + (ps_proc->i4_ht_mbs - 2) * MB_SIZE * i4_rec_strd; |
| 1578 | |
| 1579 | pu1_pad_src_chroma = ps_proc->pu1_rec_buf_chroma_base + (ps_proc->i4_ht_mbs - 2) * BLK8x8SIZE * i4_rec_strd; |
| 1580 | |
| 1581 | /* padding left luma */ |
| 1582 | ps_codec->pf_pad_left_luma(pu1_pad_src_luma, i4_rec_strd, MB_SIZE, PAD_LEFT); |
| 1583 | |
| 1584 | /* padding left chroma */ |
| 1585 | ps_codec->pf_pad_left_chroma(pu1_pad_src_chroma, i4_rec_strd, BLK8x8SIZE, PAD_LEFT); |
| 1586 | |
| 1587 | pu1_pad_src_luma += i4_rec_strd * MB_SIZE; |
| 1588 | pu1_pad_src_chroma += i4_rec_strd * BLK8x8SIZE; |
| 1589 | |
| 1590 | /* padding left luma */ |
| 1591 | ps_codec->pf_pad_left_luma(pu1_pad_src_luma, i4_rec_strd, MB_SIZE, PAD_LEFT); |
| 1592 | |
| 1593 | /* padding left chroma */ |
| 1594 | ps_codec->pf_pad_left_chroma(pu1_pad_src_chroma, i4_rec_strd, BLK8x8SIZE, PAD_LEFT); |
| 1595 | |
| 1596 | pu1_pad_src_luma = ps_proc->pu1_rec_buf_luma_base + (ps_proc->i4_ht_mbs - 2) * MB_SIZE * i4_rec_strd + (ps_proc->i4_wd_mbs) * MB_SIZE; |
| 1597 | |
| 1598 | pu1_pad_src_chroma = ps_proc->pu1_rec_buf_chroma_base + (ps_proc->i4_ht_mbs - 2) * BLK8x8SIZE * i4_rec_strd + (ps_proc->i4_wd_mbs) * MB_SIZE; |
| 1599 | |
| 1600 | /* padding right luma */ |
| 1601 | ps_codec->pf_pad_right_luma(pu1_pad_src_luma, i4_rec_strd, MB_SIZE, PAD_RIGHT); |
| 1602 | |
| 1603 | /* padding right chroma */ |
| 1604 | ps_codec->pf_pad_right_chroma(pu1_pad_src_chroma, i4_rec_strd, BLK8x8SIZE, PAD_RIGHT); |
| 1605 | |
| 1606 | pu1_pad_src_luma += i4_rec_strd * MB_SIZE; |
| 1607 | pu1_pad_src_chroma += i4_rec_strd * BLK8x8SIZE; |
| 1608 | |
| 1609 | /* padding right luma */ |
| 1610 | ps_codec->pf_pad_right_luma(pu1_pad_src_luma, i4_rec_strd, MB_SIZE, PAD_RIGHT); |
| 1611 | |
| 1612 | /* padding right chroma */ |
| 1613 | ps_codec->pf_pad_right_chroma(pu1_pad_src_chroma, i4_rec_strd, BLK8x8SIZE, PAD_RIGHT); |
| 1614 | |
| 1615 | } |
| 1616 | |
Harish Mahendrakar | c72323e | 2015-04-28 19:07:40 +0530 | [diff] [blame^] | 1617 | /* In case height is less than 2 MBs pad top */ |
| 1618 | if (ps_proc->i4_ht_mbs <= 2) |
| 1619 | { |
| 1620 | UWORD8 *pu1_pad_top_src; |
| 1621 | /* padding top luma */ |
| 1622 | pu1_pad_top_src = ps_proc->pu1_rec_buf_luma_base - PAD_LEFT; |
| 1623 | ps_codec->pf_pad_top(pu1_pad_top_src, i4_rec_strd, i4_rec_strd, PAD_TOP); |
| 1624 | |
| 1625 | /* padding top chroma */ |
| 1626 | pu1_pad_top_src = ps_proc->pu1_rec_buf_chroma_base - PAD_LEFT; |
| 1627 | ps_codec->pf_pad_top(pu1_pad_top_src, i4_rec_strd, i4_rec_strd, (PAD_TOP >> 1)); |
| 1628 | } |
| 1629 | |
Hamsalekha S | 8d3d303 | 2015-03-13 21:24:58 +0530 | [diff] [blame] | 1630 | /* padding bottom luma */ |
| 1631 | pu1_pad_bottom_src = ps_proc->pu1_rec_buf_luma_base + ps_proc->i4_ht_mbs * MB_SIZE * i4_rec_strd - PAD_LEFT; |
| 1632 | ps_codec->pf_pad_bottom(pu1_pad_bottom_src, i4_rec_strd, i4_rec_strd, PAD_BOT); |
| 1633 | |
| 1634 | /* padding bottom chroma */ |
| 1635 | pu1_pad_bottom_src = ps_proc->pu1_rec_buf_chroma_base + ps_proc->i4_ht_mbs * (MB_SIZE >> 1) * i4_rec_strd - PAD_LEFT; |
| 1636 | ps_codec->pf_pad_bottom(pu1_pad_bottom_src, i4_rec_strd, i4_rec_strd, (PAD_BOT >> 1)); |
| 1637 | } |
| 1638 | } |
| 1639 | } |
| 1640 | } |
| 1641 | |
| 1642 | return IH264E_SUCCESS; |
| 1643 | } |
| 1644 | |
| 1645 | |
| 1646 | /** |
| 1647 | ******************************************************************************* |
| 1648 | * |
| 1649 | * @brief This function performs luma & chroma core coding for a set of mb's. |
| 1650 | * |
| 1651 | * @par Description: |
| 1652 | * The mb to be coded is taken and is evaluated over a predefined set of modes |
| 1653 | * (intra (i16, i4, i8)/inter (mv, skip)) for best cost. The mode with least cost |
| 1654 | * is selected and using intra/inter prediction filters, prediction is carried out. |
| 1655 | * The deviation between src and pred signal constitutes error signal. This error |
| 1656 | * signal is transformed (hierarchical transform if necessary) and quantized. The |
