/src/libhevc/encoder/hme_refine.c
Line | Count | Source (jump to first uncovered line) |
1 | | /****************************************************************************** |
2 | | * |
3 | | * Copyright (C) 2018 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 | | * @file hme_refine.c |
23 | | * |
24 | | * @brief |
25 | | * Contains the implementation of the refinement layer searches and related |
26 | | * functionality like CU merge. |
27 | | * |
28 | | * @author |
29 | | * Ittiam |
30 | | * |
31 | | * |
32 | | * List of Functions |
33 | | * |
34 | | * |
35 | | ****************************************************************************** |
36 | | */ |
37 | | |
38 | | /*****************************************************************************/ |
39 | | /* File Includes */ |
40 | | /*****************************************************************************/ |
41 | | /* System include files */ |
42 | | #include <stdio.h> |
43 | | #include <string.h> |
44 | | #include <stdlib.h> |
45 | | #include <assert.h> |
46 | | #include <stdarg.h> |
47 | | #include <math.h> |
48 | | #include <limits.h> |
49 | | |
50 | | /* User include files */ |
51 | | #include "ihevc_typedefs.h" |
52 | | #include "itt_video_api.h" |
53 | | #include "ihevce_api.h" |
54 | | |
55 | | #include "rc_cntrl_param.h" |
56 | | #include "rc_frame_info_collector.h" |
57 | | #include "rc_look_ahead_params.h" |
58 | | |
59 | | #include "ihevc_defs.h" |
60 | | #include "ihevc_structs.h" |
61 | | #include "ihevc_platform_macros.h" |
62 | | #include "ihevc_deblk.h" |
63 | | #include "ihevc_itrans_recon.h" |
64 | | #include "ihevc_chroma_itrans_recon.h" |
65 | | #include "ihevc_chroma_intra_pred.h" |
66 | | #include "ihevc_intra_pred.h" |
67 | | #include "ihevc_inter_pred.h" |
68 | | #include "ihevc_mem_fns.h" |
69 | | #include "ihevc_padding.h" |
70 | | #include "ihevc_weighted_pred.h" |
71 | | #include "ihevc_sao.h" |
72 | | #include "ihevc_resi_trans.h" |
73 | | #include "ihevc_quant_iquant_ssd.h" |
74 | | #include "ihevc_cabac_tables.h" |
75 | | |
76 | | #include "ihevce_defs.h" |
77 | | #include "ihevce_lap_enc_structs.h" |
78 | | #include "ihevce_multi_thrd_structs.h" |
79 | | #include "ihevce_multi_thrd_funcs.h" |
80 | | #include "ihevce_me_common_defs.h" |
81 | | #include "ihevce_had_satd.h" |
82 | | #include "ihevce_error_codes.h" |
83 | | #include "ihevce_bitstream.h" |
84 | | #include "ihevce_cabac.h" |
85 | | #include "ihevce_rdoq_macros.h" |
86 | | #include "ihevce_function_selector.h" |
87 | | #include "ihevce_enc_structs.h" |
88 | | #include "ihevce_entropy_structs.h" |
89 | | #include "ihevce_cmn_utils_instr_set_router.h" |
90 | | #include "ihevce_enc_loop_structs.h" |
91 | | #include "ihevce_bs_compute_ctb.h" |
92 | | #include "ihevce_global_tables.h" |
93 | | #include "ihevce_dep_mngr_interface.h" |
94 | | #include "hme_datatype.h" |
95 | | #include "hme_interface.h" |
96 | | #include "hme_common_defs.h" |
97 | | #include "hme_defs.h" |
98 | | #include "ihevce_me_instr_set_router.h" |
99 | | #include "hme_globals.h" |
100 | | #include "hme_utils.h" |
101 | | #include "hme_coarse.h" |
102 | | #include "hme_fullpel.h" |
103 | | #include "hme_subpel.h" |
104 | | #include "hme_refine.h" |
105 | | #include "hme_err_compute.h" |
106 | | #include "hme_common_utils.h" |
107 | | #include "hme_search_algo.h" |
108 | | #include "ihevce_stasino_helpers.h" |
109 | | #include "ihevce_common_utils.h" |
110 | | |
111 | | /*****************************************************************************/ |
112 | | /* Globals */ |
113 | | /*****************************************************************************/ |
114 | | |
115 | | /* brief: mapping buffer to convert raster scan indices into z-scan oder in a ctb */ |
116 | | UWORD8 gau1_raster_scan_to_ctb[4][4] = { |
117 | | { 0, 4, 16, 20 }, { 8, 12, 24, 28 }, { 32, 36, 48, 52 }, { 40, 44, 56, 60 } |
118 | | }; |
119 | | |
120 | | /*****************************************************************************/ |
121 | | /* Extern Fucntion declaration */ |
122 | | /*****************************************************************************/ |
123 | | extern ctb_boundary_attrs_t * |
124 | | get_ctb_attrs(S32 ctb_start_x, S32 ctb_start_y, S32 pic_wd, S32 pic_ht, me_frm_ctxt_t *ps_ctxt); |
125 | | |
126 | | typedef void (*PF_HME_PROJECT_COLOC_CANDT_FXN)( |
127 | | search_node_t *ps_search_node, |
128 | | layer_ctxt_t *ps_curr_layer, |
129 | | layer_ctxt_t *ps_coarse_layer, |
130 | | S32 i4_pos_x, |
131 | | S32 i4_pos_y, |
132 | | S08 i1_ref_id, |
133 | | S32 i4_result_id); |
134 | | |
135 | | typedef void (*PF_HME_PROJECT_COLOC_CANDT_L0_ME_FXN)( |
136 | | search_node_t *ps_search_node, |
137 | | layer_ctxt_t *ps_curr_layer, |
138 | | layer_ctxt_t *ps_coarse_layer, |
139 | | S32 i4_pos_x, |
140 | | S32 i4_pos_y, |
141 | | S32 i4_num_act_ref_l0, |
142 | | U08 u1_pred_dir, |
143 | | U08 u1_default_ref_id, |
144 | | S32 i4_result_id); |
145 | | |
146 | | /*****************************************************************************/ |
147 | | /* Function Definitions */ |
148 | | /*****************************************************************************/ |
149 | | |
150 | | void ihevce_no_wt_copy( |
151 | | coarse_me_ctxt_t *ps_ctxt, |
152 | | layer_ctxt_t *ps_curr_layer, |
153 | | pu_t *ps_pu, |
154 | | UWORD8 *pu1_temp_pred, |
155 | | WORD32 temp_stride, |
156 | | WORD32 blk_x, |
157 | | WORD32 blk_y) |
158 | 0 | { |
159 | 0 | UWORD8 *pu1_ref; |
160 | 0 | WORD32 ref_stride, ref_offset; |
161 | 0 | WORD32 row, col, i4_tmp; |
162 | |
|
163 | 0 | ASSERT((ps_pu->b2_pred_mode == PRED_L0) || (ps_pu->b2_pred_mode == PRED_L1)); |
164 | | |
165 | 0 | if(ps_pu->b2_pred_mode == PRED_L0) |
166 | 0 | { |
167 | 0 | WORD8 i1_ref_idx; |
168 | |
|
169 | 0 | i1_ref_idx = ps_pu->mv.i1_l0_ref_idx; |
170 | 0 | pu1_ref = ps_curr_layer->ppu1_list_inp[i1_ref_idx]; |
171 | |
|
172 | 0 | ref_stride = ps_curr_layer->i4_inp_stride; |
173 | |
|
174 | 0 | ref_offset = ((blk_y << 3) + ps_pu->mv.s_l0_mv.i2_mvy) * ref_stride; |
175 | 0 | ref_offset += (blk_x << 3) + ps_pu->mv.s_l0_mv.i2_mvx; |
176 | |
|
177 | 0 | pu1_ref += ref_offset; |
178 | |
|
179 | 0 | for(row = 0; row < temp_stride; row++) |
180 | 0 | { |
181 | 0 | for(col = 0; col < temp_stride; col++) |
182 | 0 | { |
183 | 0 | i4_tmp = pu1_ref[col]; |
184 | 0 | pu1_temp_pred[col] = CLIP_U8(i4_tmp); |
185 | 0 | } |
186 | |
|
187 | 0 | pu1_ref += ref_stride; |
188 | 0 | pu1_temp_pred += temp_stride; |
189 | 0 | } |
190 | 0 | } |
191 | 0 | else |
192 | 0 | { |
193 | 0 | WORD8 i1_ref_idx; |
194 | |
|
195 | 0 | i1_ref_idx = ps_pu->mv.i1_l1_ref_idx; |
196 | 0 | pu1_ref = ps_curr_layer->ppu1_list_inp[i1_ref_idx]; |
197 | |
|
198 | 0 | ref_stride = ps_curr_layer->i4_inp_stride; |
199 | |
|
200 | 0 | ref_offset = ((blk_y << 3) + ps_pu->mv.s_l1_mv.i2_mvy) * ref_stride; |
201 | 0 | ref_offset += (blk_x << 3) + ps_pu->mv.s_l1_mv.i2_mvx; |
202 | |
|
203 | 0 | pu1_ref += ref_offset; |
204 | |
|
205 | 0 | for(row = 0; row < temp_stride; row++) |
206 | 0 | { |
207 | 0 | for(col = 0; col < temp_stride; col++) |
208 | 0 | { |
209 | 0 | i4_tmp = pu1_ref[col]; |
210 | 0 | pu1_temp_pred[col] = CLIP_U8(i4_tmp); |
211 | 0 | } |
212 | |
|
213 | 0 | pu1_ref += ref_stride; |
214 | 0 | pu1_temp_pred += temp_stride; |
215 | 0 | } |
216 | 0 | } |
217 | 0 | } |
218 | | |
219 | | static WORD32 hme_add_clustered_mvs_as_merge_cands( |
220 | | cluster_data_t *ps_cluster_base, |
221 | | search_node_t *ps_merge_cand, |
222 | | range_prms_t **pps_range_prms, |
223 | | U08 *pu1_refid_to_pred_dir_list, |
224 | | WORD32 i4_num_clusters, |
225 | | U08 u1_pred_dir) |
226 | 0 | { |
227 | 0 | WORD32 i, j, k; |
228 | 0 | WORD32 i4_num_cands_added = 0; |
229 | 0 | WORD32 i4_num_mvs_in_cluster; |
230 | |
|
231 | 0 | for(i = 0; i < i4_num_clusters; i++) |
232 | 0 | { |
233 | 0 | cluster_data_t *ps_data = &ps_cluster_base[i]; |
234 | |
|
235 | 0 | if(u1_pred_dir == !pu1_refid_to_pred_dir_list[ps_data->ref_id]) |
236 | 0 | { |
237 | 0 | i4_num_mvs_in_cluster = ps_data->num_mvs; |
238 | |
|
239 | 0 | for(j = 0; j < i4_num_mvs_in_cluster; j++) |
240 | 0 | { |
241 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvx = ps_data->as_mv[j].mvx; |
242 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvy = ps_data->as_mv[j].mvy; |
243 | 0 | ps_merge_cand[i4_num_cands_added].i1_ref_idx = ps_data->ref_id; |
244 | |
|
245 | 0 | CLIP_MV_WITHIN_RANGE( |
246 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvx, |
247 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvy, |
248 | 0 | pps_range_prms[ps_data->ref_id], |
249 | 0 | 0, |
250 | 0 | 0, |
251 | 0 | 0); |
252 | |
|
253 | 0 | for(k = 0; k < i4_num_cands_added; k++) |
254 | 0 | { |
255 | 0 | if((ps_merge_cand[k].s_mv.i2_mvx == ps_data->as_mv[j].mvx) && |
256 | 0 | (ps_merge_cand[k].s_mv.i2_mvy == ps_data->as_mv[j].mvy) && |
257 | 0 | (ps_merge_cand[k].i1_ref_idx == ps_data->ref_id)) |
258 | 0 | { |
259 | 0 | break; |
260 | 0 | } |
261 | 0 | } |
262 | |
|
263 | 0 | if(k == i4_num_cands_added) |
264 | 0 | { |
265 | 0 | i4_num_cands_added++; |
266 | 0 | } |
267 | 0 | } |
268 | 0 | } |
269 | 0 | } |
270 | |
|
271 | 0 | return i4_num_cands_added; |
272 | 0 | } |
273 | | |
274 | | static WORD32 hme_add_me_best_as_merge_cands( |
275 | | search_results_t **pps_child_data_array, |
276 | | inter_cu_results_t *ps_8x8cu_results, |
277 | | search_node_t *ps_merge_cand, |
278 | | range_prms_t **pps_range_prms, |
279 | | U08 *pu1_refid_to_pred_dir_list, |
280 | | S08 *pi1_past_list, |
281 | | S08 *pi1_future_list, |
282 | | BLK_SIZE_T e_blk_size, |
283 | | ME_QUALITY_PRESETS_T e_quality_preset, |
284 | | S32 i4_num_cands_added, |
285 | | U08 u1_pred_dir) |
286 | 0 | { |
287 | 0 | WORD32 i, j, k; |
288 | 0 | WORD32 i4_max_cands_to_add; |
289 | |
|
290 | 0 | WORD32 i4_result_id = 0; |
291 | |
|
292 | 0 | ASSERT(!pps_child_data_array[0]->u1_split_flag || (BLK_64x64 != e_blk_size)); |
293 | 0 | ASSERT(!pps_child_data_array[1]->u1_split_flag || (BLK_64x64 != e_blk_size)); |
294 | 0 | ASSERT(!pps_child_data_array[2]->u1_split_flag || (BLK_64x64 != e_blk_size)); |
295 | 0 | ASSERT(!pps_child_data_array[3]->u1_split_flag || (BLK_64x64 != e_blk_size)); |
296 | | |
297 | 0 | switch(e_quality_preset) |
298 | 0 | { |
299 | 0 | case ME_PRISTINE_QUALITY: |
300 | 0 | { |
301 | 0 | i4_max_cands_to_add = MAX_MERGE_CANDTS; |
302 | |
|
303 | 0 | break; |
304 | 0 | } |
305 | 0 | case ME_HIGH_QUALITY: |
306 | 0 | { |
307 | | /* All 4 children are split and each grandchild contributes an MV */ |
308 | | /* and 2 best results per grandchild */ |
309 | 0 | i4_max_cands_to_add = 4 * 4 * 2; |
310 | |
|
311 | 0 | break; |
312 | 0 | } |
313 | 0 | case ME_MEDIUM_SPEED: |
314 | 0 | { |
315 | 0 | i4_max_cands_to_add = 4 * 2 * 2; |
316 | |
|
317 | 0 | break; |
318 | 0 | } |
319 | 0 | case ME_HIGH_SPEED: |
320 | 0 | case ME_XTREME_SPEED: |
321 | 0 | case ME_XTREME_SPEED_25: |
322 | 0 | { |
323 | 0 | i4_max_cands_to_add = 4 * 2 * 1; |
324 | |
|
325 | 0 | break; |
326 | 0 | } |
327 | 0 | } |
328 | | |
329 | 0 | while(i4_result_id < 4) |
330 | 0 | { |
331 | 0 | for(i = 0; i < 4; i++) |
332 | 0 | { |
333 | 0 | inter_cu_results_t *ps_child_data = pps_child_data_array[i]->ps_cu_results; |
334 | 0 | inter_cu_results_t *ps_grandchild_data = &ps_8x8cu_results[i << 2]; |
335 | |
|
336 | 0 | if(!pps_child_data_array[i]->u1_split_flag) |
337 | 0 | { |
338 | 0 | part_type_results_t *ps_data = &ps_child_data->ps_best_results[i4_result_id]; |
339 | |
|
340 | 0 | if(ps_child_data->u1_num_best_results <= i4_result_id) |
341 | 0 | { |
342 | 0 | continue; |
343 | 0 | } |
344 | | |
345 | 0 | if(ps_data->as_pu_results->pu.b1_intra_flag) |
346 | 0 | { |
347 | 0 | continue; |
348 | 0 | } |
349 | | |
350 | 0 | for(j = 0; j <= (ps_data->u1_part_type != PRT_2Nx2N); j++) |
351 | 0 | { |
352 | 0 | mv_t *ps_mv; |
353 | |
|
354 | 0 | S08 i1_ref_idx; |
355 | |
|
356 | 0 | pu_t *ps_pu = &ps_data->as_pu_results[j].pu; |
357 | |
|
358 | 0 | if(u1_pred_dir != |
359 | 0 | ((ps_pu->b2_pred_mode == 2) ? u1_pred_dir : ps_pu->b2_pred_mode)) |
360 | 0 | { |
361 | 0 | continue; |
362 | 0 | } |
363 | | |
364 | 0 | if(u1_pred_dir) |
365 | 0 | { |
366 | 0 | ps_mv = &ps_pu->mv.s_l1_mv; |
367 | 0 | i1_ref_idx = pi1_future_list[ps_pu->mv.i1_l1_ref_idx]; |
368 | 0 | } |
369 | 0 | else |
370 | 0 | { |
371 | 0 | ps_mv = &ps_pu->mv.s_l0_mv; |
372 | 0 | i1_ref_idx = pi1_past_list[ps_pu->mv.i1_l0_ref_idx]; |
373 | 0 | } |
374 | |
|
375 | 0 | if(-1 == i1_ref_idx) |
376 | 0 | { |
377 | 0 | continue; |
378 | 0 | } |
379 | | |
380 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvx = ps_mv->i2_mvx; |
381 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvy = ps_mv->i2_mvy; |
382 | 0 | ps_merge_cand[i4_num_cands_added].i1_ref_idx = i1_ref_idx; |
383 | |
|
384 | 0 | CLIP_MV_WITHIN_RANGE( |
385 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvx, |
386 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvy, |
387 | 0 | pps_range_prms[i1_ref_idx], |
388 | 0 | 0, |
389 | 0 | 0, |
390 | 0 | 0); |
391 | |
|
392 | 0 | for(k = 0; k < i4_num_cands_added; k++) |
393 | 0 | { |
394 | 0 | if((ps_merge_cand[k].s_mv.i2_mvx == ps_mv->i2_mvx) && |
395 | 0 | (ps_merge_cand[k].s_mv.i2_mvy == ps_mv->i2_mvy) && |
396 | 0 | (ps_merge_cand[k].i1_ref_idx == i1_ref_idx)) |
397 | 0 | { |
398 | 0 | break; |
399 | 0 | } |
400 | 0 | } |
401 | |
|
402 | 0 | if(k == i4_num_cands_added) |
403 | 0 | { |
404 | 0 | i4_num_cands_added++; |
405 | |
|
406 | 0 | if(i4_max_cands_to_add <= i4_num_cands_added) |
407 | 0 | { |
408 | 0 | return i4_num_cands_added; |
409 | 0 | } |
410 | 0 | } |
411 | 0 | } |
412 | 0 | } |
413 | 0 | else |
414 | 0 | { |
415 | 0 | for(j = 0; j < 4; j++) |
416 | 0 | { |
417 | 0 | mv_t *ps_mv; |
418 | |
|
419 | 0 | S08 i1_ref_idx; |
420 | |
|
421 | 0 | part_type_results_t *ps_data = ps_grandchild_data[j].ps_best_results; |
422 | 0 | pu_t *ps_pu = &ps_data->as_pu_results[0].pu; |
423 | |
|
424 | 0 | ASSERT(ps_data->u1_part_type == PRT_2Nx2N); |
425 | | |
426 | 0 | if(ps_grandchild_data[j].u1_num_best_results <= i4_result_id) |
427 | 0 | { |
428 | 0 | continue; |
429 | 0 | } |
430 | | |
431 | 0 | if(ps_data->as_pu_results->pu.b1_intra_flag) |
432 | 0 | { |
433 | 0 | continue; |
434 | 0 | } |
435 | | |
436 | 0 | if(u1_pred_dir != |
437 | 0 | ((ps_pu->b2_pred_mode == 2) ? u1_pred_dir : ps_pu->b2_pred_mode)) |
438 | 0 | { |
439 | 0 | continue; |
440 | 0 | } |
441 | | |
442 | 0 | if(u1_pred_dir) |
443 | 0 | { |
444 | 0 | ps_mv = &ps_pu->mv.s_l1_mv; |
445 | 0 | i1_ref_idx = pi1_future_list[ps_pu->mv.i1_l1_ref_idx]; |
446 | 0 | } |
447 | 0 | else |
448 | 0 | { |
449 | 0 | ps_mv = &ps_pu->mv.s_l0_mv; |
450 | 0 | i1_ref_idx = pi1_past_list[ps_pu->mv.i1_l0_ref_idx]; |
451 | 0 | } |
452 | |
|
453 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvx = ps_mv->i2_mvx; |
454 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvy = ps_mv->i2_mvy; |
455 | 0 | ps_merge_cand[i4_num_cands_added].i1_ref_idx = i1_ref_idx; |
456 | |
|
457 | 0 | CLIP_MV_WITHIN_RANGE( |
458 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvx, |
459 | 0 | ps_merge_cand[i4_num_cands_added].s_mv.i2_mvy, |
460 | 0 | pps_range_prms[i1_ref_idx], |
461 | 0 | 0, |
462 | 0 | 0, |
463 | 0 | 0); |
464 | |
|
465 | 0 | for(k = 0; k < i4_num_cands_added; k++) |
466 | 0 | { |
467 | 0 | if((ps_merge_cand[k].s_mv.i2_mvx == ps_mv->i2_mvx) && |
468 | 0 | (ps_merge_cand[k].s_mv.i2_mvy == ps_mv->i2_mvy) && |
469 | 0 | (ps_merge_cand[k].i1_ref_idx == i1_ref_idx)) |
470 | 0 | { |
471 | 0 | break; |
472 | 0 | } |
473 | 0 | } |
474 | |
|
475 | 0 | if(k == i4_num_cands_added) |
476 | 0 | { |
477 | 0 | i4_num_cands_added++; |
478 | |
|
479 | 0 | if(i4_max_cands_to_add <= i4_num_cands_added) |
480 | 0 | { |
481 | 0 | return i4_num_cands_added; |
482 | 0 | } |
483 | 0 | } |
484 | 0 | } |
485 | 0 | } |
486 | 0 | } |
487 | | |
488 | 0 | i4_result_id++; |
489 | 0 | } |
490 | | |
491 | 0 | return i4_num_cands_added; |
492 | 0 | } |
493 | | |
494 | | WORD32 hme_add_cands_for_merge_eval( |
495 | | ctb_cluster_info_t *ps_cluster_info, |
496 | | search_results_t **pps_child_data_array, |
497 | | inter_cu_results_t *ps_8x8cu_results, |
498 | | range_prms_t **pps_range_prms, |
499 | | search_node_t *ps_merge_cand, |
500 | | U08 *pu1_refid_to_pred_dir_list, |
501 | | S08 *pi1_past_list, |
502 | | S08 *pi1_future_list, |
503 | | ME_QUALITY_PRESETS_T e_quality_preset, |
504 | | BLK_SIZE_T e_blk_size, |
505 | | U08 u1_pred_dir, |
506 | | U08 u1_blk_id) |
507 | 0 | { |
508 | 0 | WORD32 i4_num_cands_added = 0; |
509 | |
|
510 | 0 | if(ME_PRISTINE_QUALITY == e_quality_preset) |
511 | 0 | { |
512 | 0 | cluster_data_t *ps_cluster_primo; |
513 | |
|
514 | 0 | WORD32 i4_num_clusters; |
515 | |
|
516 | 0 | if(BLK_32x32 == e_blk_size) |
517 | 0 | { |
518 | 0 | ps_cluster_primo = ps_cluster_info->ps_32x32_blk[u1_blk_id].as_cluster_data; |
519 | 0 | i4_num_clusters = ps_cluster_info->ps_32x32_blk[u1_blk_id].num_clusters; |
520 | 0 | } |
521 | 0 | else |
522 | 0 | { |
523 | 0 | ps_cluster_primo = ps_cluster_info->ps_64x64_blk->as_cluster_data; |
524 | 0 | i4_num_clusters = ps_cluster_info->ps_64x64_blk->num_clusters; |
525 | 0 | } |
526 | |
|
527 | 0 | i4_num_cands_added = hme_add_clustered_mvs_as_merge_cands( |
528 | 0 | ps_cluster_primo, |
529 | 0 | ps_merge_cand, |
530 | 0 | pps_range_prms, |
531 | 0 | pu1_refid_to_pred_dir_list, |
532 | 0 | i4_num_clusters, |
533 | 0 | u1_pred_dir); |
534 | 0 | } |
535 | |
|
536 | 0 | i4_num_cands_added = hme_add_me_best_as_merge_cands( |
537 | 0 | pps_child_data_array, |
538 | 0 | ps_8x8cu_results, |
539 | 0 | ps_merge_cand, |
540 | 0 | pps_range_prms, |
541 | 0 | pu1_refid_to_pred_dir_list, |
542 | 0 | pi1_past_list, |
543 | 0 | pi1_future_list, |
544 | 0 | e_blk_size, |
545 | 0 | e_quality_preset, |
546 | 0 | i4_num_cands_added, |
547 | 0 | u1_pred_dir); |
548 | |
|
549 | 0 | return i4_num_cands_added; |
550 | 0 | } |
551 | | |
552 | | /** |
553 | | ******************************************************************************** |
554 | | * @fn void hme_pick_refine_merge_candts(hme_merge_prms_t *ps_merge_prms, |
555 | | * S08 i1_ref_idx, |
556 | | * S32 i4_best_part_type, |
557 | | * S32 i4_is_vert) |
558 | | * |
559 | | * @brief Given a target partition orientation in the merged CU, and the |
560 | | * partition type of most likely partition this fxn picks up |
561 | | * candidates from the 4 constituent CUs and does refinement search |
562 | | * to identify best results for the merge CU across active partitions |
563 | | * |
564 | | * @param[in,out] ps_merge_prms : Parameters sent from higher layers. Out of |
565 | | * these params, the search result structure is also derived and |
566 | | * updated during the search |
567 | | * |
568 | | * @param[in] i1_ref_idx : ID of the buffer within the search results to update. |
569 | | * Will be 0 if all refidx collapsed to one buf, else it'll be 0/1 |
570 | | * |
571 | | * @param[in] i4_best_part_type : partition type of potential partition in the |
572 | | * merged CU, -1 if the merge process has not yet been able to |
573 | | * determine this. |
574 | | * |
575 | | * @param[in] i4_is_vert : Whether target partition of merged CU is vertical |
576 | | * orientation or horizontal orientation. |
577 | | * |
578 | | * @return Number of merge candidates |
579 | | ******************************************************************************** |
580 | | */ |
581 | | WORD32 hme_pick_eval_merge_candts( |
582 | | hme_merge_prms_t *ps_merge_prms, |
583 | | hme_subpel_prms_t *ps_subpel_prms, |
584 | | S32 i4_search_idx, |
585 | | S32 i4_best_part_type, |
586 | | S32 i4_is_vert, |
587 | | wgt_pred_ctxt_t *ps_wt_inp_prms, |
588 | | S32 i4_frm_qstep, |
589 | | ihevce_cmn_opt_func_t *ps_cmn_utils_optimised_function_list, |
590 | | ihevce_me_optimised_function_list_t *ps_me_optimised_function_list) |
591 | 0 | { |
592 | 0 | S32 x_off, y_off; |
593 | 0 | search_node_t *ps_search_node; |
594 | 0 | S32 ai4_valid_part_ids[TOT_NUM_PARTS + 1]; |
595 | 0 | S32 i4_num_valid_parts; |
596 | 0 | pred_ctxt_t *ps_pred_ctxt; |
597 | |
|
598 | 0 | search_node_t as_merge_unique_node[MAX_MERGE_CANDTS]; |
599 | 0 | S32 num_unique_nodes_cu_merge = 0; |
600 | |
|
601 | 0 | search_results_t *ps_search_results = ps_merge_prms->ps_results_merge; |
602 | 0 | CU_SIZE_T e_cu_size = ps_search_results->e_cu_size; |
603 | 0 | S32 i4_part_mask = ps_search_results->i4_part_mask; |
604 | |
|
605 | 0 | search_results_t *aps_child_results[4]; |
606 | 0 | layer_ctxt_t *ps_curr_layer = ps_merge_prms->ps_layer_ctxt; |
607 | |
|
608 | 0 | S32 i4_ref_stride, i, j; |
609 | 0 | result_upd_prms_t s_result_prms; |
610 | |
|
611 | 0 | BLK_SIZE_T e_blk_size = ge_cu_size_to_blk_size[e_cu_size]; |
612 | 0 | S32 i4_offset; |
613 | | |
614 | | /*************************************************************************/ |
615 | | /* Function pointer for SAD/SATD, array and prms structure to pass to */ |
616 | | /* This function */ |
617 | | /*************************************************************************/ |
618 | 0 | PF_SAD_FXN_T pf_err_compute; |
619 | 0 | S32 ai4_sad_grid[9][17]; |
620 | 0 | err_prms_t s_err_prms; |
621 | | |
622 | | /*************************************************************************/ |
623 | | /* Allowed MV RANGE */ |
624 | | /*************************************************************************/ |
625 | 0 | range_prms_t **pps_range_prms = ps_merge_prms->aps_mv_range; |
626 | 0 | PF_INTERP_FXN_T pf_qpel_interp; |
627 | 0 | PF_MV_COST_FXN pf_mv_cost_compute; |
628 | 0 | WORD32 pred_lx; |
629 | 0 | U08 *apu1_hpel_ref[4]; |
630 | |
|
631 | 0 | interp_prms_t s_interp_prms; |
632 | 0 | S32 i4_interp_buf_id; |
633 | |
|
634 | 0 | S32 i4_ctb_x_off = ps_merge_prms->i4_ctb_x_off; |
635 | 0 | S32 i4_ctb_y_off = ps_merge_prms->i4_ctb_y_off; |
636 | | |
637 | | /* Sanity checks */ |
638 | 0 | ASSERT((e_blk_size == BLK_64x64) || (e_blk_size == BLK_32x32)); |
639 | | |
640 | 0 | s_err_prms.ps_cmn_utils_optimised_function_list = ps_cmn_utils_optimised_function_list; |
641 | | |
642 | | /* Initialize all the ptrs to child CUs for merge decision */ |
643 | 0 | aps_child_results[0] = ps_merge_prms->ps_results_tl; |
644 | 0 | aps_child_results[1] = ps_merge_prms->ps_results_tr; |
645 | 0 | aps_child_results[2] = ps_merge_prms->ps_results_bl; |
646 | 0 | aps_child_results[3] = ps_merge_prms->ps_results_br; |
647 | |
|
648 | 0 | num_unique_nodes_cu_merge = 0; |
649 | |
|
650 | 0 | pf_mv_cost_compute = compute_mv_cost_implicit_high_speed; |
651 | |
|
652 | 0 | if(ME_PRISTINE_QUALITY == ps_merge_prms->e_quality_preset) |
653 | 0 | { |
654 | 0 | num_unique_nodes_cu_merge = hme_add_cands_for_merge_eval( |
655 | 0 | ps_merge_prms->ps_cluster_info, |
656 | 0 | aps_child_results, |
657 | 0 | ps_merge_prms->ps_8x8_cu_results, |
658 | 0 | pps_range_prms, |
659 | 0 | as_merge_unique_node, |
660 | 0 | ps_search_results->pu1_is_past, |
661 | 0 | ps_merge_prms->pi1_past_list, |
662 | 0 | ps_merge_prms->pi1_future_list, |
663 | 0 | ps_merge_prms->e_quality_preset, |
664 | 0 | e_blk_size, |
665 | 0 | i4_search_idx, |
666 | 0 | (ps_merge_prms->ps_results_merge->u1_x_off >> 5) + |
667 | 0 | (ps_merge_prms->ps_results_merge->u1_y_off >> 4)); |
668 | 0 | } |
669 | 0 | else |
670 | 0 | { |
671 | | /*************************************************************************/ |
672 | | /* Populate the list of unique search nodes in the child CUs for merge */ |
673 | | /* evaluation */ |
674 | | /*************************************************************************/ |
675 | 0 | for(i = 0; i < 4; i++) |
676 | 0 | { |
677 | 0 | search_node_t s_search_node; |
678 | |
|
679 | 0 | PART_TYPE_T e_part_type; |
680 | 0 | PART_ID_T e_part_id; |
681 | |
|
682 | 0 | WORD32 part_num; |
683 | |
|
684 | 0 | search_results_t *ps_child = aps_child_results[i]; |
685 | |
|
686 | 0 | if(ps_child->ps_cu_results->u1_num_best_results) |
687 | 0 | { |
688 | 0 | if(!((ps_child->ps_cu_results->ps_best_results->as_pu_results->pu.b1_intra_flag) && |
689 | 0 | (1 == ps_child->ps_cu_results->u1_num_best_results))) |
690 | 0 | { |
691 | 0 | e_part_type = |
692 | 0 | (PART_TYPE_T)ps_child->ps_cu_results->ps_best_results[0].u1_part_type; |
693 | |
|
694 | 0 | ASSERT(num_unique_nodes_cu_merge < MAX_MERGE_CANDTS); |
695 | | |
696 | | /* Insert mvs of NxN partitions. */ |
697 | 0 | for(part_num = 0; part_num < gau1_num_parts_in_part_type[((S32)e_part_type)]; |
698 | 0 | part_num++) |
699 | 0 | { |
700 | 0 | e_part_id = ge_part_type_to_part_id[e_part_type][part_num]; |
701 | |
|
702 | 0 | if(ps_child->aps_part_results[i4_search_idx][e_part_id]->i1_ref_idx != -1) |
703 | 0 | { |
704 | 0 | s_search_node = *ps_child->aps_part_results[i4_search_idx][e_part_id]; |
705 | 0 | if(s_search_node.s_mv.i2_mvx != INTRA_MV) |
706 | 0 | { |
707 | 0 | CLIP_MV_WITHIN_RANGE( |
708 | 0 | s_search_node.s_mv.i2_mvx, |
709 | 0 | s_search_node.s_mv.i2_mvy, |
710 | 0 | pps_range_prms[s_search_node.i1_ref_idx], |
711 | 0 | 0, |
712 | 0 | 0, |
713 | 0 | 0); |
714 | |
|
715 | 0 | INSERT_NEW_NODE_NOMAP( |
716 | 0 | as_merge_unique_node, |
717 | 0 | num_unique_nodes_cu_merge, |
718 | 0 | s_search_node, |
719 | 0 | 1); |
720 | 0 | } |
721 | 0 | } |
722 | 0 | } |
723 | 0 | } |
724 | 0 | } |
725 | 0 | else if(!((ps_merge_prms->ps_results_grandchild[(i << 2)] |
726 | 0 | .ps_cu_results->ps_best_results->as_pu_results->pu.b1_intra_flag) && |
727 | 0 | (1 == ps_merge_prms->ps_results_grandchild[(i << 2)] |
728 | 0 | .ps_cu_results->u1_num_best_results))) |
729 | 0 | { |
730 | 0 | search_results_t *ps_results_root = &ps_merge_prms->ps_results_grandchild[(i << 2)]; |
731 | |
|
732 | 0 | for(j = 0; j < 4; j++) |
733 | 0 | { |
734 | 0 | e_part_type = (PART_TYPE_T)ps_results_root[j] |
735 | 0 | .ps_cu_results->ps_best_results[0] |
736 | 0 | .u1_part_type; |
737 | |
|
738 | 0 | ASSERT(num_unique_nodes_cu_merge < MAX_MERGE_CANDTS); |
739 | | |
740 | | /* Insert mvs of NxN partitions. */ |
741 | 0 | for(part_num = 0; part_num < gau1_num_parts_in_part_type[((S32)e_part_type)]; |
742 | 0 | part_num++) |
743 | 0 | { |
744 | 0 | e_part_id = ge_part_type_to_part_id[e_part_type][part_num]; |
745 | |
|
746 | 0 | if((ps_results_root[j] |
747 | 0 | .aps_part_results[i4_search_idx][e_part_id] |
748 | 0 | ->i1_ref_idx != -1) && |
749 | 0 | (!ps_child->ps_cu_results->ps_best_results->as_pu_results->pu |
750 | 0 | .b1_intra_flag)) |
751 | 0 | { |
752 | 0 | s_search_node = |
753 | 0 | *ps_results_root[j].aps_part_results[i4_search_idx][e_part_id]; |
754 | 0 | if(s_search_node.s_mv.i2_mvx != INTRA_MV) |
755 | 0 | { |
756 | 0 | CLIP_MV_WITHIN_RANGE( |
757 | 0 | s_search_node.s_mv.i2_mvx, |
758 | 0 | s_search_node.s_mv.i2_mvy, |
759 | 0 | pps_range_prms[s_search_node.i1_ref_idx], |
760 | 0 | 0, |
761 | 0 | 0, |
762 | 0 | 0); |
763 | |
|
764 | 0 | INSERT_NEW_NODE_NOMAP( |
765 | 0 | as_merge_unique_node, |
766 | 0 | num_unique_nodes_cu_merge, |
767 | 0 | s_search_node, |
768 | 0 | 1); |
769 | 0 | } |
770 | 0 | } |
771 | 0 | } |
772 | 0 | } |
773 | 0 | } |
774 | 0 | } |
775 | 0 | } |
776 | | |
777 | 0 | if(0 == num_unique_nodes_cu_merge) |
778 | 0 | { |
779 | 0 | return 0; |
780 | 0 | } |
781 | | |
782 | | /*************************************************************************/ |
783 | | /* Appropriate Err compute fxn, depends on SAD/SATD, blk size and remains*/ |
784 | | /* fixed through this subpel refinement for this partition. */ |
785 | | /* Note, we do not enable grid sads since one pt is evaluated per node */ |
786 | | /* Hence, part mask is also nearly dont care and we use 2Nx2N enabled. */ |
787 | | /*************************************************************************/ |
788 | 0 | i4_part_mask = ps_search_results->i4_part_mask; |
789 | | |
790 | | /* Need to add the corresponding SAD functions for EXTREME SPEED : Lokesh */ |
791 | 0 | if(ps_subpel_prms->i4_use_satd) |
792 | 0 | { |
793 | 0 | if(BLK_32x32 == e_blk_size) |
794 | 0 | { |
795 | 0 | pf_err_compute = hme_evalsatd_pt_pu_32x32; |
796 | 0 | } |
797 | 0 | else |
798 | 0 | { |
799 | 0 | pf_err_compute = hme_evalsatd_pt_pu_64x64; |
800 | 0 | } |
801 | 0 | } |
802 | 0 | else |
803 | 0 | { |
804 | 0 | pf_err_compute = (PF_SAD_FXN_T)hme_evalsad_grid_pu_MxM; |
805 | 0 | } |
806 | |
|
807 | 0 | i4_ref_stride = ps_curr_layer->i4_rec_stride; |
808 | |
|
809 | 0 | x_off = ps_merge_prms->ps_results_tl->u1_x_off; |
810 | 0 | y_off = ps_merge_prms->ps_results_tl->u1_y_off; |
811 | 0 | i4_offset = x_off + i4_ctb_x_off + ((y_off + i4_ctb_y_off) * i4_ref_stride); |
812 | | |
813 | | /*************************************************************************/ |
814 | | /* This array stores the ids of the partitions whose */ |
815 | | /* SADs are updated. Since the partitions whose SADs are updated may not */ |
816 | | /* be in contiguous order, we supply another level of indirection. */ |
817 | | /*************************************************************************/ |
818 | 0 | i4_num_valid_parts = hme_create_valid_part_ids(i4_part_mask, ai4_valid_part_ids); |
819 | | |
820 | | /* Initialize result params used for partition update */ |
821 | 0 | s_result_prms.pf_mv_cost_compute = NULL; |
822 | 0 | s_result_prms.ps_search_results = ps_search_results; |
823 | 0 | s_result_prms.pi4_valid_part_ids = ai4_valid_part_ids; |
824 | 0 | s_result_prms.i1_ref_idx = i4_search_idx; |
825 | 0 | s_result_prms.i4_part_mask = i4_part_mask; |
826 | 0 | s_result_prms.pi4_sad_grid = &ai4_sad_grid[0][0]; |
827 | 0 | s_result_prms.i4_grid_mask = 1; |
828 | | |
829 | | /* One time Initialization of error params used for SAD/SATD compute */ |
830 | 0 | s_err_prms.i4_inp_stride = ps_subpel_prms->i4_inp_stride; |
831 | 0 | s_err_prms.i4_ref_stride = i4_ref_stride; |
832 | 0 | s_err_prms.i4_part_mask = (ENABLE_2Nx2N); |
833 | 0 | s_err_prms.i4_grid_mask = 1; |
834 | 0 | s_err_prms.pi4_sad_grid = &ai4_sad_grid[0][0]; |
835 | 0 | s_err_prms.i4_blk_wd = gau1_blk_size_to_wd[e_blk_size]; |
836 | 0 | s_err_prms.i4_blk_ht = gau1_blk_size_to_ht[e_blk_size]; |
837 | 0 | s_err_prms.i4_step = 1; |
838 | | |
839 | | /*************************************************************************/ |
840 | | /* One time preparation of non changing interpolation params. */ |
841 | | /*************************************************************************/ |
842 | 0 | s_interp_prms.i4_ref_stride = i4_ref_stride; |
843 | 0 | s_interp_prms.i4_blk_wd = gau1_blk_size_to_wd[e_blk_size]; |
844 | 0 | s_interp_prms.i4_blk_ht = gau1_blk_size_to_ht[e_blk_size]; |
845 | 0 | s_interp_prms.apu1_interp_out[0] = ps_subpel_prms->pu1_wkg_mem; |
846 | 0 | s_interp_prms.i4_out_stride = gau1_blk_size_to_wd[e_blk_size]; |
847 | 0 | i4_interp_buf_id = 0; |
848 | |
|
849 | 0 | pf_qpel_interp = ps_subpel_prms->pf_qpel_interp; |
850 | | |
851 | | /***************************************************************************/ |
852 | | /* Compute SATD/SAD for all unique nodes of children CUs to get best merge */ |
853 | | /* results */ |
854 | | /***************************************************************************/ |
855 | 0 | for(i = 0; i < num_unique_nodes_cu_merge; i++) |
856 | 0 | { |
857 | 0 | WORD8 i1_ref_idx; |
858 | 0 | ps_search_node = &as_merge_unique_node[i]; |
859 | | |
860 | | /*********************************************************************/ |
861 | | /* Compute the base pointer for input, interpolated buffers */ |
862 | | /* The base pointers point as follows: */ |
863 | | /* fx fy : 0, 0 :: fx, hy : 0, 0.5, hx, fy: 0.5, 0, hx, fy: 0.5, 0.5 */ |
864 | | /* To these, we need to add the offset of the current node */ |
865 | | /*********************************************************************/ |
866 | 0 | i1_ref_idx = ps_search_node->i1_ref_idx; |
867 | 0 | apu1_hpel_ref[0] = ps_curr_layer->ppu1_list_rec_fxfy[i1_ref_idx] + i4_offset; |
868 | 0 | apu1_hpel_ref[1] = ps_curr_layer->ppu1_list_rec_hxfy[i1_ref_idx] + i4_offset; |
869 | 0 | apu1_hpel_ref[2] = ps_curr_layer->ppu1_list_rec_fxhy[i1_ref_idx] + i4_offset; |
870 | 0 | apu1_hpel_ref[3] = ps_curr_layer->ppu1_list_rec_hxhy[i1_ref_idx] + i4_offset; |
871 | |
|
872 | 0 | s_interp_prms.ppu1_ref = &apu1_hpel_ref[0]; |
873 | |
|
874 | 0 | pf_qpel_interp( |
875 | 0 | &s_interp_prms, |
876 | 0 | ps_search_node->s_mv.i2_mvx, |
877 | 0 | ps_search_node->s_mv.i2_mvy, |
878 | 0 | i4_interp_buf_id); |
879 | |
|
880 | 0 | pred_lx = i4_search_idx; |
881 | 0 | ps_pred_ctxt = &ps_search_results->as_pred_ctxt[pred_lx]; |
882 | |
|
883 | 0 | s_result_prms.u1_pred_lx = pred_lx; |
884 | 0 | s_result_prms.ps_search_node_base = ps_search_node; |
885 | 0 | s_err_prms.pu1_inp = |
886 | 0 | ps_wt_inp_prms->apu1_wt_inp[i1_ref_idx] + x_off + y_off * ps_subpel_prms->i4_inp_stride; |
887 | 0 | s_err_prms.pu1_ref = s_interp_prms.pu1_final_out; |
888 | 0 | s_err_prms.i4_ref_stride = s_interp_prms.i4_final_out_stride; |
889 | | |
890 | | /* Carry out the SAD/SATD. This call also does the TU RECURSION. |
891 | | Here the tu recursion logic is restricted with the size of the PU*/ |
892 | 0 | pf_err_compute(&s_err_prms); |
893 | |
|
894 | 0 | if(ps_subpel_prms->u1_is_cu_noisy && |
895 | 0 | ps_merge_prms->ps_inter_ctb_prms->i4_alpha_stim_multiplier) |
896 | 0 | { |
897 | 0 | ps_me_optimised_function_list->pf_compute_stim_injected_distortion_for_all_parts( |
898 | 0 | s_err_prms.pu1_ref, |
899 | 0 | s_err_prms.i4_ref_stride, |
900 | 0 | ai4_valid_part_ids, |
901 | 0 | ps_merge_prms->ps_inter_ctb_prms->pu8_part_src_sigmaX, |
902 | 0 | ps_merge_prms->ps_inter_ctb_prms->pu8_part_src_sigmaXSquared, |
903 | 0 | s_err_prms.pi4_sad_grid, |
904 | 0 | ps_merge_prms->ps_inter_ctb_prms->i4_alpha_stim_multiplier, |
905 | 0 | ps_wt_inp_prms->a_inv_wpred_wt[i1_ref_idx], |
906 | 0 | ps_wt_inp_prms->ai4_shift_val[i1_ref_idx], |
907 | 0 | i4_num_valid_parts, |
908 | 0 | ps_wt_inp_prms->wpred_log_wdc, |
909 | 0 | (BLK_32x32 == e_blk_size) ? 32 : 64); |
910 | 0 | } |
911 | | |
912 | | /* Update the mv's */ |
913 | 0 | s_result_prms.i2_mv_x = ps_search_node->s_mv.i2_mvx; |
914 | 0 | s_result_prms.i2_mv_y = ps_search_node->s_mv.i2_mvy; |
915 | | |
916 | | /* Update best results */ |
917 | 0 | hme_update_results_pt_pu_best1_subpel_hs(&s_err_prms, &s_result_prms); |
918 | 0 | } |
919 | | |
920 | | /************************************************************************/ |
921 | | /* Update mv cost and total cost for each valid partition in the CU */ |
922 | | /************************************************************************/ |
923 | 0 | for(i = 0; i < TOT_NUM_PARTS; i++) |
924 | 0 | { |
925 | 0 | if(i4_part_mask & (1 << i)) |
926 | 0 | { |
927 | 0 | WORD32 j; |
928 | 0 | WORD32 i4_mv_cost; |
929 | |
|
930 | 0 | ps_search_node = ps_search_results->aps_part_results[i4_search_idx][i]; |
931 | |
|
932 | 0 | for(j = 0; |
933 | 0 | j < MIN(ps_search_results->u1_num_results_per_part, num_unique_nodes_cu_merge); |
934 | 0 | j++) |
935 | 0 | { |
936 | 0 | if(ps_search_node->i1_ref_idx != -1) |
937 | 0 | { |
938 | 0 | pred_lx = i4_search_idx; |
939 | 0 | ps_pred_ctxt = &ps_search_results->as_pred_ctxt[pred_lx]; |
940 | | |
941 | | /* Prediction context should now deal with qpel units */ |
942 | 0 | HME_SET_MVPRED_RES(ps_pred_ctxt, MV_RES_QPEL); |
943 | |
|
944 | 0 | ps_search_node->u1_subpel_done = 1; |
945 | 0 | ps_search_node->u1_is_avail = 1; |
946 | |
|
947 | 0 | i4_mv_cost = |
948 | 0 | pf_mv_cost_compute(ps_search_node, ps_pred_ctxt, (PART_ID_T)i, MV_RES_QPEL); |
949 | |
|
950 | 0 | ps_search_node->i4_tot_cost = i4_mv_cost + ps_search_node->i4_sad; |
951 | 0 | ps_search_node->i4_mv_cost = i4_mv_cost; |
952 | |
|
953 | 0 | ps_search_node++; |
954 | 0 | } |
955 | 0 | } |
956 | 0 | } |
957 | 0 | } |
958 | |
|
959 | 0 | return num_unique_nodes_cu_merge; |
960 | 0 | } |
961 | | |
962 | 0 | #define CU_MERGE_MAX_INTRA_PARTS 4 |
963 | | |
964 | | /** |
965 | | ******************************************************************************** |
966 | | * @fn hme_try_merge_high_speed |
967 | | * |
968 | | * @brief Attempts to merge 4 NxN candts to a 2Nx2N candt, either as a single |
969 | | entity or with partititons for high speed preset |
970 | | * |
971 | | * @param[in,out] hme_merge_prms_t: Params for CU merge |
972 | | * |
973 | | * @return MERGE_RESULT_T type result of merge (CU_MERGED/CU_SPLIT) |
974 | | ******************************************************************************** |
975 | | */ |
976 | | CU_MERGE_RESULT_T hme_try_merge_high_speed( |
977 | | me_ctxt_t *ps_thrd_ctxt, |
978 | | me_frm_ctxt_t *ps_ctxt, |
979 | | ipe_l0_ctb_analyse_for_me_t *ps_cur_ipe_ctb, |
980 | | hme_subpel_prms_t *ps_subpel_prms, |
981 | | hme_merge_prms_t *ps_merge_prms, |
982 | | inter_pu_results_t *ps_pu_results, |
983 | | pu_result_t *ps_pu_result) |
984 | 0 | { |
985 | 0 | search_results_t *ps_results_tl, *ps_results_tr; |
986 | 0 | search_results_t *ps_results_bl, *ps_results_br; |
987 | |
|
988 | 0 | S32 i; |
989 | 0 | S32 i4_search_idx; |
990 | 0 | S32 i4_cost_parent; |
991 | 0 | S32 intra_cu_size; |
992 | 0 | ULWORD64 au8_final_src_sigmaX[17], au8_final_src_sigmaXSquared[17]; |
993 | |
|
994 | 0 | search_results_t *ps_results_merge = ps_merge_prms->ps_results_merge; |
995 | 0 | wgt_pred_ctxt_t *ps_wt_inp_prms = &ps_ctxt->s_wt_pred; |
996 | |
|
997 | 0 | S32 i4_part_mask = ENABLE_ALL_PARTS - ENABLE_NxN; |
998 | 0 | S32 is_vert = 0, i4_best_part_type = -1; |
999 | 0 | S32 i4_intra_parts = 0; /* Keeps track of intra percentage before merge */ |
1000 | 0 | S32 i4_cost_children = 0; |
1001 | 0 | S32 i4_frm_qstep = ps_ctxt->frm_qstep; |
1002 | 0 | S32 i4_num_merge_cands_evaluated = 0; |
1003 | 0 | U08 u1_x_off = ps_results_merge->u1_x_off; |
1004 | 0 | U08 u1_y_off = ps_results_merge->u1_y_off; |
1005 | 0 | S32 i4_32x32_id = (u1_y_off >> 4) + (u1_x_off >> 5); |
1006 | |
|
1007 | 0 | ihevce_cmn_opt_func_t *ps_cmn_utils_optimised_function_list = |
1008 | 0 | ps_thrd_ctxt->ps_cmn_utils_optimised_function_list; |
1009 | 0 | ihevce_me_optimised_function_list_t *ps_me_optimised_function_list = |
1010 | 0 | ((ihevce_me_optimised_function_list_t *)ps_thrd_ctxt->pv_me_optimised_function_list); |
1011 | 0 | ps_results_tl = ps_merge_prms->ps_results_tl; |
1012 | 0 | ps_results_tr = ps_merge_prms->ps_results_tr; |
1013 | 0 | ps_results_bl = ps_merge_prms->ps_results_bl; |
1014 | 0 | ps_results_br = ps_merge_prms->ps_results_br; |
1015 | |
|
1016 | 0 | if(ps_merge_prms->e_quality_preset == ME_XTREME_SPEED) |
1017 | 0 | { |
1018 | 0 | i4_part_mask &= ~ENABLE_AMP; |
1019 | 0 | } |
1020 | |
|
1021 | 0 | if(ps_merge_prms->e_quality_preset == ME_XTREME_SPEED_25) |
1022 | 0 | { |
1023 | 0 | i4_part_mask &= ~ENABLE_AMP; |
1024 | |
|
1025 | 0 | i4_part_mask &= ~ENABLE_SMP; |
1026 | 0 | } |
1027 | |
|
1028 | 0 | ps_merge_prms->i4_num_pred_dir_actual = 0; |
1029 | | |
1030 | | /*************************************************************************/ |
1031 | | /* The logic for High speed CU merge goes as follows: */ |
1032 | | /* */ |
1033 | | /* 1. Early exit with CU_SPLIT if sum of best partitions of children CUs */ |
1034 | | /* exceed 7 */ |
1035 | | /* 2. Early exit with CU_MERGE if mvs of best partitions of children CUs */ |
1036 | | /* are identical */ |
1037 | | /* 3. Find the all unique mvs of best partitions of children CUs and */ |
1038 | | /* evaluate partial SATDs (all 17 partitions) for each unique mv. If */ |
1039 | | /* best parent cost is lower than sum of the best children costs */ |
1040 | | /* return CU_MERGE after seeding the best results else return CU_SPLIT*/ |
1041 | | /* */ |
1042 | | /*************************************************************************/ |
1043 | | |
1044 | | /* Count the number of best partitions in child CUs, early exit if > 7 */ |
1045 | 0 | if((ps_merge_prms->e_quality_preset != ME_PRISTINE_QUALITY) || |
1046 | 0 | (CU_32x32 == ps_results_merge->e_cu_size)) |
1047 | 0 | { |
1048 | 0 | S32 num_parts_in_32x32 = 0; |
1049 | 0 | WORD32 i4_part_type; |
1050 | |
|
1051 | 0 | if(ps_results_tl->u1_split_flag) |
1052 | 0 | { |
1053 | 0 | num_parts_in_32x32 += 4; |
1054 | |
|
1055 | 0 | #define COST_INTERCHANGE 0 |
1056 | 0 | i4_cost_children = ps_merge_prms->ps_8x8_cu_results[0].ps_best_results->i4_tot_cost + |
1057 | 0 | ps_merge_prms->ps_8x8_cu_results[1].ps_best_results->i4_tot_cost + |
1058 | 0 | ps_merge_prms->ps_8x8_cu_results[2].ps_best_results->i4_tot_cost + |
1059 | 0 | ps_merge_prms->ps_8x8_cu_results[3].ps_best_results->i4_tot_cost; |
1060 | 0 | } |
1061 | 0 | else |
1062 | 0 | { |
1063 | 0 | i4_part_type = ps_results_tl->ps_cu_results->ps_best_results[0].u1_part_type; |
1064 | 0 | num_parts_in_32x32 += gau1_num_parts_in_part_type[i4_part_type]; |
1065 | 0 | i4_cost_children = ps_results_tl->ps_cu_results->ps_best_results[0].i4_tot_cost; |
1066 | 0 | } |
1067 | |
|
1068 | 0 | if(ps_results_tr->u1_split_flag) |
1069 | 0 | { |
1070 | 0 | num_parts_in_32x32 += 4; |
1071 | |
|
1072 | 0 | i4_cost_children += ps_merge_prms->ps_8x8_cu_results[4].ps_best_results->i4_tot_cost + |
1073 | 0 | ps_merge_prms->ps_8x8_cu_results[5].ps_best_results->i4_tot_cost + |
1074 | 0 | ps_merge_prms->ps_8x8_cu_results[6].ps_best_results->i4_tot_cost + |
1075 | 0 | ps_merge_prms->ps_8x8_cu_results[7].ps_best_results->i4_tot_cost; |
1076 | 0 | } |
1077 | 0 | else |
1078 | 0 | { |
1079 | 0 | i4_part_type = ps_results_tr->ps_cu_results->ps_best_results[0].u1_part_type; |
1080 | 0 | num_parts_in_32x32 += gau1_num_parts_in_part_type[i4_part_type]; |
1081 | 0 | i4_cost_children += ps_results_tr->ps_cu_results->ps_best_results[0].i4_tot_cost; |
1082 | 0 | } |
1083 | |
|
1084 | 0 | if(ps_results_bl->u1_split_flag) |
1085 | 0 | { |
1086 | 0 | num_parts_in_32x32 += 4; |
1087 | |
|
1088 | 0 | i4_cost_children += ps_merge_prms->ps_8x8_cu_results[8].ps_best_results->i4_tot_cost + |
1089 | 0 | ps_merge_prms->ps_8x8_cu_results[9].ps_best_results->i4_tot_cost + |
1090 | 0 | ps_merge_prms->ps_8x8_cu_results[10].ps_best_results->i4_tot_cost + |
1091 | 0 | ps_merge_prms->ps_8x8_cu_results[11].ps_best_results->i4_tot_cost; |
1092 | 0 | } |
1093 | 0 | else |
1094 | 0 | { |
1095 | 0 | i4_part_type = ps_results_bl->ps_cu_results->ps_best_results[0].u1_part_type; |
1096 | 0 | num_parts_in_32x32 += gau1_num_parts_in_part_type[i4_part_type]; |
1097 | 0 | i4_cost_children += ps_results_bl->ps_cu_results->ps_best_results[0].i4_tot_cost; |
1098 | 0 | } |
1099 | |
|
1100 | 0 | if(ps_results_br->u1_split_flag) |
1101 | 0 | { |
1102 | 0 | num_parts_in_32x32 += 4; |
1103 | |
|
1104 | 0 | i4_cost_children += ps_merge_prms->ps_8x8_cu_results[12].ps_best_results->i4_tot_cost + |
1105 | 0 | ps_merge_prms->ps_8x8_cu_results[13].ps_best_results->i4_tot_cost + |
1106 | 0 | ps_merge_prms->ps_8x8_cu_results[14].ps_best_results->i4_tot_cost + |
1107 | 0 | ps_merge_prms->ps_8x8_cu_results[15].ps_best_results->i4_tot_cost; |
1108 | 0 | } |
1109 | 0 | else |
1110 | 0 | { |
1111 | 0 | i4_part_type = ps_results_br->ps_cu_results->ps_best_results[0].u1_part_type; |
1112 | 0 | num_parts_in_32x32 += gau1_num_parts_in_part_type[i4_part_type]; |
1113 | 0 | i4_cost_children += ps_results_br->ps_cu_results->ps_best_results[0].i4_tot_cost; |
1114 | 0 | } |
1115 | |
|
1116 | 0 | if((num_parts_in_32x32 > 7) && (ps_merge_prms->e_quality_preset != ME_PRISTINE_QUALITY)) |
1117 | 0 | { |
1118 | 0 | return CU_SPLIT; |
1119 | 0 | } |
1120 | | |
1121 | 0 | if((num_parts_in_32x32 > MAX_NUM_CONSTITUENT_MVS_TO_ENABLE_32MERGE_IN_XS25) && |
1122 | 0 | (ps_merge_prms->e_quality_preset == ME_XTREME_SPEED_25)) |
1123 | 0 | { |
1124 | 0 | return CU_SPLIT; |
1125 | 0 | } |
1126 | 0 | } |
1127 | | |
1128 | | /* Accumulate intra percentage before merge for early CU_SPLIT decision */ |
1129 | | /* Note : Each intra part represent a NxN unit of the children CUs */ |
1130 | | /* This is essentially 1/16th of the CUsize under consideration for merge */ |
1131 | 0 | if(ME_PRISTINE_QUALITY == ps_merge_prms->e_quality_preset) |
1132 | 0 | { |
1133 | 0 | if(CU_64x64 == ps_results_merge->e_cu_size) |
1134 | 0 | { |
1135 | 0 | i4_intra_parts = |
1136 | 0 | (!ps_merge_prms->ps_cluster_info->ps_cu_tree_root->u1_inter_eval_enable) |
1137 | 0 | ? 16 |
1138 | 0 | : ps_merge_prms->ps_cluster_info->ps_cu_tree_root->u1_intra_eval_enable; |
1139 | 0 | } |
1140 | 0 | else |
1141 | 0 | { |
1142 | 0 | switch((ps_results_merge->u1_x_off >> 5) + ((ps_results_merge->u1_y_off >> 4))) |
1143 | 0 | { |
1144 | 0 | case 0: |
1145 | 0 | { |
1146 | 0 | i4_intra_parts = (!ps_merge_prms->ps_cluster_info->ps_cu_tree_root->ps_child_node_tl |
1147 | 0 | ->u1_inter_eval_enable) |
1148 | 0 | ? 16 |
1149 | 0 | : (ps_merge_prms->ps_cluster_info->ps_cu_tree_root |
1150 | 0 | ->ps_child_node_tl->u1_intra_eval_enable); |
1151 | |
|
1152 | 0 | break; |
1153 | 0 | } |
1154 | 0 | case 1: |
1155 | 0 | { |
1156 | 0 | i4_intra_parts = (!ps_merge_prms->ps_cluster_info->ps_cu_tree_root->ps_child_node_tr |
1157 | 0 | ->u1_inter_eval_enable) |
1158 | 0 | ? 16 |
1159 | 0 | : (ps_merge_prms->ps_cluster_info->ps_cu_tree_root |
1160 | 0 | ->ps_child_node_tr->u1_intra_eval_enable); |
1161 | |
|
1162 | 0 | break; |
1163 | 0 | } |
1164 | 0 | case 2: |
1165 | 0 | { |
1166 | 0 | i4_intra_parts = (!ps_merge_prms->ps_cluster_info->ps_cu_tree_root->ps_child_node_bl |
1167 | 0 | ->u1_inter_eval_enable) |
1168 | 0 | ? 16 |
1169 | 0 | : (ps_merge_prms->ps_cluster_info->ps_cu_tree_root |
1170 | 0 | ->ps_child_node_bl->u1_intra_eval_enable); |
1171 | |
|
1172 | 0 | break; |
1173 | 0 | } |
1174 | 0 | case 3: |
1175 | 0 | { |
1176 | 0 | i4_intra_parts = (!ps_merge_prms->ps_cluster_info->ps_cu_tree_root->ps_child_node_br |
1177 | 0 | ->u1_inter_eval_enable) |
1178 | 0 | ? 16 |
1179 | 0 | : (ps_merge_prms->ps_cluster_info->ps_cu_tree_root |
1180 | 0 | ->ps_child_node_br->u1_intra_eval_enable); |
1181 | |
|
1182 | 0 | break; |
1183 | 0 | } |
1184 | 0 | } |
1185 | 0 | } |
1186 | 0 | } |
1187 | 0 | else |
1188 | 0 | { |
1189 | 0 | for(i = 0; i < 4; i++) |
1190 | 0 | { |
1191 | 0 | search_results_t *ps_results = |
1192 | 0 | (i == 0) ? ps_results_tl |
1193 | 0 | : ((i == 1) ? ps_results_tr : ((i == 2) ? ps_results_bl : ps_results_br)); |
1194 | |
|
1195 | 0 | part_type_results_t *ps_best_res = &ps_results->ps_cu_results->ps_best_results[0]; |
1196 | |
|
1197 | 0 | if(ps_results->u1_split_flag) |
1198 | 0 | { |
1199 | 0 | U08 u1_x_off = ps_results->u1_x_off; |
1200 | 0 | U08 u1_y_off = ps_results->u1_y_off; |
1201 | 0 | U08 u1_8x8_zscan_id = gau1_ctb_raster_to_zscan[(u1_x_off >> 2) + (u1_y_off << 2)] >> |
1202 | 0 | 2; |
1203 | | |
1204 | | /* Special case to handle 8x8 CUs when 16x16 is split */ |
1205 | 0 | ASSERT(ps_results->e_cu_size == CU_16x16); |
1206 | | |
1207 | 0 | ps_best_res = &ps_ctxt->as_cu8x8_results[u1_8x8_zscan_id].ps_best_results[0]; |
1208 | |
|
1209 | 0 | if(ps_best_res->as_pu_results[0].pu.b1_intra_flag) |
1210 | 0 | i4_intra_parts += 1; |
1211 | |
|
1212 | 0 | ps_best_res = &ps_ctxt->as_cu8x8_results[u1_8x8_zscan_id + 1].ps_best_results[0]; |
1213 | |
|
1214 | 0 | if(ps_best_res->as_pu_results[0].pu.b1_intra_flag) |
1215 | 0 | i4_intra_parts += 1; |
1216 | |
|
1217 | 0 | ps_best_res = &ps_ctxt->as_cu8x8_results[u1_8x8_zscan_id + 2].ps_best_results[0]; |
1218 | |
|
1219 | 0 | if(ps_best_res->as_pu_results[0].pu.b1_intra_flag) |
1220 | 0 | i4_intra_parts += 1; |
1221 | |
|
1222 | 0 | ps_best_res = &ps_ctxt->as_cu8x8_results[u1_8x8_zscan_id + 3].ps_best_results[0]; |
1223 | |
|
1224 | 0 | if(ps_best_res->as_pu_results[0].pu.b1_intra_flag) |
1225 | 0 | i4_intra_parts += 1; |
1226 | 0 | } |
1227 | 0 | else if(ps_best_res[0].as_pu_results[0].pu.b1_intra_flag) |
1228 | 0 | { |
1229 | 0 | i4_intra_parts += 4; |
1230 | 0 | } |
1231 | 0 | } |
1232 | 0 | } |
1233 | | |
1234 | | /* Determine the max intra CU size indicated by IPE */ |
1235 | 0 | intra_cu_size = CU_64x64; |
1236 | 0 | if(ps_cur_ipe_ctb->u1_split_flag) |
1237 | 0 | { |
1238 | 0 | intra_cu_size = CU_32x32; |
1239 | 0 | if(ps_cur_ipe_ctb->as_intra32_analyse[i4_32x32_id].b1_split_flag) |
1240 | 0 | { |
1241 | 0 | intra_cu_size = CU_16x16; |
1242 | 0 | } |
1243 | 0 | } |
1244 | |
|
1245 | 0 | if(((i4_intra_parts > CU_MERGE_MAX_INTRA_PARTS) && |
1246 | 0 | (intra_cu_size < ps_results_merge->e_cu_size) && |
1247 | 0 | (ME_PRISTINE_QUALITY != ps_merge_prms->e_quality_preset)) || |
1248 | 0 | (i4_intra_parts == 16)) |
1249 | 0 | { |
1250 | 0 | S32 i4_merge_outcome; |
1251 | |
|
1252 | 0 | i4_merge_outcome = (CU_32x32 == ps_results_merge->e_cu_size) |
1253 | 0 | ? (!ps_cur_ipe_ctb->as_intra32_analyse[i4_32x32_id].b1_split_flag && |
1254 | 0 | ps_cur_ipe_ctb->as_intra32_analyse[i4_32x32_id].b1_valid_cu) |
1255 | 0 | : (!ps_cur_ipe_ctb->u1_split_flag); |
1256 | |
|
1257 | 0 | i4_merge_outcome = i4_merge_outcome || |
1258 | 0 | (ME_PRISTINE_QUALITY == ps_merge_prms->e_quality_preset); |
1259 | |
|
1260 | 0 | i4_merge_outcome = i4_merge_outcome && |
1261 | 0 | !(ps_subpel_prms->u1_is_cu_noisy && DISABLE_INTRA_WHEN_NOISY); |
1262 | |
|
1263 | 0 | if(i4_merge_outcome) |
1264 | 0 | { |
1265 | 0 | inter_cu_results_t *ps_cu_results = ps_results_merge->ps_cu_results; |
1266 | 0 | part_type_results_t *ps_best_result = ps_cu_results->ps_best_results; |
1267 | 0 | pu_t *ps_pu = &ps_best_result->as_pu_results->pu; |
1268 | |
|
1269 | 0 | ps_cu_results->u1_num_best_results = 1; |
1270 | 0 | ps_cu_results->u1_cu_size = ps_results_merge->e_cu_size; |
1271 | 0 | ps_cu_results->u1_x_off = u1_x_off; |
1272 | 0 | ps_cu_results->u1_y_off = u1_y_off; |
1273 | |
|
1274 | 0 | ps_best_result->u1_part_type = PRT_2Nx2N; |
1275 | 0 | ps_best_result->ai4_tu_split_flag[0] = 0; |
1276 | 0 | ps_best_result->ai4_tu_split_flag[1] = 0; |
1277 | 0 | ps_best_result->ai4_tu_split_flag[2] = 0; |
1278 | 0 | ps_best_result->ai4_tu_split_flag[3] = 0; |
1279 | 0 | ps_best_result->i4_tot_cost = |
1280 | 0 | (CU_64x64 == ps_results_merge->e_cu_size) |
1281 | 0 | ? ps_cur_ipe_ctb->i4_best64x64_intra_cost |
1282 | 0 | : ps_cur_ipe_ctb->ai4_best32x32_intra_cost[i4_32x32_id]; |
1283 | |
|
1284 | 0 | ps_pu->b1_intra_flag = 1; |
1285 | 0 | ps_pu->b4_pos_x = u1_x_off >> 2; |
1286 | 0 | ps_pu->b4_pos_y = u1_y_off >> 2; |
1287 | 0 | ps_pu->b4_wd = (1 << (ps_results_merge->e_cu_size + 1)) - 1; |
1288 | 0 | ps_pu->b4_ht = ps_pu->b4_wd; |
1289 | 0 | ps_pu->mv.i1_l0_ref_idx = -1; |
1290 | 0 | ps_pu->mv.i1_l1_ref_idx = -1; |
1291 | 0 | ps_pu->mv.s_l0_mv.i2_mvx = INTRA_MV; |
1292 | 0 | ps_pu->mv.s_l0_mv.i2_mvy = INTRA_MV; |
1293 | 0 | ps_pu->mv.s_l1_mv.i2_mvx = INTRA_MV; |
1294 | 0 | ps_pu->mv.s_l1_mv.i2_mvy = INTRA_MV; |
1295 | |
|
1296 | 0 | return CU_MERGED; |
1297 | 0 | } |
1298 | 0 | else |
1299 | 0 | { |
1300 | 0 | return CU_SPLIT; |
1301 | 0 | } |
1302 | 0 | } |
1303 | | |
1304 | 0 | if(i4_intra_parts) |
1305 | 0 | { |
1306 | 0 | i4_part_mask = ENABLE_2Nx2N; |
1307 | 0 | } |
1308 | |
|
1309 | 0 | ps_results_merge->u1_num_active_ref = (ps_ctxt->s_frm_prms.bidir_enabled) ? 2 : 1; |
1310 | |
|
1311 | 0 | hme_reset_search_results(ps_results_merge, i4_part_mask, MV_RES_QPEL); |
1312 | |
|
1313 | 0 | ps_results_merge->u1_num_active_ref = ps_merge_prms->i4_num_ref; |
1314 | 0 | ps_merge_prms->i4_num_pred_dir_actual = 0; |
1315 | |
|
1316 | 0 | if(ps_subpel_prms->u1_is_cu_noisy && ps_merge_prms->ps_inter_ctb_prms->i4_alpha_stim_multiplier) |
1317 | 0 | { |
1318 | 0 | S32 ai4_valid_part_ids[TOT_NUM_PARTS + 1]; |
1319 | 0 | S32 i4_num_valid_parts; |
1320 | 0 | S32 i4_sigma_array_offset; |
1321 | |
|
1322 | 0 | i4_num_valid_parts = hme_create_valid_part_ids(i4_part_mask, ai4_valid_part_ids); |
1323 | | |
1324 | | /*********************************************************************************************************************************************/ |
1325 | | /* i4_sigma_array_offset : takes care of pointing to the appropriate 4x4 block's sigmaX and sigmaX-squared value in a CTB out of 256 values */ |
1326 | | /* Logic is x/4 + ((y/4) x 16) : every 4 pixel increase in x equals one 4x4 block increment, every 4 pixel increase in y equals 16 4x4 block */ |
1327 | | /* increment as there will be 256 4x4 blocks in a CTB */ |
1328 | | /*********************************************************************************************************************************************/ |
1329 | 0 | i4_sigma_array_offset = (ps_merge_prms->ps_results_merge->u1_x_off / 4) + |
1330 | 0 | (ps_merge_prms->ps_results_merge->u1_y_off * 4); |
1331 | |
|
1332 | 0 | for(i = 0; i < i4_num_valid_parts; i++) |
1333 | 0 | { |
1334 | 0 | S32 i4_part_id = ai4_valid_part_ids[i]; |
1335 | |
|
1336 | 0 | hme_compute_final_sigma_of_pu_from_base_blocks( |
1337 | 0 | ps_ctxt->au4_4x4_src_sigmaX + i4_sigma_array_offset, |
1338 | 0 | ps_ctxt->au4_4x4_src_sigmaXSquared + i4_sigma_array_offset, |
1339 | 0 | au8_final_src_sigmaX, |
1340 | 0 | au8_final_src_sigmaXSquared, |
1341 | 0 | (CU_32x32 == ps_results_merge->e_cu_size) ? 32 : 64, |
1342 | 0 | 4, |
1343 | 0 | i4_part_id, |
