1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/encoder/rdopt.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,2642 @@
1.4 +/*
1.5 + * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license
1.8 + * that can be found in the LICENSE file in the root of the source
1.9 + * tree. An additional intellectual property rights grant can be found
1.10 + * in the file PATENTS. All contributing project authors may
1.11 + * be found in the AUTHORS file in the root of the source tree.
1.12 + */
1.13 +
1.14 +
1.15 +#include <stdio.h>
1.16 +#include <math.h>
1.17 +#include <limits.h>
1.18 +#include <assert.h>
1.19 +#include "vpx_config.h"
1.20 +#include "vp8_rtcd.h"
1.21 +#include "vp8/common/pragmas.h"
1.22 +#include "tokenize.h"
1.23 +#include "treewriter.h"
1.24 +#include "onyx_int.h"
1.25 +#include "modecosts.h"
1.26 +#include "encodeintra.h"
1.27 +#include "pickinter.h"
1.28 +#include "vp8/common/entropymode.h"
1.29 +#include "vp8/common/reconinter.h"
1.30 +#include "vp8/common/reconintra4x4.h"
1.31 +#include "vp8/common/findnearmv.h"
1.32 +#include "vp8/common/quant_common.h"
1.33 +#include "encodemb.h"
1.34 +#include "quantize.h"
1.35 +#include "vp8/common/variance.h"
1.36 +#include "mcomp.h"
1.37 +#include "rdopt.h"
1.38 +#include "vpx_mem/vpx_mem.h"
1.39 +#include "vp8/common/systemdependent.h"
1.40 +#if CONFIG_TEMPORAL_DENOISING
1.41 +#include "denoising.h"
1.42 +#endif
1.43 +extern void vp8_update_zbin_extra(VP8_COMP *cpi, MACROBLOCK *x);
1.44 +
1.45 +#define MAXF(a,b) (((a) > (b)) ? (a) : (b))
1.46 +
1.47 +typedef struct rate_distortion_struct
1.48 +{
1.49 + int rate2;
1.50 + int rate_y;
1.51 + int rate_uv;
1.52 + int distortion2;
1.53 + int distortion_uv;
1.54 +} RATE_DISTORTION;
1.55 +
1.56 +typedef struct best_mode_struct
1.57 +{
1.58 + int yrd;
1.59 + int rd;
1.60 + int intra_rd;
1.61 + MB_MODE_INFO mbmode;
1.62 + union b_mode_info bmodes[16];
1.63 + PARTITION_INFO partition;
1.64 +} BEST_MODE;
1.65 +
1.66 +static const int auto_speed_thresh[17] =
1.67 +{
1.68 + 1000,
1.69 + 200,
1.70 + 150,
1.71 + 130,
1.72 + 150,
1.73 + 125,
1.74 + 120,
1.75 + 115,
1.76 + 115,
1.77 + 115,
1.78 + 115,
1.79 + 115,
1.80 + 115,
1.81 + 115,
1.82 + 115,
1.83 + 115,
1.84 + 105
1.85 +};
1.86 +
1.87 +const MB_PREDICTION_MODE vp8_mode_order[MAX_MODES] =
1.88 +{
1.89 + ZEROMV,
1.90 + DC_PRED,
1.91 +
1.92 + NEARESTMV,
1.93 + NEARMV,
1.94 +
1.95 + ZEROMV,
1.96 + NEARESTMV,
1.97 +
1.98 + ZEROMV,
1.99 + NEARESTMV,
1.100 +
1.101 + NEARMV,
1.102 + NEARMV,
1.103 +
1.104 + V_PRED,
1.105 + H_PRED,
1.106 + TM_PRED,
1.107 +
1.108 + NEWMV,
1.109 + NEWMV,
1.110 + NEWMV,
1.111 +
1.112 + SPLITMV,
1.113 + SPLITMV,
1.114 + SPLITMV,
1.115 +
1.116 + B_PRED,
1.117 +};
1.118 +
1.119 +/* This table determines the search order in reference frame priority order,
1.120 + * which may not necessarily match INTRA,LAST,GOLDEN,ARF
1.121 + */
1.122 +const int vp8_ref_frame_order[MAX_MODES] =
1.123 +{
1.124 + 1,
1.125 + 0,
1.126 +
1.127 + 1,
1.128 + 1,
1.129 +
1.130 + 2,
1.131 + 2,
1.132 +
1.133 + 3,
1.134 + 3,
1.135 +
1.136 + 2,
1.137 + 3,
1.138 +
1.139 + 0,
1.140 + 0,
1.141 + 0,
1.142 +
1.143 + 1,
1.144 + 2,
1.145 + 3,
1.146 +
1.147 + 1,
1.148 + 2,
1.149 + 3,
1.150 +
1.151 + 0,
1.152 +};
1.153 +
1.154 +static void fill_token_costs(
1.155 + int c[BLOCK_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS],
1.156 + const vp8_prob p[BLOCK_TYPES][COEF_BANDS][PREV_COEF_CONTEXTS][ENTROPY_NODES]
1.157 +)
1.158 +{
1.159 + int i, j, k;
1.160 +
1.161 +
1.162 + for (i = 0; i < BLOCK_TYPES; i++)
1.163 + for (j = 0; j < COEF_BANDS; j++)
1.164 + for (k = 0; k < PREV_COEF_CONTEXTS; k++)
1.165 +
1.166 + /* check for pt=0 and band > 1 if block type 0
1.167 + * and 0 if blocktype 1
1.168 + */
1.169 + if (k == 0 && j > (i == 0))
1.170 + vp8_cost_tokens2(c[i][j][k], p [i][j][k], vp8_coef_tree, 2);
1.171 + else
1.172 + vp8_cost_tokens(c[i][j][k], p [i][j][k], vp8_coef_tree);
1.173 +}
1.174 +
1.175 +static const int rd_iifactor[32] =
1.176 +{
1.177 + 4, 4, 3, 2, 1, 0, 0, 0,
1.178 + 0, 0, 0, 0, 0, 0, 0, 0,
1.179 + 0, 0, 0, 0, 0, 0, 0, 0,
1.180 + 0, 0, 0, 0, 0, 0, 0, 0
1.181 +};
1.182 +
1.183 +/* values are now correlated to quantizer */
1.184 +static const int sad_per_bit16lut[QINDEX_RANGE] =
1.185 +{
1.186 + 2, 2, 2, 2, 2, 2, 2, 2,
1.187 + 2, 2, 2, 2, 2, 2, 2, 2,
1.188 + 3, 3, 3, 3, 3, 3, 3, 3,
1.189 + 3, 3, 3, 3, 3, 3, 4, 4,
1.190 + 4, 4, 4, 4, 4, 4, 4, 4,
1.191 + 4, 4, 5, 5, 5, 5, 5, 5,
1.192 + 5, 5, 5, 5, 5, 5, 6, 6,
1.193 + 6, 6, 6, 6, 6, 6, 6, 6,
1.194 + 6, 6, 7, 7, 7, 7, 7, 7,
1.195 + 7, 7, 7, 7, 7, 7, 8, 8,
1.196 + 8, 8, 8, 8, 8, 8, 8, 8,
1.197 + 8, 8, 9, 9, 9, 9, 9, 9,
1.198 + 9, 9, 9, 9, 9, 9, 10, 10,
1.199 + 10, 10, 10, 10, 10, 10, 11, 11,
1.200 + 11, 11, 11, 11, 12, 12, 12, 12,
1.201 + 12, 12, 13, 13, 13, 13, 14, 14
1.202 +};
1.203 +static const int sad_per_bit4lut[QINDEX_RANGE] =
1.204 +{
1.205 + 2, 2, 2, 2, 2, 2, 3, 3,
1.206 + 3, 3, 3, 3, 3, 3, 3, 3,
1.207 + 3, 3, 3, 3, 4, 4, 4, 4,
1.208 + 4, 4, 4, 4, 4, 4, 5, 5,
1.209 + 5, 5, 5, 5, 6, 6, 6, 6,
1.210 + 6, 6, 6, 6, 6, 6, 6, 6,
1.211 + 7, 7, 7, 7, 7, 7, 7, 7,
1.212 + 7, 7, 7, 7, 7, 8, 8, 8,
1.213 + 8, 8, 9, 9, 9, 9, 9, 9,
1.214 + 10, 10, 10, 10, 10, 10, 10, 10,
1.215 + 11, 11, 11, 11, 11, 11, 11, 11,
1.216 + 12, 12, 12, 12, 12, 12, 12, 12,
1.217 + 13, 13, 13, 13, 13, 13, 13, 14,
1.218 + 14, 14, 14, 14, 15, 15, 15, 15,
1.219 + 16, 16, 16, 16, 17, 17, 17, 18,
1.220 + 18, 18, 19, 19, 19, 20, 20, 20,
1.221 +};
1.222 +
1.223 +void vp8cx_initialize_me_consts(VP8_COMP *cpi, int QIndex)
1.224 +{
1.225 + cpi->mb.sadperbit16 = sad_per_bit16lut[QIndex];
1.226 + cpi->mb.sadperbit4 = sad_per_bit4lut[QIndex];
1.227 +}
1.228 +
1.229 +void vp8_initialize_rd_consts(VP8_COMP *cpi, MACROBLOCK *x, int Qvalue)
1.230 +{
1.231 + int q;
1.232 + int i;
1.233 + double capped_q = (Qvalue < 160) ? (double)Qvalue : 160.0;
1.234 + double rdconst = 2.80;
1.235 +
1.236 + vp8_clear_system_state();
1.237 +
1.238 + /* Further tests required to see if optimum is different
1.239 + * for key frames, golden frames and arf frames.
1.240 + */
1.241 + cpi->RDMULT = (int)(rdconst * (capped_q * capped_q));
1.242 +
1.243 + /* Extend rate multiplier along side quantizer zbin increases */
1.244 + if (cpi->mb.zbin_over_quant > 0)
1.245 + {
1.246 + double oq_factor;
1.247 + double modq;
1.248 +
1.249 + /* Experimental code using the same basic equation as used for Q above
1.250 + * The units of cpi->mb.zbin_over_quant are 1/128 of Q bin size
1.251 + */
1.252 + oq_factor = 1.0 + ((double)0.0015625 * cpi->mb.zbin_over_quant);
1.253 + modq = (int)((double)capped_q * oq_factor);
1.254 + cpi->RDMULT = (int)(rdconst * (modq * modq));
1.255 + }
1.256 +
1.257 + if (cpi->pass == 2 && (cpi->common.frame_type != KEY_FRAME))
1.258 + {
1.259 + if (cpi->twopass.next_iiratio > 31)
1.260 + cpi->RDMULT += (cpi->RDMULT * rd_iifactor[31]) >> 4;
1.261 + else
1.262 + cpi->RDMULT +=
1.263 + (cpi->RDMULT * rd_iifactor[cpi->twopass.next_iiratio]) >> 4;
1.264 + }
1.265 +
1.266 + cpi->mb.errorperbit = (cpi->RDMULT / 110);
1.267 + cpi->mb.errorperbit += (cpi->mb.errorperbit==0);
1.268 +
1.269 + vp8_set_speed_features(cpi);
1.270 +
1.271 + for (i = 0; i < MAX_MODES; i++)
1.272 + {
1.273 + x->mode_test_hit_counts[i] = 0;
1.274 + }
1.275 +
1.276 + q = (int)pow(Qvalue, 1.25);
1.277 +
1.278 + if (q < 8)
1.279 + q = 8;
1.280 +
1.281 + if (cpi->RDMULT > 1000)
1.282 + {
1.283 + cpi->RDDIV = 1;
1.284 + cpi->RDMULT /= 100;
1.285 +
1.286 + for (i = 0; i < MAX_MODES; i++)
1.287 + {
1.288 + if (cpi->sf.thresh_mult[i] < INT_MAX)
1.289 + {
1.290 + x->rd_threshes[i] = cpi->sf.thresh_mult[i] * q / 100;
1.291 + }
1.292 + else
1.293 + {
1.294 + x->rd_threshes[i] = INT_MAX;
1.295 + }
1.296 +
1.297 + cpi->rd_baseline_thresh[i] = x->rd_threshes[i];
1.298 + }
1.299 + }
1.300 + else
1.301 + {
1.302 + cpi->RDDIV = 100;
1.303 +
1.304 + for (i = 0; i < MAX_MODES; i++)
1.305 + {
1.306 + if (cpi->sf.thresh_mult[i] < (INT_MAX / q))
1.307 + {
1.308 + x->rd_threshes[i] = cpi->sf.thresh_mult[i] * q;
1.309 + }
1.310 + else
1.311 + {
1.312 + x->rd_threshes[i] = INT_MAX;
1.313 + }
1.314 +
1.315 + cpi->rd_baseline_thresh[i] = x->rd_threshes[i];
1.316 + }
1.317 + }
1.318 +
1.319 + {
1.320 + /* build token cost array for the type of frame we have now */
1.321 + FRAME_CONTEXT *l = &cpi->lfc_n;
1.322 +
1.323 + if(cpi->common.refresh_alt_ref_frame)
1.324 + l = &cpi->lfc_a;
1.325 + else if(cpi->common.refresh_golden_frame)
1.326 + l = &cpi->lfc_g;
1.327 +
1.328 + fill_token_costs(
1.329 + cpi->mb.token_costs,
1.330 + (const vp8_prob( *)[8][3][11]) l->coef_probs
1.331 + );
1.332 + /*
1.333 + fill_token_costs(
1.334 + cpi->mb.token_costs,
1.335 + (const vp8_prob( *)[8][3][11]) cpi->common.fc.coef_probs);
1.336 + */
1.337 +
1.338 +
1.339 + /* TODO make these mode costs depend on last,alt or gold too. (jbb) */
1.340 + vp8_init_mode_costs(cpi);
1.341 + }
1.342 +
1.343 +}
1.344 +
1.345 +void vp8_auto_select_speed(VP8_COMP *cpi)
1.346 +{
1.347 + int milliseconds_for_compress = (int)(1000000 / cpi->framerate);
1.348 +
1.349 + milliseconds_for_compress = milliseconds_for_compress * (16 - cpi->oxcf.cpu_used) / 16;
1.350 +
1.351 +#if 0
1.352 +
1.353 + if (0)
1.354 + {
1.355 + FILE *f;
1.356 +
1.357 + f = fopen("speed.stt", "a");
1.358 + fprintf(f, " %8ld %10ld %10ld %10ld\n",
1.359 + cpi->common.current_video_frame, cpi->Speed, milliseconds_for_compress, cpi->avg_pick_mode_time);
1.360 + fclose(f);
1.361 + }
1.362 +
1.363 +#endif
1.364 +
1.365 + if (cpi->avg_pick_mode_time < milliseconds_for_compress && (cpi->avg_encode_time - cpi->avg_pick_mode_time) < milliseconds_for_compress)
1.366 + {
1.367 + if (cpi->avg_pick_mode_time == 0)
1.368 + {
1.369 + cpi->Speed = 4;
1.370 + }
1.371 + else
1.372 + {
1.373 + if (milliseconds_for_compress * 100 < cpi->avg_encode_time * 95)
1.374 + {
1.375 + cpi->Speed += 2;
1.376 + cpi->avg_pick_mode_time = 0;
1.377 + cpi->avg_encode_time = 0;
1.378 +
1.379 + if (cpi->Speed > 16)
1.380 + {
1.381 + cpi->Speed = 16;
1.382 + }
1.383 + }
1.384 +
1.385 + if (milliseconds_for_compress * 100 > cpi->avg_encode_time * auto_speed_thresh[cpi->Speed])
1.386 + {
1.387 + cpi->Speed -= 1;
1.388 + cpi->avg_pick_mode_time = 0;
1.389 + cpi->avg_encode_time = 0;
1.390 +
1.391 + /* In real-time mode, cpi->speed is in [4, 16]. */
1.392 + if (cpi->Speed < 4)
1.393 + {
1.394 + cpi->Speed = 4;
1.395 + }
1.396 + }
1.397 + }
1.398 + }
1.399 + else
1.400 + {
1.401 + cpi->Speed += 4;
1.402 +
1.403 + if (cpi->Speed > 16)
1.404 + cpi->Speed = 16;
1.405 +
1.406 +
1.407 + cpi->avg_pick_mode_time = 0;
1.408 + cpi->avg_encode_time = 0;
1.409 + }
1.410 +}
1.411 +
1.412 +int vp8_block_error_c(short *coeff, short *dqcoeff)
1.413 +{
1.414 + int i;
1.415 + int error = 0;
1.416 +
1.417 + for (i = 0; i < 16; i++)
1.418 + {
1.419 + int this_diff = coeff[i] - dqcoeff[i];
1.420 + error += this_diff * this_diff;
1.421 + }
1.422 +
1.423 + return error;
1.424 +}
1.425 +
1.426 +int vp8_mbblock_error_c(MACROBLOCK *mb, int dc)
1.427 +{
1.428 + BLOCK *be;
1.429 + BLOCKD *bd;
1.430 + int i, j;
1.431 + int berror, error = 0;
1.432 +
1.433 + for (i = 0; i < 16; i++)
1.434 + {
1.435 + be = &mb->block[i];
1.436 + bd = &mb->e_mbd.block[i];
1.437 +
1.438 + berror = 0;
1.439 +
1.440 + for (j = dc; j < 16; j++)
1.441 + {
1.442 + int this_diff = be->coeff[j] - bd->dqcoeff[j];
1.443 + berror += this_diff * this_diff;
1.444 + }
1.445 +
1.446 + error += berror;
1.447 + }
1.448 +
1.449 + return error;
1.450 +}
1.451 +
1.452 +int vp8_mbuverror_c(MACROBLOCK *mb)
1.453 +{
1.454 +
1.455 + BLOCK *be;