| 1657 | * quantized residue is packed in to entropy buffer for entropy coding. This is |
| 1658 | * repeated for all the mb's enlisted under the job. |
| 1659 | * |
| 1660 | * @param[in] ps_proc |
| 1661 | * Process context corresponding to the job |
| 1662 | * |
| 1663 | * @returns error status |
| 1664 | * |
| 1665 | * @remarks none |
| 1666 | * |
| 1667 | ******************************************************************************* |
| 1668 | */ |
| 1669 | WORD32 ih264e_process(process_ctxt_t *ps_proc) |
| 1670 | { |
| 1671 | /* error status */ |
| 1672 | WORD32 error_status = IH264_SUCCESS; |
| 1673 | |
| 1674 | /* codec context */ |
| 1675 | codec_t *ps_codec = ps_proc->ps_codec; |
| 1676 | |
| 1677 | /* cbp luma, chroma */ |
| 1678 | UWORD32 u4_cbp_l, u4_cbp_c; |
| 1679 | |
| 1680 | /* width in mbs */ |
| 1681 | WORD32 i4_wd_mbs = ps_proc->i4_wd_mbs; |
| 1682 | |
| 1683 | /* loop var */ |
| 1684 | WORD32 i4_mb_idx, i4_mb_cnt = ps_proc->i4_mb_cnt; |
| 1685 | |
| 1686 | /* valid modes */ |
| 1687 | UWORD32 u4_valid_modes = 0; |
| 1688 | |
| 1689 | /* gate threshold */ |
| 1690 | WORD32 i4_gate_threshold = 0; |
| 1691 | |
| 1692 | /* is intra */ |
| 1693 | WORD32 luma_idx, chroma_idx, is_intra; |
| 1694 | |
| 1695 | /* temp variables */ |
| 1696 | WORD32 ctxt_sel = ps_proc->i4_encode_api_call_cnt & 1; |
| 1697 | |
| 1698 | /* list of modes for evaluation */ |
| 1699 | if (ps_proc->i4_slice_type == ISLICE) |
| 1700 | { |
| 1701 | /* enable intra 16x16 */ |
| 1702 | u4_valid_modes |= ps_codec->s_cfg.u4_enable_intra_16x16 ? (1 << I16x16) : 0; |
| 1703 | |
| 1704 | /* enable intra 8x8 */ |
| 1705 | u4_valid_modes |= ps_codec->s_cfg.u4_enable_intra_8x8 ? (1 << I8x8) : 0; |
| 1706 | |
| 1707 | /* enable intra 4x4 */ |
| 1708 | u4_valid_modes |= ps_codec->s_cfg.u4_enable_intra_4x4 ? (1 << I4x4) : 0; |
| 1709 | } |
| 1710 | else if (ps_proc->i4_slice_type == PSLICE) |
| 1711 | { |
| 1712 | /* enable intra 16x16 */ |
| 1713 | u4_valid_modes |= ps_codec->s_cfg.u4_enable_intra_16x16 ? (1 << I16x16) : 0; |
| 1714 | |
| 1715 | /* enable intra 4x4 */ |
| 1716 | if (ps_codec->s_cfg.u4_enc_speed_preset == IVE_SLOWEST) |
| 1717 | { |
| 1718 | u4_valid_modes |= ps_codec->s_cfg.u4_enable_intra_4x4 ? (1 << I4x4) : 0; |
| 1719 | } |
| 1720 | |
| 1721 | /* enable inter 16x16 */ |
| 1722 | u4_valid_modes |= (1 << P16x16); |
| 1723 | } |
| 1724 | |
| 1725 | |
| 1726 | /* init entropy */ |
| 1727 | ps_proc->s_entropy.i4_mb_x = ps_proc->i4_mb_x; |
| 1728 | ps_proc->s_entropy.i4_mb_y = ps_proc->i4_mb_y; |
| 1729 | ps_proc->s_entropy.i4_mb_cnt = MIN(ps_proc->i4_nmb_ntrpy, i4_wd_mbs - ps_proc->i4_mb_x); |
| 1730 | |
| 1731 | /* compute recon when : |
| 1732 | * 1. current frame is to be used as a reference |
| 1733 | * 2. dump recon for bit stream sanity check |
| 1734 | */ |
| 1735 | ps_proc->u4_compute_recon = ps_codec->u4_is_curr_frm_ref || |
| 1736 | ps_codec->s_cfg.u4_enable_recon; |
| 1737 | |
| 1738 | /* Encode 'n' macroblocks, |
| 1739 | * 'n' being the number of mbs dictated by current proc ctxt */ |
| 1740 | for (i4_mb_idx = 0; i4_mb_idx < i4_mb_cnt; i4_mb_idx ++) |
| 1741 | { |
| 1742 | /* since we have not yet found sad, we have not yet got min sad */ |
| 1743 | /* we need to initialize these variables for each MB */ |
| 1744 | /* TODO how to get the min sad into the codec */ |
| 1745 | ps_proc->u4_min_sad = ps_codec->s_cfg.i4_min_sad; |
| 1746 | ps_proc->u4_min_sad_reached = 0; |
| 1747 | |
| 1748 | /* mb analysis */ |
| 1749 | { |
| 1750 | /* temp var */ |
| 1751 | WORD32 i4_mb_id = ps_proc->i4_mb_x + ps_proc->i4_mb_y * i4_wd_mbs; |
| 1752 | |
| 1753 | /* force intra refresh ? */ |
| 1754 | WORD32 i4_air_enable_inter = (ps_codec->s_cfg.e_air_mode == IVE_AIR_MODE_NONE) || |
| 1755 | (ps_proc->pu1_is_intra_coded[i4_mb_id] != 0) || |
| 1756 | (ps_codec->pu2_intr_rfrsh_map[i4_mb_id] != ps_codec->i4_air_pic_cnt); |
| 1757 | |
| 1758 | /* evaluate inter 16x16 modes */ |