1344 | 0 | 16); |
1345 | 0 | } |
1346 | |
|
1347 | 0 | ps_merge_prms->ps_inter_ctb_prms->pu8_part_src_sigmaX = au8_final_src_sigmaX; |
1348 | 0 | ps_merge_prms->ps_inter_ctb_prms->pu8_part_src_sigmaXSquared = au8_final_src_sigmaXSquared; |
1349 | 0 | } |
1350 | | |
1351 | | /*************************************************************************/ |
1352 | | /* Loop through all ref idx and pick the merge candts and refine based */ |
1353 | | /* on the active partitions. At this stage num ref will be 1 or 2 */ |
1354 | | /*************************************************************************/ |
1355 | 0 | for(i4_search_idx = 0; i4_search_idx < ps_merge_prms->i4_num_ref; i4_search_idx++) |
1356 | 0 | { |
1357 | 0 | S32 i4_cands; |
1358 | 0 | U08 u1_pred_dir = 0; |
1359 | |
|
1360 | 0 | if((2 == ps_merge_prms->i4_num_ref) || (!ps_ctxt->s_frm_prms.bidir_enabled)) |
1361 | 0 | { |
1362 | 0 | u1_pred_dir = i4_search_idx; |
1363 | 0 | } |
1364 | 0 | else if(ps_ctxt->s_frm_prms.u1_num_active_ref_l0 == 0) |
1365 | 0 | { |
1366 | 0 | u1_pred_dir = 1; |
1367 | 0 | } |
1368 | 0 | else if(ps_ctxt->s_frm_prms.u1_num_active_ref_l1 == 0) |
1369 | 0 | { |
1370 | 0 | u1_pred_dir = 0; |
1371 | 0 | } |
1372 | 0 | else |
1373 | 0 | { |
1374 | 0 | ASSERT(0); |
1375 | 0 | } |
1376 | | |
1377 | | /* call the function to pick and evaluate the merge candts, given */ |
1378 | | /* a ref id and a part mask. */ |
1379 | 0 | i4_cands = hme_pick_eval_merge_candts( |
1380 | 0 | ps_merge_prms, |
1381 | 0 | ps_subpel_prms, |
1382 | 0 | u1_pred_dir, |
1383 | 0 | i4_best_part_type, |
1384 | 0 | is_vert, |
1385 | 0 | ps_wt_inp_prms, |
1386 | 0 | i4_frm_qstep, |
1387 | 0 | ps_cmn_utils_optimised_function_list, |
1388 | 0 | ps_me_optimised_function_list); |
1389 | |
|
1390 | 0 | if(i4_cands) |
1391 | 0 | { |
1392 | 0 | ps_merge_prms->au1_pred_dir_searched[ps_merge_prms->i4_num_pred_dir_actual] = |
1393 | 0 | u1_pred_dir; |
1394 | 0 | ps_merge_prms->i4_num_pred_dir_actual++; |
1395 | 0 | } |
1396 | |
|
1397 | 0 | i4_num_merge_cands_evaluated += i4_cands; |
1398 | 0 | } |
1399 | | |
1400 | | /* Call the decide_part_types function here */ |
1401 | | /* Populate the new PU struct with the results post subpel refinement*/ |
1402 | 0 | if(i4_num_merge_cands_evaluated) |
1403 | 0 | { |
1404 | 0 | inter_cu_results_t *ps_cu_results = ps_results_merge->ps_cu_results; |
1405 | |
|
1406 | 0 | hme_reset_wkg_mem(&ps_ctxt->s_buf_mgr); |
1407 | |
|
1408 | 0 | ps_merge_prms->ps_inter_ctb_prms->i4_ctb_x_off = ps_merge_prms->i4_ctb_x_off; |
1409 | 0 | ps_merge_prms->ps_inter_ctb_prms->i4_ctb_y_off = ps_merge_prms->i4_ctb_y_off; |
1410 | |
|
1411 | 0 | hme_populate_pus( |
1412 | 0 | ps_thrd_ctxt, |
1413 | 0 | ps_ctxt, |
1414 | 0 | ps_subpel_prms, |
1415 | 0 | ps_results_merge, |
1416 | 0 | ps_cu_results, |
1417 | 0 | ps_pu_results, |
1418 | 0 | ps_pu_result, |
1419 | 0 | ps_merge_prms->ps_inter_ctb_prms, |
1420 | 0 | &ps_ctxt->s_wt_pred, |
1421 | 0 | ps_merge_prms->ps_layer_ctxt, |
1422 | 0 | ps_merge_prms->au1_pred_dir_searched, |
1423 | 0 | ps_merge_prms->i4_num_pred_dir_actual); |
1424 | |
|
1425 | 0 | ps_cu_results->i4_inp_offset = (ps_cu_results->u1_x_off) + (ps_cu_results->u1_y_off * 64); |
1426 | |
|
1427 | 0 | hme_decide_part_types( |
1428 | 0 | ps_cu_results, |
1429 | 0 | ps_pu_results, |
1430 | 0 | ps_merge_prms->ps_inter_ctb_prms, |
1431 | 0 | ps_ctxt, |
1432 | 0 | ps_cmn_utils_optimised_function_list, |
1433 | 0 | ps_me_optimised_function_list |
1434 | |
|
1435 | 0 | ); |
1436 | | |
1437 | | /*****************************************************************/ |
1438 | | /* INSERT INTRA RESULTS AT 32x32/64x64 LEVEL. */ |
1439 | | /*****************************************************************/ |
1440 | 0 | #if DISABLE_INTRA_IN_BPICS |
1441 | 0 | if(1 != ((ME_XTREME_SPEED_25 == ps_merge_prms->e_quality_preset) && |
1442 | 0 | (ps_ctxt->s_frm_prms.i4_temporal_layer_id > TEMPORAL_LAYER_DISABLE))) |
1443 | 0 | #endif |
1444 | 0 | { |
1445 | 0 | if(!(DISABLE_INTRA_WHEN_NOISY && ps_merge_prms->ps_inter_ctb_prms->u1_is_cu_noisy)) |
1446 | 0 | { |
1447 | 0 | hme_insert_intra_nodes_post_bipred( |
1448 | 0 | ps_cu_results, ps_cur_ipe_ctb, ps_ctxt->frm_qstep); |
1449 | 0 | } |
1450 | 0 | } |
1451 | 0 | } |
1452 | 0 | else |
1453 | 0 | { |
1454 | 0 | return CU_SPLIT; |
1455 | 0 | } |
1456 | | |
1457 | | /* We check the best result of ref idx 0 and compare for parent vs child */ |
1458 | 0 | if((ps_merge_prms->e_quality_preset != ME_PRISTINE_QUALITY) || |
1459 | 0 | (CU_32x32 == ps_results_merge->e_cu_size)) |
1460 | 0 | { |
1461 | 0 | i4_cost_parent = ps_results_merge->ps_cu_results->ps_best_results[0].i4_tot_cost; |
1462 | | /*********************************************************************/ |
1463 | | /* Add the cost of signaling the CU tree bits. */ |
1464 | | /* Assuming parent is not split, then we signal 1 bit for this parent*/ |
1465 | | /* CU. If split, then 1 bit for parent CU + 4 bits for each child CU */ |
1466 | | /* So, 4*lambda is extra for children cost. :Lokesh */ |
1467 | | /*********************************************************************/ |
1468 | 0 | { |
1469 | 0 | pred_ctxt_t *ps_pred_ctxt = &ps_results_merge->as_pred_ctxt[0]; |
1470 | |
|
1471 | 0 | i4_cost_children += ((4 * ps_pred_ctxt->lambda) >> (ps_pred_ctxt->lambda_q_shift)); |
1472 | 0 | } |
1473 | |
|
1474 | 0 | if(i4_cost_parent < i4_cost_children) |
1475 | 0 | { |
1476 | 0 | return CU_MERGED; |
1477 | 0 | } |
1478 | | |
1479 | 0 | return CU_SPLIT; |
1480 | 0 | } |
1481 | 0 | else |
1482 | 0 | { |
1483 | 0 | return CU_MERGED; |
1484 | 0 | } |
1485 | 0 | } |
1486 | | |
1487 | | #define COPY_SEARCH_RESULT(ps_mv, pi1_ref_idx, ps_search_node, shift) \ |
1488 | 0 | { \ |
1489 | 0 | (ps_mv)->i2_mv_x = (ps_search_node)->s_mv.i2_mvx >> (shift); \ |
1490 | 0 | (ps_mv)->i2_mv_y = (ps_search_node)->s_mv.i2_mvy >> (shift); \ |
1491 | 0 | *(pi1_ref_idx) = (ps_search_node)->i1_ref_idx; \ |
1492 | 0 | } |
1493 | | |
1494 | | /** |
1495 | | ******************************************************************************** |
1496 | | * @fn hme_update_mv_bank_noencode(search_results_t *ps_search_results, |
1497 | | * layer_mv_t *ps_layer_mv, |
1498 | | * S32 i4_search_blk_x, |
1499 | | * S32 i4_search_blk_y, |
1500 | | * mvbank_update_prms_t *ps_prms) |
1501 | | * |
1502 | | * @brief Updates the mv bank in case there is no further encodign to be done |
1503 | | * |
1504 | | * @param[in] ps_search_results: contains results for the block just searched |
1505 | | * |
1506 | | * @param[in,out] ps_layer_mv : Has pointer to mv bank amongst other things |
1507 | | * |
1508 | | * @param[in] i4_search_blk_x : col num of blk being searched |
1509 | | * |
1510 | | * @param[in] i4_search_blk_y : row num of blk being searched |
1511 | | * |
1512 | | * @param[in] ps_prms : contains certain parameters which govern how updatedone |
1513 | | * |
1514 | | * @return None |
1515 | | ******************************************************************************** |
1516 | | */ |
1517 | | |
1518 | | void hme_update_mv_bank_noencode( |
1519 | | search_results_t *ps_search_results, |
1520 | | layer_mv_t *ps_layer_mv, |
1521 | | S32 i4_search_blk_x, |
1522 | | S32 i4_search_blk_y, |
1523 | | mvbank_update_prms_t *ps_prms) |
1524 | 0 | { |
1525 | 0 | hme_mv_t *ps_mv; |
1526 | 0 | hme_mv_t *ps_mv1, *ps_mv2, *ps_mv3, *ps_mv4; |
1527 | 0 | S08 *pi1_ref_idx, *pi1_ref_idx1, *pi1_ref_idx2, *pi1_ref_idx3, *pi1_ref_idx4; |
1528 | 0 | S32 i4_blk_x, i4_blk_y, i4_offset; |
1529 | 0 | S32 i4_j, i4_ref_id; |
1530 | 0 | search_node_t *ps_search_node; |
1531 | 0 | search_node_t *ps_search_node_8x8, *ps_search_node_4x4_1; |
1532 | 0 | search_node_t *ps_search_node_4x4_2, *ps_search_node_4x4_3; |
1533 | 0 | search_node_t *ps_search_node_4x4_4; |
1534 | |
|
1535 | 0 | i4_blk_x = i4_search_blk_x << ps_prms->i4_shift; |
1536 | 0 | i4_blk_y = i4_search_blk_y << ps_prms->i4_shift; |
1537 | 0 | i4_offset = i4_blk_x + i4_blk_y * ps_layer_mv->i4_num_blks_per_row; |
1538 | |
|
1539 | 0 | i4_offset *= ps_layer_mv->i4_num_mvs_per_blk; |
1540 | | |
1541 | | /* Identify the correct offset in the mvbank and the reference id buf */ |
1542 | 0 | ps_mv = ps_layer_mv->ps_mv + i4_offset; |
1543 | 0 | pi1_ref_idx = ps_layer_mv->pi1_ref_idx + i4_offset; |
1544 | | |
1545 | | /*************************************************************************/ |
1546 | | /* Supposing we store the mvs in the same blk size as we searched (e.g. */ |
1547 | | /* we searched 8x8 blks and store results for 8x8 blks), then we can */ |
1548 | | /* do a straightforward single update of results. This will have a 1-1 */ |
1549 | | /* correspondence. */ |
1550 | | /*************************************************************************/ |
1551 | 0 | if(ps_layer_mv->e_blk_size == ps_prms->e_search_blk_size) |
1552 | 0 | { |
1553 | 0 | for(i4_ref_id = 0; i4_ref_id < (S32)ps_prms->i4_num_ref; i4_ref_id++) |
1554 | 0 | { |
1555 | 0 | ps_search_node = ps_search_results->aps_part_results[i4_ref_id][PART_ID_2Nx2N]; |
1556 | 0 | for(i4_j = 0; i4_j < ps_layer_mv->i4_num_mvs_per_ref; i4_j++) |
1557 | 0 | { |
1558 | 0 | COPY_SEARCH_RESULT(ps_mv, pi1_ref_idx, ps_search_node, 0); |
1559 | 0 | ps_mv++; |
1560 | 0 | pi1_ref_idx++; |
1561 | 0 | ps_search_node++; |
1562 | 0 | } |
1563 | 0 | } |
1564 | 0 | return; |
1565 | 0 | } |
1566 | | |
1567 | | /*************************************************************************/ |
1568 | | /* Case where search blk size is 8x8, but we update 4x4 results. In this */ |
1569 | | /* case, we need to have NxN partitions enabled in search. */ |
1570 | | /* Further, we update on a 1-1 basis the 4x4 blk mvs from the respective */ |
1571 | | /* NxN partition. We also update the 8x8 result into each of the 4x4 bank*/ |
1572 | | /*************************************************************************/ |
1573 | 0 | ASSERT(ps_layer_mv->e_blk_size == BLK_4x4); |
1574 | 0 | ASSERT(ps_prms->e_search_blk_size == BLK_8x8); |
1575 | 0 | ASSERT((ps_search_results->i4_part_mask & (ENABLE_NxN)) == (ENABLE_NxN)); |
1576 | | |
1577 | | /*************************************************************************/ |
1578 | | /* For every 4x4 blk we store corresponding 4x4 results and 1 8x8 result */ |
1579 | | /* hence the below check. */ |
1580 | | /*************************************************************************/ |
1581 | 0 | ASSERT(ps_layer_mv->i4_num_mvs_per_ref <= ps_search_results->u1_num_results_per_part + 1); |
1582 | | |
1583 | 0 | ps_mv1 = ps_mv; |
1584 | 0 | ps_mv2 = ps_mv1 + ps_layer_mv->i4_num_mvs_per_blk; |
1585 | 0 | ps_mv3 = ps_mv1 + (ps_layer_mv->i4_num_mvs_per_row); |
1586 | 0 | ps_mv4 = ps_mv3 + (ps_layer_mv->i4_num_mvs_per_blk); |
1587 | 0 | pi1_ref_idx1 = pi1_ref_idx; |
1588 | 0 | pi1_ref_idx2 = pi1_ref_idx1 + ps_layer_mv->i4_num_mvs_per_blk; |
1589 | 0 | pi1_ref_idx3 = pi1_ref_idx1 + (ps_layer_mv->i4_num_mvs_per_row); |
1590 | 0 | pi1_ref_idx4 = pi1_ref_idx3 + (ps_layer_mv->i4_num_mvs_per_blk); |
1591 | |
|
1592 | 0 | for(i4_ref_id = 0; i4_ref_id < (S32)ps_search_results->u1_num_active_ref; i4_ref_id++) |
1593 | 0 | { |
1594 | 0 | ps_search_node_8x8 = ps_search_results->aps_part_results[i4_ref_id][PART_ID_2Nx2N]; |
1595 | |
|
1596 | 0 | ps_search_node_4x4_1 = ps_search_results->aps_part_results[i4_ref_id][PART_ID_NxN_TL]; |
1597 | |
|
1598 | 0 | ps_search_node_4x4_2 = ps_search_results->aps_part_results[i4_ref_id][PART_ID_NxN_TR]; |
1599 | |
|
1600 | 0 | ps_search_node_4x4_3 = ps_search_results->aps_part_results[i4_ref_id][PART_ID_NxN_BL]; |
1601 | |
|
1602 | 0 | ps_search_node_4x4_4 = ps_search_results->aps_part_results[i4_ref_id][PART_ID_NxN_BR]; |
1603 | |
|
1604 | 0 | COPY_SEARCH_RESULT(ps_mv1, pi1_ref_idx1, ps_search_node_4x4_1, 0); |
1605 | 0 | ps_mv1++; |
1606 | 0 | pi1_ref_idx1++; |
1607 | 0 | ps_search_node_4x4_1++; |
1608 | 0 | COPY_SEARCH_RESULT(ps_mv2, pi1_ref_idx2, ps_search_node_4x4_2, 0); |
1609 | 0 | ps_mv2++; |
1610 | 0 | pi1_ref_idx2++; |
1611 | 0 | ps_search_node_4x4_2++; |
1612 | 0 | COPY_SEARCH_RESULT(ps_mv3, pi1_ref_idx3, ps_search_node_4x4_3, 0); |
1613 | 0 | ps_mv3++; |
1614 | 0 | pi1_ref_idx3++; |
1615 | 0 | ps_search_node_4x4_3++; |
1616 | 0 | COPY_SEARCH_RESULT(ps_mv4, pi1_ref_idx4, ps_search_node_4x4_4, 0); |
1617 | 0 | ps_mv4++; |
1618 | 0 | pi1_ref_idx4++; |
1619 | 0 | ps_search_node_4x4_4++; |
1620 | |
|
1621 | 0 | if(ps_layer_mv->i4_num_mvs_per_ref > 1) |
1622 | 0 | { |
1623 | 0 | COPY_SEARCH_RESULT(ps_mv1, pi1_ref_idx1, ps_search_node_8x8, 0); |
1624 | 0 | ps_mv1++; |
1625 | 0 | pi1_ref_idx1++; |
1626 | 0 | COPY_SEARCH_RESULT(ps_mv2, pi1_ref_idx2, ps_search_node_8x8, 0); |
1627 | 0 | ps_mv2++; |
1628 | 0 | pi1_ref_idx2++; |
1629 | 0 | COPY_SEARCH_RESULT(ps_mv3, pi1_ref_idx3, ps_search_node_8x8, 0); |
1630 | 0 | ps_mv3++; |
1631 | 0 | pi1_ref_idx3++; |
1632 | 0 | COPY_SEARCH_RESULT(ps_mv4, pi1_ref_idx4, ps_search_node_8x8, 0); |
1633 | 0 | ps_mv4++; |
1634 | 0 | pi1_ref_idx4++; |
1635 | 0 | } |
1636 | |
|
1637 | 0 | for(i4_j = 2; i4_j < ps_layer_mv->i4_num_mvs_per_ref; i4_j++) |
1638 | 0 | { |
1639 | 0 | COPY_SEARCH_RESULT(ps_mv1, pi1_ref_idx1, ps_search_node_4x4_1, 0); |
1640 | 0 | ps_mv1++; |
1641 | 0 | pi1_ref_idx1++; |
1642 | 0 | ps_search_node_4x4_1++; |
1643 | 0 | COPY_SEARCH_RESULT(ps_mv2, pi1_ref_idx2, ps_search_node_4x4_2, 0); |
1644 | 0 | ps_mv2++; |
1645 | 0 | pi1_ref_idx2++; |
1646 | 0 | ps_search_node_4x4_2++; |
1647 | 0 | COPY_SEARCH_RESULT(ps_mv3, pi1_ref_idx3, ps_search_node_4x4_3, 0); |
1648 | 0 | ps_mv3++; |
1649 | 0 | pi1_ref_idx3++; |
1650 | 0 | ps_search_node_4x4_3++; |
1651 | 0 | COPY_SEARCH_RESULT(ps_mv4, pi1_ref_idx4, ps_search_node_4x4_4, 0); |
1652 | 0 | ps_mv4++; |
1653 | 0 | pi1_ref_idx4++; |
1654 | 0 | ps_search_node_4x4_4++; |
1655 | 0 | } |
1656 | 0 | } |
1657 | 0 | } |
1658 | | |
1659 | | void hme_update_mv_bank_encode( |
1660 | | search_results_t *ps_search_results, |
1661 | | layer_mv_t *ps_layer_mv, |
1662 | | S32 i4_search_blk_x, |
1663 | | S32 i4_search_blk_y, |
1664 | | mvbank_update_prms_t *ps_prms, |
1665 | | U08 *pu1_pred_dir_searched, |
1666 | | S32 i4_num_act_ref_l0) |
1667 | 0 | { |
1668 | 0 | hme_mv_t *ps_mv; |
1669 | 0 | hme_mv_t *ps_mv1, *ps_mv2, *ps_mv3, *ps_mv4; |
1670 | 0 | S08 *pi1_ref_idx, *pi1_ref_idx1, *pi1_ref_idx2, *pi1_ref_idx3, *pi1_ref_idx4; |
1671 | 0 | S32 i4_blk_x, i4_blk_y, i4_offset; |
1672 | 0 | S32 j, i, num_parts; |
1673 | 0 | search_node_t *ps_search_node_tl, *ps_search_node_tr; |
1674 | 0 | search_node_t *ps_search_node_bl, *ps_search_node_br; |
1675 | 0 | search_node_t s_zero_mv; |
1676 | 0 | WORD32 i4_part_type = ps_search_results->ps_cu_results->ps_best_results[0].u1_part_type; |
1677 | |
|
1678 | 0 | i4_blk_x = i4_search_blk_x << ps_prms->i4_shift; |
1679 | 0 | i4_blk_y = i4_search_blk_y << ps_prms->i4_shift; |
1680 | 0 | i4_offset = i4_blk_x + i4_blk_y * ps_layer_mv->i4_num_blks_per_row; |
1681 | |
|
1682 | 0 | i4_offset *= ps_layer_mv->i4_num_mvs_per_blk; |
1683 | | |
1684 | | /* Identify the correct offset in the mvbank and the reference id buf */ |
1685 | 0 | ps_mv = ps_layer_mv->ps_mv + i4_offset; |
1686 | 0 | pi1_ref_idx = ps_layer_mv->pi1_ref_idx + i4_offset; |
1687 | |
|
1688 | 0 | ASSERT(ps_layer_mv->e_blk_size == BLK_8x8); |
1689 | 0 | ASSERT(ps_prms->e_search_blk_size == BLK_16x16); |
1690 | | |
1691 | | /*************************************************************************/ |
1692 | | /* For every 4x4 blk we store corresponding 4x4 results and 1 8x8 result */ |
1693 | | /* hence the below check. */ |
1694 | | /*************************************************************************/ |
1695 | 0 | ASSERT(ps_layer_mv->i4_num_mvs_per_ref <= ps_search_results->u1_num_best_results); |
1696 | | |
1697 | 0 | ps_mv1 = ps_mv; |
1698 | 0 | ps_mv2 = ps_mv1 + ps_layer_mv->i4_num_mvs_per_blk; |
1699 | 0 | ps_mv3 = ps_mv1 + (ps_layer_mv->i4_num_mvs_per_row); |
1700 | 0 | ps_mv4 = ps_mv3 + (ps_layer_mv->i4_num_mvs_per_blk); |
1701 | 0 | pi1_ref_idx1 = pi1_ref_idx; |
1702 | 0 | pi1_ref_idx2 = pi1_ref_idx1 + ps_layer_mv->i4_num_mvs_per_blk; |
1703 | 0 | pi1_ref_idx3 = pi1_ref_idx1 + (ps_layer_mv->i4_num_mvs_per_row); |
1704 | 0 | pi1_ref_idx4 = pi1_ref_idx3 + (ps_layer_mv->i4_num_mvs_per_blk); |
1705 | | |
1706 | | /* Initialize zero mv: default mv used for intra mvs */ |
1707 | 0 | s_zero_mv.s_mv.i2_mvx = 0; |
1708 | 0 | s_zero_mv.s_mv.i2_mvy = 0; |
1709 | 0 | s_zero_mv.i1_ref_idx = 0; |
1710 | |
|
1711 | 0 | if((ps_search_results->e_cu_size == CU_16x16) && (ps_search_results->u1_split_flag) && |
1712 | 0 | (ps_search_results->i4_part_mask & ENABLE_NxN)) |
1713 | 0 | { |
1714 | 0 | i4_part_type = PRT_NxN; |
1715 | 0 | } |
1716 | |
|
1717 | 0 | for(i = 0; i < ps_prms->i4_num_ref; i++) |
1718 | 0 | { |
1719 | 0 | for(j = 0; j < ps_layer_mv->i4_num_mvs_per_ref; j++) |
1720 | 0 | { |
1721 | 0 | WORD32 i4_part_id = ge_part_type_to_part_id[i4_part_type][0]; |
1722 | |
|
1723 | 0 | num_parts = gau1_num_parts_in_part_type[i4_part_type]; |
1724 | |
|
1725 | 0 | ps_search_node_tl = |
1726 | 0 | ps_search_results->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id]; |
1727 | |
|
1728 | 0 | if(num_parts == 1) |
1729 | 0 | { |
1730 | 0 | ps_search_node_tr = ps_search_node_tl; |
1731 | 0 | ps_search_node_bl = ps_search_node_tl; |
1732 | 0 | ps_search_node_br = ps_search_node_tl; |
1733 | 0 | } |
1734 | 0 | else if(num_parts == 2) |
1735 | 0 | { |
1736 | | /* For vertically oriented partitions, tl, bl pt to same result */ |
1737 | | /* For horizontally oriented partition, tl, tr pt to same result */ |
1738 | | /* This means for AMP, 2 of the 8x8 blks in mv bank have ambiguous */ |
1739 | | /* result, e.g. for 4x16L. Here left 2 8x8 have the 4x16L partition */ |
1740 | | /* and right 2 8x8 have 12x16R partition */ |
1741 | 0 | if(gau1_is_vert_part[i4_part_type]) |
1742 | 0 | { |
1743 | 0 | ps_search_node_tr = |
1744 | 0 | ps_search_results |
1745 | 0 | ->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 1]; |
1746 | 0 | ps_search_node_bl = ps_search_node_tl; |
1747 | 0 | } |
1748 | 0 | else |
1749 | 0 | { |
1750 | 0 | ps_search_node_tr = ps_search_node_tl; |
1751 | 0 | ps_search_node_bl = |
1752 | 0 | ps_search_results |
1753 | 0 | ->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 1]; |
1754 | 0 | } |
1755 | 0 | ps_search_node_br = |
1756 | 0 | ps_search_results->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 1]; |
1757 | 0 | } |
1758 | 0 | else |
1759 | 0 | { |
1760 | | /* 4 unique results */ |
1761 | 0 | ps_search_node_tr = |
1762 | 0 | ps_search_results->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 1]; |
1763 | 0 | ps_search_node_bl = |
1764 | 0 | ps_search_results->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 2]; |
1765 | 0 | ps_search_node_br = |
1766 | 0 | ps_search_results->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 3]; |
1767 | 0 | } |
1768 | |
|
1769 | 0 | if(ps_search_node_tl->s_mv.i2_mvx == INTRA_MV) |
1770 | 0 | ps_search_node_tl++; |
1771 | 0 | if(ps_search_node_tr->s_mv.i2_mvx == INTRA_MV) |
1772 | 0 | ps_search_node_tr++; |
1773 | 0 | if(ps_search_node_bl->s_mv.i2_mvx == INTRA_MV) |
1774 | 0 | ps_search_node_bl++; |
1775 | 0 | if(ps_search_node_br->s_mv.i2_mvx == INTRA_MV) |
1776 | 0 | ps_search_node_br++; |
1777 | |
|
1778 | 0 | COPY_SEARCH_RESULT(ps_mv1, pi1_ref_idx1, ps_search_node_tl, 0); |
1779 | 0 | ps_mv1++; |
1780 | 0 | pi1_ref_idx1++; |
1781 | 0 | COPY_SEARCH_RESULT(ps_mv2, pi1_ref_idx2, ps_search_node_tr, 0); |
1782 | 0 | ps_mv2++; |
1783 | 0 | pi1_ref_idx2++; |
1784 | 0 | COPY_SEARCH_RESULT(ps_mv3, pi1_ref_idx3, ps_search_node_bl, 0); |
1785 | 0 | ps_mv3++; |
1786 | 0 | pi1_ref_idx3++; |
1787 | 0 | COPY_SEARCH_RESULT(ps_mv4, pi1_ref_idx4, ps_search_node_br, 0); |
1788 | 0 | ps_mv4++; |
1789 | 0 | pi1_ref_idx4++; |
1790 | |
|
1791 | 0 | if(ps_prms->i4_num_results_to_store > 1) |
1792 | 0 | { |
1793 | 0 | ps_search_node_tl = |
1794 | 0 | &ps_search_results->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id][1]; |
1795 | |
|
1796 | 0 | if(num_parts == 1) |
1797 | 0 | { |
1798 | 0 | ps_search_node_tr = ps_search_node_tl; |
1799 | 0 | ps_search_node_bl = ps_search_node_tl; |
1800 | 0 | ps_search_node_br = ps_search_node_tl; |
1801 | 0 | } |
1802 | 0 | else if(num_parts == 2) |
1803 | 0 | { |
1804 | | /* For vertically oriented partitions, tl, bl pt to same result */ |
1805 | | /* For horizontally oriented partition, tl, tr pt to same result */ |
1806 | | /* This means for AMP, 2 of the 8x8 blks in mv bank have ambiguous */ |
1807 | | /* result, e.g. for 4x16L. Here left 2 8x8 have the 4x16L partition */ |
1808 | | /* and right 2 8x8 have 12x16R partition */ |
1809 | 0 | if(gau1_is_vert_part[i4_part_type]) |
1810 | 0 | { |
1811 | 0 | ps_search_node_tr = |
1812 | 0 | &ps_search_results |
1813 | 0 | ->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 1][1]; |
1814 | 0 | ps_search_node_bl = ps_search_node_tl; |
1815 | 0 | } |
1816 | 0 | else |
1817 | 0 | { |
1818 | 0 | ps_search_node_tr = ps_search_node_tl; |
1819 | 0 | ps_search_node_bl = |
1820 | 0 | &ps_search_results |
1821 | 0 | ->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 1][1]; |
1822 | 0 | } |
1823 | 0 | ps_search_node_br = |
1824 | 0 | &ps_search_results |
1825 | 0 | ->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 1][1]; |
1826 | 0 | } |
1827 | 0 | else |
1828 | 0 | { |
1829 | | /* 4 unique results */ |
1830 | 0 | ps_search_node_tr = |
1831 | 0 | &ps_search_results |
1832 | 0 | ->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 1][1]; |
1833 | 0 | ps_search_node_bl = |
1834 | 0 | &ps_search_results |
1835 | 0 | ->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 2][1]; |
1836 | 0 | ps_search_node_br = |
1837 | 0 | &ps_search_results |
1838 | 0 | ->aps_part_results[pu1_pred_dir_searched[i]][i4_part_id + 3][1]; |
1839 | 0 | } |
1840 | |
|
1841 | 0 | if(ps_search_node_tl->s_mv.i2_mvx == INTRA_MV) |
1842 | 0 | ps_search_node_tl++; |
1843 | 0 | if(ps_search_node_tr->s_mv.i2_mvx == INTRA_MV) |
1844 | 0 | ps_search_node_tr++; |
1845 | 0 | if(ps_search_node_bl->s_mv.i2_mvx == INTRA_MV) |
1846 | 0 | ps_search_node_bl++; |
1847 | 0 | if(ps_search_node_br->s_mv.i2_mvx == INTRA_MV) |
1848 | 0 | ps_search_node_br++; |
1849 | |
|
1850 | 0 | COPY_SEARCH_RESULT(ps_mv1, pi1_ref_idx1, ps_search_node_tl, 0); |
1851 | 0 | ps_mv1++; |
1852 | 0 | pi1_ref_idx1++; |
1853 | 0 | COPY_SEARCH_RESULT(ps_mv2, pi1_ref_idx2, ps_search_node_tr, 0); |
1854 | 0 | ps_mv2++; |
1855 | 0 | pi1_ref_idx2++; |
1856 | 0 | COPY_SEARCH_RESULT(ps_mv3, pi1_ref_idx3, ps_search_node_bl, 0); |
1857 | 0 | ps_mv3++; |
1858 | 0 | pi1_ref_idx3++; |
1859 | 0 | COPY_SEARCH_RESULT(ps_mv4, pi1_ref_idx4, ps_search_node_br, 0); |
1860 | 0 | ps_mv4++; |
1861 | 0 | pi1_ref_idx4++; |
1862 | 0 | } |
1863 | 0 | } |
1864 | 0 | } |
1865 | 0 | } |
1866 | | |
1867 | | /** |
1868 | | ******************************************************************************** |
1869 | | * @fn hme_update_mv_bank_noencode(search_results_t *ps_search_results, |
1870 | | * layer_mv_t *ps_layer_mv, |
1871 | | * S32 i4_search_blk_x, |
1872 | | * S32 i4_search_blk_y, |
1873 | | * mvbank_update_prms_t *ps_prms) |
1874 | | * |
1875 | | * @brief Updates the mv bank in case there is no further encodign to be done |
1876 | | * |
1877 | | * @param[in] ps_search_results: contains results for the block just searched |
1878 | | * |
1879 | | * @param[in,out] ps_layer_mv : Has pointer to mv bank amongst other things |
1880 | | * |
1881 | | * @param[in] i4_search_blk_x : col num of blk being searched |
1882 | | * |
1883 | | * @param[in] i4_search_blk_y : row num of blk being searched |
1884 | | * |
1885 | | * @param[in] ps_prms : contains certain parameters which govern how updatedone |
1886 | | * |
1887 | | * @return None |
1888 | | ******************************************************************************** |
1889 | | */ |
1890 | | |
1891 | | void hme_update_mv_bank_in_l1_me( |
1892 | | search_results_t *ps_search_results, |
1893 | | layer_mv_t *ps_layer_mv, |
1894 | | S32 i4_search_blk_x, |
1895 | | S32 i4_search_blk_y, |
1896 | | mvbank_update_prms_t *ps_prms) |
1897 | 0 | { |
1898 | 0 | hme_mv_t *ps_mv; |
1899 | 0 | hme_mv_t *ps_mv1, *ps_mv2, *ps_mv3, *ps_mv4; |
1900 | 0 | S08 *pi1_ref_idx, *pi1_ref_idx1, *pi1_ref_idx2, *pi1_ref_idx3, *pi1_ref_idx4; |
1901 | 0 | S32 i4_blk_x, i4_blk_y, i4_offset; |
1902 | 0 | S32 i4_j, i4_ref_id; |
1903 | 0 | search_node_t *ps_search_node; |
1904 | 0 | search_node_t *ps_search_node_8x8, *ps_search_node_4x4; |
1905 | |
|
1906 | 0 | i4_blk_x = i4_search_blk_x << ps_prms->i4_shift; |
1907 | 0 | i4_blk_y = i4_search_blk_y << ps_prms->i4_shift; |
1908 | 0 | i4_offset = i4_blk_x + i4_blk_y * ps_layer_mv->i4_num_blks_per_row; |
1909 | |
|
1910 | 0 | i4_offset *= ps_layer_mv->i4_num_mvs_per_blk; |
1911 | | |
1912 | | /* Identify the correct offset in the mvbank and the reference id buf */ |
1913 | 0 | ps_mv = ps_layer_mv->ps_mv + i4_offset; |
1914 | 0 | pi1_ref_idx = ps_layer_mv->pi1_ref_idx + i4_offset; |
1915 | | |
1916 | | /*************************************************************************/ |
1917 | | /* Supposing we store the mvs in the same blk size as we searched (e.g. */ |
1918 | | /* we searched 8x8 blks and store results for 8x8 blks), then we can */ |
1919 | | /* do a straightforward single update of results. This will have a 1-1 */ |
1920 | | /* correspondence. */ |
1921 | | /*************************************************************************/ |
1922 | 0 | if(ps_layer_mv->e_blk_size == ps_prms->e_search_blk_size) |
1923 | 0 | { |
1924 | 0 | search_node_t *aps_result_nodes_sorted[2][MAX_NUM_REF * 2]; |
1925 | |
|
1926 | 0 | hme_mv_t *ps_mv_l0_root = ps_mv; |
1927 | 0 | hme_mv_t *ps_mv_l1_root = |
1928 | 0 | ps_mv + (ps_prms->i4_num_active_ref_l0 * ps_layer_mv->i4_num_mvs_per_ref); |
1929 | |
|
1930 | 0 | U32 u4_num_l0_results_updated = 0; |
1931 | 0 | U32 u4_num_l1_results_updated = 0; |
1932 | |
|
1933 | 0 | S08 *pi1_ref_idx_l0_root = pi1_ref_idx; |
1934 | 0 | S08 *pi1_ref_idx_l1_root = |
1935 | 0 | pi1_ref_idx_l0_root + (ps_prms->i4_num_active_ref_l0 * ps_layer_mv->i4_num_mvs_per_ref); |
1936 | |
|
1937 | 0 | for(i4_ref_id = 0; i4_ref_id < (S32)ps_prms->i4_num_ref; i4_ref_id++) |
1938 | 0 | { |
1939 | 0 | U32 *pu4_num_results_updated; |
1940 | 0 | search_node_t **pps_result_nodes; |
1941 | |
|
1942 | 0 | U08 u1_pred_dir_of_cur_ref = !ps_search_results->pu1_is_past[i4_ref_id]; |
1943 | |
|
1944 | 0 | if(u1_pred_dir_of_cur_ref) |
1945 | 0 | { |
1946 | 0 | pu4_num_results_updated = &u4_num_l1_results_updated; |
1947 | 0 | pps_result_nodes = &aps_result_nodes_sorted[1][0]; |
1948 | 0 | } |
1949 | 0 | else |
1950 | 0 | { |
1951 | 0 | pu4_num_results_updated = &u4_num_l0_results_updated; |
1952 | 0 | pps_result_nodes = &aps_result_nodes_sorted[0][0]; |
1953 | 0 | } |
1954 | |
|
1955 | 0 | ps_search_node = ps_search_results->aps_part_results[i4_ref_id][PART_ID_2Nx2N]; |
1956 | |
|
1957 | 0 | for(i4_j = 0; i4_j < ps_layer_mv->i4_num_mvs_per_ref; i4_j++) |
1958 | 0 | { |
1959 | 0 | hme_add_new_node_to_a_sorted_array( |
1960 | 0 | &ps_search_node[i4_j], pps_result_nodes, NULL, *pu4_num_results_updated, 0); |
1961 | |
|
1962 | 0 | ASSERT(ps_search_node[i4_j].i1_ref_idx == i4_ref_id); |
1963 | 0 | (*pu4_num_results_updated)++; |
1964 | 0 | } |
1965 | 0 | } |
1966 | | |
1967 | 0 | for(i4_j = 0; i4_j < (S32)u4_num_l0_results_updated; i4_j++) |
1968 | 0 | { |
1969 | 0 | COPY_SEARCH_RESULT( |
1970 | 0 | &ps_mv_l0_root[i4_j], |
1971 | 0 | &pi1_ref_idx_l0_root[i4_j], |
1972 | 0 | aps_result_nodes_sorted[0][i4_j], |
1973 | 0 | 0); |
1974 | 0 | } |
1975 | |
|
1976 | 0 | for(i4_j = 0; i4_j < (S32)u4_num_l1_results_updated; i4_j++) |
1977 | 0 | { |
1978 | 0 | COPY_SEARCH_RESULT( |
1979 | 0 | &ps_mv_l1_root[i4_j], |
1980 | 0 | &pi1_ref_idx_l1_root[i4_j], |
1981 | 0 | aps_result_nodes_sorted[1][i4_j], |
1982 | 0 | 0); |
1983 | 0 | } |
1984 | |
|
1985 | 0 | return; |
1986 | 0 | } |
1987 | | |
1988 | | /*************************************************************************/ |
1989 | | /* Case where search blk size is 8x8, but we update 4x4 results. In this */ |
1990 | | /* case, we need to have NxN partitions enabled in search. */ |
1991 | | /* Further, we update on a 1-1 basis the 4x4 blk mvs from the respective */ |
1992 | | /* NxN partition. We also update the 8x8 result into each of the 4x4 bank*/ |
1993 | | /*************************************************************************/ |
1994 | 0 | ASSERT(ps_layer_mv->e_blk_size == BLK_4x4); |
1995 | 0 | ASSERT(ps_prms->e_search_blk_size == BLK_8x8); |
1996 | 0 | ASSERT((ps_search_results->i4_part_mask & (ENABLE_NxN)) == (ENABLE_NxN)); |
1997 | | |
1998 | | /*************************************************************************/ |
1999 | | /* For every 4x4 blk we store corresponding 4x4 results and 1 8x8 result */ |
2000 | | /* hence the below check. */ |
2001 | | /*************************************************************************/ |
2002 | 0 | ASSERT(ps_layer_mv->i4_num_mvs_per_ref <= ps_search_results->u1_num_results_per_part + 1); |
2003 | | |
2004 | 0 | ps_mv1 = ps_mv; |
2005 | 0 | ps_mv2 = ps_mv1 + ps_layer_mv->i4_num_mvs_per_blk; |
2006 | 0 | ps_mv3 = ps_mv1 + (ps_layer_mv->i4_num_mvs_per_row); |
2007 | 0 | ps_mv4 = ps_mv3 + (ps_layer_mv->i4_num_mvs_per_blk); |
2008 | 0 | pi1_ref_idx1 = pi1_ref_idx; |
2009 | 0 | pi1_ref_idx2 = pi1_ref_idx1 + ps_layer_mv->i4_num_mvs_per_blk; |
2010 | 0 | pi1_ref_idx3 = pi1_ref_idx1 + (ps_layer_mv->i4_num_mvs_per_row); |
2011 | 0 | pi1_ref_idx4 = pi1_ref_idx3 + (ps_layer_mv->i4_num_mvs_per_blk); |
2012 | |
|
2013 | 0 | { |
2014 | | /* max ref frames * max results per partition * number of partitions (4x4, 8x8) */ |
2015 | 0 | search_node_t *aps_result_nodes_sorted[2][MAX_NUM_REF * MAX_RESULTS_PER_PART * 2]; |
2016 | 0 | U08 au1_cost_shifts_for_sorted_node[2][MAX_NUM_REF * MAX_RESULTS_PER_PART * 2]; |
2017 | |
|
2018 | 0 | S32 i; |
2019 | |
|
2020 | 0 | hme_mv_t *ps_mv1_l0_root = ps_mv1; |
2021 | 0 | hme_mv_t *ps_mv1_l1_root = |
2022 | 0 | ps_mv1 + (ps_prms->i4_num_active_ref_l0 * ps_layer_mv->i4_num_mvs_per_ref); |
2023 | 0 | hme_mv_t *ps_mv2_l0_root = ps_mv2; |
2024 | 0 | hme_mv_t *ps_mv2_l1_root = |
2025 | 0 | ps_mv2 + (ps_prms->i4_num_active_ref_l0 * ps_layer_mv->i4_num_mvs_per_ref); |
2026 | 0 | hme_mv_t *ps_mv3_l0_root = ps_mv3; |
2027 | 0 | hme_mv_t *ps_mv3_l1_root = |
2028 | 0 | ps_mv3 + (ps_prms->i4_num_active_ref_l0 * ps_layer_mv->i4_num_mvs_per_ref); |
2029 | 0 | hme_mv_t *ps_mv4_l0_root = ps_mv4; |
2030 | 0 | hme_mv_t *ps_mv4_l1_root = |
2031 | 0 | ps_mv4 + (ps_prms->i4_num_active_ref_l0 * ps_layer_mv->i4_num_mvs_per_ref); |
2032 | |
|
2033 | 0 | U32 u4_num_l0_results_updated = 0; |
2034 | 0 | U32 u4_num_l1_results_updated = 0; |
2035 | |
|
2036 | 0 | S08 *pi1_ref_idx1_l0_root = pi1_ref_idx1; |
2037 | 0 | S08 *pi1_ref_idx1_l1_root = pi1_ref_idx1_l0_root + (ps_prms->i4_num_active_ref_l0 * |
2038 | 0 | ps_layer_mv->i4_num_mvs_per_ref); |
2039 | 0 | S08 *pi1_ref_idx2_l0_root = pi1_ref_idx2; |
2040 | 0 | S08 *pi1_ref_idx2_l1_root = pi1_ref_idx2_l0_root + (ps_prms->i4_num_active_ref_l0 * |
2041 | 0 | ps_layer_mv->i4_num_mvs_per_ref); |
2042 | 0 | S08 *pi1_ref_idx3_l0_root = pi1_ref_idx3; |
2043 | 0 | S08 *pi1_ref_idx3_l1_root = pi1_ref_idx3_l0_root + (ps_prms->i4_num_active_ref_l0 * |
2044 | 0 | ps_layer_mv->i4_num_mvs_per_ref); |
2045 | 0 | S08 *pi1_ref_idx4_l0_root = pi1_ref_idx4; |
2046 | 0 | S08 *pi1_ref_idx4_l1_root = pi1_ref_idx4_l0_root + (ps_prms->i4_num_active_ref_l0 * |
2047 | 0 | ps_layer_mv->i4_num_mvs_per_ref); |
2048 | |
|
2049 | 0 | for(i = 0; i < 4; i++) |
2050 | 0 | { |
2051 | 0 | hme_mv_t *ps_mv_l0_root; |
2052 | 0 | hme_mv_t *ps_mv_l1_root; |
2053 | |
|
2054 | 0 | S08 *pi1_ref_idx_l0_root; |
2055 | 0 | S08 *pi1_ref_idx_l1_root; |
2056 | |
|
2057 | 0 | for(i4_ref_id = 0; i4_ref_id < ps_search_results->u1_num_active_ref; i4_ref_id++) |
2058 | 0 | { |
2059 | 0 | U32 *pu4_num_results_updated; |
2060 | 0 | search_node_t **pps_result_nodes; |
2061 | 0 | U08 *pu1_cost_shifts_for_sorted_node; |
2062 | |
|
2063 | 0 | U08 u1_pred_dir_of_cur_ref = !ps_search_results->pu1_is_past[i4_ref_id]; |
2064 | |
|
2065 | 0 | if(u1_pred_dir_of_cur_ref) |
2066 | 0 | { |
2067 | 0 | pu4_num_results_updated = &u4_num_l1_results_updated; |
2068 | 0 | pps_result_nodes = &aps_result_nodes_sorted[1][0]; |
2069 | 0 | pu1_cost_shifts_for_sorted_node = &au1_cost_shifts_for_sorted_node[1][0]; |
2070 | 0 | } |
2071 | 0 | else |
2072 | 0 | { |
2073 | 0 | pu4_num_results_updated = &u4_num_l0_results_updated; |
2074 | 0 | pps_result_nodes = &aps_result_nodes_sorted[0][0]; |
2075 | 0 | pu1_cost_shifts_for_sorted_node = &au1_cost_shifts_for_sorted_node[1][0]; |
2076 | 0 | } |
2077 | |
|
2078 | 0 | ps_search_node_8x8 = ps_search_results->aps_part_results[i4_ref_id][PART_ID_2Nx2N]; |
2079 | |
|
2080 | 0 | ps_search_node_4x4 = |
2081 | 0 | ps_search_results->aps_part_results[i4_ref_id][PART_ID_NxN_TL + i]; |
2082 | |
|
2083 | 0 | for(i4_j = 0; i4_j < ps_layer_mv->i4_num_mvs_per_ref; i4_j++) |
2084 | 0 | { |
2085 | 0 | hme_add_new_node_to_a_sorted_array( |
2086 | 0 | &ps_search_node_4x4[i4_j], |
2087 | 0 | pps_result_nodes, |
2088 | 0 | pu1_cost_shifts_for_sorted_node, |
2089 | 0 | *pu4_num_results_updated, |
2090 | 0 | 0); |
2091 | |
|
2092 | 0 | (*pu4_num_results_updated)++; |
2093 | |
|
2094 | 0 | hme_add_new_node_to_a_sorted_array( |
2095 | 0 | &ps_search_node_8x8[i4_j], |
2096 | 0 | pps_result_nodes, |
2097 | 0 | pu1_cost_shifts_for_sorted_node, |
2098 | 0 | *pu4_num_results_updated, |
2099 | 0 | 2); |
2100 | |
|
2101 | 0 | (*pu4_num_results_updated)++; |
2102 | 0 | } |
2103 | 0 | } |
2104 | |
|
2105 | 0 | switch(i) |
2106 | 0 | { |
2107 | 0 | case 0: |
2108 | 0 | { |
2109 | 0 | ps_mv_l0_root = ps_mv1_l0_root; |
2110 | 0 | ps_mv_l1_root = ps_mv1_l1_root; |
2111 | |
|
2112 | 0 | pi1_ref_idx_l0_root = pi1_ref_idx1_l0_root; |
2113 | 0 | pi1_ref_idx_l1_root = pi1_ref_idx1_l1_root; |
2114 | |
|
2115 | 0 | break; |
2116 | 0 | } |
2117 | 0 | case 1: |
2118 | 0 | { |
2119 | 0 | ps_mv_l0_root = ps_mv2_l0_root; |
2120 | 0 | ps_mv_l1_root = ps_mv2_l1_root; |
2121 | |
|
2122 | 0 | pi1_ref_idx_l0_root = pi1_ref_idx2_l0_root; |
2123 | 0 | pi1_ref_idx_l1_root = pi1_ref_idx2_l1_root; |
2124 | |
|
2125 | 0 | break; |
2126 | 0 | } |
2127 | 0 | case 2: |
2128 | 0 | { |
2129 | 0 | ps_mv_l0_root = ps_mv3_l0_root; |
2130 | 0 | ps_mv_l1_root = ps_mv3_l1_root; |
2131 | |
|
2132 | 0 | pi1_ref_idx_l0_root = pi1_ref_idx3_l0_root; |
2133 | 0 | pi1_ref_idx_l1_root = pi1_ref_idx3_l1_root; |
2134 | |
|
2135 | 0 | break; |
2136 | 0 | } |
2137 | 0 | case 3: |
2138 | 0 | { |
2139 | 0 | ps_mv_l0_root = ps_mv4_l0_root; |
2140 | 0 | ps_mv_l1_root = ps_mv4_l1_root; |
2141 | |
|
2142 | 0 | pi1_ref_idx_l0_root = pi1_ref_idx4_l0_root; |
2143 | 0 | pi1_ref_idx_l1_root = pi1_ref_idx4_l1_root; |
2144 | |
|
2145 | 0 | break; |
2146 | 0 | } |
2147 | 0 | } |
2148 | | |
2149 | 0 | u4_num_l0_results_updated = |
2150 | 0 | MIN((S32)u4_num_l0_results_updated, |
2151 | 0 | ps_prms->i4_num_active_ref_l0 * ps_layer_mv->i4_num_mvs_per_ref); |
2152 | |
|
2153 | 0 | u4_num_l1_results_updated = |
2154 | 0 | MIN((S32)u4_num_l1_results_updated, |
2155 | 0 | ps_prms->i4_num_active_ref_l1 * ps_layer_mv->i4_num_mvs_per_ref); |
2156 | |
|
2157 | 0 | for(i4_j = 0; i4_j < (S32)u4_num_l0_results_updated; i4_j++) |
2158 | 0 | { |
2159 | 0 | COPY_SEARCH_RESULT( |
2160 | 0 | &ps_mv_l0_root[i4_j], |
2161 | 0 | &pi1_ref_idx_l0_root[i4_j], |
2162 | 0 | aps_result_nodes_sorted[0][i4_j], |
2163 | 0 | 0); |
2164 | 0 | } |
2165 | |
|
2166 | 0 | for(i4_j = 0; i4_j < (S32)u4_num_l1_results_updated; i4_j++) |
2167 | 0 | { |
2168 | 0 | COPY_SEARCH_RESULT( |
2169 | 0 | &ps_mv_l1_root[i4_j], |
2170 | 0 | &pi1_ref_idx_l1_root[i4_j], |
2171 | 0 | aps_result_nodes_sorted[1][i4_j], |
2172 | 0 | 0); |
2173 | 0 | } |
2174 | 0 | } |
2175 | 0 | } |
2176 | 0 | } |
2177 | | |
2178 | | /** |
2179 | | ****************************************************************************** |
2180 | | * @brief Scales motion vector component projecte from a diff layer in same |
2181 | | * picture (so no ref id related delta poc scaling required) |
2182 | | ****************************************************************************** |
2183 | | */ |
2184 | | |
2185 | | #define SCALE_MV_COMP_RES(mvcomp_p, dim_c, dim_p) \ |
2186 | 0 | ((((mvcomp_p) * (dim_c)) + ((SIGN((mvcomp_p)) * (dim_p)) >> 1)) / (dim_p)) |
2187 | | /** |
2188 | | ******************************************************************************** |
2189 | | * @fn hme_project_coloc_candt(search_node_t *ps_search_node, |
2190 | | * layer_ctxt_t *ps_curr_layer, |
2191 | | * layer_ctxt_t *ps_coarse_layer, |
2192 | | * S32 i4_pos_x, |
2193 | | * S32 i4_pos_y, |
2194 | | * S08 i1_ref_id, |
2195 | | * S08 i1_result_id) |
2196 | | * |
2197 | | * @brief From a coarser layer, projects a candidated situated at "colocated" |
2198 | | * position in the picture (e.g. given x, y it will be x/2, y/2 dyadic |
2199 | | * |
2200 | | * @param[out] ps_search_node : contains the projected result |
2201 | | * |
2202 | | * @param[in] ps_curr_layer : current layer context |
2203 | | * |
2204 | | * @param[in] ps_coarse_layer : coarser layer context |
2205 | | * |
2206 | | * @param[in] i4_pos_x : x Position where mv is required (w.r.t. curr layer) |
2207 | | * |
2208 | | * @param[in] i4_pos_y : y Position where mv is required (w.r.t. curr layer) |
2209 | | * |
2210 | | * @param[in] i1_ref_id : reference id for which the candidate required |
2211 | | * |
2212 | | * @param[in] i4_result_id : result id for which the candidate required |
2213 | | * (0 : best result, 1 : next best) |
2214 | | * |
2215 | | * @return None |
2216 | | ******************************************************************************** |
2217 | | */ |
2218 | | |
2219 | | void hme_project_coloc_candt( |
2220 | | search_node_t *ps_search_node, |
2221 | | layer_ctxt_t *ps_curr_layer, |
2222 | | layer_ctxt_t *ps_coarse_layer, |
2223 | | S32 i4_pos_x, |
2224 | | S32 i4_pos_y, |
2225 | | S08 i1_ref_id, |
2226 | | S32 i4_result_id) |
2227 | 0 | { |
2228 | 0 | S32 wd_c, ht_c, wd_p, ht_p; |
2229 | 0 | S32 blksize_p, blk_x, blk_y, i4_offset; |
2230 | 0 | layer_mv_t *ps_layer_mvbank; |
2231 | 0 | hme_mv_t *ps_mv; |
2232 | 0 | S08 *pi1_ref_idx; |
2233 | | |
2234 | | /* Width and ht of current and prev layers */ |
2235 | 0 | wd_c = ps_curr_layer->i4_wd; |
2236 | 0 | ht_c = ps_curr_layer->i4_ht; |
2237 | 0 | wd_p = ps_coarse_layer->i4_wd; |
2238 | 0 | ht_p = ps_coarse_layer->i4_ht; |
2239 | |
|
2240 | 0 | ps_layer_mvbank = ps_coarse_layer->ps_layer_mvbank; |
2241 | 0 | blksize_p = (S32)gau1_blk_size_to_wd[ps_layer_mvbank->e_blk_size]; |
2242 | | |
2243 | | /* Safety check to avoid uninitialized access across temporal layers */ |
2244 | 0 | i4_pos_x = CLIP3(i4_pos_x, 0, (wd_c - blksize_p)); |
2245 | 0 | i4_pos_y = CLIP3(i4_pos_y, 0, (ht_c - blksize_p)); |
2246 | | |
2247 | | /* Project the positions to prev layer */ |
2248 | | /* TODO: convert these to scale factors at pic level */ |
2249 | 0 | blk_x = (i4_pos_x * wd_p) / (wd_c * blksize_p); |
2250 | 0 | blk_y = (i4_pos_y * ht_p) / (ht_c * blksize_p); |
2251 | | |
2252 | | /* Pick up the mvs from the location */ |
2253 | 0 | i4_offset = (blk_x * ps_layer_mvbank->i4_num_mvs_per_blk); |
2254 | 0 | i4_offset += (ps_layer_mvbank->i4_num_mvs_per_row * blk_y); |
2255 | |
|
2256 | 0 | ps_mv = ps_layer_mvbank->ps_mv + i4_offset; |
2257 | 0 | pi1_ref_idx = ps_layer_mvbank->pi1_ref_idx + i4_offset; |
2258 | |
|
2259 | 0 | ps_mv += (i1_ref_id * ps_layer_mvbank->i4_num_mvs_per_ref); |
2260 | 0 | pi1_ref_idx += (i1_ref_id * ps_layer_mvbank->i4_num_mvs_per_ref); |
2261 | |
|
2262 | 0 | ps_search_node->s_mv.i2_mvx = SCALE_MV_COMP_RES(ps_mv[i4_result_id].i2_mv_x, wd_c, wd_p); |
2263 | 0 | ps_search_node->s_mv.i2_mvy = SCALE_MV_COMP_RES(ps_mv[i4_result_id].i2_mv_y, ht_c, ht_p); |
2264 | 0 | ps_search_node->i1_ref_idx = pi1_ref_idx[i4_result_id]; |
2265 | 0 | ps_search_node->u1_subpel_done = 0; |
2266 | 0 | if((ps_search_node->i1_ref_idx < 0) || (ps_search_node->s_mv.i2_mvx == INTRA_MV)) |
2267 | 0 | { |
2268 | 0 | ps_search_node->i1_ref_idx = i1_ref_id; |
2269 | 0 | ps_search_node->s_mv.i2_mvx = 0; |
2270 | 0 | ps_search_node->s_mv.i2_mvy = 0; |
2271 | 0 | } |
2272 | 0 | } |
2273 | | |
2274 | | /** |
2275 | | ******************************************************************************** |
2276 | | * @fn hme_project_coloc_candt_dyadic(search_node_t *ps_search_node, |
2277 | | * layer_ctxt_t *ps_curr_layer, |
2278 | | * layer_ctxt_t *ps_coarse_layer, |
2279 | | * S32 i4_pos_x, |
2280 | | * S32 i4_pos_y, |
2281 | | * S08 i1_ref_id, |
2282 | | * S08 i1_result_id) |
2283 | | * |
2284 | | * @brief From a coarser layer, projects a candidated situated at "colocated" |
2285 | | * position in the picture when the ratios are dyadic |
2286 | | * |
2287 | | * @param[out] ps_search_node : contains the projected result |
2288 | | * |
2289 | | * @param[in] ps_curr_layer : current layer context |
2290 | | * |
2291 | | * @param[in] ps_coarse_layer : coarser layer context |
2292 | | * |
2293 | | * @param[in] i4_pos_x : x Position where mv is required (w.r.t. curr layer) |
2294 | | * |
2295 | | * @param[in] i4_pos_y : y Position where mv is required (w.r.t. curr layer) |
2296 | | * |
2297 | | * @param[in] i1_ref_id : reference id for which the candidate required |
2298 | | * |
2299 | | * @param[in] i4_result_id : result id for which the candidate required |
2300 | | * (0 : best result, 1 : next best) |
2301 | | * |
2302 | | * @return None |
2303 | | ******************************************************************************** |
2304 | | */ |
2305 | | |
2306 | | void hme_project_coloc_candt_dyadic( |
2307 | | search_node_t *ps_search_node, |
2308 | | layer_ctxt_t *ps_curr_layer, |
2309 | | layer_ctxt_t *ps_coarse_layer, |
2310 | | S32 i4_pos_x, |
2311 | | S32 i4_pos_y, |
2312 | | S08 i1_ref_id, |
2313 | | S32 i4_result_id) |
2314 | 0 | { |
2315 | 0 | S32 wd_c, ht_c, wd_p, ht_p; |
2316 | 0 | S32 blksize_p, blk_x, blk_y, i4_offset; |
2317 | 0 | layer_mv_t *ps_layer_mvbank; |
2318 | 0 | hme_mv_t *ps_mv; |
2319 | 0 | S08 *pi1_ref_idx; |
2320 | | |
2321 | | /* Width and ht of current and prev layers */ |
2322 | 0 | wd_c = ps_curr_layer->i4_wd; |
2323 | 0 | ht_c = ps_curr_layer->i4_ht; |
2324 | 0 | wd_p = ps_coarse_layer->i4_wd; |
2325 | 0 | ht_p = ps_coarse_layer->i4_ht; |
2326 | |
|
2327 | 0 | ps_layer_mvbank = ps_coarse_layer->ps_layer_mvbank; |
2328 | | /* blksize_p = log2(wd) + 1 */ |
2329 | 0 | blksize_p = (S32)gau1_blk_size_to_wd_shift[ps_layer_mvbank->e_blk_size]; |
2330 | | |
2331 | | /* ASSERT for valid sizes */ |
2332 | 0 | ASSERT((blksize_p == 3) || (blksize_p == 4) || (blksize_p == 5)); |
2333 | | |
2334 | | /* Safety check to avoid uninitialized access across temporal layers */ |
2335 | 0 | i4_pos_x = CLIP3(i4_pos_x, 0, (wd_c - blksize_p)); |
2336 | 0 | i4_pos_y = CLIP3(i4_pos_y, 0, (ht_c - blksize_p)); |
2337 | | |
2338 | | /* Project the positions to prev layer */ |
2339 | | /* TODO: convert these to scale factors at pic level */ |
2340 | 0 | blk_x = i4_pos_x >> blksize_p; // (2 * blksize_p); |
2341 | 0 | blk_y = i4_pos_y >> blksize_p; // (2 * blksize_p); |
2342 | | |
2343 | | /* Pick up the mvs from the location */ |
2344 | 0 | i4_offset = (blk_x * ps_layer_mvbank->i4_num_mvs_per_blk); |
2345 | 0 | i4_offset += (ps_layer_mvbank->i4_num_mvs_per_row * blk_y); |
2346 | |
|
2347 | 0 | ps_mv = ps_layer_mvbank->ps_mv + i4_offset; |
2348 | 0 | pi1_ref_idx = ps_layer_mvbank->pi1_ref_idx + i4_offset; |
2349 | |
|
2350 | 0 | ps_mv += (i1_ref_id * ps_layer_mvbank->i4_num_mvs_per_ref); |
2351 | 0 | pi1_ref_idx += (i1_ref_id * ps_layer_mvbank->i4_num_mvs_per_ref); |
2352 | |
|
2353 | 0 | ps_search_node->s_mv.i2_mvx = ps_mv[i4_result_id].i2_mv_x << 1; |
2354 | 0 | ps_search_node->s_mv.i2_mvy = ps_mv[i4_result_id].i2_mv_y << 1; |
2355 | 0 | ps_search_node->i1_ref_idx = pi1_ref_idx[i4_result_id]; |
2356 | 0 | if((ps_search_node->i1_ref_idx < 0) || (ps_search_node->s_mv.i2_mvx == INTRA_MV)) |
2357 | 0 | { |
2358 | 0 | ps_search_node->i1_ref_idx = i1_ref_id; |
2359 | 0 | ps_search_node->s_mv.i2_mvx = 0; |
2360 | 0 | ps_search_node->s_mv.i2_mvy = 0; |
2361 | 0 | } |
2362 | 0 | } |
2363 | | |
2364 | | void hme_project_coloc_candt_dyadic_implicit( |
2365 | | search_node_t *ps_search_node, |
2366 | | layer_ctxt_t *ps_curr_layer, |
2367 | | layer_ctxt_t *ps_coarse_layer, |
2368 | | S32 i4_pos_x, |
2369 | | S32 i4_pos_y, |
2370 | | S32 i4_num_act_ref_l0, |
2371 | | U08 u1_pred_dir, |
2372 | | U08 u1_default_ref_id, |
2373 | | S32 i4_result_id) |
2374 | 0 | { |
2375 | 0 | S32 wd_c, ht_c, wd_p, ht_p; |
2376 | 0 | S32 blksize_p, blk_x, blk_y, i4_offset; |
2377 | 0 | layer_mv_t *ps_layer_mvbank; |
2378 | 0 | hme_mv_t *ps_mv; |
2379 | 0 | S08 *pi1_ref_idx; |
2380 | | |
2381 | | /* Width and ht of current and prev layers */ |
2382 | 0 | wd_c = ps_curr_layer->i4_wd; |
2383 | 0 | ht_c = ps_curr_layer->i4_ht; |
2384 | 0 | wd_p = ps_coarse_layer->i4_wd; |
2385 | 0 | ht_p = ps_coarse_layer->i4_ht; |
2386 | |
|
2387 | 0 | ps_layer_mvbank = ps_coarse_layer->ps_layer_mvbank; |
2388 | 0 | blksize_p = (S32)gau1_blk_size_to_wd_shift[ps_layer_mvbank->e_blk_size]; |
2389 | | |
2390 | | /* ASSERT for valid sizes */ |
2391 | 0 | ASSERT((blksize_p == 3) || (blksize_p == 4) || (blksize_p == 5)); |
2392 | | |
2393 | | /* Safety check to avoid uninitialized access across temporal layers */ |
2394 | 0 | i4_pos_x = CLIP3(i4_pos_x, 0, (wd_c - blksize_p)); |
2395 | 0 | i4_pos_y = CLIP3(i4_pos_y, 0, (ht_c - blksize_p)); |
2396 | | /* Project the positions to prev layer */ |
2397 | | /* TODO: convert these to scale factors at pic level */ |
2398 | 0 | blk_x = i4_pos_x >> blksize_p; // (2 * blksize_p); |
2399 | 0 | blk_y = i4_pos_y >> blksize_p; // (2 * blksize_p); |
2400 | | |
2401 | | /* Pick up the mvs from the location */ |
2402 | 0 | i4_offset = (blk_x * ps_layer_mvbank->i4_num_mvs_per_blk); |
2403 | 0 | i4_offset += (ps_layer_mvbank->i4_num_mvs_per_row * blk_y); |
2404 | |
|
2405 | 0 | ps_mv = ps_layer_mvbank->ps_mv + i4_offset; |
2406 | 0 | pi1_ref_idx = ps_layer_mvbank->pi1_ref_idx + i4_offset; |
2407 | |
|
2408 | 0 | if(u1_pred_dir == 1) |
2409 | 0 | { |
2410 | 0 | ps_mv += (i4_num_act_ref_l0 * ps_layer_mvbank->i4_num_mvs_per_ref); |
2411 | 0 | pi1_ref_idx += (i4_num_act_ref_l0 * ps_layer_mvbank->i4_num_mvs_per_ref); |
2412 | 0 | } |
2413 | |
|
2414 | 0 | ps_search_node->s_mv.i2_mvx = ps_mv[i4_result_id].i2_mv_x << 1; |
2415 | 0 | ps_search_node->s_mv.i2_mvy = ps_mv[i4_result_id].i2_mv_y << 1; |
2416 | 0 | ps_search_node->i1_ref_idx = pi1_ref_idx[i4_result_id]; |
2417 | 0 | if((ps_search_node->i1_ref_idx < 0) || (ps_search_node->s_mv.i2_mvx == INTRA_MV)) |
2418 | 0 | { |
2419 | 0 | ps_search_node->i1_ref_idx = u1_default_ref_id; |
2420 | 0 | ps_search_node->s_mv.i2_mvx = 0; |
2421 | 0 | ps_search_node->s_mv.i2_mvy = 0; |
2422 | 0 | } |
2423 | 0 | } |
2424 | | |
2425 | | #define SCALE_RANGE_PRMS(prm1, prm2, shift) \ |
2426 | 0 | { \ |
2427 | 0 | prm1.i2_min_x = prm2.i2_min_x << shift; \ |
2428 | 0 | prm1.i2_max_x = prm2.i2_max_x << shift; \ |
2429 | 0 | prm1.i2_min_y = prm2.i2_min_y << shift; \ |
2430 | 0 | prm1.i2_max_y = prm2.i2_max_y << shift; \ |
2431 | 0 | } |
2432 | | |
2433 | | #define SCALE_RANGE_PRMS_POINTERS(prm1, prm2, shift) \ |
2434 | 0 | { \ |
2435 | 0 | prm1->i2_min_x = prm2->i2_min_x << shift; \ |
2436 | 0 | prm1->i2_max_x = prm2->i2_max_x << shift; \ |
2437 | 0 | prm1->i2_min_y = prm2->i2_min_y << shift; \ |
2438 | 0 | prm1->i2_max_y = prm2->i2_max_y << shift; \ |
2439 | 0 | } |
2440 | | |
2441 | | /** |
2442 | | ******************************************************************************** |
2443 | | * @fn void hme_refine_frm_init(me_ctxt_t *ps_ctxt, |
2444 | | * refine_layer_prms_t *ps_refine_prms) |
2445 | | * |
2446 | | * @brief Frame init of refinemnet layers in ME |
2447 | | * |
2448 | | * @param[in,out] ps_ctxt: ME Handle |
2449 | | * |
2450 | | * @param[in] ps_refine_prms : refinement layer prms |
2451 | | * |
2452 | | * @return None |
2453 | | ******************************************************************************** |
2454 | | */ |
2455 | | void hme_refine_frm_init( |
2456 | | layer_ctxt_t *ps_curr_layer, refine_prms_t *ps_refine_prms, layer_ctxt_t *ps_coarse_layer) |
2457 | 0 | { |
2458 | | /* local variables */ |
2459 | 0 | BLK_SIZE_T e_result_blk_size = BLK_8x8; |
2460 | 0 | S32 i4_num_ref_fpel, i4_num_ref_prev_layer; |
2461 | |
|
2462 | 0 | i4_num_ref_prev_layer = ps_coarse_layer->ps_layer_mvbank->i4_num_ref; |
2463 | |
|
2464 | 0 | if(ps_refine_prms->explicit_ref) |
2465 | 0 | { |
2466 | 0 | i4_num_ref_fpel = i4_num_ref_prev_layer; |
2467 | 0 | } |
2468 | 0 | else |
2469 | 0 | { |
2470 | 0 | i4_num_ref_fpel = 2; |
2471 | 0 | } |
2472 | |
|
2473 | 0 | if(ps_refine_prms->i4_enable_4x4_part) |
2474 | 0 | { |
2475 | 0 | e_result_blk_size = BLK_4x4; |
2476 | 0 | } |
2477 | |
|
2478 | 0 | i4_num_ref_fpel = MIN(i4_num_ref_fpel, i4_num_ref_prev_layer); |
2479 | |
|
2480 | 0 | hme_init_mv_bank( |
2481 | 0 | ps_curr_layer, |
2482 | 0 | e_result_blk_size, |
2483 | 0 | i4_num_ref_fpel, |
2484 | 0 | ps_refine_prms->i4_num_mvbank_results, |
2485 | 0 | ps_refine_prms->i4_layer_id > 0 ? 0 : 1); |
2486 | 0 | } |
2487 | | |
2488 | | #if 1 //ENABLE_CU_RECURSION || TEST_AND_EVALUATE_CU_RECURSION |
2489 | | /** |
2490 | | ******************************************************************************** |
2491 | | * @fn void hme_init_clusters_16x16 |
2492 | | * ( |
2493 | | * cluster_16x16_blk_t *ps_cluster_blk_16x16 |
2494 | | * ) |
2495 | | * |
2496 | | * @brief Intialisations for the structs used in clustering algorithm |
2497 | | * |
2498 | | * @param[in/out] ps_cluster_blk_16x16: pointer to structure containing clusters |
2499 | | * of 16x16 block |
2500 | | * |
2501 | | * @return None |
2502 | | ******************************************************************************** |
2503 | | */ |
2504 | | static __inline void |
2505 | | hme_init_clusters_16x16(cluster_16x16_blk_t *ps_cluster_blk_16x16, S32 bidir_enabled) |
2506 | 0 | { |
2507 | 0 | S32 i; |
2508 | |
|
2509 | 0 | ps_cluster_blk_16x16->num_clusters = 0; |
2510 | 0 | ps_cluster_blk_16x16->intra_mv_area = 0; |
2511 | 0 | ps_cluster_blk_16x16->best_inter_cost = 0; |
2512 | |
|
2513 | 0 | for(i = 0; i < MAX_NUM_CLUSTERS_16x16; i++) |
2514 | 0 | { |
2515 | 0 | ps_cluster_blk_16x16->as_cluster_data[i].max_dist_from_centroid = |
2516 | 0 | bidir_enabled ? MAX_DISTANCE_FROM_CENTROID_16x16_B : MAX_DISTANCE_FROM_CENTROID_16x16; |
2517 | |
|
2518 | 0 | ps_cluster_blk_16x16->as_cluster_data[i].is_valid_cluster = 0; |
2519 | |
|
2520 | 0 | ps_cluster_blk_16x16->as_cluster_data[i].bi_mv_pixel_area = 0; |
2521 | 0 | ps_cluster_blk_16x16->as_cluster_data[i].uni_mv_pixel_area = 0; |
2522 | 0 | } |
2523 | 0 | for(i = 0; i < MAX_NUM_REF; i++) |
2524 | 0 | { |
2525 | 0 | ps_cluster_blk_16x16->au1_num_clusters[i] = 0; |
2526 | 0 | } |
2527 | 0 | } |
2528 | | |
2529 | | /** |
2530 | | ******************************************************************************** |
2531 | | * @fn void hme_init_clusters_32x32 |
2532 | | * ( |
2533 | | * cluster_32x32_blk_t *ps_cluster_blk_32x32 |
2534 | | * ) |
2535 | | * |
2536 | | * @brief Intialisations for the structs used in clustering algorithm |
2537 | | * |
2538 | | * @param[in/out] ps_cluster_blk_32x32: pointer to structure containing clusters |
2539 | | * of 32x32 block |
2540 | | * |
2541 | | * @return None |
2542 | | ******************************************************************************** |
2543 | | */ |
2544 | | static __inline void |
2545 | | hme_init_clusters_32x32(cluster_32x32_blk_t *ps_cluster_blk_32x32, S32 bidir_enabled) |
2546 | 0 | { |
2547 | 0 | S32 i; |
2548 | |
|
2549 | 0 | ps_cluster_blk_32x32->num_clusters = 0; |
2550 | 0 | ps_cluster_blk_32x32->intra_mv_area = 0; |
2551 | 0 | ps_cluster_blk_32x32->best_alt_ref = -1; |
2552 | 0 | ps_cluster_blk_32x32->best_uni_ref = -1; |
2553 | 0 | ps_cluster_blk_32x32->best_inter_cost = 0; |
2554 | 0 | ps_cluster_blk_32x32->num_clusters_with_weak_sdi_density = 0; |
2555 | |
|
2556 | 0 | for(i = 0; i < MAX_NUM_CLUSTERS_32x32; i++) |
2557 | 0 | { |
2558 | 0 | ps_cluster_blk_32x32->as_cluster_data[i].max_dist_from_centroid = |
2559 | 0 | bidir_enabled ? MAX_DISTANCE_FROM_CENTROID_32x32_B : MAX_DISTANCE_FROM_CENTROID_32x32; |
2560 | 0 | ps_cluster_blk_32x32->as_cluster_data[i].is_valid_cluster = 0; |
2561 | |
|
2562 | 0 | ps_cluster_blk_32x32->as_cluster_data[i].bi_mv_pixel_area = 0; |
2563 | 0 | ps_cluster_blk_32x32->as_cluster_data[i].uni_mv_pixel_area = 0; |
2564 | 0 | } |
2565 | 0 | for(i = 0; i < MAX_NUM_REF; i++) |
2566 | 0 | { |
2567 | 0 | ps_cluster_blk_32x32->au1_num_clusters[i] = 0; |
2568 | 0 | } |