1.456 + BLOCKD *bd;
1.457 +
1.458 +
1.459 + int i;
1.460 + int error = 0;
1.461 +
1.462 + for (i = 16; i < 24; i++)
1.463 + {
1.464 + be = &mb->block[i];
1.465 + bd = &mb->e_mbd.block[i];
1.466 +
1.467 + error += vp8_block_error_c(be->coeff, bd->dqcoeff);
1.468 + }
1.469 +
1.470 + return error;
1.471 +}
1.472 +
1.473 +int VP8_UVSSE(MACROBLOCK *x)
1.474 +{
1.475 + unsigned char *uptr, *vptr;
1.476 + unsigned char *upred_ptr = (*(x->block[16].base_src) + x->block[16].src);
1.477 + unsigned char *vpred_ptr = (*(x->block[20].base_src) + x->block[20].src);
1.478 + int uv_stride = x->block[16].src_stride;
1.479 +
1.480 + unsigned int sse1 = 0;
1.481 + unsigned int sse2 = 0;
1.482 + int mv_row = x->e_mbd.mode_info_context->mbmi.mv.as_mv.row;
1.483 + int mv_col = x->e_mbd.mode_info_context->mbmi.mv.as_mv.col;
1.484 + int offset;
1.485 + int pre_stride = x->e_mbd.pre.uv_stride;
1.486 +
1.487 + if (mv_row < 0)
1.488 + mv_row -= 1;
1.489 + else
1.490 + mv_row += 1;
1.491 +
1.492 + if (mv_col < 0)
1.493 + mv_col -= 1;
1.494 + else
1.495 + mv_col += 1;
1.496 +
1.497 + mv_row /= 2;
1.498 + mv_col /= 2;
1.499 +
1.500 + offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
1.501 + uptr = x->e_mbd.pre.u_buffer + offset;
1.502 + vptr = x->e_mbd.pre.v_buffer + offset;
1.503 +
1.504 + if ((mv_row | mv_col) & 7)
1.505 + {
1.506 + vp8_sub_pixel_variance8x8(uptr, pre_stride,
1.507 + mv_col & 7, mv_row & 7, upred_ptr, uv_stride, &sse2);
1.508 + vp8_sub_pixel_variance8x8(vptr, pre_stride,
1.509 + mv_col & 7, mv_row & 7, vpred_ptr, uv_stride, &sse1);
1.510 + sse2 += sse1;
1.511 + }
1.512 + else
1.513 + {
1.514 + vp8_variance8x8(uptr, pre_stride,
1.515 + upred_ptr, uv_stride, &sse2);
1.516 + vp8_variance8x8(vptr, pre_stride,
1.517 + vpred_ptr, uv_stride, &sse1);
1.518 + sse2 += sse1;
1.519 + }
1.520 + return sse2;
1.521 +
1.522 +}
1.523 +
1.524 +static int cost_coeffs(MACROBLOCK *mb, BLOCKD *b, int type, ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l)
1.525 +{
1.526 + int c = !type; /* start at coef 0, unless Y with Y2 */
1.527 + int eob = (int)(*b->eob);
1.528 + int pt ; /* surrounding block/prev coef predictor */
1.529 + int cost = 0;
1.530 + short *qcoeff_ptr = b->qcoeff;
1.531 +
1.532 + VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
1.533 +
1.534 +# define QC( I) ( qcoeff_ptr [vp8_default_zig_zag1d[I]] )
1.535 +
1.536 + for (; c < eob; c++)
1.537 + {
1.538 + int v = QC(c);
1.539 + int t = vp8_dct_value_tokens_ptr[v].Token;
1.540 + cost += mb->token_costs [type] [vp8_coef_bands[c]] [pt] [t];
1.541 + cost += vp8_dct_value_cost_ptr[v];
1.542 + pt = vp8_prev_token_class[t];
1.543 + }
1.544 +
1.545 +# undef QC
1.546 +
1.547 + if (c < 16)
1.548 + cost += mb->token_costs [type] [vp8_coef_bands[c]] [pt] [DCT_EOB_TOKEN];
1.549 +
1.550 + pt = (c != !type); /* is eob first coefficient; */
1.551 + *a = *l = pt;
1.552 +
1.553 + return cost;
1.554 +}
1.555 +
1.556 +static int vp8_rdcost_mby(MACROBLOCK *mb)
1.557 +{
1.558 + int cost = 0;
1.559 + int b;
1.560 + MACROBLOCKD *x = &mb->e_mbd;
1.561 + ENTROPY_CONTEXT_PLANES t_above, t_left;
1.562 + ENTROPY_CONTEXT *ta;
1.563 + ENTROPY_CONTEXT *tl;
1.564 +
1.565 + vpx_memcpy(&t_above, mb->e_mbd.above_context, sizeof(ENTROPY_CONTEXT_PLANES));
1.566 + vpx_memcpy(&t_left, mb->e_mbd.left_context, sizeof(ENTROPY_CONTEXT_PLANES));
1.567 +
1.568 + ta = (ENTROPY_CONTEXT *)&t_above;
1.569 + tl = (ENTROPY_CONTEXT *)&t_left;
1.570 +
1.571 + for (b = 0; b < 16; b++)
1.572 + cost += cost_coeffs(mb, x->block + b, PLANE_TYPE_Y_NO_DC,
1.573 + ta + vp8_block2above[b], tl + vp8_block2left[b]);
1.574 +
1.575 + cost += cost_coeffs(mb, x->block + 24, PLANE_TYPE_Y2,
1.576 + ta + vp8_block2above[24], tl + vp8_block2left[24]);
1.577 +
1.578 + return cost;
1.579 +}
1.580 +
1.581 +static void macro_block_yrd( MACROBLOCK *mb,
1.582 + int *Rate,
1.583 + int *Distortion)
1.584 +{
1.585 + int b;
1.586 + MACROBLOCKD *const x = &mb->e_mbd;
1.587 + BLOCK *const mb_y2 = mb->block + 24;
1.588 + BLOCKD *const x_y2 = x->block + 24;
1.589 + short *Y2DCPtr = mb_y2->src_diff;
1.590 + BLOCK *beptr;
1.591 + int d;
1.592 +
1.593 + vp8_subtract_mby( mb->src_diff, *(mb->block[0].base_src),
1.594 + mb->block[0].src_stride, mb->e_mbd.predictor, 16);
1.595 +
1.596 + /* Fdct and building the 2nd order block */
1.597 + for (beptr = mb->block; beptr < mb->block + 16; beptr += 2)
1.598 + {
1.599 + mb->short_fdct8x4(beptr->src_diff, beptr->coeff, 32);
1.600 + *Y2DCPtr++ = beptr->coeff[0];
1.601 + *Y2DCPtr++ = beptr->coeff[16];
1.602 + }
1.603 +
1.604 + /* 2nd order fdct */
1.605 + mb->short_walsh4x4(mb_y2->src_diff, mb_y2->coeff, 8);
1.606 +
1.607 + /* Quantization */
1.608 + for (b = 0; b < 16; b++)
1.609 + {
1.610 + mb->quantize_b(&mb->block[b], &mb->e_mbd.block[b]);
1.611 + }
1.612 +
1.613 + /* DC predication and Quantization of 2nd Order block */
1.614 + mb->quantize_b(mb_y2, x_y2);
1.615 +
1.616 + /* Distortion */
1.617 + d = vp8_mbblock_error(mb, 1) << 2;
1.618 + d += vp8_block_error(mb_y2->coeff, x_y2->dqcoeff);
1.619 +
1.620 + *Distortion = (d >> 4);
1.621 +
1.622 + /* rate */
1.623 + *Rate = vp8_rdcost_mby(mb);
1.624 +}
1.625 +
1.626 +static void copy_predictor(unsigned char *dst, const unsigned char *predictor)
1.627 +{
1.628 + const unsigned int *p = (const unsigned int *)predictor;
1.629 + unsigned int *d = (unsigned int *)dst;
1.630 + d[0] = p[0];
1.631 + d[4] = p[4];
1.632 + d[8] = p[8];
1.633 + d[12] = p[12];
1.634 +}
1.635 +static int rd_pick_intra4x4block(
1.636 + MACROBLOCK *x,
1.637 + BLOCK *be,
1.638 + BLOCKD *b,
1.639 + B_PREDICTION_MODE *best_mode,
1.640 + const int *bmode_costs,
1.641 + ENTROPY_CONTEXT *a,
1.642 + ENTROPY_CONTEXT *l,
1.643 +
1.644 + int *bestrate,
1.645 + int *bestratey,
1.646 + int *bestdistortion)
1.647 +{
1.648 + B_PREDICTION_MODE mode;
1.649 + int best_rd = INT_MAX;
1.650 + int rate = 0;
1.651 + int distortion;
1.652 +
1.653 + ENTROPY_CONTEXT ta = *a, tempa = *a;
1.654 + ENTROPY_CONTEXT tl = *l, templ = *l;
1.655 + /*
1.656 + * The predictor buffer is a 2d buffer with a stride of 16. Create
1.657 + * a temp buffer that meets the stride requirements, but we are only
1.658 + * interested in the left 4x4 block
1.659 + * */
1.660 + DECLARE_ALIGNED_ARRAY(16, unsigned char, best_predictor, 16*4);
1.661 + DECLARE_ALIGNED_ARRAY(16, short, best_dqcoeff, 16);
1.662 + int dst_stride = x->e_mbd.dst.y_stride;
1.663 + unsigned char *dst = x->e_mbd.dst.y_buffer + b->offset;
1.664 +
1.665 + unsigned char *Above = dst - dst_stride;
1.666 + unsigned char *yleft = dst - 1;
1.667 + unsigned char top_left = Above[-1];
1.668 +
1.669 + for (mode = B_DC_PRED; mode <= B_HU_PRED; mode++)
1.670 + {
1.671 + int this_rd;
1.672 + int ratey;
1.673 +
1.674 + rate = bmode_costs[mode];
1.675 +
1.676 + vp8_intra4x4_predict(Above, yleft, dst_stride, mode,
1.677 + b->predictor, 16, top_left);
1.678 + vp8_subtract_b(be, b, 16);
1.679 + x->short_fdct4x4(be->src_diff, be->coeff, 32);
1.680 + x->quantize_b(be, b);
1.681 +
1.682 + tempa = ta;
1.683 + templ = tl;
1.684 +
1.685 + ratey = cost_coeffs(x, b, PLANE_TYPE_Y_WITH_DC, &tempa, &templ);
1.686 + rate += ratey;
1.687 + distortion = vp8_block_error(be->coeff, b->dqcoeff) >> 2;
1.688 +
1.689 + this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);
1.690 +
1.691 + if (this_rd < best_rd)
1.692 + {
1.693 + *bestrate = rate;
1.694 + *bestratey = ratey;
1.695 + *bestdistortion = distortion;
1.696 + best_rd = this_rd;
1.697 + *best_mode = mode;
1.698 + *a = tempa;
1.699 + *l = templ;
1.700 + copy_predictor(best_predictor, b->predictor);
1.701 + vpx_memcpy(best_dqcoeff, b->dqcoeff, 32);
1.702 + }
1.703 + }
1.704 + b->bmi.as_mode = *best_mode;
1.705 +
1.706 + vp8_short_idct4x4llm(best_dqcoeff, best_predictor, 16, dst, dst_stride);
1.707 +
1.708 + return best_rd;
1.709 +}
1.710 +
1.711 +static int rd_pick_intra4x4mby_modes(MACROBLOCK *mb, int *Rate,
1.712 + int *rate_y, int *Distortion, int best_rd)
1.713 +{
1.714 + MACROBLOCKD *const xd = &mb->e_mbd;
1.715 + int i;
1.716 + int cost = mb->mbmode_cost [xd->frame_type] [B_PRED];
1.717 + int distortion = 0;
1.718 + int tot_rate_y = 0;
1.719 + int64_t total_rd = 0;
1.720 + ENTROPY_CONTEXT_PLANES t_above, t_left;
1.721 + ENTROPY_CONTEXT *ta;
1.722 + ENTROPY_CONTEXT *tl;
1.723 + const int *bmode_costs;
1.724 +
1.725 + vpx_memcpy(&t_above, mb->e_mbd.above_context, sizeof(ENTROPY_CONTEXT_PLANES));
1.726 + vpx_memcpy(&t_left, mb->e_mbd.left_context, sizeof(ENTROPY_CONTEXT_PLANES));
1.727 +
1.728 + ta = (ENTROPY_CONTEXT *)&t_above;
1.729 + tl = (ENTROPY_CONTEXT *)&t_left;
1.730 +
1.731 + intra_prediction_down_copy(xd, xd->dst.y_buffer - xd->dst.y_stride + 16);
1.732 +
1.733 + bmode_costs = mb->inter_bmode_costs;
1.734 +
1.735 + for (i = 0; i < 16; i++)
1.736 + {
1.737 + MODE_INFO *const mic = xd->mode_info_context;
1.738 + const int mis = xd->mode_info_stride;
1.739 + B_PREDICTION_MODE UNINITIALIZED_IS_SAFE(best_mode);
1.740 + int UNINITIALIZED_IS_SAFE(r), UNINITIALIZED_IS_SAFE(ry), UNINITIALIZED_IS_SAFE(d);
1.741 +
1.742 + if (mb->e_mbd.frame_type == KEY_FRAME)
1.743 + {
1.744 + const B_PREDICTION_MODE A = above_block_mode(mic, i, mis);
1.745 + const B_PREDICTION_MODE L = left_block_mode(mic, i);
1.746 +
1.747 + bmode_costs = mb->bmode_costs[A][L];
1.748 + }
1.749 +
1.750 + total_rd += rd_pick_intra4x4block(
1.751 + mb, mb->block + i, xd->block + i, &best_mode, bmode_costs,
1.752 + ta + vp8_block2above[i],
1.753 + tl + vp8_block2left[i], &r, &ry, &d);
1.754 +
1.755 + cost += r;
1.756 + distortion += d;
1.757 + tot_rate_y += ry;
1.758 +
1.759 + mic->bmi[i].as_mode = best_mode;
1.760 +
1.761 + if(total_rd >= (int64_t)best_rd)
1.762 + break;
1.763 + }
1.764 +
1.765 + if(total_rd >= (int64_t)best_rd)
1.766 + return INT_MAX;
1.767 +
1.768 + *Rate = cost;
1.769 + *rate_y = tot_rate_y;
1.770 + *Distortion = distortion;
1.771 +
1.772 + return RDCOST(mb->rdmult, mb->rddiv, cost, distortion);
1.773 +}
1.774 +
1.775 +
1.776 +static int rd_pick_intra16x16mby_mode(MACROBLOCK *x,
1.777 + int *Rate,
1.778 + int *rate_y,
1.779 + int *Distortion)
1.780 +{
1.781 + MB_PREDICTION_MODE mode;
1.782 + MB_PREDICTION_MODE UNINITIALIZED_IS_SAFE(mode_selected);
1.783 + int rate, ratey;
1.784 + int distortion;
1.785 + int best_rd = INT_MAX;
1.786 + int this_rd;
1.787 + MACROBLOCKD *xd = &x->e_mbd;
1.788 +
1.789 + /* Y Search for 16x16 intra prediction mode */
1.790 + for (mode = DC_PRED; mode <= TM_PRED; mode++)
1.791 + {
1.792 + xd->mode_info_context->mbmi.mode = mode;
1.793 +
1.794 + vp8_build_intra_predictors_mby_s(xd,
1.795 + xd->dst.y_buffer - xd->dst.y_stride,
1.796 + xd->dst.y_buffer - 1,
1.797 + xd->dst.y_stride,
1.798 + xd->predictor,
1.799 + 16);
1.800 +
1.801 + macro_block_yrd(x, &ratey, &distortion);
1.802 + rate = ratey + x->mbmode_cost[xd->frame_type]
1.803 + [xd->mode_info_context->mbmi.mode];
1.804 +
1.805 + this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);
1.806 +
1.807 + if (this_rd < best_rd)
1.808 + {
1.809 + mode_selected = mode;
1.810 + best_rd = this_rd;
1.811 + *Rate = rate;
1.812 + *rate_y = ratey;
1.813 + *Distortion = distortion;
1.814 + }
1.815 + }
1.816 +
1.817 + xd->mode_info_context->mbmi.mode = mode_selected;
1.818 + return best_rd;
1.819 +}
1.820 +
1.821 +static int rd_cost_mbuv(MACROBLOCK *mb)
1.822 +{
1.823 + int b;
1.824 + int cost = 0;
1.825 + MACROBLOCKD *x = &mb->e_mbd;
1.826 + ENTROPY_CONTEXT_PLANES t_above, t_left;
1.827 + ENTROPY_CONTEXT *ta;
1.828 + ENTROPY_CONTEXT *tl;
1.829 +
1.830 + vpx_memcpy(&t_above, mb->e_mbd.above_context, sizeof(ENTROPY_CONTEXT_PLANES));