| 1759 | if (u4_valid_modes & (1 << P16x16)) |
| 1760 | { |
| 1761 | /* compute nmb me */ |
| 1762 | if (ps_proc->i4_mb_x % ps_proc->u4_nmb_me == 0) |
| 1763 | { |
| 1764 | ih264e_compute_me_nmb(ps_proc, MIN((WORD32)ps_proc->u4_nmb_me, |
| 1765 | i4_wd_mbs - ps_proc->i4_mb_x)); |
| 1766 | } |
| 1767 | |
| 1768 | /* set pointers to ME data appropriately for other modules to use */ |
| 1769 | { |
| 1770 | UWORD32 u4_mb_index = ps_proc->i4_mb_x % ps_proc->u4_nmb_me ; |
| 1771 | |
| 1772 | /* get the min sad condition for current mb */ |
| 1773 | ps_proc->u4_min_sad_reached = ps_proc->ps_nmb_info[u4_mb_index].u4_min_sad_reached; |
| 1774 | ps_proc->u4_min_sad = ps_proc->ps_nmb_info[u4_mb_index].u4_min_sad; |
| 1775 | |
| 1776 | ps_proc->ps_skip_mv = &(ps_proc->ps_nmb_info[u4_mb_index].s_skip_mv); |
| 1777 | ps_proc->ps_ngbr_avbl = &(ps_proc->ps_nmb_info[u4_mb_index].s_ngbr_avbl); |
| 1778 | ps_proc->ps_pred_mv = &(ps_proc->ps_nmb_info[u4_mb_index].s_pred_mv); |
| 1779 | |
| 1780 | ps_proc->i4_mb_distortion = ps_proc->ps_nmb_info[u4_mb_index].i4_mb_distortion; |
| 1781 | ps_proc->i4_mb_cost = ps_proc->ps_nmb_info[u4_mb_index].i4_mb_cost; |
| 1782 | ps_proc->u4_min_sad = ps_proc->ps_nmb_info[u4_mb_index].u4_min_sad; |
| 1783 | ps_proc->u4_min_sad_reached = ps_proc->ps_nmb_info[u4_mb_index].u4_min_sad_reached; |
| 1784 | ps_proc->u4_mb_type = ps_proc->ps_nmb_info[u4_mb_index].u4_mb_type; |
| 1785 | |
| 1786 | /* get the best sub pel buffer */ |
| 1787 | ps_proc->pu1_best_subpel_buf = ps_proc->ps_nmb_info[u4_mb_index].pu1_best_sub_pel_buf; |
| 1788 | ps_proc->u4_bst_spel_buf_strd = ps_proc->ps_nmb_info[u4_mb_index].u4_bst_spel_buf_strd; |
| 1789 | } |
| 1790 | ih264e_derive_nghbr_avbl_of_mbs(ps_proc); |
| 1791 | } |
| 1792 | else |
| 1793 | { |
| 1794 | /* Derive neighbor availability for the current macroblock */ |
| 1795 | ps_proc->ps_ngbr_avbl = &ps_proc->s_ngbr_avbl; |
| 1796 | |
| 1797 | ih264e_derive_nghbr_avbl_of_mbs(ps_proc); |
| 1798 | } |
| 1799 | |
| 1800 | /* |
| 1801 | * If air says intra, we need to force the following code path to evaluate intra |
| 1802 | * The easy way is just to say that the inter cost is too much |
| 1803 | */ |
| 1804 | if (!i4_air_enable_inter) |
| 1805 | { |
| 1806 | ps_proc->u4_min_sad_reached = 0; |
| 1807 | ps_proc->i4_mb_cost = INT_MAX; |
| 1808 | ps_proc->i4_mb_distortion = INT_MAX; |
| 1809 | } |
| 1810 | else if (ps_proc->u4_mb_type == PSKIP) |
| 1811 | { |
| 1812 | goto UPDATE_MB_INFO; |
| 1813 | } |
| 1814 | |
| 1815 | /* wait until the proc of [top + 1] mb is computed. |
| 1816 | * We wait till the proc dependencies are satisfied */ |
| 1817 | if(ps_proc->i4_mb_y > 0) |
| 1818 | { |
| 1819 | /* proc map */ |
| 1820 | UWORD8 *pu1_proc_map_top; |
| 1821 | |
| 1822 | pu1_proc_map_top = ps_proc->pu1_proc_map + ((ps_proc->i4_mb_y - 1) * i4_wd_mbs); |
| 1823 | |
| 1824 | while (1) |
| 1825 | { |
| 1826 | volatile UWORD8 *pu1_buf; |
| 1827 | WORD32 idx = i4_mb_idx + 1; |
| 1828 | |
| 1829 | idx = MIN(idx, ((WORD32)ps_codec->s_cfg.i4_wd_mbs - 1)); |
| 1830 | pu1_buf = pu1_proc_map_top + idx; |
| 1831 | if(*pu1_buf) |
| 1832 | break; |
| 1833 | ithread_yield(); |
| 1834 | } |
| 1835 | } |
| 1836 | |
| 1837 | /* If we already have the minimum sad, there is no point in searching for sad again */ |
| 1838 | if (ps_proc->u4_min_sad_reached == 0) |
| 1839 | { |
| 1840 | /* intra gating in inter slices */ |
| 1841 | /* No need of gating if we want to force intra, we need to find the threshold only if inter is enabled by AIR*/ |
| 1842 | if (i4_air_enable_inter && ps_proc->i4_slice_type == PSLICE && ps_codec->u4_inter_gate) |
| 1843 | { |
| 1844 | /* distortion of neighboring blocks */ |
| 1845 | WORD32 i4_distortion[4]; |
| 1846 | |
| 1847 | i4_distortion[0] = ps_proc->s_left_mb_syntax_ele.i4_mb_distortion; |
| 1848 | |
| 1849 | i4_distortion[1] = ps_proc->ps_top_row_mb_syntax_ele[ps_proc->i4_mb_x].i4_mb_distortion; |
| 1850 | |
| 1851 | i4_distortion[2] = ps_proc->ps_top_row_mb_syntax_ele[ps_proc->i4_mb_x + 1].i4_mb_distortion; |