2569 | 0 | } |
2570 | | |
2571 | | /** |
2572 | | ******************************************************************************** |
2573 | | * @fn void hme_init_clusters_64x64 |
2574 | | * ( |
2575 | | * cluster_64x64_blk_t *ps_cluster_blk_64x64 |
2576 | | * ) |
2577 | | * |
2578 | | * @brief Intialisations for the structs used in clustering algorithm |
2579 | | * |
2580 | | * @param[in/out] ps_cluster_blk_64x64: pointer to structure containing clusters |
2581 | | * of 64x64 block |
2582 | | * |
2583 | | * @return None |
2584 | | ******************************************************************************** |
2585 | | */ |
2586 | | static __inline void |
2587 | | hme_init_clusters_64x64(cluster_64x64_blk_t *ps_cluster_blk_64x64, S32 bidir_enabled) |
2588 | 0 | { |
2589 | 0 | S32 i; |
2590 | |
|
2591 | 0 | ps_cluster_blk_64x64->num_clusters = 0; |
2592 | 0 | ps_cluster_blk_64x64->intra_mv_area = 0; |
2593 | 0 | ps_cluster_blk_64x64->best_alt_ref = -1; |
2594 | 0 | ps_cluster_blk_64x64->best_uni_ref = -1; |
2595 | 0 | ps_cluster_blk_64x64->best_inter_cost = 0; |
2596 | |
|
2597 | 0 | for(i = 0; i < MAX_NUM_CLUSTERS_64x64; i++) |
2598 | 0 | { |
2599 | 0 | ps_cluster_blk_64x64->as_cluster_data[i].max_dist_from_centroid = |
2600 | 0 | bidir_enabled ? MAX_DISTANCE_FROM_CENTROID_64x64_B : MAX_DISTANCE_FROM_CENTROID_64x64; |
2601 | 0 | ps_cluster_blk_64x64->as_cluster_data[i].is_valid_cluster = 0; |
2602 | |
|
2603 | 0 | ps_cluster_blk_64x64->as_cluster_data[i].bi_mv_pixel_area = 0; |
2604 | 0 | ps_cluster_blk_64x64->as_cluster_data[i].uni_mv_pixel_area = 0; |
2605 | 0 | } |
2606 | 0 | for(i = 0; i < MAX_NUM_REF; i++) |
2607 | 0 | { |
2608 | 0 | ps_cluster_blk_64x64->au1_num_clusters[i] = 0; |
2609 | 0 | } |
2610 | 0 | } |
2611 | | |
2612 | | /** |
2613 | | ******************************************************************************** |
2614 | | * @fn void hme_sort_and_assign_top_ref_ids_areawise |
2615 | | * ( |
2616 | | * ctb_cluster_info_t *ps_ctb_cluster_info |
2617 | | * ) |
2618 | | * |
2619 | | * @brief Finds best_uni_ref and best_alt_ref |
2620 | | * |
2621 | | * @param[in/out] ps_ctb_cluster_info: structure that points to ctb data |
2622 | | * |
2623 | | * @param[in] bidir_enabled: flag that indicates whether or not bi-pred is |
2624 | | * enabled |
2625 | | * |
2626 | | * @param[in] block_width: width of the block in pels |
2627 | | * |
2628 | | * @param[in] e_cu_pos: position of the block within the CTB |
2629 | | * |
2630 | | * @return None |
2631 | | ******************************************************************************** |
2632 | | */ |
2633 | | void hme_sort_and_assign_top_ref_ids_areawise( |
2634 | | ctb_cluster_info_t *ps_ctb_cluster_info, S32 bidir_enabled, S32 block_width, CU_POS_T e_cu_pos) |
2635 | 0 | { |
2636 | 0 | cluster_32x32_blk_t *ps_32x32 = NULL; |
2637 | 0 | cluster_64x64_blk_t *ps_64x64 = NULL; |
2638 | 0 | cluster_data_t *ps_data; |
2639 | |
|
2640 | 0 | S32 j, k; |
2641 | |
|
2642 | 0 | S32 ai4_uni_area[MAX_NUM_REF]; |
2643 | 0 | S32 ai4_bi_area[MAX_NUM_REF]; |
2644 | 0 | S32 ai4_ref_id_found[MAX_NUM_REF]; |
2645 | 0 | S32 ai4_ref_id[MAX_NUM_REF]; |
2646 | |
|
2647 | 0 | S32 best_uni_ref = -1, best_alt_ref = -1; |
2648 | 0 | S32 num_clusters; |
2649 | 0 | S32 num_ref = 0; |
2650 | 0 | S32 num_clusters_evaluated = 0; |
2651 | 0 | S32 is_cur_blk_valid; |
2652 | |
|
2653 | 0 | if(32 == block_width) |
2654 | 0 | { |
2655 | 0 | is_cur_blk_valid = (ps_ctb_cluster_info->blk_32x32_mask & (1 << e_cu_pos)) || 0; |
2656 | 0 | ps_32x32 = &ps_ctb_cluster_info->ps_32x32_blk[e_cu_pos]; |
2657 | 0 | num_clusters = ps_32x32->num_clusters; |
2658 | 0 | ps_data = &ps_32x32->as_cluster_data[0]; |
2659 | 0 | } |
2660 | 0 | else |
2661 | 0 | { |
2662 | 0 | is_cur_blk_valid = (ps_ctb_cluster_info->blk_32x32_mask == 0xf); |
2663 | 0 | ps_64x64 = ps_ctb_cluster_info->ps_64x64_blk; |
2664 | 0 | num_clusters = ps_64x64->num_clusters; |
2665 | 0 | ps_data = &ps_64x64->as_cluster_data[0]; |
2666 | 0 | } |
2667 | |
|
2668 | | #if !ENABLE_4CTB_EVALUATION |
2669 | | if((num_clusters > MAX_NUM_CLUSTERS_IN_VALID_32x32_BLK)) |
2670 | | { |
2671 | | return; |
2672 | | } |
2673 | | #endif |
2674 | 0 | if(num_clusters == 0) |
2675 | 0 | { |
2676 | 0 | return; |
2677 | 0 | } |
2678 | 0 | else if(!is_cur_blk_valid) |
2679 | 0 | { |
2680 | 0 | return; |
2681 | 0 | } |
2682 | | |
2683 | 0 | memset(ai4_uni_area, 0, sizeof(S32) * MAX_NUM_REF); |
2684 | 0 | memset(ai4_bi_area, 0, sizeof(S32) * MAX_NUM_REF); |
2685 | 0 | memset(ai4_ref_id_found, 0, sizeof(S32) * MAX_NUM_REF); |
2686 | 0 | memset(ai4_ref_id, -1, sizeof(S32) * MAX_NUM_REF); |
2687 | |
|
2688 | 0 | for(j = 0; num_clusters_evaluated < num_clusters; j++, ps_data++) |
2689 | 0 | { |
2690 | 0 | S32 ref_id; |
2691 | |
|
2692 | 0 | if(!ps_data->is_valid_cluster) |
2693 | 0 | { |
2694 | 0 | continue; |
2695 | 0 | } |
2696 | | |
2697 | 0 | ref_id = ps_data->ref_id; |
2698 | |
|
2699 | 0 | num_clusters_evaluated++; |
2700 | |
|
2701 | 0 | ai4_uni_area[ref_id] += ps_data->uni_mv_pixel_area; |
2702 | 0 | ai4_bi_area[ref_id] += ps_data->bi_mv_pixel_area; |
2703 | |
|
2704 | 0 | if(!ai4_ref_id_found[ref_id]) |
2705 | 0 | { |
2706 | 0 | ai4_ref_id[ref_id] = ref_id; |
2707 | 0 | ai4_ref_id_found[ref_id] = 1; |
2708 | 0 | num_ref++; |
2709 | 0 | } |
2710 | 0 | } |
2711 | |
|
2712 | 0 | { |
2713 | 0 | S32 ai4_ref_id_temp[MAX_NUM_REF]; |
2714 | |
|
2715 | 0 | memcpy(ai4_ref_id_temp, ai4_ref_id, sizeof(S32) * MAX_NUM_REF); |
2716 | |
|
2717 | 0 | for(k = 1; k < MAX_NUM_REF; k++) |
2718 | 0 | { |
2719 | 0 | if(ai4_uni_area[k] > ai4_uni_area[0]) |
2720 | 0 | { |
2721 | 0 | SWAP_HME(ai4_uni_area[k], ai4_uni_area[0], S32); |
2722 | 0 | SWAP_HME(ai4_ref_id_temp[k], ai4_ref_id_temp[0], S32); |
2723 | 0 | } |
2724 | 0 | } |
2725 | |
|
2726 | 0 | best_uni_ref = ai4_ref_id_temp[0]; |
2727 | 0 | } |
2728 | |
|
2729 | 0 | if(bidir_enabled) |
2730 | 0 | { |
2731 | 0 | for(k = 1; k < MAX_NUM_REF; k++) |
2732 | 0 | { |
2733 | 0 | if(ai4_bi_area[k] > ai4_bi_area[0]) |
2734 | 0 | { |
2735 | 0 | SWAP_HME(ai4_bi_area[k], ai4_bi_area[0], S32); |
2736 | 0 | SWAP_HME(ai4_ref_id[k], ai4_ref_id[0], S32); |
2737 | 0 | } |
2738 | 0 | } |
2739 | |
|
2740 | 0 | if(!ai4_bi_area[0]) |
2741 | 0 | { |
2742 | 0 | best_alt_ref = -1; |
2743 | |
|
2744 | 0 | if(32 == block_width) |
2745 | 0 | { |
2746 | 0 | SET_VALUES_FOR_TOP_REF_IDS(ps_32x32, best_uni_ref, best_alt_ref, num_ref); |
2747 | 0 | } |
2748 | 0 | else |
2749 | 0 | { |
2750 | 0 | SET_VALUES_FOR_TOP_REF_IDS(ps_64x64, best_uni_ref, best_alt_ref, num_ref); |
2751 | 0 | } |
2752 | |
|
2753 | 0 | return; |
2754 | 0 | } |
2755 | | |
2756 | 0 | if(best_uni_ref == ai4_ref_id[0]) |
2757 | 0 | { |
2758 | 0 | for(k = 2; k < MAX_NUM_REF; k++) |
2759 | 0 | { |
2760 | 0 | if(ai4_bi_area[k] > ai4_bi_area[1]) |
2761 | 0 | { |
2762 | 0 | SWAP_HME(ai4_bi_area[k], ai4_bi_area[1], S32); |
2763 | 0 | SWAP_HME(ai4_ref_id[k], ai4_ref_id[1], S32); |
2764 | 0 | } |
2765 | 0 | } |
2766 | |
|
2767 | 0 | best_alt_ref = ai4_ref_id[1]; |
2768 | 0 | } |
2769 | 0 | else |
2770 | 0 | { |
2771 | 0 | best_alt_ref = ai4_ref_id[0]; |
2772 | 0 | } |
2773 | 0 | } |
2774 | | |
2775 | 0 | if(32 == block_width) |
2776 | 0 | { |
2777 | 0 | SET_VALUES_FOR_TOP_REF_IDS(ps_32x32, best_uni_ref, best_alt_ref, num_ref); |
2778 | 0 | } |
2779 | 0 | else |
2780 | 0 | { |
2781 | 0 | SET_VALUES_FOR_TOP_REF_IDS(ps_64x64, best_uni_ref, best_alt_ref, num_ref); |
2782 | 0 | } |
2783 | 0 | } |
2784 | | |
2785 | | /** |
2786 | | ******************************************************************************** |
2787 | | * @fn void hme_find_top_ref_ids |
2788 | | * ( |
2789 | | * ctb_cluster_info_t *ps_ctb_cluster_info |
2790 | | * ) |
2791 | | * |
2792 | | * @brief Finds best_uni_ref and best_alt_ref |
2793 | | * |
2794 | | * @param[in/out] ps_ctb_cluster_info: structure that points to ctb data |
2795 | | * |
2796 | | * @return None |
2797 | | ******************************************************************************** |
2798 | | */ |
2799 | | void hme_find_top_ref_ids( |
2800 | | ctb_cluster_info_t *ps_ctb_cluster_info, S32 bidir_enabled, S32 block_width) |
2801 | 0 | { |
2802 | 0 | S32 i; |
2803 | |
|
2804 | 0 | if(32 == block_width) |
2805 | 0 | { |
2806 | 0 | for(i = 0; i < 4; i++) |
2807 | 0 | { |
2808 | 0 | hme_sort_and_assign_top_ref_ids_areawise( |
2809 | 0 | ps_ctb_cluster_info, bidir_enabled, block_width, (CU_POS_T)i); |
2810 | 0 | } |
2811 | 0 | } |
2812 | 0 | else if(64 == block_width) |
2813 | 0 | { |
2814 | 0 | hme_sort_and_assign_top_ref_ids_areawise( |
2815 | 0 | ps_ctb_cluster_info, bidir_enabled, block_width, POS_NA); |
2816 | 0 | } |
2817 | 0 | } |
2818 | | |
2819 | | /** |
2820 | | ******************************************************************************** |
2821 | | * @fn void hme_boot_out_outlier |
2822 | | * ( |
2823 | | * ctb_cluster_info_t *ps_ctb_cluster_info |
2824 | | * ) |
2825 | | * |
2826 | | * @brief Removes outlier clusters before CU tree population |
2827 | | * |
2828 | | * @param[in/out] ps_ctb_cluster_info: structure that points to ctb data |
2829 | | * |
2830 | | * @return None |
2831 | | ******************************************************************************** |
2832 | | */ |
2833 | | void hme_boot_out_outlier(ctb_cluster_info_t *ps_ctb_cluster_info, S32 blk_width) |
2834 | 0 | { |
2835 | 0 | cluster_32x32_blk_t *ps_32x32; |
2836 | |
|
2837 | 0 | S32 i; |
2838 | |
|
2839 | 0 | cluster_64x64_blk_t *ps_64x64 = &ps_ctb_cluster_info->ps_64x64_blk[0]; |
2840 | |
|
2841 | 0 | S32 sdi_threshold = ps_ctb_cluster_info->sdi_threshold; |
2842 | |
|
2843 | 0 | if(32 == blk_width) |
2844 | 0 | { |
2845 | | /* 32x32 clusters */ |
2846 | 0 | for(i = 0; i < 4; i++) |
2847 | 0 | { |
2848 | 0 | ps_32x32 = &ps_ctb_cluster_info->ps_32x32_blk[i]; |
2849 | |
|
2850 | 0 | if(ps_32x32->num_clusters > MAX_NUM_CLUSTERS_IN_ONE_REF_IDX) |
2851 | 0 | { |
2852 | 0 | BUMP_OUTLIER_CLUSTERS(ps_32x32, sdi_threshold); |
2853 | 0 | } |
2854 | 0 | } |
2855 | 0 | } |
2856 | 0 | else if(64 == blk_width) |
2857 | 0 | { |
2858 | | /* 64x64 clusters */ |
2859 | 0 | if(ps_64x64->num_clusters > MAX_NUM_CLUSTERS_IN_ONE_REF_IDX) |
2860 | 0 | { |
2861 | 0 | BUMP_OUTLIER_CLUSTERS(ps_64x64, sdi_threshold); |
2862 | 0 | } |
2863 | 0 | } |
2864 | 0 | } |
2865 | | |
2866 | | /** |
2867 | | ******************************************************************************** |
2868 | | * @fn void hme_update_cluster_attributes |
2869 | | * ( |
2870 | | * cluster_data_t *ps_cluster_data, |
2871 | | * S32 mvx, |
2872 | | * S32 mvy, |
2873 | | * PART_ID_T e_part_id |
2874 | | * ) |
2875 | | * |
2876 | | * @brief Implementation fo the clustering algorithm |
2877 | | * |
2878 | | * @param[in/out] ps_cluster_data: pointer to cluster_data_t struct |
2879 | | * |
2880 | | * @param[in] mvx : x co-ordinate of the motion vector |
2881 | | * |
2882 | | * @param[in] mvy : y co-ordinate of the motion vector |
2883 | | * |
2884 | | * @param[in] ref_idx : ref_id of the motion vector |
2885 | | * |
2886 | | * @param[in] e_part_id : partition id of the motion vector |
2887 | | * |
2888 | | * @return None |
2889 | | ******************************************************************************** |
2890 | | */ |
2891 | | static __inline void hme_update_cluster_attributes( |
2892 | | cluster_data_t *ps_cluster_data, |
2893 | | S32 mvx, |
2894 | | S32 mvy, |
2895 | | S32 mvdx, |
2896 | | S32 mvdy, |
2897 | | S32 ref_id, |
2898 | | S32 sdi, |
2899 | | U08 is_part_of_bi, |
2900 | | PART_ID_T e_part_id) |
2901 | 0 | { |
2902 | 0 | LWORD64 i8_mvx_sum_q8; |
2903 | 0 | LWORD64 i8_mvy_sum_q8; |
2904 | |
|
2905 | 0 | S32 centroid_posx_q8 = ps_cluster_data->s_centroid.i4_pos_x_q8; |
2906 | 0 | S32 centroid_posy_q8 = ps_cluster_data->s_centroid.i4_pos_y_q8; |
2907 | |
|
2908 | 0 | if((mvdx > 0) && (ps_cluster_data->min_x > mvx)) |
2909 | 0 | { |
2910 | 0 | ps_cluster_data->min_x = mvx; |
2911 | 0 | } |
2912 | 0 | else if((mvdx < 0) && (ps_cluster_data->max_x < mvx)) |
2913 | 0 | { |
2914 | 0 | ps_cluster_data->max_x = mvx; |
2915 | 0 | } |
2916 | |
|
2917 | 0 | if((mvdy > 0) && (ps_cluster_data->min_y > mvy)) |
2918 | 0 | { |
2919 | 0 | ps_cluster_data->min_y = mvy; |
2920 | 0 | } |
2921 | 0 | else if((mvdy < 0) && (ps_cluster_data->max_y < mvy)) |
2922 | 0 | { |
2923 | 0 | ps_cluster_data->max_y = mvy; |
2924 | 0 | } |
2925 | |
|
2926 | 0 | { |
2927 | 0 | S32 num_mvs = ps_cluster_data->num_mvs; |
2928 | |
|
2929 | 0 | ps_cluster_data->as_mv[num_mvs].pixel_count = gai4_partition_area[e_part_id]; |
2930 | 0 | ps_cluster_data->as_mv[num_mvs].mvx = mvx; |
2931 | 0 | ps_cluster_data->as_mv[num_mvs].mvy = mvy; |
2932 | | |
2933 | | /***************************/ |
2934 | 0 | ps_cluster_data->as_mv[num_mvs].is_uni = !is_part_of_bi; |
2935 | 0 | ps_cluster_data->as_mv[num_mvs].sdi = sdi; |
2936 | | /**************************/ |
2937 | 0 | } |
2938 | | |
2939 | | /* Updation of centroid */ |
2940 | 0 | { |
2941 | 0 | i8_mvx_sum_q8 = (LWORD64)centroid_posx_q8 * ps_cluster_data->num_mvs + (mvx << 8); |
2942 | 0 | i8_mvy_sum_q8 = (LWORD64)centroid_posy_q8 * ps_cluster_data->num_mvs + (mvy << 8); |
2943 | |
|
2944 | 0 | ps_cluster_data->num_mvs++; |
2945 | |
|
2946 | 0 | ps_cluster_data->s_centroid.i4_pos_x_q8 = |
2947 | 0 | (WORD32)((i8_mvx_sum_q8) / ps_cluster_data->num_mvs); |
2948 | 0 | ps_cluster_data->s_centroid.i4_pos_y_q8 = |
2949 | 0 | (WORD32)((i8_mvy_sum_q8) / ps_cluster_data->num_mvs); |
2950 | 0 | } |
2951 | |
|
2952 | 0 | ps_cluster_data->area_in_pixels += gai4_partition_area[e_part_id]; |
2953 | |
|
2954 | 0 | if(is_part_of_bi) |
2955 | 0 | { |
2956 | 0 | ps_cluster_data->bi_mv_pixel_area += gai4_partition_area[e_part_id]; |
2957 | 0 | } |
2958 | 0 | else |
2959 | 0 | { |
2960 | 0 | ps_cluster_data->uni_mv_pixel_area += gai4_partition_area[e_part_id]; |
2961 | 0 | } |
2962 | 0 | } |
2963 | | |
2964 | | /** |
2965 | | ******************************************************************************** |
2966 | | * @fn void hme_try_cluster_merge |
2967 | | * ( |
2968 | | * cluster_data_t *ps_cluster_data, |
2969 | | * S32 *pi4_num_clusters, |
2970 | | * S32 idx_of_updated_cluster |
2971 | | * ) |
2972 | | * |
2973 | | * @brief Implementation fo the clustering algorithm |
2974 | | * |
2975 | | * @param[in/out] ps_cluster_data: pointer to cluster_data_t struct |
2976 | | * |
2977 | | * @param[in/out] pi4_num_clusters : pointer to number of clusters |
2978 | | * |
2979 | | * @param[in] idx_of_updated_cluster : index of the cluster most recently |
2980 | | * updated |
2981 | | * |
2982 | | * @return Nothing |
2983 | | ******************************************************************************** |
2984 | | */ |
2985 | | void hme_try_cluster_merge( |
2986 | | cluster_data_t *ps_cluster_data, U08 *pu1_num_clusters, S32 idx_of_updated_cluster) |
2987 | 0 | { |
2988 | 0 | centroid_t *ps_centroid; |
2989 | |
|
2990 | 0 | S32 cur_pos_x_q8; |
2991 | 0 | S32 cur_pos_y_q8; |
2992 | 0 | S32 i; |
2993 | 0 | S32 max_dist_from_centroid; |
2994 | 0 | S32 mvd; |
2995 | 0 | S32 mvdx_q8; |
2996 | 0 | S32 mvdx; |
2997 | 0 | S32 mvdy_q8; |
2998 | 0 | S32 mvdy; |
2999 | 0 | S32 num_clusters, num_clusters_evaluated; |
3000 | 0 | S32 other_pos_x_q8; |
3001 | 0 | S32 other_pos_y_q8; |
3002 | |
|
3003 | 0 | cluster_data_t *ps_root = ps_cluster_data; |
3004 | 0 | cluster_data_t *ps_cur_cluster = &ps_cluster_data[idx_of_updated_cluster]; |
3005 | 0 | centroid_t *ps_cur_centroid = &ps_cur_cluster->s_centroid; |
3006 | | |
3007 | | /* Merge is superfluous if num_clusters is 1 */ |
3008 | 0 | if(*pu1_num_clusters == 1) |
3009 | 0 | { |
3010 | 0 | return; |
3011 | 0 | } |
3012 | | |
3013 | 0 | cur_pos_x_q8 = ps_cur_centroid->i4_pos_x_q8; |
3014 | 0 | cur_pos_y_q8 = ps_cur_centroid->i4_pos_y_q8; |
3015 | |
|
3016 | 0 | max_dist_from_centroid = ps_cur_cluster->max_dist_from_centroid; |
3017 | |
|
3018 | 0 | num_clusters = *pu1_num_clusters; |
3019 | 0 | num_clusters_evaluated = 0; |
3020 | |
|
3021 | 0 | for(i = 0; num_clusters_evaluated < num_clusters; i++, ps_cluster_data++) |
3022 | 0 | { |
3023 | 0 | if(!ps_cluster_data->is_valid_cluster) |
3024 | 0 | { |
3025 | 0 | continue; |
3026 | 0 | } |
3027 | 0 | if((ps_cluster_data->ref_id != ps_cur_cluster->ref_id) || (i == idx_of_updated_cluster)) |
3028 | 0 | { |
3029 | 0 | num_clusters_evaluated++; |
3030 | 0 | continue; |
3031 | 0 | } |
3032 | | |
3033 | 0 | ps_centroid = &ps_cluster_data->s_centroid; |
3034 | |
|
3035 | 0 | other_pos_x_q8 = ps_centroid->i4_pos_x_q8; |
3036 | 0 | other_pos_y_q8 = ps_centroid->i4_pos_y_q8; |
3037 | |
|
3038 | 0 | mvdx_q8 = (cur_pos_x_q8 - other_pos_x_q8); |
3039 | 0 | mvdy_q8 = (cur_pos_y_q8 - other_pos_y_q8); |
3040 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3041 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3042 | |
|
3043 | 0 | mvd = ABS(mvdx) + ABS(mvdy); |
3044 | |
|
3045 | 0 | if(mvd <= (max_dist_from_centroid >> 1)) |
3046 | 0 | { |
3047 | | /* 0 => no updates */ |
3048 | | /* 1 => min updated */ |
3049 | | /* 2 => max updated */ |
3050 | 0 | S32 minmax_x_update_id; |
3051 | 0 | S32 minmax_y_update_id; |
3052 | |
|
3053 | 0 | LWORD64 i8_mv_x_sum_self = (LWORD64)cur_pos_x_q8 * ps_cur_cluster->num_mvs; |
3054 | 0 | LWORD64 i8_mv_y_sum_self = (LWORD64)cur_pos_y_q8 * ps_cur_cluster->num_mvs; |
3055 | 0 | LWORD64 i8_mv_x_sum_cousin = (LWORD64)other_pos_x_q8 * ps_cluster_data->num_mvs; |
3056 | 0 | LWORD64 i8_mv_y_sum_cousin = (LWORD64)other_pos_y_q8 * ps_cluster_data->num_mvs; |
3057 | |
|
3058 | 0 | (*pu1_num_clusters)--; |
3059 | |
|
3060 | 0 | ps_cluster_data->is_valid_cluster = 0; |
3061 | |
|
3062 | 0 | memcpy( |
3063 | 0 | &ps_cur_cluster->as_mv[ps_cur_cluster->num_mvs], |
3064 | 0 | ps_cluster_data->as_mv, |
3065 | 0 | sizeof(mv_data_t) * ps_cluster_data->num_mvs); |
3066 | |
|
3067 | 0 | ps_cur_cluster->num_mvs += ps_cluster_data->num_mvs; |
3068 | 0 | ps_cur_cluster->area_in_pixels += ps_cluster_data->area_in_pixels; |
3069 | 0 | ps_cur_cluster->bi_mv_pixel_area += ps_cluster_data->bi_mv_pixel_area; |
3070 | 0 | ps_cur_cluster->uni_mv_pixel_area += ps_cluster_data->uni_mv_pixel_area; |
3071 | 0 | i8_mv_x_sum_self += i8_mv_x_sum_cousin; |
3072 | 0 | i8_mv_y_sum_self += i8_mv_y_sum_cousin; |
3073 | |
|
3074 | 0 | ps_cur_centroid->i4_pos_x_q8 = (WORD32)(i8_mv_x_sum_self / ps_cur_cluster->num_mvs); |
3075 | 0 | ps_cur_centroid->i4_pos_y_q8 = (WORD32)(i8_mv_y_sum_self / ps_cur_cluster->num_mvs); |
3076 | |
|
3077 | 0 | minmax_x_update_id = (ps_cur_cluster->min_x < ps_cluster_data->min_x) |
3078 | 0 | ? ((ps_cur_cluster->max_x > ps_cluster_data->max_x) ? 0 : 2) |
3079 | 0 | : 1; |
3080 | 0 | minmax_y_update_id = (ps_cur_cluster->min_y < ps_cluster_data->min_y) |
3081 | 0 | ? ((ps_cur_cluster->max_y > ps_cluster_data->max_y) ? 0 : 2) |
3082 | 0 | : 1; |
3083 | | |
3084 | | /* Updation of centroid spread */ |
3085 | 0 | switch(minmax_x_update_id + (minmax_y_update_id << 2)) |
3086 | 0 | { |
3087 | 0 | case 1: |
3088 | 0 | { |
3089 | 0 | S32 mvd, mvd_q8; |
3090 | |
|
3091 | 0 | ps_cur_cluster->min_x = ps_cluster_data->min_x; |
3092 | |
|
3093 | 0 | mvd_q8 = ps_centroid->i4_pos_x_q8 - (ps_cur_cluster->min_x << 8); |
3094 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3095 | |
|
3096 | 0 | if(mvd > (max_dist_from_centroid)) |
3097 | 0 | { |
3098 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
3099 | 0 | } |
3100 | 0 | break; |
3101 | 0 | } |
3102 | 0 | case 2: |
3103 | 0 | { |
3104 | 0 | S32 mvd, mvd_q8; |
3105 | |
|
3106 | 0 | ps_cur_cluster->max_x = ps_cluster_data->max_x; |
3107 | |
|
3108 | 0 | mvd_q8 = (ps_cur_cluster->max_x << 8) - ps_centroid->i4_pos_x_q8; |
3109 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3110 | |
|
3111 | 0 | if(mvd > (max_dist_from_centroid)) |
3112 | 0 | { |
3113 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
3114 | 0 | } |
3115 | 0 | break; |
3116 | 0 | } |
3117 | 0 | case 4: |
3118 | 0 | { |
3119 | 0 | S32 mvd, mvd_q8; |
3120 | |
|
3121 | 0 | ps_cur_cluster->min_y = ps_cluster_data->min_y; |
3122 | |
|
3123 | 0 | mvd_q8 = ps_centroid->i4_pos_y_q8 - (ps_cur_cluster->min_y << 8); |
3124 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3125 | |
|
3126 | 0 | if(mvd > (max_dist_from_centroid)) |
3127 | 0 | { |
3128 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
3129 | 0 | } |
3130 | 0 | break; |
3131 | 0 | } |
3132 | 0 | case 5: |
3133 | 0 | { |
3134 | 0 | S32 mvd; |
3135 | 0 | S32 mvdx, mvdx_q8; |
3136 | 0 | S32 mvdy, mvdy_q8; |
3137 | |
|
3138 | 0 | mvdy_q8 = ps_centroid->i4_pos_y_q8 - (ps_cur_cluster->min_y << 8); |
3139 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3140 | |
|
3141 | 0 | mvdx_q8 = ps_centroid->i4_pos_x_q8 - (ps_cur_cluster->min_x << 8); |
3142 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3143 | |
|
3144 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
3145 | |
|
3146 | 0 | ps_cur_cluster->min_x = ps_cluster_data->min_x; |
3147 | 0 | ps_cur_cluster->min_y = ps_cluster_data->min_y; |
3148 | |
|
3149 | 0 | if(mvd > max_dist_from_centroid) |
3150 | 0 | { |
3151 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
3152 | 0 | } |
3153 | 0 | break; |
3154 | 0 | } |
3155 | 0 | case 6: |
3156 | 0 | { |
3157 | 0 | S32 mvd; |
3158 | 0 | S32 mvdx, mvdx_q8; |
3159 | 0 | S32 mvdy, mvdy_q8; |
3160 | |
|
3161 | 0 | mvdy_q8 = ps_centroid->i4_pos_y_q8 - (ps_cur_cluster->min_y << 8); |
3162 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3163 | |
|
3164 | 0 | mvdx_q8 = (ps_cur_cluster->max_x << 8) - ps_centroid->i4_pos_x_q8; |
3165 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3166 | |
|
3167 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
3168 | |
|
3169 | 0 | ps_cur_cluster->max_x = ps_cluster_data->max_x; |
3170 | 0 | ps_cur_cluster->min_y = ps_cluster_data->min_y; |
3171 | |
|
3172 | 0 | if(mvd > max_dist_from_centroid) |
3173 | 0 | { |
3174 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
3175 | 0 | } |
3176 | 0 | break; |
3177 | 0 | } |
3178 | 0 | case 8: |
3179 | 0 | { |
3180 | 0 | S32 mvd, mvd_q8; |
3181 | |
|
3182 | 0 | ps_cur_cluster->max_y = ps_cluster_data->max_y; |
3183 | |
|
3184 | 0 | mvd_q8 = (ps_cur_cluster->max_y << 8) - ps_centroid->i4_pos_y_q8; |
3185 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3186 | |
|
3187 | 0 | if(mvd > (max_dist_from_centroid)) |
3188 | 0 | { |
3189 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
3190 | 0 | } |
3191 | 0 | break; |
3192 | 0 | } |
3193 | 0 | case 9: |
3194 | 0 | { |
3195 | 0 | S32 mvd; |
3196 | 0 | S32 mvdx, mvdx_q8; |
3197 | 0 | S32 mvdy, mvdy_q8; |
3198 | |
|
3199 | 0 | mvdx_q8 = ps_centroid->i4_pos_x_q8 - (ps_cur_cluster->min_x << 8); |
3200 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3201 | |
|
3202 | 0 | mvdy_q8 = (ps_cur_cluster->max_y << 8) - ps_centroid->i4_pos_y_q8; |
3203 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3204 | |
|
3205 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
3206 | |
|
3207 | 0 | ps_cur_cluster->min_x = ps_cluster_data->min_x; |
3208 | 0 | ps_cur_cluster->max_y = ps_cluster_data->max_y; |
3209 | |
|
3210 | 0 | if(mvd > max_dist_from_centroid) |
3211 | 0 | { |
3212 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
3213 | 0 | } |
3214 | 0 | break; |
3215 | 0 | } |
3216 | 0 | case 10: |
3217 | 0 | { |
3218 | 0 | S32 mvd; |
3219 | 0 | S32 mvdx, mvdx_q8; |
3220 | 0 | S32 mvdy, mvdy_q8; |
3221 | |
|
3222 | 0 | mvdx_q8 = (ps_cur_cluster->max_x << 8) - ps_centroid->i4_pos_x_q8; |
3223 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3224 | |
|
3225 | 0 | mvdy_q8 = (ps_cur_cluster->max_y << 8) - ps_centroid->i4_pos_y_q8; |
3226 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3227 | |
|
3228 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
3229 | |
|
3230 | 0 | ps_cur_cluster->max_x = ps_cluster_data->max_x; |
3231 | 0 | ps_cur_cluster->max_y = ps_cluster_data->max_y; |
3232 | |
|
3233 | 0 | if(mvd > ps_cluster_data->max_dist_from_centroid) |
3234 | 0 | { |
3235 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
3236 | 0 | } |
3237 | 0 | break; |
3238 | 0 | } |
3239 | 0 | default: |
3240 | 0 | { |
3241 | 0 | break; |
3242 | 0 | } |
3243 | 0 | } |
3244 | | |
3245 | 0 | hme_try_cluster_merge(ps_root, pu1_num_clusters, idx_of_updated_cluster); |
3246 | |
|
3247 | 0 | return; |
3248 | 0 | } |
3249 | | |
3250 | 0 | num_clusters_evaluated++; |
3251 | 0 | } |
3252 | 0 | } |
3253 | | |
3254 | | /** |
3255 | | ******************************************************************************** |
3256 | | * @fn void hme_find_and_update_clusters |
3257 | | * ( |
3258 | | * cluster_data_t *ps_cluster_data, |
3259 | | * S32 *pi4_num_clusters, |
3260 | | * S32 mvx, |
3261 | | * S32 mvy, |
3262 | | * S32 ref_idx, |
3263 | | * PART_ID_T e_part_id |
3264 | | * ) |
3265 | | * |
3266 | | * @brief Implementation fo the clustering algorithm |
3267 | | * |
3268 | | * @param[in/out] ps_cluster_data: pointer to cluster_data_t struct |
3269 | | * |
3270 | | * @param[in/out] pi4_num_clusters : pointer to number of clusters |
3271 | | * |
3272 | | * @param[in] mvx : x co-ordinate of the motion vector |
3273 | | * |
3274 | | * @param[in] mvy : y co-ordinate of the motion vector |
3275 | | * |
3276 | | * @param[in] ref_idx : ref_id of the motion vector |
3277 | | * |
3278 | | * @param[in] e_part_id : partition id of the motion vector |
3279 | | * |
3280 | | * @return None |
3281 | | ******************************************************************************** |
3282 | | */ |
3283 | | void hme_find_and_update_clusters( |
3284 | | cluster_data_t *ps_cluster_data, |
3285 | | U08 *pu1_num_clusters, |
3286 | | S16 i2_mv_x, |
3287 | | S16 i2_mv_y, |
3288 | | U08 i1_ref_idx, |
3289 | | S32 i4_sdi, |
3290 | | PART_ID_T e_part_id, |
3291 | | U08 is_part_of_bi) |
3292 | 0 | { |
3293 | 0 | S32 i; |
3294 | 0 | S32 min_mvd_cluster_id = -1; |
3295 | 0 | S32 mvd, mvd_limit, mvdx, mvdy; |
3296 | 0 | S32 min_mvdx, min_mvdy; |
3297 | |
|
3298 | 0 | S32 min_mvd = MAX_32BIT_VAL; |
3299 | 0 | S32 num_clusters = *pu1_num_clusters; |
3300 | |
|
3301 | 0 | S32 mvx = i2_mv_x; |
3302 | 0 | S32 mvy = i2_mv_y; |
3303 | 0 | S32 ref_idx = i1_ref_idx; |
3304 | 0 | S32 sdi = i4_sdi; |
3305 | 0 | S32 new_cluster_idx = MAX_NUM_CLUSTERS_16x16; |
3306 | |
|
3307 | 0 | if(num_clusters == 0) |
3308 | 0 | { |
3309 | 0 | cluster_data_t *ps_data = &ps_cluster_data[num_clusters]; |
3310 | |
|
3311 | 0 | ps_data->num_mvs = 1; |
3312 | 0 | ps_data->s_centroid.i4_pos_x_q8 = mvx << 8; |
3313 | 0 | ps_data->s_centroid.i4_pos_y_q8 = mvy << 8; |
3314 | 0 | ps_data->ref_id = ref_idx; |
3315 | 0 | ps_data->area_in_pixels = gai4_partition_area[e_part_id]; |
3316 | 0 | ps_data->as_mv[0].pixel_count = gai4_partition_area[e_part_id]; |
3317 | 0 | ps_data->as_mv[0].mvx = mvx; |
3318 | 0 | ps_data->as_mv[0].mvy = mvy; |
3319 | | |
3320 | | /***************************/ |
3321 | 0 | ps_data->as_mv[0].is_uni = !is_part_of_bi; |
3322 | 0 | ps_data->as_mv[0].sdi = sdi; |
3323 | 0 | if(is_part_of_bi) |
3324 | 0 | { |
3325 | 0 | ps_data->bi_mv_pixel_area += ps_data->area_in_pixels; |
3326 | 0 | } |
3327 | 0 | else |
3328 | 0 | { |
3329 | 0 | ps_data->uni_mv_pixel_area += ps_data->area_in_pixels; |
3330 | 0 | } |
3331 | | /**************************/ |
3332 | 0 | ps_data->max_x = mvx; |
3333 | 0 | ps_data->min_x = mvx; |
3334 | 0 | ps_data->max_y = mvy; |
3335 | 0 | ps_data->min_y = mvy; |
3336 | |
|
3337 | 0 | ps_data->is_valid_cluster = 1; |
3338 | |
|
3339 | 0 | *pu1_num_clusters = 1; |
3340 | 0 | } |
3341 | 0 | else |
3342 | 0 | { |
3343 | 0 | S32 num_clusters_evaluated = 0; |
3344 | |
|
3345 | 0 | for(i = 0; num_clusters_evaluated < num_clusters; i++) |
3346 | 0 | { |
3347 | 0 | cluster_data_t *ps_data = &ps_cluster_data[i]; |
3348 | |
|
3349 | 0 | centroid_t *ps_centroid; |
3350 | |
|
3351 | 0 | S32 mvx_q8; |
3352 | 0 | S32 mvy_q8; |
3353 | 0 | S32 posx_q8; |
3354 | 0 | S32 posy_q8; |
3355 | 0 | S32 mvdx_q8; |
3356 | 0 | S32 mvdy_q8; |
3357 | | |
3358 | | /* In anticipation of a possible merging of clusters */ |
3359 | 0 | if(ps_data->is_valid_cluster == 0) |
3360 | 0 | { |
3361 | 0 | new_cluster_idx = i; |
3362 | 0 | continue; |
3363 | 0 | } |
3364 | | |
3365 | 0 | if(ref_idx != ps_data->ref_id) |
3366 | 0 | { |
3367 | 0 | num_clusters_evaluated++; |
3368 | 0 | continue; |
3369 | 0 | } |
3370 | | |
3371 | 0 | ps_centroid = &ps_data->s_centroid; |
3372 | 0 | posx_q8 = ps_centroid->i4_pos_x_q8; |
3373 | 0 | posy_q8 = ps_centroid->i4_pos_y_q8; |
3374 | |
|
3375 | 0 | mvx_q8 = mvx << 8; |
3376 | 0 | mvy_q8 = mvy << 8; |
3377 | |
|
3378 | 0 | mvdx_q8 = posx_q8 - mvx_q8; |
3379 | 0 | mvdy_q8 = posy_q8 - mvy_q8; |
3380 | |
|
3381 | 0 | mvdx = (((mvdx_q8 + (1 << 7)) >> 8)); |
3382 | 0 | mvdy = (((mvdy_q8 + (1 << 7)) >> 8)); |
3383 | |
|
3384 | 0 | mvd = ABS(mvdx) + ABS(mvdy); |
3385 | |
|
3386 | 0 | if(mvd < min_mvd) |
3387 | 0 | { |
3388 | 0 | min_mvd = mvd; |
3389 | 0 | min_mvdx = mvdx; |
3390 | 0 | min_mvdy = mvdy; |
3391 | 0 | min_mvd_cluster_id = i; |
3392 | 0 | } |
3393 | |
|
3394 | 0 | num_clusters_evaluated++; |
3395 | 0 | } |
3396 | |
|
3397 | 0 | mvd_limit = (min_mvd_cluster_id == -1) |
3398 | 0 | ? ps_cluster_data[0].max_dist_from_centroid |
3399 | 0 | : ps_cluster_data[min_mvd_cluster_id].max_dist_from_centroid; |
3400 | | |
3401 | | /* This condition implies that min_mvd has been updated */ |
3402 | 0 | if(min_mvd <= mvd_limit) |
3403 | 0 | { |
3404 | 0 | hme_update_cluster_attributes( |
3405 | 0 | &ps_cluster_data[min_mvd_cluster_id], |
3406 | 0 | mvx, |
3407 | 0 | mvy, |
3408 | 0 | min_mvdx, |
3409 | 0 | min_mvdy, |
3410 | 0 | ref_idx, |
3411 | 0 | sdi, |
3412 | 0 | is_part_of_bi, |
3413 | 0 | e_part_id); |
3414 | |
|
3415 | 0 | if(PRT_NxN == ge_part_id_to_part_type[e_part_id]) |
3416 | 0 | { |
3417 | 0 | hme_try_cluster_merge(ps_cluster_data, pu1_num_clusters, min_mvd_cluster_id); |
3418 | 0 | } |
3419 | 0 | } |
3420 | 0 | else |
3421 | 0 | { |
3422 | 0 | cluster_data_t *ps_data = (new_cluster_idx == MAX_NUM_CLUSTERS_16x16) |
3423 | 0 | ? &ps_cluster_data[num_clusters] |
3424 | 0 | : &ps_cluster_data[new_cluster_idx]; |
3425 | |
|
3426 | 0 | ps_data->num_mvs = 1; |
3427 | 0 | ps_data->s_centroid.i4_pos_x_q8 = mvx << 8; |
3428 | 0 | ps_data->s_centroid.i4_pos_y_q8 = mvy << 8; |
3429 | 0 | ps_data->ref_id = ref_idx; |
3430 | 0 | ps_data->area_in_pixels = gai4_partition_area[e_part_id]; |
3431 | 0 | ps_data->as_mv[0].pixel_count = gai4_partition_area[e_part_id]; |
3432 | 0 | ps_data->as_mv[0].mvx = mvx; |
3433 | 0 | ps_data->as_mv[0].mvy = mvy; |
3434 | | |
3435 | | /***************************/ |
3436 | 0 | ps_data->as_mv[0].is_uni = !is_part_of_bi; |
3437 | 0 | ps_data->as_mv[0].sdi = sdi; |
3438 | 0 | if(is_part_of_bi) |
3439 | 0 | { |
3440 | 0 | ps_data->bi_mv_pixel_area += ps_data->area_in_pixels; |
3441 | 0 | } |
3442 | 0 | else |
3443 | 0 | { |
3444 | 0 | ps_data->uni_mv_pixel_area += ps_data->area_in_pixels; |
3445 | 0 | } |
3446 | | /**************************/ |
3447 | 0 | ps_data->max_x = mvx; |
3448 | 0 | ps_data->min_x = mvx; |
3449 | 0 | ps_data->max_y = mvy; |
3450 | 0 | ps_data->min_y = mvy; |
3451 | |
|
3452 | 0 | ps_data->is_valid_cluster = 1; |
3453 | |
|
3454 | 0 | num_clusters++; |
3455 | 0 | *pu1_num_clusters = num_clusters; |
3456 | 0 | } |
3457 | 0 | } |
3458 | 0 | } |
3459 | | |
3460 | | /** |
3461 | | ******************************************************************************** |
3462 | | * @fn void hme_update_32x32_cluster_attributes |
3463 | | * ( |
3464 | | * cluster_32x32_blk_t *ps_blk_32x32, |
3465 | | * cluster_data_t *ps_cluster_data |
3466 | | * ) |
3467 | | * |
3468 | | * @brief Updates attributes for 32x32 clusters based on the attributes of |
3469 | | * the constituent 16x16 clusters |
3470 | | * |
3471 | | * @param[out] ps_blk_32x32: structure containing 32x32 block results |
3472 | | * |
3473 | | * @param[in] ps_cluster_data : structure containing 16x16 block results |
3474 | | * |
3475 | | * @return None |
3476 | | ******************************************************************************** |
3477 | | */ |
3478 | | void hme_update_32x32_cluster_attributes( |
3479 | | cluster_32x32_blk_t *ps_blk_32x32, cluster_data_t *ps_cluster_data) |
3480 | 0 | { |
3481 | 0 | cluster_data_t *ps_cur_cluster_32; |
3482 | |
|
3483 | 0 | S32 i; |
3484 | 0 | S32 mvd_limit; |
3485 | |
|
3486 | 0 | S32 num_clusters = ps_blk_32x32->num_clusters; |
3487 | |
|
3488 | 0 | if(0 == num_clusters) |
3489 | 0 | { |
3490 | 0 | ps_cur_cluster_32 = &ps_blk_32x32->as_cluster_data[0]; |
3491 | |
|
3492 | 0 | ps_blk_32x32->num_clusters++; |
3493 | 0 | ps_blk_32x32->au1_num_clusters[ps_cluster_data->ref_id]++; |
3494 | |
|
3495 | 0 | ps_cur_cluster_32->is_valid_cluster = 1; |
3496 | |
|
3497 | 0 | ps_cur_cluster_32->area_in_pixels = ps_cluster_data->area_in_pixels; |
3498 | 0 | ps_cur_cluster_32->bi_mv_pixel_area += ps_cluster_data->bi_mv_pixel_area; |
3499 | 0 | ps_cur_cluster_32->uni_mv_pixel_area += ps_cluster_data->uni_mv_pixel_area; |
3500 | |
|
3501 | 0 | memcpy( |
3502 | 0 | ps_cur_cluster_32->as_mv, |
3503 | 0 | ps_cluster_data->as_mv, |
3504 | 0 | sizeof(mv_data_t) * ps_cluster_data->num_mvs); |
3505 | |
|
3506 | 0 | ps_cur_cluster_32->num_mvs = ps_cluster_data->num_mvs; |
3507 | |
|
3508 | 0 | ps_cur_cluster_32->ref_id = ps_cluster_data->ref_id; |
3509 | |
|
3510 | 0 | ps_cur_cluster_32->max_x = ps_cluster_data->max_x; |
3511 | 0 | ps_cur_cluster_32->max_y = ps_cluster_data->max_y; |
3512 | 0 | ps_cur_cluster_32->min_x = ps_cluster_data->min_x; |
3513 | 0 | ps_cur_cluster_32->min_y = ps_cluster_data->min_y; |
3514 | |
|
3515 | 0 | ps_cur_cluster_32->s_centroid = ps_cluster_data->s_centroid; |
3516 | 0 | } |
3517 | 0 | else |
3518 | 0 | { |
3519 | 0 | centroid_t *ps_centroid; |
3520 | |
|
3521 | 0 | S32 cur_posx_q8, cur_posy_q8; |
3522 | 0 | S32 min_mvd_cluster_id = -1; |
3523 | 0 | S32 mvd; |
3524 | 0 | S32 mvdx; |
3525 | 0 | S32 mvdy; |
3526 | 0 | S32 mvdx_min; |
3527 | 0 | S32 mvdy_min; |
3528 | 0 | S32 mvdx_q8; |
3529 | 0 | S32 mvdy_q8; |
3530 | |
|
3531 | 0 | S32 num_clusters_evaluated = 0; |
3532 | |
|
3533 | 0 | S32 mvd_min = MAX_32BIT_VAL; |
3534 | |
|
3535 | 0 | S32 mvx_inp_q8 = ps_cluster_data->s_centroid.i4_pos_x_q8; |
3536 | 0 | S32 mvy_inp_q8 = ps_cluster_data->s_centroid.i4_pos_y_q8; |
3537 | |
|
3538 | 0 | for(i = 0; num_clusters_evaluated < num_clusters; i++) |
3539 | 0 | { |
3540 | 0 | ps_cur_cluster_32 = &ps_blk_32x32->as_cluster_data[i]; |
3541 | |
|
3542 | 0 | if(ps_cur_cluster_32->ref_id != ps_cluster_data->ref_id) |
3543 | 0 | { |
3544 | 0 | num_clusters_evaluated++; |
3545 | 0 | continue; |
3546 | 0 | } |
3547 | 0 | if(!ps_cluster_data->is_valid_cluster) |
3548 | 0 | { |
3549 | 0 | continue; |
3550 | 0 | } |
3551 | | |
3552 | 0 | num_clusters_evaluated++; |
3553 | |
|
3554 | 0 | ps_centroid = &ps_cur_cluster_32->s_centroid; |
3555 | |
|
3556 | 0 | cur_posx_q8 = ps_centroid->i4_pos_x_q8; |
3557 | 0 | cur_posy_q8 = ps_centroid->i4_pos_y_q8; |
3558 | |
|
3559 | 0 | mvdx_q8 = cur_posx_q8 - mvx_inp_q8; |
3560 | 0 | mvdy_q8 = cur_posy_q8 - mvy_inp_q8; |
3561 | |
|
3562 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3563 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3564 | |
|
3565 | 0 | mvd = ABS(mvdx) + ABS(mvdy); |
3566 | |
|
3567 | 0 | if(mvd < mvd_min) |
3568 | 0 | { |
3569 | 0 | mvd_min = mvd; |
3570 | 0 | mvdx_min = mvdx; |
3571 | 0 | mvdy_min = mvdy; |
3572 | 0 | min_mvd_cluster_id = i; |
3573 | 0 | } |
3574 | 0 | } |
3575 | |
|
3576 | 0 | ps_cur_cluster_32 = &ps_blk_32x32->as_cluster_data[0]; |
3577 | |
|
3578 | 0 | mvd_limit = (min_mvd_cluster_id == -1) |
3579 | 0 | ? ps_cur_cluster_32[0].max_dist_from_centroid |
3580 | 0 | : ps_cur_cluster_32[min_mvd_cluster_id].max_dist_from_centroid; |
3581 | |
|
3582 | 0 | if(mvd_min <= mvd_limit) |
3583 | 0 | { |
3584 | 0 | LWORD64 i8_updated_posx; |
3585 | 0 | LWORD64 i8_updated_posy; |
3586 | 0 | WORD32 minmax_updated_x = 0; |
3587 | 0 | WORD32 minmax_updated_y = 0; |
3588 | |
|
3589 | 0 | ps_cur_cluster_32 = &ps_blk_32x32->as_cluster_data[min_mvd_cluster_id]; |
3590 | |
|
3591 | 0 | ps_centroid = &ps_cur_cluster_32->s_centroid; |
3592 | |
|
3593 | 0 | ps_cur_cluster_32->is_valid_cluster = 1; |
3594 | |
|
3595 | 0 | ps_cur_cluster_32->area_in_pixels += ps_cluster_data->area_in_pixels; |
3596 | 0 | ps_cur_cluster_32->bi_mv_pixel_area += ps_cluster_data->bi_mv_pixel_area; |
3597 | 0 | ps_cur_cluster_32->uni_mv_pixel_area += ps_cluster_data->uni_mv_pixel_area; |
3598 | |
|
3599 | 0 | memcpy( |
3600 | 0 | &ps_cur_cluster_32->as_mv[ps_cur_cluster_32->num_mvs], |
3601 | 0 | ps_cluster_data->as_mv, |
3602 | 0 | sizeof(mv_data_t) * ps_cluster_data->num_mvs); |
3603 | |
|
3604 | 0 | if((mvdx_min > 0) && ((ps_cur_cluster_32->min_x << 8) > mvx_inp_q8)) |
3605 | 0 | { |
3606 | 0 | ps_cur_cluster_32->min_x = (mvx_inp_q8 + ((1 << 7))) >> 8; |
3607 | 0 | minmax_updated_x = 1; |
3608 | 0 | } |
3609 | 0 | else if((mvdx_min < 0) && ((ps_cur_cluster_32->max_x << 8) < mvx_inp_q8)) |
3610 | 0 | { |
3611 | 0 | ps_cur_cluster_32->max_x = (mvx_inp_q8 + (1 << 7)) >> 8; |
3612 | 0 | minmax_updated_x = 2; |
3613 | 0 | } |
3614 | |
|
3615 | 0 | if((mvdy_min > 0) && ((ps_cur_cluster_32->min_y << 8) > mvy_inp_q8)) |
3616 | 0 | { |
3617 | 0 | ps_cur_cluster_32->min_y = (mvy_inp_q8 + (1 << 7)) >> 8; |
3618 | 0 | minmax_updated_y = 1; |
3619 | 0 | } |
3620 | 0 | else if((mvdy_min < 0) && ((ps_cur_cluster_32->max_y << 8) < mvy_inp_q8)) |
3621 | 0 | { |
3622 | 0 | ps_cur_cluster_32->max_y = (mvy_inp_q8 + (1 << 7)) >> 8; |
3623 | 0 | minmax_updated_y = 2; |
3624 | 0 | } |
3625 | |
|
3626 | 0 | switch((minmax_updated_y << 2) + minmax_updated_x) |
3627 | 0 | { |
3628 | 0 | case 1: |
3629 | 0 | { |
3630 | 0 | S32 mvd, mvd_q8; |
3631 | |
|
3632 | 0 | mvd_q8 = ps_centroid->i4_pos_x_q8 - (ps_cur_cluster_32->min_x << 8); |
3633 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3634 | |
|
3635 | 0 | if(mvd > (mvd_limit)) |
3636 | 0 | { |
3637 | 0 | ps_cur_cluster_32->max_dist_from_centroid = mvd; |
3638 | 0 | } |
3639 | 0 | break; |
3640 | 0 | } |
3641 | 0 | case 2: |
3642 | 0 | { |
3643 | 0 | S32 mvd, mvd_q8; |
3644 | |
|
3645 | 0 | mvd_q8 = (ps_cur_cluster_32->max_x << 8) - ps_centroid->i4_pos_x_q8; |
3646 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3647 | |
|
3648 | 0 | if(mvd > (mvd_limit)) |
3649 | 0 | { |
3650 | 0 | ps_cur_cluster_32->max_dist_from_centroid = mvd; |
3651 | 0 | } |
3652 | 0 | break; |
3653 | 0 | } |
3654 | 0 | case 4: |
3655 | 0 | { |
3656 | 0 | S32 mvd, mvd_q8; |
3657 | |
|
3658 | 0 | mvd_q8 = ps_centroid->i4_pos_y_q8 - (ps_cur_cluster_32->min_y << 8); |
3659 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3660 | |
|
3661 | 0 | if(mvd > (mvd_limit)) |
3662 | 0 | { |
3663 | 0 | ps_cur_cluster_32->max_dist_from_centroid = mvd; |
3664 | 0 | } |
3665 | 0 | break; |
3666 | 0 | } |
3667 | 0 | case 5: |
3668 | 0 | { |
3669 | 0 | S32 mvd; |
3670 | 0 | S32 mvdx, mvdx_q8; |
3671 | 0 | S32 mvdy, mvdy_q8; |
3672 | |
|
3673 | 0 | mvdy_q8 = ps_centroid->i4_pos_y_q8 - (ps_cur_cluster_32->min_y << 8); |
3674 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3675 | |
|
3676 | 0 | mvdx_q8 = ps_centroid->i4_pos_x_q8 - (ps_cur_cluster_32->min_x << 8); |
3677 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3678 | |
|
3679 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
3680 | |
|
3681 | 0 | if(mvd > mvd_limit) |
3682 | 0 | { |
3683 | 0 | ps_cur_cluster_32->max_dist_from_centroid = mvd; |
3684 | 0 | } |
3685 | 0 | break; |
3686 | 0 | } |
3687 | 0 | case 6: |
3688 | 0 | { |
3689 | 0 | S32 mvd; |
3690 | 0 | S32 mvdx, mvdx_q8; |
3691 | 0 | S32 mvdy, mvdy_q8; |
3692 | |
|
3693 | 0 | mvdy_q8 = ps_centroid->i4_pos_y_q8 - (ps_cur_cluster_32->min_y << 8); |
3694 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3695 | |
|
3696 | 0 | mvdx_q8 = (ps_cur_cluster_32->max_x << 8) - ps_centroid->i4_pos_x_q8; |
3697 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3698 | |
|
3699 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
3700 | |
|
3701 | 0 | if(mvd > mvd_limit) |
3702 | 0 | { |
3703 | 0 | ps_cur_cluster_32->max_dist_from_centroid = mvd; |
3704 | 0 | } |
3705 | 0 | break; |
3706 | 0 | } |
3707 | 0 | case 8: |
3708 | 0 | { |
3709 | 0 | S32 mvd, mvd_q8; |
3710 | |
|
3711 | 0 | mvd_q8 = (ps_cur_cluster_32->max_y << 8) - ps_centroid->i4_pos_y_q8; |
3712 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3713 | |
|
3714 | 0 | if(mvd > (mvd_limit)) |
3715 | 0 | { |
3716 | 0 | ps_cur_cluster_32->max_dist_from_centroid = mvd; |
3717 | 0 | } |
3718 | 0 | break; |
3719 | 0 | } |
3720 | 0 | case 9: |
3721 | 0 | { |
3722 | 0 | S32 mvd; |
3723 | 0 | S32 mvdx, mvdx_q8; |
3724 | 0 | S32 mvdy, mvdy_q8; |
3725 | |
|
3726 | 0 | mvdx_q8 = ps_centroid->i4_pos_x_q8 - (ps_cur_cluster_32->min_x << 8); |
3727 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3728 | |
|
3729 | 0 | mvdy_q8 = (ps_cur_cluster_32->max_y << 8) - ps_centroid->i4_pos_y_q8; |
3730 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3731 | |
|
3732 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
3733 | |
|
3734 | 0 | if(mvd > mvd_limit) |
3735 | 0 | { |
3736 | 0 | ps_cur_cluster_32->max_dist_from_centroid = mvd; |
3737 | 0 | } |
3738 | 0 | break; |
3739 | 0 | } |
3740 | 0 | case 10: |
3741 | 0 | { |
3742 | 0 | S32 mvd; |
3743 | 0 | S32 mvdx, mvdx_q8; |
3744 | 0 | S32 mvdy, mvdy_q8; |
3745 | |
|
3746 | 0 | mvdx_q8 = (ps_cur_cluster_32->max_x << 8) - ps_centroid->i4_pos_x_q8; |
3747 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3748 | |
|
3749 | 0 | mvdy_q8 = (ps_cur_cluster_32->max_y << 8) - ps_centroid->i4_pos_y_q8; |
3750 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3751 | |
|
3752 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
3753 | |
|
3754 | 0 | if(mvd > ps_cur_cluster_32->max_dist_from_centroid) |
3755 | 0 | { |
3756 | 0 | ps_cur_cluster_32->max_dist_from_centroid = mvd; |
3757 | 0 | } |
3758 | 0 | break; |
3759 | 0 | } |
3760 | 0 | default: |
3761 | 0 | { |
3762 | 0 | break; |
3763 | 0 | } |
3764 | 0 | } |
3765 | | |
3766 | 0 | i8_updated_posx = ((LWORD64)ps_centroid->i4_pos_x_q8 * ps_cur_cluster_32->num_mvs) + |
3767 | 0 | ((LWORD64)mvx_inp_q8 * ps_cluster_data->num_mvs); |
3768 | 0 | i8_updated_posy = ((LWORD64)ps_centroid->i4_pos_y_q8 * ps_cur_cluster_32->num_mvs) + |
3769 | 0 | ((LWORD64)mvy_inp_q8 * ps_cluster_data->num_mvs); |
3770 | |
|
3771 | 0 | ps_cur_cluster_32->num_mvs += ps_cluster_data->num_mvs; |
3772 | |
|
3773 | 0 | ps_centroid->i4_pos_x_q8 = (WORD32)(i8_updated_posx / ps_cur_cluster_32->num_mvs); |
3774 | 0 | ps_centroid->i4_pos_y_q8 = (WORD32)(i8_updated_posy / ps_cur_cluster_32->num_mvs); |
3775 | 0 | } |
3776 | 0 | else if(num_clusters < MAX_NUM_CLUSTERS_32x32) |
3777 | 0 | { |
3778 | 0 | ps_cur_cluster_32 = &ps_blk_32x32->as_cluster_data[num_clusters]; |
3779 | |
|
3780 | 0 | ps_blk_32x32->num_clusters++; |
3781 | 0 | ps_blk_32x32->au1_num_clusters[ps_cluster_data->ref_id]++; |
3782 | |
|
3783 | 0 | ps_cur_cluster_32->is_valid_cluster = 1; |
3784 | |
|
3785 | 0 | ps_cur_cluster_32->area_in_pixels = ps_cluster_data->area_in_pixels; |
3786 | 0 | ps_cur_cluster_32->bi_mv_pixel_area += ps_cluster_data->bi_mv_pixel_area; |
3787 | 0 | ps_cur_cluster_32->uni_mv_pixel_area += ps_cluster_data->uni_mv_pixel_area; |
3788 | |
|
3789 | 0 | memcpy( |
3790 | 0 | ps_cur_cluster_32->as_mv, |
3791 | 0 | ps_cluster_data->as_mv, |
3792 | 0 | sizeof(mv_data_t) * ps_cluster_data->num_mvs); |
3793 | |
|
3794 | 0 | ps_cur_cluster_32->num_mvs = ps_cluster_data->num_mvs; |
3795 | |
|
3796 | 0 | ps_cur_cluster_32->ref_id = ps_cluster_data->ref_id; |
3797 | |
|
3798 | 0 | ps_cur_cluster_32->max_x = ps_cluster_data->max_x; |
3799 | 0 | ps_cur_cluster_32->max_y = ps_cluster_data->max_y; |
3800 | 0 | ps_cur_cluster_32->min_x = ps_cluster_data->min_x; |
3801 | 0 | ps_cur_cluster_32->min_y = ps_cluster_data->min_y; |
3802 | |
|
3803 | 0 | ps_cur_cluster_32->s_centroid = ps_cluster_data->s_centroid; |
3804 | 0 | } |
3805 | 0 | } |
3806 | 0 | } |
3807 | | |
3808 | | /** |
3809 | | ******************************************************************************** |
3810 | | * @fn void hme_update_64x64_cluster_attributes |
3811 | | * ( |
3812 | | * cluster_64x64_blk_t *ps_blk_32x32, |
3813 | | * cluster_data_t *ps_cluster_data |
3814 | | * ) |
3815 | | * |
3816 | | * @brief Updates attributes for 64x64 clusters based on the attributes of |
3817 | | * the constituent 16x16 clusters |
3818 | | * |
3819 | | * @param[out] ps_blk_64x64: structure containing 64x64 block results |
3820 | | * |
3821 | | * @param[in] ps_cluster_data : structure containing 32x32 block results |
3822 | | * |
3823 | | * @return None |
3824 | | ******************************************************************************** |
3825 | | */ |
3826 | | void hme_update_64x64_cluster_attributes( |
3827 | | cluster_64x64_blk_t *ps_blk_64x64, cluster_data_t *ps_cluster_data) |
3828 | 0 | { |
3829 | 0 | cluster_data_t *ps_cur_cluster_64; |
3830 | |
|
3831 | 0 | S32 i; |
3832 | 0 | S32 mvd_limit; |
3833 | |
|
3834 | 0 | S32 num_clusters = ps_blk_64x64->num_clusters; |
3835 | |
|
3836 | 0 | if(0 == num_clusters) |
3837 | 0 | { |
3838 | 0 | ps_cur_cluster_64 = &ps_blk_64x64->as_cluster_data[0]; |
3839 | |
|
3840 | 0 | ps_blk_64x64->num_clusters++; |
3841 | 0 | ps_blk_64x64->au1_num_clusters[ps_cluster_data->ref_id]++; |
3842 | |
|
3843 | 0 | ps_cur_cluster_64->is_valid_cluster = 1; |
3844 | |
|
3845 | 0 | ps_cur_cluster_64->area_in_pixels = ps_cluster_data->area_in_pixels; |
3846 | 0 | ps_cur_cluster_64->bi_mv_pixel_area += ps_cluster_data->bi_mv_pixel_area; |
3847 | 0 | ps_cur_cluster_64->uni_mv_pixel_area += ps_cluster_data->uni_mv_pixel_area; |
3848 | |
|
3849 | 0 | memcpy( |
3850 | 0 | ps_cur_cluster_64->as_mv, |
3851 | 0 | ps_cluster_data->as_mv, |
3852 | 0 | sizeof(mv_data_t) * ps_cluster_data->num_mvs); |
3853 | |
|
3854 | 0 | ps_cur_cluster_64->num_mvs = ps_cluster_data->num_mvs; |
3855 | |
|
3856 | 0 | ps_cur_cluster_64->ref_id = ps_cluster_data->ref_id; |
3857 | |
|
3858 | 0 | ps_cur_cluster_64->max_x = ps_cluster_data->max_x; |
3859 | 0 | ps_cur_cluster_64->max_y = ps_cluster_data->max_y; |
3860 | 0 | ps_cur_cluster_64->min_x = ps_cluster_data->min_x; |
3861 | 0 | ps_cur_cluster_64->min_y = ps_cluster_data->min_y; |
3862 | |
|
3863 | 0 | ps_cur_cluster_64->s_centroid = ps_cluster_data->s_centroid; |
3864 | 0 | } |
3865 | 0 | else |
3866 | 0 | { |
3867 | 0 | centroid_t *ps_centroid; |
3868 | |
|
3869 | 0 | S32 cur_posx_q8, cur_posy_q8; |
3870 | 0 | S32 min_mvd_cluster_id = -1; |
3871 | 0 | S32 mvd; |
3872 | 0 | S32 mvdx; |
3873 | 0 | S32 mvdy; |
3874 | 0 | S32 mvdx_min; |
3875 | 0 | S32 mvdy_min; |
3876 | 0 | S32 mvdx_q8; |
3877 | 0 | S32 mvdy_q8; |
3878 | |
|
3879 | 0 | S32 num_clusters_evaluated = 0; |
3880 | |
|
3881 | 0 | S32 mvd_min = MAX_32BIT_VAL; |
3882 | |
|
3883 | 0 | S32 mvx_inp_q8 = ps_cluster_data->s_centroid.i4_pos_x_q8; |
3884 | 0 | S32 mvy_inp_q8 = ps_cluster_data->s_centroid.i4_pos_y_q8; |
3885 | |
|
3886 | 0 | for(i = 0; num_clusters_evaluated < num_clusters; i++) |
3887 | 0 | { |
3888 | 0 | ps_cur_cluster_64 = &ps_blk_64x64->as_cluster_data[i]; |
3889 | |
|
3890 | 0 | if(ps_cur_cluster_64->ref_id != ps_cluster_data->ref_id) |
3891 | 0 | { |
3892 | 0 | num_clusters_evaluated++; |
3893 | 0 | continue; |
3894 | 0 | } |
3895 | | |
3896 | 0 | if(!ps_cur_cluster_64->is_valid_cluster) |
3897 | 0 | { |
3898 | 0 | continue; |
3899 | 0 | } |
3900 | | |
3901 | 0 | num_clusters_evaluated++; |
3902 | |
|
3903 | 0 | ps_centroid = &ps_cur_cluster_64->s_centroid; |
3904 | |
|
3905 | 0 | cur_posx_q8 = ps_centroid->i4_pos_x_q8; |
3906 | 0 | cur_posy_q8 = ps_centroid->i4_pos_y_q8; |
3907 | |
|
3908 | 0 | mvdx_q8 = cur_posx_q8 - mvx_inp_q8; |
3909 | 0 | mvdy_q8 = cur_posy_q8 - mvy_inp_q8; |
3910 | |
|
3911 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
3912 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
3913 | |
|
3914 | 0 | mvd = ABS(mvdx) + ABS(mvdy); |
3915 | |
|
3916 | 0 | if(mvd < mvd_min) |
3917 | 0 | { |
3918 | 0 | mvd_min = mvd; |
3919 | 0 | mvdx_min = mvdx; |
3920 | 0 | mvdy_min = mvdy; |
3921 | 0 | min_mvd_cluster_id = i; |
3922 | 0 | } |
3923 | 0 | } |
3924 | |
|
3925 | 0 | ps_cur_cluster_64 = ps_blk_64x64->as_cluster_data; |
3926 | |
|
3927 | 0 | mvd_limit = (min_mvd_cluster_id == -1) |
3928 | 0 | ? ps_cur_cluster_64[0].max_dist_from_centroid |
3929 | 0 | : ps_cur_cluster_64[min_mvd_cluster_id].max_dist_from_centroid; |
3930 | |
|
3931 | 0 | if(mvd_min <= mvd_limit) |
3932 | 0 | { |
3933 | 0 | LWORD64 i8_updated_posx; |
3934 | 0 | LWORD64 i8_updated_posy; |
3935 | 0 | WORD32 minmax_updated_x = 0; |
3936 | 0 | WORD32 minmax_updated_y = 0; |
3937 | |
|
3938 | 0 | ps_cur_cluster_64 = &ps_blk_64x64->as_cluster_data[min_mvd_cluster_id]; |
3939 | |
|
3940 | 0 | ps_centroid = &ps_cur_cluster_64->s_centroid; |
3941 | |
|
3942 | 0 | ps_cur_cluster_64->is_valid_cluster = 1; |
3943 | |
|
3944 | 0 | ps_cur_cluster_64->area_in_pixels += ps_cluster_data->area_in_pixels; |
3945 | 0 | ps_cur_cluster_64->bi_mv_pixel_area += ps_cluster_data->bi_mv_pixel_area; |
3946 | 0 | ps_cur_cluster_64->uni_mv_pixel_area += ps_cluster_data->uni_mv_pixel_area; |
3947 | |
|
3948 | 0 | memcpy( |
3949 | 0 | &ps_cur_cluster_64->as_mv[ps_cur_cluster_64->num_mvs], |
3950 | 0 | ps_cluster_data->as_mv, |
3951 | 0 | sizeof(mv_data_t) * ps_cluster_data->num_mvs); |
3952 | |
|
3953 | 0 | if((mvdx_min > 0) && ((ps_cur_cluster_64->min_x << 8) > mvx_inp_q8)) |
3954 | 0 | { |
3955 | 0 | ps_cur_cluster_64->min_x = (mvx_inp_q8 + (1 << 7)) >> 8; |
3956 | 0 | minmax_updated_x = 1; |
3957 | 0 | } |
3958 | 0 | else if((mvdx_min < 0) && ((ps_cur_cluster_64->max_x << 8) < mvx_inp_q8)) |
3959 | 0 | { |
3960 | 0 | ps_cur_cluster_64->max_x = (mvx_inp_q8 + (1 << 7)) >> 8; |
3961 | 0 | minmax_updated_x = 2; |
3962 | 0 | } |
3963 | |
|
3964 | 0 | if((mvdy_min > 0) && ((ps_cur_cluster_64->min_y << 8) > mvy_inp_q8)) |
3965 | 0 | { |
3966 | 0 | ps_cur_cluster_64->min_y = (mvy_inp_q8 + (1 << 7)) >> 8; |
3967 | 0 | minmax_updated_y = 1; |
3968 | 0 | } |
3969 | 0 | else if((mvdy_min < 0) && ((ps_cur_cluster_64->max_y << 8) < mvy_inp_q8)) |
3970 | 0 | { |
3971 | 0 | ps_cur_cluster_64->max_y = (mvy_inp_q8 + (1 << 7)) >> 8; |
3972 | 0 | minmax_updated_y = 2; |
3973 | 0 | } |
3974 | |
|
3975 | 0 | switch((minmax_updated_y << 2) + minmax_updated_x) |
3976 | 0 | { |
3977 | 0 | case 1: |
3978 | 0 | { |
3979 | 0 | S32 mvd, mvd_q8; |
3980 | |
|
3981 | 0 | mvd_q8 = ps_centroid->i4_pos_x_q8 - (ps_cur_cluster_64->min_x << 8); |
3982 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3983 | |
|
3984 | 0 | if(mvd > (mvd_limit)) |
3985 | 0 | { |
3986 | 0 | ps_cur_cluster_64->max_dist_from_centroid = mvd; |
3987 | 0 | } |
3988 | 0 | break; |
3989 | 0 | } |
3990 | 0 | case 2: |
3991 | 0 | { |
3992 | 0 | S32 mvd, mvd_q8; |
3993 | |
|
3994 | 0 | mvd_q8 = (ps_cur_cluster_64->max_x << 8) - ps_centroid->i4_pos_x_q8; |
3995 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
3996 | |
|
3997 | 0 | if(mvd > (mvd_limit)) |
3998 | 0 | { |
3999 | 0 | ps_cur_cluster_64->max_dist_from_centroid = mvd; |
4000 | 0 | } |
4001 | 0 | break; |
4002 | 0 | } |
4003 | 0 | case 4: |
4004 | 0 | { |
4005 | 0 | S32 mvd, mvd_q8; |
4006 | |
|
4007 | 0 | mvd_q8 = ps_centroid->i4_pos_y_q8 - (ps_cur_cluster_64->min_y << 8); |
4008 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
4009 | |
|
4010 | 0 | if(mvd > (mvd_limit)) |
4011 | 0 | { |
4012 | 0 | ps_cur_cluster_64->max_dist_from_centroid = mvd; |
4013 | 0 | } |
4014 | 0 | break; |
4015 | 0 | } |
4016 | 0 | case 5: |
4017 | 0 | { |
4018 | 0 | S32 mvd; |
4019 | 0 | S32 mvdx, mvdx_q8; |
4020 | 0 | S32 mvdy, mvdy_q8; |
4021 | |
|
4022 | 0 | mvdy_q8 = ps_centroid->i4_pos_y_q8 - (ps_cur_cluster_64->min_y << 8); |
4023 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
4024 | |
|
4025 | 0 | mvdx_q8 = ps_centroid->i4_pos_x_q8 - (ps_cur_cluster_64->min_x << 8); |
4026 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
4027 | |
|
4028 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
4029 | |
|
4030 | 0 | if(mvd > mvd_limit) |
4031 | 0 | { |
4032 | 0 | ps_cur_cluster_64->max_dist_from_centroid = mvd; |
4033 | 0 | } |
4034 | 0 | break; |
4035 | 0 | } |
4036 | 0 | case 6: |
4037 | 0 | { |
4038 | 0 | S32 mvd; |
4039 | 0 | S32 mvdx, mvdx_q8; |
4040 | 0 | S32 mvdy, mvdy_q8; |
4041 | |
|
4042 | 0 | mvdy_q8 = ps_centroid->i4_pos_y_q8 - (ps_cur_cluster_64->min_y << 8); |
4043 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
4044 | |
|
4045 | 0 | mvdx_q8 = (ps_cur_cluster_64->max_x << 8) - ps_centroid->i4_pos_x_q8; |
4046 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
4047 | |
|
4048 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
4049 | |
|
4050 | 0 | if(mvd > mvd_limit) |
4051 | 0 | { |
4052 | 0 | ps_cur_cluster_64->max_dist_from_centroid = mvd; |
4053 | 0 | } |