1.831 + vpx_memcpy(&t_left, mb->e_mbd.left_context, sizeof(ENTROPY_CONTEXT_PLANES));
1.832 +
1.833 + ta = (ENTROPY_CONTEXT *)&t_above;
1.834 + tl = (ENTROPY_CONTEXT *)&t_left;
1.835 +
1.836 + for (b = 16; b < 24; b++)
1.837 + cost += cost_coeffs(mb, x->block + b, PLANE_TYPE_UV,
1.838 + ta + vp8_block2above[b], tl + vp8_block2left[b]);
1.839 +
1.840 + return cost;
1.841 +}
1.842 +
1.843 +
1.844 +static int rd_inter16x16_uv(VP8_COMP *cpi, MACROBLOCK *x, int *rate,
1.845 + int *distortion, int fullpixel)
1.846 +{
1.847 + vp8_build_inter16x16_predictors_mbuv(&x->e_mbd);
1.848 + vp8_subtract_mbuv(x->src_diff,
1.849 + x->src.u_buffer, x->src.v_buffer, x->src.uv_stride,
1.850 + &x->e_mbd.predictor[256], &x->e_mbd.predictor[320], 8);
1.851 +
1.852 + vp8_transform_mbuv(x);
1.853 + vp8_quantize_mbuv(x);
1.854 +
1.855 + *rate = rd_cost_mbuv(x);
1.856 + *distortion = vp8_mbuverror(x) / 4;
1.857 +
1.858 + return RDCOST(x->rdmult, x->rddiv, *rate, *distortion);
1.859 +}
1.860 +
1.861 +static int rd_inter4x4_uv(VP8_COMP *cpi, MACROBLOCK *x, int *rate,
1.862 + int *distortion, int fullpixel)
1.863 +{
1.864 + vp8_build_inter4x4_predictors_mbuv(&x->e_mbd);
1.865 + vp8_subtract_mbuv(x->src_diff,
1.866 + x->src.u_buffer, x->src.v_buffer, x->src.uv_stride,
1.867 + &x->e_mbd.predictor[256], &x->e_mbd.predictor[320], 8);
1.868 +
1.869 + vp8_transform_mbuv(x);
1.870 + vp8_quantize_mbuv(x);
1.871 +
1.872 + *rate = rd_cost_mbuv(x);
1.873 + *distortion = vp8_mbuverror(x) / 4;
1.874 +
1.875 + return RDCOST(x->rdmult, x->rddiv, *rate, *distortion);
1.876 +}
1.877 +
1.878 +static void rd_pick_intra_mbuv_mode(MACROBLOCK *x, int *rate,
1.879 + int *rate_tokenonly, int *distortion)
1.880 +{
1.881 + MB_PREDICTION_MODE mode;
1.882 + MB_PREDICTION_MODE UNINITIALIZED_IS_SAFE(mode_selected);
1.883 + int best_rd = INT_MAX;
1.884 + int UNINITIALIZED_IS_SAFE(d), UNINITIALIZED_IS_SAFE(r);
1.885 + int rate_to;
1.886 + MACROBLOCKD *xd = &x->e_mbd;
1.887 +
1.888 + for (mode = DC_PRED; mode <= TM_PRED; mode++)
1.889 + {
1.890 + int this_rate;
1.891 + int this_distortion;
1.892 + int this_rd;
1.893 +
1.894 + xd->mode_info_context->mbmi.uv_mode = mode;
1.895 +
1.896 + vp8_build_intra_predictors_mbuv_s(xd,
1.897 + xd->dst.u_buffer - xd->dst.uv_stride,
1.898 + xd->dst.v_buffer - xd->dst.uv_stride,
1.899 + xd->dst.u_buffer - 1,
1.900 + xd->dst.v_buffer - 1,
1.901 + xd->dst.uv_stride,
1.902 + &xd->predictor[256], &xd->predictor[320],
1.903 + 8);
1.904 +
1.905 +
1.906 + vp8_subtract_mbuv(x->src_diff,
1.907 + x->src.u_buffer, x->src.v_buffer, x->src.uv_stride,
1.908 + &xd->predictor[256], &xd->predictor[320], 8);
1.909 + vp8_transform_mbuv(x);
1.910 + vp8_quantize_mbuv(x);
1.911 +
1.912 + rate_to = rd_cost_mbuv(x);
1.913 + this_rate = rate_to + x->intra_uv_mode_cost[xd->frame_type][xd->mode_info_context->mbmi.uv_mode];
1.914 +
1.915 + this_distortion = vp8_mbuverror(x) / 4;
1.916 +
1.917 + this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
1.918 +
1.919 + if (this_rd < best_rd)
1.920 + {
1.921 + best_rd = this_rd;
1.922 + d = this_distortion;
1.923 + r = this_rate;
1.924 + *rate_tokenonly = rate_to;
1.925 + mode_selected = mode;
1.926 + }
1.927 + }
1.928 +
1.929 + *rate = r;
1.930 + *distortion = d;
1.931 +
1.932 + xd->mode_info_context->mbmi.uv_mode = mode_selected;
1.933 +}
1.934 +
1.935 +int vp8_cost_mv_ref(MB_PREDICTION_MODE m, const int near_mv_ref_ct[4])
1.936 +{
1.937 + vp8_prob p [VP8_MVREFS-1];
1.938 + assert(NEARESTMV <= m && m <= SPLITMV);
1.939 + vp8_mv_ref_probs(p, near_mv_ref_ct);
1.940 + return vp8_cost_token(vp8_mv_ref_tree, p,
1.941 + vp8_mv_ref_encoding_array + (m - NEARESTMV));
1.942 +}
1.943 +
1.944 +void vp8_set_mbmode_and_mvs(MACROBLOCK *x, MB_PREDICTION_MODE mb, int_mv *mv)
1.945 +{
1.946 + x->e_mbd.mode_info_context->mbmi.mode = mb;
1.947 + x->e_mbd.mode_info_context->mbmi.mv.as_int = mv->as_int;
1.948 +}
1.949 +
1.950 +static int labels2mode(
1.951 + MACROBLOCK *x,
1.952 + int const *labelings, int which_label,
1.953 + B_PREDICTION_MODE this_mode,
1.954 + int_mv *this_mv, int_mv *best_ref_mv,
1.955 + int *mvcost[2]
1.956 +)
1.957 +{
1.958 + MACROBLOCKD *const xd = & x->e_mbd;
1.959 + MODE_INFO *const mic = xd->mode_info_context;
1.960 + const int mis = xd->mode_info_stride;
1.961 +
1.962 + int cost = 0;
1.963 + int thismvcost = 0;
1.964 +
1.965 + /* We have to be careful retrieving previously-encoded motion vectors.
1.966 + Ones from this macroblock have to be pulled from the BLOCKD array
1.967 + as they have not yet made it to the bmi array in our MB_MODE_INFO. */
1.968 +
1.969 + int i = 0;
1.970 +
1.971 + do
1.972 + {
1.973 + BLOCKD *const d = xd->block + i;
1.974 + const int row = i >> 2, col = i & 3;
1.975 +
1.976 + B_PREDICTION_MODE m;
1.977 +
1.978 + if (labelings[i] != which_label)
1.979 + continue;
1.980 +
1.981 + if (col && labelings[i] == labelings[i-1])
1.982 + m = LEFT4X4;
1.983 + else if (row && labelings[i] == labelings[i-4])
1.984 + m = ABOVE4X4;
1.985 + else
1.986 + {
1.987 + /* the only time we should do costing for new motion vector
1.988 + * or mode is when we are on a new label (jbb May 08, 2007)
1.989 + */
1.990 + switch (m = this_mode)
1.991 + {
1.992 + case NEW4X4 :
1.993 + thismvcost = vp8_mv_bit_cost(this_mv, best_ref_mv, mvcost, 102);
1.994 + break;
1.995 + case LEFT4X4:
1.996 + this_mv->as_int = col ? d[-1].bmi.mv.as_int : left_block_mv(mic, i);
1.997 + break;
1.998 + case ABOVE4X4:
1.999 + this_mv->as_int = row ? d[-4].bmi.mv.as_int : above_block_mv(mic, i, mis);
1.1000 + break;
1.1001 + case ZERO4X4:
1.1002 + this_mv->as_int = 0;
1.1003 + break;
1.1004 + default:
1.1005 + break;
1.1006 + }
1.1007 +
1.1008 + if (m == ABOVE4X4) /* replace above with left if same */
1.1009 + {
1.1010 + int_mv left_mv;
1.1011 +
1.1012 + left_mv.as_int = col ? d[-1].bmi.mv.as_int :
1.1013 + left_block_mv(mic, i);
1.1014 +
1.1015 + if (left_mv.as_int == this_mv->as_int)
1.1016 + m = LEFT4X4;
1.1017 + }
1.1018 +
1.1019 + cost = x->inter_bmode_costs[ m];
1.1020 + }
1.1021 +
1.1022 + d->bmi.mv.as_int = this_mv->as_int;
1.1023 +
1.1024 + x->partition_info->bmi[i].mode = m;
1.1025 + x->partition_info->bmi[i].mv.as_int = this_mv->as_int;
1.1026 +
1.1027 + }
1.1028 + while (++i < 16);
1.1029 +
1.1030 + cost += thismvcost ;
1.1031 + return cost;
1.1032 +}
1.1033 +
1.1034 +static int rdcost_mbsegment_y(MACROBLOCK *mb, const int *labels,
1.1035 + int which_label, ENTROPY_CONTEXT *ta,
1.1036 + ENTROPY_CONTEXT *tl)
1.1037 +{
1.1038 + int cost = 0;
1.1039 + int b;
1.1040 + MACROBLOCKD *x = &mb->e_mbd;
1.1041 +
1.1042 + for (b = 0; b < 16; b++)
1.1043 + if (labels[ b] == which_label)
1.1044 + cost += cost_coeffs(mb, x->block + b, PLANE_TYPE_Y_WITH_DC,
1.1045 + ta + vp8_block2above[b],
1.1046 + tl + vp8_block2left[b]);
1.1047 +
1.1048 + return cost;
1.1049 +
1.1050 +}
1.1051 +static unsigned int vp8_encode_inter_mb_segment(MACROBLOCK *x, int const *labels, int which_label)
1.1052 +{
1.1053 + int i;
1.1054 + unsigned int distortion = 0;
1.1055 + int pre_stride = x->e_mbd.pre.y_stride;
1.1056 + unsigned char *base_pre = x->e_mbd.pre.y_buffer;
1.1057 +
1.1058 +
1.1059 + for (i = 0; i < 16; i++)
1.1060 + {
1.1061 + if (labels[i] == which_label)
1.1062 + {
1.1063 + BLOCKD *bd = &x->e_mbd.block[i];
1.1064 + BLOCK *be = &x->block[i];
1.1065 +
1.1066 + vp8_build_inter_predictors_b(bd, 16, base_pre, pre_stride, x->e_mbd.subpixel_predict);
1.1067 + vp8_subtract_b(be, bd, 16);
1.1068 + x->short_fdct4x4(be->src_diff, be->coeff, 32);
1.1069 + x->quantize_b(be, bd);
1.1070 +
1.1071 + distortion += vp8_block_error(be->coeff, bd->dqcoeff);
1.1072 + }
1.1073 + }
1.1074 +
1.1075 + return distortion;
1.1076 +}
1.1077 +
1.1078 +
1.1079 +static const unsigned int segmentation_to_sseshift[4] = {3, 3, 2, 0};
1.1080 +
1.1081 +
1.1082 +typedef struct
1.1083 +{
1.1084 + int_mv *ref_mv;
1.1085 + int_mv mvp;
1.1086 +
1.1087 + int segment_rd;
1.1088 + int segment_num;
1.1089 + int r;
1.1090 + int d;
1.1091 + int segment_yrate;
1.1092 + B_PREDICTION_MODE modes[16];
1.1093 + int_mv mvs[16];
1.1094 + unsigned char eobs[16];
1.1095 +
1.1096 + int mvthresh;
1.1097 + int *mdcounts;
1.1098 +
1.1099 + int_mv sv_mvp[4]; /* save 4 mvp from 8x8 */
1.1100 + int sv_istep[2]; /* save 2 initial step_param for 16x8/8x16 */
1.1101 +
1.1102 +} BEST_SEG_INFO;
1.1103 +
1.1104 +
1.1105 +static void rd_check_segment(VP8_COMP *cpi, MACROBLOCK *x,
1.1106 + BEST_SEG_INFO *bsi, unsigned int segmentation)
1.1107 +{
1.1108 + int i;
1.1109 + int const *labels;
1.1110 + int br = 0;
1.1111 + int bd = 0;
1.1112 + B_PREDICTION_MODE this_mode;
1.1113 +
1.1114 +
1.1115 + int label_count;
1.1116 + int this_segment_rd = 0;
1.1117 + int label_mv_thresh;
1.1118 + int rate = 0;
1.1119 + int sbr = 0;
1.1120 + int sbd = 0;
1.1121 + int segmentyrate = 0;
1.1122 +
1.1123 + vp8_variance_fn_ptr_t *v_fn_ptr;
1.1124 +
1.1125 + ENTROPY_CONTEXT_PLANES t_above, t_left;
1.1126 + ENTROPY_CONTEXT *ta;
1.1127 + ENTROPY_CONTEXT *tl;
1.1128 + ENTROPY_CONTEXT_PLANES t_above_b, t_left_b;
1.1129 + ENTROPY_CONTEXT *ta_b;
1.1130 + ENTROPY_CONTEXT *tl_b;
1.1131 +
1.1132 + vpx_memcpy(&t_above, x->e_mbd.above_context, sizeof(ENTROPY_CONTEXT_PLANES));
1.1133 + vpx_memcpy(&t_left, x->e_mbd.left_context, sizeof(ENTROPY_CONTEXT_PLANES));
1.1134 +
1.1135 + ta = (ENTROPY_CONTEXT *)&t_above;
1.1136 + tl = (ENTROPY_CONTEXT *)&t_left;
1.1137 + ta_b = (ENTROPY_CONTEXT *)&t_above_b;
1.1138 + tl_b = (ENTROPY_CONTEXT *)&t_left_b;
1.1139 +
1.1140 + br = 0;
1.1141 + bd = 0;
1.1142 +
1.1143 + v_fn_ptr = &cpi->fn_ptr[segmentation];
1.1144 + labels = vp8_mbsplits[segmentation];
1.1145 + label_count = vp8_mbsplit_count[segmentation];
1.1146 +
1.1147 + /* 64 makes this threshold really big effectively making it so that we
1.1148 + * very rarely check mvs on segments. setting this to 1 would make mv
1.1149 + * thresh roughly equal to what it is for macroblocks
1.1150 + */
1.1151 + label_mv_thresh = 1 * bsi->mvthresh / label_count ;
1.1152 +
1.1153 + /* Segmentation method overheads */
1.1154 + rate = vp8_cost_token(vp8_mbsplit_tree, vp8_mbsplit_probs, vp8_mbsplit_encodings + segmentation);
1.1155 + rate += vp8_cost_mv_ref(SPLITMV, bsi->mdcounts);
1.1156 + this_segment_rd += RDCOST(x->rdmult, x->rddiv, rate, 0);
1.1157 + br += rate;
1.1158 +
1.1159 + for (i = 0; i < label_count; i++)
1.1160 + {
1.1161 + int_mv mode_mv[B_MODE_COUNT];
1.1162 + int best_label_rd = INT_MAX;
1.1163 + B_PREDICTION_MODE mode_selected = ZERO4X4;
1.1164 + int bestlabelyrate = 0;
1.1165 +
1.1166 + /* search for the best motion vector on this segment */
1.1167 + for (this_mode = LEFT4X4; this_mode <= NEW4X4 ; this_mode ++)
1.1168 + {
1.1169 + int this_rd;
1.1170 + int distortion;
1.1171 + int labelyrate;
1.1172 + ENTROPY_CONTEXT_PLANES t_above_s, t_left_s;
1.1173 + ENTROPY_CONTEXT *ta_s;
1.1174 + ENTROPY_CONTEXT *tl_s;
1.1175 +
1.1176 + vpx_memcpy(&t_above_s, &t_above, sizeof(ENTROPY_CONTEXT_PLANES));
1.1177 + vpx_memcpy(&t_left_s, &t_left, sizeof(ENTROPY_CONTEXT_PLANES));
1.1178 +
1.1179 + ta_s = (ENTROPY_CONTEXT *)&t_above_s;
1.1180 + tl_s = (ENTROPY_CONTEXT *)&t_left_s;
1.1181 +
1.1182 + if (this_mode == NEW4X4)
1.1183 + {
1.1184 + int sseshift;
1.1185 + int num00;
1.1186 + int step_param = 0;
1.1187 + int further_steps;
1.1188 + int n;
1.1189 + int thissme;
1.1190 + int bestsme = INT_MAX;
1.1191 + int_mv temp_mv;
1.1192 + BLOCK *c;
1.1193 + BLOCKD *e;
1.1194 +
1.1195 + /* Is the best so far sufficiently good that we cant justify
1.1196 + * doing a new motion search.