| 1852 | |
| 1853 | i4_distortion[3] = ps_proc->s_top_left_mb_syntax_ele.i4_mb_distortion; |
| 1854 | |
| 1855 | i4_gate_threshold = (i4_distortion[0] + i4_distortion[1] + i4_distortion[2] + i4_distortion[3]) >> 2; |
| 1856 | |
| 1857 | } |
| 1858 | |
| 1859 | /* If we are going to force intra we need to evaluate intra irrespective of gating */ |
| 1860 | if ( (!i4_air_enable_inter) || ((i4_gate_threshold + 16 *((WORD32) ps_proc->u4_lambda)) < ps_proc->i4_mb_distortion)) |
| 1861 | { |
| 1862 | /* evaluate intra 4x4 modes */ |
| 1863 | if (u4_valid_modes & (1 << I4x4)) |
| 1864 | { |
| 1865 | if (ps_codec->s_cfg.u4_enc_speed_preset == IVE_SLOWEST) |
| 1866 | { |
| 1867 | ih264e_evaluate_intra4x4_modes_for_least_cost_rdopton(ps_proc); |
| 1868 | } |
| 1869 | else |
| 1870 | { |
| 1871 | ih264e_evaluate_intra4x4_modes_for_least_cost_rdoptoff(ps_proc); |
| 1872 | } |
| 1873 | } |
| 1874 | |
| 1875 | /* evaluate intra 16x16 modes */ |
| 1876 | if (u4_valid_modes & (1 << I16x16)) |
| 1877 | { |
| 1878 | ih264e_evaluate_intra16x16_modes_for_least_cost_rdoptoff(ps_proc); |
| 1879 | } |
| 1880 | |
| 1881 | /* evaluate intra 8x8 modes */ |
| 1882 | if (u4_valid_modes & (1 << I8x8)) |
| 1883 | { |
| 1884 | ih264e_evaluate_intra8x8_modes_for_least_cost_rdoptoff(ps_proc); |
| 1885 | } |
| 1886 | } |
| 1887 | |
| 1888 | } |
| 1889 | } |
| 1890 | |
| 1891 | /* is intra */ |
| 1892 | if (ps_proc->u4_mb_type == I4x4 || ps_proc->u4_mb_type == I16x16 || ps_proc->u4_mb_type == I8x8) |
| 1893 | { |
| 1894 | luma_idx = ps_proc->u4_mb_type; |
| 1895 | chroma_idx = 0; |
| 1896 | is_intra = 1; |
| 1897 | |
| 1898 | /* evaluate chroma blocks for intra */ |
| 1899 | ih264e_evaluate_chroma_intra8x8_modes_for_least_cost_rdoptoff(ps_proc); |
| 1900 | } |
| 1901 | else |
| 1902 | { |
| 1903 | luma_idx = 3; |
| 1904 | chroma_idx = 1; |
| 1905 | is_intra = 0; |
| 1906 | } |
| 1907 | ps_proc->u4_is_intra = is_intra; |
| 1908 | |
| 1909 | /* redo MV pred of neighbors in the case intra mb */ |
| 1910 | /* TODO : currently called unconditionally, needs to be called only in the case of intra |
| 1911 | * to modify neighbors */ |
| 1912 | if (ps_proc->i4_slice_type != ISLICE) |
| 1913 | { |
| 1914 | ih264e_mv_pred(ps_proc); |
| 1915 | } |
| 1916 | |
| 1917 | /* Perform luma mb core coding */ |
| 1918 | u4_cbp_l = (ps_codec->luma_energy_compaction)[luma_idx](ps_proc); |
| 1919 | |
| 1920 | /* Perform luma mb core coding */ |
| 1921 | u4_cbp_c = (ps_codec->chroma_energy_compaction)[chroma_idx](ps_proc); |
| 1922 | |
| 1923 | /* coded block pattern */ |
| 1924 | ps_proc->u4_cbp = (u4_cbp_c << 4) | u4_cbp_l; |
| 1925 | |
| 1926 | /* mb skip */ |
| 1927 | if (is_intra == 0) |
| 1928 | { |
| 1929 | if (ps_proc->u4_cbp == 0) |
| 1930 | { |
| 1931 | /* get skip mv */ |
| 1932 | UWORD32 u4_for_me = 0; |
| 1933 | ih264e_find_skip_motion_vector(ps_proc,u4_for_me); |
| 1934 | |
| 1935 | /* skip ? */ |
| 1936 | if (ps_proc->ps_skip_mv->i2_mvx == ps_proc->ps_pu->s_l0_mv.i2_mvx && |
| 1937 | ps_proc->ps_skip_mv->i2_mvy == ps_proc->ps_pu->s_l0_mv.i2_mvy) |
| 1938 | { |
| 1939 | ps_proc->u4_mb_type = PSKIP; |
| 1940 | } |
| 1941 | } |
| 1942 | } |
| 1943 | |
| 1944 | UPDATE_MB_INFO: |
| 1945 | |
| 1946 | /* Update mb sad, mb qp and intra mb cost. Will be used by rate control */ |
| 1947 | ih264e_update_rc_mb_info(&ps_proc->s_frame_info, ps_proc); |
| 1948 | |
| 1949 | /**********************************************************************/ |
| 1950 | /* if disable deblock level is '0' this implies enable deblocking for */ |
| 1951 | /* all edges of all macroblocks with out any restrictions */ |
| 1952 | /* */ |
| 1953 | /* if disable deblock level is '1' this implies disable deblocking for*/ |
| 1954 | /* all edges of all macroblocks with out any restrictions */ |
| 1955 | /* */ |
| 1956 | /* if disable deblock level is '2' this implies enable deblocking for */ |