4054 | 0 | break; |
4055 | 0 | } |
4056 | 0 | case 8: |
4057 | 0 | { |
4058 | 0 | S32 mvd, mvd_q8; |
4059 | |
|
4060 | 0 | mvd_q8 = (ps_cur_cluster_64->max_y << 8) - ps_centroid->i4_pos_y_q8; |
4061 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
4062 | |
|
4063 | 0 | if(mvd > (mvd_limit)) |
4064 | 0 | { |
4065 | 0 | ps_cur_cluster_64->max_dist_from_centroid = mvd; |
4066 | 0 | } |
4067 | 0 | break; |
4068 | 0 | } |
4069 | 0 | case 9: |
4070 | 0 | { |
4071 | 0 | S32 mvd; |
4072 | 0 | S32 mvdx, mvdx_q8; |
4073 | 0 | S32 mvdy, mvdy_q8; |
4074 | |
|
4075 | 0 | mvdx_q8 = ps_centroid->i4_pos_x_q8 - (ps_cur_cluster_64->min_x << 8); |
4076 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
4077 | |
|
4078 | 0 | mvdy_q8 = (ps_cur_cluster_64->max_y << 8) - ps_centroid->i4_pos_y_q8; |
4079 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
4080 | |
|
4081 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
4082 | |
|
4083 | 0 | if(mvd > mvd_limit) |
4084 | 0 | { |
4085 | 0 | ps_cur_cluster_64->max_dist_from_centroid = mvd; |
4086 | 0 | } |
4087 | 0 | break; |
4088 | 0 | } |
4089 | 0 | case 10: |
4090 | 0 | { |
4091 | 0 | S32 mvd; |
4092 | 0 | S32 mvdx, mvdx_q8; |
4093 | 0 | S32 mvdy, mvdy_q8; |
4094 | |
|
4095 | 0 | mvdx_q8 = (ps_cur_cluster_64->max_x << 8) - ps_centroid->i4_pos_x_q8; |
4096 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
4097 | |
|
4098 | 0 | mvdy_q8 = (ps_cur_cluster_64->max_y << 8) - ps_centroid->i4_pos_y_q8; |
4099 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
4100 | |
|
4101 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
4102 | |
|
4103 | 0 | if(mvd > ps_cur_cluster_64->max_dist_from_centroid) |
4104 | 0 | { |
4105 | 0 | ps_cur_cluster_64->max_dist_from_centroid = mvd; |
4106 | 0 | } |
4107 | 0 | break; |
4108 | 0 | } |
4109 | 0 | default: |
4110 | 0 | { |
4111 | 0 | break; |
4112 | 0 | } |
4113 | 0 | } |
4114 | | |
4115 | 0 | i8_updated_posx = ((LWORD64)ps_centroid->i4_pos_x_q8 * ps_cur_cluster_64->num_mvs) + |
4116 | 0 | ((LWORD64)mvx_inp_q8 * ps_cluster_data->num_mvs); |
4117 | 0 | i8_updated_posy = ((LWORD64)ps_centroid->i4_pos_y_q8 * ps_cur_cluster_64->num_mvs) + |
4118 | 0 | ((LWORD64)mvy_inp_q8 * ps_cluster_data->num_mvs); |
4119 | |
|
4120 | 0 | ps_cur_cluster_64->num_mvs += ps_cluster_data->num_mvs; |
4121 | |
|
4122 | 0 | ps_centroid->i4_pos_x_q8 = (WORD32)(i8_updated_posx / ps_cur_cluster_64->num_mvs); |
4123 | 0 | ps_centroid->i4_pos_y_q8 = (WORD32)(i8_updated_posy / ps_cur_cluster_64->num_mvs); |
4124 | 0 | } |
4125 | 0 | else if(num_clusters < MAX_NUM_CLUSTERS_64x64) |
4126 | 0 | { |
4127 | 0 | ps_cur_cluster_64 = &ps_blk_64x64->as_cluster_data[num_clusters]; |
4128 | |
|
4129 | 0 | ps_blk_64x64->num_clusters++; |
4130 | 0 | ps_blk_64x64->au1_num_clusters[ps_cluster_data->ref_id]++; |
4131 | |
|
4132 | 0 | ps_cur_cluster_64->is_valid_cluster = 1; |
4133 | |
|
4134 | 0 | ps_cur_cluster_64->area_in_pixels = ps_cluster_data->area_in_pixels; |
4135 | 0 | ps_cur_cluster_64->bi_mv_pixel_area += ps_cluster_data->bi_mv_pixel_area; |
4136 | 0 | ps_cur_cluster_64->uni_mv_pixel_area += ps_cluster_data->uni_mv_pixel_area; |
4137 | |
|
4138 | 0 | memcpy( |
4139 | 0 | &ps_cur_cluster_64->as_mv[0], |
4140 | 0 | ps_cluster_data->as_mv, |
4141 | 0 | sizeof(mv_data_t) * ps_cluster_data->num_mvs); |
4142 | |
|
4143 | 0 | ps_cur_cluster_64->num_mvs = ps_cluster_data->num_mvs; |
4144 | |
|
4145 | 0 | ps_cur_cluster_64->ref_id = ps_cluster_data->ref_id; |
4146 | |
|
4147 | 0 | ps_cur_cluster_64->max_x = ps_cluster_data->max_x; |
4148 | 0 | ps_cur_cluster_64->max_y = ps_cluster_data->max_y; |
4149 | 0 | ps_cur_cluster_64->min_x = ps_cluster_data->min_x; |
4150 | 0 | ps_cur_cluster_64->min_y = ps_cluster_data->min_y; |
4151 | |
|
4152 | 0 | ps_cur_cluster_64->s_centroid = ps_cluster_data->s_centroid; |
4153 | 0 | } |
4154 | 0 | } |
4155 | 0 | } |
4156 | | |
4157 | | /** |
4158 | | ******************************************************************************** |
4159 | | * @fn void hme_update_32x32_clusters |
4160 | | * ( |
4161 | | * cluster_32x32_blk_t *ps_blk_32x32, |
4162 | | * cluster_16x16_blk_t *ps_blk_16x16 |
4163 | | * ) |
4164 | | * |
4165 | | * @brief Updates attributes for 32x32 clusters based on the attributes of |
4166 | | * the constituent 16x16 clusters |
4167 | | * |
4168 | | * @param[out] ps_blk_32x32: structure containing 32x32 block results |
4169 | | * |
4170 | | * @param[in] ps_blk_16x16 : structure containing 16x16 block results |
4171 | | * |
4172 | | * @return None |
4173 | | ******************************************************************************** |
4174 | | */ |
4175 | | static __inline void |
4176 | | hme_update_32x32_clusters(cluster_32x32_blk_t *ps_blk_32x32, cluster_16x16_blk_t *ps_blk_16x16) |
4177 | 0 | { |
4178 | 0 | cluster_16x16_blk_t *ps_blk_16x16_cur; |
4179 | 0 | cluster_data_t *ps_cur_cluster; |
4180 | |
|
4181 | 0 | S32 i, j; |
4182 | 0 | S32 num_clusters_cur_16x16_blk; |
4183 | |
|
4184 | 0 | for(i = 0; i < 4; i++) |
4185 | 0 | { |
4186 | 0 | S32 num_clusters_evaluated = 0; |
4187 | |
|
4188 | 0 | ps_blk_16x16_cur = &ps_blk_16x16[i]; |
4189 | |
|
4190 | 0 | num_clusters_cur_16x16_blk = ps_blk_16x16_cur->num_clusters; |
4191 | |
|
4192 | 0 | ps_blk_32x32->intra_mv_area += ps_blk_16x16_cur->intra_mv_area; |
4193 | |
|
4194 | 0 | ps_blk_32x32->best_inter_cost += ps_blk_16x16_cur->best_inter_cost; |
4195 | |
|
4196 | 0 | for(j = 0; num_clusters_evaluated < num_clusters_cur_16x16_blk; j++) |
4197 | 0 | { |
4198 | 0 | ps_cur_cluster = &ps_blk_16x16_cur->as_cluster_data[j]; |
4199 | |
|
4200 | 0 | if(!ps_cur_cluster->is_valid_cluster) |
4201 | 0 | { |
4202 | 0 | continue; |
4203 | 0 | } |
4204 | | |
4205 | 0 | hme_update_32x32_cluster_attributes(ps_blk_32x32, ps_cur_cluster); |
4206 | |
|
4207 | 0 | num_clusters_evaluated++; |
4208 | 0 | } |
4209 | 0 | } |
4210 | 0 | } |
4211 | | |
4212 | | /** |
4213 | | ******************************************************************************** |
4214 | | * @fn void hme_update_64x64_clusters |
4215 | | * ( |
4216 | | * cluster_64x64_blk_t *ps_blk_64x64, |
4217 | | * cluster_32x32_blk_t *ps_blk_32x32 |
4218 | | * ) |
4219 | | * |
4220 | | * @brief Updates attributes for 64x64 clusters based on the attributes of |
4221 | | * the constituent 16x16 clusters |
4222 | | * |
4223 | | * @param[out] ps_blk_64x64: structure containing 32x32 block results |
4224 | | * |
4225 | | * @param[in] ps_blk_32x32 : structure containing 16x16 block results |
4226 | | * |
4227 | | * @return None |
4228 | | ******************************************************************************** |
4229 | | */ |
4230 | | static __inline void |
4231 | | hme_update_64x64_clusters(cluster_64x64_blk_t *ps_blk_64x64, cluster_32x32_blk_t *ps_blk_32x32) |
4232 | 0 | { |
4233 | 0 | cluster_32x32_blk_t *ps_blk_32x32_cur; |
4234 | 0 | cluster_data_t *ps_cur_cluster; |
4235 | |
|
4236 | 0 | S32 i, j; |
4237 | 0 | S32 num_clusters_cur_32x32_blk; |
4238 | |
|
4239 | 0 | for(i = 0; i < 4; i++) |
4240 | 0 | { |
4241 | 0 | S32 num_clusters_evaluated = 0; |
4242 | |
|
4243 | 0 | ps_blk_32x32_cur = &ps_blk_32x32[i]; |
4244 | |
|
4245 | 0 | num_clusters_cur_32x32_blk = ps_blk_32x32_cur->num_clusters; |
4246 | |
|
4247 | 0 | ps_blk_64x64->intra_mv_area += ps_blk_32x32_cur->intra_mv_area; |
4248 | 0 | ps_blk_64x64->best_inter_cost += ps_blk_32x32_cur->best_inter_cost; |
4249 | |
|
4250 | 0 | for(j = 0; num_clusters_evaluated < num_clusters_cur_32x32_blk; j++) |
4251 | 0 | { |
4252 | 0 | ps_cur_cluster = &ps_blk_32x32_cur->as_cluster_data[j]; |
4253 | |
|
4254 | 0 | if(!ps_cur_cluster->is_valid_cluster) |
4255 | 0 | { |
4256 | 0 | continue; |
4257 | 0 | } |
4258 | | |
4259 | 0 | hme_update_64x64_cluster_attributes(ps_blk_64x64, ps_cur_cluster); |
4260 | |
|
4261 | 0 | num_clusters_evaluated++; |
4262 | 0 | } |
4263 | 0 | } |
4264 | 0 | } |
4265 | | |
4266 | | /** |
4267 | | ******************************************************************************** |
4268 | | * @fn void hme_try_merge_clusters_blksize_gt_16 |
4269 | | * ( |
4270 | | * cluster_data_t *ps_cluster_data, |
4271 | | * S32 num_clusters |
4272 | | * ) |
4273 | | * |
4274 | | * @brief Merging clusters from blocks of size 32x32 and greater |
4275 | | * |
4276 | | * @param[in/out] ps_cluster_data: structure containing cluster data |
4277 | | * |
4278 | | * @param[in/out] pi4_num_clusters : pointer to number of clusters |
4279 | | * |
4280 | | * @return Success or failure |
4281 | | ******************************************************************************** |
4282 | | */ |
4283 | | S32 hme_try_merge_clusters_blksize_gt_16(cluster_data_t *ps_cluster_data, S32 num_clusters) |
4284 | 0 | { |
4285 | 0 | centroid_t *ps_cur_centroid; |
4286 | 0 | cluster_data_t *ps_cur_cluster; |
4287 | |
|
4288 | 0 | S32 i, mvd; |
4289 | 0 | S32 mvdx, mvdy, mvdx_q8, mvdy_q8; |
4290 | |
|
4291 | 0 | centroid_t *ps_centroid = &ps_cluster_data->s_centroid; |
4292 | |
|
4293 | 0 | S32 mvd_limit = ps_cluster_data->max_dist_from_centroid; |
4294 | 0 | S32 ref_id = ps_cluster_data->ref_id; |
4295 | |
|
4296 | 0 | S32 node0_posx_q8 = ps_centroid->i4_pos_x_q8; |
4297 | 0 | S32 node0_posy_q8 = ps_centroid->i4_pos_y_q8; |
4298 | 0 | S32 num_clusters_evaluated = 1; |
4299 | 0 | S32 ret_value = 0; |
4300 | |
|
4301 | 0 | if(1 >= num_clusters) |
4302 | 0 | { |
4303 | 0 | return ret_value; |
4304 | 0 | } |
4305 | | |
4306 | 0 | for(i = 1; num_clusters_evaluated < num_clusters; i++) |
4307 | 0 | { |
4308 | 0 | S32 cur_posx_q8; |
4309 | 0 | S32 cur_posy_q8; |
4310 | |
|
4311 | 0 | ps_cur_cluster = &ps_cluster_data[i]; |
4312 | |
|
4313 | 0 | if((ref_id != ps_cur_cluster->ref_id)) |
4314 | 0 | { |
4315 | 0 | num_clusters_evaluated++; |
4316 | 0 | continue; |
4317 | 0 | } |
4318 | | |
4319 | 0 | if((!ps_cur_cluster->is_valid_cluster)) |
4320 | 0 | { |
4321 | 0 | continue; |
4322 | 0 | } |
4323 | | |
4324 | 0 | num_clusters_evaluated++; |
4325 | |
|
4326 | 0 | ps_cur_centroid = &ps_cur_cluster->s_centroid; |
4327 | |
|
4328 | 0 | cur_posx_q8 = ps_cur_centroid->i4_pos_x_q8; |
4329 | 0 | cur_posy_q8 = ps_cur_centroid->i4_pos_y_q8; |
4330 | |
|
4331 | 0 | mvdx_q8 = cur_posx_q8 - node0_posx_q8; |
4332 | 0 | mvdy_q8 = cur_posy_q8 - node0_posy_q8; |
4333 | |
|
4334 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
4335 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
4336 | |
|
4337 | 0 | mvd = ABS(mvdx) + ABS(mvdy); |
4338 | |
|
4339 | 0 | if(mvd <= (mvd_limit >> 1)) |
4340 | 0 | { |
4341 | 0 | LWORD64 i8_updated_posx; |
4342 | 0 | LWORD64 i8_updated_posy; |
4343 | 0 | WORD32 minmax_updated_x = 0; |
4344 | 0 | WORD32 minmax_updated_y = 0; |
4345 | |
|
4346 | 0 | ps_cur_cluster->is_valid_cluster = 0; |
4347 | |
|
4348 | 0 | ps_cluster_data->area_in_pixels += ps_cur_cluster->area_in_pixels; |
4349 | 0 | ps_cluster_data->bi_mv_pixel_area += ps_cur_cluster->bi_mv_pixel_area; |
4350 | 0 | ps_cluster_data->uni_mv_pixel_area += ps_cur_cluster->uni_mv_pixel_area; |
4351 | |
|
4352 | 0 | memcpy( |
4353 | 0 | &ps_cluster_data->as_mv[ps_cluster_data->num_mvs], |
4354 | 0 | ps_cur_cluster->as_mv, |
4355 | 0 | sizeof(mv_data_t) * ps_cur_cluster->num_mvs); |
4356 | |
|
4357 | 0 | if(mvdx > 0) |
4358 | 0 | { |
4359 | 0 | ps_cluster_data->min_x = (cur_posx_q8 + (1 << 7)) >> 8; |
4360 | 0 | minmax_updated_x = 1; |
4361 | 0 | } |
4362 | 0 | else |
4363 | 0 | { |
4364 | 0 | ps_cluster_data->max_x = (cur_posx_q8 + (1 << 7)) >> 8; |
4365 | 0 | minmax_updated_x = 2; |
4366 | 0 | } |
4367 | |
|
4368 | 0 | if(mvdy > 0) |
4369 | 0 | { |
4370 | 0 | ps_cluster_data->min_y = (cur_posy_q8 + (1 << 7)) >> 8; |
4371 | 0 | minmax_updated_y = 1; |
4372 | 0 | } |
4373 | 0 | else |
4374 | 0 | { |
4375 | 0 | ps_cluster_data->max_y = (cur_posy_q8 + (1 << 7)) >> 8; |
4376 | 0 | minmax_updated_y = 2; |
4377 | 0 | } |
4378 | |
|
4379 | 0 | switch((minmax_updated_y << 2) + minmax_updated_x) |
4380 | 0 | { |
4381 | 0 | case 1: |
4382 | 0 | { |
4383 | 0 | S32 mvd, mvd_q8; |
4384 | |
|
4385 | 0 | mvd_q8 = ps_cur_centroid->i4_pos_x_q8 - (ps_cluster_data->min_x << 8); |
4386 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
4387 | |
|
4388 | 0 | if(mvd > (mvd_limit)) |
4389 | 0 | { |
4390 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
4391 | 0 | } |
4392 | 0 | break; |
4393 | 0 | } |
4394 | 0 | case 2: |
4395 | 0 | { |
4396 | 0 | S32 mvd, mvd_q8; |
4397 | |
|
4398 | 0 | mvd_q8 = (ps_cluster_data->max_x << 8) - ps_cur_centroid->i4_pos_x_q8; |
4399 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
4400 | |
|
4401 | 0 | if(mvd > (mvd_limit)) |
4402 | 0 | { |
4403 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
4404 | 0 | } |
4405 | 0 | break; |
4406 | 0 | } |
4407 | 0 | case 4: |
4408 | 0 | { |
4409 | 0 | S32 mvd, mvd_q8; |
4410 | |
|
4411 | 0 | mvd_q8 = ps_cur_centroid->i4_pos_y_q8 - (ps_cluster_data->min_y << 8); |
4412 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
4413 | |
|
4414 | 0 | if(mvd > (mvd_limit)) |
4415 | 0 | { |
4416 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
4417 | 0 | } |
4418 | 0 | break; |
4419 | 0 | } |
4420 | 0 | case 5: |
4421 | 0 | { |
4422 | 0 | S32 mvd; |
4423 | 0 | S32 mvdx, mvdx_q8; |
4424 | 0 | S32 mvdy, mvdy_q8; |
4425 | |
|
4426 | 0 | mvdy_q8 = ps_cur_centroid->i4_pos_y_q8 - (ps_cluster_data->min_y << 8); |
4427 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
4428 | |
|
4429 | 0 | mvdx_q8 = ps_cur_centroid->i4_pos_x_q8 - (ps_cluster_data->min_x << 8); |
4430 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
4431 | |
|
4432 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
4433 | |
|
4434 | 0 | if(mvd > mvd_limit) |
4435 | 0 | { |
4436 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
4437 | 0 | } |
4438 | 0 | break; |
4439 | 0 | } |
4440 | 0 | case 6: |
4441 | 0 | { |
4442 | 0 | S32 mvd; |
4443 | 0 | S32 mvdx, mvdx_q8; |
4444 | 0 | S32 mvdy, mvdy_q8; |
4445 | |
|
4446 | 0 | mvdy_q8 = ps_cur_centroid->i4_pos_y_q8 - (ps_cluster_data->min_y << 8); |
4447 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
4448 | |
|
4449 | 0 | mvdx_q8 = (ps_cluster_data->max_x << 8) - ps_cur_centroid->i4_pos_x_q8; |
4450 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
4451 | |
|
4452 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
4453 | |
|
4454 | 0 | if(mvd > mvd_limit) |
4455 | 0 | { |
4456 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
4457 | 0 | } |
4458 | 0 | break; |
4459 | 0 | } |
4460 | 0 | case 8: |
4461 | 0 | { |
4462 | 0 | S32 mvd, mvd_q8; |
4463 | |
|
4464 | 0 | mvd_q8 = (ps_cluster_data->max_y << 8) - ps_cur_centroid->i4_pos_y_q8; |
4465 | 0 | mvd = (mvd_q8 + (1 << 7)) >> 8; |
4466 | |
|
4467 | 0 | if(mvd > (mvd_limit)) |
4468 | 0 | { |
4469 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
4470 | 0 | } |
4471 | 0 | break; |
4472 | 0 | } |
4473 | 0 | case 9: |
4474 | 0 | { |
4475 | 0 | S32 mvd; |
4476 | 0 | S32 mvdx, mvdx_q8; |
4477 | 0 | S32 mvdy, mvdy_q8; |
4478 | |
|
4479 | 0 | mvdx_q8 = ps_cur_centroid->i4_pos_x_q8 - (ps_cluster_data->min_x << 8); |
4480 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
4481 | |
|
4482 | 0 | mvdy_q8 = (ps_cluster_data->max_y << 8) - ps_cur_centroid->i4_pos_y_q8; |
4483 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
4484 | |
|
4485 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
4486 | |
|
4487 | 0 | if(mvd > mvd_limit) |
4488 | 0 | { |
4489 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
4490 | 0 | } |
4491 | 0 | break; |
4492 | 0 | } |
4493 | 0 | case 10: |
4494 | 0 | { |
4495 | 0 | S32 mvd; |
4496 | 0 | S32 mvdx, mvdx_q8; |
4497 | 0 | S32 mvdy, mvdy_q8; |
4498 | |
|
4499 | 0 | mvdx_q8 = (ps_cluster_data->max_x << 8) - ps_cur_centroid->i4_pos_x_q8; |
4500 | 0 | mvdx = (mvdx_q8 + (1 << 7)) >> 8; |
4501 | |
|
4502 | 0 | mvdy_q8 = (ps_cluster_data->max_y << 8) - ps_cur_centroid->i4_pos_y_q8; |
4503 | 0 | mvdy = (mvdy_q8 + (1 << 7)) >> 8; |
4504 | |
|
4505 | 0 | mvd = (mvdx > mvdy) ? mvdx : mvdy; |
4506 | |
|
4507 | 0 | if(mvd > ps_cluster_data->max_dist_from_centroid) |
4508 | 0 | { |
4509 | 0 | ps_cluster_data->max_dist_from_centroid = mvd; |
4510 | 0 | } |
4511 | 0 | break; |
4512 | 0 | } |
4513 | 0 | default: |
4514 | 0 | { |
4515 | 0 | break; |
4516 | 0 | } |
4517 | 0 | } |
4518 | | |
4519 | 0 | i8_updated_posx = ((LWORD64)ps_centroid->i4_pos_x_q8 * ps_cluster_data->num_mvs) + |
4520 | 0 | ((LWORD64)cur_posx_q8 * ps_cur_cluster->num_mvs); |
4521 | 0 | i8_updated_posy = ((LWORD64)ps_centroid->i4_pos_y_q8 * ps_cluster_data->num_mvs) + |
4522 | 0 | ((LWORD64)cur_posy_q8 * ps_cur_cluster->num_mvs); |
4523 | |
|
4524 | 0 | ps_cluster_data->num_mvs += ps_cur_cluster->num_mvs; |
4525 | |
|
4526 | 0 | ps_centroid->i4_pos_x_q8 = (WORD32)(i8_updated_posx / ps_cluster_data->num_mvs); |
4527 | 0 | ps_centroid->i4_pos_y_q8 = (WORD32)(i8_updated_posy / ps_cluster_data->num_mvs); |
4528 | |
|
4529 | 0 | if(MAX_NUM_CLUSTERS_IN_VALID_64x64_BLK >= num_clusters) |
4530 | 0 | { |
4531 | 0 | num_clusters--; |
4532 | 0 | num_clusters_evaluated = 1; |
4533 | 0 | i = 0; |
4534 | 0 | ret_value++; |
4535 | 0 | } |
4536 | 0 | else |
4537 | 0 | { |
4538 | 0 | ret_value++; |
4539 | |
|
4540 | 0 | return ret_value; |
4541 | 0 | } |
4542 | 0 | } |
4543 | 0 | } |
4544 | | |
4545 | 0 | if(ret_value) |
4546 | 0 | { |
4547 | 0 | for(i = 1; i < (num_clusters + ret_value); i++) |
4548 | 0 | { |
4549 | 0 | if(ps_cluster_data[i].is_valid_cluster) |
4550 | 0 | { |
4551 | 0 | break; |
4552 | 0 | } |
4553 | 0 | } |
4554 | 0 | if(i == (num_clusters + ret_value)) |
4555 | 0 | { |
4556 | 0 | return ret_value; |
4557 | 0 | } |
4558 | 0 | } |
4559 | 0 | else |
4560 | 0 | { |
4561 | 0 | i = 1; |
4562 | 0 | } |
4563 | | |
4564 | 0 | return (hme_try_merge_clusters_blksize_gt_16(&ps_cluster_data[i], num_clusters - 1)) + |
4565 | 0 | ret_value; |
4566 | 0 | } |
4567 | | |
4568 | | /** |
4569 | | ******************************************************************************** |
4570 | | * @fn S32 hme_determine_validity_32x32 |
4571 | | * ( |
4572 | | * ctb_cluster_info_t *ps_ctb_cluster_info |
4573 | | * ) |
4574 | | * |
4575 | | * @brief Determines whther current 32x32 block needs to be evaluated in enc_loop |
4576 | | * while recursing through the CU tree or not |
4577 | | * |
4578 | | * @param[in] ps_cluster_data: structure containing cluster data |
4579 | | * |
4580 | | * @return Success or failure |
4581 | | ******************************************************************************** |
4582 | | */ |
4583 | | __inline S32 hme_determine_validity_32x32( |
4584 | | ctb_cluster_info_t *ps_ctb_cluster_info, |
4585 | | S32 *pi4_children_nodes_required, |
4586 | | S32 blk_validity_wrt_pic_bndry, |
4587 | | S32 parent_blk_validity_wrt_pic_bndry) |
4588 | 0 | { |
4589 | 0 | cluster_data_t *ps_data; |
4590 | 0 |
|
4591 | 0 | cluster_32x32_blk_t *ps_32x32_blk = ps_ctb_cluster_info->ps_32x32_blk; |
4592 | 0 | cluster_64x64_blk_t *ps_64x64_blk = ps_ctb_cluster_info->ps_64x64_blk; |
4593 | 0 |
|
4594 | 0 | S32 num_clusters = ps_32x32_blk->num_clusters; |
4595 | 0 | S32 num_clusters_parent = ps_64x64_blk->num_clusters; |
4596 | 0 |
|
4597 | 0 | if(!blk_validity_wrt_pic_bndry) |
4598 | 0 | { |
4599 | 0 | *pi4_children_nodes_required = 1; |
4600 | 0 | return 0; |
4601 | 0 | } |
4602 | 0 |
|
4603 | 0 | if(!parent_blk_validity_wrt_pic_bndry) |
4604 | 0 | { |
4605 | 0 | *pi4_children_nodes_required = 1; |
4606 | 0 | return 1; |
4607 | 0 | } |
4608 | 0 |
|
4609 | 0 | if(num_clusters > MAX_NUM_CLUSTERS_IN_VALID_32x32_BLK) |
4610 | 0 | { |
4611 | 0 | *pi4_children_nodes_required = 1; |
4612 | 0 | return 0; |
4613 | 0 | } |
4614 | 0 |
|
4615 | 0 | if(num_clusters_parent > MAX_NUM_CLUSTERS_IN_VALID_64x64_BLK) |
4616 | 0 | { |
4617 | 0 | *pi4_children_nodes_required = 1; |
4618 | 0 |
|
4619 | 0 | return 1; |
4620 | 0 | } |
4621 | 0 | else if(num_clusters_parent < MAX_NUM_CLUSTERS_IN_VALID_64x64_BLK) |
4622 | 0 | { |
4623 | 0 | *pi4_children_nodes_required = 0; |
4624 | 0 |
|
4625 | 0 | return 1; |
4626 | 0 | } |
4627 | 0 | else |
4628 | 0 | { |
4629 | 0 | if(num_clusters < MAX_NUM_CLUSTERS_IN_VALID_32x32_BLK) |
4630 | 0 | { |
4631 | 0 | *pi4_children_nodes_required = 0; |
4632 | 0 | return 1; |
4633 | 0 | } |
4634 | 0 | else |
4635 | 0 | { |
4636 | 0 | S32 i; |
4637 | 0 |
|
4638 | 0 | S32 area_of_parent = gai4_partition_area[PART_ID_2Nx2N] << 4; |
4639 | 0 | S32 min_area = MAX_32BIT_VAL; |
4640 | 0 | S32 num_clusters_evaluated = 0; |
4641 | 0 |
|
4642 | 0 | for(i = 0; num_clusters_evaluated < num_clusters; i++) |
4643 | 0 | { |
4644 | 0 | ps_data = &ps_32x32_blk->as_cluster_data[i]; |
4645 | 0 |
|
4646 | 0 | if(!ps_data->is_valid_cluster) |
4647 | 0 | { |
4648 | 0 | continue; |
4649 | 0 | } |
4650 | 0 |
|
4651 | 0 | num_clusters_evaluated++; |
4652 | 0 |
|
4653 | 0 | if(ps_data->area_in_pixels < min_area) |
4654 | 0 | { |
4655 | 0 | min_area = ps_data->area_in_pixels; |
4656 | 0 | } |
4657 | 0 | } |
4658 | 0 |
|
4659 | 0 | if((min_area << 4) < area_of_parent) |
4660 | 0 | { |
4661 | 0 | *pi4_children_nodes_required = 1; |
4662 | 0 | return 0; |
4663 | 0 | } |
4664 | 0 | else |
4665 | 0 | { |
4666 | 0 | *pi4_children_nodes_required = 0; |
4667 | 0 | return 1; |
4668 | 0 | } |
4669 | 0 | } |
4670 | 0 | } |
4671 | 0 | } |
4672 | | |
4673 | | /** |
4674 | | ******************************************************************************** |
4675 | | * @fn S32 hme_determine_validity_16x16 |
4676 | | * ( |
4677 | | * ctb_cluster_info_t *ps_ctb_cluster_info |
4678 | | * ) |
4679 | | * |
4680 | | * @brief Determines whther current 16x16 block needs to be evaluated in enc_loop |
4681 | | * while recursing through the CU tree or not |
4682 | | * |
4683 | | * @param[in] ps_cluster_data: structure containing cluster data |
4684 | | * |
4685 | | * @return Success or failure |
4686 | | ******************************************************************************** |
4687 | | */ |
4688 | | __inline S32 hme_determine_validity_16x16( |
4689 | | ctb_cluster_info_t *ps_ctb_cluster_info, |
4690 | | S32 *pi4_children_nodes_required, |
4691 | | S32 blk_validity_wrt_pic_bndry, |
4692 | | S32 parent_blk_validity_wrt_pic_bndry) |
4693 | 0 | { |
4694 | 0 | cluster_data_t *ps_data; |
4695 | 0 |
|
4696 | 0 | cluster_16x16_blk_t *ps_16x16_blk = ps_ctb_cluster_info->ps_16x16_blk; |
4697 | 0 | cluster_32x32_blk_t *ps_32x32_blk = ps_ctb_cluster_info->ps_32x32_blk; |
4698 | 0 | cluster_64x64_blk_t *ps_64x64_blk = ps_ctb_cluster_info->ps_64x64_blk; |
4699 | 0 |
|
4700 | 0 | S32 num_clusters = ps_16x16_blk->num_clusters; |
4701 | 0 | S32 num_clusters_parent = ps_32x32_blk->num_clusters; |
4702 | 0 | S32 num_clusters_grandparent = ps_64x64_blk->num_clusters; |
4703 | 0 |
|
4704 | 0 | if(!blk_validity_wrt_pic_bndry) |
4705 | 0 | { |
4706 | 0 | *pi4_children_nodes_required = 1; |
4707 | 0 | return 0; |
4708 | 0 | } |
4709 | 0 |
|
4710 | 0 | if(!parent_blk_validity_wrt_pic_bndry) |
4711 | 0 | { |
4712 | 0 | *pi4_children_nodes_required = 1; |
4713 | 0 | return 1; |
4714 | 0 | } |
4715 | 0 |
|
4716 | 0 | if((num_clusters_parent > MAX_NUM_CLUSTERS_IN_VALID_32x32_BLK) && |
4717 | 0 | (num_clusters_grandparent > MAX_NUM_CLUSTERS_IN_VALID_64x64_BLK)) |
4718 | 0 | { |
4719 | 0 | *pi4_children_nodes_required = 1; |
4720 | 0 | return 1; |
4721 | 0 | } |
4722 | 0 |
|
4723 | 0 | /* Implies nc_64 <= 3 when num_clusters_parent > 3 & */ |
4724 | 0 | /* implies nc_64 > 3 when num_clusters_parent < 3 & */ |
4725 | 0 | if(num_clusters_parent != MAX_NUM_CLUSTERS_IN_VALID_32x32_BLK) |
4726 | 0 | { |
4727 | 0 | if(num_clusters <= MAX_NUM_CLUSTERS_IN_VALID_16x16_BLK) |
4728 | 0 | { |
4729 | 0 | *pi4_children_nodes_required = 0; |
4730 | 0 |
|
4731 | 0 | return 1; |
4732 | 0 | } |
4733 | 0 | else |
4734 | 0 | { |
4735 | 0 | *pi4_children_nodes_required = 1; |
4736 | 0 |
|
4737 | 0 | return 0; |
4738 | 0 | } |
4739 | 0 | } |
4740 | 0 | /* Implies nc_64 >= 3 */ |
4741 | 0 | else |
4742 | 0 | { |
4743 | 0 | if(num_clusters < MAX_NUM_CLUSTERS_IN_VALID_16x16_BLK) |
4744 | 0 | { |
4745 | 0 | *pi4_children_nodes_required = 0; |
4746 | 0 | return 1; |
4747 | 0 | } |
4748 | 0 | else if(num_clusters > MAX_NUM_CLUSTERS_IN_VALID_16x16_BLK) |
4749 | 0 | { |
4750 | 0 | *pi4_children_nodes_required = 1; |
4751 | 0 | return 0; |
4752 | 0 | } |
4753 | 0 | else |
4754 | 0 | { |
4755 | 0 | S32 i; |
4756 | 0 |
|
4757 | 0 | S32 area_of_parent = gai4_partition_area[PART_ID_2Nx2N] << 2; |
4758 | 0 | S32 min_area = MAX_32BIT_VAL; |
4759 | 0 | S32 num_clusters_evaluated = 0; |
4760 | 0 |
|
4761 | 0 | for(i = 0; num_clusters_evaluated < num_clusters; i++) |
4762 | 0 | { |
4763 | 0 | ps_data = &ps_16x16_blk->as_cluster_data[i]; |
4764 | 0 |
|
4765 | 0 | if(!ps_data->is_valid_cluster) |
4766 | 0 | { |
4767 | 0 | continue; |
4768 | 0 | } |
4769 | 0 |
|
4770 | 0 | num_clusters_evaluated++; |
4771 | 0 |
|
4772 | 0 | if(ps_data->area_in_pixels < min_area) |
4773 | 0 | { |
4774 | 0 | min_area = ps_data->area_in_pixels; |
4775 | 0 | } |
4776 | 0 | } |
4777 | 0 |
|
4778 | 0 | if((min_area << 4) < area_of_parent) |
4779 | 0 | { |
4780 | 0 | *pi4_children_nodes_required = 1; |
4781 | 0 | return 0; |
4782 | 0 | } |
4783 | 0 | else |
4784 | 0 | { |
4785 | 0 | *pi4_children_nodes_required = 0; |
4786 | 0 | return 1; |
4787 | 0 | } |
4788 | 0 | } |
4789 | 0 | } |
4790 | 0 | } |
4791 | | |
4792 | | /** |
4793 | | ******************************************************************************** |
4794 | | * @fn void hme_build_cu_tree |
4795 | | * ( |
4796 | | * ctb_cluster_info_t *ps_ctb_cluster_info, |
4797 | | * cur_ctb_cu_tree_t *ps_cu_tree, |
4798 | | * S32 tree_depth, |
4799 | | * CU_POS_T e_grand_parent_blk_pos, |
4800 | | * CU_POS_T e_parent_blk_pos, |
4801 | | * CU_POS_T e_cur_blk_pos |
4802 | | * ) |
4803 | | * |
4804 | | * @brief Recursive function for CU tree initialisation |
4805 | | * |
4806 | | * @param[in] ps_ctb_cluster_info: structure containing pointers to clusters |
4807 | | * corresponding to all block sizes from 64x64 |
4808 | | * to 16x16 |
4809 | | * |
4810 | | * @param[in] e_parent_blk_pos: position of parent block wrt its parent, if |
4811 | | * applicable |
4812 | | * |
4813 | | * @param[in] e_cur_blk_pos: position of current block wrt parent |
4814 | | * |
4815 | | * @param[out] ps_cu_tree : represents CU tree used in CU recursion |
4816 | | * |
4817 | | * @param[in] tree_depth : specifies depth of the CU tree |
4818 | | * |
4819 | | * @return Nothing |
4820 | | ******************************************************************************** |
4821 | | */ |
4822 | | void hme_build_cu_tree( |
4823 | | ctb_cluster_info_t *ps_ctb_cluster_info, |
4824 | | cur_ctb_cu_tree_t *ps_cu_tree, |
4825 | | S32 tree_depth, |
4826 | | CU_POS_T e_grandparent_blk_pos, |
4827 | | CU_POS_T e_parent_blk_pos, |
4828 | | CU_POS_T e_cur_blk_pos) |
4829 | 0 | { |
4830 | 0 | ihevce_cu_tree_init( |
4831 | 0 | ps_cu_tree, |
4832 | 0 | ps_ctb_cluster_info->ps_cu_tree_root, |
4833 | 0 | &ps_ctb_cluster_info->nodes_created_in_cu_tree, |
4834 | 0 | tree_depth, |
4835 | 0 | e_grandparent_blk_pos, |
4836 | 0 | e_parent_blk_pos, |
4837 | 0 | e_cur_blk_pos); |
4838 | 0 | } |
4839 | | |
4840 | | /** |
4841 | | ******************************************************************************** |
4842 | | * @fn S32 hme_sdi_based_cluster_spread_eligibility |
4843 | | * ( |
4844 | | * cluster_32x32_blk_t *ps_blk_32x32 |
4845 | | * ) |
4846 | | * |
4847 | | * @brief Determines whether the spread of high SDI MV's around each cluster |
4848 | | * center is below a pre-determined threshold |
4849 | | * |
4850 | | * @param[in] ps_blk_32x32: structure containing pointers to clusters |
4851 | | * corresponding to all block sizes from 64x64 |
4852 | | * to 16x16 |
4853 | | * |
4854 | | * @return 1 if the spread is constrained, else 0 |
4855 | | ******************************************************************************** |
4856 | | */ |
4857 | | __inline S32 |
4858 | | hme_sdi_based_cluster_spread_eligibility(cluster_32x32_blk_t *ps_blk_32x32, S32 sdi_threshold) |
4859 | 0 | { |
4860 | 0 | S32 cumulative_mv_distance; |
4861 | 0 | S32 i, j; |
4862 | 0 | S32 num_high_sdi_mvs; |
4863 | 0 |
|
4864 | 0 | S32 num_clusters = ps_blk_32x32->num_clusters; |
4865 | 0 |
|
4866 | 0 | for(i = 0; i < num_clusters; i++) |
4867 | 0 | { |
4868 | 0 | cluster_data_t *ps_data = &ps_blk_32x32->as_cluster_data[i]; |
4869 | 0 |
|
4870 | 0 | num_high_sdi_mvs = 0; |
4871 | 0 | cumulative_mv_distance = 0; |
4872 | 0 |
|
4873 | 0 | for(j = 0; j < ps_data->num_mvs; j++) |
4874 | 0 | { |
4875 | 0 | mv_data_t *ps_mv = &ps_data->as_mv[j]; |
4876 | 0 |
|
4877 | 0 | if(ps_mv->sdi >= sdi_threshold) |
4878 | 0 | { |
4879 | 0 | num_high_sdi_mvs++; |
4880 | 0 |
|
4881 | 0 | COMPUTE_MVD(ps_mv, ps_data, cumulative_mv_distance); |
4882 | 0 | } |
4883 | 0 | } |
4884 | 0 |
|
4885 | 0 | if(cumulative_mv_distance > ((ps_data->max_dist_from_centroid >> 1) * num_high_sdi_mvs)) |
4886 | 0 | { |
4887 | 0 | return 0; |
4888 | 0 | } |
4889 | 0 | } |
4890 | 0 |
|
4891 | 0 | return 1; |
4892 | 0 | } |
4893 | | |
4894 | | /** |
4895 | | ******************************************************************************** |
4896 | | * @fn S32 hme_populate_cu_tree |
4897 | | * ( |
4898 | | * ctb_cluster_info_t *ps_ctb_cluster_info, |
4899 | | * ipe_l0_ctb_analyse_for_me_t *ps_cur_ipe_ctb, |
4900 | | * cur_ctb_cu_tree_t *ps_cu_tree, |
4901 | | * S32 tree_depth, |
4902 | | * CU_POS_T e_parent_blk_pos, |
4903 | | * CU_POS_T e_cur_blk_pos |
4904 | | * ) |
4905 | | * |
4906 | | * @brief Recursive function for CU tree population based on output of |
4907 | | * clustering algorithm |
4908 | | * |
4909 | | * @param[in] ps_ctb_cluster_info: structure containing pointers to clusters |
4910 | | * corresponding to all block sizes from 64x64 |
4911 | | * to 16x16 |
4912 | | * |
4913 | | * @param[in] e_parent_blk_pos: position of parent block wrt its parent, if |
4914 | | applicable |
4915 | | * |
4916 | | * @param[in] e_cur_blk_pos: position of current block wrt parent |
4917 | | * |
4918 | | * @param[in] ps_cur_ipe_ctb : output container for ipe analyses |
4919 | | * |
4920 | | * @param[out] ps_cu_tree : represents CU tree used in CU recursion |
4921 | | * |
4922 | | * @param[in] tree_depth : specifies depth of the CU tree |
4923 | | * |
4924 | | * @param[in] ipe_decision_precedence : specifies whether precedence should |
4925 | | * be given to decisions made either by IPE(1) or clustering algos. |
4926 | | * |
4927 | | * @return 1 if re-evaluation of parent node's validity is not required, |
4928 | | else 0 |
4929 | | ******************************************************************************** |
4930 | | */ |
4931 | | void hme_populate_cu_tree( |
4932 | | ctb_cluster_info_t *ps_ctb_cluster_info, |
4933 | | cur_ctb_cu_tree_t *ps_cu_tree, |
4934 | | S32 tree_depth, |
4935 | | ME_QUALITY_PRESETS_T e_quality_preset, |
4936 | | CU_POS_T e_grandparent_blk_pos, |
4937 | | CU_POS_T e_parent_blk_pos, |
4938 | | CU_POS_T e_cur_blk_pos) |
4939 | 0 | { |
4940 | 0 | S32 area_of_cur_blk; |
4941 | 0 | S32 area_limit_for_me_decision_precedence; |
4942 | 0 | S32 children_nodes_required; |
4943 | 0 | S32 intra_mv_area; |
4944 | 0 | S32 intra_eval_enable; |
4945 | 0 | S32 inter_eval_enable; |
4946 | 0 | S32 ipe_decision_precedence; |
4947 | 0 | S32 node_validity; |
4948 | 0 | S32 num_clusters; |
4949 | |
|
4950 | 0 | ipe_l0_ctb_analyse_for_me_t *ps_cur_ipe_ctb = ps_ctb_cluster_info->ps_cur_ipe_ctb; |
4951 | |
|
4952 | 0 | if(NULL == ps_cu_tree) |
4953 | 0 | { |
4954 | 0 | return; |
4955 | 0 | } |
4956 | | |
4957 | 0 | switch(tree_depth) |
4958 | 0 | { |
4959 | 0 | case 0: |
4960 | 0 | { |
4961 | | /* 64x64 block */ |
4962 | 0 | S32 blk_32x32_mask = ps_ctb_cluster_info->blk_32x32_mask; |
4963 | |
|
4964 | 0 | cluster_64x64_blk_t *ps_blk_64x64 = ps_ctb_cluster_info->ps_64x64_blk; |
4965 | |
|
4966 | 0 | area_of_cur_blk = gai4_partition_area[PART_ID_2Nx2N] << 4; |
4967 | 0 | area_limit_for_me_decision_precedence = (area_of_cur_blk * MAX_INTRA_PERCENTAGE) / 100; |
4968 | 0 | children_nodes_required = 0; |
4969 | 0 | intra_mv_area = ps_blk_64x64->intra_mv_area; |
4970 | |
|
4971 | 0 | ipe_decision_precedence = (intra_mv_area >= area_limit_for_me_decision_precedence); |
4972 | |
|
4973 | 0 | intra_eval_enable = ipe_decision_precedence; |
4974 | 0 | inter_eval_enable = !!ps_blk_64x64->num_clusters; |
4975 | |
|
4976 | 0 | #if 1 //!PROCESS_GT_1CTB_VIA_CU_RECUR_IN_FAST_PRESETS |
4977 | 0 | if(e_quality_preset >= ME_HIGH_QUALITY) |
4978 | 0 | { |
4979 | 0 | inter_eval_enable = 1; |
4980 | 0 | node_validity = (blk_32x32_mask == 0xf); |
4981 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
4982 | 0 | ps_cu_tree->u1_inter_eval_enable = !(intra_mv_area == area_of_cur_blk); |
4983 | 0 | #endif |
4984 | 0 | break; |
4985 | 0 | } |
4986 | 0 | #endif |
4987 | | |
4988 | 0 | #if ENABLE_4CTB_EVALUATION |
4989 | 0 | node_validity = (blk_32x32_mask == 0xf); |
4990 | |
|
4991 | 0 | break; |
4992 | | #else |
4993 | | { |
4994 | | S32 i; |
4995 | | |
4996 | | num_clusters = ps_blk_64x64->num_clusters; |
4997 | | |
4998 | | node_validity = (ipe_decision_precedence) |
4999 | | ? (!ps_cur_ipe_ctb->u1_split_flag) |
5000 | | : (num_clusters <= MAX_NUM_CLUSTERS_IN_VALID_64x64_BLK); |
5001 | | |
5002 | | for(i = 0; i < MAX_NUM_REF; i++) |
5003 | | { |
5004 | | node_validity = node_validity && (ps_blk_64x64->au1_num_clusters[i] <= |
5005 | | MAX_NUM_CLUSTERS_IN_ONE_REF_IDX); |
5006 | | } |
5007 | | |
5008 | | node_validity = node_validity && (blk_32x32_mask == 0xf); |
5009 | | } |
5010 | | break; |
5011 | | #endif |
5012 | 0 | } |
5013 | 0 | case 1: |
5014 | 0 | { |
5015 | | /* 32x32 block */ |
5016 | 0 | S32 is_percent_intra_area_gt_threshold; |
5017 | |
|
5018 | 0 | cluster_32x32_blk_t *ps_blk_32x32 = &ps_ctb_cluster_info->ps_32x32_blk[e_cur_blk_pos]; |
5019 | |
|
5020 | 0 | S32 blk_32x32_mask = ps_ctb_cluster_info->blk_32x32_mask; |
5021 | |
|
5022 | | #if !ENABLE_4CTB_EVALUATION |
5023 | | S32 best_inter_cost = ps_blk_32x32->best_inter_cost; |
5024 | | S32 best_intra_cost = |
5025 | | ((ps_ctb_cluster_info->ps_cur_ipe_ctb->ai4_best32x32_intra_cost[e_cur_blk_pos] + |
5026 | | ps_ctb_cluster_info->i4_frame_qstep * ps_ctb_cluster_info->i4_frame_qstep_multiplier * |
5027 | | 4) < 0) |
5028 | | ? MAX_32BIT_VAL |
5029 | | : (ps_ctb_cluster_info->ps_cur_ipe_ctb->ai4_best32x32_intra_cost[e_cur_blk_pos] + |
5030 | | ps_ctb_cluster_info->i4_frame_qstep * |
5031 | | ps_ctb_cluster_info->i4_frame_qstep_multiplier * 4); |
5032 | | S32 best_cost = (best_inter_cost > best_intra_cost) ? best_intra_cost : best_inter_cost; |
5033 | | S32 cost_differential = (best_inter_cost - best_cost); |
5034 | | #endif |
5035 | |
|
5036 | 0 | area_of_cur_blk = gai4_partition_area[PART_ID_2Nx2N] << 2; |
5037 | 0 | area_limit_for_me_decision_precedence = (area_of_cur_blk * MAX_INTRA_PERCENTAGE) / 100; |
5038 | 0 | intra_mv_area = ps_blk_32x32->intra_mv_area; |
5039 | 0 | is_percent_intra_area_gt_threshold = |
5040 | 0 | (intra_mv_area > area_limit_for_me_decision_precedence); |
5041 | 0 | ipe_decision_precedence = (intra_mv_area >= area_limit_for_me_decision_precedence); |
5042 | |
|
5043 | 0 | intra_eval_enable = ipe_decision_precedence; |
5044 | 0 | inter_eval_enable = !!ps_blk_32x32->num_clusters; |
5045 | 0 | children_nodes_required = 1; |
5046 | |
|
5047 | 0 | #if 1 //!PROCESS_GT_1CTB_VIA_CU_RECUR_IN_FAST_PRESETS |
5048 | 0 | if(e_quality_preset >= ME_HIGH_QUALITY) |
5049 | 0 | { |
5050 | 0 | inter_eval_enable = 1; |
5051 | 0 | node_validity = (((blk_32x32_mask) & (1 << e_cur_blk_pos)) || 0); |
5052 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
5053 | 0 | ps_cu_tree->u1_inter_eval_enable = !(intra_mv_area == area_of_cur_blk); |
5054 | 0 | #endif |
5055 | 0 | break; |
5056 | 0 | } |
5057 | 0 | #endif |
5058 | | |
5059 | 0 | #if ENABLE_4CTB_EVALUATION |
5060 | 0 | node_validity = (((blk_32x32_mask) & (1 << e_cur_blk_pos)) || 0); |
5061 | |
|
5062 | 0 | break; |
5063 | | #else |
5064 | | { |
5065 | | S32 i; |
5066 | | num_clusters = ps_blk_32x32->num_clusters; |
5067 | | |
5068 | | if(ipe_decision_precedence) |
5069 | | { |
5070 | | node_validity = (ps_cur_ipe_ctb->as_intra32_analyse[e_cur_blk_pos].b1_merge_flag); |
5071 | | node_validity = node_validity && (((blk_32x32_mask) & (1 << e_cur_blk_pos)) || 0); |
5072 | | } |
5073 | | else |
5074 | | { |
5075 | | node_validity = |
5076 | | ((ALL_INTER_COST_DIFF_THR * best_cost) >= (100 * cost_differential)) && |
5077 | | (num_clusters <= MAX_NUM_CLUSTERS_IN_VALID_32x32_BLK) && |
5078 | | (((blk_32x32_mask) & (1 << e_cur_blk_pos)) || 0); |
5079 | | |
5080 | | for(i = 0; (i < MAX_NUM_REF) && (node_validity); i++) |
5081 | | { |
5082 | | node_validity = node_validity && (ps_blk_32x32->au1_num_clusters[i] <= |
5083 | | MAX_NUM_CLUSTERS_IN_ONE_REF_IDX); |
5084 | | } |
5085 | | |
5086 | | if(node_validity) |
5087 | | { |
5088 | | node_validity = node_validity && |
5089 | | hme_sdi_based_cluster_spread_eligibility( |
5090 | | ps_blk_32x32, ps_ctb_cluster_info->sdi_threshold); |
5091 | | } |
5092 | | } |
5093 | | } |
5094 | | |
5095 | | break; |
5096 | | #endif |
5097 | 0 | } |
5098 | 0 | case 2: |
5099 | 0 | { |
5100 | 0 | cluster_16x16_blk_t *ps_blk_16x16 = |
5101 | 0 | &ps_ctb_cluster_info->ps_16x16_blk[e_cur_blk_pos + (e_parent_blk_pos << 2)]; |
5102 | |
|
5103 | 0 | S32 blk_8x8_mask = |
5104 | 0 | ps_ctb_cluster_info->pi4_blk_8x8_mask[(S32)(e_parent_blk_pos << 2) + e_cur_blk_pos]; |
5105 | |
|
5106 | 0 | area_of_cur_blk = gai4_partition_area[PART_ID_2Nx2N]; |
5107 | 0 | area_limit_for_me_decision_precedence = (area_of_cur_blk * MAX_INTRA_PERCENTAGE) / 100; |
5108 | 0 | children_nodes_required = 1; |
5109 | 0 | intra_mv_area = ps_blk_16x16->intra_mv_area; |
5110 | 0 | ipe_decision_precedence = (intra_mv_area >= area_limit_for_me_decision_precedence); |
5111 | 0 | num_clusters = ps_blk_16x16->num_clusters; |
5112 | |
|
5113 | 0 | intra_eval_enable = ipe_decision_precedence; |
5114 | 0 | inter_eval_enable = 1; |
5115 | |
|
5116 | 0 | #if 1 //!PROCESS_GT_1CTB_VIA_CU_RECUR_IN_FAST_PRESETS |
5117 | 0 | if(e_quality_preset >= ME_HIGH_QUALITY) |
5118 | 0 | { |
5119 | 0 | node_validity = |
5120 | 0 | !ps_ctb_cluster_info |
5121 | 0 | ->au1_is_16x16_blk_split[(S32)(e_parent_blk_pos << 2) + e_cur_blk_pos]; |
5122 | 0 | children_nodes_required = !node_validity; |
5123 | 0 | break; |
5124 | 0 | } |
5125 | 0 | #endif |
5126 | | |
5127 | 0 | #if ENABLE_4CTB_EVALUATION |
5128 | 0 | node_validity = (blk_8x8_mask == 0xf); |
5129 | |
|
5130 | 0 | #if ENABLE_CU_TREE_CULLING |
5131 | 0 | { |
5132 | 0 | cur_ctb_cu_tree_t *ps_32x32_root = NULL; |
5133 | |
|
5134 | 0 | switch(e_parent_blk_pos) |
5135 | 0 | { |
5136 | 0 | case POS_TL: |
5137 | 0 | { |
5138 | 0 | ps_32x32_root = ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_tl; |
5139 | |
|
5140 | 0 | break; |
5141 | 0 | } |
5142 | 0 | case POS_TR: |
5143 | 0 | { |
5144 | 0 | ps_32x32_root = ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_tr; |
5145 | |
|
5146 | 0 | break; |
5147 | 0 | } |
5148 | 0 | case POS_BL: |
5149 | 0 | { |
5150 | 0 | ps_32x32_root = ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_bl; |
5151 | |
|
5152 | 0 | break; |
5153 | 0 | } |
5154 | 0 | case POS_BR: |
5155 | 0 | { |
5156 | 0 | ps_32x32_root = ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_br; |
5157 | |
|
5158 | 0 | break; |
5159 | 0 | } |
5160 | 0 | default: |
5161 | 0 | { |
5162 | 0 | DBG_PRINTF("Invalid block position %d\n", e_parent_blk_pos); |
5163 | 0 | break; |
5164 | 0 | } |
5165 | 0 | } |
5166 | | |
5167 | 0 | if(ps_32x32_root->is_node_valid) |
5168 | 0 | { |
5169 | 0 | node_validity = |
5170 | 0 | node_validity && |
5171 | 0 | !ps_ctb_cluster_info |
5172 | 0 | ->au1_is_16x16_blk_split[(S32)(e_parent_blk_pos << 2) + e_cur_blk_pos]; |
5173 | 0 | children_nodes_required = !node_validity; |
5174 | 0 | } |
5175 | 0 | } |
5176 | 0 | #endif |
5177 | | |
5178 | 0 | break; |
5179 | | #else |
5180 | | |
5181 | | if(ipe_decision_precedence) |
5182 | | { |
5183 | | S32 merge_flag_16 = (ps_cur_ipe_ctb->as_intra32_analyse[e_parent_blk_pos] |
5184 | | .as_intra16_analyse[e_cur_blk_pos] |
5185 | | .b1_merge_flag); |
5186 | | S32 valid_flag = (blk_8x8_mask == 0xf); |
5187 | | |
5188 | | node_validity = merge_flag_16 && valid_flag; |
5189 | | } |
5190 | | else |
5191 | | { |
5192 | | node_validity = (blk_8x8_mask == 0xf); |
5193 | | } |
5194 | | |
5195 | | break; |
5196 | | #endif |
5197 | 0 | } |
5198 | 0 | case 3: |
5199 | 0 | { |
5200 | 0 | S32 blk_8x8_mask = |
5201 | 0 | ps_ctb_cluster_info |
5202 | 0 | ->pi4_blk_8x8_mask[(S32)(e_grandparent_blk_pos << 2) + e_parent_blk_pos]; |
5203 | 0 | S32 merge_flag_16 = (ps_cur_ipe_ctb->as_intra32_analyse[e_grandparent_blk_pos] |
5204 | 0 | .as_intra16_analyse[e_parent_blk_pos] |
5205 | 0 | .b1_merge_flag); |
5206 | 0 | S32 merge_flag_32 = |
5207 | 0 | (ps_cur_ipe_ctb->as_intra32_analyse[e_grandparent_blk_pos].b1_merge_flag); |
5208 | |
|
5209 | 0 | intra_eval_enable = !merge_flag_16 || !merge_flag_32; |
5210 | 0 | inter_eval_enable = 1; |
5211 | 0 | children_nodes_required = 0; |
5212 | |
|
5213 | 0 | #if 1 //!PROCESS_GT_1CTB_VIA_CU_RECUR_IN_FAST_PRESETS |
5214 | 0 | if(e_quality_preset >= ME_HIGH_QUALITY) |
5215 | 0 | { |
5216 | 0 | node_validity = ((blk_8x8_mask & (1 << e_cur_blk_pos)) || 0); |
5217 | 0 | break; |
5218 | 0 | } |
5219 | 0 | #endif |
5220 | | |
5221 | 0 | #if ENABLE_4CTB_EVALUATION |
5222 | 0 | node_validity = ((blk_8x8_mask & (1 << e_cur_blk_pos)) || 0); |
5223 | |
|
5224 | 0 | break; |
5225 | | #else |
5226 | | { |
5227 | | cur_ctb_cu_tree_t *ps_32x32_root; |
5228 | | cur_ctb_cu_tree_t *ps_16x16_root; |
5229 | | cluster_32x32_blk_t *ps_32x32_blk; |
5230 | | |
5231 | | switch(e_grandparent_blk_pos) |
5232 | | { |
5233 | | case POS_TL: |
5234 | | { |
5235 | | ps_32x32_root = ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_tl; |
5236 | | |
5237 | | break; |
5238 | | } |
5239 | | case POS_TR: |
5240 | | { |
5241 | | ps_32x32_root = ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_tr; |
5242 | | |
5243 | | break; |
5244 | | } |
5245 | | case POS_BL: |
5246 | | { |
5247 | | ps_32x32_root = ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_bl; |
5248 | | |
5249 | | break; |
5250 | | } |
5251 | | case POS_BR: |
5252 | | { |
5253 | | ps_32x32_root = ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_br; |
5254 | | |
5255 | | break; |
5256 | | } |
5257 | | default: |
5258 | | { |
5259 | | DBG_PRINTF("Invalid block position %d\n", e_grandparent_blk_pos); |
5260 | | break; |
5261 | | } |
5262 | | } |
5263 | | |
5264 | | switch(e_parent_blk_pos) |
5265 | | { |
5266 | | case POS_TL: |
5267 | | { |
5268 | | ps_16x16_root = ps_32x32_root->ps_child_node_tl; |
5269 | | |
5270 | | break; |
5271 | | } |
5272 | | case POS_TR: |
5273 | | { |
5274 | | ps_16x16_root = ps_32x32_root->ps_child_node_tr; |
5275 | | |
5276 | | break; |
5277 | | } |
5278 | | case POS_BL: |
5279 | | { |
5280 | | ps_16x16_root = ps_32x32_root->ps_child_node_bl; |
5281 | | |
5282 | | break; |
5283 | | } |
5284 | | case POS_BR: |
5285 | | { |
5286 | | ps_16x16_root = ps_32x32_root->ps_child_node_br; |
5287 | | |
5288 | | break; |
5289 | | } |
5290 | | default: |
5291 | | { |
5292 | | DBG_PRINTF("Invalid block position %d\n", e_parent_blk_pos); |
5293 | | break; |
5294 | | } |
5295 | | } |
5296 | | |
5297 | | ps_32x32_blk = &ps_ctb_cluster_info->ps_32x32_blk[e_grandparent_blk_pos]; |
5298 | | |
5299 | | node_validity = ((blk_8x8_mask & (1 << e_cur_blk_pos)) || 0) && |
5300 | | ((!ps_32x32_root->is_node_valid) || |
5301 | | (ps_32x32_blk->num_clusters_with_weak_sdi_density > 0) || |
5302 | | (!ps_16x16_root->is_node_valid)); |
5303 | | |
5304 | | break; |
5305 | | } |
5306 | | #endif |
5307 | 0 | } |
5308 | 0 | } |
5309 | | |
5310 | | /* Fill the current cu_tree node */ |
5311 | 0 | ps_cu_tree->is_node_valid = node_validity; |
5312 | 0 | ps_cu_tree->u1_intra_eval_enable = intra_eval_enable; |
5313 | 0 | ps_cu_tree->u1_inter_eval_enable = inter_eval_enable; |
5314 | |
|
5315 | 0 | if(children_nodes_required) |
5316 | 0 | { |
5317 | 0 | tree_depth++; |
5318 | |
|
5319 | 0 | hme_populate_cu_tree( |
5320 | 0 | ps_ctb_cluster_info, |
5321 | 0 | ps_cu_tree->ps_child_node_tl, |
5322 | 0 | tree_depth, |
5323 | 0 | e_quality_preset, |
5324 | 0 | e_parent_blk_pos, |
5325 | 0 | e_cur_blk_pos, |
5326 | 0 | POS_TL); |
5327 | |
|
5328 | 0 | hme_populate_cu_tree( |
5329 | 0 | ps_ctb_cluster_info, |
5330 | 0 | ps_cu_tree->ps_child_node_tr, |
5331 | 0 | tree_depth, |
5332 | 0 | e_quality_preset, |
5333 | 0 | e_parent_blk_pos, |
5334 | 0 | e_cur_blk_pos, |
5335 | 0 | POS_TR); |
5336 | |
|
5337 | 0 | hme_populate_cu_tree( |
5338 | 0 | ps_ctb_cluster_info, |
5339 | 0 | ps_cu_tree->ps_child_node_bl, |
5340 | 0 | tree_depth, |
5341 | 0 | e_quality_preset, |
5342 | 0 | e_parent_blk_pos, |
5343 | 0 | e_cur_blk_pos, |
5344 | 0 | POS_BL); |
5345 | |
|
5346 | 0 | hme_populate_cu_tree( |
5347 | 0 | ps_ctb_cluster_info, |
5348 | 0 | ps_cu_tree->ps_child_node_br, |
5349 | 0 | tree_depth, |
5350 | 0 | e_quality_preset, |
5351 | 0 | e_parent_blk_pos, |
5352 | 0 | e_cur_blk_pos, |
5353 | 0 | POS_BR); |
5354 | 0 | } |
5355 | 0 | } |
5356 | | |
5357 | | /** |
5358 | | ******************************************************************************** |
5359 | | * @fn void hme_analyse_mv_clustering |
5360 | | * ( |
5361 | | * search_results_t *ps_search_results, |
5362 | | * ipe_l0_ctb_analyse_for_me_t *ps_cur_ipe_ctb, |
5363 | | * cur_ctb_cu_tree_t *ps_cu_tree |
5364 | | * ) |
5365 | | * |
5366 | | * @brief Implementation for the clustering algorithm |
5367 | | * |
5368 | | * @param[in] ps_search_results: structure containing 16x16 block results |
5369 | | * |
5370 | | * @param[in] ps_cur_ipe_ctb : output container for ipe analyses |
5371 | | * |
5372 | | * @param[out] ps_cu_tree : represents CU tree used in CU recursion |
5373 | | * |
5374 | | * @return None |
5375 | | ******************************************************************************** |
5376 | | */ |
5377 | | void hme_analyse_mv_clustering( |
5378 | | search_results_t *ps_search_results, |
5379 | | inter_cu_results_t *ps_16x16_cu_results, |
5380 | | inter_cu_results_t *ps_8x8_cu_results, |
5381 | | ctb_cluster_info_t *ps_ctb_cluster_info, |
5382 | | S08 *pi1_future_list, |
5383 | | S08 *pi1_past_list, |
5384 | | S32 bidir_enabled, |
5385 | | ME_QUALITY_PRESETS_T e_quality_preset) |
5386 | 0 | { |
5387 | 0 | cluster_16x16_blk_t *ps_blk_16x16; |
5388 | 0 | cluster_32x32_blk_t *ps_blk_32x32; |
5389 | 0 | cluster_64x64_blk_t *ps_blk_64x64; |
5390 | |
|
5391 | 0 | part_type_results_t *ps_best_result; |
5392 | 0 | pu_result_t *aps_part_result[MAX_NUM_PARTS]; |
5393 | 0 | pu_result_t *aps_inferior_parts[MAX_NUM_PARTS]; |
5394 | |
|
5395 | 0 | PART_ID_T e_part_id; |
5396 | 0 | PART_TYPE_T e_part_type; |
5397 | |
|
5398 | 0 | S32 enable_64x64_merge; |
5399 | 0 | S32 i, j, k; |
5400 | 0 | S32 mvx, mvy; |
5401 | 0 | S32 num_parts; |
5402 | 0 | S32 ref_idx; |
5403 | 0 | S32 ai4_pred_mode[MAX_NUM_PARTS]; |
5404 | |
|
5405 | 0 | S32 num_32x32_merges = 0; |
5406 | | |
5407 | | /*****************************************/ |
5408 | | /*****************************************/ |
5409 | | /********* Enter ye who is HQ ************/ |
5410 | | /*****************************************/ |
5411 | | /*****************************************/ |
5412 | |
|
5413 | 0 | ps_blk_64x64 = ps_ctb_cluster_info->ps_64x64_blk; |
5414 | | |
5415 | | /* Initialise data in each of the clusters */ |
5416 | 0 | for(i = 0; i < 16; i++) |
5417 | 0 | { |
5418 | 0 | ps_blk_16x16 = &ps_ctb_cluster_info->ps_16x16_blk[i]; |
5419 | |
|
5420 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
5421 | 0 | if(e_quality_preset < ME_HIGH_QUALITY) |
5422 | 0 | { |
5423 | 0 | hme_init_clusters_16x16(ps_blk_16x16, bidir_enabled); |
5424 | 0 | } |
5425 | 0 | else |
5426 | 0 | { |
5427 | 0 | ps_blk_16x16->best_inter_cost = 0; |
5428 | 0 | ps_blk_16x16->intra_mv_area = 0; |
5429 | 0 | } |
5430 | | #else |
5431 | | hme_init_clusters_16x16(ps_blk_16x16, bidir_enabled); |
5432 | | #endif |
5433 | 0 | } |
5434 | |
|
5435 | 0 | for(i = 0; i < 4; i++) |
5436 | 0 | { |
5437 | 0 | ps_blk_32x32 = &ps_ctb_cluster_info->ps_32x32_blk[i]; |
5438 | |
|
5439 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
5440 | 0 | if(e_quality_preset < ME_HIGH_QUALITY) |
5441 | 0 | { |
5442 | 0 | hme_init_clusters_32x32(ps_blk_32x32, bidir_enabled); |
5443 | 0 | } |
5444 | 0 | else |
5445 | 0 | { |
5446 | 0 | ps_blk_32x32->best_inter_cost = 0; |
5447 | 0 | ps_blk_32x32->intra_mv_area = 0; |
5448 | 0 | } |
5449 | | #else |
5450 | | hme_init_clusters_32x32(ps_blk_32x32, bidir_enabled); |
5451 | | #endif |
5452 | 0 | } |
5453 | |
|
5454 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
5455 | 0 | if(e_quality_preset < ME_HIGH_QUALITY) |
5456 | 0 | { |
5457 | 0 | hme_init_clusters_64x64(ps_blk_64x64, bidir_enabled); |
5458 | 0 | } |
5459 | 0 | else |
5460 | 0 | { |
5461 | 0 | ps_blk_64x64->best_inter_cost = 0; |
5462 | 0 | ps_blk_64x64->intra_mv_area = 0; |
5463 | 0 | } |
5464 | | #else |
5465 | | hme_init_clusters_64x64(ps_blk_64x64, bidir_enabled); |
5466 | | #endif |
5467 | | |
5468 | | /* Initialise data for all nodes in the CU tree */ |
5469 | 0 | hme_build_cu_tree( |
5470 | 0 | ps_ctb_cluster_info, ps_ctb_cluster_info->ps_cu_tree_root, 0, POS_NA, POS_NA, POS_NA); |
5471 | |
|
5472 | 0 | if(e_quality_preset >= ME_HIGH_QUALITY) |
5473 | 0 | { |
5474 | 0 | memset(ps_ctb_cluster_info->au1_is_16x16_blk_split, 1, 16 * sizeof(U08)); |
5475 | 0 | } |
5476 | |
|
5477 | | #if ENABLE_UNIFORM_CU_SIZE_16x16 || ENABLE_UNIFORM_CU_SIZE_8x8 |
5478 | | return; |
5479 | | #endif |
5480 | |
|
5481 | 0 | for(i = 0; i < 16; i++) |
5482 | 0 | { |
5483 | 0 | S32 blk_8x8_mask; |
5484 | 0 | S32 is_16x16_blk_valid; |
5485 | 0 | S32 num_clusters_updated; |
5486 | 0 | S32 num_clusters; |
5487 | |
|
5488 | 0 | blk_8x8_mask = ps_ctb_cluster_info->pi4_blk_8x8_mask[i]; |
5489 | |
|
5490 | 0 | ps_blk_16x16 = &ps_ctb_cluster_info->ps_16x16_blk[i]; |
5491 | |
|
5492 | 0 | is_16x16_blk_valid = (blk_8x8_mask == 0xf); |
5493 | |
|
5494 | 0 | if(is_16x16_blk_valid) |
5495 | 0 | { |
5496 | | /* Use 8x8 data when 16x16 CU is split */ |
5497 | 0 | if(ps_search_results[i].u1_split_flag) |
5498 | 0 | { |
5499 | 0 | S32 blk_8x8_idx = i << 2; |
5500 | |
|
5501 | 0 | num_parts = 4; |
5502 | 0 | e_part_type = PRT_NxN; |
5503 | |
|
5504 | 0 | for(j = 0; j < num_parts; j++, blk_8x8_idx++) |
5505 | 0 | { |
5506 | | /* Only 2Nx2N partition supported for 8x8 block */ |
5507 | 0 | ASSERT( |
5508 | 0 | ps_8x8_cu_results[blk_8x8_idx].ps_best_results[0].u1_part_type == |
5509 | 0 | ((PART_TYPE_T)PRT_2Nx2N)); |
5510 | | |
5511 | 0 | aps_part_result[j] = |
5512 | 0 | &ps_8x8_cu_results[blk_8x8_idx].ps_best_results[0].as_pu_results[0]; |
5513 | 0 | aps_inferior_parts[j] = |
5514 | 0 | &ps_8x8_cu_results[blk_8x8_idx].ps_best_results[1].as_pu_results[0]; |
5515 | 0 | ai4_pred_mode[j] = (aps_part_result[j]->pu.b2_pred_mode); |
5516 | 0 | } |
5517 | 0 | } |
5518 | 0 | else |
5519 | 0 | { |
5520 | 0 | ps_best_result = &ps_16x16_cu_results[i].ps_best_results[0]; |
5521 | |
|
5522 | 0 | e_part_type = (PART_TYPE_T)ps_best_result->u1_part_type; |
5523 | 0 | num_parts = gau1_num_parts_in_part_type[e_part_type]; |
5524 | |
|
5525 | 0 | for(j = 0; j < num_parts; j++) |
5526 | 0 | { |
5527 | 0 | aps_part_result[j] = &ps_best_result->as_pu_results[j]; |
5528 | 0 | aps_inferior_parts[j] = &ps_best_result[1].as_pu_results[j]; |
5529 | 0 | ai4_pred_mode[j] = (aps_part_result[j]->pu.b2_pred_mode); |
5530 | 0 | } |
5531 | |
|
5532 | 0 | ps_ctb_cluster_info->au1_is_16x16_blk_split[i] = 0; |
5533 | 0 | } |
5534 | | |
5535 | 0 | for(j = 0; j < num_parts; j++) |
5536 | 0 | { |
5537 | 0 | pu_result_t *ps_part_result = aps_part_result[j]; |
5538 | |
|
5539 | 0 | S32 num_mvs = ((ai4_pred_mode[j] > 1) + 1); |
5540 | |
|
5541 | 0 | e_part_id = ge_part_type_to_part_id[e_part_type][j]; |
5542 | | |
5543 | | /* Skip clustering if best mode is intra */ |
5544 | 0 | if((ps_part_result->pu.b1_intra_flag)) |
5545 | 0 | { |
5546 | 0 | ps_blk_16x16->intra_mv_area += gai4_partition_area[e_part_id]; |
5547 | 0 | ps_blk_16x16->best_inter_cost += aps_inferior_parts[j]->i4_tot_cost; |
5548 | 0 | continue; |
5549 | 0 | } |
5550 | 0 | else |
5551 | 0 | { |
5552 | 0 | ps_blk_16x16->best_inter_cost += ps_part_result->i4_tot_cost; |
5553 | 0 | } |
5554 | | |
5555 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
5556 | 0 | if(e_quality_preset >= ME_HIGH_QUALITY) |
5557 | 0 | { |
5558 | 0 | continue; |
5559 | 0 | } |
5560 | 0 | #endif |
5561 | | |
5562 | 0 | for(k = 0; k < num_mvs; k++) |
5563 | 0 | { |
5564 | 0 | mv_t *ps_mv; |
5565 | |
|
5566 | 0 | pu_mv_t *ps_pu_mv = &ps_part_result->pu.mv; |
5567 | |
|