1.1197 + */
1.1198 + if (best_label_rd < label_mv_thresh)
1.1199 + break;
1.1200 +
1.1201 + if(cpi->compressor_speed)
1.1202 + {
1.1203 + if (segmentation == BLOCK_8X16 || segmentation == BLOCK_16X8)
1.1204 + {
1.1205 + bsi->mvp.as_int = bsi->sv_mvp[i].as_int;
1.1206 + if (i==1 && segmentation == BLOCK_16X8)
1.1207 + bsi->mvp.as_int = bsi->sv_mvp[2].as_int;
1.1208 +
1.1209 + step_param = bsi->sv_istep[i];
1.1210 + }
1.1211 +
1.1212 + /* use previous block's result as next block's MV
1.1213 + * predictor.
1.1214 + */
1.1215 + if (segmentation == BLOCK_4X4 && i>0)
1.1216 + {
1.1217 + bsi->mvp.as_int = x->e_mbd.block[i-1].bmi.mv.as_int;
1.1218 + if (i==4 || i==8 || i==12)
1.1219 + bsi->mvp.as_int = x->e_mbd.block[i-4].bmi.mv.as_int;
1.1220 + step_param = 2;
1.1221 + }
1.1222 + }
1.1223 +
1.1224 + further_steps = (MAX_MVSEARCH_STEPS - 1) - step_param;
1.1225 +
1.1226 + {
1.1227 + int sadpb = x->sadperbit4;
1.1228 + int_mv mvp_full;
1.1229 +
1.1230 + mvp_full.as_mv.row = bsi->mvp.as_mv.row >>3;
1.1231 + mvp_full.as_mv.col = bsi->mvp.as_mv.col >>3;
1.1232 +
1.1233 + /* find first label */
1.1234 + n = vp8_mbsplit_offset[segmentation][i];
1.1235 +
1.1236 + c = &x->block[n];
1.1237 + e = &x->e_mbd.block[n];
1.1238 +
1.1239 + {
1.1240 + bestsme = cpi->diamond_search_sad(x, c, e, &mvp_full,
1.1241 + &mode_mv[NEW4X4], step_param,
1.1242 + sadpb, &num00, v_fn_ptr,
1.1243 + x->mvcost, bsi->ref_mv);
1.1244 +
1.1245 + n = num00;
1.1246 + num00 = 0;
1.1247 +
1.1248 + while (n < further_steps)
1.1249 + {
1.1250 + n++;
1.1251 +
1.1252 + if (num00)
1.1253 + num00--;
1.1254 + else
1.1255 + {
1.1256 + thissme = cpi->diamond_search_sad(x, c, e,
1.1257 + &mvp_full, &temp_mv,
1.1258 + step_param + n, sadpb,
1.1259 + &num00, v_fn_ptr,
1.1260 + x->mvcost, bsi->ref_mv);
1.1261 +
1.1262 + if (thissme < bestsme)
1.1263 + {
1.1264 + bestsme = thissme;
1.1265 + mode_mv[NEW4X4].as_int = temp_mv.as_int;
1.1266 + }
1.1267 + }
1.1268 + }
1.1269 + }
1.1270 +
1.1271 + sseshift = segmentation_to_sseshift[segmentation];
1.1272 +
1.1273 + /* Should we do a full search (best quality only) */
1.1274 + if ((cpi->compressor_speed == 0) && (bestsme >> sseshift) > 4000)
1.1275 + {
1.1276 + /* Check if mvp_full is within the range. */
1.1277 + vp8_clamp_mv(&mvp_full, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
1.1278 +
1.1279 + thissme = cpi->full_search_sad(x, c, e, &mvp_full,
1.1280 + sadpb, 16, v_fn_ptr,
1.1281 + x->mvcost, bsi->ref_mv);
1.1282 +
1.1283 + if (thissme < bestsme)
1.1284 + {
1.1285 + bestsme = thissme;
1.1286 + mode_mv[NEW4X4].as_int = e->bmi.mv.as_int;
1.1287 + }
1.1288 + else
1.1289 + {
1.1290 + /* The full search result is actually worse so
1.1291 + * re-instate the previous best vector
1.1292 + */
1.1293 + e->bmi.mv.as_int = mode_mv[NEW4X4].as_int;
1.1294 + }
1.1295 + }
1.1296 + }
1.1297 +
1.1298 + if (bestsme < INT_MAX)
1.1299 + {
1.1300 + int disto;
1.1301 + unsigned int sse;
1.1302 + cpi->find_fractional_mv_step(x, c, e, &mode_mv[NEW4X4],
1.1303 + bsi->ref_mv, x->errorperbit, v_fn_ptr, x->mvcost,
1.1304 + &disto, &sse);
1.1305 + }
1.1306 + } /* NEW4X4 */
1.1307 +
1.1308 + rate = labels2mode(x, labels, i, this_mode, &mode_mv[this_mode],
1.1309 + bsi->ref_mv, x->mvcost);
1.1310 +
1.1311 + /* Trap vectors that reach beyond the UMV borders */
1.1312 + if (((mode_mv[this_mode].as_mv.row >> 3) < x->mv_row_min) || ((mode_mv[this_mode].as_mv.row >> 3) > x->mv_row_max) ||
1.1313 + ((mode_mv[this_mode].as_mv.col >> 3) < x->mv_col_min) || ((mode_mv[this_mode].as_mv.col >> 3) > x->mv_col_max))
1.1314 + {
1.1315 + continue;
1.1316 + }
1.1317 +
1.1318 + distortion = vp8_encode_inter_mb_segment(x, labels, i) / 4;
1.1319 +
1.1320 + labelyrate = rdcost_mbsegment_y(x, labels, i, ta_s, tl_s);
1.1321 + rate += labelyrate;
1.1322 +
1.1323 + this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion);
1.1324 +
1.1325 + if (this_rd < best_label_rd)
1.1326 + {
1.1327 + sbr = rate;
1.1328 + sbd = distortion;
1.1329 + bestlabelyrate = labelyrate;
1.1330 + mode_selected = this_mode;
1.1331 + best_label_rd = this_rd;
1.1332 +
1.1333 + vpx_memcpy(ta_b, ta_s, sizeof(ENTROPY_CONTEXT_PLANES));
1.1334 + vpx_memcpy(tl_b, tl_s, sizeof(ENTROPY_CONTEXT_PLANES));
1.1335 +
1.1336 + }
1.1337 + } /*for each 4x4 mode*/
1.1338 +
1.1339 + vpx_memcpy(ta, ta_b, sizeof(ENTROPY_CONTEXT_PLANES));
1.1340 + vpx_memcpy(tl, tl_b, sizeof(ENTROPY_CONTEXT_PLANES));
1.1341 +
1.1342 + labels2mode(x, labels, i, mode_selected, &mode_mv[mode_selected],
1.1343 + bsi->ref_mv, x->mvcost);
1.1344 +
1.1345 + br += sbr;
1.1346 + bd += sbd;
1.1347 + segmentyrate += bestlabelyrate;
1.1348 + this_segment_rd += best_label_rd;
1.1349 +
1.1350 + if (this_segment_rd >= bsi->segment_rd)
1.1351 + break;
1.1352 +
1.1353 + } /* for each label */
1.1354 +
1.1355 + if (this_segment_rd < bsi->segment_rd)
1.1356 + {
1.1357 + bsi->r = br;
1.1358 + bsi->d = bd;
1.1359 + bsi->segment_yrate = segmentyrate;
1.1360 + bsi->segment_rd = this_segment_rd;
1.1361 + bsi->segment_num = segmentation;
1.1362 +
1.1363 + /* store everything needed to come back to this!! */
1.1364 + for (i = 0; i < 16; i++)
1.1365 + {
1.1366 + bsi->mvs[i].as_mv = x->partition_info->bmi[i].mv.as_mv;
1.1367 + bsi->modes[i] = x->partition_info->bmi[i].mode;
1.1368 + bsi->eobs[i] = x->e_mbd.eobs[i];
1.1369 + }
1.1370 + }
1.1371 +}
1.1372 +
1.1373 +static
1.1374 +void vp8_cal_step_param(int sr, int *sp)
1.1375 +{
1.1376 + int step = 0;
1.1377 +
1.1378 + if (sr > MAX_FIRST_STEP) sr = MAX_FIRST_STEP;
1.1379 + else if (sr < 1) sr = 1;
1.1380 +
1.1381 + while (sr>>=1)
1.1382 + step++;
1.1383 +
1.1384 + *sp = MAX_MVSEARCH_STEPS - 1 - step;
1.1385 +}
1.1386 +
1.1387 +static int vp8_rd_pick_best_mbsegmentation(VP8_COMP *cpi, MACROBLOCK *x,
1.1388 + int_mv *best_ref_mv, int best_rd,
1.1389 + int *mdcounts, int *returntotrate,
1.1390 + int *returnyrate, int *returndistortion,
1.1391 + int mvthresh)
1.1392 +{
1.1393 + int i;
1.1394 + BEST_SEG_INFO bsi;
1.1395 +
1.1396 + vpx_memset(&bsi, 0, sizeof(bsi));
1.1397 +
1.1398 + bsi.segment_rd = best_rd;
1.1399 + bsi.ref_mv = best_ref_mv;
1.1400 + bsi.mvp.as_int = best_ref_mv->as_int;
1.1401 + bsi.mvthresh = mvthresh;
1.1402 + bsi.mdcounts = mdcounts;
1.1403 +
1.1404 + for(i = 0; i < 16; i++)
1.1405 + {
1.1406 + bsi.modes[i] = ZERO4X4;
1.1407 + }
1.1408 +
1.1409 + if(cpi->compressor_speed == 0)
1.1410 + {
1.1411 + /* for now, we will keep the original segmentation order
1.1412 + when in best quality mode */
1.1413 + rd_check_segment(cpi, x, &bsi, BLOCK_16X8);
1.1414 + rd_check_segment(cpi, x, &bsi, BLOCK_8X16);
1.1415 + rd_check_segment(cpi, x, &bsi, BLOCK_8X8);
1.1416 + rd_check_segment(cpi, x, &bsi, BLOCK_4X4);
1.1417 + }
1.1418 + else
1.1419 + {
1.1420 + int sr;
1.1421 +
1.1422 + rd_check_segment(cpi, x, &bsi, BLOCK_8X8);
1.1423 +
1.1424 + if (bsi.segment_rd < best_rd)
1.1425 + {
1.1426 + int col_min = ((best_ref_mv->as_mv.col+7)>>3) - MAX_FULL_PEL_VAL;
1.1427 + int row_min = ((best_ref_mv->as_mv.row+7)>>3) - MAX_FULL_PEL_VAL;
1.1428 + int col_max = (best_ref_mv->as_mv.col>>3) + MAX_FULL_PEL_VAL;
1.1429 + int row_max = (best_ref_mv->as_mv.row>>3) + MAX_FULL_PEL_VAL;
1.1430 +
1.1431 + int tmp_col_min = x->mv_col_min;
1.1432 + int tmp_col_max = x->mv_col_max;
1.1433 + int tmp_row_min = x->mv_row_min;
1.1434 + int tmp_row_max = x->mv_row_max;
1.1435 +
1.1436 + /* Get intersection of UMV window and valid MV window to reduce # of checks in diamond search. */
1.1437 + if (x->mv_col_min < col_min )
1.1438 + x->mv_col_min = col_min;
1.1439 + if (x->mv_col_max > col_max )
1.1440 + x->mv_col_max = col_max;
1.1441 + if (x->mv_row_min < row_min )
1.1442 + x->mv_row_min = row_min;
1.1443 + if (x->mv_row_max > row_max )
1.1444 + x->mv_row_max = row_max;
1.1445 +
1.1446 + /* Get 8x8 result */
1.1447 + bsi.sv_mvp[0].as_int = bsi.mvs[0].as_int;
1.1448 + bsi.sv_mvp[1].as_int = bsi.mvs[2].as_int;
1.1449 + bsi.sv_mvp[2].as_int = bsi.mvs[8].as_int;
1.1450 + bsi.sv_mvp[3].as_int = bsi.mvs[10].as_int;
1.1451 +