| 1957 | /* all edges of all macroblocks except edges overlapping with slice */ |
| 1958 | /* boundaries. This option is not currently supported by the encoder */ |
| 1959 | /* hence the slice map should be of no significance to perform debloc */ |
| 1960 | /* king */ |
| 1961 | /**********************************************************************/ |
| 1962 | |
| 1963 | if (ps_proc->u4_compute_recon) |
| 1964 | { |
| 1965 | /* deblk context */ |
| 1966 | /* src pointers */ |
| 1967 | UWORD8 *pu1_cur_pic_luma = ps_proc->pu1_rec_buf_luma; |
| 1968 | UWORD8 *pu1_cur_pic_chroma = ps_proc->pu1_rec_buf_chroma; |
| 1969 | |
| 1970 | /* src indices */ |
| 1971 | UWORD32 i4_mb_x = ps_proc->i4_mb_x; |
| 1972 | UWORD32 i4_mb_y = ps_proc->i4_mb_y; |
| 1973 | |
| 1974 | /* compute blocking strength */ |
| 1975 | if (ps_proc->u4_disable_deblock_level != 1) |
| 1976 | { |
| 1977 | ih264e_compute_bs(ps_proc); |
| 1978 | } |
| 1979 | |
| 1980 | /* nmb deblocking and hpel and padding */ |
| 1981 | ih264e_dblk_pad_hpel_processing_n_mbs(ps_proc, pu1_cur_pic_luma, |
| 1982 | pu1_cur_pic_chroma, i4_mb_x, |
| 1983 | i4_mb_y); |
| 1984 | } |
| 1985 | |
| 1986 | /* update the context after for coding next mb */ |
| 1987 | error_status |= ih264e_update_proc_ctxt(ps_proc); |
| 1988 | |
| 1989 | /* Once the last row is processed, mark the buffer status appropriately */ |
| 1990 | if (ps_proc->i4_ht_mbs == ps_proc->i4_mb_y) |
| 1991 | { |
| 1992 | /* Pointer to current picture buffer structure */ |
| 1993 | pic_buf_t *ps_cur_pic = ps_proc->ps_cur_pic; |
| 1994 | |
| 1995 | /* Pointer to current picture's mv buffer structure */ |
| 1996 | mv_buf_t *ps_cur_mv_buf = ps_proc->ps_cur_mv_buf; |
| 1997 | |
| 1998 | /**********************************************************************/ |
| 1999 | /* if disable deblock level is '0' this implies enable deblocking for */ |
| 2000 | /* all edges of all macroblocks with out any restrictions */ |
| 2001 | /* */ |
| 2002 | /* if disable deblock level is '1' this implies disable deblocking for*/ |
| 2003 | /* all edges of all macroblocks with out any restrictions */ |
| 2004 | /* */ |
| 2005 | /* if disable deblock level is '2' this implies enable deblocking for */ |
| 2006 | /* all edges of all macroblocks except edges overlapping with slice */ |
| 2007 | /* boundaries. This option is not currently supported by the encoder */ |
| 2008 | /* hence the slice map should be of no significance to perform debloc */ |
| 2009 | /* king */ |
| 2010 | /**********************************************************************/ |
| 2011 | error_status |= ih264_buf_mgr_release(ps_codec->pv_mv_buf_mgr, ps_cur_mv_buf->i4_buf_id , BUF_MGR_CODEC); |
| 2012 | |
| 2013 | error_status |= ih264_buf_mgr_release(ps_codec->pv_ref_buf_mgr, ps_cur_pic->i4_buf_id , BUF_MGR_CODEC); |
| 2014 | |
| 2015 | if (ps_codec->s_cfg.u4_enable_recon) |
| 2016 | { |
| 2017 | /* pic cnt */ |
| 2018 | ps_codec->as_rec_buf[ctxt_sel].i4_pic_cnt = ps_proc->i4_pic_cnt; |
| 2019 | |
| 2020 | /* rec buffers */ |
| 2021 | ps_codec->as_rec_buf[ctxt_sel].s_pic_buf = *ps_proc->ps_cur_pic; |
| 2022 | |
| 2023 | /* is last? */ |
| 2024 | ps_codec->as_rec_buf[ctxt_sel].u4_is_last = ps_proc->s_entropy.u4_is_last; |
| 2025 | |
| 2026 | /* frame time stamp */ |
| 2027 | ps_codec->as_rec_buf[ctxt_sel].u4_timestamp_high = ps_proc->s_entropy.u4_timestamp_high; |
| 2028 | ps_codec->as_rec_buf[ctxt_sel].u4_timestamp_low = ps_proc->s_entropy.u4_timestamp_low; |
| 2029 | } |
| 2030 | |
| 2031 | } |
| 2032 | } |
| 2033 | |
| 2034 | DEBUG_HISTOGRAM_DUMP(ps_codec->s_cfg.i4_ht_mbs == ps_proc->i4_mb_y); |
| 2035 | |
| 2036 | return error_status; |
| 2037 | } |
| 2038 | |
| 2039 | /** |
| 2040 | ******************************************************************************* |
| 2041 | * |
| 2042 | * @brief |
| 2043 | * function to receive frame qp and pic type before encoding |
| 2044 | * |
| 2045 | * @par Description: |
| 2046 | * Before encoding the frame, this function calls the rc library for frame qp |