5568 | 0 | S32 is_l0_mv = ((ai4_pred_mode[j] == 2) && !k) || (ai4_pred_mode[j] == 0); |
5569 | |
|
5570 | 0 | ps_mv = (is_l0_mv) ? (&ps_pu_mv->s_l0_mv) : (&ps_pu_mv->s_l1_mv); |
5571 | |
|
5572 | 0 | mvx = ps_mv->i2_mvx; |
5573 | 0 | mvy = ps_mv->i2_mvy; |
5574 | |
|
5575 | 0 | ref_idx = (is_l0_mv) ? pi1_past_list[ps_pu_mv->i1_l0_ref_idx] |
5576 | 0 | : pi1_future_list[ps_pu_mv->i1_l1_ref_idx]; |
5577 | |
|
5578 | 0 | num_clusters = ps_blk_16x16->num_clusters; |
5579 | |
|
5580 | 0 | hme_find_and_update_clusters( |
5581 | 0 | ps_blk_16x16->as_cluster_data, |
5582 | 0 | &(ps_blk_16x16->num_clusters), |
5583 | 0 | mvx, |
5584 | 0 | mvy, |
5585 | 0 | ref_idx, |
5586 | 0 | ps_part_result->i4_sdi, |
5587 | 0 | e_part_id, |
5588 | 0 | (ai4_pred_mode[j] == 2)); |
5589 | |
|
5590 | 0 | num_clusters_updated = (ps_blk_16x16->num_clusters); |
5591 | |
|
5592 | 0 | ps_blk_16x16->au1_num_clusters[ref_idx] += |
5593 | 0 | (num_clusters_updated - num_clusters); |
5594 | 0 | } |
5595 | 0 | } |
5596 | 0 | } |
5597 | 0 | } |
5598 | | |
5599 | | /* Search for 32x32 clusters */ |
5600 | 0 | for(i = 0; i < 4; i++) |
5601 | 0 | { |
5602 | 0 | S32 num_clusters_merged; |
5603 | |
|
5604 | 0 | S32 is_32x32_blk_valid = (ps_ctb_cluster_info->blk_32x32_mask & (1 << i)) || 0; |
5605 | |
|
5606 | 0 | if(is_32x32_blk_valid) |
5607 | 0 | { |
5608 | 0 | ps_blk_32x32 = &ps_ctb_cluster_info->ps_32x32_blk[i]; |
5609 | 0 | ps_blk_16x16 = &ps_ctb_cluster_info->ps_16x16_blk[i << 2]; |
5610 | |
|
5611 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
5612 | 0 | if(e_quality_preset >= ME_HIGH_QUALITY) |
5613 | 0 | { |
5614 | 0 | for(j = 0; j < 4; j++, ps_blk_16x16++) |
5615 | 0 | { |
5616 | 0 | ps_blk_32x32->intra_mv_area += ps_blk_16x16->intra_mv_area; |
5617 | |
|
5618 | 0 | ps_blk_32x32->best_inter_cost += ps_blk_16x16->best_inter_cost; |
5619 | 0 | } |
5620 | 0 | continue; |
5621 | 0 | } |
5622 | 0 | #endif |
5623 | | |
5624 | 0 | hme_update_32x32_clusters(ps_blk_32x32, ps_blk_16x16); |
5625 | |
|
5626 | 0 | if((ps_blk_32x32->num_clusters >= MAX_NUM_CLUSTERS_IN_VALID_32x32_BLK)) |
5627 | 0 | { |
5628 | 0 | num_clusters_merged = hme_try_merge_clusters_blksize_gt_16( |
5629 | 0 | ps_blk_32x32->as_cluster_data, (ps_blk_32x32->num_clusters)); |
5630 | |
|
5631 | 0 | if(num_clusters_merged) |
5632 | 0 | { |
5633 | 0 | ps_blk_32x32->num_clusters -= num_clusters_merged; |
5634 | |
|
5635 | 0 | UPDATE_CLUSTER_METADATA_POST_MERGE(ps_blk_32x32); |
5636 | 0 | } |
5637 | 0 | } |
5638 | 0 | } |
5639 | 0 | } |
5640 | |
|
5641 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
5642 | | /* Eliminate outlier 32x32 clusters */ |
5643 | 0 | if(e_quality_preset < ME_HIGH_QUALITY) |
5644 | 0 | #endif |
5645 | 0 | { |
5646 | 0 | hme_boot_out_outlier(ps_ctb_cluster_info, 32); |
5647 | | |
5648 | | /* Find best_uni_ref and best_alt_ref */ |
5649 | 0 | hme_find_top_ref_ids(ps_ctb_cluster_info, bidir_enabled, 32); |
5650 | 0 | } |
5651 | | |
5652 | | /* Populate the CU tree for depths 1 and higher */ |
5653 | 0 | { |
5654 | 0 | cur_ctb_cu_tree_t *ps_tree_root = ps_ctb_cluster_info->ps_cu_tree_root; |
5655 | 0 | cur_ctb_cu_tree_t *ps_tl = ps_tree_root->ps_child_node_tl; |
5656 | 0 | cur_ctb_cu_tree_t *ps_tr = ps_tree_root->ps_child_node_tr; |
5657 | 0 | cur_ctb_cu_tree_t *ps_bl = ps_tree_root->ps_child_node_bl; |
5658 | 0 | cur_ctb_cu_tree_t *ps_br = ps_tree_root->ps_child_node_br; |
5659 | |
|
5660 | 0 | hme_populate_cu_tree( |
5661 | 0 | ps_ctb_cluster_info, ps_tl, 1, e_quality_preset, POS_NA, POS_NA, POS_TL); |
5662 | |
|
5663 | 0 | num_32x32_merges += (ps_tl->is_node_valid == 1); |
5664 | |
|
5665 | 0 | hme_populate_cu_tree( |
5666 | 0 | ps_ctb_cluster_info, ps_tr, 1, e_quality_preset, POS_NA, POS_NA, POS_TR); |
5667 | |
|
5668 | 0 | num_32x32_merges += (ps_tr->is_node_valid == 1); |
5669 | |
|
5670 | 0 | hme_populate_cu_tree( |
5671 | 0 | ps_ctb_cluster_info, ps_bl, 1, e_quality_preset, POS_NA, POS_NA, POS_BL); |
5672 | |
|
5673 | 0 | num_32x32_merges += (ps_bl->is_node_valid == 1); |
5674 | |
|
5675 | 0 | hme_populate_cu_tree( |
5676 | 0 | ps_ctb_cluster_info, ps_br, 1, e_quality_preset, POS_NA, POS_NA, POS_BR); |
5677 | |
|
5678 | 0 | num_32x32_merges += (ps_br->is_node_valid == 1); |
5679 | 0 | } |
5680 | |
|
5681 | | #if !ENABLE_4CTB_EVALUATION |
5682 | | if(e_quality_preset < ME_HIGH_QUALITY) |
5683 | | { |
5684 | | enable_64x64_merge = (num_32x32_merges >= 3); |
5685 | | } |
5686 | | #else |
5687 | 0 | if(e_quality_preset < ME_HIGH_QUALITY) |
5688 | 0 | { |
5689 | 0 | enable_64x64_merge = 1; |
5690 | 0 | } |
5691 | 0 | #endif |
5692 | |
|
5693 | 0 | #if 1 //!PROCESS_GT_1CTB_VIA_CU_RECUR_IN_FAST_PRESETS |
5694 | 0 | if(e_quality_preset >= ME_HIGH_QUALITY) |
5695 | 0 | { |
5696 | 0 | enable_64x64_merge = 1; |
5697 | 0 | } |
5698 | | #else |
5699 | | if(e_quality_preset >= ME_HIGH_QUALITY) |
5700 | | { |
5701 | | enable_64x64_merge = (num_32x32_merges >= 3); |
5702 | | } |
5703 | | #endif |
5704 | |
|
5705 | 0 | if(enable_64x64_merge) |
5706 | 0 | { |
5707 | 0 | S32 num_clusters_merged; |
5708 | |
|
5709 | 0 | ps_blk_32x32 = &ps_ctb_cluster_info->ps_32x32_blk[0]; |
5710 | |
|
5711 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
5712 | 0 | if(e_quality_preset >= ME_HIGH_QUALITY) |
5713 | 0 | { |
5714 | 0 | for(j = 0; j < 4; j++, ps_blk_32x32++) |
5715 | 0 | { |
5716 | 0 | ps_blk_64x64->intra_mv_area += ps_blk_32x32->intra_mv_area; |
5717 | |
|
5718 | 0 | ps_blk_64x64->best_inter_cost += ps_blk_32x32->best_inter_cost; |
5719 | 0 | } |
5720 | 0 | } |
5721 | 0 | else |
5722 | 0 | #endif |
5723 | 0 | { |
5724 | 0 | hme_update_64x64_clusters(ps_blk_64x64, ps_blk_32x32); |
5725 | |
|
5726 | 0 | if((ps_blk_64x64->num_clusters >= MAX_NUM_CLUSTERS_IN_VALID_64x64_BLK)) |
5727 | 0 | { |
5728 | 0 | num_clusters_merged = hme_try_merge_clusters_blksize_gt_16( |
5729 | 0 | ps_blk_64x64->as_cluster_data, (ps_blk_64x64->num_clusters)); |
5730 | |
|
5731 | 0 | if(num_clusters_merged) |
5732 | 0 | { |
5733 | 0 | ps_blk_64x64->num_clusters -= num_clusters_merged; |
5734 | |
|
5735 | 0 | UPDATE_CLUSTER_METADATA_POST_MERGE(ps_blk_64x64); |
5736 | 0 | } |
5737 | 0 | } |
5738 | 0 | } |
5739 | |
|
5740 | | #if !ENABLE_4CTB_EVALUATION |
5741 | | if(e_quality_preset < ME_HIGH_QUALITY) |
5742 | | { |
5743 | | S32 best_inter_cost = ps_blk_64x64->best_inter_cost; |
5744 | | S32 best_intra_cost = |
5745 | | ((ps_ctb_cluster_info->ps_cur_ipe_ctb->i4_best64x64_intra_cost + |
5746 | | ps_ctb_cluster_info->i4_frame_qstep * |
5747 | | ps_ctb_cluster_info->i4_frame_qstep_multiplier * 16) < 0) |
5748 | | ? MAX_32BIT_VAL |
5749 | | : (ps_ctb_cluster_info->ps_cur_ipe_ctb->i4_best64x64_intra_cost + |
5750 | | ps_ctb_cluster_info->i4_frame_qstep * |
5751 | | ps_ctb_cluster_info->i4_frame_qstep_multiplier * 16); |
5752 | | S32 best_cost = (best_inter_cost > best_intra_cost) ? best_intra_cost : best_inter_cost; |
5753 | | S32 cost_differential = (best_inter_cost - best_cost); |
5754 | | |
5755 | | enable_64x64_merge = |
5756 | | ((ALL_INTER_COST_DIFF_THR * best_cost) >= (100 * cost_differential)); |
5757 | | } |
5758 | | #endif |
5759 | 0 | } |
5760 | |
|
5761 | 0 | if(enable_64x64_merge) |
5762 | 0 | { |
5763 | 0 | #if !USE_CLUSTER_DATA_AS_BLK_MERGE_CANDTS |
5764 | 0 | if(e_quality_preset < ME_HIGH_QUALITY) |
5765 | 0 | #endif |
5766 | 0 | { |
5767 | 0 | hme_boot_out_outlier(ps_ctb_cluster_info, 64); |
5768 | |
|
5769 | 0 | hme_find_top_ref_ids(ps_ctb_cluster_info, bidir_enabled, 64); |
5770 | 0 | } |
5771 | |
|
5772 | 0 | hme_populate_cu_tree( |
5773 | 0 | ps_ctb_cluster_info, |
5774 | 0 | ps_ctb_cluster_info->ps_cu_tree_root, |
5775 | 0 | 0, |
5776 | 0 | e_quality_preset, |
5777 | 0 | POS_NA, |
5778 | 0 | POS_NA, |
5779 | 0 | POS_NA); |
5780 | 0 | } |
5781 | 0 | } |
5782 | | #endif |
5783 | | |
5784 | | static __inline void hme_merge_prms_init( |
5785 | | hme_merge_prms_t *ps_prms, |
5786 | | layer_ctxt_t *ps_curr_layer, |
5787 | | refine_prms_t *ps_refine_prms, |
5788 | | me_frm_ctxt_t *ps_me_ctxt, |
5789 | | range_prms_t *ps_range_prms_rec, |
5790 | | range_prms_t *ps_range_prms_inp, |
5791 | | mv_grid_t **pps_mv_grid, |
5792 | | inter_ctb_prms_t *ps_inter_ctb_prms, |
5793 | | S32 i4_num_pred_dir, |
5794 | | S32 i4_32x32_id, |
5795 | | BLK_SIZE_T e_blk_size, |
5796 | | ME_QUALITY_PRESETS_T e_me_quality_presets) |
5797 | 0 | { |
5798 | 0 | S32 i4_use_rec = ps_refine_prms->i4_use_rec_in_fpel; |
5799 | 0 | S32 i4_cu_16x16 = (BLK_32x32 == e_blk_size) ? (i4_32x32_id << 2) : 0; |
5800 | | |
5801 | | /* Currently not enabling segmentation info from prev layers */ |
5802 | 0 | ps_prms->i4_seg_info_avail = 0; |
5803 | 0 | ps_prms->i4_part_mask = 0; |
5804 | | |
5805 | | /* Number of reference pics in which to do merge */ |
5806 | 0 | ps_prms->i4_num_ref = i4_num_pred_dir; |
5807 | | |
5808 | | /* Layer ctxt info */ |
5809 | 0 | ps_prms->ps_layer_ctxt = ps_curr_layer; |
5810 | |
|
5811 | 0 | ps_prms->ps_inter_ctb_prms = ps_inter_ctb_prms; |
5812 | | |
5813 | | /* Top left, top right, bottom left and bottom right 16x16 units */ |
5814 | 0 | if(BLK_32x32 == e_blk_size) |
5815 | 0 | { |
5816 | 0 | ps_prms->ps_results_tl = &ps_me_ctxt->as_search_results_16x16[i4_cu_16x16]; |
5817 | 0 | ps_prms->ps_results_tr = &ps_me_ctxt->as_search_results_16x16[i4_cu_16x16 + 1]; |
5818 | 0 | ps_prms->ps_results_bl = &ps_me_ctxt->as_search_results_16x16[i4_cu_16x16 + 2]; |
5819 | 0 | ps_prms->ps_results_br = &ps_me_ctxt->as_search_results_16x16[i4_cu_16x16 + 3]; |
5820 | | |
5821 | | /* Merge results stored here */ |
5822 | 0 | ps_prms->ps_results_merge = &ps_me_ctxt->as_search_results_32x32[i4_32x32_id]; |
5823 | | |
5824 | | /* This could be lesser than the number of 16x16results generated*/ |
5825 | | /* For now, keeping it to be same */ |
5826 | 0 | ps_prms->i4_num_inp_results = ps_refine_prms->i4_num_fpel_results; |
5827 | 0 | ps_prms->ps_8x8_cu_results = &ps_me_ctxt->as_cu8x8_results[i4_32x32_id << 4]; |
5828 | 0 | ps_prms->ps_results_grandchild = NULL; |
5829 | 0 | } |
5830 | 0 | else |
5831 | 0 | { |
5832 | 0 | ps_prms->ps_results_tl = &ps_me_ctxt->as_search_results_32x32[0]; |
5833 | 0 | ps_prms->ps_results_tr = &ps_me_ctxt->as_search_results_32x32[1]; |
5834 | 0 | ps_prms->ps_results_bl = &ps_me_ctxt->as_search_results_32x32[2]; |
5835 | 0 | ps_prms->ps_results_br = &ps_me_ctxt->as_search_results_32x32[3]; |
5836 | | |
5837 | | /* Merge results stored here */ |
5838 | 0 | ps_prms->ps_results_merge = &ps_me_ctxt->s_search_results_64x64; |
5839 | |
|
5840 | 0 | ps_prms->i4_num_inp_results = ps_refine_prms->i4_num_32x32_merge_results; |
5841 | 0 | ps_prms->ps_8x8_cu_results = &ps_me_ctxt->as_cu8x8_results[0]; |
5842 | 0 | ps_prms->ps_results_grandchild = ps_me_ctxt->as_search_results_16x16; |
5843 | 0 | } |
5844 | |
|
5845 | 0 | if(i4_use_rec) |
5846 | 0 | { |
5847 | 0 | WORD32 ref_ctr; |
5848 | |
|
5849 | 0 | for(ref_ctr = 0; ref_ctr < MAX_NUM_REF; ref_ctr++) |
5850 | 0 | { |
5851 | 0 | ps_prms->aps_mv_range[ref_ctr] = &ps_range_prms_rec[ref_ctr]; |
5852 | 0 | } |
5853 | 0 | } |
5854 | 0 | else |
5855 | 0 | { |
5856 | 0 | WORD32 ref_ctr; |
5857 | |
|
5858 | 0 | for(ref_ctr = 0; ref_ctr < MAX_NUM_REF; ref_ctr++) |
5859 | 0 | { |
5860 | 0 | ps_prms->aps_mv_range[ref_ctr] = &ps_range_prms_inp[ref_ctr]; |
5861 | 0 | } |
5862 | 0 | } |
5863 | 0 | ps_prms->i4_use_rec = i4_use_rec; |
5864 | |
|
5865 | 0 | ps_prms->pf_mv_cost_compute = compute_mv_cost_implicit_high_speed; |
5866 | |
|
5867 | 0 | ps_prms->pps_mv_grid = pps_mv_grid; |
5868 | |
|
5869 | 0 | ps_prms->log_ctb_size = ps_me_ctxt->log_ctb_size; |
5870 | |
|
5871 | 0 | ps_prms->e_quality_preset = e_me_quality_presets; |
5872 | 0 | ps_prms->pi1_future_list = ps_me_ctxt->ai1_future_list; |
5873 | 0 | ps_prms->pi1_past_list = ps_me_ctxt->ai1_past_list; |
5874 | 0 | ps_prms->ps_cluster_info = ps_me_ctxt->ps_ctb_cluster_info; |
5875 | 0 | } |
5876 | | |
5877 | | /** |
5878 | | ******************************************************************************** |
5879 | | * @fn void hme_refine(me_ctxt_t *ps_ctxt, |
5880 | | * refine_layer_prms_t *ps_refine_prms) |
5881 | | * |
5882 | | * @brief Top level entry point for refinement ME |
5883 | | * |
5884 | | * @param[in,out] ps_ctxt: ME Handle |
5885 | | * |
5886 | | * @param[in] ps_refine_prms : refinement layer prms |
5887 | | * |
5888 | | * @return None |
5889 | | ******************************************************************************** |
5890 | | */ |
5891 | | void hme_refine( |
5892 | | me_ctxt_t *ps_thrd_ctxt, |
5893 | | refine_prms_t *ps_refine_prms, |
5894 | | PF_EXT_UPDATE_FXN_T pf_ext_update_fxn, |
5895 | | layer_ctxt_t *ps_coarse_layer, |
5896 | | multi_thrd_ctxt_t *ps_multi_thrd_ctxt, |
5897 | | S32 lyr_job_type, |
5898 | | S32 thrd_id, |
5899 | | S32 me_frm_id, |
5900 | | pre_enc_L0_ipe_encloop_ctxt_t *ps_l0_ipe_input) |
5901 | 0 | { |
5902 | 0 | inter_ctb_prms_t s_common_frm_prms; |
5903 | |
|
5904 | 0 | BLK_SIZE_T e_search_blk_size, e_result_blk_size; |
5905 | 0 | WORD32 i4_me_frm_id = me_frm_id % MAX_NUM_ME_PARALLEL; |
5906 | 0 | me_frm_ctxt_t *ps_ctxt = ps_thrd_ctxt->aps_me_frm_prms[i4_me_frm_id]; |
5907 | 0 | ME_QUALITY_PRESETS_T e_me_quality_presets = |
5908 | 0 | ps_thrd_ctxt->s_init_prms.s_me_coding_tools.e_me_quality_presets; |
5909 | |
|
5910 | 0 | WORD32 num_rows_proc = 0; |
5911 | 0 | WORD32 num_act_ref_pics; |
5912 | 0 | WORD16 i2_prev_enc_frm_max_mv_y; |
5913 | 0 | WORD32 i4_idx_dvsr_p = ps_multi_thrd_ctxt->i4_idx_dvsr_p; |
5914 | | |
5915 | | /*************************************************************************/ |
5916 | | /* Complexity of search: Low to High */ |
5917 | | /*************************************************************************/ |
5918 | 0 | SEARCH_COMPLEXITY_T e_search_complexity; |
5919 | | |
5920 | | /*************************************************************************/ |
5921 | | /* to store the PU results which are passed to the decide_part_types */ |
5922 | | /* as input prms. Multiplied by 4 as the max number of Ref in a List is 4*/ |
5923 | | /*************************************************************************/ |
5924 | |
|
5925 | 0 | pu_result_t as_pu_results[2][TOT_NUM_PARTS][MAX_NUM_RESULTS_PER_PART_LIST]; |
5926 | 0 | inter_pu_results_t as_inter_pu_results[4]; |
5927 | 0 | inter_pu_results_t *ps_pu_results = as_inter_pu_results; |
5928 | | |
5929 | | /*************************************************************************/ |
5930 | | /* Config parameter structures for varius ME submodules */ |
5931 | | /*************************************************************************/ |
5932 | 0 | hme_merge_prms_t s_merge_prms_32x32_tl, s_merge_prms_32x32_tr; |
5933 | 0 | hme_merge_prms_t s_merge_prms_32x32_bl, s_merge_prms_32x32_br; |
5934 | 0 | hme_merge_prms_t s_merge_prms_64x64; |
5935 | 0 | hme_search_prms_t s_search_prms_blk; |
5936 | 0 | mvbank_update_prms_t s_mv_update_prms; |
5937 | 0 | hme_ctb_prms_t s_ctb_prms; |
5938 | 0 | hme_subpel_prms_t s_subpel_prms; |
5939 | 0 | fullpel_refine_ctxt_t *ps_fullpel_refine_ctxt = ps_ctxt->ps_fullpel_refine_ctxt; |
5940 | 0 | ctb_cluster_info_t *ps_ctb_cluster_info; |
5941 | 0 | fpel_srch_cand_init_data_t s_srch_cand_init_data; |
5942 | | |
5943 | | /* 4 bits (LSBs) of this variable control merge of 4 32x32 CUs in CTB */ |
5944 | 0 | S32 en_merge_32x32; |
5945 | | /* 5 lsb's specify whether or not merge algorithm is required */ |
5946 | | /* to be executed or not. Relevant only in PQ. Ought to be */ |
5947 | | /* used in conjunction with en_merge_32x32 and */ |
5948 | | /* ps_ctb_bound_attrs->u1_merge_to_64x64_flag. This is */ |
5949 | | /* required when all children are deemed to be intras */ |
5950 | 0 | S32 en_merge_execution; |
5951 | | |
5952 | | /*************************************************************************/ |
5953 | | /* All types of search candidates for predictor based search. */ |
5954 | | /*************************************************************************/ |
5955 | 0 | S32 num_init_candts = 0; |
5956 | 0 | S32 i4_num_act_ref_l0 = ps_ctxt->s_frm_prms.u1_num_active_ref_l0; |
5957 | 0 | S32 i4_num_act_ref_l1 = ps_ctxt->s_frm_prms.u1_num_active_ref_l1; |
5958 | 0 | search_candt_t *ps_search_candts, as_search_candts[MAX_INIT_CANDTS]; |
5959 | 0 | search_node_t as_top_neighbours[4], as_left_neighbours[3]; |
5960 | |
|
5961 | 0 | pf_get_wt_inp fp_get_wt_inp; |
5962 | |
|
5963 | 0 | search_node_t as_unique_search_nodes[MAX_INIT_CANDTS * 9]; |
5964 | 0 | U32 au4_unique_node_map[MAP_X_MAX * 2]; |
5965 | | |
5966 | | /* Controls the boundary attributes of CTB, whether it has 64x64 or not */ |
5967 | 0 | ctb_boundary_attrs_t *ps_ctb_bound_attrs; |
5968 | | |
5969 | | /*************************************************************************/ |
5970 | | /* points ot the search results for the blk level search (8x8/16x16) */ |
5971 | | /*************************************************************************/ |
5972 | 0 | search_results_t *ps_search_results; |
5973 | | |
5974 | | /*************************************************************************/ |
5975 | | /* Coordinates */ |
5976 | | /*************************************************************************/ |
5977 | 0 | S32 blk_x, blk_y, i4_ctb_x, i4_ctb_y, tile_col_idx, blk_id_in_ctb; |
5978 | 0 | S32 pos_x, pos_y; |
5979 | 0 | S32 blk_id_in_full_ctb; |
5980 | | |
5981 | | /*************************************************************************/ |
5982 | | /* Related to dimensions of block being searched and pic dimensions */ |
5983 | | /*************************************************************************/ |
5984 | 0 | S32 blk_4x4_to_16x16; |
5985 | 0 | S32 blk_wd, blk_ht, blk_size_shift; |
5986 | 0 | S32 i4_pic_wd, i4_pic_ht, num_blks_in_this_ctb; |
5987 | 0 | S32 num_results_prev_layer; |
5988 | | |
5989 | | /*************************************************************************/ |
5990 | | /* Size of a basic unit for this layer. For non encode layers, we search */ |
5991 | | /* in block sizes of 8x8. For encode layers, though we search 16x16s the */ |
5992 | | /* basic unit size is the ctb size. */ |
5993 | | /*************************************************************************/ |
5994 | 0 | S32 unit_size; |
5995 | | |
5996 | | /*************************************************************************/ |
5997 | | /* Local variable storing results of any 4 CU merge to bigger CU */ |
5998 | | /*************************************************************************/ |
5999 | 0 | CU_MERGE_RESULT_T e_merge_result; |
6000 | | |
6001 | | /*************************************************************************/ |
6002 | | /* This mv grid stores results during and after fpel search, during */ |
6003 | | /* merge, subpel and bidirect refinements stages. 2 instances of this are*/ |
6004 | | /* meant for the 2 directions of search (l0 and l1). */ |
6005 | | /*************************************************************************/ |
6006 | 0 | mv_grid_t *aps_mv_grid[2]; |
6007 | | |
6008 | | /*************************************************************************/ |
6009 | | /* Pointers to context in current and coarser layers */ |
6010 | | /*************************************************************************/ |
6011 | 0 | layer_ctxt_t *ps_curr_layer, *ps_prev_layer; |
6012 | | |
6013 | | /*************************************************************************/ |
6014 | | /* to store mv range per blk, and picture limit, allowed search range */ |
6015 | | /* range prms in hpel and qpel units as well */ |
6016 | | /*************************************************************************/ |
6017 | 0 | range_prms_t as_range_prms_inp[MAX_NUM_REF], as_range_prms_rec[MAX_NUM_REF]; |
6018 | 0 | range_prms_t s_pic_limit_inp, s_pic_limit_rec, as_mv_limit[MAX_NUM_REF]; |
6019 | 0 | range_prms_t as_range_prms_hpel[MAX_NUM_REF], as_range_prms_qpel[MAX_NUM_REF]; |
6020 | | |
6021 | | /*************************************************************************/ |
6022 | | /* These variables are used to track number of references at different */ |
6023 | | /* stages of ME. */ |
6024 | | /*************************************************************************/ |
6025 | 0 | S32 i4_num_pred_dir; |
6026 | 0 | S32 i4_num_ref_each_dir, i, i4_num_ref_prev_layer; |
6027 | 0 | S32 lambda_recon = ps_refine_prms->lambda_recon; |
6028 | | |
6029 | | /* Counts successful merge to 32x32 every CTB (0-4) */ |
6030 | 0 | S32 merge_count_32x32; |
6031 | |
|
6032 | 0 | S32 ai4_id_coloc[14], ai4_id_Z[2]; |
6033 | 0 | U08 au1_search_candidate_list_index[2]; |
6034 | 0 | S32 ai4_num_coloc_cands[2]; |
6035 | 0 | U08 u1_pred_dir, u1_pred_dir_ctr; |
6036 | | |
6037 | | /*************************************************************************/ |
6038 | | /* Input pointer and stride */ |
6039 | | /*************************************************************************/ |
6040 | 0 | U08 *pu1_inp; |
6041 | 0 | S32 i4_inp_stride; |
6042 | 0 | S32 end_of_frame; |
6043 | 0 | S32 num_sync_units_in_row, num_sync_units_in_tile; |
6044 | | |
6045 | | /*************************************************************************/ |
6046 | | /* Indicates whether the all 4 8x8 blks are valid in the 16x16 blk in the*/ |
6047 | | /* encode layer. If not 15, then 1 or more 8x8 blks not valid. Means that*/ |
6048 | | /* we need to stop merges and force 8x8 CUs for that 16x16 blk */ |
6049 | | /*************************************************************************/ |
6050 | 0 | S32 blk_8x8_mask; |
6051 | 0 | S32 ai4_blk_8x8_mask[16]; |
6052 | 0 | U08 au1_is_64x64Blk_noisy[1]; |
6053 | 0 | U08 au1_is_32x32Blk_noisy[4]; |
6054 | 0 | U08 au1_is_16x16Blk_noisy[16]; |
6055 | |
|
6056 | 0 | ihevce_cmn_opt_func_t *ps_cmn_utils_optimised_function_list = |
6057 | 0 | ps_thrd_ctxt->ps_cmn_utils_optimised_function_list; |
6058 | 0 | ihevce_me_optimised_function_list_t *ps_me_optimised_function_list = |
6059 | 0 | ((ihevce_me_optimised_function_list_t *)ps_thrd_ctxt->pv_me_optimised_function_list); |
6060 | |
|
6061 | 0 | ASSERT(ps_refine_prms->i4_layer_id < ps_ctxt->num_layers - 1); |
6062 | | |
6063 | | /*************************************************************************/ |
6064 | | /* Pointers to current and coarse layer are needed for projection */ |
6065 | | /* Pointer to prev layer are needed for other candts like coloc */ |
6066 | | /*************************************************************************/ |
6067 | 0 | ps_curr_layer = ps_ctxt->ps_curr_descr->aps_layers[ps_refine_prms->i4_layer_id]; |
6068 | |
|
6069 | 0 | ps_prev_layer = hme_get_past_layer_ctxt( |
6070 | 0 | ps_thrd_ctxt, ps_ctxt, ps_refine_prms->i4_layer_id, ps_multi_thrd_ctxt->i4_num_me_frm_pllel); |
6071 | |
|
6072 | 0 | num_results_prev_layer = ps_coarse_layer->ps_layer_mvbank->i4_num_mvs_per_ref; |
6073 | | |
6074 | | /* Function pointer is selected based on the C vc X86 macro */ |
6075 | |
|
6076 | 0 | fp_get_wt_inp = ps_me_optimised_function_list->pf_get_wt_inp_ctb; |
6077 | |
|
6078 | 0 | i4_inp_stride = ps_curr_layer->i4_inp_stride; |
6079 | 0 | i4_pic_wd = ps_curr_layer->i4_wd; |
6080 | 0 | i4_pic_ht = ps_curr_layer->i4_ht; |
6081 | 0 | e_search_complexity = ps_refine_prms->e_search_complexity; |
6082 | 0 | end_of_frame = 0; |
6083 | | |
6084 | | /* This points to all the initial candts */ |
6085 | 0 | ps_search_candts = &as_search_candts[0]; |
6086 | | |
6087 | | /* mv grid being huge strucutre is part of context */ |
6088 | 0 | aps_mv_grid[0] = &ps_ctxt->as_mv_grid[0]; |
6089 | 0 | aps_mv_grid[1] = &ps_ctxt->as_mv_grid[1]; |
6090 | | |
6091 | | /*************************************************************************/ |
6092 | | /* If the current layer is encoded (since it may be multicast or final */ |
6093 | | /* layer (finest)), then we use 16x16 blk size with some selected parts */ |
6094 | | /* If the current layer is not encoded, then we use 8x8 blk size, with */ |
6095 | | /* enable or disable of 4x4 partitions depending on the input prms */ |
6096 | | /*************************************************************************/ |
6097 | 0 | e_search_blk_size = BLK_16x16; |
6098 | 0 | blk_wd = blk_ht = 16; |
6099 | 0 | blk_size_shift = 4; |
6100 | 0 | e_result_blk_size = BLK_8x8; |
6101 | 0 | s_mv_update_prms.i4_shift = 1; |
6102 | |
|
6103 | 0 | if(ps_coarse_layer->ps_layer_mvbank->e_blk_size == BLK_4x4) |
6104 | 0 | { |
6105 | 0 | blk_4x4_to_16x16 = 1; |
6106 | 0 | } |
6107 | 0 | else |
6108 | 0 | { |
6109 | 0 | blk_4x4_to_16x16 = 0; |
6110 | 0 | } |
6111 | |
|
6112 | 0 | unit_size = 1 << ps_ctxt->log_ctb_size; |
6113 | 0 | s_search_prms_blk.i4_inp_stride = unit_size; |
6114 | | |
6115 | | /* This is required to properly update the layer mv bank */ |
6116 | 0 | s_mv_update_prms.e_search_blk_size = e_search_blk_size; |
6117 | 0 | s_search_prms_blk.e_blk_size = e_search_blk_size; |
6118 | | |
6119 | | /*************************************************************************/ |
6120 | | /* If current layer is explicit, then the number of ref frames are to */ |
6121 | | /* be same as previous layer. Else it will be 2 */ |
6122 | | /*************************************************************************/ |
6123 | 0 | i4_num_ref_prev_layer = ps_coarse_layer->ps_layer_mvbank->i4_num_ref; |
6124 | 0 | i4_num_pred_dir = |
6125 | 0 | (ps_ctxt->s_frm_prms.bidir_enabled && (i4_num_act_ref_l0 > 0) && (i4_num_act_ref_l1 > 0)) + |
6126 | 0 | 1; |
6127 | |
|
6128 | 0 | #if USE_MODIFIED == 1 |
6129 | 0 | s_search_prms_blk.pf_mv_cost_compute = compute_mv_cost_implicit_high_speed_modified; |
6130 | | #else |
6131 | | s_search_prms_blk.pf_mv_cost_compute = compute_mv_cost_implicit_high_speed; |
6132 | | #endif |
6133 | |
|
6134 | 0 | i4_num_pred_dir = MIN(i4_num_pred_dir, i4_num_ref_prev_layer); |
6135 | 0 | if(i4_num_ref_prev_layer <= 2) |
6136 | 0 | { |
6137 | 0 | i4_num_ref_each_dir = 1; |
6138 | 0 | } |
6139 | 0 | else |
6140 | 0 | { |
6141 | 0 | i4_num_ref_each_dir = i4_num_ref_prev_layer >> 1; |
6142 | 0 | } |
6143 | |
|
6144 | 0 | s_mv_update_prms.i4_num_ref = i4_num_pred_dir; |
6145 | 0 | s_mv_update_prms.i4_num_results_to_store = |
6146 | 0 | MIN((ps_ctxt->s_frm_prms.bidir_enabled) ? ps_curr_layer->ps_layer_mvbank->i4_num_mvs_per_ref |
6147 | 0 | : (i4_num_act_ref_l0 > 1) + 1, |
6148 | 0 | ps_refine_prms->i4_num_results_per_part); |
6149 | | |
6150 | | /*************************************************************************/ |
6151 | | /* Initialization of merge params for 16x16 to 32x32 merge. */ |
6152 | | /* There are 4 32x32 units in a CTB, so 4 param structures initialized */ |
6153 | | /*************************************************************************/ |
6154 | 0 | { |
6155 | 0 | hme_merge_prms_t *aps_merge_prms[4]; |
6156 | 0 | aps_merge_prms[0] = &s_merge_prms_32x32_tl; |
6157 | 0 | aps_merge_prms[1] = &s_merge_prms_32x32_tr; |
6158 | 0 | aps_merge_prms[2] = &s_merge_prms_32x32_bl; |
6159 | 0 | aps_merge_prms[3] = &s_merge_prms_32x32_br; |
6160 | 0 | for(i = 0; i < 4; i++) |
6161 | 0 | { |
6162 | 0 | hme_merge_prms_init( |
6163 | 0 | aps_merge_prms[i], |
6164 | 0 | ps_curr_layer, |
6165 | 0 | ps_refine_prms, |
6166 | 0 | ps_ctxt, |
6167 | 0 | as_range_prms_rec, |
6168 | 0 | as_range_prms_inp, |
6169 | 0 | &aps_mv_grid[0], |
6170 | 0 | &s_common_frm_prms, |
6171 | 0 | i4_num_pred_dir, |
6172 | 0 | i, |
6173 | 0 | BLK_32x32, |
6174 | 0 | e_me_quality_presets); |
6175 | 0 | } |
6176 | 0 | } |
6177 | | |
6178 | | /*************************************************************************/ |
6179 | | /* Initialization of merge params for 32x32 to 64x64 merge. */ |
6180 | | /* There are 4 32x32 units in a CTB, so only 1 64x64 CU can be in CTB */ |
6181 | | /*************************************************************************/ |
6182 | 0 | { |
6183 | 0 | hme_merge_prms_init( |
6184 | 0 | &s_merge_prms_64x64, |
6185 | 0 | ps_curr_layer, |
6186 | 0 | ps_refine_prms, |
6187 | 0 | ps_ctxt, |
6188 | 0 | as_range_prms_rec, |
6189 | 0 | as_range_prms_inp, |
6190 | 0 | &aps_mv_grid[0], |
6191 | 0 | &s_common_frm_prms, |
6192 | 0 | i4_num_pred_dir, |
6193 | 0 | 0, |
6194 | 0 | BLK_64x64, |
6195 | 0 | e_me_quality_presets); |
6196 | 0 | } |
6197 | | |
6198 | | /* Pointers to cu_results are initialised here */ |
6199 | 0 | { |
6200 | 0 | WORD32 i; |
6201 | |
|
6202 | 0 | ps_ctxt->s_search_results_64x64.ps_cu_results = &ps_ctxt->s_cu64x64_results; |
6203 | |
|
6204 | 0 | for(i = 0; i < 4; i++) |
6205 | 0 | { |
6206 | 0 | ps_ctxt->as_search_results_32x32[i].ps_cu_results = &ps_ctxt->as_cu32x32_results[i]; |
6207 | 0 | } |
6208 | |
|
6209 | 0 | for(i = 0; i < 16; i++) |
6210 | 0 | { |
6211 | 0 | ps_ctxt->as_search_results_16x16[i].ps_cu_results = &ps_ctxt->as_cu16x16_results[i]; |
6212 | 0 | } |
6213 | 0 | } |
6214 | | |
6215 | | /*************************************************************************/ |
6216 | | /* SUBPEL Params initialized here */ |
6217 | | /*************************************************************************/ |
6218 | 0 | { |
6219 | 0 | s_subpel_prms.ps_search_results_16x16 = &ps_ctxt->as_search_results_16x16[0]; |
6220 | 0 | s_subpel_prms.ps_search_results_32x32 = &ps_ctxt->as_search_results_32x32[0]; |
6221 | 0 | s_subpel_prms.ps_search_results_64x64 = &ps_ctxt->s_search_results_64x64; |
6222 | |
|
6223 | 0 | s_subpel_prms.i4_num_16x16_candts = ps_refine_prms->i4_num_fpel_results; |
6224 | 0 | s_subpel_prms.i4_num_32x32_candts = ps_refine_prms->i4_num_32x32_merge_results; |
6225 | 0 | s_subpel_prms.i4_num_64x64_candts = ps_refine_prms->i4_num_64x64_merge_results; |
6226 | |
|
6227 | 0 | s_subpel_prms.i4_num_steps_hpel_refine = ps_refine_prms->i4_num_steps_hpel_refine; |
6228 | 0 | s_subpel_prms.i4_num_steps_qpel_refine = ps_refine_prms->i4_num_steps_qpel_refine; |
6229 | |
|
6230 | 0 | s_subpel_prms.i4_use_satd = ps_refine_prms->i4_use_satd_subpel; |
6231 | |
|
6232 | 0 | s_subpel_prms.i4_inp_stride = unit_size; |
6233 | |
|
6234 | 0 | s_subpel_prms.u1_max_subpel_candts_2Nx2N = ps_refine_prms->u1_max_subpel_candts_2Nx2N; |
6235 | 0 | s_subpel_prms.u1_max_subpel_candts_NxN = ps_refine_prms->u1_max_subpel_candts_NxN; |
6236 | 0 | s_subpel_prms.u1_subpel_candt_threshold = ps_refine_prms->u1_subpel_candt_threshold; |
6237 | |
|
6238 | 0 | s_subpel_prms.pf_qpel_interp = ps_me_optimised_function_list->pf_qpel_interp_avg_generic; |
6239 | |
|
6240 | 0 | { |
6241 | 0 | WORD32 ref_ctr; |
6242 | 0 | for(ref_ctr = 0; ref_ctr < MAX_NUM_REF; ref_ctr++) |
6243 | 0 | { |
6244 | 0 | s_subpel_prms.aps_mv_range_hpel[ref_ctr] = &as_range_prms_hpel[ref_ctr]; |
6245 | 0 | s_subpel_prms.aps_mv_range_qpel[ref_ctr] = &as_range_prms_qpel[ref_ctr]; |
6246 | 0 | } |
6247 | 0 | } |
6248 | 0 | s_subpel_prms.pi2_inp_bck = ps_ctxt->pi2_inp_bck; |
6249 | |
|
6250 | | #if USE_MODIFIED == 0 |
6251 | | s_subpel_prms.pf_mv_cost_compute = compute_mv_cost_implicit_high_speed; |
6252 | | #else |
6253 | 0 | s_subpel_prms.pf_mv_cost_compute = compute_mv_cost_implicit_high_speed_modified; |
6254 | 0 | #endif |
6255 | 0 | s_subpel_prms.e_me_quality_presets = e_me_quality_presets; |
6256 | | |
6257 | | /* BI Refinement done only if this field is 1 */ |
6258 | 0 | s_subpel_prms.bidir_enabled = ps_refine_prms->bidir_enabled; |
6259 | |
|
6260 | 0 | s_subpel_prms.u1_num_ref = ps_ctxt->num_ref_future + ps_ctxt->num_ref_past; |
6261 | |
|
6262 | 0 | s_subpel_prms.i4_num_act_ref_l0 = ps_ctxt->s_frm_prms.u1_num_active_ref_l0; |
6263 | 0 | s_subpel_prms.i4_num_act_ref_l1 = ps_ctxt->s_frm_prms.u1_num_active_ref_l1; |
6264 | 0 | s_subpel_prms.u1_max_num_subpel_refine_centers = |
6265 | 0 | ps_refine_prms->u1_max_num_subpel_refine_centers; |
6266 | 0 | } |
6267 | | |
6268 | | /* inter_ctb_prms_t struct initialisation */ |
6269 | 0 | { |
6270 | 0 | inter_ctb_prms_t *ps_inter_ctb_prms = &s_common_frm_prms; |
6271 | 0 | hme_subpel_prms_t *ps_subpel_prms = &s_subpel_prms; |
6272 | |
|
6273 | 0 | ps_inter_ctb_prms->pps_rec_list_l0 = ps_ctxt->ps_hme_ref_map->pps_rec_list_l0; |
6274 | 0 | ps_inter_ctb_prms->pps_rec_list_l1 = ps_ctxt->ps_hme_ref_map->pps_rec_list_l1; |
6275 | 0 | ps_inter_ctb_prms->wpred_log_wdc = ps_ctxt->s_wt_pred.wpred_log_wdc; |
6276 | 0 | ps_inter_ctb_prms->u1_max_tr_depth = ps_thrd_ctxt->s_init_prms.u1_max_tr_depth; |
6277 | 0 | ps_inter_ctb_prms->i1_quality_preset = e_me_quality_presets; |
6278 | 0 | ps_inter_ctb_prms->i4_bidir_enabled = ps_subpel_prms->bidir_enabled; |
6279 | 0 | ps_inter_ctb_prms->i4_inp_stride = ps_subpel_prms->i4_inp_stride; |
6280 | 0 | ps_inter_ctb_prms->u1_num_ref = ps_subpel_prms->u1_num_ref; |
6281 | 0 | ps_inter_ctb_prms->u1_use_satd = ps_subpel_prms->i4_use_satd; |
6282 | 0 | ps_inter_ctb_prms->i4_rec_stride = ps_curr_layer->i4_rec_stride; |
6283 | 0 | ps_inter_ctb_prms->u1_num_active_ref_l0 = ps_ctxt->s_frm_prms.u1_num_active_ref_l0; |
6284 | 0 | ps_inter_ctb_prms->u1_num_active_ref_l1 = ps_ctxt->s_frm_prms.u1_num_active_ref_l1; |
6285 | 0 | ps_inter_ctb_prms->i4_lamda = lambda_recon; |
6286 | 0 | ps_inter_ctb_prms->u1_lamda_qshift = ps_refine_prms->lambda_q_shift; |
6287 | 0 | ps_inter_ctb_prms->i4_qstep_ls8 = ps_ctxt->ps_hme_frm_prms->qstep_ls8; |
6288 | 0 | ps_inter_ctb_prms->pi4_inv_wt = ps_ctxt->s_wt_pred.a_inv_wpred_wt; |
6289 | 0 | ps_inter_ctb_prms->pi1_past_list = ps_ctxt->ai1_past_list; |
6290 | 0 | ps_inter_ctb_prms->pi1_future_list = ps_ctxt->ai1_future_list; |
6291 | 0 | ps_inter_ctb_prms->pu4_src_variance = s_search_prms_blk.au4_src_variance; |
6292 | 0 | ps_inter_ctb_prms->u1_max_2nx2n_tu_recur_cands = |
6293 | 0 | ps_refine_prms->u1_max_2nx2n_tu_recur_cands; |
6294 | 0 | } |
6295 | |
|
6296 | 0 | for(i = 0; i < MAX_INIT_CANDTS; i++) |
6297 | 0 | { |
6298 | 0 | ps_search_candts[i].ps_search_node = &ps_ctxt->s_init_search_node[i]; |
6299 | 0 | ps_search_candts[i].ps_search_node->ps_mv = &ps_ctxt->as_search_cand_mv[i]; |
6300 | |
|
6301 | 0 | INIT_SEARCH_NODE(ps_search_candts[i].ps_search_node, 0); |
6302 | 0 | } |
6303 | 0 | num_act_ref_pics = |
6304 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l0 + ps_ctxt->s_frm_prms.u1_num_active_ref_l1; |
6305 | |
|
6306 | 0 | if(num_act_ref_pics) |
6307 | 0 | { |
6308 | 0 | hme_search_cand_data_init( |
6309 | 0 | ai4_id_Z, |
6310 | 0 | ai4_id_coloc, |
6311 | 0 | ai4_num_coloc_cands, |
6312 | 0 | au1_search_candidate_list_index, |
6313 | 0 | i4_num_act_ref_l0, |
6314 | 0 | i4_num_act_ref_l1, |
6315 | 0 | ps_ctxt->s_frm_prms.bidir_enabled, |
6316 | 0 | blk_4x4_to_16x16); |
6317 | 0 | } |
6318 | |
|
6319 | 0 | if(!ps_ctxt->s_frm_prms.bidir_enabled && (i4_num_act_ref_l0 > 1)) |
6320 | 0 | { |
6321 | 0 | ps_search_candts[ai4_id_Z[0]].ps_search_node->i1_ref_idx = ps_ctxt->ai1_past_list[0]; |
6322 | 0 | ps_search_candts[ai4_id_Z[1]].ps_search_node->i1_ref_idx = ps_ctxt->ai1_past_list[1]; |
6323 | 0 | } |
6324 | 0 | else if(!ps_ctxt->s_frm_prms.bidir_enabled && (i4_num_act_ref_l0 == 1)) |
6325 | 0 | { |
6326 | 0 | ps_search_candts[ai4_id_Z[0]].ps_search_node->i1_ref_idx = ps_ctxt->ai1_past_list[0]; |
6327 | 0 | } |
6328 | |
|
6329 | 0 | for(i = 0; i < 3; i++) |
6330 | 0 | { |
6331 | 0 | search_node_t *ps_search_node; |
6332 | 0 | ps_search_node = &as_left_neighbours[i]; |
6333 | 0 | INIT_SEARCH_NODE(ps_search_node, 0); |
6334 | 0 | ps_search_node = &as_top_neighbours[i]; |
6335 | 0 | INIT_SEARCH_NODE(ps_search_node, 0); |
6336 | 0 | } |
6337 | |
|
6338 | 0 | INIT_SEARCH_NODE(&as_top_neighbours[3], 0); |
6339 | 0 | as_left_neighbours[2].u1_is_avail = 0; |
6340 | | |
6341 | | /*************************************************************************/ |
6342 | | /* Initialize all the search results structure here. We update all the */ |
6343 | | /* search results to default values, and configure things like blk sizes */ |
6344 | | /*************************************************************************/ |
6345 | 0 | if(num_act_ref_pics) |
6346 | 0 | { |
6347 | 0 | S32 i4_x, i4_y; |
6348 | | /* 16x16 results */ |
6349 | 0 | for(i = 0; i < 16; i++) |
6350 | 0 | { |
6351 | 0 | search_results_t *ps_search_results; |
6352 | 0 | S32 pred_lx; |
6353 | 0 | ps_search_results = &ps_ctxt->as_search_results_16x16[i]; |
6354 | 0 | i4_x = (S32)gau1_encode_to_raster_x[i]; |
6355 | 0 | i4_y = (S32)gau1_encode_to_raster_y[i]; |
6356 | 0 | i4_x <<= 4; |
6357 | 0 | i4_y <<= 4; |
6358 | |
|
6359 | 0 | hme_init_search_results( |
6360 | 0 | ps_search_results, |
6361 | 0 | i4_num_pred_dir, |
6362 | 0 | ps_refine_prms->i4_num_fpel_results, |
6363 | 0 | ps_refine_prms->i4_num_results_per_part, |
6364 | 0 | e_search_blk_size, |
6365 | 0 | i4_x, |
6366 | 0 | i4_y, |
6367 | 0 | &ps_ctxt->au1_is_past[0]); |
6368 | |
|
6369 | 0 | for(pred_lx = 0; pred_lx < 2; pred_lx++) |
6370 | 0 | { |
6371 | 0 | pred_ctxt_t *ps_pred_ctxt; |
6372 | |
|
6373 | 0 | ps_pred_ctxt = &ps_search_results->as_pred_ctxt[pred_lx]; |
6374 | |
|
6375 | 0 | hme_init_pred_ctxt_encode( |
6376 | 0 | ps_pred_ctxt, |
6377 | 0 | ps_search_results, |
6378 | 0 | ps_search_candts[ai4_id_coloc[0]].ps_search_node, |
6379 | 0 | ps_search_candts[ai4_id_Z[0]].ps_search_node, |
6380 | 0 | aps_mv_grid[pred_lx], |
6381 | 0 | pred_lx, |
6382 | 0 | lambda_recon, |
6383 | 0 | ps_refine_prms->lambda_q_shift, |
6384 | 0 | &ps_ctxt->apu1_ref_bits_tlu_lc[0], |
6385 | 0 | &ps_ctxt->ai2_ref_scf[0]); |
6386 | 0 | } |
6387 | 0 | } |
6388 | |
|
6389 | 0 | for(i = 0; i < 4; i++) |
6390 | 0 | { |
6391 | 0 | search_results_t *ps_search_results; |
6392 | 0 | S32 pred_lx; |
6393 | 0 | ps_search_results = &ps_ctxt->as_search_results_32x32[i]; |
6394 | |
|
6395 | 0 | i4_x = (S32)gau1_encode_to_raster_x[i]; |
6396 | 0 | i4_y = (S32)gau1_encode_to_raster_y[i]; |
6397 | 0 | i4_x <<= 5; |
6398 | 0 | i4_y <<= 5; |
6399 | |
|
6400 | 0 | hme_init_search_results( |
6401 | 0 | ps_search_results, |
6402 | 0 | i4_num_pred_dir, |
6403 | 0 | ps_refine_prms->i4_num_32x32_merge_results, |
6404 | 0 | ps_refine_prms->i4_num_results_per_part, |
6405 | 0 | BLK_32x32, |
6406 | 0 | i4_x, |
6407 | 0 | i4_y, |
6408 | 0 | &ps_ctxt->au1_is_past[0]); |
6409 | |
|
6410 | 0 | for(pred_lx = 0; pred_lx < 2; pred_lx++) |
6411 | 0 | { |
6412 | 0 | pred_ctxt_t *ps_pred_ctxt; |
6413 | |
|
6414 | 0 | ps_pred_ctxt = &ps_search_results->as_pred_ctxt[pred_lx]; |
6415 | |
|
6416 | 0 | hme_init_pred_ctxt_encode( |
6417 | 0 | ps_pred_ctxt, |
6418 | 0 | ps_search_results, |
6419 | 0 | ps_search_candts[ai4_id_coloc[0]].ps_search_node, |
6420 | 0 | ps_search_candts[ai4_id_Z[0]].ps_search_node, |
6421 | 0 | aps_mv_grid[pred_lx], |
6422 | 0 | pred_lx, |
6423 | 0 | lambda_recon, |
6424 | 0 | ps_refine_prms->lambda_q_shift, |
6425 | 0 | &ps_ctxt->apu1_ref_bits_tlu_lc[0], |
6426 | 0 | &ps_ctxt->ai2_ref_scf[0]); |
6427 | 0 | } |
6428 | 0 | } |
6429 | |
|
6430 | 0 | { |
6431 | 0 | search_results_t *ps_search_results; |
6432 | 0 | S32 pred_lx; |
6433 | 0 | ps_search_results = &ps_ctxt->s_search_results_64x64; |
6434 | |
|
6435 | 0 | hme_init_search_results( |
6436 | 0 | ps_search_results, |
6437 | 0 | i4_num_pred_dir, |
6438 | 0 | ps_refine_prms->i4_num_64x64_merge_results, |
6439 | 0 | ps_refine_prms->i4_num_results_per_part, |
6440 | 0 | BLK_64x64, |
6441 | 0 | 0, |
6442 | 0 | 0, |
6443 | 0 | &ps_ctxt->au1_is_past[0]); |
6444 | |
|
6445 | 0 | for(pred_lx = 0; pred_lx < 2; pred_lx++) |
6446 | 0 | { |
6447 | 0 | pred_ctxt_t *ps_pred_ctxt; |
6448 | |
|
6449 | 0 | ps_pred_ctxt = &ps_search_results->as_pred_ctxt[pred_lx]; |
6450 | |
|
6451 | 0 | hme_init_pred_ctxt_encode( |
6452 | 0 | ps_pred_ctxt, |
6453 | 0 | ps_search_results, |
6454 | 0 | ps_search_candts[ai4_id_coloc[0]].ps_search_node, |
6455 | 0 | ps_search_candts[ai4_id_Z[0]].ps_search_node, |
6456 | 0 | aps_mv_grid[pred_lx], |
6457 | 0 | pred_lx, |
6458 | 0 | lambda_recon, |
6459 | 0 | ps_refine_prms->lambda_q_shift, |
6460 | 0 | &ps_ctxt->apu1_ref_bits_tlu_lc[0], |
6461 | 0 | &ps_ctxt->ai2_ref_scf[0]); |
6462 | 0 | } |
6463 | 0 | } |
6464 | 0 | } |
6465 | | |
6466 | | /* Initialise the structure used in clustering */ |
6467 | 0 | if(ME_PRISTINE_QUALITY == e_me_quality_presets) |
6468 | 0 | { |
6469 | 0 | ps_ctb_cluster_info = ps_ctxt->ps_ctb_cluster_info; |
6470 | |
|
6471 | 0 | ps_ctb_cluster_info->ps_16x16_blk = ps_ctxt->ps_blk_16x16; |
6472 | 0 | ps_ctb_cluster_info->ps_32x32_blk = ps_ctxt->ps_blk_32x32; |
6473 | 0 | ps_ctb_cluster_info->ps_64x64_blk = ps_ctxt->ps_blk_64x64; |
6474 | 0 | ps_ctb_cluster_info->pi4_blk_8x8_mask = ai4_blk_8x8_mask; |
6475 | 0 | ps_ctb_cluster_info->sdi_threshold = ps_refine_prms->sdi_threshold; |
6476 | 0 | ps_ctb_cluster_info->i4_frame_qstep = ps_ctxt->frm_qstep; |
6477 | 0 | ps_ctb_cluster_info->i4_frame_qstep_multiplier = 16; |
6478 | 0 | } |
6479 | | |
6480 | | /*********************************************************************/ |
6481 | | /* Initialize the dyn. search range params. for each reference index */ |
6482 | | /* in current layer ctxt */ |
6483 | | /*********************************************************************/ |
6484 | | |
6485 | | /* Only for P pic. For P, both are 0, I&B has them mut. exclusive */ |
6486 | 0 | if(ps_ctxt->s_frm_prms.is_i_pic == ps_ctxt->s_frm_prms.bidir_enabled) |
6487 | 0 | { |
6488 | 0 | WORD32 ref_ctr; |
6489 | | /* set no. of act ref in L0 for further use at frame level */ |
6490 | 0 | ps_ctxt->as_l0_dyn_range_prms[i4_idx_dvsr_p].i4_num_act_ref_in_l0 = |
6491 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l0; |
6492 | |
|
6493 | 0 | for(ref_ctr = 0; ref_ctr < ps_ctxt->s_frm_prms.u1_num_active_ref_l0; ref_ctr++) |
6494 | 0 | { |
6495 | 0 | INIT_DYN_SEARCH_PRMS( |
6496 | 0 | &ps_ctxt->as_l0_dyn_range_prms[i4_idx_dvsr_p].as_dyn_range_prms[ref_ctr], |
6497 | 0 | ps_ctxt->ai4_ref_idx_to_poc_lc[ref_ctr]); |
6498 | 0 | } |
6499 | 0 | } |
6500 | | /*************************************************************************/ |
6501 | | /* Now that the candidates have been ordered, to choose the right number */ |
6502 | | /* of initial candidates. */ |
6503 | | /*************************************************************************/ |
6504 | 0 | if(blk_4x4_to_16x16) |
6505 | 0 | { |
6506 | 0 | if(i4_num_ref_prev_layer > 2) |
6507 | 0 | { |
6508 | 0 | if(e_search_complexity == SEARCH_CX_LOW) |
6509 | 0 | num_init_candts = 7 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6510 | 0 | else if(e_search_complexity == SEARCH_CX_MED) |
6511 | 0 | num_init_candts = 14 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6512 | 0 | else if(e_search_complexity == SEARCH_CX_HIGH) |
6513 | 0 | num_init_candts = 21 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6514 | 0 | else |
6515 | 0 | ASSERT(0); |
6516 | 0 | } |
6517 | 0 | else if(i4_num_ref_prev_layer == 2) |
6518 | 0 | { |
6519 | 0 | if(e_search_complexity == SEARCH_CX_LOW) |
6520 | 0 | num_init_candts = 5 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6521 | 0 | else if(e_search_complexity == SEARCH_CX_MED) |
6522 | 0 | num_init_candts = 12 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6523 | 0 | else if(e_search_complexity == SEARCH_CX_HIGH) |
6524 | 0 | num_init_candts = 19 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6525 | 0 | else |
6526 | 0 | ASSERT(0); |
6527 | 0 | } |
6528 | 0 | else |
6529 | 0 | { |
6530 | 0 | if(e_search_complexity == SEARCH_CX_LOW) |
6531 | 0 | num_init_candts = 5; |
6532 | 0 | else if(e_search_complexity == SEARCH_CX_MED) |
6533 | 0 | num_init_candts = 12; |
6534 | 0 | else if(e_search_complexity == SEARCH_CX_HIGH) |
6535 | 0 | num_init_candts = 19; |
6536 | 0 | else |
6537 | 0 | ASSERT(0); |
6538 | 0 | } |
6539 | 0 | } |
6540 | 0 | else |
6541 | 0 | { |
6542 | 0 | if(i4_num_ref_prev_layer > 2) |
6543 | 0 | { |
6544 | 0 | if(e_search_complexity == SEARCH_CX_LOW) |
6545 | 0 | num_init_candts = 7 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6546 | 0 | else if(e_search_complexity == SEARCH_CX_MED) |
6547 | 0 | num_init_candts = 13 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6548 | 0 | else if(e_search_complexity == SEARCH_CX_HIGH) |
6549 | 0 | num_init_candts = 18 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6550 | 0 | else |
6551 | 0 | ASSERT(0); |
6552 | 0 | } |
6553 | 0 | else if(i4_num_ref_prev_layer == 2) |
6554 | 0 | { |
6555 | 0 | if(e_search_complexity == SEARCH_CX_LOW) |
6556 | 0 | num_init_candts = 5 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6557 | 0 | else if(e_search_complexity == SEARCH_CX_MED) |
6558 | 0 | num_init_candts = 11 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6559 | 0 | else if(e_search_complexity == SEARCH_CX_HIGH) |
6560 | 0 | num_init_candts = 16 * (!ps_ctxt->s_frm_prms.bidir_enabled + 1); |
6561 | 0 | else |
6562 | 0 | ASSERT(0); |
6563 | 0 | } |
6564 | 0 | else |
6565 | 0 | { |
6566 | 0 | if(e_search_complexity == SEARCH_CX_LOW) |
6567 | 0 | num_init_candts = 5; |
6568 | 0 | else if(e_search_complexity == SEARCH_CX_MED) |
6569 | 0 | num_init_candts = 11; |
6570 | 0 | else if(e_search_complexity == SEARCH_CX_HIGH) |
6571 | 0 | num_init_candts = 16; |
6572 | 0 | else |
6573 | 0 | ASSERT(0); |
6574 | 0 | } |
6575 | 0 | } |
6576 | | |
6577 | | /*************************************************************************/ |
6578 | | /* The following search parameters are fixed throughout the search across*/ |
6579 | | /* all blks. So these are configured outside processing loop */ |
6580 | | /*************************************************************************/ |
6581 | 0 | s_search_prms_blk.i4_num_init_candts = num_init_candts; |
6582 | 0 | s_search_prms_blk.i4_start_step = 1; |
6583 | 0 | s_search_prms_blk.i4_use_satd = 0; |
6584 | 0 | s_search_prms_blk.i4_num_steps_post_refine = ps_refine_prms->i4_num_steps_post_refine_fpel; |
6585 | | /* we use recon only for encoded layers, otherwise it is not available */ |
6586 | 0 | s_search_prms_blk.i4_use_rec = ps_refine_prms->i4_encode & ps_refine_prms->i4_use_rec_in_fpel; |
6587 | |
|
6588 | 0 | s_search_prms_blk.ps_search_candts = ps_search_candts; |
6589 | 0 | if(s_search_prms_blk.i4_use_rec) |
6590 | 0 | { |
6591 | 0 | WORD32 ref_ctr; |
6592 | 0 | for(ref_ctr = 0; ref_ctr < MAX_NUM_REF; ref_ctr++) |
6593 | 0 | s_search_prms_blk.aps_mv_range[ref_ctr] = &as_range_prms_rec[ref_ctr]; |
6594 | 0 | } |
6595 | 0 | else |
6596 | 0 | { |
6597 | 0 | WORD32 ref_ctr; |
6598 | 0 | for(ref_ctr = 0; ref_ctr < MAX_NUM_REF; ref_ctr++) |
6599 | 0 | s_search_prms_blk.aps_mv_range[ref_ctr] = &as_range_prms_inp[ref_ctr]; |
6600 | 0 | } |
6601 | | |
6602 | | /*************************************************************************/ |
6603 | | /* Initialize coordinates. Meaning as follows */ |
6604 | | /* blk_x : x coordinate of the 16x16 blk, in terms of number of blks */ |
6605 | | /* blk_y : same as above, y coord. */ |
6606 | | /* num_blks_in_this_ctb : number of blks in this given ctb that starts */ |
6607 | | /* at i4_ctb_x, i4_ctb_y. This may not be 16 at picture boundaries. */ |
6608 | | /* i4_ctb_x, i4_ctb_y: pixel coordinate of the ctb realtive to top left */ |
6609 | | /* corner of the picture. Always multiple of 64. */ |
6610 | | /* blk_id_in_ctb : encode order id of the blk in the ctb. */ |
6611 | | /*************************************************************************/ |
6612 | 0 | blk_y = 0; |
6613 | 0 | blk_id_in_ctb = 0; |
6614 | 0 | i4_ctb_y = 0; |
6615 | | |
6616 | | /*************************************************************************/ |
6617 | | /* Picture limit on all 4 sides. This will be used to set mv limits for */ |
6618 | | /* every block given its coordinate. Note thsi assumes that the min amt */ |
6619 | | /* of padding to right of pic is equal to the blk size. If we go all the */ |
6620 | | /* way upto 64x64, then the min padding on right size of picture should */ |
6621 | | /* be 64, and also on bottom side of picture. */ |
6622 | | /*************************************************************************/ |
6623 | 0 | SET_PIC_LIMIT( |
6624 | 0 | s_pic_limit_inp, |
6625 | 0 | ps_curr_layer->i4_pad_x_rec, |
6626 | 0 | ps_curr_layer->i4_pad_y_rec, |
6627 | 0 | ps_curr_layer->i4_wd, |
6628 | 0 | ps_curr_layer->i4_ht, |
6629 | 0 | s_search_prms_blk.i4_num_steps_post_refine); |
6630 | |
|
6631 | 0 | SET_PIC_LIMIT( |
6632 | 0 | s_pic_limit_rec, |
6633 | 0 | ps_curr_layer->i4_pad_x_rec, |
6634 | 0 | ps_curr_layer->i4_pad_y_rec, |
6635 | 0 | ps_curr_layer->i4_wd, |
6636 | 0 | ps_curr_layer->i4_ht, |
6637 | 0 | s_search_prms_blk.i4_num_steps_post_refine); |
6638 | | |
6639 | | /*************************************************************************/ |
6640 | | /* set the MV limit per ref. pic. */ |
6641 | | /* - P pic. : Based on the config params. */ |
6642 | | /* - B/b pic: Based on the Max/Min MV from prev. P and config. param. */ |
6643 | | /*************************************************************************/ |
6644 | 0 | hme_set_mv_limit_using_dvsr_data( |
6645 | 0 | ps_ctxt, ps_curr_layer, as_mv_limit, &i2_prev_enc_frm_max_mv_y, num_act_ref_pics); |
6646 | 0 | s_srch_cand_init_data.pu1_num_fpel_search_cands = ps_refine_prms->au1_num_fpel_search_cands; |
6647 | 0 | s_srch_cand_init_data.i4_num_act_ref_l0 = ps_ctxt->s_frm_prms.u1_num_active_ref_l0; |
6648 | 0 | s_srch_cand_init_data.i4_num_act_ref_l1 = ps_ctxt->s_frm_prms.u1_num_active_ref_l1; |
6649 | 0 | s_srch_cand_init_data.ps_coarse_layer = ps_coarse_layer; |
6650 | 0 | s_srch_cand_init_data.ps_curr_layer = ps_curr_layer; |
6651 | 0 | s_srch_cand_init_data.i4_max_num_init_cands = num_init_candts; |
6652 | 0 | s_srch_cand_init_data.ps_search_cands = ps_search_candts; |
6653 | 0 | s_srch_cand_init_data.u1_num_results_in_mvbank = s_mv_update_prms.i4_num_results_to_store; |
6654 | 0 | s_srch_cand_init_data.pi4_ref_id_lc_to_l0_map = ps_ctxt->a_ref_idx_lc_to_l0; |
6655 | 0 | s_srch_cand_init_data.pi4_ref_id_lc_to_l1_map = ps_ctxt->a_ref_idx_lc_to_l1; |
6656 | 0 | s_srch_cand_init_data.e_search_blk_size = e_search_blk_size; |
6657 | |
|
6658 | 0 | while(0 == end_of_frame) |
6659 | 0 | { |
6660 | 0 | job_queue_t *ps_job; |
6661 | 0 | frm_ctb_ctxt_t *ps_frm_ctb_prms; |
6662 | 0 | ipe_l0_ctb_analyse_for_me_t *ps_cur_ipe_ctb; |
6663 | |
|
6664 | 0 | WORD32 i4_max_mv_x_in_ctb; |
6665 | 0 | WORD32 i4_max_mv_y_in_ctb; |
6666 | 0 | void *pv_dep_mngr_encloop_dep_me; |
6667 | 0 | WORD32 offset_val, check_dep_pos, set_dep_pos; |
6668 | 0 | WORD32 left_ctb_in_diff_tile, i4_first_ctb_x = 0; |
6669 | |
|
6670 | 0 | pv_dep_mngr_encloop_dep_me = ps_ctxt->pv_dep_mngr_encloop_dep_me; |
6671 | |
|
6672 | 0 | ps_frm_ctb_prms = (frm_ctb_ctxt_t *)ps_thrd_ctxt->pv_ext_frm_prms; |
6673 | | |
6674 | | /* Get the current row from the job queue */ |
6675 | 0 | ps_job = (job_queue_t *)ihevce_enc_grp_get_next_job( |
6676 | 0 | ps_multi_thrd_ctxt, lyr_job_type, 1, me_frm_id); |
6677 | | |
6678 | | /* If all rows are done, set the end of process flag to 1, */ |
6679 | | /* and the current row to -1 */ |
6680 | 0 | if(NULL == ps_job) |
6681 | 0 | { |
6682 | 0 | blk_y = -1; |
6683 | 0 | i4_ctb_y = -1; |
6684 | 0 | tile_col_idx = -1; |
6685 | 0 | end_of_frame = 1; |
6686 | |
|
6687 | 0 | continue; |
6688 | 0 | } |
6689 | | |
6690 | | /* set the output dependency after picking up the row */ |
6691 | 0 | ihevce_enc_grp_job_set_out_dep(ps_multi_thrd_ctxt, ps_job, me_frm_id); |
6692 | | |
6693 | | /* Obtain the current row's details from the job */ |
6694 | 0 | { |
6695 | 0 | ihevce_tile_params_t *ps_col_tile_params; |
6696 | |
|
6697 | 0 | i4_ctb_y = ps_job->s_job_info.s_me_job_info.i4_vert_unit_row_no; |
6698 | | /* Obtain the current colum tile index from the job */ |
6699 | 0 | tile_col_idx = ps_job->s_job_info.s_me_job_info.i4_tile_col_idx; |
6700 | | |
6701 | | /* in encode layer block are 16x16 and CTB is 64 x 64 */ |
6702 | | /* note if ctb is 32x32 the this calc needs to be changed */ |
6703 | 0 | num_sync_units_in_row = (i4_pic_wd + ((1 << ps_ctxt->log_ctb_size) - 1)) >> |
6704 | 0 | ps_ctxt->log_ctb_size; |
6705 | | |
6706 | | /* The tile parameter for the col. idx. Use only the properties |
6707 | | which is same for all the bottom tiles like width, start_x, etc. |
6708 | | Don't use height, start_y, etc. */ |
6709 | 0 | ps_col_tile_params = |
6710 | 0 | ((ihevce_tile_params_t *)ps_thrd_ctxt->pv_tile_params_base + tile_col_idx); |
6711 | | /* in encode layer block are 16x16 and CTB is 64 x 64 */ |
6712 | | /* note if ctb is 32x32 the this calc needs to be changed */ |
6713 | 0 | num_sync_units_in_tile = |
6714 | 0 | (ps_col_tile_params->i4_curr_tile_width + ((1 << ps_ctxt->log_ctb_size) - 1)) >> |
6715 | 0 | ps_ctxt->log_ctb_size; |
6716 | |
|
6717 | 0 | i4_first_ctb_x = ps_col_tile_params->i4_first_ctb_x; |
6718 | 0 | i4_ctb_x = i4_first_ctb_x; |
6719 | |
|
6720 | 0 | if(!num_act_ref_pics) |
6721 | 0 | { |
6722 | 0 | for(i4_ctb_x = i4_first_ctb_x; |
6723 | 0 | i4_ctb_x < (ps_col_tile_params->i4_first_ctb_x + num_sync_units_in_tile); |
6724 | 0 | i4_ctb_x++) |
6725 | 0 | { |
6726 | 0 | S32 blk_i = 0, blk_j = 0; |