1.1452 + /* Use 8x8 result as 16x8/8x16's predictor MV. Adjust search range according to the closeness of 2 MV. */
1.1453 + /* block 8X16 */
1.1454 + {
1.1455 + sr = MAXF((abs(bsi.sv_mvp[0].as_mv.row - bsi.sv_mvp[2].as_mv.row))>>3, (abs(bsi.sv_mvp[0].as_mv.col - bsi.sv_mvp[2].as_mv.col))>>3);
1.1456 + vp8_cal_step_param(sr, &bsi.sv_istep[0]);
1.1457 +
1.1458 + sr = MAXF((abs(bsi.sv_mvp[1].as_mv.row - bsi.sv_mvp[3].as_mv.row))>>3, (abs(bsi.sv_mvp[1].as_mv.col - bsi.sv_mvp[3].as_mv.col))>>3);
1.1459 + vp8_cal_step_param(sr, &bsi.sv_istep[1]);
1.1460 +
1.1461 + rd_check_segment(cpi, x, &bsi, BLOCK_8X16);
1.1462 + }
1.1463 +
1.1464 + /* block 16X8 */
1.1465 + {
1.1466 + sr = MAXF((abs(bsi.sv_mvp[0].as_mv.row - bsi.sv_mvp[1].as_mv.row))>>3, (abs(bsi.sv_mvp[0].as_mv.col - bsi.sv_mvp[1].as_mv.col))>>3);
1.1467 + vp8_cal_step_param(sr, &bsi.sv_istep[0]);
1.1468 +
1.1469 + sr = MAXF((abs(bsi.sv_mvp[2].as_mv.row - bsi.sv_mvp[3].as_mv.row))>>3, (abs(bsi.sv_mvp[2].as_mv.col - bsi.sv_mvp[3].as_mv.col))>>3);
1.1470 + vp8_cal_step_param(sr, &bsi.sv_istep[1]);
1.1471 +
1.1472 + rd_check_segment(cpi, x, &bsi, BLOCK_16X8);
1.1473 + }
1.1474 +
1.1475 + /* If 8x8 is better than 16x8/8x16, then do 4x4 search */
1.1476 + /* Not skip 4x4 if speed=0 (good quality) */
1.1477 + if (cpi->sf.no_skip_block4x4_search || bsi.segment_num == BLOCK_8X8) /* || (sv_segment_rd8x8-bsi.segment_rd) < sv_segment_rd8x8>>5) */
1.1478 + {
1.1479 + bsi.mvp.as_int = bsi.sv_mvp[0].as_int;
1.1480 + rd_check_segment(cpi, x, &bsi, BLOCK_4X4);
1.1481 + }
1.1482 +
1.1483 + /* restore UMV window */
1.1484 + x->mv_col_min = tmp_col_min;
1.1485 + x->mv_col_max = tmp_col_max;
1.1486 + x->mv_row_min = tmp_row_min;
1.1487 + x->mv_row_max = tmp_row_max;
1.1488 + }
1.1489 + }
1.1490 +
1.1491 + /* set it to the best */
1.1492 + for (i = 0; i < 16; i++)
1.1493 + {
1.1494 + BLOCKD *bd = &x->e_mbd.block[i];
1.1495 +
1.1496 + bd->bmi.mv.as_int = bsi.mvs[i].as_int;
1.1497 + *bd->eob = bsi.eobs[i];
1.1498 + }
1.1499 +
1.1500 + *returntotrate = bsi.r;
1.1501 + *returndistortion = bsi.d;
1.1502 + *returnyrate = bsi.segment_yrate;
1.1503 +
1.1504 + /* save partitions */
1.1505 + x->e_mbd.mode_info_context->mbmi.partitioning = bsi.segment_num;
1.1506 + x->partition_info->count = vp8_mbsplit_count[bsi.segment_num];
1.1507 +
1.1508 + for (i = 0; i < x->partition_info->count; i++)
1.1509 + {
1.1510 + int j;
1.1511 +
1.1512 + j = vp8_mbsplit_offset[bsi.segment_num][i];
1.1513 +
1.1514 + x->partition_info->bmi[i].mode = bsi.modes[j];
1.1515 + x->partition_info->bmi[i].mv.as_mv = bsi.mvs[j].as_mv;
1.1516 + }
1.1517 + /*
1.1518 + * used to set x->e_mbd.mode_info_context->mbmi.mv.as_int
1.1519 + */
1.1520 + x->partition_info->bmi[15].mv.as_int = bsi.mvs[15].as_int;
1.1521 +
1.1522 + return bsi.segment_rd;
1.1523 +}
1.1524 +
1.1525 +/* The improved MV prediction */
1.1526 +void vp8_mv_pred
1.1527 +(
1.1528 + VP8_COMP *cpi,
1.1529 + MACROBLOCKD *xd,
1.1530 + const MODE_INFO *here,
1.1531 + int_mv *mvp,
1.1532 + int refframe,
1.1533 + int *ref_frame_sign_bias,
1.1534 + int *sr,
1.1535 + int near_sadidx[]
1.1536 +)
1.1537 +{
1.1538 + const MODE_INFO *above = here - xd->mode_info_stride;
1.1539 + const MODE_INFO *left = here - 1;
1.1540 + const MODE_INFO *aboveleft = above - 1;
1.1541 + int_mv near_mvs[8];
1.1542 + int near_ref[8];
1.1543 + int_mv mv;
1.1544 + int vcnt=0;
1.1545 + int find=0;
1.1546 + int mb_offset;
1.1547 +
1.1548 + int mvx[8];
1.1549 + int mvy[8];
1.1550 + int i;
1.1551 +
1.1552 + mv.as_int = 0;
1.1553 +
1.1554 + if(here->mbmi.ref_frame != INTRA_FRAME)
1.1555 + {
1.1556 + near_mvs[0].as_int = near_mvs[1].as_int = near_mvs[2].as_int = near_mvs[3].as_int = near_mvs[4].as_int = near_mvs[5].as_int = near_mvs[6].as_int = near_mvs[7].as_int = 0;
1.1557 + near_ref[0] = near_ref[1] = near_ref[2] = near_ref[3] = near_ref[4] = near_ref[5] = near_ref[6] = near_ref[7] = 0;
1.1558 +
1.1559 + /* read in 3 nearby block's MVs from current frame as prediction
1.1560 + * candidates.
1.1561 + */
1.1562 + if (above->mbmi.ref_frame != INTRA_FRAME)
1.1563 + {
1.1564 + near_mvs[vcnt].as_int = above->mbmi.mv.as_int;
1.1565 + mv_bias(ref_frame_sign_bias[above->mbmi.ref_frame], refframe, &near_mvs[vcnt], ref_frame_sign_bias);
1.1566 + near_ref[vcnt] = above->mbmi.ref_frame;
1.1567 + }
1.1568 + vcnt++;
1.1569 + if (left->mbmi.ref_frame != INTRA_FRAME)
1.1570 + {
1.1571 + near_mvs[vcnt].as_int = left->mbmi.mv.as_int;
1.1572 + mv_bias(ref_frame_sign_bias[left->mbmi.ref_frame], refframe, &near_mvs[vcnt], ref_frame_sign_bias);
1.1573 + near_ref[vcnt] = left->mbmi.ref_frame;
1.1574 + }
1.1575 + vcnt++;
1.1576 + if (aboveleft->mbmi.ref_frame != INTRA_FRAME)
1.1577 + {
1.1578 + near_mvs[vcnt].as_int = aboveleft->mbmi.mv.as_int;
1.1579 + mv_bias(ref_frame_sign_bias[aboveleft->mbmi.ref_frame], refframe, &near_mvs[vcnt], ref_frame_sign_bias);
1.1580 + near_ref[vcnt] = aboveleft->mbmi.ref_frame;
1.1581 + }
1.1582 + vcnt++;
1.1583 +
1.1584 + /* read in 5 nearby block's MVs from last frame. */
1.1585 + if(cpi->common.last_frame_type != KEY_FRAME)
1.1586 + {
1.1587 + mb_offset = (-xd->mb_to_top_edge/128 + 1) * (xd->mode_info_stride +1) + (-xd->mb_to_left_edge/128 +1) ;
1.1588 +
1.1589 + /* current in last frame */
1.1590 + if (cpi->lf_ref_frame[mb_offset] != INTRA_FRAME)
1.1591 + {
1.1592 + near_mvs[vcnt].as_int = cpi->lfmv[mb_offset].as_int;
1.1593 + mv_bias(cpi->lf_ref_frame_sign_bias[mb_offset], refframe, &near_mvs[vcnt], ref_frame_sign_bias);
1.1594 + near_ref[vcnt] = cpi->lf_ref_frame[mb_offset];
1.1595 + }
1.1596 + vcnt++;
1.1597 +
1.1598 + /* above in last frame */
1.1599 + if (cpi->lf_ref_frame[mb_offset - xd->mode_info_stride-1] != INTRA_FRAME)
1.1600 + {
1.1601 + near_mvs[vcnt].as_int = cpi->lfmv[mb_offset - xd->mode_info_stride-1].as_int;
1.1602 + mv_bias(cpi->lf_ref_frame_sign_bias[mb_offset - xd->mode_info_stride-1], refframe, &near_mvs[vcnt], ref_frame_sign_bias);
1.1603 + near_ref[vcnt] = cpi->lf_ref_frame[mb_offset - xd->mode_info_stride-1];
1.1604 + }
1.1605 + vcnt++;
1.1606 +
1.1607 + /* left in last frame */
1.1608 + if (cpi->lf_ref_frame[mb_offset-1] != INTRA_FRAME)
1.1609 + {
1.1610 + near_mvs[vcnt].as_int = cpi->lfmv[mb_offset -1].as_int;
1.1611 + mv_bias(cpi->lf_ref_frame_sign_bias[mb_offset -1], refframe, &near_mvs[vcnt], ref_frame_sign_bias);
1.1612 + near_ref[vcnt] = cpi->lf_ref_frame[mb_offset - 1];
1.1613 + }
1.1614 + vcnt++;
1.1615 +
1.1616 + /* right in last frame */
1.1617 + if (cpi->lf_ref_frame[mb_offset +1] != INTRA_FRAME)
1.1618 + {
1.1619 + near_mvs[vcnt].as_int = cpi->lfmv[mb_offset +1].as_int;
1.1620 + mv_bias(cpi->lf_ref_frame_sign_bias[mb_offset +1], refframe, &near_mvs[vcnt], ref_frame_sign_bias);
1.1621 + near_ref[vcnt] = cpi->lf_ref_frame[mb_offset +1];
1.1622 + }
1.1623 + vcnt++;
1.1624 +
1.1625 + /* below in last frame */
1.1626 + if (cpi->lf_ref_frame[mb_offset + xd->mode_info_stride +1] != INTRA_FRAME)
1.1627 + {
1.1628 + near_mvs[vcnt].as_int = cpi->lfmv[mb_offset + xd->mode_info_stride +1].as_int;
1.1629 + mv_bias(cpi->lf_ref_frame_sign_bias[mb_offset + xd->mode_info_stride +1], refframe, &near_mvs[vcnt], ref_frame_sign_bias);
1.1630 + near_ref[vcnt] = cpi->lf_ref_frame[mb_offset + xd->mode_info_stride +1];
1.1631 + }
1.1632 + vcnt++;
1.1633 + }
1.1634 +
1.1635 + for(i=0; i< vcnt; i++)
1.1636 + {
1.1637 + if(near_ref[near_sadidx[i]] != INTRA_FRAME)
1.1638 + {
1.1639 + if(here->mbmi.ref_frame == near_ref[near_sadidx[i]])
1.1640 + {
1.1641 + mv.as_int = near_mvs[near_sadidx[i]].as_int;
1.1642 + find = 1;
1.1643 + if (i < 3)
1.1644 + *sr = 3;
1.1645 + else
1.1646 + *sr = 2;
1.1647 + break;
1.1648 + }
1.1649 + }
1.1650 + }
1.1651 +
1.1652 + if(!find)
1.1653 + {
1.1654 + for(i=0; i<vcnt; i++)
1.1655 + {
1.1656 + mvx[i] = near_mvs[i].as_mv.row;
1.1657 + mvy[i] = near_mvs[i].as_mv.col;
1.1658 + }
1.1659 +
1.1660 + insertsortmv(mvx, vcnt);
1.1661 + insertsortmv(mvy, vcnt);
1.1662 + mv.as_mv.row = mvx[vcnt/2];
1.1663 + mv.as_mv.col = mvy[vcnt/2];
1.1664 +
1.1665 + find = 1;
1.1666 + /* sr is set to 0 to allow calling function to decide the search
1.1667 + * range.