| 2047 | * and picture type |
| 2048 | * |
| 2049 | * @param[in] ps_codec |
| 2050 | * Pointer to codec context |
| 2051 | * |
| 2052 | * @param[in] pic_cnt |
| 2053 | * pic count |
| 2054 | * |
| 2055 | * @param[out] pi4_pic_type |
| 2056 | * pic type |
| 2057 | |
| 2058 | * @returns skip_src |
| 2059 | * if the source frame rate and target frame rate are not identical, the encoder |
| 2060 | * skips few source frames. skip_src is set when the source need not be encoded. |
| 2061 | * |
| 2062 | * @remarks none |
| 2063 | * |
| 2064 | ******************************************************************************* |
| 2065 | */ |
| 2066 | WORD32 ih264e_set_rc_pic_params(codec_t *ps_codec, WORD32 cur_pic_cnt, WORD32 *pi4_pic_type) |
| 2067 | { |
| 2068 | /* rate control context */ |
| 2069 | rate_control_ctxt_t *ps_rate_control = &ps_codec->s_rate_control; |
| 2070 | |
| 2071 | /* frame qp */ |
| 2072 | UWORD8 u1_frame_qp; |
| 2073 | |
| 2074 | /* pic type */ |
| 2075 | PIC_TYPE_T pic_type = PIC_NA; |
| 2076 | |
| 2077 | /* should src be skipped */ |
| 2078 | WORD32 skip_src = 0; |
| 2079 | |
| 2080 | /* temp var */ |
| 2081 | WORD32 delta_time_stamp = 1; |
| 2082 | |
| 2083 | /* see if the app requires any specific frame */ |
| 2084 | if (ps_codec->force_curr_frame_type == IV_IDR_FRAME || ps_codec->force_curr_frame_type == IV_I_FRAME) |
| 2085 | { |
| 2086 | irc_force_I_frame(ps_codec->s_rate_control.pps_rate_control_api); |
| 2087 | } |
| 2088 | |
| 2089 | /* call rate control lib to get curr pic type and qp to be used */ |
| 2090 | skip_src = ih264e_rc_pre_enc(ps_rate_control->pps_rate_control_api, |
| 2091 | ps_rate_control->pps_pd_frm_rate, |
| 2092 | ps_rate_control->pps_time_stamp, |
| 2093 | ps_rate_control->pps_frame_time, |
| 2094 | delta_time_stamp, |
| 2095 | (ps_codec->s_cfg.i4_wd_mbs * ps_codec->s_cfg.i4_ht_mbs), |
| 2096 | &ps_rate_control->e_pic_type, |
| 2097 | &u1_frame_qp); |
| 2098 | |
| 2099 | switch (ps_rate_control->e_pic_type) |
| 2100 | { |
| 2101 | case I_PIC: |
| 2102 | pic_type = PIC_I; |
| 2103 | break; |
| 2104 | |
| 2105 | case P_PIC: |
| 2106 | pic_type = PIC_P; |
| 2107 | break; |
| 2108 | |
| 2109 | case B_PIC: |
| 2110 | pic_type = PIC_B; |
| 2111 | break; |
| 2112 | |
| 2113 | default: |
| 2114 | break; |
| 2115 | } |
| 2116 | |
| 2117 | /* is idr? */ |
| 2118 | if ((0 == cur_pic_cnt % ps_codec->s_cfg.u4_idr_frm_interval) || |
| 2119 | ps_codec->force_curr_frame_type == IV_IDR_FRAME) |
| 2120 | { |
| 2121 | pic_type = PIC_IDR; |
| 2122 | } |
| 2123 | |
| 2124 | /* force frame tag is not sticky */ |
| 2125 | if (ps_codec->force_curr_frame_type == IV_IDR_FRAME || ps_codec->force_curr_frame_type == IV_I_FRAME) |
| 2126 | { |
| 2127 | ps_codec->force_curr_frame_type = IV_NA_FRAME; |
| 2128 | } |
| 2129 | |
| 2130 | /* qp */ |
| 2131 | ps_codec->u4_frame_qp = gau1_mpeg2_to_h264_qmap[u1_frame_qp]; |
| 2132 | |
| 2133 | /* pic type */ |
| 2134 | *pi4_pic_type = pic_type; |
| 2135 | |
| 2136 | return skip_src; |
| 2137 | } |
| 2138 | |
| 2139 | /** |
| 2140 | ******************************************************************************* |
| 2141 | * |
| 2142 | * @brief |
| 2143 | * Function to update rc context after encoding |
| 2144 | * |
| 2145 | * @par Description |
| 2146 | * This function updates the rate control context after the frame is encoded. |
| 2147 | * Number of bits consumed by the current frame, frame distortion, frame cost, |
| 2148 | * number of intra/inter mb's, ... are passed on to rate control context for |
| 2149 | * updating the rc model. |
| 2150 | * |
| 2151 | * @param[in] ps_codec |
| 2152 | * Handle to codec context |
| 2153 | * |
| 2154 | * @param[in] ctxt_sel |
| 2155 | * frame context selector |
| 2156 | * |
| 2157 | * @param[in] pic_cnt |
| 2158 | * pic count |
| 2159 | * |
| 2160 | * @returns i4_stuffing_byte |
| 2161 | * number of stuffing bytes (if necessary) |
| 2162 | * |
| 2163 | * @remarks |
| 2164 | * |