6727 | | /* set the dependency for the corresponding row in enc loop */ |
6728 | 0 | ihevce_dmgr_set_row_row_sync( |
6729 | 0 | pv_dep_mngr_encloop_dep_me, |
6730 | 0 | (i4_ctb_x + 1), |
6731 | 0 | i4_ctb_y, |
6732 | 0 | tile_col_idx /* Col Tile No. */); |
6733 | 0 | } |
6734 | |
|
6735 | 0 | continue; |
6736 | 0 | } |
6737 | | |
6738 | | /* increment the number of rows proc */ |
6739 | 0 | num_rows_proc++; |
6740 | | |
6741 | | /* Set Variables for Dep. Checking and Setting */ |
6742 | 0 | set_dep_pos = i4_ctb_y + 1; |
6743 | 0 | if(i4_ctb_y > 0) |
6744 | 0 | { |
6745 | 0 | offset_val = 2; |
6746 | 0 | check_dep_pos = i4_ctb_y - 1; |
6747 | 0 | } |
6748 | 0 | else |
6749 | 0 | { |
6750 | | /* First row should run without waiting */ |
6751 | 0 | offset_val = -1; |
6752 | 0 | check_dep_pos = 0; |
6753 | 0 | } |
6754 | | |
6755 | | /* row ctb out pointer */ |
6756 | 0 | ps_ctxt->ps_ctb_analyse_curr_row = |
6757 | 0 | ps_ctxt->ps_ctb_analyse_base + i4_ctb_y * ps_frm_ctb_prms->i4_num_ctbs_horz; |
6758 | | |
6759 | | /* Row level CU Tree buffer */ |
6760 | 0 | ps_ctxt->ps_cu_tree_curr_row = |
6761 | 0 | ps_ctxt->ps_cu_tree_base + |
6762 | 0 | i4_ctb_y * ps_frm_ctb_prms->i4_num_ctbs_horz * MAX_NUM_NODES_CU_TREE; |
6763 | |
|
6764 | 0 | ps_ctxt->ps_me_ctb_data_curr_row = |
6765 | 0 | ps_ctxt->ps_me_ctb_data_base + i4_ctb_y * ps_frm_ctb_prms->i4_num_ctbs_horz; |
6766 | 0 | } |
6767 | | |
6768 | | /* This flag says the CTB under processing is at the start of tile in horz dir.*/ |
6769 | 0 | left_ctb_in_diff_tile = 1; |
6770 | | |
6771 | | /* To make sure no 64-bit overflow happens when inv_wt is multiplied with un-normalized src_var, */ |
6772 | | /* the shift value will be passed onto the functions wherever inv_wt isused so that inv_wt is appropriately shift and multiplied */ |
6773 | 0 | { |
6774 | 0 | S32 i4_ref_id, i4_bits_req; |
6775 | |
|
6776 | 0 | for(i4_ref_id = 0; i4_ref_id < (ps_ctxt->s_frm_prms.u1_num_active_ref_l0 + |
6777 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l1); |
6778 | 0 | i4_ref_id++) |
6779 | 0 | { |
6780 | 0 | GETRANGE(i4_bits_req, ps_ctxt->s_wt_pred.a_inv_wpred_wt[i4_ref_id]); |
6781 | |
|
6782 | 0 | if(i4_bits_req > 12) |
6783 | 0 | { |
6784 | 0 | ps_ctxt->s_wt_pred.ai4_shift_val[i4_ref_id] = (i4_bits_req - 12); |
6785 | 0 | } |
6786 | 0 | else |
6787 | 0 | { |
6788 | 0 | ps_ctxt->s_wt_pred.ai4_shift_val[i4_ref_id] = 0; |
6789 | 0 | } |
6790 | 0 | } |
6791 | |
|
6792 | 0 | s_common_frm_prms.pi4_inv_wt_shift_val = ps_ctxt->s_wt_pred.ai4_shift_val; |
6793 | 0 | } |
6794 | | |
6795 | | /* if non-encode layer then i4_ctb_x will be same as blk_x */ |
6796 | | /* loop over all the units is a row */ |
6797 | 0 | for(i4_ctb_x = i4_first_ctb_x; i4_ctb_x < (i4_first_ctb_x + num_sync_units_in_tile); |
6798 | 0 | i4_ctb_x++) |
6799 | 0 | { |
6800 | 0 | ihevce_ctb_noise_params *ps_ctb_noise_params = |
6801 | 0 | &ps_ctxt->ps_ctb_analyse_curr_row[i4_ctb_x].s_ctb_noise_params; |
6802 | |
|
6803 | 0 | s_common_frm_prms.i4_ctb_x_off = i4_ctb_x << 6; |
6804 | 0 | s_common_frm_prms.i4_ctb_y_off = i4_ctb_y << 6; |
6805 | |
|
6806 | 0 | ps_ctxt->s_mc_ctxt.i4_ctb_frm_pos_y = i4_ctb_y << 6; |
6807 | 0 | ps_ctxt->s_mc_ctxt.i4_ctb_frm_pos_x = i4_ctb_x << 6; |
6808 | | /* Initialize ptr to current IPE CTB */ |
6809 | 0 | ps_cur_ipe_ctb = ps_ctxt->ps_ipe_l0_ctb_frm_base + i4_ctb_x + |
6810 | 0 | i4_ctb_y * ps_frm_ctb_prms->i4_num_ctbs_horz; |
6811 | 0 | { |
6812 | 0 | ps_ctb_bound_attrs = |
6813 | 0 | get_ctb_attrs(i4_ctb_x << 6, i4_ctb_y << 6, i4_pic_wd, i4_pic_ht, ps_ctxt); |
6814 | |
|
6815 | 0 | en_merge_32x32 = ps_ctb_bound_attrs->u1_merge_to_32x32_flag; |
6816 | 0 | num_blks_in_this_ctb = ps_ctb_bound_attrs->u1_num_blks_in_ctb; |
6817 | 0 | } |
6818 | | |
6819 | | /* Block to initialise pointers to part_type_results_t */ |
6820 | | /* in each size-specific inter_cu_results_t */ |
6821 | 0 | { |
6822 | 0 | WORD32 i; |
6823 | |
|
6824 | 0 | for(i = 0; i < 64; i++) |
6825 | 0 | { |
6826 | 0 | ps_ctxt->as_cu8x8_results[i].ps_best_results = |
6827 | 0 | ps_ctxt->ps_me_ctb_data_curr_row[i4_ctb_x] |
6828 | 0 | .as_8x8_block_data[i] |
6829 | 0 | .as_best_results; |
6830 | 0 | ps_ctxt->as_cu8x8_results[i].u1_num_best_results = 0; |
6831 | 0 | } |
6832 | |
|
6833 | 0 | for(i = 0; i < 16; i++) |
6834 | 0 | { |
6835 | 0 | ps_ctxt->as_cu16x16_results[i].ps_best_results = |
6836 | 0 | ps_ctxt->ps_me_ctb_data_curr_row[i4_ctb_x].as_block_data[i].as_best_results; |
6837 | 0 | ps_ctxt->as_cu16x16_results[i].u1_num_best_results = 0; |
6838 | 0 | } |
6839 | |
|
6840 | 0 | for(i = 0; i < 4; i++) |
6841 | 0 | { |
6842 | 0 | ps_ctxt->as_cu32x32_results[i].ps_best_results = |
6843 | 0 | ps_ctxt->ps_me_ctb_data_curr_row[i4_ctb_x] |
6844 | 0 | .as_32x32_block_data[i] |
6845 | 0 | .as_best_results; |
6846 | 0 | ps_ctxt->as_cu32x32_results[i].u1_num_best_results = 0; |
6847 | 0 | } |
6848 | |
|
6849 | 0 | ps_ctxt->s_cu64x64_results.ps_best_results = |
6850 | 0 | ps_ctxt->ps_me_ctb_data_curr_row[i4_ctb_x].s_64x64_block_data.as_best_results; |
6851 | 0 | ps_ctxt->s_cu64x64_results.u1_num_best_results = 0; |
6852 | 0 | } |
6853 | |
|
6854 | 0 | if(ME_PRISTINE_QUALITY == e_me_quality_presets) |
6855 | 0 | { |
6856 | 0 | ps_ctb_cluster_info->blk_32x32_mask = en_merge_32x32; |
6857 | 0 | ps_ctb_cluster_info->ps_cur_ipe_ctb = ps_cur_ipe_ctb; |
6858 | 0 | ps_ctb_cluster_info->ps_cu_tree_root = |
6859 | 0 | ps_ctxt->ps_cu_tree_curr_row + (i4_ctb_x * MAX_NUM_NODES_CU_TREE); |
6860 | 0 | ps_ctb_cluster_info->nodes_created_in_cu_tree = 1; |
6861 | 0 | } |
6862 | |
|
6863 | 0 | if(ME_PRISTINE_QUALITY != e_me_quality_presets) |
6864 | 0 | { |
6865 | 0 | S32 i4_nodes_created_in_cu_tree = 1; |
6866 | |
|
6867 | 0 | ihevce_cu_tree_init( |
6868 | 0 | (ps_ctxt->ps_cu_tree_curr_row + (i4_ctb_x * MAX_NUM_NODES_CU_TREE)), |
6869 | 0 | (ps_ctxt->ps_cu_tree_curr_row + (i4_ctb_x * MAX_NUM_NODES_CU_TREE)), |
6870 | 0 | &i4_nodes_created_in_cu_tree, |
6871 | 0 | 0, |
6872 | 0 | POS_NA, |
6873 | 0 | POS_NA, |
6874 | 0 | POS_NA); |
6875 | 0 | } |
6876 | |
|
6877 | 0 | memset(ai4_blk_8x8_mask, 0, 16 * sizeof(S32)); |
6878 | |
|
6879 | 0 | if(ps_refine_prms->u1_use_lambda_derived_from_min_8x8_act_in_ctb) |
6880 | 0 | { |
6881 | 0 | S32 j; |
6882 | |
|
6883 | 0 | ipe_l0_ctb_analyse_for_me_t *ps_cur_ipe_ctb; |
6884 | |
|
6885 | 0 | ps_cur_ipe_ctb = |
6886 | 0 | ps_ctxt->ps_ipe_l0_ctb_frm_base + i4_ctb_x + i4_ctb_y * num_sync_units_in_row; |
6887 | 0 | lambda_recon = |
6888 | 0 | hme_recompute_lambda_from_min_8x8_act_in_ctb(ps_ctxt, ps_cur_ipe_ctb); |
6889 | |
|
6890 | 0 | lambda_recon = ((float)lambda_recon * (100.0f - ME_LAMBDA_DISCOUNT) / 100.0f); |
6891 | |
|
6892 | 0 | for(i = 0; i < 4; i++) |
6893 | 0 | { |
6894 | 0 | ps_search_results = &ps_ctxt->as_search_results_32x32[i]; |
6895 | |
|
6896 | 0 | for(j = 0; j < 2; j++) |
6897 | 0 | { |
6898 | 0 | ps_search_results->as_pred_ctxt[j].lambda = lambda_recon; |
6899 | 0 | } |
6900 | 0 | } |
6901 | 0 | ps_search_results = &ps_ctxt->s_search_results_64x64; |
6902 | |
|
6903 | 0 | for(j = 0; j < 2; j++) |
6904 | 0 | { |
6905 | 0 | ps_search_results->as_pred_ctxt[j].lambda = lambda_recon; |
6906 | 0 | } |
6907 | |
|
6908 | 0 | s_common_frm_prms.i4_lamda = lambda_recon; |
6909 | 0 | } |
6910 | 0 | else |
6911 | 0 | { |
6912 | 0 | lambda_recon = ps_refine_prms->lambda_recon; |
6913 | 0 | } |
6914 | | |
6915 | | /*********************************************************************/ |
6916 | | /* replicate the inp buffer at blk or ctb level for each ref id, */ |
6917 | | /* Instead of searching with wk * ref(k), we search with Ik = I / wk */ |
6918 | | /* thereby avoiding a bloat up of memory. If we did all references */ |
6919 | | /* weighted pred, we will end up with a duplicate copy of each ref */ |
6920 | | /* at each layer, since we need to preserve the original reference. */ |
6921 | | /* ToDo: Need to observe performance with this mechanism and compare */ |
6922 | | /* with case where ref is weighted. */ |
6923 | | /*********************************************************************/ |
6924 | 0 | fp_get_wt_inp( |
6925 | 0 | ps_curr_layer, |
6926 | 0 | &ps_ctxt->s_wt_pred, |
6927 | 0 | unit_size, |
6928 | 0 | s_common_frm_prms.i4_ctb_x_off, |
6929 | 0 | s_common_frm_prms.i4_ctb_y_off, |
6930 | 0 | unit_size, |
6931 | 0 | ps_ctxt->num_ref_future + ps_ctxt->num_ref_past, |
6932 | 0 | ps_ctxt->i4_wt_pred_enable_flag); |
6933 | |
|
6934 | 0 | if(ps_thrd_ctxt->s_init_prms.u1_is_stasino_enabled) |
6935 | 0 | { |
6936 | 0 | #if TEMPORAL_NOISE_DETECT |
6937 | 0 | { |
6938 | 0 | WORD32 had_block_size = 16; |
6939 | 0 | WORD32 ctb_width = ((i4_pic_wd - s_common_frm_prms.i4_ctb_x_off) >= 64) |
6940 | 0 | ? 64 |
6941 | 0 | : i4_pic_wd - s_common_frm_prms.i4_ctb_x_off; |
6942 | 0 | WORD32 ctb_height = ((i4_pic_ht - s_common_frm_prms.i4_ctb_y_off) >= 64) |
6943 | 0 | ? 64 |
6944 | 0 | : i4_pic_ht - s_common_frm_prms.i4_ctb_y_off; |
6945 | 0 | WORD32 num_pred_dir = i4_num_pred_dir; |
6946 | 0 | WORD32 i4_x_off = s_common_frm_prms.i4_ctb_x_off; |
6947 | 0 | WORD32 i4_y_off = s_common_frm_prms.i4_ctb_y_off; |
6948 | |
|
6949 | 0 | WORD32 i; |
6950 | 0 | WORD32 noise_detected; |
6951 | 0 | WORD32 ctb_size; |
6952 | 0 | WORD32 num_comp_had_blocks; |
6953 | 0 | WORD32 noisy_block_cnt; |
6954 | 0 | WORD32 index_8x8_block; |
6955 | 0 | WORD32 num_8x8_in_ctb_row; |
6956 | |
|
6957 | 0 | WORD32 ht_offset; |
6958 | 0 | WORD32 wd_offset; |
6959 | 0 | WORD32 block_ht; |
6960 | 0 | WORD32 block_wd; |
6961 | |
|
6962 | 0 | WORD32 num_horz_blocks; |
6963 | 0 | WORD32 num_vert_blocks; |
6964 | |
|
6965 | 0 | WORD32 mean; |
6966 | 0 | UWORD32 variance_8x8; |
6967 | |
|
6968 | 0 | WORD32 hh_energy_percent; |
6969 | | |
6970 | | /* variables to hold the constant values. The variable values held are decided by the HAD block size */ |
6971 | 0 | WORD32 min_noisy_block_cnt; |
6972 | 0 | WORD32 min_coeffs_above_avg; |
6973 | 0 | WORD32 min_coeff_avg_energy; |
6974 | | |
6975 | | /* to store the mean and variance of each 8*8 block and find the variance of any higher block sizes later on. block */ |
6976 | 0 | WORD32 i4_cu_x_off, i4_cu_y_off; |
6977 | 0 | WORD32 is_noisy; |
6978 | | |
6979 | | /* intialise the variables holding the constants */ |
6980 | 0 | if(had_block_size == 8) |
6981 | 0 | { |
6982 | 0 | min_noisy_block_cnt = MIN_NOISY_BLOCKS_CNT_8x8; //6;// |
6983 | 0 | min_coeffs_above_avg = MIN_NUM_COEFFS_ABOVE_AVG_8x8; |
6984 | 0 | min_coeff_avg_energy = MIN_COEFF_AVG_ENERGY_8x8; |
6985 | 0 | } |
6986 | 0 | else |
6987 | 0 | { |
6988 | 0 | min_noisy_block_cnt = MIN_NOISY_BLOCKS_CNT_16x16; //7;// |
6989 | 0 | min_coeffs_above_avg = MIN_NUM_COEFFS_ABOVE_AVG_16x16; |
6990 | 0 | min_coeff_avg_energy = MIN_COEFF_AVG_ENERGY_16x16; |
6991 | 0 | } |
6992 | | |
6993 | | /* initialize the variables */ |
6994 | 0 | noise_detected = 0; |
6995 | 0 | noisy_block_cnt = 0; |
6996 | 0 | hh_energy_percent = 0; |
6997 | 0 | variance_8x8 = 0; |
6998 | 0 | block_ht = ctb_height; |
6999 | 0 | block_wd = ctb_width; |
7000 | |
|
7001 | 0 | mean = 0; |
7002 | |
|
7003 | 0 | ctb_size = block_ht * block_wd; //ctb_width * ctb_height; |
7004 | 0 | num_comp_had_blocks = ctb_size / (had_block_size * had_block_size); |
7005 | |
|
7006 | 0 | num_horz_blocks = block_wd / had_block_size; //ctb_width / had_block_size; |
7007 | 0 | num_vert_blocks = block_ht / had_block_size; //ctb_height / had_block_size; |
7008 | |
|
7009 | 0 | ht_offset = -had_block_size; |
7010 | 0 | wd_offset = -had_block_size; |
7011 | |
|
7012 | 0 | num_8x8_in_ctb_row = block_wd / 8; // number of 8x8 in this ctb |
7013 | 0 | for(i = 0; i < num_comp_had_blocks; i++) |
7014 | 0 | { |
7015 | 0 | if(i % num_horz_blocks == 0) |
7016 | 0 | { |
7017 | 0 | wd_offset = -had_block_size; |
7018 | 0 | ht_offset += had_block_size; |
7019 | 0 | } |
7020 | 0 | wd_offset += had_block_size; |
7021 | | |
7022 | | /* CU level offsets */ |
7023 | 0 | i4_cu_x_off = i4_x_off + (i % 4) * 16; //+ (i % 4) * 16 |
7024 | 0 | i4_cu_y_off = i4_y_off + (i / 4) * 16; |
7025 | | |
7026 | | /* if 50 % or more of the CU is noisy then the return value is 1 */ |
7027 | 0 | is_noisy = ihevce_determine_cu_noise_based_on_8x8Blk_data( |
7028 | 0 | ps_ctb_noise_params->au1_is_8x8Blk_noisy, |
7029 | 0 | (i % 4) * 16, |
7030 | 0 | (i / 4) * 16, |
7031 | 0 | 16); |
7032 | | |
7033 | | /* only if the CU is noisy then check the temporal noise detect call is made on the CU */ |
7034 | 0 | if(is_noisy) |
7035 | 0 | { |
7036 | 0 | index_8x8_block = (i / num_horz_blocks) * 2 * num_8x8_in_ctb_row + |
7037 | 0 | (i % num_horz_blocks) * 2; |
7038 | 0 | noisy_block_cnt += ihevce_16x16block_temporal_noise_detect( |
7039 | 0 | 16, |
7040 | 0 | ((i4_pic_wd - s_common_frm_prms.i4_ctb_x_off) >= 64) |
7041 | 0 | ? 64 |
7042 | 0 | : i4_pic_wd - s_common_frm_prms.i4_ctb_x_off, |
7043 | 0 | ((i4_pic_ht - s_common_frm_prms.i4_ctb_y_off) >= 64) |
7044 | 0 | ? 64 |
7045 | 0 | : i4_pic_ht - s_common_frm_prms.i4_ctb_y_off, |
7046 | 0 | ps_ctb_noise_params, |
7047 | 0 | &s_srch_cand_init_data, |
7048 | 0 | &s_search_prms_blk, |
7049 | 0 | ps_ctxt, |
7050 | 0 | num_pred_dir, |
7051 | 0 | i4_num_act_ref_l0, |
7052 | 0 | i4_num_act_ref_l1, |
7053 | 0 | i4_cu_x_off, |
7054 | 0 | i4_cu_y_off, |
7055 | 0 | &ps_ctxt->s_wt_pred, |
7056 | 0 | unit_size, |
7057 | 0 | index_8x8_block, |
7058 | 0 | num_horz_blocks, |
7059 | 0 | /*num_8x8_in_ctb_row*/ 8, // this should be a variable extra |
7060 | 0 | i); |
7061 | 0 | } /* if 16x16 is noisy */ |
7062 | 0 | } /* loop over for all 16x16*/ |
7063 | |
|
7064 | 0 | if(noisy_block_cnt >= min_noisy_block_cnt) |
7065 | 0 | { |
7066 | 0 | noise_detected = 1; |
7067 | 0 | } |
7068 | | |
7069 | | /* write back the noise presence detected for the current CTB to the structure */ |
7070 | 0 | ps_ctb_noise_params->i4_noise_present = noise_detected; |
7071 | 0 | } |
7072 | 0 | #endif |
7073 | |
|
7074 | | #if EVERYWHERE_NOISY && USE_NOISE_TERM_IN_L0_ME |
7075 | | if(ps_thrd_ctxt->s_init_prms.u1_is_stasino_enabled && |
7076 | | ps_ctb_noise_params->i4_noise_present) |
7077 | | { |
7078 | | memset( |
7079 | | ps_ctb_noise_params->au1_is_8x8Blk_noisy, |
7080 | | 1, |
7081 | | sizeof(ps_ctb_noise_params->au1_is_8x8Blk_noisy)); |
7082 | | } |
7083 | | #endif |
7084 | |
|
7085 | 0 | for(i = 0; i < 16; i++) |
7086 | 0 | { |
7087 | 0 | au1_is_16x16Blk_noisy[i] = ihevce_determine_cu_noise_based_on_8x8Blk_data( |
7088 | 0 | ps_ctb_noise_params->au1_is_8x8Blk_noisy, (i % 4) * 16, (i / 4) * 16, 16); |
7089 | 0 | } |
7090 | |
|
7091 | 0 | for(i = 0; i < 4; i++) |
7092 | 0 | { |
7093 | 0 | au1_is_32x32Blk_noisy[i] = ihevce_determine_cu_noise_based_on_8x8Blk_data( |
7094 | 0 | ps_ctb_noise_params->au1_is_8x8Blk_noisy, (i % 2) * 32, (i / 2) * 32, 32); |
7095 | 0 | } |
7096 | |
|
7097 | 0 | for(i = 0; i < 1; i++) |
7098 | 0 | { |
7099 | 0 | au1_is_64x64Blk_noisy[i] = ihevce_determine_cu_noise_based_on_8x8Blk_data( |
7100 | 0 | ps_ctb_noise_params->au1_is_8x8Blk_noisy, 0, 0, 64); |
7101 | 0 | } |
7102 | |
|
7103 | 0 | if(ps_ctxt->s_frm_prms.bidir_enabled && |
7104 | 0 | (ps_ctxt->s_frm_prms.i4_temporal_layer_id <= |
7105 | 0 | MAX_LAYER_ID_OF_B_PICS_WITHOUT_NOISE_DETECTION)) |
7106 | 0 | { |
7107 | 0 | ps_ctb_noise_params->i4_noise_present = 0; |
7108 | 0 | memset( |
7109 | 0 | ps_ctb_noise_params->au1_is_8x8Blk_noisy, |
7110 | 0 | 0, |
7111 | 0 | sizeof(ps_ctb_noise_params->au1_is_8x8Blk_noisy)); |
7112 | 0 | } |
7113 | |
|
7114 | 0 | #if ME_LAMBDA_DISCOUNT_WHEN_NOISY |
7115 | 0 | for(i = 0; i < 4; i++) |
7116 | 0 | { |
7117 | 0 | S32 j; |
7118 | 0 | S32 lambda; |
7119 | |
|
7120 | 0 | if(au1_is_32x32Blk_noisy[i]) |
7121 | 0 | { |
7122 | 0 | lambda = lambda_recon; |
7123 | 0 | lambda = |
7124 | 0 | ((float)lambda * (100.0f - ME_LAMBDA_DISCOUNT_WHEN_NOISY) / 100.0f); |
7125 | |
|
7126 | 0 | ps_search_results = &ps_ctxt->as_search_results_32x32[i]; |
7127 | |
|
7128 | 0 | for(j = 0; j < 2; j++) |
7129 | 0 | { |
7130 | 0 | ps_search_results->as_pred_ctxt[j].lambda = lambda; |
7131 | 0 | } |
7132 | 0 | } |
7133 | 0 | } |
7134 | |
|
7135 | 0 | { |
7136 | 0 | S32 j; |
7137 | 0 | S32 lambda; |
7138 | |
|
7139 | 0 | if(au1_is_64x64Blk_noisy[0]) |
7140 | 0 | { |
7141 | 0 | lambda = lambda_recon; |
7142 | 0 | lambda = |
7143 | 0 | ((float)lambda * (100.0f - ME_LAMBDA_DISCOUNT_WHEN_NOISY) / 100.0f); |
7144 | |
|
7145 | 0 | ps_search_results = &ps_ctxt->s_search_results_64x64; |
7146 | |
|
7147 | 0 | for(j = 0; j < 2; j++) |
7148 | 0 | { |
7149 | 0 | ps_search_results->as_pred_ctxt[j].lambda = lambda; |
7150 | 0 | } |
7151 | 0 | } |
7152 | 0 | } |
7153 | 0 | #endif |
7154 | 0 | if(au1_is_64x64Blk_noisy[0]) |
7155 | 0 | { |
7156 | 0 | U08 *pu1_inp = ps_curr_layer->pu1_inp + (s_common_frm_prms.i4_ctb_x_off + |
7157 | 0 | (s_common_frm_prms.i4_ctb_y_off * |
7158 | 0 | ps_curr_layer->i4_inp_stride)); |
7159 | |
|
7160 | 0 | hme_compute_sigmaX_and_sigmaXSquared( |
7161 | 0 | pu1_inp, |
7162 | 0 | ps_curr_layer->i4_inp_stride, |
7163 | 0 | ps_ctxt->au4_4x4_src_sigmaX, |
7164 | 0 | ps_ctxt->au4_4x4_src_sigmaXSquared, |
7165 | 0 | 4, |
7166 | 0 | 4, |
7167 | 0 | 64, |
7168 | 0 | 64, |
7169 | 0 | 1, |
7170 | 0 | 16); |
7171 | 0 | } |
7172 | 0 | else |
7173 | 0 | { |
7174 | 0 | for(i = 0; i < 4; i++) |
7175 | 0 | { |
7176 | 0 | if(au1_is_32x32Blk_noisy[i]) |
7177 | 0 | { |
7178 | 0 | U08 *pu1_inp = |
7179 | 0 | ps_curr_layer->pu1_inp + |
7180 | 0 | (s_common_frm_prms.i4_ctb_x_off + |
7181 | 0 | (s_common_frm_prms.i4_ctb_y_off * ps_curr_layer->i4_inp_stride)); |
7182 | |
|
7183 | 0 | U08 u1_cu_size = 32; |
7184 | 0 | WORD32 i4_inp_buf_offset = |
7185 | 0 | (((i / 2) * (u1_cu_size * ps_curr_layer->i4_inp_stride)) + |
7186 | 0 | ((i % 2) * u1_cu_size)); |
7187 | |
|
7188 | 0 | U16 u2_sigma_arr_start_index_of_3rd_32x32_blk_in_ctb = 128; |
7189 | 0 | U16 u2_sigma_arr_start_index_of_2nd_32x32_blk_in_ctb = 8; |
7190 | 0 | S32 i4_sigma_arr_offset = |
7191 | 0 | (((i / 2) * u2_sigma_arr_start_index_of_3rd_32x32_blk_in_ctb) + |
7192 | 0 | ((i % 2) * u2_sigma_arr_start_index_of_2nd_32x32_blk_in_ctb)); |
7193 | |
|
7194 | 0 | hme_compute_sigmaX_and_sigmaXSquared( |
7195 | 0 | pu1_inp + i4_inp_buf_offset, |
7196 | 0 | ps_curr_layer->i4_inp_stride, |
7197 | 0 | ps_ctxt->au4_4x4_src_sigmaX + i4_sigma_arr_offset, |
7198 | 0 | ps_ctxt->au4_4x4_src_sigmaXSquared + i4_sigma_arr_offset, |
7199 | 0 | 4, |
7200 | 0 | 4, |
7201 | 0 | 32, |
7202 | 0 | 32, |
7203 | 0 | 1, |
7204 | 0 | 16); |
7205 | 0 | } |
7206 | 0 | else |
7207 | 0 | { |
7208 | 0 | S32 j; |
7209 | |
|
7210 | 0 | U08 u1_16x16_blk_start_index_in_3rd_32x32_blk_of_ctb = 8; |
7211 | 0 | U08 u1_16x16_blk_start_index_in_2nd_32x32_blk_of_ctb = 2; |
7212 | 0 | S32 i4_16x16_blk_start_index_in_i_th_32x32_blk = |
7213 | 0 | (((i / 2) * u1_16x16_blk_start_index_in_3rd_32x32_blk_of_ctb) + |
7214 | 0 | ((i % 2) * u1_16x16_blk_start_index_in_2nd_32x32_blk_of_ctb)); |
7215 | |
|
7216 | 0 | for(j = 0; j < 4; j++) |
7217 | 0 | { |
7218 | 0 | U08 u1_3rd_16x16_blk_index_in_32x32_blk = 4; |
7219 | 0 | U08 u1_2nd_16x16_blk_index_in_32x32_blk = 1; |
7220 | 0 | S32 i4_16x16_blk_index_in_ctb = |
7221 | 0 | i4_16x16_blk_start_index_in_i_th_32x32_blk + |
7222 | 0 | ((j % 2) * u1_2nd_16x16_blk_index_in_32x32_blk) + |
7223 | 0 | ((j / 2) * u1_3rd_16x16_blk_index_in_32x32_blk); |
7224 | | |
7225 | | //S32 k = (((i / 2) * 8) + ((i % 2) * 2)) + ((j % 2) * 1) + ((j / 2) * 4); |
7226 | |
|
7227 | 0 | if(au1_is_16x16Blk_noisy[i4_16x16_blk_index_in_ctb]) |
7228 | 0 | { |
7229 | 0 | U08 *pu1_inp = |
7230 | 0 | ps_curr_layer->pu1_inp + (s_common_frm_prms.i4_ctb_x_off + |
7231 | 0 | (s_common_frm_prms.i4_ctb_y_off * |
7232 | 0 | ps_curr_layer->i4_inp_stride)); |
7233 | |
|
7234 | 0 | U08 u1_cu_size = 16; |
7235 | 0 | WORD32 i4_inp_buf_offset = |
7236 | 0 | (((i4_16x16_blk_index_in_ctb % 4) * u1_cu_size) + |
7237 | 0 | ((i4_16x16_blk_index_in_ctb / 4) * |
7238 | 0 | (u1_cu_size * ps_curr_layer->i4_inp_stride))); |
7239 | |
|
7240 | 0 | U16 u2_sigma_arr_start_index_of_3rd_16x16_blk_in_32x32_blk = 64; |
7241 | 0 | U16 u2_sigma_arr_start_index_of_2nd_16x16_blk_in_32x32_blk = 4; |
7242 | 0 | S32 i4_sigma_arr_offset = |
7243 | 0 | (((i4_16x16_blk_index_in_ctb % 4) * |
7244 | 0 | u2_sigma_arr_start_index_of_2nd_16x16_blk_in_32x32_blk) + |
7245 | 0 | ((i4_16x16_blk_index_in_ctb / 4) * |
7246 | 0 | u2_sigma_arr_start_index_of_3rd_16x16_blk_in_32x32_blk)); |
7247 | |
|
7248 | 0 | hme_compute_sigmaX_and_sigmaXSquared( |
7249 | 0 | pu1_inp + i4_inp_buf_offset, |
7250 | 0 | ps_curr_layer->i4_inp_stride, |
7251 | 0 | (ps_ctxt->au4_4x4_src_sigmaX + i4_sigma_arr_offset), |
7252 | 0 | (ps_ctxt->au4_4x4_src_sigmaXSquared + i4_sigma_arr_offset), |
7253 | 0 | 4, |
7254 | 0 | 4, |
7255 | 0 | 16, |
7256 | 0 | 16, |
7257 | 0 | 1, |
7258 | 0 | 16); |
7259 | 0 | } |
7260 | 0 | } |
7261 | 0 | } |
7262 | 0 | } |
7263 | 0 | } |
7264 | 0 | } |
7265 | 0 | else |
7266 | 0 | { |
7267 | 0 | memset(au1_is_16x16Blk_noisy, 0, sizeof(au1_is_16x16Blk_noisy)); |
7268 | |
|
7269 | 0 | memset(au1_is_32x32Blk_noisy, 0, sizeof(au1_is_32x32Blk_noisy)); |
7270 | |
|
7271 | 0 | memset(au1_is_64x64Blk_noisy, 0, sizeof(au1_is_64x64Blk_noisy)); |
7272 | 0 | } |
7273 | |
|
7274 | 0 | for(blk_id_in_ctb = 0; blk_id_in_ctb < num_blks_in_this_ctb; blk_id_in_ctb++) |
7275 | 0 | { |
7276 | 0 | S32 ref_ctr; |
7277 | 0 | U08 au1_pred_dir_searched[2]; |
7278 | 0 | U08 u1_is_cu_noisy; |
7279 | 0 | ULWORD64 au8_final_src_sigmaX[17], au8_final_src_sigmaXSquared[17]; |
7280 | |
|
7281 | 0 | { |
7282 | 0 | blk_x = (i4_ctb_x << 2) + |
7283 | 0 | (ps_ctb_bound_attrs->as_blk_attrs[blk_id_in_ctb].u1_blk_x); |
7284 | 0 | blk_y = (i4_ctb_y << 2) + |
7285 | 0 | (ps_ctb_bound_attrs->as_blk_attrs[blk_id_in_ctb].u1_blk_y); |
7286 | |
|
7287 | 0 | blk_id_in_full_ctb = |
7288 | 0 | ps_ctb_bound_attrs->as_blk_attrs[blk_id_in_ctb].u1_blk_id_in_full_ctb; |
7289 | 0 | blk_8x8_mask = ps_ctb_bound_attrs->as_blk_attrs[blk_id_in_ctb].u1_blk_8x8_mask; |
7290 | 0 | ai4_blk_8x8_mask[blk_id_in_full_ctb] = blk_8x8_mask; |
7291 | 0 | s_search_prms_blk.i4_cu_x_off = (blk_x << blk_size_shift) - (i4_ctb_x << 6); |
7292 | 0 | s_search_prms_blk.i4_cu_y_off = (blk_y << blk_size_shift) - (i4_ctb_y << 6); |
7293 | 0 | } |
7294 | | |
7295 | | /* get the current input blk point */ |
7296 | 0 | pos_x = blk_x << blk_size_shift; |
7297 | 0 | pos_y = blk_y << blk_size_shift; |
7298 | 0 | pu1_inp = ps_curr_layer->pu1_inp + pos_x + (pos_y * i4_inp_stride); |
7299 | | |
7300 | | /*********************************************************************/ |
7301 | | /* For every blk in the picture, the search range needs to be derived*/ |
7302 | | /* Any blk can have any mv, but practical search constraints are */ |
7303 | | /* imposed by the picture boundary and amt of padding. */ |
7304 | | /*********************************************************************/ |
7305 | | /* MV limit is different based on ref. PIC */ |
7306 | 0 | for(ref_ctr = 0; ref_ctr < num_act_ref_pics; ref_ctr++) |
7307 | 0 | { |
7308 | 0 | if(!s_search_prms_blk.i4_use_rec) |
7309 | 0 | { |
7310 | 0 | hme_derive_search_range( |
7311 | 0 | &as_range_prms_inp[ref_ctr], |
7312 | 0 | &s_pic_limit_inp, |
7313 | 0 | &as_mv_limit[ref_ctr], |
7314 | 0 | pos_x, |
7315 | 0 | pos_y, |
7316 | 0 | blk_wd, |
7317 | 0 | blk_ht); |
7318 | 0 | } |
7319 | 0 | else |
7320 | 0 | { |
7321 | 0 | hme_derive_search_range( |
7322 | 0 | &as_range_prms_rec[ref_ctr], |
7323 | 0 | &s_pic_limit_rec, |
7324 | 0 | &as_mv_limit[ref_ctr], |
7325 | 0 | pos_x, |
7326 | 0 | pos_y, |
7327 | 0 | blk_wd, |
7328 | 0 | blk_ht); |
7329 | 0 | } |
7330 | 0 | } |
7331 | 0 | s_search_prms_blk.i4_x_off = blk_x << blk_size_shift; |
7332 | 0 | s_search_prms_blk.i4_y_off = blk_y << blk_size_shift; |
7333 | | /* Select search results from a suitable search result in the context */ |
7334 | 0 | { |
7335 | 0 | ps_search_results = &ps_ctxt->as_search_results_16x16[blk_id_in_full_ctb]; |
7336 | |
|
7337 | 0 | if(ps_refine_prms->u1_use_lambda_derived_from_min_8x8_act_in_ctb) |
7338 | 0 | { |
7339 | 0 | S32 i; |
7340 | |
|
7341 | 0 | for(i = 0; i < 2; i++) |
7342 | 0 | { |
7343 | 0 | ps_search_results->as_pred_ctxt[i].lambda = lambda_recon; |
7344 | 0 | } |
7345 | 0 | } |
7346 | 0 | } |
7347 | |
|
7348 | 0 | u1_is_cu_noisy = au1_is_16x16Blk_noisy |
7349 | 0 | [(s_search_prms_blk.i4_cu_x_off >> 4) + (s_search_prms_blk.i4_cu_y_off >> 2)]; |
7350 | |
|
7351 | 0 | s_subpel_prms.u1_is_cu_noisy = u1_is_cu_noisy; |
7352 | |
|
7353 | 0 | #if ME_LAMBDA_DISCOUNT_WHEN_NOISY |
7354 | 0 | if(u1_is_cu_noisy) |
7355 | 0 | { |
7356 | 0 | S32 j; |
7357 | 0 | S32 lambda; |
7358 | |
|
7359 | 0 | lambda = lambda_recon; |
7360 | 0 | lambda = ((float)lambda * (100.0f - ME_LAMBDA_DISCOUNT_WHEN_NOISY) / 100.0f); |
7361 | |
|
7362 | 0 | for(j = 0; j < 2; j++) |
7363 | 0 | { |
7364 | 0 | ps_search_results->as_pred_ctxt[j].lambda = lambda; |
7365 | 0 | } |
7366 | 0 | } |
7367 | 0 | else |
7368 | 0 | { |
7369 | 0 | S32 j; |
7370 | 0 | S32 lambda; |
7371 | |
|
7372 | 0 | lambda = lambda_recon; |
7373 | |
|
7374 | 0 | for(j = 0; j < 2; j++) |
7375 | 0 | { |
7376 | 0 | ps_search_results->as_pred_ctxt[j].lambda = lambda; |
7377 | 0 | } |
7378 | 0 | } |
7379 | 0 | #endif |
7380 | |
|
7381 | 0 | s_search_prms_blk.ps_search_results = ps_search_results; |
7382 | |
|
7383 | 0 | s_search_prms_blk.i4_part_mask = hme_part_mask_populator( |
7384 | 0 | pu1_inp, |
7385 | 0 | i4_inp_stride, |
7386 | 0 | ps_refine_prms->limit_active_partitions, |
7387 | 0 | ps_ctxt->ps_hme_frm_prms->bidir_enabled, |
7388 | 0 | ps_ctxt->u1_is_curFrame_a_refFrame, |
7389 | 0 | blk_8x8_mask, |
7390 | 0 | e_me_quality_presets); |
7391 | |
|
7392 | 0 | if(ME_PRISTINE_QUALITY == e_me_quality_presets) |
7393 | 0 | { |
7394 | 0 | ps_ctb_cluster_info->ai4_part_mask[blk_id_in_full_ctb] = |
7395 | 0 | s_search_prms_blk.i4_part_mask; |
7396 | 0 | } |
7397 | | |
7398 | | /* RESET ALL SEARCH RESULTS FOR THE NEW BLK */ |
7399 | 0 | { |
7400 | | /* Setting u1_num_active_refs to 2 */ |
7401 | | /* for the sole purpose of the */ |
7402 | | /* function called below */ |
7403 | 0 | ps_search_results->u1_num_active_ref = (ps_refine_prms->bidir_enabled) ? 2 : 1; |
7404 | |
|
7405 | 0 | hme_reset_search_results( |
7406 | 0 | ps_search_results, s_search_prms_blk.i4_part_mask, MV_RES_FPEL); |
7407 | |
|
7408 | 0 | ps_search_results->u1_num_active_ref = i4_num_pred_dir; |
7409 | 0 | } |
7410 | |
|
7411 | 0 | if(0 == blk_id_in_ctb) |
7412 | 0 | { |
7413 | 0 | UWORD8 u1_ctr; |
7414 | 0 | for(u1_ctr = 0; u1_ctr < (ps_ctxt->s_frm_prms.u1_num_active_ref_l0 + |
7415 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l1); |
7416 | 0 | u1_ctr++) |
7417 | 0 | { |
7418 | 0 | WORD32 i4_max_dep_ctb_y; |
7419 | 0 | WORD32 i4_max_dep_ctb_x; |
7420 | | |
7421 | | /* Set max mv in ctb units */ |
7422 | 0 | i4_max_mv_x_in_ctb = |
7423 | 0 | (ps_curr_layer->i2_max_mv_x + ((1 << ps_ctxt->log_ctb_size) - 1)) >> |
7424 | 0 | ps_ctxt->log_ctb_size; |
7425 | |
|
7426 | 0 | i4_max_mv_y_in_ctb = |
7427 | 0 | (as_mv_limit[u1_ctr].i2_max_y + ((1 << ps_ctxt->log_ctb_size) - 1)) >> |
7428 | 0 | ps_ctxt->log_ctb_size; |
7429 | | /********************************************************************/ |
7430 | | /* Set max ctb_x and ctb_y dependency on reference picture */ |
7431 | | /* Note +1 is due to delayed deblock, SAO, subpel plan dependency */ |
7432 | | /********************************************************************/ |
7433 | 0 | i4_max_dep_ctb_x = CLIP3( |
7434 | 0 | (i4_ctb_x + i4_max_mv_x_in_ctb + 1), |
7435 | 0 | 0, |
7436 | 0 | ps_frm_ctb_prms->i4_num_ctbs_horz - 1); |
7437 | 0 | i4_max_dep_ctb_y = CLIP3( |
7438 | 0 | (i4_ctb_y + i4_max_mv_y_in_ctb + 1), |
7439 | 0 | 0, |
7440 | 0 | ps_frm_ctb_prms->i4_num_ctbs_vert - 1); |
7441 | |
|
7442 | 0 | ihevce_dmgr_map_chk_sync( |
7443 | 0 | ps_curr_layer->ppv_dep_mngr_recon[u1_ctr], |
7444 | 0 | ps_ctxt->thrd_id, |
7445 | 0 | i4_ctb_x, |
7446 | 0 | i4_ctb_y, |
7447 | 0 | i4_max_mv_x_in_ctb, |
7448 | 0 | i4_max_mv_y_in_ctb); |
7449 | 0 | } |
7450 | 0 | } |
7451 | | |
7452 | | /* Loop across different Ref IDx */ |
7453 | 0 | for(u1_pred_dir_ctr = 0; u1_pred_dir_ctr < i4_num_pred_dir; u1_pred_dir_ctr++) |
7454 | 0 | { |
7455 | 0 | S32 resultid; |
7456 | 0 | S08 u1_default_ref_id; |
7457 | 0 | S32 i4_num_srch_cands = 0; |
7458 | 0 | S32 i4_num_refinement_iterations; |
7459 | 0 | S32 i4_refine_iter_ctr; |
7460 | |
|
7461 | 0 | if((i4_num_pred_dir == 2) || (!ps_ctxt->s_frm_prms.bidir_enabled) || |
7462 | 0 | (ps_ctxt->s_frm_prms.u1_num_active_ref_l1 == 0)) |
7463 | 0 | { |
7464 | 0 | u1_pred_dir = u1_pred_dir_ctr; |
7465 | 0 | } |
7466 | 0 | else if(ps_ctxt->s_frm_prms.u1_num_active_ref_l0 == 0) |
7467 | 0 | { |
7468 | 0 | u1_pred_dir = 1; |
7469 | 0 | } |
7470 | |
|
7471 | 0 | u1_default_ref_id = (u1_pred_dir == 0) ? ps_ctxt->ai1_past_list[0] |
7472 | 0 | : ps_ctxt->ai1_future_list[0]; |
7473 | 0 | au1_pred_dir_searched[u1_pred_dir_ctr] = u1_pred_dir; |
7474 | |
|
7475 | 0 | i4_num_srch_cands = 0; |
7476 | 0 | resultid = 0; |
7477 | | |
7478 | | /* START OF NEW CTB MEANS FILL UP NEOGHBOURS IN 18x18 GRID */ |
7479 | 0 | if(0 == blk_id_in_ctb) |
7480 | 0 | { |
7481 | | /*****************************************************************/ |
7482 | | /* Initialize the mv grid with results of neighbours for the next*/ |
7483 | | /* ctb. */ |
7484 | | /*****************************************************************/ |
7485 | 0 | hme_fill_ctb_neighbour_mvs( |
7486 | 0 | ps_curr_layer, |
7487 | 0 | blk_x, |
7488 | 0 | blk_y, |
7489 | 0 | aps_mv_grid[u1_pred_dir], |
7490 | 0 | u1_pred_dir_ctr, |
7491 | 0 | u1_default_ref_id, |
7492 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l0); |
7493 | 0 | } |
7494 | |
|
7495 | 0 | s_search_prms_blk.i1_ref_idx = u1_pred_dir; |
7496 | |
|
7497 | 0 | { |
7498 | 0 | if((blk_id_in_full_ctb % 4) == 0) |
7499 | 0 | { |
7500 | 0 | ps_ctxt->as_search_results_32x32[blk_id_in_full_ctb >> 2] |
7501 | 0 | .as_pred_ctxt[u1_pred_dir] |
7502 | 0 | .proj_used = (blk_id_in_full_ctb == 8) ? 0 : 1; |
7503 | 0 | } |
7504 | |
|
7505 | 0 | if(blk_id_in_full_ctb == 0) |
7506 | 0 | { |
7507 | 0 | ps_ctxt->s_search_results_64x64.as_pred_ctxt[u1_pred_dir].proj_used = 1; |
7508 | 0 | } |
7509 | |
|
7510 | 0 | ps_search_results->as_pred_ctxt[u1_pred_dir].proj_used = |
7511 | 0 | !gau1_encode_to_raster_y[blk_id_in_full_ctb]; |
7512 | 0 | } |
7513 | |
|
7514 | 0 | { |
7515 | 0 | S32 x = gau1_encode_to_raster_x[blk_id_in_full_ctb]; |
7516 | 0 | S32 y = gau1_encode_to_raster_y[blk_id_in_full_ctb]; |
7517 | 0 | U08 u1_is_blk_at_ctb_boundary = !y; |
7518 | |
|
7519 | 0 | s_srch_cand_init_data.u1_is_left_available = |
7520 | 0 | !(left_ctb_in_diff_tile && !s_search_prms_blk.i4_cu_x_off); |
7521 | |
|
7522 | 0 | if(u1_is_blk_at_ctb_boundary) |
7523 | 0 | { |
7524 | 0 | s_srch_cand_init_data.u1_is_topRight_available = 0; |
7525 | 0 | s_srch_cand_init_data.u1_is_topLeft_available = 0; |
7526 | 0 | s_srch_cand_init_data.u1_is_top_available = 0; |
7527 | 0 | } |
7528 | 0 | else |
7529 | 0 | { |
7530 | 0 | s_srch_cand_init_data.u1_is_topRight_available = |
7531 | 0 | gau1_cu_tr_valid[y][x] && ((pos_x + blk_wd) < i4_pic_wd); |
7532 | 0 | s_srch_cand_init_data.u1_is_top_available = 1; |
7533 | 0 | s_srch_cand_init_data.u1_is_topLeft_available = |
7534 | 0 | s_srch_cand_init_data.u1_is_left_available; |
7535 | 0 | } |
7536 | 0 | } |
7537 | |
|
7538 | 0 | s_srch_cand_init_data.i1_default_ref_id = u1_default_ref_id; |
7539 | 0 | s_srch_cand_init_data.i1_alt_default_ref_id = ps_ctxt->ai1_past_list[1]; |
7540 | 0 | s_srch_cand_init_data.i4_pos_x = pos_x; |
7541 | 0 | s_srch_cand_init_data.i4_pos_y = pos_y; |
7542 | 0 | s_srch_cand_init_data.u1_pred_dir = u1_pred_dir; |
7543 | 0 | s_srch_cand_init_data.u1_pred_dir_ctr = u1_pred_dir_ctr; |
7544 | 0 | s_srch_cand_init_data.u1_search_candidate_list_index = |
7545 | 0 | au1_search_candidate_list_index[u1_pred_dir]; |
7546 | |
|
7547 | 0 | i4_num_srch_cands = hme_populate_search_candidates(&s_srch_cand_init_data); |
7548 | | |
7549 | | /* Note this block also clips the MV range for all candidates */ |
7550 | 0 | { |
7551 | 0 | S08 i1_check_for_mult_refs; |
7552 | |
|
7553 | 0 | i1_check_for_mult_refs = u1_pred_dir ? (ps_ctxt->num_ref_future > 1) |
7554 | 0 | : (ps_ctxt->num_ref_past > 1); |
7555 | |
|
7556 | 0 | ps_me_optimised_function_list->pf_mv_clipper( |
7557 | 0 | &s_search_prms_blk, |
7558 | 0 | i4_num_srch_cands, |
7559 | 0 | i1_check_for_mult_refs, |
7560 | 0 | ps_refine_prms->i4_num_steps_fpel_refine, |
7561 | 0 | ps_refine_prms->i4_num_steps_hpel_refine, |
7562 | 0 | ps_refine_prms->i4_num_steps_qpel_refine); |
7563 | 0 | } |
7564 | |
|
7565 | 0 | #if ENABLE_EXPLICIT_SEARCH_IN_P_IN_L0 |
7566 | 0 | i4_num_refinement_iterations = |
7567 | 0 | ((!ps_ctxt->s_frm_prms.bidir_enabled) && (i4_num_act_ref_l0 > 1)) |
7568 | 0 | ? ((e_me_quality_presets == ME_HIGH_QUALITY) ? 2 : i4_num_act_ref_l0) |
7569 | 0 | : 1; |
7570 | | #else |
7571 | | i4_num_refinement_iterations = |
7572 | | ((!ps_ctxt->s_frm_prms.bidir_enabled) && (i4_num_act_ref_l0 > 1)) ? 2 : 1; |
7573 | | #endif |
7574 | |
|
7575 | | #if ENABLE_EXPLICIT_SEARCH_IN_PQ |
7576 | | if(e_me_quality_presets == ME_PRISTINE_QUALITY) |
7577 | | { |
7578 | | i4_num_refinement_iterations = (u1_pred_dir == 0) ? i4_num_act_ref_l0 |
7579 | | : i4_num_act_ref_l1; |
7580 | | } |
7581 | | #endif |
7582 | |
|
7583 | 0 | for(i4_refine_iter_ctr = 0; i4_refine_iter_ctr < i4_num_refinement_iterations; |
7584 | 0 | i4_refine_iter_ctr++) |
7585 | 0 | { |
7586 | 0 | S32 center_x; |
7587 | 0 | S32 center_y; |
7588 | 0 | S32 center_ref_idx; |
7589 | |
|
7590 | 0 | S08 *pi1_pred_dir_to_ref_idx = |
7591 | 0 | (u1_pred_dir == 0) ? ps_ctxt->ai1_past_list : ps_ctxt->ai1_future_list; |
7592 | |
|
7593 | 0 | { |
7594 | 0 | WORD32 i4_i; |
7595 | |
|
7596 | 0 | for(i4_i = 0; i4_i < TOT_NUM_PARTS; i4_i++) |
7597 | 0 | { |
7598 | 0 | ps_fullpel_refine_ctxt->i2_tot_cost[0][i4_i] = MAX_SIGNED_16BIT_VAL; |
7599 | 0 | ps_fullpel_refine_ctxt->i2_mv_cost[0][i4_i] = MAX_SIGNED_16BIT_VAL; |
7600 | 0 | ps_fullpel_refine_ctxt->i2_stim_injected_cost[0][i4_i] = |
7601 | 0 | MAX_SIGNED_16BIT_VAL; |
7602 | 0 | ps_fullpel_refine_ctxt->i2_mv_x[0][i4_i] = 0; |
7603 | 0 | ps_fullpel_refine_ctxt->i2_mv_y[0][i4_i] = 0; |
7604 | 0 | ps_fullpel_refine_ctxt->i2_ref_idx[0][i4_i] = u1_default_ref_id; |
7605 | |
|
7606 | 0 | if(ps_refine_prms->i4_num_results_per_part == 2) |
7607 | 0 | { |
7608 | 0 | ps_fullpel_refine_ctxt->i2_tot_cost[1][i4_i] = |
7609 | 0 | MAX_SIGNED_16BIT_VAL; |
7610 | 0 | ps_fullpel_refine_ctxt->i2_mv_cost[1][i4_i] = |
7611 | 0 | MAX_SIGNED_16BIT_VAL; |
7612 | 0 | ps_fullpel_refine_ctxt->i2_stim_injected_cost[1][i4_i] = |
7613 | 0 | MAX_SIGNED_16BIT_VAL; |
7614 | 0 | ps_fullpel_refine_ctxt->i2_mv_x[1][i4_i] = 0; |
7615 | 0 | ps_fullpel_refine_ctxt->i2_mv_y[1][i4_i] = 0; |
7616 | 0 | ps_fullpel_refine_ctxt->i2_ref_idx[1][i4_i] = u1_default_ref_id; |
7617 | 0 | } |
7618 | 0 | } |
7619 | |
|
7620 | 0 | s_search_prms_blk.ps_fullpel_refine_ctxt = ps_fullpel_refine_ctxt; |
7621 | 0 | s_subpel_prms.ps_subpel_refine_ctxt = ps_fullpel_refine_ctxt; |
7622 | 0 | } |
7623 | |
|
7624 | 0 | { |
7625 | 0 | search_node_t *ps_coloc_node; |
7626 | |
|
7627 | 0 | S32 i = 0; |
7628 | |
|
7629 | 0 | if(i4_num_refinement_iterations > 1) |
7630 | 0 | { |
7631 | 0 | for(i = 0; i < ai4_num_coloc_cands[u1_pred_dir]; i++) |
7632 | 0 | { |
7633 | 0 | ps_coloc_node = |
7634 | 0 | s_search_prms_blk.ps_search_candts[ai4_id_coloc[i]] |
7635 | 0 | .ps_search_node; |
7636 | |
|
7637 | 0 | if(pi1_pred_dir_to_ref_idx[i4_refine_iter_ctr] == |
7638 | 0 | ps_coloc_node->i1_ref_idx) |
7639 | 0 | { |
7640 | 0 | break; |
7641 | 0 | } |
7642 | 0 | } |
7643 | |
|
7644 | 0 | if(i == ai4_num_coloc_cands[u1_pred_dir]) |
7645 | 0 | { |
7646 | 0 | i = 0; |
7647 | 0 | } |
7648 | 0 | } |
7649 | 0 | else |
7650 | 0 | { |
7651 | 0 | ps_coloc_node = s_search_prms_blk.ps_search_candts[ai4_id_coloc[0]] |
7652 | 0 | .ps_search_node; |
7653 | 0 | } |
7654 | |
|
7655 | 0 | hme_set_mvp_node( |
7656 | 0 | ps_search_results, |
7657 | 0 | ps_coloc_node, |
7658 | 0 | u1_pred_dir, |
7659 | 0 | (i4_num_refinement_iterations > 1) |
7660 | 0 | ? pi1_pred_dir_to_ref_idx[i4_refine_iter_ctr] |
7661 | 0 | : u1_default_ref_id); |
7662 | |
|
7663 | 0 | center_x = ps_coloc_node->ps_mv->i2_mvx; |
7664 | 0 | center_y = ps_coloc_node->ps_mv->i2_mvy; |
7665 | 0 | center_ref_idx = ps_coloc_node->i1_ref_idx; |
7666 | 0 | } |
7667 | | |
7668 | | /* Full-Pel search */ |
7669 | 0 | { |
7670 | 0 | S32 num_unique_nodes; |
7671 | |
|
7672 | 0 | memset(au4_unique_node_map, 0, sizeof(au4_unique_node_map)); |
7673 | |
|
7674 | 0 | num_unique_nodes = hme_remove_duplicate_fpel_search_candidates( |
7675 | 0 | as_unique_search_nodes, |
7676 | 0 | s_search_prms_blk.ps_search_candts, |
7677 | 0 | au4_unique_node_map, |
7678 | 0 | pi1_pred_dir_to_ref_idx, |
7679 | 0 | i4_num_srch_cands, |
7680 | 0 | s_search_prms_blk.i4_num_init_candts, |
7681 | 0 | i4_refine_iter_ctr, |
7682 | 0 | i4_num_refinement_iterations, |
7683 | 0 | i4_num_act_ref_l0, |
7684 | 0 | center_ref_idx, |
7685 | 0 | center_x, |
7686 | 0 | center_y, |
7687 | 0 | ps_ctxt->s_frm_prms.bidir_enabled, |
7688 | 0 | e_me_quality_presets); |
7689 | | |
7690 | | /*************************************************************************/ |
7691 | | /* This array stores the ids of the partitions whose */ |
7692 | | /* SADs are updated. Since the partitions whose SADs are updated may not */ |
7693 | | /* be in contiguous order, we supply another level of indirection. */ |
7694 | | /*************************************************************************/ |
7695 | 0 | ps_fullpel_refine_ctxt->i4_num_valid_parts = hme_create_valid_part_ids( |
7696 | 0 | s_search_prms_blk.i4_part_mask, |
7697 | 0 | &ps_fullpel_refine_ctxt->ai4_part_id[0]); |
7698 | |
|
7699 | 0 | if(!i4_refine_iter_ctr && !u1_pred_dir_ctr && u1_is_cu_noisy) |
7700 | 0 | { |
7701 | 0 | S32 i; |
7702 | | /*i4_sigma_array_offset : takes care of pointing to the appropriate 4x4 block's sigmaX and sigmaX-squared value in a CTB out of 256 values*/ |
7703 | 0 | S32 i4_sigma_array_offset = (s_search_prms_blk.i4_cu_x_off / 4) + |
7704 | 0 | (s_search_prms_blk.i4_cu_y_off * 4); |
7705 | |
|
7706 | 0 | for(i = 0; i < ps_fullpel_refine_ctxt->i4_num_valid_parts; i++) |
7707 | 0 | { |
7708 | 0 | S32 i4_part_id = ps_fullpel_refine_ctxt->ai4_part_id[i]; |
7709 | |
|
7710 | 0 | hme_compute_final_sigma_of_pu_from_base_blocks( |
7711 | 0 | ps_ctxt->au4_4x4_src_sigmaX + i4_sigma_array_offset, |
7712 | 0 | ps_ctxt->au4_4x4_src_sigmaXSquared + i4_sigma_array_offset, |
7713 | 0 | au8_final_src_sigmaX, |
7714 | 0 | au8_final_src_sigmaXSquared, |
7715 | 0 | 16, |
7716 | 0 | 4, |
7717 | 0 | i4_part_id, |
7718 | 0 | 16); |
7719 | 0 | } |
7720 | |
|
7721 | 0 | s_common_frm_prms.pu8_part_src_sigmaX = au8_final_src_sigmaX; |
7722 | 0 | s_common_frm_prms.pu8_part_src_sigmaXSquared = |
7723 | 0 | au8_final_src_sigmaXSquared; |
7724 | |
|
7725 | 0 | s_search_prms_blk.pu8_part_src_sigmaX = au8_final_src_sigmaX; |
7726 | 0 | s_search_prms_blk.pu8_part_src_sigmaXSquared = |
7727 | 0 | au8_final_src_sigmaXSquared; |
7728 | 0 | } |
7729 | |
|
7730 | 0 | if(0 == num_unique_nodes) |
7731 | 0 | { |
7732 | 0 | continue; |
7733 | 0 | } |
7734 | | |
7735 | 0 | if(num_unique_nodes >= 2) |
7736 | 0 | { |
7737 | 0 | s_search_prms_blk.ps_search_nodes = &as_unique_search_nodes[0]; |
7738 | 0 | s_search_prms_blk.i4_num_search_nodes = num_unique_nodes; |
7739 | 0 | if(ps_ctxt->i4_pic_type != IV_P_FRAME) |
7740 | 0 | { |
7741 | 0 | if(ps_ctxt->i4_temporal_layer == 1) |
7742 | 0 | { |
7743 | 0 | hme_fullpel_cand_sifter( |
7744 | 0 | &s_search_prms_blk, |
7745 | 0 | ps_curr_layer, |
7746 | 0 | &ps_ctxt->s_wt_pred, |
7747 | 0 | ALPHA_FOR_NOISE_TERM_IN_ME, |
7748 | 0 | u1_is_cu_noisy, |
7749 | 0 | ps_me_optimised_function_list); |
7750 | 0 | } |
7751 | 0 | else |
7752 | 0 | { |
7753 | 0 | hme_fullpel_cand_sifter( |
7754 | 0 | &s_search_prms_blk, |
7755 | 0 | ps_curr_layer, |
7756 | 0 | &ps_ctxt->s_wt_pred, |
7757 | 0 | ALPHA_FOR_NOISE_TERM_IN_ME, |
7758 | 0 | u1_is_cu_noisy, |
7759 | 0 | ps_me_optimised_function_list); |
7760 | 0 | } |
7761 | 0 | } |
7762 | 0 | else |
7763 | 0 | { |
7764 | 0 | hme_fullpel_cand_sifter( |
7765 | 0 | &s_search_prms_blk, |
7766 | 0 | ps_curr_layer, |
7767 | 0 | &ps_ctxt->s_wt_pred, |
7768 | 0 | ALPHA_FOR_NOISE_TERM_IN_ME_P, |
7769 | 0 | u1_is_cu_noisy, |
7770 | 0 | ps_me_optimised_function_list); |
7771 | 0 | } |
7772 | 0 | } |
7773 | |
|
7774 | 0 | s_search_prms_blk.ps_search_nodes = &as_unique_search_nodes[0]; |
7775 | |
|
7776 | 0 | hme_fullpel_refine( |
7777 | 0 | ps_refine_prms, |
7778 | 0 | &s_search_prms_blk, |
7779 | 0 | ps_curr_layer, |
7780 | 0 | &ps_ctxt->s_wt_pred, |
7781 | 0 | au4_unique_node_map, |
7782 | 0 | num_unique_nodes, |
7783 | 0 | blk_8x8_mask, |
7784 | 0 | center_x, |
7785 | 0 | center_y, |
7786 | 0 | center_ref_idx, |
7787 | 0 | e_me_quality_presets, |
7788 | 0 | ps_me_optimised_function_list); |
7789 | 0 | } |
7790 | | |
7791 | | /* Sub-Pel search */ |
7792 | 0 | { |
7793 | 0 | hme_reset_wkg_mem(&ps_ctxt->s_buf_mgr); |
7794 | |
|
7795 | 0 | s_subpel_prms.pu1_wkg_mem = (U08 *)hme_get_wkg_mem( |
7796 | 0 | &ps_ctxt->s_buf_mgr, |
7797 | 0 | INTERP_INTERMED_BUF_SIZE + INTERP_OUT_BUF_SIZE); |
7798 | | /* MV limit is different based on ref. PIC */ |
7799 | 0 | for(ref_ctr = 0; ref_ctr < num_act_ref_pics; ref_ctr++) |
7800 | 0 | { |
7801 | 0 | SCALE_RANGE_PRMS( |
7802 | 0 | as_range_prms_hpel[ref_ctr], as_range_prms_rec[ref_ctr], 1); |
7803 | 0 | SCALE_RANGE_PRMS( |
7804 | 0 | as_range_prms_qpel[ref_ctr], as_range_prms_rec[ref_ctr], 2); |
7805 | 0 | } |
7806 | 0 | s_subpel_prms.i4_ctb_x_off = i4_ctb_x << 6; |
7807 | 0 | s_subpel_prms.i4_ctb_y_off = i4_ctb_y << 6; |
7808 | |
|
7809 | 0 | hme_subpel_refine_cu_hs( |
7810 | 0 | &s_subpel_prms, |
7811 | 0 | ps_curr_layer, |
7812 | 0 | ps_search_results, |
7813 | 0 | u1_pred_dir, |
7814 | 0 | &ps_ctxt->s_wt_pred, |
7815 | 0 | blk_8x8_mask, |
7816 | 0 | ps_ctxt->ps_func_selector, |
7817 | 0 | ps_cmn_utils_optimised_function_list, |
7818 | 0 | ps_me_optimised_function_list); |
7819 | 0 | } |
7820 | 0 | } |
7821 | 0 | } |
7822 | | /* Populate the new PU struct with the results post subpel refinement*/ |
7823 | 0 | { |
7824 | 0 | inter_cu_results_t *ps_cu_results; |
7825 | 0 | WORD32 best_inter_cost, intra_cost, posx, posy; |
7826 | |
|
7827 | 0 | UWORD8 intra_8x8_enabled = 0; |
7828 | | |
7829 | | /* cost of 16x16 cu parent */ |
7830 | 0 | WORD32 parent_cost = MAX_32BIT_VAL; |
7831 | | |
7832 | | /* cost of 8x8 cu children */ |
7833 | | /*********************************************************************/ |
7834 | | /* Assuming parent is not split, then we signal 1 bit for this parent*/ |
7835 | | /* CU. If split, then 1 bit for parent CU + 4 bits for each child CU */ |
7836 | | /* So, 4*lambda is extra for children cost. */ |
7837 | | /*********************************************************************/ |
7838 | 0 | WORD32 child_cost = 0; |
7839 | |
|
7840 | 0 | ps_cu_results = ps_search_results->ps_cu_results; |
7841 | | |
7842 | | /* Initialize the pu_results pointers to the first struct in the stack array */ |
7843 | 0 | ps_pu_results = as_inter_pu_results; |
7844 | |
|
7845 | 0 | hme_reset_wkg_mem(&ps_ctxt->s_buf_mgr); |
7846 | |
|
7847 | 0 | hme_populate_pus( |
7848 | 0 | ps_thrd_ctxt, |
7849 | 0 | ps_ctxt, |
7850 | 0 | &s_subpel_prms, |
7851 | 0 | ps_search_results, |
7852 | 0 | ps_cu_results, |
7853 | 0 | ps_pu_results, |
7854 | 0 | &(as_pu_results[0][0][0]), |
7855 | 0 | &s_common_frm_prms, |
7856 | 0 | &ps_ctxt->s_wt_pred, |
7857 | 0 | ps_curr_layer, |
7858 | 0 | au1_pred_dir_searched, |
7859 | 0 | i4_num_pred_dir); |
7860 | |
|
7861 | 0 | ps_cu_results->i4_inp_offset = |
7862 | 0 | (ps_cu_results->u1_x_off) + (ps_cu_results->u1_y_off * 64); |
7863 | |
|
7864 | 0 | hme_decide_part_types( |
7865 | 0 | ps_cu_results, |
7866 | 0 | ps_pu_results, |
7867 | 0 | &s_common_frm_prms, |
7868 | 0 | ps_ctxt, |
7869 | 0 | ps_cmn_utils_optimised_function_list, |
7870 | 0 | ps_me_optimised_function_list |
7871 | |
|
7872 | 0 | ); |
7873 | | |
7874 | | /* UPDATE the MIN and MAX MVs for Dynamical Search Range for each ref. pic. */ |
7875 | | /* Only for P pic. For P, both are 0, I&B has them mut. exclusive */ |
7876 | 0 | if(ps_ctxt->s_frm_prms.is_i_pic == ps_ctxt->s_frm_prms.bidir_enabled) |
7877 | 0 | { |
7878 | 0 | WORD32 res_ctr; |
7879 | |
|
7880 | 0 | for(res_ctr = 0; res_ctr < ps_cu_results->u1_num_best_results; res_ctr++) |
7881 | 0 | { |
7882 | 0 | WORD32 num_part = 2, part_ctr; |
7883 | 0 | part_type_results_t *ps_best_results = |
7884 | 0 | &ps_cu_results->ps_best_results[res_ctr]; |
7885 | |
|
7886 | 0 | if(PRT_2Nx2N == ps_best_results->u1_part_type) |
7887 | 0 | num_part = 1; |
7888 | |
|
7889 | 0 | for(part_ctr = 0; part_ctr < num_part; part_ctr++) |
7890 | 0 | { |
7891 | 0 | pu_result_t *ps_pu_results = |
7892 | 0 | &ps_best_results->as_pu_results[part_ctr]; |
7893 | |
|
7894 | 0 | ASSERT(PRED_L0 == ps_pu_results->pu.b2_pred_mode); |
7895 | | |
7896 | 0 | hme_update_dynamic_search_params( |
7897 | 0 | &ps_ctxt->as_l0_dyn_range_prms[i4_idx_dvsr_p] |
7898 | 0 | .as_dyn_range_prms[ps_pu_results->pu.mv.i1_l0_ref_idx], |
7899 | 0 | ps_pu_results->pu.mv.s_l0_mv.i2_mvy); |
7900 | | |
7901 | | /* Sanity Check */ |
7902 | 0 | ASSERT( |
7903 | 0 | ps_pu_results->pu.mv.i1_l0_ref_idx < |
7904 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l0); |
7905 | | |
7906 | | /* No L1 for P Pic. */ |
7907 | 0 | ASSERT(PRED_L1 != ps_pu_results->pu.b2_pred_mode); |
7908 | | /* No BI for P Pic. */ |
7909 | 0 | ASSERT(PRED_BI != ps_pu_results->pu.b2_pred_mode); |
7910 | 0 | } |
7911 | 0 | } |
7912 | 0 | } |
7913 | | |
7914 | | /*****************************************************************/ |
7915 | | /* INSERT INTRA RESULTS AT 16x16 LEVEL. */ |
7916 | | /*****************************************************************/ |
7917 | | |
7918 | 0 | #if DISABLE_INTRA_IN_BPICS |
7919 | 0 | if(1 != ((ME_XTREME_SPEED_25 == e_me_quality_presets) && |
7920 | 0 | (ps_ctxt->s_frm_prms.i4_temporal_layer_id > TEMPORAL_LAYER_DISABLE))) |
7921 | 0 | #endif |
7922 | 0 | { |
7923 | 0 | if(!(DISABLE_INTRA_WHEN_NOISY && s_common_frm_prms.u1_is_cu_noisy)) |
7924 | 0 | { |
7925 | 0 | hme_insert_intra_nodes_post_bipred( |
7926 | 0 | ps_cu_results, ps_cur_ipe_ctb, ps_ctxt->frm_qstep); |
7927 | 0 | } |
7928 | 0 | } |
7929 | |
|
7930 | 0 | #if DISABLE_INTRA_IN_BPICS |
7931 | 0 | if((ME_XTREME_SPEED_25 == e_me_quality_presets) && |
7932 | 0 | (ps_ctxt->s_frm_prms.i4_temporal_layer_id > TEMPORAL_LAYER_DISABLE)) |
7933 | 0 | { |
7934 | 0 | intra_8x8_enabled = 0; |
7935 | 0 | } |
7936 | 0 | else |
7937 | 0 | #endif |
7938 | 0 | { |
7939 | | /*TRAQO intra flag updation*/ |
7940 | 0 | if(1 == ps_cu_results->ps_best_results->as_pu_results[0].pu.b1_intra_flag) |
7941 | 0 | { |
7942 | 0 | best_inter_cost = |
7943 | 0 | ps_cu_results->ps_best_results->as_pu_results[1].i4_tot_cost; |
7944 | 0 | intra_cost = |
7945 | 0 | ps_cu_results->ps_best_results->as_pu_results[0].i4_tot_cost; |
7946 | | /*@16x16 level*/ |
7947 | 0 | posx = (ps_cu_results->ps_best_results->as_pu_results[1].pu.b4_pos_x |
7948 | 0 | << 2) >> |
7949 | 0 | 4; |
7950 | 0 | posy = (ps_cu_results->ps_best_results->as_pu_results[1].pu.b4_pos_y |
7951 | 0 | << 2) >> |
7952 | 0 | 4; |
7953 | 0 | } |
7954 | 0 | else |
7955 | 0 | { |
7956 | 0 | best_inter_cost = |
7957 | 0 | ps_cu_results->ps_best_results->as_pu_results[0].i4_tot_cost; |
7958 | 0 | posx = (ps_cu_results->ps_best_results->as_pu_results[0].pu.b4_pos_x |
7959 | 0 | << 2) >> |
7960 | 0 | 3; |
7961 | 0 | posy = (ps_cu_results->ps_best_results->as_pu_results[0].pu.b4_pos_y |
7962 | 0 | << 2) >> |
7963 | 0 | 3; |
7964 | 0 | } |
7965 | | |
7966 | | /* Disable intra16/32/64 flags based on split flags recommended by IPE */ |
7967 | 0 | if(ps_cur_ipe_ctb->u1_split_flag) |
7968 | 0 | { |
7969 | | /* Id of the 32x32 block, 16x16 block in a CTB */ |
7970 | 0 | WORD32 i4_32x32_id = |
7971 | 0 | (ps_cu_results->u1_y_off >> 5) * 2 + (ps_cu_results->u1_x_off >> 5); |
7972 | 0 | WORD32 i4_16x16_id = ((ps_cu_results->u1_y_off >> 4) & 0x1) * 2 + |
7973 | 0 | ((ps_cu_results->u1_x_off >> 4) & 0x1); |
7974 | |
|
7975 | 0 | if(ps_cur_ipe_ctb->as_intra32_analyse[i4_32x32_id].b1_split_flag) |
7976 | 0 | { |
7977 | 0 | if(ps_cur_ipe_ctb->as_intra32_analyse[i4_32x32_id] |
7978 | 0 | .as_intra16_analyse[i4_16x16_id] |
7979 | 0 | .b1_split_flag) |
7980 | 0 | { |
7981 | 0 | intra_8x8_enabled = |
7982 | 0 | ps_cur_ipe_ctb->as_intra32_analyse[i4_32x32_id] |
7983 | 0 | .as_intra16_analyse[i4_16x16_id] |
7984 | 0 | .as_intra8_analyse[0] |
7985 | 0 | .b1_valid_cu; |
7986 | 0 | intra_8x8_enabled &= |
7987 | 0 | ps_cur_ipe_ctb->as_intra32_analyse[i4_32x32_id] |
7988 | 0 | .as_intra16_analyse[i4_16x16_id] |
7989 | 0 | .as_intra8_analyse[1] |
7990 | 0 | .b1_valid_cu; |
7991 | 0 | intra_8x8_enabled &= |
7992 | 0 | ps_cur_ipe_ctb->as_intra32_analyse[i4_32x32_id] |
7993 | 0 | .as_intra16_analyse[i4_16x16_id] |
7994 | 0 | .as_intra8_analyse[2] |
7995 | 0 | .b1_valid_cu; |
7996 | 0 | intra_8x8_enabled &= |
7997 | 0 | ps_cur_ipe_ctb->as_intra32_analyse[i4_32x32_id] |
7998 | 0 | .as_intra16_analyse[i4_16x16_id] |
7999 | 0 | .as_intra8_analyse[3] |
8000 | 0 | .b1_valid_cu; |
8001 | 0 | } |
8002 | 0 | } |
8003 | 0 | } |
8004 | 0 | } |
8005 | |
|
8006 | 0 | if(blk_8x8_mask == 0xf) |
8007 | 0 | { |
8008 | 0 | parent_cost = |
8009 | 0 | ps_search_results->ps_cu_results->ps_best_results[0].i4_tot_cost; |
8010 | 0 | ps_search_results->u1_split_flag = 0; |
8011 | 0 | } |
8012 | 0 | else |
8013 | 0 | { |
8014 | 0 | ps_search_results->u1_split_flag = 1; |
8015 | 0 | } |
8016 | |
|
8017 | 0 | ps_cu_results = &ps_ctxt->as_cu8x8_results[blk_id_in_full_ctb << 2]; |
8018 | |
|
8019 | 0 | if(s_common_frm_prms.u1_is_cu_noisy) |
8020 | 0 | { |
8021 | 0 | intra_8x8_enabled = 0; |
8022 | 0 | } |
8023 | | |
8024 | | /* Evalaute 8x8 if NxN part id is enabled */ |
8025 | 0 | if((ps_search_results->i4_part_mask & ENABLE_NxN) || intra_8x8_enabled) |
8026 | 0 | { |
8027 | | /* Populates the PU's for the 4 8x8's in one call */ |
8028 | 0 | hme_populate_pus_8x8_cu( |
8029 | 0 | ps_thrd_ctxt, |
8030 | 0 | ps_ctxt, |
8031 | 0 | &s_subpel_prms, |
8032 | 0 | ps_search_results, |
8033 | 0 | ps_cu_results, |
8034 | 0 | ps_pu_results, |
8035 | 0 | &(as_pu_results[0][0][0]), |
8036 | 0 | &s_common_frm_prms, |
8037 | 0 | au1_pred_dir_searched, |
8038 | 0 | i4_num_pred_dir, |
8039 | 0 | blk_8x8_mask); |
8040 | | |
8041 | | /* Re-initialize the pu_results pointers to the first struct in the stack array */ |
8042 | 0 | ps_pu_results = as_inter_pu_results; |
8043 | |
|
8044 | 0 | for(i = 0; i < 4; i++) |
8045 | 0 | { |
8046 | 0 | if((blk_8x8_mask & (1 << i))) |
8047 | 0 | { |
8048 | 0 | if(ps_cu_results->i4_part_mask) |
8049 | 0 | { |
8050 | 0 | hme_decide_part_types( |
8051 | 0 | ps_cu_results, |
8052 | 0 | ps_pu_results, |
8053 | 0 | &s_common_frm_prms, |
8054 | 0 | ps_ctxt, |
8055 | 0 | ps_cmn_utils_optimised_function_list, |
8056 | 0 | ps_me_optimised_function_list |
8057 | |
|
8058 | 0 | ); |
8059 | 0 | } |
8060 | | /*****************************************************************/ |
8061 | | /* INSERT INTRA RESULTS AT 8x8 LEVEL. */ |
8062 | | /*****************************************************************/ |
8063 | 0 | #if DISABLE_INTRA_IN_BPICS |
8064 | 0 | if(1 != ((ME_XTREME_SPEED_25 == e_me_quality_presets) && |
8065 | 0 | (ps_ctxt->s_frm_prms.i4_temporal_layer_id > |
8066 | 0 | TEMPORAL_LAYER_DISABLE))) |
8067 | 0 | #endif |
8068 | 0 | { |
8069 | 0 | if(!(DISABLE_INTRA_WHEN_NOISY && |
8070 | 0 | s_common_frm_prms.u1_is_cu_noisy)) |
8071 | 0 | { |
8072 | 0 | hme_insert_intra_nodes_post_bipred( |
8073 | 0 | ps_cu_results, ps_cur_ipe_ctb, ps_ctxt->frm_qstep); |
8074 | 0 | } |
8075 | 0 | } |
8076 | |
|
8077 | 0 | child_cost += ps_cu_results->ps_best_results[0].i4_tot_cost; |
8078 | 0 | } |
8079 | |
|
8080 | 0 | ps_cu_results++; |
8081 | 0 | ps_pu_results++; |
8082 | 0 | } |
8083 | | |
8084 | | /* Compare 16x16 vs 8x8 cost */ |
8085 | 0 | if(child_cost < parent_cost) |
8086 | 0 | { |
8087 | 0 | ps_search_results->best_cu_cost = child_cost; |
8088 | 0 | ps_search_results->u1_split_flag = 1; |
8089 | 0 | } |
8090 | 0 | } |
8091 | 0 | } |
8092 | | |
8093 | 0 | hme_update_mv_bank_encode( |
8094 | 0 | ps_search_results, |
8095 | 0 | ps_curr_layer->ps_layer_mvbank, |
8096 | 0 | blk_x, |
8097 | 0 | blk_y, |
8098 | 0 | &s_mv_update_prms, |
8099 | 0 | au1_pred_dir_searched, |
8100 | 0 | i4_num_act_ref_l0); |
8101 | | |
8102 | | /*********************************************************************/ |
8103 | | /* Map the best results to an MV Grid. This is a 18x18 grid that is */ |
8104 | | /* useful for doing things like predictor for cost calculation or */ |
8105 | | /* also for merge calculations if need be. */ |