1.1668 + */
1.1669 + *sr = 0;
1.1670 + }
1.1671 + }
1.1672 +
1.1673 + /* Set up return values */
1.1674 + mvp->as_int = mv.as_int;
1.1675 + vp8_clamp_mv2(mvp, xd);
1.1676 +}
1.1677 +
1.1678 +void vp8_cal_sad(VP8_COMP *cpi, MACROBLOCKD *xd, MACROBLOCK *x, int recon_yoffset, int near_sadidx[])
1.1679 +{
1.1680 + /* near_sad indexes:
1.1681 + * 0-cf above, 1-cf left, 2-cf aboveleft,
1.1682 + * 3-lf current, 4-lf above, 5-lf left, 6-lf right, 7-lf below
1.1683 + */
1.1684 + int near_sad[8] = {0};
1.1685 + BLOCK *b = &x->block[0];
1.1686 + unsigned char *src_y_ptr = *(b->base_src);
1.1687 +
1.1688 + /* calculate sad for current frame 3 nearby MBs. */
1.1689 + if( xd->mb_to_top_edge==0 && xd->mb_to_left_edge ==0)
1.1690 + {
1.1691 + near_sad[0] = near_sad[1] = near_sad[2] = INT_MAX;
1.1692 + }else if(xd->mb_to_top_edge==0)
1.1693 + { /* only has left MB for sad calculation. */
1.1694 + near_sad[0] = near_sad[2] = INT_MAX;
1.1695 + near_sad[1] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, xd->dst.y_buffer - 16,xd->dst.y_stride, UINT_MAX);
1.1696 + }else if(xd->mb_to_left_edge ==0)
1.1697 + { /* only has left MB for sad calculation. */
1.1698 + near_sad[1] = near_sad[2] = INT_MAX;
1.1699 + near_sad[0] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, xd->dst.y_buffer - xd->dst.y_stride *16,xd->dst.y_stride, UINT_MAX);
1.1700 + }else
1.1701 + {
1.1702 + near_sad[0] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, xd->dst.y_buffer - xd->dst.y_stride *16,xd->dst.y_stride, UINT_MAX);
1.1703 + near_sad[1] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, xd->dst.y_buffer - 16,xd->dst.y_stride, UINT_MAX);
1.1704 + near_sad[2] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, xd->dst.y_buffer - xd->dst.y_stride *16 -16,xd->dst.y_stride, UINT_MAX);
1.1705 + }
1.1706 +
1.1707 + if(cpi->common.last_frame_type != KEY_FRAME)
1.1708 + {
1.1709 + /* calculate sad for last frame 5 nearby MBs. */
1.1710 + unsigned char *pre_y_buffer = cpi->common.yv12_fb[cpi->common.lst_fb_idx].y_buffer + recon_yoffset;
1.1711 + int pre_y_stride = cpi->common.yv12_fb[cpi->common.lst_fb_idx].y_stride;
1.1712 +
1.1713 + if(xd->mb_to_top_edge==0) near_sad[4] = INT_MAX;
1.1714 + if(xd->mb_to_left_edge ==0) near_sad[5] = INT_MAX;
1.1715 + if(xd->mb_to_right_edge ==0) near_sad[6] = INT_MAX;
1.1716 + if(xd->mb_to_bottom_edge==0) near_sad[7] = INT_MAX;
1.1717 +
1.1718 + if(near_sad[4] != INT_MAX)
1.1719 + near_sad[4] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, pre_y_buffer - pre_y_stride *16, pre_y_stride, UINT_MAX);
1.1720 + if(near_sad[5] != INT_MAX)
1.1721 + near_sad[5] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, pre_y_buffer - 16, pre_y_stride, UINT_MAX);
1.1722 + near_sad[3] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, pre_y_buffer, pre_y_stride, UINT_MAX);
1.1723 + if(near_sad[6] != INT_MAX)
1.1724 + near_sad[6] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, pre_y_buffer + 16, pre_y_stride, UINT_MAX);
1.1725 + if(near_sad[7] != INT_MAX)
1.1726 + near_sad[7] = cpi->fn_ptr[BLOCK_16X16].sdf(src_y_ptr, b->src_stride, pre_y_buffer + pre_y_stride *16, pre_y_stride, UINT_MAX);
1.1727 + }
1.1728 +
1.1729 + if(cpi->common.last_frame_type != KEY_FRAME)
1.1730 + {
1.1731 + insertsortsad(near_sad, near_sadidx, 8);
1.1732 + }else
1.1733 + {
1.1734 + insertsortsad(near_sad, near_sadidx, 3);
1.1735 + }
1.1736 +}
1.1737 +
1.1738 +static void rd_update_mvcount(MACROBLOCK *x, int_mv *best_ref_mv)
1.1739 +{
1.1740 + if (x->e_mbd.mode_info_context->mbmi.mode == SPLITMV)
1.1741 + {
1.1742 + int i;
1.1743 +
1.1744 + for (i = 0; i < x->partition_info->count; i++)
1.1745 + {
1.1746 + if (x->partition_info->bmi[i].mode == NEW4X4)
1.1747 + {
1.1748 + x->MVcount[0][mv_max+((x->partition_info->bmi[i].mv.as_mv.row
1.1749 + - best_ref_mv->as_mv.row) >> 1)]++;
1.1750 + x->MVcount[1][mv_max+((x->partition_info->bmi[i].mv.as_mv.col
1.1751 + - best_ref_mv->as_mv.col) >> 1)]++;
1.1752 + }
1.1753 + }
1.1754 + }
1.1755 + else if (x->e_mbd.mode_info_context->mbmi.mode == NEWMV)
1.1756 + {
1.1757 + x->MVcount[0][mv_max+((x->e_mbd.mode_info_context->mbmi.mv.as_mv.row
1.1758 + - best_ref_mv->as_mv.row) >> 1)]++;
1.1759 + x->MVcount[1][mv_max+((x->e_mbd.mode_info_context->mbmi.mv.as_mv.col
1.1760 + - best_ref_mv->as_mv.col) >> 1)]++;
1.1761 + }
1.1762 +}
1.1763 +
1.1764 +static int evaluate_inter_mode_rd(int mdcounts[4],
1.1765 + RATE_DISTORTION* rd,
1.1766 + int* disable_skip,
1.1767 + VP8_COMP *cpi, MACROBLOCK *x)
1.1768 +{
1.1769 + MB_PREDICTION_MODE this_mode = x->e_mbd.mode_info_context->mbmi.mode;
1.1770 + BLOCK *b = &x->block[0];
1.1771 + MACROBLOCKD *xd = &x->e_mbd;
1.1772 + int distortion;
1.1773 + vp8_build_inter16x16_predictors_mby(&x->e_mbd, x->e_mbd.predictor, 16);
1.1774 +
1.1775 + if (cpi->active_map_enabled && x->active_ptr[0] == 0) {
1.1776 + x->skip = 1;
1.1777 + }
1.1778 + else if (x->encode_breakout)
1.1779 + {
1.1780 + unsigned int sse;
1.1781 + unsigned int var;
1.1782 + unsigned int threshold = (xd->block[0].dequant[1]
1.1783 + * xd->block[0].dequant[1] >>4);
1.1784 +
1.1785 + if(threshold < x->encode_breakout)
1.1786 + threshold = x->encode_breakout;
1.1787 +
1.1788 + var = vp8_variance16x16
1.1789 + (*(b->base_src), b->src_stride,
1.1790 + x->e_mbd.predictor, 16, &sse);
1.1791 +
1.1792 + if (sse < threshold)
1.1793 + {
1.1794 + unsigned int q2dc = xd->block[24].dequant[0];
1.1795 + /* If theres is no codeable 2nd order dc
1.1796 + or a very small uniform pixel change change */
1.1797 + if ((sse - var < q2dc * q2dc >>4) ||
1.1798 + (sse /2 > var && sse-var < 64))
1.1799 + {
1.1800 + /* Check u and v to make sure skip is ok */
1.1801 + unsigned int sse2 = VP8_UVSSE(x);
1.1802 + if (sse2 * 2 < threshold)
1.1803 + {
1.1804 + x->skip = 1;
1.1805 + rd->distortion2 = sse + sse2;
1.1806 + rd->rate2 = 500;
1.1807 +
1.1808 + /* for best_yrd calculation */
1.1809 + rd->rate_uv = 0;
1.1810 + rd->distortion_uv = sse2;
1.1811 +
1.1812 + *disable_skip = 1;
1.1813 + return RDCOST(x->rdmult, x->rddiv, rd->rate2,
1.1814 + rd->distortion2);
1.1815 + }
1.1816 + }
1.1817 + }
1.1818 + }
1.1819 +
1.1820 +
1.1821 + /* Add in the Mv/mode cost */
1.1822 + rd->rate2 += vp8_cost_mv_ref(this_mode, mdcounts);
1.1823 +
1.1824 + /* Y cost and distortion */
1.1825 + macro_block_yrd(x, &rd->rate_y, &distortion);
1.1826 + rd->rate2 += rd->rate_y;
1.1827 + rd->distortion2 += distortion;
1.1828 +
1.1829 + /* UV cost and distortion */
1.1830 + rd_inter16x16_uv(cpi, x, &rd->rate_uv, &rd->distortion_uv,
1.1831 + cpi->common.full_pixel);
1.1832 + rd->rate2 += rd->rate_uv;
1.1833 + rd->distortion2 += rd->distortion_uv;
1.1834 + return INT_MAX;
1.1835 +}
1.1836 +
1.1837 +static int calculate_final_rd_costs(int this_rd,
1.1838 + RATE_DISTORTION* rd,
1.1839 + int* other_cost,
1.1840 + int disable_skip,
1.1841 + int uv_intra_tteob,
1.1842 + int intra_rd_penalty,
1.1843 + VP8_COMP *cpi, MACROBLOCK *x)
1.1844 +{
1.1845 + MB_PREDICTION_MODE this_mode = x->e_mbd.mode_info_context->mbmi.mode;
1.1846 +
1.1847 + /* Where skip is allowable add in the default per mb cost for the no
1.1848 + * skip case. where we then decide to skip we have to delete this and
1.1849 + * replace it with the cost of signalling a skip
1.1850 + */
1.1851 + if (cpi->common.mb_no_coeff_skip)
1.1852 + {
1.1853 + *other_cost += vp8_cost_bit(cpi->prob_skip_false, 0);
1.1854 + rd->rate2 += *other_cost;
1.1855 + }
1.1856 +
1.1857 + /* Estimate the reference frame signaling cost and add it
1.1858 + * to the rolling cost variable.
1.1859 + */
1.1860 + rd->rate2 +=
1.1861 + x->ref_frame_cost[x->e_mbd.mode_info_context->mbmi.ref_frame];
1.1862 +
1.1863 + if (!disable_skip)
1.1864 + {
1.1865 + /* Test for the condition where skip block will be activated
1.1866 + * because there are no non zero coefficients and make any
1.1867 + * necessary adjustment for rate
1.1868 + */
1.1869 + if (cpi->common.mb_no_coeff_skip)
1.1870 + {
1.1871 + int i;
1.1872 + int tteob;
1.1873 + int has_y2_block = (this_mode!=SPLITMV && this_mode!=B_PRED);
1.1874 +
1.1875 + tteob = 0;
1.1876 + if(has_y2_block)
1.1877 + tteob += x->e_mbd.eobs[24];
1.1878 +
1.1879 + for (i = 0; i < 16; i++)
1.1880 + tteob += (x->e_mbd.eobs[i] > has_y2_block);
1.1881 +
1.1882 + if (x->e_mbd.mode_info_context->mbmi.ref_frame)
1.1883 + {
1.1884 + for (i = 16; i < 24; i++)
1.1885 + tteob += x->e_mbd.eobs[i];
1.1886 + }
1.1887 + else
1.1888 + tteob += uv_intra_tteob;
1.1889 +
1.1890 + if (tteob == 0)
1.1891 + {
1.1892 + rd->rate2 -= (rd->rate_y + rd->rate_uv);
1.1893 + /* for best_yrd calculation */
1.1894 + rd->rate_uv = 0;
1.1895 +
1.1896 + /* Back out no skip flag costing and add in skip flag costing */
1.1897 + if (cpi->prob_skip_false)
1.1898 + {
1.1899 + int prob_skip_cost;
1.1900 +
1.1901 + prob_skip_cost = vp8_cost_bit(cpi->prob_skip_false, 1);
1.1902 + prob_skip_cost -= vp8_cost_bit(cpi->prob_skip_false, 0);
1.1903 + rd->rate2 += prob_skip_cost;
1.1904 + *other_cost += prob_skip_cost;
1.1905 + }
1.1906 + }
1.1907 + }
1.1908 + /* Calculate the final RD estimate for this mode */
1.1909 + this_rd = RDCOST(x->rdmult, x->rddiv, rd->rate2, rd->distortion2);
1.1910 + if (this_rd < INT_MAX && x->e_mbd.mode_info_context->mbmi.ref_frame
1.1911 + == INTRA_FRAME)
1.1912 + this_rd += intra_rd_penalty;
1.1913 + }
1.1914 + return this_rd;
1.1915 +}
1.1916 +
1.1917 +static void update_best_mode(BEST_MODE* best_mode, int this_rd,
1.1918 + RATE_DISTORTION* rd, int other_cost, MACROBLOCK *x)
1.1919 +{
1.1920 + MB_PREDICTION_MODE this_mode = x->e_mbd.mode_info_context->mbmi.mode;
1.1921 +
1.1922 + other_cost +=
1.1923 + x->ref_frame_cost[x->e_mbd.mode_info_context->mbmi.ref_frame];
1.1924 +
1.1925 + /* Calculate the final y RD estimate for this mode */
1.1926 + best_mode->yrd = RDCOST(x->rdmult, x->rddiv, (rd->rate2-rd->rate_uv-other_cost),
1.1927 + (rd->distortion2-rd->distortion_uv));
1.1928 +
1.1929 + best_mode->rd = this_rd;
1.1930 + vpx_memcpy(&best_mode->mbmode, &x->e_mbd.mode_info_context->mbmi, sizeof(MB_MODE_INFO));
1.1931 + vpx_memcpy(&best_mode->partition, x->partition_info, sizeof(PARTITION_INFO));
1.1932 +
1.1933 + if ((this_mode == B_PRED) || (this_mode == SPLITMV))
1.1934 + {
1.1935 + int i;
1.1936 + for (i = 0; i < 16; i++)
1.1937 + {
1.1938 + best_mode->bmodes[i] = x->e_mbd.block[i].bmi;
1.1939 + }
1.1940 + }
1.1941 +}
1.1942 +
1.1943 +void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
1.1944 + int recon_uvoffset, int *returnrate,
1.1945 + int *returndistortion, int *returnintra)
1.1946 +{
1.1947 + BLOCK *b = &x->block[0];
1.1948 + BLOCKD *d = &x->e_mbd.block[0];
1.1949 + MACROBLOCKD *xd = &x->e_mbd;
1.1950 + int_mv best_ref_mv_sb[2];
1.1951 + int_mv mode_mv_sb[2][MB_MODE_COUNT];
1.1952 + int_mv best_ref_mv;
1.1953 + int_mv *mode_mv;
1.1954 + MB_PREDICTION_MODE this_mode;
1.1955 + int num00;
1.1956 + int best_mode_index = 0;
1.1957 + BEST_MODE best_mode;
1.1958 +
1.1959 + int i;
1.1960 + int mode_index;
1.1961 + int mdcounts[4];
1.1962 + int rate;
1.1963 + RATE_DISTORTION rd;
1.1964 + int uv_intra_rate, uv_intra_distortion, uv_intra_rate_tokenonly;
1.1965 + int uv_intra_tteob = 0;
1.1966 + int uv_intra_done = 0;
1.1967 +
1.1968 + MB_PREDICTION_MODE uv_intra_mode = 0;
1.1969 + int_mv mvp;
1.1970 + int near_sadidx[8] = {0, 1, 2, 3, 4, 5, 6, 7};
1.1971 + int saddone=0;
1.1972 + /* search range got from mv_pred(). It uses step_param levels. (0-7) */
1.1973 + int sr=0;
1.1974 +
1.1975 + unsigned char *plane[4][3];
1.1976 + int ref_frame_map[4];
1.1977 + int sign_bias = 0;
1.1978 +
1.1979 + int intra_rd_penalty = 10* vp8_dc_quant(cpi->common.base_qindex,
1.1980 + cpi->common.y1dc_delta_q);
1.1981 +
1.1982 +#if CONFIG_TEMPORAL_DENOISING
1.1983 + unsigned int zero_mv_sse = INT_MAX, best_sse = INT_MAX,
1.1984 + best_rd_sse = INT_MAX;
1.1985 +#endif
1.1986 +
1.1987 + mode_mv = mode_mv_sb[sign_bias];
1.1988 + best_ref_mv.as_int = 0;
1.1989 + best_mode.rd = INT_MAX;
1.1990 + best_mode.yrd = INT_MAX;
1.1991 + best_mode.intra_rd = INT_MAX;
1.1992 + vpx_memset(mode_mv_sb, 0, sizeof(mode_mv_sb));
1.1993 + vpx_memset(&best_mode.mbmode, 0, sizeof(best_mode.mbmode));
1.1994 + vpx_memset(&best_mode.bmodes, 0, sizeof(best_mode.bmodes));
1.1995 +
1.1996 + /* Setup search priorities */
1.1997 + get_reference_search_order(cpi, ref_frame_map);
1.1998 +
1.1999 + /* Check to see if there is at least 1 valid reference frame that we need
1.2000 + * to calculate near_mvs.