| 2165 | ******************************************************************************* |
| 2166 | */ |
| 2167 | WORD32 ih264e_update_rc_post_enc(codec_t *ps_codec, WORD32 ctxt_sel, WORD32 pic_cnt) |
| 2168 | { |
| 2169 | /* proc set base idx */ |
| 2170 | WORD32 i4_proc_ctxt_sel_base = ctxt_sel ? (MAX_PROCESS_CTXT / 2) : 0; |
| 2171 | |
| 2172 | /* proc ctxt */ |
| 2173 | process_ctxt_t *ps_proc = &ps_codec->as_process[i4_proc_ctxt_sel_base]; |
| 2174 | |
| 2175 | /* frame qp */ |
| 2176 | UWORD8 u1_frame_qp = ps_codec->u4_frame_qp; |
| 2177 | |
| 2178 | /* cbr rc return status */ |
| 2179 | WORD32 i4_stuffing_byte = 0; |
| 2180 | |
| 2181 | /* current frame stats */ |
| 2182 | frame_info_t s_frame_info; |
| 2183 | picture_type_e rc_pic_type; |
| 2184 | |
| 2185 | /* temp var */ |
| 2186 | WORD32 i, j; |
| 2187 | |
| 2188 | /********************************************************************/ |
| 2189 | /* BEGIN INIT */ |
| 2190 | /********************************************************************/ |
| 2191 | |
| 2192 | /* init frame info */ |
| 2193 | irc_init_frame_info(&s_frame_info); |
| 2194 | |
| 2195 | /* get frame info */ |
| 2196 | for (i = 0; i < (WORD32)ps_codec->s_cfg.u4_num_cores; i++) |
| 2197 | { |
| 2198 | /*****************************************************************/ |
| 2199 | /* One frame can be encoded by max of u4_num_cores threads */ |
| 2200 | /* Accumulating the num mbs, sad, qp and intra_mb_cost from */ |
| 2201 | /* u4_num_cores threads */ |
| 2202 | /*****************************************************************/ |
| 2203 | for (j = 0; j< MAX_MB_TYPE; j++) |
| 2204 | { |
| 2205 | s_frame_info.num_mbs[j] += ps_proc[i].s_frame_info.num_mbs[j]; |
| 2206 | |
| 2207 | s_frame_info.tot_mb_sad[j] += ps_proc[i].s_frame_info.tot_mb_sad[j]; |
| 2208 | |
| 2209 | s_frame_info.qp_sum[j] += ps_proc[i].s_frame_info.qp_sum[j]; |
| 2210 | } |
| 2211 | |
| 2212 | s_frame_info.intra_mb_cost_sum += ps_proc[i].s_frame_info.intra_mb_cost_sum; |
| 2213 | |
| 2214 | s_frame_info.activity_sum += ps_proc[i].s_frame_info.activity_sum; |
| 2215 | |
| 2216 | /*****************************************************************/ |
| 2217 | /* gather number of residue and header bits consumed by the frame*/ |
| 2218 | /*****************************************************************/ |
| 2219 | ih264e_update_rc_bits_info(&s_frame_info, &ps_proc[i].s_entropy); |
| 2220 | } |
| 2221 | |
| 2222 | /* get pic type */ |
| 2223 | switch (ps_codec->pic_type) |
| 2224 | { |
| 2225 | case PIC_I: |
| 2226 | case PIC_IDR: |
| 2227 | rc_pic_type = I_PIC; |
| 2228 | break; |
| 2229 | case PIC_P: |
| 2230 | rc_pic_type = P_PIC; |
| 2231 | break; |
| 2232 | case PIC_B: |
| 2233 | rc_pic_type = B_PIC; |
| 2234 | break; |
| 2235 | default: |
| 2236 | assert(0); |
| 2237 | break; |
| 2238 | } |
| 2239 | |
| 2240 | /* update rc lib with current frame stats */ |
| 2241 | i4_stuffing_byte = ih264e_rc_post_enc(ps_codec->s_rate_control.pps_rate_control_api, |
| 2242 | &(s_frame_info), |
| 2243 | ps_codec->s_rate_control.pps_pd_frm_rate, |
| 2244 | ps_codec->s_rate_control.pps_time_stamp, |
| 2245 | ps_codec->s_rate_control.pps_frame_time, |
| 2246 | (ps_proc->i4_wd_mbs * ps_proc->i4_ht_mbs), |
| 2247 | &rc_pic_type, |
| 2248 | pic_cnt, |
| 2249 | &ps_codec->s_rate_control.post_encode_skip[ctxt_sel], |
| 2250 | u1_frame_qp, |
| 2251 | &ps_codec->s_rate_control.num_intra_in_prev_frame, |
| 2252 | &ps_codec->s_rate_control.i4_avg_activity); |
| 2253 | |
| 2254 | /* in case the frame needs to be skipped, the frame num should not be incremented */ |
| 2255 | if (ps_codec->s_rate_control.post_encode_skip[ctxt_sel]) |
| 2256 | { |
| 2257 | ps_codec->i4_frame_num --; |
| 2258 | } |
| 2259 | |
| 2260 | return i4_stuffing_byte; |
| 2261 | } |
| 2262 | |
| 2263 | /** |
| 2264 | ******************************************************************************* |
| 2265 | * |
| 2266 | * @brief |
| 2267 | * entry point of a spawned encoder thread |