8106 | | /*********************************************************************/ |
8107 | 0 | hme_map_mvs_to_grid( |
8108 | 0 | &aps_mv_grid[0], ps_search_results, au1_pred_dir_searched, i4_num_pred_dir); |
8109 | 0 | } |
8110 | | |
8111 | | /* Set the CU tree nodes appropriately */ |
8112 | 0 | if(e_me_quality_presets != ME_PRISTINE_QUALITY) |
8113 | 0 | { |
8114 | 0 | WORD32 i, j; |
8115 | |
|
8116 | 0 | for(i = 0; i < 16; i++) |
8117 | 0 | { |
8118 | 0 | cur_ctb_cu_tree_t *ps_tree_node = |
8119 | 0 | ps_ctxt->ps_cu_tree_curr_row + (i4_ctb_x * MAX_NUM_NODES_CU_TREE); |
8120 | 0 | search_results_t *ps_results = &ps_ctxt->as_search_results_16x16[i]; |
8121 | |
|
8122 | 0 | switch(i >> 2) |
8123 | 0 | { |
8124 | 0 | case 0: |
8125 | 0 | { |
8126 | 0 | ps_tree_node = ps_tree_node->ps_child_node_tl; |
8127 | |
|
8128 | 0 | break; |
8129 | 0 | } |
8130 | 0 | case 1: |
8131 | 0 | { |
8132 | 0 | ps_tree_node = ps_tree_node->ps_child_node_tr; |
8133 | |
|
8134 | 0 | break; |
8135 | 0 | } |
8136 | 0 | case 2: |
8137 | 0 | { |
8138 | 0 | ps_tree_node = ps_tree_node->ps_child_node_bl; |
8139 | |
|
8140 | 0 | break; |
8141 | 0 | } |
8142 | 0 | case 3: |
8143 | 0 | { |
8144 | 0 | ps_tree_node = ps_tree_node->ps_child_node_br; |
8145 | |
|
8146 | 0 | break; |
8147 | 0 | } |
8148 | 0 | } |
8149 | | |
8150 | 0 | switch(i % 4) |
8151 | 0 | { |
8152 | 0 | case 0: |
8153 | 0 | { |
8154 | 0 | ps_tree_node = ps_tree_node->ps_child_node_tl; |
8155 | |
|
8156 | 0 | break; |
8157 | 0 | } |
8158 | 0 | case 1: |
8159 | 0 | { |
8160 | 0 | ps_tree_node = ps_tree_node->ps_child_node_tr; |
8161 | |
|
8162 | 0 | break; |
8163 | 0 | } |
8164 | 0 | case 2: |
8165 | 0 | { |
8166 | 0 | ps_tree_node = ps_tree_node->ps_child_node_bl; |
8167 | |
|
8168 | 0 | break; |
8169 | 0 | } |
8170 | 0 | case 3: |
8171 | 0 | { |
8172 | 0 | ps_tree_node = ps_tree_node->ps_child_node_br; |
8173 | |
|
8174 | 0 | break; |
8175 | 0 | } |
8176 | 0 | } |
8177 | | |
8178 | 0 | if(ai4_blk_8x8_mask[i] == 15) |
8179 | 0 | { |
8180 | 0 | if(!ps_results->u1_split_flag) |
8181 | 0 | { |
8182 | 0 | ps_tree_node->is_node_valid = 1; |
8183 | 0 | NULLIFY_THE_CHILDREN_NODES(ps_tree_node); |
8184 | 0 | } |
8185 | 0 | else |
8186 | 0 | { |
8187 | 0 | ps_tree_node->is_node_valid = 0; |
8188 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree_node); |
8189 | 0 | } |
8190 | 0 | } |
8191 | 0 | else |
8192 | 0 | { |
8193 | 0 | cur_ctb_cu_tree_t *ps_tree_child; |
8194 | |
|
8195 | 0 | ps_tree_node->is_node_valid = 0; |
8196 | |
|
8197 | 0 | for(j = 0; j < 4; j++) |
8198 | 0 | { |
8199 | 0 | switch(j) |
8200 | 0 | { |
8201 | 0 | case 0: |
8202 | 0 | { |
8203 | 0 | ps_tree_child = ps_tree_node->ps_child_node_tl; |
8204 | |
|
8205 | 0 | break; |
8206 | 0 | } |
8207 | 0 | case 1: |
8208 | 0 | { |
8209 | 0 | ps_tree_child = ps_tree_node->ps_child_node_tr; |
8210 | |
|
8211 | 0 | break; |
8212 | 0 | } |
8213 | 0 | case 2: |
8214 | 0 | { |
8215 | 0 | ps_tree_child = ps_tree_node->ps_child_node_bl; |
8216 | |
|
8217 | 0 | break; |
8218 | 0 | } |
8219 | 0 | case 3: |
8220 | 0 | { |
8221 | 0 | ps_tree_child = ps_tree_node->ps_child_node_br; |
8222 | |
|
8223 | 0 | break; |
8224 | 0 | } |
8225 | 0 | } |
8226 | | |
8227 | 0 | ps_tree_child->is_node_valid = !!(ai4_blk_8x8_mask[i] & (1 << j)); |
8228 | 0 | } |
8229 | 0 | } |
8230 | 0 | } |
8231 | 0 | } |
8232 | | |
8233 | 0 | if(ME_PRISTINE_QUALITY == e_me_quality_presets) |
8234 | 0 | { |
8235 | 0 | cur_ctb_cu_tree_t *ps_tree = ps_ctb_cluster_info->ps_cu_tree_root; |
8236 | |
|
8237 | 0 | hme_analyse_mv_clustering( |
8238 | 0 | ps_ctxt->as_search_results_16x16, |
8239 | 0 | ps_ctxt->as_cu16x16_results, |
8240 | 0 | ps_ctxt->as_cu8x8_results, |
8241 | 0 | ps_ctxt->ps_ctb_cluster_info, |
8242 | 0 | ps_ctxt->ai1_future_list, |
8243 | 0 | ps_ctxt->ai1_past_list, |
8244 | 0 | ps_ctxt->s_frm_prms.bidir_enabled, |
8245 | 0 | e_me_quality_presets); |
8246 | |
|
8247 | | #if DISABLE_BLK_MERGE_WHEN_NOISY |
8248 | | ps_tree->ps_child_node_tl->is_node_valid = !au1_is_32x32Blk_noisy[0]; |
8249 | | ps_tree->ps_child_node_tr->is_node_valid = !au1_is_32x32Blk_noisy[1]; |
8250 | | ps_tree->ps_child_node_bl->is_node_valid = !au1_is_32x32Blk_noisy[2]; |
8251 | | ps_tree->ps_child_node_br->is_node_valid = !au1_is_32x32Blk_noisy[3]; |
8252 | | ps_tree->ps_child_node_tl->u1_inter_eval_enable = !au1_is_32x32Blk_noisy[0]; |
8253 | | ps_tree->ps_child_node_tr->u1_inter_eval_enable = !au1_is_32x32Blk_noisy[1]; |
8254 | | ps_tree->ps_child_node_bl->u1_inter_eval_enable = !au1_is_32x32Blk_noisy[2]; |
8255 | | ps_tree->ps_child_node_br->u1_inter_eval_enable = !au1_is_32x32Blk_noisy[3]; |
8256 | | ps_tree->is_node_valid = !au1_is_64x64Blk_noisy[0]; |
8257 | | ps_tree->u1_inter_eval_enable = !au1_is_64x64Blk_noisy[0]; |
8258 | | #endif |
8259 | |
|
8260 | 0 | en_merge_32x32 = (ps_tree->ps_child_node_tl->is_node_valid << 0) | |
8261 | 0 | (ps_tree->ps_child_node_tr->is_node_valid << 1) | |
8262 | 0 | (ps_tree->ps_child_node_bl->is_node_valid << 2) | |
8263 | 0 | (ps_tree->ps_child_node_br->is_node_valid << 3); |
8264 | |
|
8265 | 0 | en_merge_execution = (ps_tree->ps_child_node_tl->u1_inter_eval_enable << 0) | |
8266 | 0 | (ps_tree->ps_child_node_tr->u1_inter_eval_enable << 1) | |
8267 | 0 | (ps_tree->ps_child_node_bl->u1_inter_eval_enable << 2) | |
8268 | 0 | (ps_tree->ps_child_node_br->u1_inter_eval_enable << 3) | |
8269 | 0 | (ps_tree->u1_inter_eval_enable << 4); |
8270 | 0 | } |
8271 | 0 | else |
8272 | 0 | { |
8273 | 0 | en_merge_execution = 0x1f; |
8274 | |
|
8275 | | #if DISABLE_BLK_MERGE_WHEN_NOISY |
8276 | | en_merge_32x32 = ((!au1_is_32x32Blk_noisy[0] << 0) & (en_merge_32x32 & 1)) | |
8277 | | ((!au1_is_32x32Blk_noisy[1] << 1) & (en_merge_32x32 & 2)) | |
8278 | | ((!au1_is_32x32Blk_noisy[2] << 2) & (en_merge_32x32 & 4)) | |
8279 | | ((!au1_is_32x32Blk_noisy[3] << 3) & (en_merge_32x32 & 8)); |
8280 | | #endif |
8281 | 0 | } |
8282 | | |
8283 | | /* Re-initialize the pu_results pointers to the first struct in the stack array */ |
8284 | 0 | ps_pu_results = as_inter_pu_results; |
8285 | |
|
8286 | 0 | { |
8287 | 0 | WORD32 ref_ctr; |
8288 | |
|
8289 | 0 | s_ctb_prms.i4_ctb_x = i4_ctb_x << 6; |
8290 | 0 | s_ctb_prms.i4_ctb_y = i4_ctb_y << 6; |
8291 | | |
8292 | | /* MV limit is different based on ref. PIC */ |
8293 | 0 | for(ref_ctr = 0; ref_ctr < num_act_ref_pics; ref_ctr++) |
8294 | 0 | { |
8295 | 0 | SCALE_RANGE_PRMS(as_range_prms_hpel[ref_ctr], as_range_prms_rec[ref_ctr], 1); |
8296 | 0 | SCALE_RANGE_PRMS(as_range_prms_qpel[ref_ctr], as_range_prms_rec[ref_ctr], 2); |
8297 | 0 | } |
8298 | |
|
8299 | 0 | e_merge_result = CU_SPLIT; |
8300 | 0 | merge_count_32x32 = 0; |
8301 | |
|
8302 | 0 | if((en_merge_32x32 & 1) && (en_merge_execution & 1)) |
8303 | 0 | { |
8304 | 0 | range_prms_t *ps_pic_limit; |
8305 | 0 | if(s_merge_prms_32x32_tl.i4_use_rec == 1) |
8306 | 0 | { |
8307 | 0 | ps_pic_limit = &s_pic_limit_rec; |
8308 | 0 | } |
8309 | 0 | else |
8310 | 0 | { |
8311 | 0 | ps_pic_limit = &s_pic_limit_inp; |
8312 | 0 | } |
8313 | | /* MV limit is different based on ref. PIC */ |
8314 | 0 | for(ref_ctr = 0; ref_ctr < num_act_ref_pics; ref_ctr++) |
8315 | 0 | { |
8316 | 0 | hme_derive_search_range( |
8317 | 0 | s_merge_prms_32x32_tl.aps_mv_range[ref_ctr], |
8318 | 0 | ps_pic_limit, |
8319 | 0 | &as_mv_limit[ref_ctr], |
8320 | 0 | i4_ctb_x << 6, |
8321 | 0 | i4_ctb_y << 6, |
8322 | 0 | 32, |
8323 | 0 | 32); |
8324 | |
|
8325 | 0 | SCALE_RANGE_PRMS_POINTERS( |
8326 | 0 | s_merge_prms_32x32_tl.aps_mv_range[ref_ctr], |
8327 | 0 | s_merge_prms_32x32_tl.aps_mv_range[ref_ctr], |
8328 | 0 | 2); |
8329 | 0 | } |
8330 | 0 | s_merge_prms_32x32_tl.i4_ctb_x_off = i4_ctb_x << 6; |
8331 | 0 | s_merge_prms_32x32_tl.i4_ctb_y_off = i4_ctb_y << 6; |
8332 | 0 | s_subpel_prms.u1_is_cu_noisy = au1_is_32x32Blk_noisy[0]; |
8333 | |
|
8334 | 0 | e_merge_result = hme_try_merge_high_speed( |
8335 | 0 | ps_thrd_ctxt, |
8336 | 0 | ps_ctxt, |
8337 | 0 | ps_cur_ipe_ctb, |
8338 | 0 | &s_subpel_prms, |
8339 | 0 | &s_merge_prms_32x32_tl, |
8340 | 0 | ps_pu_results, |
8341 | 0 | &as_pu_results[0][0][0]); |
8342 | |
|
8343 | 0 | if(e_merge_result == CU_MERGED) |
8344 | 0 | { |
8345 | 0 | inter_cu_results_t *ps_cu_results = |
8346 | 0 | s_merge_prms_32x32_tl.ps_results_merge->ps_cu_results; |
8347 | |
|
8348 | 0 | if(!((ps_cu_results->u1_num_best_results == 1) && |
8349 | 0 | (ps_cu_results->ps_best_results->as_pu_results->pu.b1_intra_flag))) |
8350 | 0 | { |
8351 | 0 | hme_map_mvs_to_grid( |
8352 | 0 | &aps_mv_grid[0], |
8353 | 0 | s_merge_prms_32x32_tl.ps_results_merge, |
8354 | 0 | s_merge_prms_32x32_tl.au1_pred_dir_searched, |
8355 | 0 | s_merge_prms_32x32_tl.i4_num_pred_dir_actual); |
8356 | 0 | } |
8357 | |
|
8358 | 0 | if(ME_PRISTINE_QUALITY != e_me_quality_presets) |
8359 | 0 | { |
8360 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8361 | 0 | .ps_child_node_tl->is_node_valid = 1; |
8362 | 0 | NULLIFY_THE_CHILDREN_NODES( |
8363 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8364 | 0 | .ps_child_node_tl); |
8365 | 0 | } |
8366 | |
|
8367 | 0 | merge_count_32x32++; |
8368 | 0 | e_merge_result = CU_SPLIT; |
8369 | 0 | } |
8370 | 0 | else if(ME_PRISTINE_QUALITY == e_me_quality_presets) |
8371 | 0 | { |
8372 | 0 | #if ENABLE_CU_TREE_CULLING |
8373 | 0 | cur_ctb_cu_tree_t *ps_tree = |
8374 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_tl; |
8375 | |
|
8376 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->is_node_valid = 0; |
8377 | 0 | en_merge_execution = (en_merge_execution & (~(1 << 4))); |
8378 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree); |
8379 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tl); |
8380 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tr); |
8381 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_bl); |
8382 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_br); |
8383 | 0 | #endif |
8384 | 0 | } |
8385 | 0 | } |
8386 | 0 | else if((en_merge_32x32 & 1) && (!(en_merge_execution & 1))) |
8387 | 0 | { |
8388 | 0 | #if ENABLE_CU_TREE_CULLING |
8389 | 0 | cur_ctb_cu_tree_t *ps_tree = |
8390 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_tl; |
8391 | |
|
8392 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree); |
8393 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tl); |
8394 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tr); |
8395 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_bl); |
8396 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_br); |
8397 | 0 | #endif |
8398 | |
|
8399 | 0 | if(au1_is_32x32Blk_noisy[0] && DISABLE_INTRA_WHEN_NOISY) |
8400 | 0 | { |
8401 | 0 | ps_tree->is_node_valid = 0; |
8402 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->is_node_valid = 0; |
8403 | 0 | en_merge_execution = (en_merge_execution & (~(1 << 4))); |
8404 | 0 | } |
8405 | 0 | } |
8406 | |
|
8407 | 0 | if((en_merge_32x32 & 2) && (en_merge_execution & 2)) |
8408 | 0 | { |
8409 | 0 | range_prms_t *ps_pic_limit; |
8410 | 0 | if(s_merge_prms_32x32_tr.i4_use_rec == 1) |
8411 | 0 | { |
8412 | 0 | ps_pic_limit = &s_pic_limit_rec; |
8413 | 0 | } |
8414 | 0 | else |
8415 | 0 | { |
8416 | 0 | ps_pic_limit = &s_pic_limit_inp; |
8417 | 0 | } |
8418 | | /* MV limit is different based on ref. PIC */ |
8419 | 0 | for(ref_ctr = 0; ref_ctr < num_act_ref_pics; ref_ctr++) |
8420 | 0 | { |
8421 | 0 | hme_derive_search_range( |
8422 | 0 | s_merge_prms_32x32_tr.aps_mv_range[ref_ctr], |
8423 | 0 | ps_pic_limit, |
8424 | 0 | &as_mv_limit[ref_ctr], |
8425 | 0 | (i4_ctb_x << 6) + 32, |
8426 | 0 | i4_ctb_y << 6, |
8427 | 0 | 32, |
8428 | 0 | 32); |
8429 | 0 | SCALE_RANGE_PRMS_POINTERS( |
8430 | 0 | s_merge_prms_32x32_tr.aps_mv_range[ref_ctr], |
8431 | 0 | s_merge_prms_32x32_tr.aps_mv_range[ref_ctr], |
8432 | 0 | 2); |
8433 | 0 | } |
8434 | 0 | s_merge_prms_32x32_tr.i4_ctb_x_off = i4_ctb_x << 6; |
8435 | 0 | s_merge_prms_32x32_tr.i4_ctb_y_off = i4_ctb_y << 6; |
8436 | 0 | s_subpel_prms.u1_is_cu_noisy = au1_is_32x32Blk_noisy[1]; |
8437 | |
|
8438 | 0 | e_merge_result = hme_try_merge_high_speed( |
8439 | 0 | ps_thrd_ctxt, |
8440 | 0 | ps_ctxt, |
8441 | 0 | ps_cur_ipe_ctb, |
8442 | 0 | &s_subpel_prms, |
8443 | 0 | &s_merge_prms_32x32_tr, |
8444 | 0 | ps_pu_results, |
8445 | 0 | &as_pu_results[0][0][0]); |
8446 | |
|
8447 | 0 | if(e_merge_result == CU_MERGED) |
8448 | 0 | { |
8449 | 0 | inter_cu_results_t *ps_cu_results = |
8450 | 0 | s_merge_prms_32x32_tr.ps_results_merge->ps_cu_results; |
8451 | |
|
8452 | 0 | if(!((ps_cu_results->u1_num_best_results == 1) && |
8453 | 0 | (ps_cu_results->ps_best_results->as_pu_results->pu.b1_intra_flag))) |
8454 | 0 | { |
8455 | 0 | hme_map_mvs_to_grid( |
8456 | 0 | &aps_mv_grid[0], |
8457 | 0 | s_merge_prms_32x32_tr.ps_results_merge, |
8458 | 0 | s_merge_prms_32x32_tr.au1_pred_dir_searched, |
8459 | 0 | s_merge_prms_32x32_tr.i4_num_pred_dir_actual); |
8460 | 0 | } |
8461 | |
|
8462 | 0 | if(ME_PRISTINE_QUALITY != e_me_quality_presets) |
8463 | 0 | { |
8464 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8465 | 0 | .ps_child_node_tr->is_node_valid = 1; |
8466 | 0 | NULLIFY_THE_CHILDREN_NODES( |
8467 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8468 | 0 | .ps_child_node_tr); |
8469 | 0 | } |
8470 | |
|
8471 | 0 | merge_count_32x32++; |
8472 | 0 | e_merge_result = CU_SPLIT; |
8473 | 0 | } |
8474 | 0 | else if(ME_PRISTINE_QUALITY == e_me_quality_presets) |
8475 | 0 | { |
8476 | 0 | #if ENABLE_CU_TREE_CULLING |
8477 | 0 | cur_ctb_cu_tree_t *ps_tree = |
8478 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_tr; |
8479 | |
|
8480 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->is_node_valid = 0; |
8481 | 0 | en_merge_execution = (en_merge_execution & (~(1 << 4))); |
8482 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree); |
8483 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tl); |
8484 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tr); |
8485 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_bl); |
8486 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_br); |
8487 | 0 | #endif |
8488 | 0 | } |
8489 | 0 | } |
8490 | 0 | else if((en_merge_32x32 & 2) && (!(en_merge_execution & 2))) |
8491 | 0 | { |
8492 | 0 | #if ENABLE_CU_TREE_CULLING |
8493 | 0 | cur_ctb_cu_tree_t *ps_tree = |
8494 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_tr; |
8495 | |
|
8496 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree); |
8497 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tl); |
8498 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tr); |
8499 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_bl); |
8500 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_br); |
8501 | 0 | #endif |
8502 | |
|
8503 | 0 | if(au1_is_32x32Blk_noisy[1] && DISABLE_INTRA_WHEN_NOISY) |
8504 | 0 | { |
8505 | 0 | ps_tree->is_node_valid = 0; |
8506 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->is_node_valid = 0; |
8507 | 0 | en_merge_execution = (en_merge_execution & (~(1 << 4))); |
8508 | 0 | } |
8509 | 0 | } |
8510 | |
|
8511 | 0 | if((en_merge_32x32 & 4) && (en_merge_execution & 4)) |
8512 | 0 | { |
8513 | 0 | range_prms_t *ps_pic_limit; |
8514 | 0 | if(s_merge_prms_32x32_bl.i4_use_rec == 1) |
8515 | 0 | { |
8516 | 0 | ps_pic_limit = &s_pic_limit_rec; |
8517 | 0 | } |
8518 | 0 | else |
8519 | 0 | { |
8520 | 0 | ps_pic_limit = &s_pic_limit_inp; |
8521 | 0 | } |
8522 | | /* MV limit is different based on ref. PIC */ |
8523 | 0 | for(ref_ctr = 0; ref_ctr < num_act_ref_pics; ref_ctr++) |
8524 | 0 | { |
8525 | 0 | hme_derive_search_range( |
8526 | 0 | s_merge_prms_32x32_bl.aps_mv_range[ref_ctr], |
8527 | 0 | ps_pic_limit, |
8528 | 0 | &as_mv_limit[ref_ctr], |
8529 | 0 | i4_ctb_x << 6, |
8530 | 0 | (i4_ctb_y << 6) + 32, |
8531 | 0 | 32, |
8532 | 0 | 32); |
8533 | 0 | SCALE_RANGE_PRMS_POINTERS( |
8534 | 0 | s_merge_prms_32x32_bl.aps_mv_range[ref_ctr], |
8535 | 0 | s_merge_prms_32x32_bl.aps_mv_range[ref_ctr], |
8536 | 0 | 2); |
8537 | 0 | } |
8538 | 0 | s_merge_prms_32x32_bl.i4_ctb_x_off = i4_ctb_x << 6; |
8539 | 0 | s_merge_prms_32x32_bl.i4_ctb_y_off = i4_ctb_y << 6; |
8540 | 0 | s_subpel_prms.u1_is_cu_noisy = au1_is_32x32Blk_noisy[2]; |
8541 | |
|
8542 | 0 | e_merge_result = hme_try_merge_high_speed( |
8543 | 0 | ps_thrd_ctxt, |
8544 | 0 | ps_ctxt, |
8545 | 0 | ps_cur_ipe_ctb, |
8546 | 0 | &s_subpel_prms, |
8547 | 0 | &s_merge_prms_32x32_bl, |
8548 | 0 | ps_pu_results, |
8549 | 0 | &as_pu_results[0][0][0]); |
8550 | |
|
8551 | 0 | if(e_merge_result == CU_MERGED) |
8552 | 0 | { |
8553 | 0 | inter_cu_results_t *ps_cu_results = |
8554 | 0 | s_merge_prms_32x32_bl.ps_results_merge->ps_cu_results; |
8555 | |
|
8556 | 0 | if(!((ps_cu_results->u1_num_best_results == 1) && |
8557 | 0 | (ps_cu_results->ps_best_results->as_pu_results->pu.b1_intra_flag))) |
8558 | 0 | { |
8559 | 0 | hme_map_mvs_to_grid( |
8560 | 0 | &aps_mv_grid[0], |
8561 | 0 | s_merge_prms_32x32_bl.ps_results_merge, |
8562 | 0 | s_merge_prms_32x32_bl.au1_pred_dir_searched, |
8563 | 0 | s_merge_prms_32x32_bl.i4_num_pred_dir_actual); |
8564 | 0 | } |
8565 | |
|
8566 | 0 | if(ME_PRISTINE_QUALITY != e_me_quality_presets) |
8567 | 0 | { |
8568 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8569 | 0 | .ps_child_node_bl->is_node_valid = 1; |
8570 | 0 | NULLIFY_THE_CHILDREN_NODES( |
8571 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8572 | 0 | .ps_child_node_bl); |
8573 | 0 | } |
8574 | |
|
8575 | 0 | merge_count_32x32++; |
8576 | 0 | e_merge_result = CU_SPLIT; |
8577 | 0 | } |
8578 | 0 | else if(ME_PRISTINE_QUALITY == e_me_quality_presets) |
8579 | 0 | { |
8580 | 0 | #if ENABLE_CU_TREE_CULLING |
8581 | 0 | cur_ctb_cu_tree_t *ps_tree = |
8582 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_bl; |
8583 | |
|
8584 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->is_node_valid = 0; |
8585 | 0 | en_merge_execution = (en_merge_execution & (~(1 << 4))); |
8586 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree); |
8587 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tl); |
8588 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tr); |
8589 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_bl); |
8590 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_br); |
8591 | 0 | #endif |
8592 | 0 | } |
8593 | 0 | } |
8594 | 0 | else if((en_merge_32x32 & 4) && (!(en_merge_execution & 4))) |
8595 | 0 | { |
8596 | 0 | #if ENABLE_CU_TREE_CULLING |
8597 | 0 | cur_ctb_cu_tree_t *ps_tree = |
8598 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_bl; |
8599 | |
|
8600 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree); |
8601 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tl); |
8602 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tr); |
8603 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_bl); |
8604 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_br); |
8605 | 0 | #endif |
8606 | |
|
8607 | 0 | if(au1_is_32x32Blk_noisy[2] && DISABLE_INTRA_WHEN_NOISY) |
8608 | 0 | { |
8609 | 0 | ps_tree->is_node_valid = 0; |
8610 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->is_node_valid = 0; |
8611 | 0 | en_merge_execution = (en_merge_execution & (~(1 << 4))); |
8612 | 0 | } |
8613 | 0 | } |
8614 | |
|
8615 | 0 | if((en_merge_32x32 & 8) && (en_merge_execution & 8)) |
8616 | 0 | { |
8617 | 0 | range_prms_t *ps_pic_limit; |
8618 | 0 | if(s_merge_prms_32x32_br.i4_use_rec == 1) |
8619 | 0 | { |
8620 | 0 | ps_pic_limit = &s_pic_limit_rec; |
8621 | 0 | } |
8622 | 0 | else |
8623 | 0 | { |
8624 | 0 | ps_pic_limit = &s_pic_limit_inp; |
8625 | 0 | } |
8626 | | /* MV limit is different based on ref. PIC */ |
8627 | 0 | for(ref_ctr = 0; ref_ctr < num_act_ref_pics; ref_ctr++) |
8628 | 0 | { |
8629 | 0 | hme_derive_search_range( |
8630 | 0 | s_merge_prms_32x32_br.aps_mv_range[ref_ctr], |
8631 | 0 | ps_pic_limit, |
8632 | 0 | &as_mv_limit[ref_ctr], |
8633 | 0 | (i4_ctb_x << 6) + 32, |
8634 | 0 | (i4_ctb_y << 6) + 32, |
8635 | 0 | 32, |
8636 | 0 | 32); |
8637 | |
|
8638 | 0 | SCALE_RANGE_PRMS_POINTERS( |
8639 | 0 | s_merge_prms_32x32_br.aps_mv_range[ref_ctr], |
8640 | 0 | s_merge_prms_32x32_br.aps_mv_range[ref_ctr], |
8641 | 0 | 2); |
8642 | 0 | } |
8643 | 0 | s_merge_prms_32x32_br.i4_ctb_x_off = i4_ctb_x << 6; |
8644 | 0 | s_merge_prms_32x32_br.i4_ctb_y_off = i4_ctb_y << 6; |
8645 | 0 | s_subpel_prms.u1_is_cu_noisy = au1_is_32x32Blk_noisy[3]; |
8646 | |
|
8647 | 0 | e_merge_result = hme_try_merge_high_speed( |
8648 | 0 | ps_thrd_ctxt, |
8649 | 0 | ps_ctxt, |
8650 | 0 | ps_cur_ipe_ctb, |
8651 | 0 | &s_subpel_prms, |
8652 | 0 | &s_merge_prms_32x32_br, |
8653 | 0 | ps_pu_results, |
8654 | 0 | &as_pu_results[0][0][0]); |
8655 | |
|
8656 | 0 | if(e_merge_result == CU_MERGED) |
8657 | 0 | { |
8658 | | /*inter_cu_results_t *ps_cu_results = s_merge_prms_32x32_br.ps_results_merge->ps_cu_results; |
8659 | | |
8660 | | if(!((ps_cu_results->u1_num_best_results == 1) && |
8661 | | (ps_cu_results->ps_best_results->as_pu_results->pu.b1_intra_flag))) |
8662 | | { |
8663 | | hme_map_mvs_to_grid |
8664 | | ( |
8665 | | &aps_mv_grid[0], |
8666 | | s_merge_prms_32x32_br.ps_results_merge, |
8667 | | s_merge_prms_32x32_br.au1_pred_dir_searched, |
8668 | | s_merge_prms_32x32_br.i4_num_pred_dir_actual |
8669 | | ); |
8670 | | }*/ |
8671 | |
|
8672 | 0 | if(ME_PRISTINE_QUALITY != e_me_quality_presets) |
8673 | 0 | { |
8674 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8675 | 0 | .ps_child_node_br->is_node_valid = 1; |
8676 | 0 | NULLIFY_THE_CHILDREN_NODES( |
8677 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8678 | 0 | .ps_child_node_br); |
8679 | 0 | } |
8680 | |
|
8681 | 0 | merge_count_32x32++; |
8682 | 0 | e_merge_result = CU_SPLIT; |
8683 | 0 | } |
8684 | 0 | else if(ME_PRISTINE_QUALITY == e_me_quality_presets) |
8685 | 0 | { |
8686 | 0 | #if ENABLE_CU_TREE_CULLING |
8687 | 0 | cur_ctb_cu_tree_t *ps_tree = |
8688 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_br; |
8689 | |
|
8690 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->is_node_valid = 0; |
8691 | 0 | en_merge_execution = (en_merge_execution & (~(1 << 4))); |
8692 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree); |
8693 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tl); |
8694 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tr); |
8695 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_bl); |
8696 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_br); |
8697 | 0 | #endif |
8698 | 0 | } |
8699 | 0 | } |
8700 | 0 | else if((en_merge_32x32 & 8) && (!(en_merge_execution & 8))) |
8701 | 0 | { |
8702 | 0 | #if ENABLE_CU_TREE_CULLING |
8703 | 0 | cur_ctb_cu_tree_t *ps_tree = |
8704 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->ps_child_node_br; |
8705 | |
|
8706 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree); |
8707 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tl); |
8708 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_tr); |
8709 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_bl); |
8710 | 0 | ENABLE_THE_CHILDREN_NODES(ps_tree->ps_child_node_br); |
8711 | 0 | #endif |
8712 | |
|
8713 | 0 | if(au1_is_32x32Blk_noisy[3] && DISABLE_INTRA_WHEN_NOISY) |
8714 | 0 | { |
8715 | 0 | ps_tree->is_node_valid = 0; |
8716 | 0 | ps_ctb_cluster_info->ps_cu_tree_root->is_node_valid = 0; |
8717 | 0 | en_merge_execution = (en_merge_execution & (~(1 << 4))); |
8718 | 0 | } |
8719 | 0 | } |
8720 | | |
8721 | | /* Try merging all 32x32 to 64x64 candts */ |
8722 | 0 | if(((en_merge_32x32 & 0xf) == 0xf) && |
8723 | 0 | (((merge_count_32x32 == 4) && (e_me_quality_presets != ME_PRISTINE_QUALITY)) || |
8724 | 0 | ((en_merge_execution & 16) && (e_me_quality_presets == ME_PRISTINE_QUALITY)))) |
8725 | 0 | if((((e_me_quality_presets == ME_XTREME_SPEED_25) && |
8726 | 0 | !DISABLE_64X64_BLOCK_MERGE_IN_ME_IN_XS25) || |
8727 | 0 | (e_me_quality_presets != ME_XTREME_SPEED_25))) |
8728 | 0 | { |
8729 | 0 | range_prms_t *ps_pic_limit; |
8730 | 0 | if(s_merge_prms_64x64.i4_use_rec == 1) |
8731 | 0 | { |
8732 | 0 | ps_pic_limit = &s_pic_limit_rec; |
8733 | 0 | } |
8734 | 0 | else |
8735 | 0 | { |
8736 | 0 | ps_pic_limit = &s_pic_limit_inp; |
8737 | 0 | } |
8738 | | /* MV limit is different based on ref. PIC */ |
8739 | 0 | for(ref_ctr = 0; ref_ctr < num_act_ref_pics; ref_ctr++) |
8740 | 0 | { |
8741 | 0 | hme_derive_search_range( |
8742 | 0 | s_merge_prms_64x64.aps_mv_range[ref_ctr], |
8743 | 0 | ps_pic_limit, |
8744 | 0 | &as_mv_limit[ref_ctr], |
8745 | 0 | i4_ctb_x << 6, |
8746 | 0 | i4_ctb_y << 6, |
8747 | 0 | 64, |
8748 | 0 | 64); |
8749 | |
|
8750 | 0 | SCALE_RANGE_PRMS_POINTERS( |
8751 | 0 | s_merge_prms_64x64.aps_mv_range[ref_ctr], |
8752 | 0 | s_merge_prms_64x64.aps_mv_range[ref_ctr], |
8753 | 0 | 2); |
8754 | 0 | } |
8755 | 0 | s_merge_prms_64x64.i4_ctb_x_off = i4_ctb_x << 6; |
8756 | 0 | s_merge_prms_64x64.i4_ctb_y_off = i4_ctb_y << 6; |
8757 | 0 | s_subpel_prms.u1_is_cu_noisy = au1_is_64x64Blk_noisy[0]; |
8758 | |
|
8759 | 0 | e_merge_result = hme_try_merge_high_speed( |
8760 | 0 | ps_thrd_ctxt, |
8761 | 0 | ps_ctxt, |
8762 | 0 | ps_cur_ipe_ctb, |
8763 | 0 | &s_subpel_prms, |
8764 | 0 | &s_merge_prms_64x64, |
8765 | 0 | ps_pu_results, |
8766 | 0 | &as_pu_results[0][0][0]); |
8767 | |
|
8768 | 0 | if((e_merge_result == CU_MERGED) && |
8769 | 0 | (ME_PRISTINE_QUALITY != e_me_quality_presets)) |
8770 | 0 | { |
8771 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8772 | 0 | .is_node_valid = 1; |
8773 | 0 | NULLIFY_THE_CHILDREN_NODES( |
8774 | 0 | ps_ctxt->ps_cu_tree_curr_row + (i4_ctb_x * MAX_NUM_NODES_CU_TREE)); |
8775 | 0 | } |
8776 | 0 | else if( |
8777 | 0 | (e_merge_result == CU_SPLIT) && |
8778 | 0 | (ME_PRISTINE_QUALITY == e_me_quality_presets)) |
8779 | 0 | { |
8780 | 0 | ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)] |
8781 | 0 | .is_node_valid = 0; |
8782 | 0 | } |
8783 | 0 | } |
8784 | | |
8785 | | /*****************************************************************/ |
8786 | | /* UPDATION OF RESULT TO EXTERNAL STRUCTURES */ |
8787 | | /*****************************************************************/ |
8788 | 0 | pf_ext_update_fxn((void *)ps_thrd_ctxt, (void *)ps_ctxt, i4_ctb_x, i4_ctb_y); |
8789 | |
|
8790 | 0 | { |
8791 | | #ifdef _DEBUG |
8792 | | S32 wd = ((i4_pic_wd - s_common_frm_prms.i4_ctb_x_off) >= 64) |
8793 | | ? 64 |
8794 | | : i4_pic_wd - s_common_frm_prms.i4_ctb_x_off; |
8795 | | S32 ht = ((i4_pic_ht - s_common_frm_prms.i4_ctb_y_off) >= 64) |
8796 | | ? 64 |
8797 | | : i4_pic_ht - s_common_frm_prms.i4_ctb_y_off; |
8798 | | ASSERT( |
8799 | | (wd * ht) == |
8800 | | ihevce_compute_area_of_valid_cus_in_ctb( |
8801 | | &ps_ctxt->ps_cu_tree_curr_row[(i4_ctb_x * MAX_NUM_NODES_CU_TREE)])); |
8802 | | #endif |
8803 | 0 | } |
8804 | 0 | } |
8805 | | |
8806 | | /* set the dependency for the corresponding row in enc loop */ |
8807 | 0 | ihevce_dmgr_set_row_row_sync( |
8808 | 0 | pv_dep_mngr_encloop_dep_me, |
8809 | 0 | (i4_ctb_x + 1), |
8810 | 0 | i4_ctb_y, |
8811 | 0 | tile_col_idx /* Col Tile No. */); |
8812 | |
|
8813 | 0 | left_ctb_in_diff_tile = 0; |
8814 | 0 | } |
8815 | 0 | } |
8816 | 0 | } |
8817 | | |
8818 | | /** |
8819 | | ******************************************************************************** |
8820 | | * @fn void hme_refine_no_encode(coarse_me_ctxt_t *ps_ctxt, |
8821 | | * refine_layer_prms_t *ps_refine_prms) |
8822 | | * |
8823 | | * @brief Top level entry point for refinement ME |
8824 | | * |
8825 | | * @param[in,out] ps_ctxt: ME Handle |
8826 | | * |
8827 | | * @param[in] ps_refine_prms : refinement layer prms |
8828 | | * |
8829 | | * @return None |
8830 | | ******************************************************************************** |
8831 | | */ |
8832 | | void hme_refine_no_encode( |
8833 | | coarse_me_ctxt_t *ps_ctxt, |
8834 | | refine_prms_t *ps_refine_prms, |
8835 | | multi_thrd_ctxt_t *ps_multi_thrd_ctxt, |
8836 | | S32 lyr_job_type, |
8837 | | WORD32 i4_ping_pong, |
8838 | | void **ppv_dep_mngr_hme_sync) |
8839 | 0 | { |
8840 | 0 | BLK_SIZE_T e_search_blk_size, e_result_blk_size; |
8841 | 0 | ME_QUALITY_PRESETS_T e_me_quality_presets = |
8842 | 0 | ps_ctxt->s_init_prms.s_me_coding_tools.e_me_quality_presets; |
8843 | | |
8844 | | /*************************************************************************/ |
8845 | | /* Complexity of search: Low to High */ |
8846 | | /*************************************************************************/ |
8847 | 0 | SEARCH_COMPLEXITY_T e_search_complexity; |
8848 | | |
8849 | | /*************************************************************************/ |
8850 | | /* Config parameter structures for varius ME submodules */ |
8851 | | /*************************************************************************/ |
8852 | 0 | hme_search_prms_t s_search_prms_blk; |
8853 | 0 | mvbank_update_prms_t s_mv_update_prms; |
8854 | | |
8855 | | /*************************************************************************/ |
8856 | | /* All types of search candidates for predictor based search. */ |
8857 | | /*************************************************************************/ |
8858 | 0 | S32 num_init_candts = 0; |
8859 | 0 | search_candt_t *ps_search_candts, as_search_candts[MAX_INIT_CANDTS]; |
8860 | 0 | search_node_t as_top_neighbours[4], as_left_neighbours[3]; |
8861 | 0 | search_node_t *ps_candt_zeromv, *ps_candt_tl, *ps_candt_tr; |
8862 | 0 | search_node_t *ps_candt_l, *ps_candt_t; |
8863 | 0 | search_node_t *ps_candt_prj_br[2], *ps_candt_prj_b[2], *ps_candt_prj_r[2]; |
8864 | 0 | search_node_t *ps_candt_prj_bl[2]; |
8865 | 0 | search_node_t *ps_candt_prj_tr[2], *ps_candt_prj_t[2], *ps_candt_prj_tl[2]; |
8866 | 0 | search_node_t *ps_candt_prj_coloc[2]; |
8867 | |
|
8868 | 0 | pf_get_wt_inp fp_get_wt_inp; |
8869 | |
|
8870 | 0 | search_node_t as_unique_search_nodes[MAX_INIT_CANDTS * 9]; |
8871 | 0 | U32 au4_unique_node_map[MAP_X_MAX * 2]; |
8872 | | |
8873 | | /*EIID */ |
8874 | 0 | WORD32 i4_num_inter_wins = 0; //debug code to find stat of |
8875 | 0 | WORD32 i4_num_comparisions = 0; //debug code |
8876 | 0 | WORD32 i4_threshold_multiplier; |
8877 | 0 | WORD32 i4_threshold_divider; |
8878 | 0 | WORD32 i4_temporal_layer = |
8879 | 0 | ps_multi_thrd_ctxt->aps_curr_inp_pre_enc[i4_ping_pong]->s_lap_out.i4_temporal_lyr_id; |
8880 | | |
8881 | | /*************************************************************************/ |
8882 | | /* points ot the search results for the blk level search (8x8/16x16) */ |
8883 | | /*************************************************************************/ |
8884 | 0 | search_results_t *ps_search_results; |
8885 | | |
8886 | | /*************************************************************************/ |
8887 | | /* Coordinates */ |
8888 | | /*************************************************************************/ |
8889 | 0 | S32 blk_x, i4_ctb_x, blk_id_in_ctb; |
8890 | | //S32 i4_ctb_y; |
8891 | 0 | S32 pos_x, pos_y; |
8892 | 0 | S32 blk_id_in_full_ctb; |
8893 | 0 | S32 i4_num_srch_cands; |
8894 | |
|
8895 | 0 | S32 blk_y; |
8896 | | |
8897 | | /*************************************************************************/ |
8898 | | /* Related to dimensions of block being searched and pic dimensions */ |
8899 | | /*************************************************************************/ |
8900 | 0 | S32 blk_wd, blk_ht, blk_size_shift, num_blks_in_row, num_blks_in_pic; |
8901 | 0 | S32 i4_pic_wd, i4_pic_ht, num_blks_in_this_ctb; |
8902 | 0 | S32 num_results_prev_layer; |
8903 | | |
8904 | | /*************************************************************************/ |
8905 | | /* Size of a basic unit for this layer. For non encode layers, we search */ |
8906 | | /* in block sizes of 8x8. For encode layers, though we search 16x16s the */ |
8907 | | /* basic unit size is the ctb size. */ |
8908 | | /*************************************************************************/ |
8909 | 0 | S32 unit_size; |
8910 | | |
8911 | | /*************************************************************************/ |
8912 | | /* Pointers to context in current and coarser layers */ |
8913 | | /*************************************************************************/ |
8914 | 0 | layer_ctxt_t *ps_curr_layer, *ps_coarse_layer; |
8915 | | |
8916 | | /*************************************************************************/ |
8917 | | /* to store mv range per blk, and picture limit, allowed search range */ |
8918 | | /* range prms in hpel and qpel units as well */ |
8919 | | /*************************************************************************/ |
8920 | 0 | range_prms_t s_range_prms_inp, s_range_prms_rec; |
8921 | 0 | range_prms_t s_pic_limit_inp, s_pic_limit_rec, as_mv_limit[MAX_NUM_REF]; |
8922 | | /*************************************************************************/ |
8923 | | /* These variables are used to track number of references at different */ |
8924 | | /* stages of ME. */ |
8925 | | /*************************************************************************/ |
8926 | 0 | S32 i4_num_ref_fpel, i4_num_ref_before_merge; |
8927 | 0 | S32 i4_num_ref_each_dir, i, i4_num_ref_prev_layer; |
8928 | 0 | S32 lambda_inp = ps_refine_prms->lambda_inp; |
8929 | | |
8930 | | /*************************************************************************/ |
8931 | | /* When a layer is implicit, it means that it searches on 1 or 2 ref idx */ |
8932 | | /* Explicit means it searches on all active ref idx. */ |
8933 | | /*************************************************************************/ |
8934 | 0 | S32 curr_layer_implicit, prev_layer_implicit; |
8935 | | |
8936 | | /*************************************************************************/ |
8937 | | /* Variables for loop counts */ |
8938 | | /*************************************************************************/ |
8939 | 0 | S32 id; |
8940 | 0 | S08 i1_ref_idx; |
8941 | | |
8942 | | /*************************************************************************/ |
8943 | | /* Input pointer and stride */ |
8944 | | /*************************************************************************/ |
8945 | 0 | U08 *pu1_inp; |
8946 | 0 | S32 i4_inp_stride; |
8947 | |
|
8948 | 0 | S32 end_of_frame; |
8949 | |
|
8950 | 0 | S32 num_sync_units_in_row; |
8951 | |
|
8952 | 0 | PF_HME_PROJECT_COLOC_CANDT_FXN pf_hme_project_coloc_candt; |
8953 | 0 | ASSERT(ps_refine_prms->i4_layer_id < ps_ctxt->num_layers - 1); |
8954 | | |
8955 | | /*************************************************************************/ |
8956 | | /* Pointers to current and coarse layer are needed for projection */ |
8957 | | /* Pointer to prev layer are needed for other candts like coloc */ |
8958 | | /*************************************************************************/ |
8959 | 0 | ps_curr_layer = ps_ctxt->ps_curr_descr->aps_layers[ps_refine_prms->i4_layer_id]; |
8960 | |
|
8961 | 0 | ps_coarse_layer = ps_ctxt->ps_curr_descr->aps_layers[ps_refine_prms->i4_layer_id + 1]; |
8962 | |
|
8963 | 0 | num_results_prev_layer = ps_coarse_layer->ps_layer_mvbank->i4_num_mvs_per_ref; |
8964 | | |
8965 | | /* Function pointer is selected based on the C vc X86 macro */ |
8966 | |
|
8967 | 0 | fp_get_wt_inp = ((ihevce_me_optimised_function_list_t *)ps_ctxt->pv_me_optimised_function_list) |
8968 | 0 | ->pf_get_wt_inp_8x8; |
8969 | |
|
8970 | 0 | i4_inp_stride = ps_curr_layer->i4_inp_stride; |
8971 | 0 | i4_pic_wd = ps_curr_layer->i4_wd; |
8972 | 0 | i4_pic_ht = ps_curr_layer->i4_ht; |
8973 | 0 | e_search_complexity = ps_refine_prms->e_search_complexity; |
8974 | |
|
8975 | 0 | end_of_frame = 0; |
8976 | | |
8977 | | /* If the previous layer is non-encode layer, then use dyadic projection */ |
8978 | 0 | if(0 == ps_ctxt->u1_encode[ps_refine_prms->i4_layer_id + 1]) |
8979 | 0 | pf_hme_project_coloc_candt = hme_project_coloc_candt_dyadic; |
8980 | 0 | else |
8981 | 0 | pf_hme_project_coloc_candt = hme_project_coloc_candt; |
8982 | | |
8983 | | /* This points to all the initial candts */ |
8984 | 0 | ps_search_candts = &as_search_candts[0]; |
8985 | |
|
8986 | 0 | { |
8987 | 0 | e_search_blk_size = BLK_8x8; |
8988 | 0 | blk_wd = blk_ht = 8; |
8989 | 0 | blk_size_shift = 3; |
8990 | 0 | s_mv_update_prms.i4_shift = 0; |
8991 | | /*********************************************************************/ |
8992 | | /* In case we do not encode this layer, we search 8x8 with or without*/ |
8993 | | /* enable 4x4 SAD. */ |
8994 | | /*********************************************************************/ |
8995 | 0 | { |
8996 | 0 | S32 i4_mask = (ENABLE_2Nx2N); |
8997 | |
|
8998 | 0 | e_result_blk_size = BLK_8x8; |
8999 | 0 | if(ps_refine_prms->i4_enable_4x4_part) |
9000 | 0 | { |
9001 | 0 | i4_mask |= (ENABLE_NxN); |
9002 | 0 | e_result_blk_size = BLK_4x4; |
9003 | 0 | s_mv_update_prms.i4_shift = 1; |
9004 | 0 | } |
9005 | |
|
9006 | 0 | s_search_prms_blk.i4_part_mask = i4_mask; |
9007 | 0 | } |
9008 | |
|
9009 | 0 | unit_size = blk_wd; |
9010 | 0 | s_search_prms_blk.i4_inp_stride = unit_size; |
9011 | 0 | } |
9012 | | |
9013 | | /* This is required to properly update the layer mv bank */ |
9014 | 0 | s_mv_update_prms.e_search_blk_size = e_search_blk_size; |
9015 | 0 | s_search_prms_blk.e_blk_size = e_search_blk_size; |
9016 | | |
9017 | | /*************************************************************************/ |
9018 | | /* If current layer is explicit, then the number of ref frames are to */ |
9019 | | /* be same as previous layer. Else it will be 2 */ |
9020 | | /*************************************************************************/ |
9021 | 0 | i4_num_ref_prev_layer = ps_coarse_layer->ps_layer_mvbank->i4_num_ref; |
9022 | 0 | if(ps_refine_prms->explicit_ref) |
9023 | 0 | { |
9024 | 0 | curr_layer_implicit = 0; |
9025 | 0 | i4_num_ref_fpel = i4_num_ref_prev_layer; |
9026 | | /* 100578 : Using same mv cost fun. for all presets. */ |
9027 | 0 | s_search_prms_blk.pf_mv_cost_compute = compute_mv_cost_refine; |
9028 | 0 | } |
9029 | 0 | else |
9030 | 0 | { |
9031 | 0 | i4_num_ref_fpel = 2; |
9032 | 0 | curr_layer_implicit = 1; |
9033 | 0 | { |
9034 | 0 | if(ME_MEDIUM_SPEED > e_me_quality_presets) |
9035 | 0 | { |
9036 | 0 | s_search_prms_blk.pf_mv_cost_compute = compute_mv_cost_implicit; |
9037 | 0 | } |
9038 | 0 | else |
9039 | 0 | { |
9040 | 0 | #if USE_MODIFIED == 1 |
9041 | 0 | s_search_prms_blk.pf_mv_cost_compute = compute_mv_cost_implicit_high_speed_modified; |
9042 | | #else |
9043 | | s_search_prms_blk.pf_mv_cost_compute = compute_mv_cost_implicit_high_speed; |
9044 | | #endif |
9045 | 0 | } |
9046 | 0 | } |
9047 | 0 | } |
9048 | |
|
9049 | 0 | i4_num_ref_fpel = MIN(i4_num_ref_fpel, i4_num_ref_prev_layer); |
9050 | 0 | if(ps_multi_thrd_ctxt->aps_curr_inp_pre_enc[i4_ping_pong]->s_lap_out.i4_pic_type == |
9051 | 0 | IV_IDR_FRAME || |
9052 | 0 | ps_multi_thrd_ctxt->aps_curr_inp_pre_enc[i4_ping_pong]->s_lap_out.i4_pic_type == IV_I_FRAME) |
9053 | 0 | { |
9054 | 0 | i4_num_ref_fpel = 1; |
9055 | 0 | } |
9056 | 0 | if(i4_num_ref_prev_layer <= 2) |
9057 | 0 | { |
9058 | 0 | prev_layer_implicit = 1; |
9059 | 0 | curr_layer_implicit = 1; |
9060 | 0 | i4_num_ref_each_dir = 1; |
9061 | 0 | } |
9062 | 0 | else |
9063 | 0 | { |
9064 | | /* It is assumed that we have equal number of references in each dir */ |
9065 | | //ASSERT(!(i4_num_ref_prev_layer & 1)); |
9066 | 0 | prev_layer_implicit = 0; |
9067 | 0 | i4_num_ref_each_dir = i4_num_ref_prev_layer >> 1; |
9068 | 0 | } |
9069 | 0 | s_mv_update_prms.i4_num_ref = i4_num_ref_fpel; |
9070 | 0 | s_mv_update_prms.i4_num_active_ref_l0 = ps_ctxt->s_frm_prms.u1_num_active_ref_l0; |
9071 | 0 | s_mv_update_prms.i4_num_active_ref_l1 = ps_ctxt->s_frm_prms.u1_num_active_ref_l1; |
9072 | | |
9073 | | /* this can be kept to 1 or 2 */ |
9074 | 0 | i4_num_ref_before_merge = 2; |
9075 | 0 | i4_num_ref_before_merge = MIN(i4_num_ref_before_merge, i4_num_ref_fpel); |
9076 | | |
9077 | | /* Set up place holders to hold the search nodes of each initial candt */ |
9078 | 0 | for(i = 0; i < MAX_INIT_CANDTS; i++) |
9079 | 0 | { |
9080 | 0 | ps_search_candts[i].ps_search_node = &ps_ctxt->s_init_search_node[i]; |
9081 | 0 | INIT_SEARCH_NODE(ps_search_candts[i].ps_search_node, 0); |
9082 | 0 | } |
9083 | | |
9084 | | /* redundant, but doing it here since it is used in pred ctxt init */ |
9085 | 0 | ps_candt_zeromv = ps_search_candts[0].ps_search_node; |
9086 | 0 | for(i = 0; i < 3; i++) |
9087 | 0 | { |
9088 | 0 | search_node_t *ps_search_node; |
9089 | 0 | ps_search_node = &as_left_neighbours[i]; |
9090 | 0 | INIT_SEARCH_NODE(ps_search_node, 0); |
9091 | 0 | ps_search_node = &as_top_neighbours[i]; |
9092 | 0 | INIT_SEARCH_NODE(ps_search_node, 0); |
9093 | 0 | } |
9094 | |
|
9095 | 0 | INIT_SEARCH_NODE(&as_top_neighbours[3], 0); |
9096 | | /* bottom left node always not available for the blk being searched */ |
9097 | 0 | as_left_neighbours[2].u1_is_avail = 0; |
9098 | | /*************************************************************************/ |
9099 | | /* Initialize all the search results structure here. We update all the */ |
9100 | | /* search results to default values, and configure things like blk sizes */ |
9101 | | /*************************************************************************/ |
9102 | 0 | if(ps_refine_prms->i4_encode == 0) |
9103 | 0 | { |
9104 | 0 | S32 pred_lx; |
9105 | 0 | search_results_t *ps_search_results; |
9106 | |
|
9107 | 0 | ps_search_results = &ps_ctxt->s_search_results_8x8; |
9108 | 0 | hme_init_search_results( |
9109 | 0 | ps_search_results, |
9110 | 0 | i4_num_ref_fpel, |
9111 | 0 | ps_refine_prms->i4_num_fpel_results, |
9112 | 0 | ps_refine_prms->i4_num_results_per_part, |
9113 | 0 | e_search_blk_size, |
9114 | 0 | 0, |
9115 | 0 | 0, |
9116 | 0 | &ps_ctxt->au1_is_past[0]); |
9117 | 0 | for(pred_lx = 0; pred_lx < 2; pred_lx++) |
9118 | 0 | { |
9119 | 0 | hme_init_pred_ctxt_no_encode( |
9120 | 0 | &ps_search_results->as_pred_ctxt[pred_lx], |
9121 | 0 | ps_search_results, |
9122 | 0 | &as_top_neighbours[0], |
9123 | 0 | &as_left_neighbours[0], |
9124 | 0 | &ps_candt_prj_coloc[0], |
9125 | 0 | ps_candt_zeromv, |
9126 | 0 | ps_candt_zeromv, |
9127 | 0 | pred_lx, |
9128 | 0 | lambda_inp, |
9129 | 0 | ps_refine_prms->lambda_q_shift, |
9130 | 0 | &ps_ctxt->apu1_ref_bits_tlu_lc[0], |
9131 | 0 | &ps_ctxt->ai2_ref_scf[0]); |
9132 | 0 | } |
9133 | 0 | } |
9134 | | |
9135 | | /*********************************************************************/ |
9136 | | /* Initialize the dyn. search range params. for each reference index */ |
9137 | | /* in current layer ctxt */ |
9138 | | /*********************************************************************/ |
9139 | | /* Only for P pic. For P, both are 0, I&B has them mut. exclusive */ |
9140 | 0 | if(ps_ctxt->s_frm_prms.is_i_pic == ps_ctxt->s_frm_prms.bidir_enabled) |
9141 | 0 | { |
9142 | 0 | WORD32 ref_ctr; |
9143 | |
|
9144 | 0 | for(ref_ctr = 0; ref_ctr < s_mv_update_prms.i4_num_ref; ref_ctr++) |
9145 | 0 | { |
9146 | 0 | INIT_DYN_SEARCH_PRMS( |
9147 | 0 | &ps_ctxt->s_coarse_dyn_range_prms |
9148 | 0 | .as_dyn_range_prms[ps_refine_prms->i4_layer_id][ref_ctr], |
9149 | 0 | ps_ctxt->ai4_ref_idx_to_poc_lc[ref_ctr]); |
9150 | 0 | } |
9151 | 0 | } |
9152 | | |
9153 | | /* Next set up initial candidates according to a given set of rules. */ |
9154 | | /* The number of initial candidates affects the quality of ME in the */ |
9155 | | /* case of motion with multiple degrees of freedom. In case of simple */ |
9156 | | /* translational motion, a current and a few causal and non causal */ |
9157 | | /* candts would suffice. More candidates help to cover more complex */ |
9158 | | /* cases like partitions, rotation/zoom, occlusion in/out, fine motion */ |
9159 | | /* where multiple ref helps etc. */ |
9160 | | /* The candidate choice also depends on the following parameters. */ |
9161 | | /* e_search_complexity: SRCH_CX_LOW, SRCH_CX_MED, SRCH_CX_HIGH */ |
9162 | | /* Whether we encode or not, and the type of search across reference */ |
9163 | | /* i.e. the previous layer may have been explicit/implicit and curr */ |
9164 | | /* layer may be explicit/implicit */ |
9165 | | |
9166 | | /* 0, 0, L, T, projected coloc best always presnt by default */ |
9167 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me(ZERO_MV, e_me_quality_presets); |
9168 | 0 | ps_candt_zeromv = ps_search_candts[id].ps_search_node; |
9169 | 0 | ps_search_candts[id].u1_num_steps_refine = 0; |
9170 | 0 | ps_candt_zeromv->s_mv.i2_mvx = 0; |
9171 | 0 | ps_candt_zeromv->s_mv.i2_mvy = 0; |
9172 | |
|
9173 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me(SPATIAL_LEFT0, e_me_quality_presets); |
9174 | 0 | ps_candt_l = ps_search_candts[id].ps_search_node; |
9175 | 0 | ps_search_candts[id].u1_num_steps_refine = 0; |
9176 | | |
9177 | | /* Even in ME_HIGH_SPEED mode, in layer 0, blocks */ |
9178 | | /* not at the CTB boundary use the causal T and */ |
9179 | | /* not the projected T, although the candidate is */ |
9180 | | /* still pointed to by ps_candt_prj_t[0] */ |
9181 | 0 | if(ME_MEDIUM_SPEED <= e_me_quality_presets) |
9182 | 0 | { |
9183 | | /* Using Projected top to eliminate sync */ |
9184 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9185 | 0 | PROJECTED_TOP0, e_me_quality_presets); |
9186 | 0 | ps_candt_prj_t[0] = ps_search_candts[id].ps_search_node; |
9187 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9188 | 0 | } |
9189 | 0 | else |
9190 | 0 | { |
9191 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9192 | 0 | SPATIAL_TOP0, e_me_quality_presets); |
9193 | 0 | ps_candt_t = ps_search_candts[id].ps_search_node; |
9194 | 0 | ps_search_candts[id].u1_num_steps_refine = 0; |
9195 | 0 | } |
9196 | |
|
9197 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9198 | 0 | PROJECTED_COLOC0, e_me_quality_presets); |
9199 | 0 | ps_candt_prj_coloc[0] = ps_search_candts[id].ps_search_node; |
9200 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9201 | |
|
9202 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9203 | 0 | PROJECTED_COLOC1, e_me_quality_presets); |
9204 | 0 | ps_candt_prj_coloc[1] = ps_search_candts[id].ps_search_node; |
9205 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9206 | |
|
9207 | 0 | if(ME_MEDIUM_SPEED <= e_me_quality_presets) |
9208 | 0 | { |
9209 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9210 | 0 | PROJECTED_TOP_RIGHT0, e_me_quality_presets); |
9211 | 0 | ps_candt_prj_tr[0] = ps_search_candts[id].ps_search_node; |
9212 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9213 | |
|
9214 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9215 | 0 | PROJECTED_TOP_LEFT0, e_me_quality_presets); |
9216 | 0 | ps_candt_prj_tl[0] = ps_search_candts[id].ps_search_node; |
9217 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9218 | 0 | } |
9219 | 0 | else |
9220 | 0 | { |
9221 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9222 | 0 | SPATIAL_TOP_RIGHT0, e_me_quality_presets); |
9223 | 0 | ps_candt_tr = ps_search_candts[id].ps_search_node; |
9224 | 0 | ps_search_candts[id].u1_num_steps_refine = 0; |
9225 | |
|
9226 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9227 | 0 | SPATIAL_TOP_LEFT0, e_me_quality_presets); |
9228 | 0 | ps_candt_tl = ps_search_candts[id].ps_search_node; |
9229 | 0 | ps_search_candts[id].u1_num_steps_refine = 0; |
9230 | 0 | } |
9231 | |
|
9232 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9233 | 0 | PROJECTED_RIGHT0, e_me_quality_presets); |
9234 | 0 | ps_candt_prj_r[0] = ps_search_candts[id].ps_search_node; |
9235 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9236 | |
|
9237 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9238 | 0 | PROJECTED_BOTTOM0, e_me_quality_presets); |
9239 | 0 | ps_candt_prj_b[0] = ps_search_candts[id].ps_search_node; |
9240 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9241 | |
|
9242 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9243 | 0 | PROJECTED_BOTTOM_RIGHT0, e_me_quality_presets); |
9244 | 0 | ps_candt_prj_br[0] = ps_search_candts[id].ps_search_node; |
9245 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9246 | |
|
9247 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9248 | 0 | PROJECTED_BOTTOM_LEFT0, e_me_quality_presets); |
9249 | 0 | ps_candt_prj_bl[0] = ps_search_candts[id].ps_search_node; |
9250 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9251 | |
|
9252 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9253 | 0 | PROJECTED_RIGHT1, e_me_quality_presets); |
9254 | 0 | ps_candt_prj_r[1] = ps_search_candts[id].ps_search_node; |
9255 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9256 | |
|
9257 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9258 | 0 | PROJECTED_BOTTOM1, e_me_quality_presets); |
9259 | 0 | ps_candt_prj_b[1] = ps_search_candts[id].ps_search_node; |
9260 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9261 | |
|
9262 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9263 | 0 | PROJECTED_BOTTOM_RIGHT1, e_me_quality_presets); |
9264 | 0 | ps_candt_prj_br[1] = ps_search_candts[id].ps_search_node; |
9265 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9266 | |
|
9267 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9268 | 0 | PROJECTED_BOTTOM_LEFT1, e_me_quality_presets); |
9269 | 0 | ps_candt_prj_bl[1] = ps_search_candts[id].ps_search_node; |
9270 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9271 | |
|
9272 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me(PROJECTED_TOP1, e_me_quality_presets); |
9273 | 0 | ps_candt_prj_t[1] = ps_search_candts[id].ps_search_node; |
9274 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9275 | |
|
9276 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9277 | 0 | PROJECTED_TOP_RIGHT1, e_me_quality_presets); |
9278 | 0 | ps_candt_prj_tr[1] = ps_search_candts[id].ps_search_node; |
9279 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9280 | |
|
9281 | 0 | id = hme_decide_search_candidate_priority_in_l1_and_l2_me( |
9282 | 0 | PROJECTED_TOP_LEFT1, e_me_quality_presets); |
9283 | 0 | ps_candt_prj_tl[1] = ps_search_candts[id].ps_search_node; |
9284 | 0 | ps_search_candts[id].u1_num_steps_refine = 1; |
9285 | | |
9286 | | /*************************************************************************/ |
9287 | | /* Now that the candidates have been ordered, to choose the right number */ |
9288 | | /* of initial candidates. */ |
9289 | | /*************************************************************************/ |
9290 | 0 | if(curr_layer_implicit && !prev_layer_implicit) |
9291 | 0 | { |
9292 | 0 | if(e_search_complexity == SEARCH_CX_LOW) |
9293 | 0 | num_init_candts = 7; |
9294 | 0 | else if(e_search_complexity == SEARCH_CX_MED) |
9295 | 0 | num_init_candts = 13; |
9296 | 0 | else if(e_search_complexity == SEARCH_CX_HIGH) |
9297 | 0 | num_init_candts = 18; |
9298 | 0 | else |
9299 | 0 | ASSERT(0); |
9300 | 0 | } |
9301 | 0 | else |
9302 | 0 | { |
9303 | 0 | if(e_search_complexity == SEARCH_CX_LOW) |
9304 | 0 | num_init_candts = 5; |
9305 | 0 | else if(e_search_complexity == SEARCH_CX_MED) |
9306 | 0 | num_init_candts = 11; |
9307 | 0 | else if(e_search_complexity == SEARCH_CX_HIGH) |
9308 | 0 | num_init_candts = 16; |
9309 | 0 | else |
9310 | 0 | ASSERT(0); |
9311 | 0 | } |
9312 | | |
9313 | 0 | if(ME_XTREME_SPEED_25 == e_me_quality_presets) |
9314 | 0 | { |
9315 | 0 | num_init_candts = NUM_INIT_SEARCH_CANDS_IN_L1_AND_L2_ME_IN_XS25; |
9316 | 0 | } |
9317 | | |
9318 | | /*************************************************************************/ |
9319 | | /* The following search parameters are fixed throughout the search across*/ |
9320 | | /* all blks. So these are configured outside processing loop */ |
9321 | | /*************************************************************************/ |
9322 | 0 | s_search_prms_blk.i4_num_init_candts = num_init_candts; |
9323 | 0 | s_search_prms_blk.i4_start_step = 1; |
9324 | 0 | s_search_prms_blk.i4_use_satd = 0; |
9325 | 0 | s_search_prms_blk.i4_num_steps_post_refine = ps_refine_prms->i4_num_steps_post_refine_fpel; |
9326 | | /* we use recon only for encoded layers, otherwise it is not available */ |
9327 | 0 | s_search_prms_blk.i4_use_rec = ps_refine_prms->i4_encode & ps_refine_prms->i4_use_rec_in_fpel; |
9328 | |
|
9329 | 0 | s_search_prms_blk.ps_search_candts = ps_search_candts; |
9330 | | /* We use the same mv_range for all ref. pic. So assign to member 0 */ |
9331 | 0 | if(s_search_prms_blk.i4_use_rec) |
9332 | 0 | s_search_prms_blk.aps_mv_range[0] = &s_range_prms_rec; |
9333 | 0 | else |
9334 | 0 | s_search_prms_blk.aps_mv_range[0] = &s_range_prms_inp; |
9335 | | /*************************************************************************/ |
9336 | | /* Initialize coordinates. Meaning as follows */ |
9337 | | /* blk_x : x coordinate of the 16x16 blk, in terms of number of blks */ |
9338 | | /* blk_y : same as above, y coord. */ |
9339 | | /* num_blks_in_this_ctb : number of blks in this given ctb that starts */ |