1.2001 + */
1.2002 + if (ref_frame_map[1] > 0)
1.2003 + {
1.2004 + sign_bias = vp8_find_near_mvs_bias(&x->e_mbd,
1.2005 + x->e_mbd.mode_info_context,
1.2006 + mode_mv_sb,
1.2007 + best_ref_mv_sb,
1.2008 + mdcounts,
1.2009 + ref_frame_map[1],
1.2010 + cpi->common.ref_frame_sign_bias);
1.2011 +
1.2012 + mode_mv = mode_mv_sb[sign_bias];
1.2013 + best_ref_mv.as_int = best_ref_mv_sb[sign_bias].as_int;
1.2014 + }
1.2015 +
1.2016 + get_predictor_pointers(cpi, plane, recon_yoffset, recon_uvoffset);
1.2017 +
1.2018 + *returnintra = INT_MAX;
1.2019 + /* Count of the number of MBs tested so far this frame */
1.2020 + x->mbs_tested_so_far++;
1.2021 +
1.2022 + x->skip = 0;
1.2023 +
1.2024 + for (mode_index = 0; mode_index < MAX_MODES; mode_index++)
1.2025 + {
1.2026 + int this_rd = INT_MAX;
1.2027 + int disable_skip = 0;
1.2028 + int other_cost = 0;
1.2029 + int this_ref_frame = ref_frame_map[vp8_ref_frame_order[mode_index]];
1.2030 +
1.2031 + /* Test best rd so far against threshold for trying this mode. */
1.2032 + if (best_mode.rd <= x->rd_threshes[mode_index])
1.2033 + continue;
1.2034 +
1.2035 + if (this_ref_frame < 0)
1.2036 + continue;
1.2037 +
1.2038 + /* These variables hold are rolling total cost and distortion for
1.2039 + * this mode
1.2040 + */
1.2041 + rd.rate2 = 0;
1.2042 + rd.distortion2 = 0;
1.2043 +
1.2044 + this_mode = vp8_mode_order[mode_index];
1.2045 +
1.2046 + x->e_mbd.mode_info_context->mbmi.mode = this_mode;
1.2047 + x->e_mbd.mode_info_context->mbmi.ref_frame = this_ref_frame;
1.2048 +
1.2049 + /* Only consider ZEROMV/ALTREF_FRAME for alt ref frame,
1.2050 + * unless ARNR filtering is enabled in which case we want
1.2051 + * an unfiltered alternative
1.2052 + */
1.2053 + if (cpi->is_src_frame_alt_ref && (cpi->oxcf.arnr_max_frames == 0))
1.2054 + {
1.2055 + if (this_mode != ZEROMV || x->e_mbd.mode_info_context->mbmi.ref_frame != ALTREF_FRAME)
1.2056 + continue;
1.2057 + }
1.2058 +
1.2059 + /* everything but intra */
1.2060 + if (x->e_mbd.mode_info_context->mbmi.ref_frame)
1.2061 + {
1.2062 + x->e_mbd.pre.y_buffer = plane[this_ref_frame][0];
1.2063 + x->e_mbd.pre.u_buffer = plane[this_ref_frame][1];
1.2064 + x->e_mbd.pre.v_buffer = plane[this_ref_frame][2];
1.2065 +
1.2066 + if (sign_bias != cpi->common.ref_frame_sign_bias[this_ref_frame])
1.2067 + {
1.2068 + sign_bias = cpi->common.ref_frame_sign_bias[this_ref_frame];
1.2069 + mode_mv = mode_mv_sb[sign_bias];
1.2070 + best_ref_mv.as_int = best_ref_mv_sb[sign_bias].as_int;
1.2071 + }
1.2072 + }
1.2073 +
1.2074 + /* Check to see if the testing frequency for this mode is at its
1.2075 + * max If so then prevent it from being tested and increase the
1.2076 + * threshold for its testing
1.2077 + */
1.2078 + if (x->mode_test_hit_counts[mode_index] && (cpi->mode_check_freq[mode_index] > 1))
1.2079 + {
1.2080 + if (x->mbs_tested_so_far <= cpi->mode_check_freq[mode_index] * x->mode_test_hit_counts[mode_index])
1.2081 + {
1.2082 + /* Increase the threshold for coding this mode to make it
1.2083 + * less likely to be chosen
1.2084 + */
1.2085 + x->rd_thresh_mult[mode_index] += 4;
1.2086 +
1.2087 + if (x->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
1.2088 + x->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
1.2089 +
1.2090 + x->rd_threshes[mode_index] =
1.2091 + (cpi->rd_baseline_thresh[mode_index] >> 7) *
1.2092 + x->rd_thresh_mult[mode_index];
1.2093 +
1.2094 + continue;
1.2095 + }
1.2096 + }
1.2097 +
1.2098 + /* We have now reached the point where we are going to test the
1.2099 + * current mode so increment the counter for the number of times
1.2100 + * it has been tested
1.2101 + */
1.2102 + x->mode_test_hit_counts[mode_index] ++;
1.2103 +
1.2104 + /* Experimental code. Special case for gf and arf zeromv modes.
1.2105 + * Increase zbin size to supress noise
1.2106 + */
1.2107 + if (x->zbin_mode_boost_enabled)
1.2108 + {
1.2109 + if ( this_ref_frame == INTRA_FRAME )
1.2110 + x->zbin_mode_boost = 0;
1.2111 + else
1.2112 + {
1.2113 + if (vp8_mode_order[mode_index] == ZEROMV)
1.2114 + {
1.2115 + if (this_ref_frame != LAST_FRAME)
1.2116 + x->zbin_mode_boost = GF_ZEROMV_ZBIN_BOOST;
1.2117 + else
1.2118 + x->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST;
1.2119 + }
1.2120 + else if (vp8_mode_order[mode_index] == SPLITMV)
1.2121 + x->zbin_mode_boost = 0;
1.2122 + else
1.2123 + x->zbin_mode_boost = MV_ZBIN_BOOST;
1.2124 + }
1.2125 +
1.2126 + vp8_update_zbin_extra(cpi, x);
1.2127 + }
1.2128 +
1.2129 + if(!uv_intra_done && this_ref_frame == INTRA_FRAME)
1.2130 + {
1.2131 + rd_pick_intra_mbuv_mode(x, &uv_intra_rate,
1.2132 + &uv_intra_rate_tokenonly,
1.2133 + &uv_intra_distortion);
1.2134 + uv_intra_mode = x->e_mbd.mode_info_context->mbmi.uv_mode;
1.2135 +
1.2136 + /*
1.2137 + * Total of the eobs is used later to further adjust rate2. Since uv
1.2138 + * block's intra eobs will be overwritten when we check inter modes,
1.2139 + * we need to save uv_intra_tteob here.
1.2140 + */
1.2141 + for (i = 16; i < 24; i++)
1.2142 + uv_intra_tteob += x->e_mbd.eobs[i];
1.2143 +
1.2144 + uv_intra_done = 1;
1.2145 + }
1.2146 +
1.2147 + switch (this_mode)
1.2148 + {
1.2149 + case B_PRED:
1.2150 + {
1.2151 + int tmp_rd;
1.2152 +
1.2153 + /* Note the rate value returned here includes the cost of
1.2154 + * coding the BPRED mode: x->mbmode_cost[x->e_mbd.frame_type][BPRED]
1.2155 + */
1.2156 + int distortion;
1.2157 + tmp_rd = rd_pick_intra4x4mby_modes(x, &rate, &rd.rate_y, &distortion, best_mode.yrd);
1.2158 + rd.rate2 += rate;
1.2159 + rd.distortion2 += distortion;
1.2160 +
1.2161 + if(tmp_rd < best_mode.yrd)
1.2162 + {
1.2163 + rd.rate2 += uv_intra_rate;
1.2164 + rd.rate_uv = uv_intra_rate_tokenonly;
1.2165 + rd.distortion2 += uv_intra_distortion;
1.2166 + rd.distortion_uv = uv_intra_distortion;
1.2167 + }
1.2168 + else
1.2169 + {
1.2170 + this_rd = INT_MAX;
1.2171 + disable_skip = 1;
1.2172 + }
1.2173 + }
1.2174 + break;
1.2175 +
1.2176 + case SPLITMV:
1.2177 + {
1.2178 + int tmp_rd;
1.2179 + int this_rd_thresh;
1.2180 + int distortion;
1.2181 +
1.2182 + this_rd_thresh = (vp8_ref_frame_order[mode_index] == 1) ?
1.2183 + x->rd_threshes[THR_NEW1] : x->rd_threshes[THR_NEW3];
1.2184 + this_rd_thresh = (vp8_ref_frame_order[mode_index] == 2) ?
1.2185 + x->rd_threshes[THR_NEW2] : this_rd_thresh;
1.2186 +
1.2187 + tmp_rd = vp8_rd_pick_best_mbsegmentation(cpi, x, &best_ref_mv,
1.2188 + best_mode.yrd, mdcounts,
1.2189 + &rate, &rd.rate_y, &distortion, this_rd_thresh) ;
1.2190 +
1.2191 + rd.rate2 += rate;
1.2192 + rd.distortion2 += distortion;
1.2193 +
1.2194 + /* If even the 'Y' rd value of split is higher than best so far
1.2195 + * then dont bother looking at UV
1.2196 + */
1.2197 + if (tmp_rd < best_mode.yrd)
1.2198 + {
1.2199 + /* Now work out UV cost and add it in */
1.2200 + rd_inter4x4_uv(cpi, x, &rd.rate_uv, &rd.distortion_uv, cpi->common.full_pixel);
1.2201 + rd.rate2 += rd.rate_uv;
1.2202 + rd.distortion2 += rd.distortion_uv;
1.2203 + }
1.2204 + else
1.2205 + {
1.2206 + this_rd = INT_MAX;
1.2207 + disable_skip = 1;
1.2208 + }
1.2209 + }
1.2210 + break;
1.2211 + case DC_PRED:
1.2212 + case V_PRED:
1.2213 + case H_PRED:
1.2214 + case TM_PRED:
1.2215 + {
1.2216 + int distortion;
1.2217 + x->e_mbd.mode_info_context->mbmi.ref_frame = INTRA_FRAME;
1.2218 +
1.2219 + vp8_build_intra_predictors_mby_s(xd,
1.2220 + xd->dst.y_buffer - xd->dst.y_stride,
1.2221 + xd->dst.y_buffer - 1,
1.2222 + xd->dst.y_stride,
1.2223 + xd->predictor,
1.2224 + 16);
1.2225 + macro_block_yrd(x, &rd.rate_y, &distortion) ;
1.2226 + rd.rate2 += rd.rate_y;
1.2227 + rd.distortion2 += distortion;
1.2228 + rd.rate2 += x->mbmode_cost[x->e_mbd.frame_type][x->e_mbd.mode_info_context->mbmi.mode];
1.2229 + rd.rate2 += uv_intra_rate;
1.2230 + rd.rate_uv = uv_intra_rate_tokenonly;
1.2231 + rd.distortion2 += uv_intra_distortion;
1.2232 + rd.distortion_uv = uv_intra_distortion;
1.2233 + }
1.2234 + break;
1.2235 +
1.2236 + case NEWMV:
1.2237 + {
1.2238 + int thissme;
1.2239 + int bestsme = INT_MAX;
1.2240 + int step_param = cpi->sf.first_step;
1.2241 + int further_steps;
1.2242 + int n;
1.2243 + int do_refine=1; /* If last step (1-away) of n-step search doesn't pick the center point as the best match,
1.2244 + we will do a final 1-away diamond refining search */
1.2245 +
1.2246 + int sadpb = x->sadperbit16;
1.2247 + int_mv mvp_full;
1.2248 +
1.2249 + int col_min = ((best_ref_mv.as_mv.col+7)>>3) - MAX_FULL_PEL_VAL;
1.2250 + int row_min = ((best_ref_mv.as_mv.row+7)>>3) - MAX_FULL_PEL_VAL;
1.2251 + int col_max = (best_ref_mv.as_mv.col>>3) + MAX_FULL_PEL_VAL;
1.2252 + int row_max = (best_ref_mv.as_mv.row>>3) + MAX_FULL_PEL_VAL;
1.2253 +
1.2254 + int tmp_col_min = x->mv_col_min;
1.2255 + int tmp_col_max = x->mv_col_max;
1.2256 + int tmp_row_min = x->mv_row_min;
1.2257 + int tmp_row_max = x->mv_row_max;
1.2258 +
1.2259 + if(!saddone)
1.2260 + {
1.2261 + vp8_cal_sad(cpi,xd,x, recon_yoffset ,&near_sadidx[0] );
1.2262 + saddone = 1;
1.2263 + }
1.2264 +
1.2265 + vp8_mv_pred(cpi, &x->e_mbd, x->e_mbd.mode_info_context, &mvp,
1.2266 + x->e_mbd.mode_info_context->mbmi.ref_frame, cpi->common.ref_frame_sign_bias, &sr, &near_sadidx[0]);
1.2267 +
1.2268 + mvp_full.as_mv.col = mvp.as_mv.col>>3;
1.2269 + mvp_full.as_mv.row = mvp.as_mv.row>>3;
1.2270 +
1.2271 + /* Get intersection of UMV window and valid MV window to
1.2272 + * reduce # of checks in diamond search.
1.2273 + */
1.2274 + if (x->mv_col_min < col_min )
1.2275 + x->mv_col_min = col_min;
1.2276 + if (x->mv_col_max > col_max )
1.2277 + x->mv_col_max = col_max;
1.2278 + if (x->mv_row_min < row_min )
1.2279 + x->mv_row_min = row_min;
1.2280 + if (x->mv_row_max > row_max )
1.2281 + x->mv_row_max = row_max;
1.2282 +
1.2283 + /* adjust search range according to sr from mv prediction */
1.2284 + if(sr > step_param)
1.2285 + step_param = sr;
1.2286 +
1.2287 + /* Initial step/diamond search */
1.2288 + {
1.2289 + bestsme = cpi->diamond_search_sad(x, b, d, &mvp_full, &d->bmi.mv,
1.2290 + step_param, sadpb, &num00,
1.2291 + &cpi->fn_ptr[BLOCK_16X16],
1.2292 + x->mvcost, &best_ref_mv);
1.2293 + mode_mv[NEWMV].as_int = d->bmi.mv.as_int;
1.2294 +
1.2295 + /* Further step/diamond searches as necessary */
1.2296 + n = 0;
1.2297 + further_steps = (cpi->sf.max_step_search_steps - 1) - step_param;
1.2298 +
1.2299 + n = num00;
1.2300 + num00 = 0;
1.2301 +
1.2302 + /* If there won't be more n-step search, check to see if refining search is needed. */
1.2303 + if (n > further_steps)
1.2304 + do_refine = 0;
1.2305 +
1.2306 + while (n < further_steps)
1.2307 + {
1.2308 + n++;
1.2309 +
1.2310 + if (num00)
1.2311 + num00--;
1.2312 + else
1.2313 + {
1.2314 + thissme = cpi->diamond_search_sad(x, b, d, &mvp_full,
1.2315 + &d->bmi.mv, step_param + n, sadpb, &num00,
1.2316 + &cpi->fn_ptr[BLOCK_16X16], x->mvcost,
1.2317 + &best_ref_mv);
1.2318 +
1.2319 + /* check to see if refining search is needed. */
1.2320 + if (num00 > (further_steps-n))
1.2321 + do_refine = 0;
1.2322 +
1.2323 + if (thissme < bestsme)
1.2324 + {
1.2325 + bestsme = thissme;
1.2326 + mode_mv[NEWMV].as_int = d->bmi.mv.as_int;
1.2327 + }
1.2328 + else
1.2329 + {
1.2330 + d->bmi.mv.as_int = mode_mv[NEWMV].as_int;
1.2331 + }
1.2332 + }
1.2333 + }
1.2334 + }
1.2335 +
1.2336 + /* final 1-away diamond refining search */
1.2337 + if (do_refine == 1)
1.2338 + {
1.2339 + int search_range;
1.2340 +
1.2341 + search_range = 8;
1.2342 +
1.2343 + thissme = cpi->refining_search_sad(x, b, d, &d->bmi.mv, sadpb,
1.2344 + search_range, &cpi->fn_ptr[BLOCK_16X16],
1.2345 + x->mvcost, &best_ref_mv);
1.2346 +
1.2347 + if (thissme < bestsme)
1.2348 + {
1.2349 + bestsme = thissme;
1.2350 + mode_mv[NEWMV].as_int = d->bmi.mv.as_int;
1.2351 + }
1.2352 + else
1.2353 + {
1.2354 + d->bmi.mv.as_int = mode_mv[NEWMV].as_int;
1.2355 + }
1.2356 + }
1.2357 +
1.2358 + x->mv_col_min = tmp_col_min;
1.2359 + x->mv_col_max = tmp_col_max;
1.2360 + x->mv_row_min = tmp_row_min;
1.2361 + x->mv_row_max = tmp_row_max;
1.2362 +
1.2363 + if (bestsme < INT_MAX)
1.2364 + {
1.2365 + int dis; /* TODO: use dis in distortion calculation later. */
1.2366 + unsigned int sse;
1.2367 + cpi->find_fractional_mv_step(x, b, d, &d->bmi.mv, &best_ref_mv,
1.2368 + x->errorperbit,
1.2369 + &cpi->fn_ptr[BLOCK_16X16],
1.2370 + x->mvcost, &dis, &sse);
1.2371 + }
1.2372 +
1.2373 + mode_mv[NEWMV].as_int = d->bmi.mv.as_int;
1.2374 +
1.2375 + /* Add the new motion vector cost to our rolling cost variable */
1.2376 + rd.rate2 += vp8_mv_bit_cost(&mode_mv[NEWMV], &best_ref_mv, x->mvcost, 96);
1.2377 + }
1.2378 +
1.2379 + case NEARESTMV:
1.2380 + case NEARMV:
1.2381 + /* Clip "next_nearest" so that it does not extend to far out
1.2382 + * of image
1.2383 + */
1.2384 + vp8_clamp_mv2(&mode_mv[this_mode], xd);
1.2385 +
1.2386 + /* Do not bother proceeding if the vector (from newmv, nearest
1.2387 + * or near) is 0,0 as this should then be coded using the zeromv
1.2388 + * mode.