| 2268 | * |
| 2269 | * @par Description: |
| 2270 | * The encoder thread dequeues a proc/entropy job from the encoder queue and |
| 2271 | * calls necessary routines. |
| 2272 | * |
| 2273 | * @param[in] pv_proc |
| 2274 | * Process context corresponding to the thread |
| 2275 | * |
| 2276 | * @returns error status |
| 2277 | * |
| 2278 | * @remarks |
| 2279 | * |
| 2280 | ******************************************************************************* |
| 2281 | */ |
| 2282 | WORD32 ih264e_process_thread(void *pv_proc) |
| 2283 | { |
| 2284 | /* error status */ |
| 2285 | IH264_ERROR_T ret = IH264_SUCCESS; |
| 2286 | WORD32 error_status = IH264_SUCCESS; |
| 2287 | |
| 2288 | /* proc ctxt */ |
| 2289 | process_ctxt_t *ps_proc = pv_proc; |
| 2290 | |
| 2291 | /* codec ctxt */ |
| 2292 | codec_t *ps_codec = ps_proc->ps_codec; |
| 2293 | |
| 2294 | /* structure to represent a processing job entry */ |
| 2295 | job_t s_job; |
| 2296 | |
| 2297 | /* blocking call : entropy dequeue is non-blocking till all |
| 2298 | * the proc jobs are processed */ |
| 2299 | WORD32 is_blocking = 0; |
| 2300 | |
| 2301 | /* set affinity */ |
| 2302 | ithread_set_affinity(ps_proc->i4_id); |
| 2303 | |
| 2304 | while(1) |
| 2305 | { |
| 2306 | /* dequeue a job from the entropy queue */ |
| 2307 | { |
| 2308 | int error = ithread_mutex_lock(ps_codec->pv_entropy_mutex); |
| 2309 | |
| 2310 | /* codec context selector */ |
| 2311 | WORD32 ctxt_sel = ps_codec->i4_encode_api_call_cnt & 1; |
| 2312 | |
| 2313 | volatile UWORD32 *pu4_buf = &ps_codec->au4_entropy_thread_active[ctxt_sel]; |
| 2314 | |
| 2315 | /* have the lock */ |
| 2316 | if (error == 0) |
| 2317 | { |
| 2318 | if (*pu4_buf == 0) |
| 2319 | { |
| 2320 | /* no entropy threads are active, try dequeuing a job from the entropy queue */ |
| 2321 | ret = ih264_list_dequeue(ps_proc->pv_entropy_jobq, &s_job, is_blocking); |
| 2322 | if (IH264_SUCCESS == ret) |
| 2323 | { |
| 2324 | *pu4_buf = 1; |
| 2325 | ithread_mutex_unlock(ps_codec->pv_entropy_mutex); |
| 2326 | goto WORKER; |
| 2327 | } |
| 2328 | else if(is_blocking) |
| 2329 | { |
| 2330 | ithread_mutex_unlock(ps_codec->pv_entropy_mutex); |
| 2331 | break; |
| 2332 | } |
| 2333 | } |
| 2334 | ithread_mutex_unlock(ps_codec->pv_entropy_mutex); |
| 2335 | } |
| 2336 | } |
| 2337 | |
| 2338 | /* dequeue a job from the process queue */ |
| 2339 | ret = ih264_list_dequeue(ps_proc->pv_proc_jobq, &s_job, 1); |
| 2340 | if (IH264_SUCCESS != ret) |
| 2341 | { |
| 2342 | if(ps_proc->i4_id) |
| 2343 | break; |
| 2344 | else |
| 2345 | { |
| 2346 | is_blocking = 1; |
| 2347 | continue; |
| 2348 | } |
| 2349 | } |
| 2350 | |
| 2351 | WORKER: |
| 2352 | /* choose appropriate proc context based on proc_base_idx */ |
| 2353 | ps_proc = &ps_codec->as_process[ps_proc->i4_id + s_job.i2_proc_base_idx]; |
| 2354 | |
| 2355 | switch (s_job.i4_cmd) |
| 2356 | { |
| 2357 | case CMD_PROCESS: |
| 2358 | ps_proc->i4_mb_cnt = s_job.i2_mb_cnt; |
| 2359 | ps_proc->i4_mb_x = s_job.i2_mb_x; |
| 2360 | ps_proc->i4_mb_y = s_job.i2_mb_y; |
| 2361 | |
| 2362 | /* init process context */ |
| 2363 | ih264e_init_proc_ctxt(ps_proc); |
| 2364 | |
| 2365 | /* core code all mbs enlisted under the current job */ |
| 2366 | error_status |= ih264e_process(ps_proc); |
| 2367 | break; |
| 2368 | |
| 2369 | case CMD_ENTROPY: |
| 2370 | ps_proc->s_entropy.i4_mb_x = s_job.i2_mb_x; |
| 2371 | ps_proc->s_entropy.i4_mb_y = s_job.i2_mb_y; |
| 2372 | ps_proc->s_entropy.i4_mb_cnt = s_job.i2_mb_cnt; |
| 2373 | |
| 2374 | /* init entropy */ |
| 2375 | ih264e_init_entropy_ctxt(ps_proc); |
| 2376 | |
| 2377 | /* entropy code all mbs enlisted under the current job */ |
| 2378 | error_status |= ih264e_entropy(ps_proc); |
| 2379 | break; |
| 2380 | |
| 2381 | default: |
| 2382 | error_status |= IH264_FAIL; |
| 2383 | break; |
| 2384 | } |
| 2385 | } |
| 2386 | |
| 2387 | /* send error code */ |
| 2388 | ps_proc->i4_error_code = error_status; |
| 2389 | return ret; |
| 2390 | } |