9340 | | /* at i4_ctb_x, i4_ctb_y. This may not be 16 at picture boundaries. */ |
9341 | | /* i4_ctb_x, i4_ctb_y: pixel coordinate of the ctb realtive to top left */ |
9342 | | /* corner of the picture. Always multiple of 64. */ |
9343 | | /* blk_id_in_ctb : encode order id of the blk in the ctb. */ |
9344 | | /*************************************************************************/ |
9345 | 0 | blk_y = 0; |
9346 | 0 | blk_id_in_ctb = 0; |
9347 | |
|
9348 | 0 | GET_NUM_BLKS_IN_PIC(i4_pic_wd, i4_pic_ht, blk_size_shift, num_blks_in_row, num_blks_in_pic); |
9349 | | |
9350 | | /* Get the number of sync units in a row based on encode/non enocde layer */ |
9351 | 0 | num_sync_units_in_row = num_blks_in_row; |
9352 | | |
9353 | | /*************************************************************************/ |
9354 | | /* Picture limit on all 4 sides. This will be used to set mv limits for */ |
9355 | | /* every block given its coordinate. Note thsi assumes that the min amt */ |
9356 | | /* of padding to right of pic is equal to the blk size. If we go all the */ |
9357 | | /* way upto 64x64, then the min padding on right size of picture should */ |
9358 | | /* be 64, and also on bottom side of picture. */ |
9359 | | /*************************************************************************/ |
9360 | 0 | SET_PIC_LIMIT( |
9361 | 0 | s_pic_limit_inp, |
9362 | 0 | ps_curr_layer->i4_pad_x_inp, |
9363 | 0 | ps_curr_layer->i4_pad_y_inp, |
9364 | 0 | ps_curr_layer->i4_wd, |
9365 | 0 | ps_curr_layer->i4_ht, |
9366 | 0 | s_search_prms_blk.i4_num_steps_post_refine); |
9367 | |
|
9368 | 0 | SET_PIC_LIMIT( |
9369 | 0 | s_pic_limit_rec, |
9370 | 0 | ps_curr_layer->i4_pad_x_rec, |
9371 | 0 | ps_curr_layer->i4_pad_y_rec, |
9372 | 0 | ps_curr_layer->i4_wd, |
9373 | 0 | ps_curr_layer->i4_ht, |
9374 | 0 | s_search_prms_blk.i4_num_steps_post_refine); |
9375 | | |
9376 | | /*************************************************************************/ |
9377 | | /* set the MV limit per ref. pic. */ |
9378 | | /* - P pic. : Based on the config params. */ |
9379 | | /* - B/b pic: Based on the Max/Min MV from prev. P and config. param. */ |
9380 | | /*************************************************************************/ |
9381 | 0 | { |
9382 | 0 | WORD32 ref_ctr; |
9383 | | /* Only for B/b pic. */ |
9384 | 0 | if(1 == ps_ctxt->s_frm_prms.bidir_enabled) |
9385 | 0 | { |
9386 | 0 | WORD16 i2_mv_y_per_poc, i2_max_mv_y; |
9387 | 0 | WORD32 cur_poc, ref_poc, abs_poc_diff; |
9388 | |
|
9389 | 0 | cur_poc = ps_ctxt->i4_curr_poc; |
9390 | | |
9391 | | /* Get abs MAX for symmetric search */ |
9392 | 0 | i2_mv_y_per_poc = MAX( |
9393 | 0 | ps_ctxt->s_coarse_dyn_range_prms.i2_dyn_max_y_per_poc[ps_refine_prms->i4_layer_id], |
9394 | 0 | (ABS(ps_ctxt->s_coarse_dyn_range_prms |
9395 | 0 | .i2_dyn_min_y_per_poc[ps_refine_prms->i4_layer_id]))); |
9396 | |
|
9397 | 0 | for(ref_ctr = 0; ref_ctr < i4_num_ref_fpel; ref_ctr++) |
9398 | 0 | { |
9399 | 0 | ref_poc = ps_ctxt->ai4_ref_idx_to_poc_lc[ref_ctr]; |
9400 | 0 | abs_poc_diff = ABS((cur_poc - ref_poc)); |
9401 | | /* Get the cur. max MV based on POC distance */ |
9402 | 0 | i2_max_mv_y = i2_mv_y_per_poc * abs_poc_diff; |
9403 | 0 | i2_max_mv_y = MIN(i2_max_mv_y, ps_curr_layer->i2_max_mv_y); |
9404 | |
|
9405 | 0 | as_mv_limit[ref_ctr].i2_min_x = -ps_curr_layer->i2_max_mv_x; |
9406 | 0 | as_mv_limit[ref_ctr].i2_min_y = -i2_max_mv_y; |
9407 | 0 | as_mv_limit[ref_ctr].i2_max_x = ps_curr_layer->i2_max_mv_x; |
9408 | 0 | as_mv_limit[ref_ctr].i2_max_y = i2_max_mv_y; |
9409 | 0 | } |
9410 | 0 | } |
9411 | 0 | else |
9412 | 0 | { |
9413 | | /* Set the Config. File Params for P pic. */ |
9414 | 0 | for(ref_ctr = 0; ref_ctr < i4_num_ref_fpel; ref_ctr++) |
9415 | 0 | { |
9416 | 0 | as_mv_limit[ref_ctr].i2_min_x = -ps_curr_layer->i2_max_mv_x; |
9417 | 0 | as_mv_limit[ref_ctr].i2_min_y = -ps_curr_layer->i2_max_mv_y; |
9418 | 0 | as_mv_limit[ref_ctr].i2_max_x = ps_curr_layer->i2_max_mv_x; |
9419 | 0 | as_mv_limit[ref_ctr].i2_max_y = ps_curr_layer->i2_max_mv_y; |
9420 | 0 | } |
9421 | 0 | } |
9422 | 0 | } |
9423 | | |
9424 | | /* EIID: Calculate threshold based on quality preset and/or temporal layers */ |
9425 | 0 | if(e_me_quality_presets == ME_MEDIUM_SPEED) |
9426 | 0 | { |
9427 | 0 | i4_threshold_multiplier = 1; |
9428 | 0 | i4_threshold_divider = 4; |
9429 | 0 | } |
9430 | 0 | else if(e_me_quality_presets == ME_HIGH_SPEED) |
9431 | 0 | { |
9432 | 0 | i4_threshold_multiplier = 1; |
9433 | 0 | i4_threshold_divider = 2; |
9434 | 0 | } |
9435 | 0 | else if((e_me_quality_presets == ME_XTREME_SPEED) || (e_me_quality_presets == ME_XTREME_SPEED_25)) |
9436 | 0 | { |
9437 | | #if OLD_XTREME_SPEED |
9438 | | /* Hard coding the temporal ID value to 1, if it is older xtreme speed */ |
9439 | | i4_temporal_layer = 1; |
9440 | | #endif |
9441 | 0 | if(i4_temporal_layer == 0) |
9442 | 0 | { |
9443 | 0 | i4_threshold_multiplier = 3; |
9444 | 0 | i4_threshold_divider = 4; |
9445 | 0 | } |
9446 | 0 | else if(i4_temporal_layer == 1) |
9447 | 0 | { |
9448 | 0 | i4_threshold_multiplier = 3; |
9449 | 0 | i4_threshold_divider = 4; |
9450 | 0 | } |
9451 | 0 | else if(i4_temporal_layer == 2) |
9452 | 0 | { |
9453 | 0 | i4_threshold_multiplier = 1; |
9454 | 0 | i4_threshold_divider = 1; |
9455 | 0 | } |
9456 | 0 | else |
9457 | 0 | { |
9458 | 0 | i4_threshold_multiplier = 5; |
9459 | 0 | i4_threshold_divider = 4; |
9460 | 0 | } |
9461 | 0 | } |
9462 | 0 | else if(e_me_quality_presets == ME_HIGH_QUALITY) |
9463 | 0 | { |
9464 | 0 | i4_threshold_multiplier = 1; |
9465 | 0 | i4_threshold_divider = 1; |
9466 | 0 | } |
9467 | | |
9468 | | /*************************************************************************/ |
9469 | | /*************************************************************************/ |
9470 | | /*************************************************************************/ |
9471 | | /* START OF THE CORE LOOP */ |
9472 | | /* If Encode is 0, then we just loop over each blk */ |
9473 | | /*************************************************************************/ |
9474 | | /*************************************************************************/ |
9475 | | /*************************************************************************/ |
9476 | 0 | while(0 == end_of_frame) |
9477 | 0 | { |
9478 | 0 | job_queue_t *ps_job; |
9479 | 0 | ihevce_ed_blk_t *ps_ed_blk_ctxt_curr_row; //EIID |
9480 | 0 | WORD32 i4_ctb_row_ctr; //counter to calculate CTB row counter. It's (row_ctr /4) |
9481 | 0 | WORD32 i4_num_ctbs_in_row = (num_blks_in_row + 3) / 4; //calculations verified for L1 only |
9482 | | //+3 to get ceil values when divided by 4 |
9483 | 0 | WORD32 i4_num_4x4_blocks_in_ctb_at_l1 = |
9484 | 0 | 8 * 8; //considering CTB size 32x32 at L1. hardcoded for now |
9485 | | //if there is variable for ctb size use that and this variable can be derived |
9486 | 0 | WORD32 offset_val, check_dep_pos, set_dep_pos; |
9487 | 0 | void *pv_hme_dep_mngr; |
9488 | 0 | ihevce_ed_ctb_l1_t *ps_ed_ctb_l1_row; |
9489 | | |
9490 | | /* Get the current layer HME Dep Mngr */ |
9491 | | /* Note : Use layer_id - 1 in HME layers */ |
9492 | |
|
9493 | 0 | pv_hme_dep_mngr = ppv_dep_mngr_hme_sync[ps_refine_prms->i4_layer_id - 1]; |
9494 | | |
9495 | | /* Get the current row from the job queue */ |
9496 | 0 | ps_job = (job_queue_t *)ihevce_pre_enc_grp_get_next_job( |
9497 | 0 | ps_multi_thrd_ctxt, lyr_job_type, 1, i4_ping_pong); |
9498 | | |
9499 | | /* If all rows are done, set the end of process flag to 1, */ |
9500 | | /* and the current row to -1 */ |
9501 | 0 | if(NULL == ps_job) |
9502 | 0 | { |
9503 | 0 | blk_y = -1; |
9504 | 0 | end_of_frame = 1; |
9505 | |
|
9506 | 0 | continue; |
9507 | 0 | } |
9508 | | |
9509 | 0 | if(1 == ps_ctxt->s_frm_prms.is_i_pic) |
9510 | 0 | { |
9511 | | /* set the output dependency of current row */ |
9512 | 0 | ihevce_pre_enc_grp_job_set_out_dep(ps_multi_thrd_ctxt, ps_job, i4_ping_pong); |
9513 | 0 | continue; |
9514 | 0 | } |
9515 | | |
9516 | 0 | blk_y = ps_job->s_job_info.s_me_job_info.i4_vert_unit_row_no; |
9517 | 0 | blk_x = 0; |
9518 | 0 | i4_ctb_x = 0; |
9519 | | |
9520 | | /* wait for Corresponding Pre intra Job to be completed */ |
9521 | 0 | if(1 == ps_refine_prms->i4_layer_id) |
9522 | 0 | { |
9523 | 0 | volatile UWORD32 i4_l1_done; |
9524 | 0 | volatile UWORD32 *pi4_l1_done; |
9525 | 0 | pi4_l1_done = (volatile UWORD32 *)&ps_multi_thrd_ctxt |
9526 | 0 | ->aai4_l1_pre_intra_done[i4_ping_pong][blk_y >> 2]; |
9527 | 0 | i4_l1_done = *pi4_l1_done; |
9528 | 0 | while(!i4_l1_done) |
9529 | 0 | { |
9530 | 0 | i4_l1_done = *pi4_l1_done; |
9531 | 0 | } |
9532 | 0 | } |
9533 | | /* Set Variables for Dep. Checking and Setting */ |
9534 | 0 | set_dep_pos = blk_y + 1; |
9535 | 0 | if(blk_y > 0) |
9536 | 0 | { |
9537 | 0 | offset_val = 2; |
9538 | 0 | check_dep_pos = blk_y - 1; |
9539 | 0 | } |
9540 | 0 | else |
9541 | 0 | { |
9542 | | /* First row should run without waiting */ |
9543 | 0 | offset_val = -1; |
9544 | 0 | check_dep_pos = 0; |
9545 | 0 | } |
9546 | | |
9547 | | /* EIID: calculate ed_blk_ctxt pointer for current row */ |
9548 | | /* valid for only layer-1. not varified and used for other layers */ |
9549 | 0 | i4_ctb_row_ctr = blk_y / 4; |
9550 | 0 | ps_ed_blk_ctxt_curr_row = |
9551 | 0 | ps_ctxt->ps_ed_blk + (i4_ctb_row_ctr * i4_num_ctbs_in_row * |
9552 | 0 | i4_num_4x4_blocks_in_ctb_at_l1); //valid for L1 only |
9553 | 0 | ps_ed_ctb_l1_row = ps_ctxt->ps_ed_ctb_l1 + (i4_ctb_row_ctr * i4_num_ctbs_in_row); |
9554 | | |
9555 | | /* if non-encode layer then i4_ctb_x will be same as blk_x */ |
9556 | | /* loop over all the units is a row */ |
9557 | 0 | for(; i4_ctb_x < num_sync_units_in_row; i4_ctb_x++) |
9558 | 0 | { |
9559 | 0 | ihevce_ed_blk_t *ps_ed_blk_ctxt_curr_ctb; //EIDD |
9560 | 0 | ihevce_ed_ctb_l1_t *ps_ed_ctb_l1_curr; |
9561 | 0 | WORD32 i4_ctb_blk_ctr = i4_ctb_x / 4; |
9562 | | |
9563 | | /* Wait till top row block is processed */ |
9564 | | /* Currently checking till top right block*/ |
9565 | | |
9566 | | /* Disabled since all candidates, except for */ |
9567 | | /* L and C, are projected from the coarser layer, */ |
9568 | | /* only in ME_HIGH_SPEED mode */ |
9569 | 0 | if((ME_MEDIUM_SPEED > e_me_quality_presets)) |
9570 | 0 | { |
9571 | 0 | if(i4_ctb_x < (num_sync_units_in_row - 1)) |
9572 | 0 | { |
9573 | 0 | ihevce_dmgr_chk_row_row_sync( |
9574 | 0 | pv_hme_dep_mngr, |
9575 | 0 | i4_ctb_x, |
9576 | 0 | offset_val, |
9577 | 0 | check_dep_pos, |
9578 | 0 | 0, /* Col Tile No. : Not supported in PreEnc*/ |
9579 | 0 | ps_ctxt->thrd_id); |
9580 | 0 | } |
9581 | 0 | } |
9582 | |
|
9583 | 0 | { |
9584 | | /* for non encoder layer only one block is processed */ |
9585 | 0 | num_blks_in_this_ctb = 1; |
9586 | 0 | } |
9587 | | |
9588 | | /* EIID: derive ed_ctxt ptr for current CTB */ |
9589 | 0 | ps_ed_blk_ctxt_curr_ctb = |
9590 | 0 | ps_ed_blk_ctxt_curr_row + |
9591 | 0 | (i4_ctb_blk_ctr * |
9592 | 0 | i4_num_4x4_blocks_in_ctb_at_l1); //currently valid for l1 layer only |
9593 | 0 | ps_ed_ctb_l1_curr = ps_ed_ctb_l1_row + i4_ctb_blk_ctr; |
9594 | | |
9595 | | /* loop over all the blocks in CTB will always be 1 */ |
9596 | 0 | for(blk_id_in_ctb = 0; blk_id_in_ctb < num_blks_in_this_ctb; blk_id_in_ctb++) |
9597 | 0 | { |
9598 | 0 | { |
9599 | | /* non encode layer */ |
9600 | 0 | blk_x = i4_ctb_x; |
9601 | 0 | blk_id_in_full_ctb = 0; |
9602 | 0 | s_search_prms_blk.i4_cu_x_off = s_search_prms_blk.i4_cu_y_off = 0; |
9603 | 0 | } |
9604 | | |
9605 | | /* get the current input blk point */ |
9606 | 0 | pos_x = blk_x << blk_size_shift; |
9607 | 0 | pos_y = blk_y << blk_size_shift; |
9608 | 0 | pu1_inp = ps_curr_layer->pu1_inp + pos_x + (pos_y * i4_inp_stride); |
9609 | | |
9610 | | /*********************************************************************/ |
9611 | | /* replicate the inp buffer at blk or ctb level for each ref id, */ |
9612 | | /* Instead of searching with wk * ref(k), we search with Ik = I / wk */ |
9613 | | /* thereby avoiding a bloat up of memory. If we did all references */ |
9614 | | /* weighted pred, we will end up with a duplicate copy of each ref */ |
9615 | | /* at each layer, since we need to preserve the original reference. */ |
9616 | | /* ToDo: Need to observe performance with this mechanism and compare */ |
9617 | | /* with case where ref is weighted. */ |
9618 | | /*********************************************************************/ |
9619 | 0 | if(blk_id_in_ctb == 0) |
9620 | 0 | { |
9621 | 0 | fp_get_wt_inp( |
9622 | 0 | ps_curr_layer, |
9623 | 0 | &ps_ctxt->s_wt_pred, |
9624 | 0 | unit_size, |
9625 | 0 | pos_x, |
9626 | 0 | pos_y, |
9627 | 0 | unit_size, |
9628 | 0 | ps_ctxt->num_ref_future + ps_ctxt->num_ref_past, |
9629 | 0 | ps_ctxt->i4_wt_pred_enable_flag); |
9630 | 0 | } |
9631 | |
|
9632 | 0 | s_search_prms_blk.i4_x_off = blk_x << blk_size_shift; |
9633 | 0 | s_search_prms_blk.i4_y_off = blk_y << blk_size_shift; |
9634 | | /* Select search results from a suitable search result in the context */ |
9635 | 0 | { |
9636 | 0 | ps_search_results = &ps_ctxt->s_search_results_8x8; |
9637 | 0 | } |
9638 | |
|
9639 | 0 | s_search_prms_blk.ps_search_results = ps_search_results; |
9640 | | |
9641 | | /* RESET ALL SEARCH RESULTS FOR THE NEW BLK */ |
9642 | 0 | hme_reset_search_results( |
9643 | 0 | ps_search_results, s_search_prms_blk.i4_part_mask, MV_RES_FPEL); |
9644 | | |
9645 | | /* Loop across different Ref IDx */ |
9646 | 0 | for(i1_ref_idx = 0; i1_ref_idx < i4_num_ref_fpel; i1_ref_idx++) |
9647 | 0 | { |
9648 | 0 | S32 next_blk_offset = (e_search_blk_size == BLK_16x16) ? 22 : 12; |
9649 | 0 | S32 prev_blk_offset = 6; |
9650 | 0 | S32 resultid; |
9651 | | |
9652 | | /*********************************************************************/ |
9653 | | /* For every blk in the picture, the search range needs to be derived*/ |
9654 | | /* Any blk can have any mv, but practical search constraints are */ |
9655 | | /* imposed by the picture boundary and amt of padding. */ |
9656 | | /*********************************************************************/ |
9657 | | /* MV limit is different based on ref. PIC */ |
9658 | 0 | hme_derive_search_range( |
9659 | 0 | &s_range_prms_inp, |
9660 | 0 | &s_pic_limit_inp, |
9661 | 0 | &as_mv_limit[i1_ref_idx], |
9662 | 0 | pos_x, |
9663 | 0 | pos_y, |
9664 | 0 | blk_wd, |
9665 | 0 | blk_ht); |
9666 | 0 | hme_derive_search_range( |
9667 | 0 | &s_range_prms_rec, |
9668 | 0 | &s_pic_limit_rec, |
9669 | 0 | &as_mv_limit[i1_ref_idx], |
9670 | 0 | pos_x, |
9671 | 0 | pos_y, |
9672 | 0 | blk_wd, |
9673 | 0 | blk_ht); |
9674 | |
|
9675 | 0 | s_search_prms_blk.i1_ref_idx = i1_ref_idx; |
9676 | 0 | ps_candt_zeromv->i1_ref_idx = i1_ref_idx; |
9677 | |
|
9678 | 0 | i4_num_srch_cands = 1; |
9679 | |
|
9680 | 0 | if(1 != ps_refine_prms->i4_layer_id) |
9681 | 0 | { |
9682 | 0 | S32 x, y; |
9683 | 0 | x = gau1_encode_to_raster_x[blk_id_in_full_ctb]; |
9684 | 0 | y = gau1_encode_to_raster_y[blk_id_in_full_ctb]; |
9685 | |
|
9686 | 0 | if(ME_MEDIUM_SPEED > e_me_quality_presets) |
9687 | 0 | { |
9688 | 0 | hme_get_spatial_candt( |
9689 | 0 | ps_curr_layer, |
9690 | 0 | e_search_blk_size, |
9691 | 0 | blk_x, |
9692 | 0 | blk_y, |
9693 | 0 | i1_ref_idx, |
9694 | 0 | &as_top_neighbours[0], |
9695 | 0 | &as_left_neighbours[0], |
9696 | 0 | 0, |
9697 | 0 | ((ps_refine_prms->i4_encode) ? gau1_cu_tr_valid[y][x] : 1), |
9698 | 0 | 0, |
9699 | 0 | ps_refine_prms->i4_encode); |
9700 | |
|
9701 | 0 | *ps_candt_tr = as_top_neighbours[3]; |
9702 | 0 | *ps_candt_t = as_top_neighbours[1]; |
9703 | 0 | *ps_candt_tl = as_top_neighbours[0]; |
9704 | 0 | i4_num_srch_cands += 3; |
9705 | 0 | } |
9706 | 0 | else |
9707 | 0 | { |
9708 | 0 | layer_mv_t *ps_layer_mvbank = ps_curr_layer->ps_layer_mvbank; |
9709 | 0 | S32 i4_blk_size1 = gau1_blk_size_to_wd[ps_layer_mvbank->e_blk_size]; |
9710 | 0 | S32 i4_blk_size2 = gau1_blk_size_to_wd[e_search_blk_size]; |
9711 | 0 | search_node_t *ps_search_node; |
9712 | 0 | S32 i4_offset, blk_x_temp = blk_x, blk_y_temp = blk_y; |
9713 | 0 | hme_mv_t *ps_mv, *ps_mv_base; |
9714 | 0 | S08 *pi1_ref_idx, *pi1_ref_idx_base; |
9715 | 0 | S32 jump = 1, mvs_in_blk, mvs_in_row; |
9716 | 0 | S32 shift = (ps_refine_prms->i4_encode ? 2 : 0); |
9717 | |
|
9718 | 0 | if(i4_blk_size1 != i4_blk_size2) |
9719 | 0 | { |
9720 | 0 | blk_x_temp <<= 1; |
9721 | 0 | blk_y_temp <<= 1; |
9722 | 0 | jump = 2; |
9723 | 0 | if((i4_blk_size1 << 2) == i4_blk_size2) |
9724 | 0 | { |
9725 | 0 | blk_x_temp <<= 1; |
9726 | 0 | blk_y_temp <<= 1; |
9727 | 0 | jump = 4; |
9728 | 0 | } |
9729 | 0 | } |
9730 | |
|
9731 | 0 | mvs_in_blk = ps_layer_mvbank->i4_num_mvs_per_blk; |
9732 | 0 | mvs_in_row = ps_layer_mvbank->i4_num_mvs_per_row; |
9733 | | |
9734 | | /* Adjust teh blk coord to point to top left locn */ |
9735 | 0 | blk_x_temp -= 1; |
9736 | 0 | blk_y_temp -= 1; |
9737 | | |
9738 | | /* Pick up the mvs from the location */ |
9739 | 0 | i4_offset = (blk_x_temp * ps_layer_mvbank->i4_num_mvs_per_blk); |
9740 | 0 | i4_offset += (ps_layer_mvbank->i4_num_mvs_per_row * blk_y_temp); |
9741 | |
|
9742 | 0 | ps_mv = ps_layer_mvbank->ps_mv + i4_offset; |
9743 | 0 | pi1_ref_idx = ps_layer_mvbank->pi1_ref_idx + i4_offset; |
9744 | |
|
9745 | 0 | ps_mv += (i1_ref_idx * ps_layer_mvbank->i4_num_mvs_per_ref); |
9746 | 0 | pi1_ref_idx += (i1_ref_idx * ps_layer_mvbank->i4_num_mvs_per_ref); |
9747 | |
|
9748 | 0 | ps_mv_base = ps_mv; |
9749 | 0 | pi1_ref_idx_base = pi1_ref_idx; |
9750 | |
|
9751 | 0 | ps_search_node = &as_left_neighbours[0]; |
9752 | 0 | ps_mv = ps_mv_base + mvs_in_row; |
9753 | 0 | pi1_ref_idx = pi1_ref_idx_base + mvs_in_row; |
9754 | 0 | COPY_MV_TO_SEARCH_NODE( |
9755 | 0 | ps_search_node, ps_mv, pi1_ref_idx, i1_ref_idx, shift); |
9756 | |
|
9757 | 0 | i4_num_srch_cands++; |
9758 | 0 | } |
9759 | 0 | } |
9760 | 0 | else |
9761 | 0 | { |
9762 | 0 | S32 x, y; |
9763 | 0 | x = gau1_encode_to_raster_x[blk_id_in_full_ctb]; |
9764 | 0 | y = gau1_encode_to_raster_y[blk_id_in_full_ctb]; |
9765 | |
|
9766 | 0 | if(ME_MEDIUM_SPEED > e_me_quality_presets) |
9767 | 0 | { |
9768 | 0 | hme_get_spatial_candt_in_l1_me( |
9769 | 0 | ps_curr_layer, |
9770 | 0 | e_search_blk_size, |
9771 | 0 | blk_x, |
9772 | 0 | blk_y, |
9773 | 0 | i1_ref_idx, |
9774 | 0 | !ps_search_results->pu1_is_past[i1_ref_idx], |
9775 | 0 | &as_top_neighbours[0], |
9776 | 0 | &as_left_neighbours[0], |
9777 | 0 | 0, |
9778 | 0 | ((ps_refine_prms->i4_encode) ? gau1_cu_tr_valid[y][x] : 1), |
9779 | 0 | 0, |
9780 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l0, |
9781 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l1); |
9782 | |
|
9783 | 0 | *ps_candt_tr = as_top_neighbours[3]; |
9784 | 0 | *ps_candt_t = as_top_neighbours[1]; |
9785 | 0 | *ps_candt_tl = as_top_neighbours[0]; |
9786 | |
|
9787 | 0 | i4_num_srch_cands += 3; |
9788 | 0 | } |
9789 | 0 | else |
9790 | 0 | { |
9791 | 0 | layer_mv_t *ps_layer_mvbank = ps_curr_layer->ps_layer_mvbank; |
9792 | 0 | S32 i4_blk_size1 = gau1_blk_size_to_wd[ps_layer_mvbank->e_blk_size]; |
9793 | 0 | S32 i4_blk_size2 = gau1_blk_size_to_wd[e_search_blk_size]; |
9794 | 0 | S32 i4_mv_pos_in_implicit_array; |
9795 | 0 | search_node_t *ps_search_node; |
9796 | 0 | S32 i4_offset, blk_x_temp = blk_x, blk_y_temp = blk_y; |
9797 | 0 | hme_mv_t *ps_mv, *ps_mv_base; |
9798 | 0 | S08 *pi1_ref_idx, *pi1_ref_idx_base; |
9799 | 0 | S32 jump = 1, mvs_in_blk, mvs_in_row; |
9800 | 0 | S32 shift = (ps_refine_prms->i4_encode ? 2 : 0); |
9801 | 0 | U08 u1_pred_dir = !ps_search_results->pu1_is_past[i1_ref_idx]; |
9802 | 0 | S32 i4_num_results_in_given_dir = |
9803 | 0 | ((u1_pred_dir == 1) ? (ps_layer_mvbank->i4_num_mvs_per_ref * |
9804 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l1) |
9805 | 0 | : (ps_layer_mvbank->i4_num_mvs_per_ref * |
9806 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l0)); |
9807 | |
|
9808 | 0 | if(i4_blk_size1 != i4_blk_size2) |
9809 | 0 | { |
9810 | 0 | blk_x_temp <<= 1; |
9811 | 0 | blk_y_temp <<= 1; |
9812 | 0 | jump = 2; |
9813 | 0 | if((i4_blk_size1 << 2) == i4_blk_size2) |
9814 | 0 | { |
9815 | 0 | blk_x_temp <<= 1; |
9816 | 0 | blk_y_temp <<= 1; |
9817 | 0 | jump = 4; |
9818 | 0 | } |
9819 | 0 | } |
9820 | |
|
9821 | 0 | mvs_in_blk = ps_layer_mvbank->i4_num_mvs_per_blk; |
9822 | 0 | mvs_in_row = ps_layer_mvbank->i4_num_mvs_per_row; |
9823 | | |
9824 | | /* Adjust teh blk coord to point to top left locn */ |
9825 | 0 | blk_x_temp -= 1; |
9826 | 0 | blk_y_temp -= 1; |
9827 | | |
9828 | | /* Pick up the mvs from the location */ |
9829 | 0 | i4_offset = (blk_x_temp * ps_layer_mvbank->i4_num_mvs_per_blk); |
9830 | 0 | i4_offset += (ps_layer_mvbank->i4_num_mvs_per_row * blk_y_temp); |
9831 | |
|
9832 | 0 | i4_offset += |
9833 | 0 | ((u1_pred_dir == 1) ? (ps_layer_mvbank->i4_num_mvs_per_ref * |
9834 | 0 | ps_ctxt->s_frm_prms.u1_num_active_ref_l0) |
9835 | 0 | : 0); |
9836 | |
|
9837 | 0 | ps_mv = ps_layer_mvbank->ps_mv + i4_offset; |
9838 | 0 | pi1_ref_idx = ps_layer_mvbank->pi1_ref_idx + i4_offset; |
9839 | |
|
9840 | 0 | ps_mv_base = ps_mv; |
9841 | 0 | pi1_ref_idx_base = pi1_ref_idx; |
9842 | |
|
9843 | 0 | { |
9844 | | /* ps_mv and pi1_ref_idx now point to the top left locn */ |
9845 | 0 | ps_search_node = &as_left_neighbours[0]; |
9846 | 0 | ps_mv = ps_mv_base + mvs_in_row; |
9847 | 0 | pi1_ref_idx = pi1_ref_idx_base + mvs_in_row; |
9848 | |
|
9849 | 0 | i4_mv_pos_in_implicit_array = |
9850 | 0 | hme_find_pos_of_implicitly_stored_ref_id( |
9851 | 0 | pi1_ref_idx, i1_ref_idx, 0, i4_num_results_in_given_dir); |
9852 | |
|
9853 | 0 | if(-1 != i4_mv_pos_in_implicit_array) |
9854 | 0 | { |
9855 | 0 | COPY_MV_TO_SEARCH_NODE( |
9856 | 0 | ps_search_node, |
9857 | 0 | &ps_mv[i4_mv_pos_in_implicit_array], |
9858 | 0 | &pi1_ref_idx[i4_mv_pos_in_implicit_array], |
9859 | 0 | i1_ref_idx, |
9860 | 0 | shift); |
9861 | 0 | } |
9862 | 0 | else |
9863 | 0 | { |
9864 | 0 | ps_search_node->u1_is_avail = 0; |
9865 | 0 | ps_search_node->s_mv.i2_mvx = 0; |
9866 | 0 | ps_search_node->s_mv.i2_mvy = 0; |
9867 | 0 | ps_search_node->i1_ref_idx = i1_ref_idx; |
9868 | 0 | } |
9869 | |
|
9870 | 0 | i4_num_srch_cands++; |
9871 | 0 | } |
9872 | 0 | } |
9873 | 0 | } |
9874 | |
|
9875 | 0 | *ps_candt_l = as_left_neighbours[0]; |
9876 | | |
9877 | | /* when 16x16 is searched in an encode layer, and the prev layer */ |
9878 | | /* stores results for 4x4 blks, we project 5 candts corresponding */ |
9879 | | /* to (2,2), (2,14), (14,2), 14,14) and 2nd best of (2,2) */ |
9880 | | /* However in other cases, only 2,2 best and 2nd best reqd */ |
9881 | 0 | resultid = 0; |
9882 | 0 | pf_hme_project_coloc_candt( |
9883 | 0 | ps_candt_prj_coloc[0], |
9884 | 0 | ps_curr_layer, |
9885 | 0 | ps_coarse_layer, |
9886 | 0 | pos_x + 2, |
9887 | 0 | pos_y + 2, |
9888 | 0 | i1_ref_idx, |
9889 | 0 | resultid); |
9890 | |
|
9891 | 0 | i4_num_srch_cands++; |
9892 | |
|
9893 | 0 | resultid = 1; |
9894 | 0 | if(num_results_prev_layer > 1) |
9895 | 0 | { |
9896 | 0 | pf_hme_project_coloc_candt( |
9897 | 0 | ps_candt_prj_coloc[1], |
9898 | 0 | ps_curr_layer, |
9899 | 0 | ps_coarse_layer, |
9900 | 0 | pos_x + 2, |
9901 | 0 | pos_y + 2, |
9902 | 0 | i1_ref_idx, |
9903 | 0 | resultid); |
9904 | |
|
9905 | 0 | i4_num_srch_cands++; |
9906 | 0 | } |
9907 | |
|
9908 | 0 | resultid = 0; |
9909 | |
|
9910 | 0 | if(ME_MEDIUM_SPEED <= e_me_quality_presets) |
9911 | 0 | { |
9912 | 0 | pf_hme_project_coloc_candt( |
9913 | 0 | ps_candt_prj_t[0], |
9914 | 0 | ps_curr_layer, |
9915 | 0 | ps_coarse_layer, |
9916 | 0 | pos_x, |
9917 | 0 | pos_y - prev_blk_offset, |
9918 | 0 | i1_ref_idx, |
9919 | 0 | resultid); |
9920 | |
|
9921 | 0 | i4_num_srch_cands++; |
9922 | 0 | } |
9923 | |
|
9924 | 0 | { |
9925 | 0 | pf_hme_project_coloc_candt( |
9926 | 0 | ps_candt_prj_br[0], |
9927 | 0 | ps_curr_layer, |
9928 | 0 | ps_coarse_layer, |
9929 | 0 | pos_x + next_blk_offset, |
9930 | 0 | pos_y + next_blk_offset, |
9931 | 0 | i1_ref_idx, |
9932 | 0 | resultid); |
9933 | 0 | pf_hme_project_coloc_candt( |
9934 | 0 | ps_candt_prj_bl[0], |
9935 | 0 | ps_curr_layer, |
9936 | 0 | ps_coarse_layer, |
9937 | 0 | pos_x - prev_blk_offset, |
9938 | 0 | pos_y + next_blk_offset, |
9939 | 0 | i1_ref_idx, |
9940 | 0 | resultid); |
9941 | 0 | pf_hme_project_coloc_candt( |
9942 | 0 | ps_candt_prj_r[0], |
9943 | 0 | ps_curr_layer, |
9944 | 0 | ps_coarse_layer, |
9945 | 0 | pos_x + next_blk_offset, |
9946 | 0 | pos_y, |
9947 | 0 | i1_ref_idx, |
9948 | 0 | resultid); |
9949 | 0 | pf_hme_project_coloc_candt( |
9950 | 0 | ps_candt_prj_b[0], |
9951 | 0 | ps_curr_layer, |
9952 | 0 | ps_coarse_layer, |
9953 | 0 | pos_x, |
9954 | 0 | pos_y + next_blk_offset, |
9955 | 0 | i1_ref_idx, |
9956 | 0 | resultid); |
9957 | |
|
9958 | 0 | i4_num_srch_cands += 4; |
9959 | |
|
9960 | 0 | if(ME_MEDIUM_SPEED <= e_me_quality_presets) |
9961 | 0 | { |
9962 | 0 | pf_hme_project_coloc_candt( |
9963 | 0 | ps_candt_prj_tr[0], |
9964 | 0 | ps_curr_layer, |
9965 | 0 | ps_coarse_layer, |
9966 | 0 | pos_x + next_blk_offset, |
9967 | 0 | pos_y - prev_blk_offset, |
9968 | 0 | i1_ref_idx, |
9969 | 0 | resultid); |
9970 | 0 | pf_hme_project_coloc_candt( |
9971 | 0 | ps_candt_prj_tl[0], |
9972 | 0 | ps_curr_layer, |
9973 | 0 | ps_coarse_layer, |
9974 | 0 | pos_x - prev_blk_offset, |
9975 | 0 | pos_y - prev_blk_offset, |
9976 | 0 | i1_ref_idx, |
9977 | 0 | resultid); |
9978 | |
|
9979 | 0 | i4_num_srch_cands += 2; |
9980 | 0 | } |
9981 | 0 | } |
9982 | 0 | if((num_results_prev_layer > 1) && (e_search_complexity >= SEARCH_CX_MED)) |
9983 | 0 | { |
9984 | 0 | resultid = 1; |
9985 | 0 | pf_hme_project_coloc_candt( |
9986 | 0 | ps_candt_prj_br[1], |
9987 | 0 | ps_curr_layer, |
9988 | 0 | ps_coarse_layer, |
9989 | 0 | pos_x + next_blk_offset, |
9990 | 0 | pos_y + next_blk_offset, |
9991 | 0 | i1_ref_idx, |
9992 | 0 | resultid); |
9993 | 0 | pf_hme_project_coloc_candt( |
9994 | 0 | ps_candt_prj_bl[1], |
9995 | 0 | ps_curr_layer, |
9996 | 0 | ps_coarse_layer, |
9997 | 0 | pos_x - prev_blk_offset, |
9998 | 0 | pos_y + next_blk_offset, |
9999 | 0 | i1_ref_idx, |
10000 | 0 | resultid); |
10001 | 0 | pf_hme_project_coloc_candt( |
10002 | 0 | ps_candt_prj_r[1], |
10003 | 0 | ps_curr_layer, |
10004 | 0 | ps_coarse_layer, |
10005 | 0 | pos_x + next_blk_offset, |
10006 | 0 | pos_y, |
10007 | 0 | i1_ref_idx, |
10008 | 0 | resultid); |
10009 | 0 | pf_hme_project_coloc_candt( |
10010 | 0 | ps_candt_prj_b[1], |
10011 | 0 | ps_curr_layer, |
10012 | 0 | ps_coarse_layer, |
10013 | 0 | pos_x, |
10014 | 0 | pos_y + next_blk_offset, |
10015 | 0 | i1_ref_idx, |
10016 | 0 | resultid); |
10017 | |
|
10018 | 0 | i4_num_srch_cands += 4; |
10019 | |
|
10020 | 0 | pf_hme_project_coloc_candt( |
10021 | 0 | ps_candt_prj_tr[1], |
10022 | 0 | ps_curr_layer, |
10023 | 0 | ps_coarse_layer, |
10024 | 0 | pos_x + next_blk_offset, |
10025 | 0 | pos_y - prev_blk_offset, |
10026 | 0 | i1_ref_idx, |
10027 | 0 | resultid); |
10028 | 0 | pf_hme_project_coloc_candt( |
10029 | 0 | ps_candt_prj_tl[1], |
10030 | 0 | ps_curr_layer, |
10031 | 0 | ps_coarse_layer, |
10032 | 0 | pos_x - prev_blk_offset, |
10033 | 0 | pos_y - prev_blk_offset, |
10034 | 0 | i1_ref_idx, |
10035 | 0 | resultid); |
10036 | 0 | pf_hme_project_coloc_candt( |
10037 | 0 | ps_candt_prj_t[1], |
10038 | 0 | ps_curr_layer, |
10039 | 0 | ps_coarse_layer, |
10040 | 0 | pos_x, |
10041 | 0 | pos_y - prev_blk_offset, |
10042 | 0 | i1_ref_idx, |
10043 | 0 | resultid); |
10044 | |
|
10045 | 0 | i4_num_srch_cands += 3; |
10046 | 0 | } |
10047 | | |
10048 | | /* Note this block also clips the MV range for all candidates */ |
10049 | | #ifdef _DEBUG |
10050 | | { |
10051 | | S32 candt; |
10052 | | range_prms_t *ps_range_prms; |
10053 | | |
10054 | | S32 num_ref_valid = ps_ctxt->num_ref_future + ps_ctxt->num_ref_past; |
10055 | | for(candt = 0; candt < i4_num_srch_cands; candt++) |
10056 | | { |
10057 | | search_node_t *ps_search_node; |
10058 | | |
10059 | | ps_search_node = |
10060 | | s_search_prms_blk.ps_search_candts[candt].ps_search_node; |
10061 | | |
10062 | | ps_range_prms = s_search_prms_blk.aps_mv_range[0]; |
10063 | | |
10064 | | if((ps_search_node->i1_ref_idx >= num_ref_valid) || |
10065 | | (ps_search_node->i1_ref_idx < 0)) |
10066 | | { |
10067 | | ASSERT(0); |
10068 | | } |
10069 | | } |
10070 | | } |
10071 | | #endif |
10072 | |
|
10073 | 0 | { |
10074 | 0 | S32 srch_cand; |
10075 | 0 | S32 num_unique_nodes = 0; |
10076 | 0 | S32 num_nodes_searched = 0; |
10077 | 0 | S32 num_best_cand = 0; |
10078 | 0 | S08 i1_grid_enable = 0; |
10079 | 0 | search_node_t as_best_two_proj_node[TOT_NUM_PARTS * 2]; |
10080 | | /* has list of valid partition to search terminated by -1 */ |
10081 | 0 | S32 ai4_valid_part_ids[TOT_NUM_PARTS + 1]; |
10082 | 0 | S32 center_x; |
10083 | 0 | S32 center_y; |
10084 | | |
10085 | | /* indicates if the centre point of grid needs to be explicitly added for search */ |
10086 | 0 | S32 add_centre = 0; |
10087 | |
|
10088 | 0 | memset(au4_unique_node_map, 0, sizeof(au4_unique_node_map)); |
10089 | 0 | center_x = ps_candt_prj_coloc[0]->s_mv.i2_mvx; |
10090 | 0 | center_y = ps_candt_prj_coloc[0]->s_mv.i2_mvy; |
10091 | |
|
10092 | 0 | for(srch_cand = 0; |
10093 | 0 | (srch_cand < i4_num_srch_cands) && |
10094 | 0 | (num_unique_nodes <= s_search_prms_blk.i4_num_init_candts); |
10095 | 0 | srch_cand++) |
10096 | 0 | { |
10097 | 0 | search_node_t s_search_node_temp = |
10098 | 0 | s_search_prms_blk.ps_search_candts[srch_cand].ps_search_node[0]; |
10099 | |
|
10100 | 0 | s_search_node_temp.i1_ref_idx = i1_ref_idx; //TEMP FIX; |
10101 | | |
10102 | | /* Clip the motion vectors as well here since after clipping |
10103 | | two candidates can become same and they will be removed during deduplication */ |
10104 | 0 | CLIP_MV_WITHIN_RANGE( |
10105 | 0 | s_search_node_temp.s_mv.i2_mvx, |
10106 | 0 | s_search_node_temp.s_mv.i2_mvy, |
10107 | 0 | s_search_prms_blk.aps_mv_range[0], |
10108 | 0 | ps_refine_prms->i4_num_steps_fpel_refine, |
10109 | 0 | ps_refine_prms->i4_num_steps_hpel_refine, |
10110 | 0 | ps_refine_prms->i4_num_steps_qpel_refine); |
10111 | | |
10112 | | /* PT_C */ |
10113 | 0 | INSERT_NEW_NODE( |
10114 | 0 | as_unique_search_nodes, |
10115 | 0 | num_unique_nodes, |
10116 | 0 | s_search_node_temp, |
10117 | 0 | 0, |
10118 | 0 | au4_unique_node_map, |
10119 | 0 | center_x, |
10120 | 0 | center_y, |
10121 | 0 | 1); |
10122 | |
|
10123 | 0 | num_nodes_searched += 1; |
10124 | 0 | } |
10125 | 0 | num_unique_nodes = |
10126 | 0 | MIN(num_unique_nodes, s_search_prms_blk.i4_num_init_candts); |
10127 | | |
10128 | | /* If number of candidates projected/number of candidates to be refined are more than 2, |
10129 | | then filter out and choose the best two here */ |
10130 | 0 | if(num_unique_nodes >= 2) |
10131 | 0 | { |
10132 | 0 | S32 num_results; |
10133 | 0 | S32 cnt; |
10134 | 0 | S32 *pi4_valid_part_ids; |
10135 | 0 | s_search_prms_blk.ps_search_nodes = &as_unique_search_nodes[0]; |
10136 | 0 | s_search_prms_blk.i4_num_search_nodes = num_unique_nodes; |
10137 | 0 | pi4_valid_part_ids = &ai4_valid_part_ids[0]; |
10138 | | |
10139 | | /* pi4_valid_part_ids is updated inside */ |
10140 | 0 | hme_pred_search_no_encode( |
10141 | 0 | &s_search_prms_blk, |
10142 | 0 | ps_curr_layer, |
10143 | 0 | &ps_ctxt->s_wt_pred, |
10144 | 0 | pi4_valid_part_ids, |
10145 | 0 | 1, |
10146 | 0 | e_me_quality_presets, |
10147 | 0 | i1_grid_enable, |
10148 | 0 | (ihevce_me_optimised_function_list_t *) |
10149 | 0 | ps_ctxt->pv_me_optimised_function_list |
10150 | |
|
10151 | 0 | ); |
10152 | |
|
10153 | 0 | num_best_cand = 0; |
10154 | 0 | cnt = 0; |
10155 | 0 | num_results = ps_search_results->u1_num_results_per_part; |
10156 | |
|
10157 | 0 | while((id = pi4_valid_part_ids[cnt++]) >= 0) |
10158 | 0 | { |
10159 | 0 | num_results = |
10160 | 0 | MIN(ps_refine_prms->pu1_num_best_results[id], num_results); |
10161 | |
|
10162 | 0 | for(i = 0; i < num_results; i++) |
10163 | 0 | { |
10164 | 0 | search_node_t s_search_node_temp; |
10165 | 0 | s_search_node_temp = |
10166 | 0 | *(ps_search_results->aps_part_results[i1_ref_idx][id] + i); |
10167 | 0 | if(s_search_node_temp.i1_ref_idx >= 0) |
10168 | 0 | { |
10169 | 0 | INSERT_NEW_NODE_NOMAP( |
10170 | 0 | as_best_two_proj_node, |
10171 | 0 | num_best_cand, |
10172 | 0 | s_search_node_temp, |
10173 | 0 | 0); |
10174 | 0 | } |
10175 | 0 | } |
10176 | 0 | } |
10177 | 0 | } |
10178 | 0 | else |
10179 | 0 | { |
10180 | 0 | add_centre = 1; |
10181 | 0 | num_best_cand = num_unique_nodes; |
10182 | 0 | as_best_two_proj_node[0] = as_unique_search_nodes[0]; |
10183 | 0 | } |
10184 | |
|
10185 | 0 | num_unique_nodes = 0; |
10186 | 0 | num_nodes_searched = 0; |
10187 | |
|
10188 | 0 | if(1 == num_best_cand) |
10189 | 0 | { |
10190 | 0 | search_node_t s_search_node_temp = as_best_two_proj_node[0]; |
10191 | 0 | S16 i2_mv_x = s_search_node_temp.s_mv.i2_mvx; |
10192 | 0 | S16 i2_mv_y = s_search_node_temp.s_mv.i2_mvy; |
10193 | 0 | S08 i1_ref_idx = s_search_node_temp.i1_ref_idx; |
10194 | |
|
10195 | 0 | i1_grid_enable = 1; |
10196 | |
|
10197 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvx = i2_mv_x - 1; |
10198 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvy = i2_mv_y - 1; |
10199 | 0 | as_unique_search_nodes[num_unique_nodes++].i1_ref_idx = i1_ref_idx; |
10200 | |
|
10201 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvx = i2_mv_x; |
10202 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvy = i2_mv_y - 1; |
10203 | 0 | as_unique_search_nodes[num_unique_nodes++].i1_ref_idx = i1_ref_idx; |
10204 | |
|
10205 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvx = i2_mv_x + 1; |
10206 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvy = i2_mv_y - 1; |
10207 | 0 | as_unique_search_nodes[num_unique_nodes++].i1_ref_idx = i1_ref_idx; |
10208 | |
|
10209 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvx = i2_mv_x - 1; |
10210 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvy = i2_mv_y; |
10211 | 0 | as_unique_search_nodes[num_unique_nodes++].i1_ref_idx = i1_ref_idx; |
10212 | |
|
10213 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvx = i2_mv_x + 1; |
10214 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvy = i2_mv_y; |
10215 | 0 | as_unique_search_nodes[num_unique_nodes++].i1_ref_idx = i1_ref_idx; |
10216 | |
|
10217 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvx = i2_mv_x - 1; |
10218 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvy = i2_mv_y + 1; |
10219 | 0 | as_unique_search_nodes[num_unique_nodes++].i1_ref_idx = i1_ref_idx; |
10220 | |
|
10221 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvx = i2_mv_x; |
10222 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvy = i2_mv_y + 1; |
10223 | 0 | as_unique_search_nodes[num_unique_nodes++].i1_ref_idx = i1_ref_idx; |
10224 | |
|
10225 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvx = i2_mv_x + 1; |
10226 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvy = i2_mv_y + 1; |
10227 | 0 | as_unique_search_nodes[num_unique_nodes++].i1_ref_idx = i1_ref_idx; |
10228 | |
|
10229 | 0 | if(add_centre) |
10230 | 0 | { |
10231 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvx = i2_mv_x; |
10232 | 0 | as_unique_search_nodes[num_unique_nodes].s_mv.i2_mvy = i2_mv_y; |
10233 | 0 | as_unique_search_nodes[num_unique_nodes++].i1_ref_idx = i1_ref_idx; |
10234 | 0 | } |
10235 | 0 | } |
10236 | 0 | else |
10237 | 0 | { |
10238 | | /* For the candidates where refinement was required, choose the best two */ |
10239 | 0 | for(srch_cand = 0; srch_cand < num_best_cand; srch_cand++) |
10240 | 0 | { |
10241 | 0 | search_node_t s_search_node_temp = as_best_two_proj_node[srch_cand]; |
10242 | 0 | WORD32 mv_x = s_search_node_temp.s_mv.i2_mvx; |
10243 | 0 | WORD32 mv_y = s_search_node_temp.s_mv.i2_mvy; |
10244 | | |
10245 | | /* Because there may not be two best unique candidates (because of clipping), |
10246 | | second best candidate can be uninitialized, ignore that */ |
10247 | 0 | if(s_search_node_temp.s_mv.i2_mvx == INTRA_MV || |
10248 | 0 | s_search_node_temp.i1_ref_idx < 0) |
10249 | 0 | { |
10250 | 0 | num_nodes_searched++; |
10251 | 0 | continue; |
10252 | 0 | } |
10253 | | |
10254 | | /* PT_C */ |
10255 | | /* Since the center point has already be evaluated and best results are persistent, |
10256 | | it will not be evaluated again */ |
10257 | 0 | if(add_centre) /* centre point added explicitly again if search results is not updated */ |
10258 | 0 | { |
10259 | 0 | INSERT_NEW_NODE( |
10260 | 0 | as_unique_search_nodes, |
10261 | 0 | num_unique_nodes, |
10262 | 0 | s_search_node_temp, |
10263 | 0 | 0, |
10264 | 0 | au4_unique_node_map, |
10265 | 0 | center_x, |
10266 | 0 | center_y, |
10267 | 0 | 1); |
10268 | 0 | } |
10269 | | |
10270 | | /* PT_L */ |
10271 | 0 | s_search_node_temp.s_mv.i2_mvx = mv_x - 1; |
10272 | 0 | s_search_node_temp.s_mv.i2_mvy = mv_y; |
10273 | 0 | INSERT_NEW_NODE( |
10274 | 0 | as_unique_search_nodes, |
10275 | 0 | num_unique_nodes, |
10276 | 0 | s_search_node_temp, |
10277 | 0 | 0, |
10278 | 0 | au4_unique_node_map, |
10279 | 0 | center_x, |
10280 | 0 | center_y, |
10281 | 0 | 1); |
10282 | | |
10283 | | /* PT_T */ |
10284 | 0 | s_search_node_temp.s_mv.i2_mvx = mv_x; |
10285 | 0 | s_search_node_temp.s_mv.i2_mvy = mv_y - 1; |
10286 | 0 | INSERT_NEW_NODE( |
10287 | 0 | as_unique_search_nodes, |
10288 | 0 | num_unique_nodes, |
10289 | 0 | s_search_node_temp, |
10290 | 0 | 0, |
10291 | 0 | au4_unique_node_map, |
10292 | 0 | center_x, |
10293 | 0 | center_y, |
10294 | 0 | 1); |
10295 | | |
10296 | | /* PT_R */ |
10297 | 0 | s_search_node_temp.s_mv.i2_mvx = mv_x + 1; |
10298 | 0 | s_search_node_temp.s_mv.i2_mvy = mv_y; |
10299 | 0 | INSERT_NEW_NODE( |
10300 | 0 | as_unique_search_nodes, |
10301 | 0 | num_unique_nodes, |
10302 | 0 | s_search_node_temp, |
10303 | 0 | 0, |
10304 | 0 | au4_unique_node_map, |
10305 | 0 | center_x, |
10306 | 0 | center_y, |
10307 | 0 | 1); |
10308 | | |
10309 | | /* PT_B */ |
10310 | 0 | s_search_node_temp.s_mv.i2_mvx = mv_x; |
10311 | 0 | s_search_node_temp.s_mv.i2_mvy = mv_y + 1; |
10312 | 0 | INSERT_NEW_NODE( |
10313 | 0 | as_unique_search_nodes, |
10314 | 0 | num_unique_nodes, |
10315 | 0 | s_search_node_temp, |
10316 | 0 | 0, |
10317 | 0 | au4_unique_node_map, |
10318 | 0 | center_x, |
10319 | 0 | center_y, |
10320 | 0 | 1); |
10321 | | |
10322 | | /* PT_TL */ |
10323 | 0 | s_search_node_temp.s_mv.i2_mvx = mv_x - 1; |
10324 | 0 | s_search_node_temp.s_mv.i2_mvy = mv_y - 1; |
10325 | 0 | INSERT_NEW_NODE( |
10326 | 0 | as_unique_search_nodes, |
10327 | 0 | num_unique_nodes, |
10328 | 0 | s_search_node_temp, |
10329 | 0 | 0, |
10330 | 0 | au4_unique_node_map, |
10331 | 0 | center_x, |
10332 | 0 | center_y, |
10333 | 0 | 1); |
10334 | | |
10335 | | /* PT_TR */ |
10336 | 0 | s_search_node_temp.s_mv.i2_mvx = mv_x + 1; |
10337 | 0 | s_search_node_temp.s_mv.i2_mvy = mv_y - 1; |
10338 | 0 | INSERT_NEW_NODE( |
10339 | 0 | as_unique_search_nodes, |
10340 | 0 | num_unique_nodes, |
10341 | 0 | s_search_node_temp, |
10342 | 0 | 0, |
10343 | 0 | au4_unique_node_map, |
10344 | 0 | center_x, |
10345 | 0 | center_y, |
10346 | 0 | 1); |
10347 | | |
10348 | | /* PT_BL */ |
10349 | 0 | s_search_node_temp.s_mv.i2_mvx = mv_x - 1; |
10350 | 0 | s_search_node_temp.s_mv.i2_mvy = mv_y + 1; |
10351 | 0 | INSERT_NEW_NODE( |
10352 | 0 | as_unique_search_nodes, |
10353 | 0 | num_unique_nodes, |
10354 | 0 | s_search_node_temp, |
10355 | 0 | 0, |
10356 | 0 | au4_unique_node_map, |
10357 | 0 | center_x, |
10358 | 0 | center_y, |
10359 | 0 | 1); |
10360 | | |
10361 | | /* PT_BR */ |
10362 | 0 | s_search_node_temp.s_mv.i2_mvx = mv_x + 1; |
10363 | 0 | s_search_node_temp.s_mv.i2_mvy = mv_y + 1; |
10364 | 0 | INSERT_NEW_NODE( |
10365 | 0 | as_unique_search_nodes, |
10366 | 0 | num_unique_nodes, |
10367 | 0 | s_search_node_temp, |
10368 | 0 | 0, |
10369 | 0 | au4_unique_node_map, |
10370 | 0 | center_x, |
10371 | 0 | center_y, |
10372 | 0 | 1); |
10373 | 0 | } |
10374 | 0 | } |
10375 | |
|
10376 | 0 | s_search_prms_blk.ps_search_nodes = &as_unique_search_nodes[0]; |
10377 | 0 | s_search_prms_blk.i4_num_search_nodes = num_unique_nodes; |
10378 | | |
10379 | | /*****************************************************************/ |
10380 | | /* Call the search algorithm, this includes: */ |
10381 | | /* Pre-Search-Refinement (for coarse candts) */ |
10382 | | /* Search on each candidate */ |
10383 | | /* Post Search Refinement on winners/other new candidates */ |
10384 | | /*****************************************************************/ |
10385 | |
|
10386 | 0 | hme_pred_search_no_encode( |
10387 | 0 | &s_search_prms_blk, |
10388 | 0 | ps_curr_layer, |
10389 | 0 | &ps_ctxt->s_wt_pred, |
10390 | 0 | ai4_valid_part_ids, |
10391 | 0 | 0, |
10392 | 0 | e_me_quality_presets, |
10393 | 0 | i1_grid_enable, |
10394 | 0 | (ihevce_me_optimised_function_list_t *) |
10395 | 0 | ps_ctxt->pv_me_optimised_function_list); |
10396 | |
|
10397 | 0 | i1_grid_enable = 0; |
10398 | 0 | } |
10399 | 0 | } |
10400 | | |
10401 | | /* for non encode layer update MV and end processing for block */ |
10402 | 0 | { |
10403 | 0 | WORD32 i4_ref_id, min_cost = 0x7fffffff, min_sad = 0; |
10404 | 0 | search_node_t *ps_search_node; |
10405 | | /* now update the reqd results back to the layer mv bank. */ |
10406 | 0 | if(1 == ps_refine_prms->i4_layer_id) |
10407 | 0 | { |
10408 | 0 | hme_update_mv_bank_in_l1_me( |
10409 | 0 | ps_search_results, |
10410 | 0 | ps_curr_layer->ps_layer_mvbank, |
10411 | 0 | blk_x, |
10412 | 0 | blk_y, |
10413 | 0 | &s_mv_update_prms); |
10414 | 0 | } |
10415 | 0 | else |
10416 | 0 | { |
10417 | 0 | hme_update_mv_bank_noencode( |
10418 | 0 | ps_search_results, |
10419 | 0 | ps_curr_layer->ps_layer_mvbank, |
10420 | 0 | blk_x, |
10421 | 0 | blk_y, |
10422 | 0 | &s_mv_update_prms); |
10423 | 0 | } |
10424 | | |
10425 | | /* UPDATE the MIN and MAX MVs for Dynamical Search Range for each ref. pic. */ |
10426 | | /* Only for P pic. For P, both are 0, I&B has them mut. exclusive */ |
10427 | 0 | if(ps_ctxt->s_frm_prms.is_i_pic == ps_ctxt->s_frm_prms.bidir_enabled) |
10428 | 0 | { |
10429 | 0 | WORD32 i4_j; |
10430 | 0 | layer_mv_t *ps_layer_mv = ps_curr_layer->ps_layer_mvbank; |
10431 | | |
10432 | | //if (ps_layer_mv->e_blk_size == s_mv_update_prms.e_search_blk_size) |
10433 | | /* Not considering this for Dyn. Search Update */ |
10434 | 0 | { |
10435 | 0 | for(i4_ref_id = 0; i4_ref_id < (S32)s_mv_update_prms.i4_num_ref; |
10436 | 0 | i4_ref_id++) |
10437 | 0 | { |
10438 | 0 | ps_search_node = |
10439 | 0 | ps_search_results->aps_part_results[i4_ref_id][PART_ID_2Nx2N]; |
10440 | |
|
10441 | 0 | for(i4_j = 0; i4_j < ps_layer_mv->i4_num_mvs_per_ref; i4_j++) |
10442 | 0 | { |
10443 | 0 | hme_update_dynamic_search_params( |
10444 | 0 | &ps_ctxt->s_coarse_dyn_range_prms |
10445 | 0 | .as_dyn_range_prms[ps_refine_prms->i4_layer_id] |
10446 | 0 | [i4_ref_id], |
10447 | 0 | ps_search_node->s_mv.i2_mvy); |
10448 | |
|
10449 | 0 | ps_search_node++; |
10450 | 0 | } |
10451 | 0 | } |
10452 | 0 | } |
10453 | 0 | } |
10454 | |
|
10455 | 0 | if(1 == ps_refine_prms->i4_layer_id) |
10456 | 0 | { |
10457 | 0 | WORD32 wt_pred_val, log_wt_pred_val; |
10458 | 0 | WORD32 ref_id_of_nearest_poc = 0; |
10459 | 0 | WORD32 max_val = 0x7fffffff; |
10460 | 0 | WORD32 max_l0_val = 0x7fffffff; |
10461 | 0 | WORD32 max_l1_val = 0x7fffffff; |
10462 | 0 | WORD32 cur_val; |
10463 | 0 | WORD32 i4_local_weighted_sad, i4_local_cost_weighted_pred; |
10464 | |
|
10465 | 0 | WORD32 bestl0_sad = 0x7fffffff; |
10466 | 0 | WORD32 bestl1_sad = 0x7fffffff; |
10467 | 0 | search_node_t *ps_best_l0_blk = NULL, *ps_best_l1_blk = NULL; |
10468 | |
|
10469 | 0 | for(i4_ref_id = 0; i4_ref_id < (S32)s_mv_update_prms.i4_num_ref; |
10470 | 0 | i4_ref_id++) |
10471 | 0 | { |
10472 | 0 | wt_pred_val = ps_ctxt->s_wt_pred.a_wpred_wt[i4_ref_id]; |
10473 | 0 | log_wt_pred_val = ps_ctxt->s_wt_pred.wpred_log_wdc; |
10474 | |
|
10475 | 0 | ps_search_node = |
10476 | 0 | ps_search_results->aps_part_results[i4_ref_id][PART_ID_2Nx2N]; |
10477 | |
|
10478 | 0 | i4_local_weighted_sad = ((ps_search_node->i4_sad * wt_pred_val) + |
10479 | 0 | ((1 << log_wt_pred_val) >> 1)) >> |
10480 | 0 | log_wt_pred_val; |
10481 | |
|
10482 | 0 | i4_local_cost_weighted_pred = |
10483 | 0 | i4_local_weighted_sad + |
10484 | 0 | (ps_search_node->i4_tot_cost - ps_search_node->i4_sad); |
10485 | | //the loop is redundant as the results are already sorted based on total cost |
10486 | | //for (i4_j = 0; i4_j < ps_curr_layer->ps_layer_mvbank->i4_num_mvs_per_ref; i4_j++) |
10487 | 0 | { |
10488 | 0 | if(i4_local_cost_weighted_pred < min_cost) |
10489 | 0 | { |
10490 | 0 | min_cost = i4_local_cost_weighted_pred; |
10491 | 0 | min_sad = i4_local_weighted_sad; |
10492 | 0 | } |
10493 | 0 | } |
10494 | | |
10495 | | /* For P frame, calculate the nearest poc which is either P or I frame*/ |
10496 | 0 | if(ps_ctxt->s_frm_prms.is_i_pic == ps_ctxt->s_frm_prms.bidir_enabled) |
10497 | 0 | { |
10498 | 0 | if(-1 != ps_coarse_layer->ai4_ref_id_to_poc_lc[i4_ref_id]) |
10499 | 0 | { |
10500 | 0 | cur_val = |
10501 | 0 | ABS(ps_ctxt->i4_curr_poc - |
10502 | 0 | ps_coarse_layer->ai4_ref_id_to_poc_lc[i4_ref_id]); |
10503 | 0 | if(cur_val < max_val) |
10504 | 0 | { |
10505 | 0 | max_val = cur_val; |
10506 | 0 | ref_id_of_nearest_poc = i4_ref_id; |
10507 | 0 | } |
10508 | 0 | } |
10509 | 0 | } |
10510 | 0 | } |
10511 | | /*Store me cost wrt. to past frame only for P frame */ |
10512 | 0 | if(ps_ctxt->s_frm_prms.is_i_pic == ps_ctxt->s_frm_prms.bidir_enabled) |
10513 | 0 | { |
10514 | 0 | if(-1 != ps_coarse_layer->ai4_ref_id_to_poc_lc[ref_id_of_nearest_poc]) |
10515 | 0 | { |
10516 | 0 | WORD16 i2_mvx, i2_mvy; |
10517 | |
|
10518 | 0 | WORD32 i4_diff_col_ctr = blk_x - (i4_ctb_blk_ctr * 4); |
10519 | 0 | WORD32 i4_diff_row_ctr = blk_y - (i4_ctb_row_ctr * 4); |
10520 | 0 | WORD32 z_scan_idx = |
10521 | 0 | gau1_raster_scan_to_ctb[i4_diff_row_ctr][i4_diff_col_ctr]; |
10522 | 0 | WORD32 wt, log_wt; |
10523 | | |
10524 | | /*ASSERT((ps_ctxt->i4_curr_poc - ps_coarse_layer->ai4_ref_id_to_poc_lc[ref_id_of_nearest_poc]) |
10525 | | <= (1 + ps_ctxt->num_b_frms));*/ |
10526 | | |
10527 | | /*obtain mvx and mvy */ |
10528 | 0 | i2_mvx = |
10529 | 0 | ps_search_results |
10530 | 0 | ->aps_part_results[ref_id_of_nearest_poc][PART_ID_2Nx2N] |
10531 | 0 | ->s_mv.i2_mvx; |
10532 | 0 | i2_mvy = |
10533 | 0 | ps_search_results |
10534 | 0 | ->aps_part_results[ref_id_of_nearest_poc][PART_ID_2Nx2N] |
10535 | 0 | ->s_mv.i2_mvy; |
10536 | | |
10537 | | /*register the min cost for l1 me in blk context */ |
10538 | 0 | wt = ps_ctxt->s_wt_pred.a_wpred_wt[ref_id_of_nearest_poc]; |
10539 | 0 | log_wt = ps_ctxt->s_wt_pred.wpred_log_wdc; |
10540 | | |
10541 | | /*register the min cost for l1 me in blk context */ |
10542 | 0 | ps_ed_ctb_l1_curr->i4_sad_me_for_ref[z_scan_idx >> 2] = |
10543 | 0 | ((ps_search_results |
10544 | 0 | ->aps_part_results[ref_id_of_nearest_poc][PART_ID_2Nx2N] |
10545 | 0 | ->i4_sad * |
10546 | 0 | wt) + |
10547 | 0 | ((1 << log_wt) >> 1)) >> |
10548 | 0 | log_wt; |
10549 | 0 | ps_ed_ctb_l1_curr->i4_sad_cost_me_for_ref[z_scan_idx >> 2] = |
10550 | 0 | ps_ed_ctb_l1_curr->i4_sad_me_for_ref[z_scan_idx >> 2] + |
10551 | 0 | (ps_search_results |
10552 | 0 | ->aps_part_results[ref_id_of_nearest_poc][PART_ID_2Nx2N] |
10553 | 0 | ->i4_tot_cost - |
10554 | 0 | ps_search_results |
10555 | 0 | ->aps_part_results[ref_id_of_nearest_poc][PART_ID_2Nx2N] |
10556 | 0 | ->i4_sad); |
10557 | | /*for complexity change detection*/ |
10558 | 0 | ps_ctxt->i4_num_blks++; |
10559 | 0 | if(ps_ed_ctb_l1_curr->i4_sad_cost_me_for_ref[z_scan_idx >> 2] > |
10560 | 0 | (8 /*blk width*/ * 8 /*blk height*/ * (1 + ps_ctxt->num_b_frms))) |
10561 | 0 | { |
10562 | 0 | ps_ctxt->i4_num_blks_high_sad++; |
10563 | 0 | } |
10564 | 0 | } |
10565 | 0 | } |
10566 | 0 | } |
10567 | | |
10568 | | /* EIID: Early inter intra decisions */ |
10569 | | /* tap L1 level SAD for inter intra decisions */ |
10570 | 0 | if((e_me_quality_presets >= ME_MEDIUM_SPEED) && |
10571 | 0 | (!ps_ctxt->s_frm_prms |
10572 | 0 | .is_i_pic)) //for high-quality preset->disable early decisions |
10573 | 0 | { |
10574 | 0 | if(1 == ps_refine_prms->i4_layer_id) |
10575 | 0 | { |
10576 | 0 | WORD32 i4_min_sad_cost_8x8_block = min_cost; |
10577 | 0 | ihevce_ed_blk_t *ps_curr_ed_blk_ctxt; |
10578 | 0 | WORD32 i4_diff_col_ctr = blk_x - (i4_ctb_blk_ctr * 4); |
10579 | 0 | WORD32 i4_diff_row_ctr = blk_y - (i4_ctb_row_ctr * 4); |
10580 | 0 | WORD32 z_scan_idx = |
10581 | 0 | gau1_raster_scan_to_ctb[i4_diff_row_ctr][i4_diff_col_ctr]; |
10582 | 0 | ps_curr_ed_blk_ctxt = ps_ed_blk_ctxt_curr_ctb + z_scan_idx; |
10583 | | |
10584 | | /*register the min cost for l1 me in blk context */ |
10585 | 0 | ps_ed_ctb_l1_curr->i4_best_sad_cost_8x8_l1_me[z_scan_idx >> 2] = |
10586 | 0 | i4_min_sad_cost_8x8_block; |
10587 | 0 | i4_num_comparisions++; |
10588 | | |
10589 | | /* take early inter-intra decision here */ |
10590 | 0 | ps_curr_ed_blk_ctxt->intra_or_inter = 3; /*init saying eval both */ |
10591 | 0 | #if DISABLE_INTRA_IN_BPICS |
10592 | 0 | if((e_me_quality_presets == ME_XTREME_SPEED_25) && |
10593 | 0 | (ps_ctxt->s_frm_prms.i4_temporal_layer_id > TEMPORAL_LAYER_DISABLE)) |
10594 | 0 | { |
10595 | 0 | ps_curr_ed_blk_ctxt->intra_or_inter = |
10596 | 0 | 2; /*eval only inter if inter cost is less */ |
10597 | 0 | i4_num_inter_wins++; |
10598 | 0 | } |
10599 | 0 | else |
10600 | 0 | #endif |
10601 | 0 | { |
10602 | 0 | if(ps_ed_ctb_l1_curr->i4_best_sad_cost_8x8_l1_me[z_scan_idx >> 2] < |
10603 | 0 | ((ps_ed_ctb_l1_curr->i4_best_sad_cost_8x8_l1_ipe[z_scan_idx >> 2] * |
10604 | 0 | i4_threshold_multiplier) / |
10605 | 0 | i4_threshold_divider)) |
10606 | 0 | { |
10607 | 0 | ps_curr_ed_blk_ctxt->intra_or_inter = |
10608 | 0 | 2; /*eval only inter if inter cost is less */ |
10609 | 0 | i4_num_inter_wins++; |
10610 | 0 | } |
10611 | 0 | } |
10612 | | |
10613 | | //{ |
10614 | | // DBG_PRINTF ("(blk x, blk y):(%d, %d)\t me:(ctb_x, ctb_y):(%d, %d)\t intra_SAD_COST: %d\tInter_SAD_COST: %d\n", |
10615 | | // blk_x,blk_y, |
10616 | | // i4_ctb_blk_ctr, i4_ctb_row_ctr, |
10617 | | // ps_curr_ed_blk_ctxt->i4_best_sad_8x8_l1_ipe, |
10618 | | // i4_min_sad_cost_8x8_block |
10619 | | // ); |
10620 | | //} |
10621 | |
|
10622 | 0 | } //end of layer-1 |
10623 | 0 | } //end of if (e_me_quality_presets >= ME_MEDIUM_SPEED) |
10624 | 0 | else |
10625 | 0 | { |
10626 | 0 | if(1 == ps_refine_prms->i4_layer_id) |
10627 | 0 | { |
10628 | 0 | WORD32 i4_min_sad_cost_8x8_block = min_cost; |
10629 | 0 | WORD32 i4_diff_col_ctr = blk_x - (i4_ctb_blk_ctr * 4); |
10630 | 0 | WORD32 i4_diff_row_ctr = blk_y - (i4_ctb_row_ctr * 4); |
10631 | 0 | WORD32 z_scan_idx = |
10632 | 0 | gau1_raster_scan_to_ctb[i4_diff_row_ctr][i4_diff_col_ctr]; |
10633 | | |
10634 | | /*register the min cost for l1 me in blk context */ |
10635 | 0 | ps_ed_ctb_l1_curr->i4_best_sad_cost_8x8_l1_me[z_scan_idx >> 2] = |
10636 | 0 | i4_min_sad_cost_8x8_block; |
10637 | 0 | } |
10638 | 0 | } |
10639 | 0 | if(1 == ps_refine_prms->i4_layer_id) |
10640 | 0 | { |
10641 | 0 | WORD32 i4_diff_col_ctr = blk_x - (i4_ctb_blk_ctr * 4); |
10642 | 0 | WORD32 i4_diff_row_ctr = blk_y - (i4_ctb_row_ctr * 4); |
10643 | 0 | WORD32 z_scan_idx = |
10644 | 0 | gau1_raster_scan_to_ctb[i4_diff_row_ctr][i4_diff_col_ctr]; |
10645 | |
|
10646 | 0 | ps_ed_ctb_l1_curr->i4_best_sad_8x8_l1_me_for_decide[z_scan_idx >> 2] = |
10647 | 0 | min_sad; |
10648 | |
|
10649 | 0 | if(min_cost < |
10650 | 0 | ps_ed_ctb_l1_curr->i4_best_sad_cost_8x8_l1_ipe[z_scan_idx >> 2]) |
10651 | 0 | { |
10652 | 0 | ps_ctxt->i4_L1_hme_best_cost += min_cost; |
10653 | 0 | ps_ctxt->i4_L1_hme_sad += min_sad; |
10654 | 0 | ps_ed_ctb_l1_curr->i4_best_sad_8x8_l1_me[z_scan_idx >> 2] = min_sad; |
10655 | 0 | } |
10656 | 0 | else |
10657 | 0 | { |
10658 | 0 | ps_ctxt->i4_L1_hme_best_cost += |
10659 | 0 | ps_ed_ctb_l1_curr->i4_best_sad_cost_8x8_l1_ipe[z_scan_idx >> 2]; |
10660 | 0 | ps_ctxt->i4_L1_hme_sad += |
10661 | 0 | ps_ed_ctb_l1_curr->i4_best_sad_8x8_l1_ipe[z_scan_idx >> 2]; |
10662 | 0 | ps_ed_ctb_l1_curr->i4_best_sad_8x8_l1_me[z_scan_idx >> 2] = |
10663 | 0 | ps_ed_ctb_l1_curr->i4_best_sad_8x8_l1_ipe[z_scan_idx >> 2]; |
10664 | 0 | } |
10665 | 0 | } |
10666 | 0 | } |
10667 | 0 | } |
10668 | | |
10669 | | /* Update the number of blocks processed in the current row */ |
10670 | 0 | if((ME_MEDIUM_SPEED > e_me_quality_presets)) |
10671 | 0 | { |
10672 | 0 | ihevce_dmgr_set_row_row_sync( |
10673 | 0 | pv_hme_dep_mngr, |
10674 | 0 | (i4_ctb_x + 1), |
10675 | 0 | blk_y, |
10676 | 0 | 0 /* Col Tile No. : Not supported in PreEnc*/); |
10677 | 0 | } |
10678 | 0 | } |
10679 | | |
10680 | | /* set the output dependency after completion of row */ |
10681 | 0 | ihevce_pre_enc_grp_job_set_out_dep(ps_multi_thrd_ctxt, ps_job, i4_ping_pong); |
10682 | 0 | } |
10683 | 0 | } |