1.2389 + */
1.2390 + if (((this_mode == NEARMV) || (this_mode == NEARESTMV)) && (mode_mv[this_mode].as_int == 0))
1.2391 + continue;
1.2392 +
1.2393 + case ZEROMV:
1.2394 +
1.2395 + /* Trap vectors that reach beyond the UMV borders
1.2396 + * Note that ALL New MV, Nearest MV Near MV and Zero MV code
1.2397 + * drops through to this point because of the lack of break
1.2398 + * statements in the previous two cases.
1.2399 + */
1.2400 + if (((mode_mv[this_mode].as_mv.row >> 3) < x->mv_row_min) || ((mode_mv[this_mode].as_mv.row >> 3) > x->mv_row_max) ||
1.2401 + ((mode_mv[this_mode].as_mv.col >> 3) < x->mv_col_min) || ((mode_mv[this_mode].as_mv.col >> 3) > x->mv_col_max))
1.2402 + continue;
1.2403 +
1.2404 + vp8_set_mbmode_and_mvs(x, this_mode, &mode_mv[this_mode]);
1.2405 + this_rd = evaluate_inter_mode_rd(mdcounts, &rd,
1.2406 + &disable_skip, cpi, x);
1.2407 + break;
1.2408 +
1.2409 + default:
1.2410 + break;
1.2411 + }
1.2412 +
1.2413 + this_rd = calculate_final_rd_costs(this_rd, &rd, &other_cost,
1.2414 + disable_skip, uv_intra_tteob,
1.2415 + intra_rd_penalty, cpi, x);
1.2416 +
1.2417 + /* Keep record of best intra distortion */
1.2418 + if ((x->e_mbd.mode_info_context->mbmi.ref_frame == INTRA_FRAME) &&
1.2419 + (this_rd < best_mode.intra_rd) )
1.2420 + {
1.2421 + best_mode.intra_rd = this_rd;
1.2422 + *returnintra = rd.distortion2 ;
1.2423 + }
1.2424 +#if CONFIG_TEMPORAL_DENOISING
1.2425 + if (cpi->oxcf.noise_sensitivity)
1.2426 + {
1.2427 + unsigned int sse;
1.2428 + vp8_get_inter_mbpred_error(x,&cpi->fn_ptr[BLOCK_16X16],&sse,
1.2429 + mode_mv[this_mode]);
1.2430 +
1.2431 + if (sse < best_rd_sse)
1.2432 + best_rd_sse = sse;
1.2433 +
1.2434 + /* Store for later use by denoiser. */
1.2435 + if (this_mode == ZEROMV && sse < zero_mv_sse )
1.2436 + {
1.2437 + zero_mv_sse = sse;
1.2438 + x->best_zeromv_reference_frame =
1.2439 + x->e_mbd.mode_info_context->mbmi.ref_frame;
1.2440 + }
1.2441 +
1.2442 + /* Store the best NEWMV in x for later use in the denoiser. */
1.2443 + if (x->e_mbd.mode_info_context->mbmi.mode == NEWMV &&
1.2444 + sse < best_sse)
1.2445 + {
1.2446 + best_sse = sse;
1.2447 + vp8_get_inter_mbpred_error(x,&cpi->fn_ptr[BLOCK_16X16],&best_sse,
1.2448 + mode_mv[this_mode]);
1.2449 + x->best_sse_inter_mode = NEWMV;
1.2450 + x->best_sse_mv = x->e_mbd.mode_info_context->mbmi.mv;
1.2451 + x->need_to_clamp_best_mvs =
1.2452 + x->e_mbd.mode_info_context->mbmi.need_to_clamp_mvs;
1.2453 + x->best_reference_frame =
1.2454 + x->e_mbd.mode_info_context->mbmi.ref_frame;
1.2455 + }
1.2456 + }
1.2457 +#endif
1.2458 +
1.2459 + /* Did this mode help.. i.i is it the new best mode */
1.2460 + if (this_rd < best_mode.rd || x->skip)
1.2461 + {
1.2462 + /* Note index of best mode so far */
1.2463 + best_mode_index = mode_index;
1.2464 + *returnrate = rd.rate2;
1.2465 + *returndistortion = rd.distortion2;
1.2466 + if (this_mode <= B_PRED)
1.2467 + {
1.2468 + x->e_mbd.mode_info_context->mbmi.uv_mode = uv_intra_mode;
1.2469 + /* required for left and above block mv */
1.2470 + x->e_mbd.mode_info_context->mbmi.mv.as_int = 0;
1.2471 + }
1.2472 + update_best_mode(&best_mode, this_rd, &rd, other_cost, x);
1.2473 +
1.2474 +
1.2475 + /* Testing this mode gave rise to an improvement in best error
1.2476 + * score. Lower threshold a bit for next time
1.2477 + */
1.2478 + x->rd_thresh_mult[mode_index] =
1.2479 + (x->rd_thresh_mult[mode_index] >= (MIN_THRESHMULT + 2)) ?
1.2480 + x->rd_thresh_mult[mode_index] - 2 : MIN_THRESHMULT;
1.2481 + }
1.2482 +
1.2483 + /* If the mode did not help improve the best error case then raise
1.2484 + * the threshold for testing that mode next time around.
1.2485 + */
1.2486 + else
1.2487 + {
1.2488 + x->rd_thresh_mult[mode_index] += 4;
1.2489 +
1.2490 + if (x->rd_thresh_mult[mode_index] > MAX_THRESHMULT)
1.2491 + x->rd_thresh_mult[mode_index] = MAX_THRESHMULT;
1.2492 + }
1.2493 + x->rd_threshes[mode_index] =
1.2494 + (cpi->rd_baseline_thresh[mode_index] >> 7) *
1.2495 + x->rd_thresh_mult[mode_index];
1.2496 +
1.2497 + if (x->skip)
1.2498 + break;
1.2499 +
1.2500 + }
1.2501 +
1.2502 + /* Reduce the activation RD thresholds for the best choice mode */
1.2503 + if ((cpi->rd_baseline_thresh[best_mode_index] > 0) && (cpi->rd_baseline_thresh[best_mode_index] < (INT_MAX >> 2)))
1.2504 + {
1.2505 + int best_adjustment = (x->rd_thresh_mult[best_mode_index] >> 2);
1.2506 +
1.2507 + x->rd_thresh_mult[best_mode_index] =
1.2508 + (x->rd_thresh_mult[best_mode_index] >=
1.2509 + (MIN_THRESHMULT + best_adjustment)) ?
1.2510 + x->rd_thresh_mult[best_mode_index] - best_adjustment :
1.2511 + MIN_THRESHMULT;
1.2512 + x->rd_threshes[best_mode_index] =
1.2513 + (cpi->rd_baseline_thresh[best_mode_index] >> 7) *
1.2514 + x->rd_thresh_mult[best_mode_index];
1.2515 + }
1.2516 +
1.2517 +#if CONFIG_TEMPORAL_DENOISING
1.2518 + if (cpi->oxcf.noise_sensitivity)
1.2519 + {
1.2520 + if (x->best_sse_inter_mode == DC_PRED)
1.2521 + {
1.2522 + /* No best MV found. */
1.2523 + x->best_sse_inter_mode = best_mode.mbmode.mode;
1.2524 + x->best_sse_mv = best_mode.mbmode.mv;
1.2525 + x->need_to_clamp_best_mvs = best_mode.mbmode.need_to_clamp_mvs;
1.2526 + x->best_reference_frame = best_mode.mbmode.ref_frame;
1.2527 + best_sse = best_rd_sse;
1.2528 + }
1.2529 + vp8_denoiser_denoise_mb(&cpi->denoiser, x, best_sse, zero_mv_sse,
1.2530 + recon_yoffset, recon_uvoffset);
1.2531 +
1.2532 +
1.2533 + /* Reevaluate ZEROMV after denoising. */
1.2534 + if (best_mode.mbmode.ref_frame == INTRA_FRAME &&
1.2535 + x->best_zeromv_reference_frame != INTRA_FRAME)
1.2536 + {
1.2537 + int this_rd = INT_MAX;
1.2538 + int disable_skip = 0;
1.2539 + int other_cost = 0;
1.2540 + int this_ref_frame = x->best_zeromv_reference_frame;
1.2541 + rd.rate2 = x->ref_frame_cost[this_ref_frame] +
1.2542 + vp8_cost_mv_ref(ZEROMV, mdcounts);
1.2543 + rd.distortion2 = 0;
1.2544 +
1.2545 + /* set up the proper prediction buffers for the frame */
1.2546 + x->e_mbd.mode_info_context->mbmi.ref_frame = this_ref_frame;
1.2547 + x->e_mbd.pre.y_buffer = plane[this_ref_frame][0];
1.2548 + x->e_mbd.pre.u_buffer = plane[this_ref_frame][1];
1.2549 + x->e_mbd.pre.v_buffer = plane[this_ref_frame][2];
1.2550 +
1.2551 + x->e_mbd.mode_info_context->mbmi.mode = ZEROMV;
1.2552 + x->e_mbd.mode_info_context->mbmi.uv_mode = DC_PRED;
1.2553 + x->e_mbd.mode_info_context->mbmi.mv.as_int = 0;
1.2554 +
1.2555 + this_rd = evaluate_inter_mode_rd(mdcounts, &rd, &disable_skip, cpi, x);
1.2556 + this_rd = calculate_final_rd_costs(this_rd, &rd, &other_cost,
1.2557 + disable_skip, uv_intra_tteob,
1.2558 + intra_rd_penalty, cpi, x);
1.2559 + if (this_rd < best_mode.rd || x->skip)
1.2560 + {
1.2561 + /* Note index of best mode so far */
1.2562 + best_mode_index = mode_index;
1.2563 + *returnrate = rd.rate2;
1.2564 + *returndistortion = rd.distortion2;
1.2565 + update_best_mode(&best_mode, this_rd, &rd, other_cost, x);
1.2566 + }
1.2567 + }
1.2568 +
1.2569 + }
1.2570 +#endif
1.2571 +
1.2572 + if (cpi->is_src_frame_alt_ref &&
1.2573 + (best_mode.mbmode.mode != ZEROMV || best_mode.mbmode.ref_frame != ALTREF_FRAME))
1.2574 + {
1.2575 + x->e_mbd.mode_info_context->mbmi.mode = ZEROMV;
1.2576 + x->e_mbd.mode_info_context->mbmi.ref_frame = ALTREF_FRAME;
1.2577 + x->e_mbd.mode_info_context->mbmi.mv.as_int = 0;
1.2578 + x->e_mbd.mode_info_context->mbmi.uv_mode = DC_PRED;
1.2579 + x->e_mbd.mode_info_context->mbmi.mb_skip_coeff =
1.2580 + (cpi->common.mb_no_coeff_skip);
1.2581 + x->e_mbd.mode_info_context->mbmi.partitioning = 0;
1.2582 + return;
1.2583 + }
1.2584 +
1.2585 +
1.2586 + /* macroblock modes */
1.2587 + vpx_memcpy(&x->e_mbd.mode_info_context->mbmi, &best_mode.mbmode, sizeof(MB_MODE_INFO));
1.2588 +
1.2589 + if (best_mode.mbmode.mode == B_PRED)
1.2590 + {
1.2591 + for (i = 0; i < 16; i++)
1.2592 + xd->mode_info_context->bmi[i].as_mode = best_mode.bmodes[i].as_mode;
1.2593 + }
1.2594 +
1.2595 + if (best_mode.mbmode.mode == SPLITMV)
1.2596 + {
1.2597 + for (i = 0; i < 16; i++)
1.2598 + xd->mode_info_context->bmi[i].mv.as_int = best_mode.bmodes[i].mv.as_int;
1.2599 +
1.2600 + vpx_memcpy(x->partition_info, &best_mode.partition, sizeof(PARTITION_INFO));
1.2601 +
1.2602 + x->e_mbd.mode_info_context->mbmi.mv.as_int =
1.2603 + x->partition_info->bmi[15].mv.as_int;
1.2604 + }
1.2605 +
1.2606 + if (sign_bias
1.2607 + != cpi->common.ref_frame_sign_bias[xd->mode_info_context->mbmi.ref_frame])
1.2608 + best_ref_mv.as_int = best_ref_mv_sb[!sign_bias].as_int;
1.2609 +
1.2610 + rd_update_mvcount(x, &best_ref_mv);
1.2611 +}
1.2612 +
1.2613 +void vp8_rd_pick_intra_mode(MACROBLOCK *x, int *rate_)
1.2614 +{
1.2615 + int error4x4, error16x16;
1.2616 + int rate4x4, rate16x16 = 0, rateuv;
1.2617 + int dist4x4, dist16x16, distuv;
1.2618 + int rate;
1.2619 + int rate4x4_tokenonly = 0;
1.2620 + int rate16x16_tokenonly = 0;
1.2621 + int rateuv_tokenonly = 0;
1.2622 +
1.2623 + x->e_mbd.mode_info_context->mbmi.ref_frame = INTRA_FRAME;
1.2624 +
1.2625 + rd_pick_intra_mbuv_mode(x, &rateuv, &rateuv_tokenonly, &distuv);
1.2626 + rate = rateuv;
1.2627 +
1.2628 + error16x16 = rd_pick_intra16x16mby_mode(x, &rate16x16, &rate16x16_tokenonly,
1.2629 + &dist16x16);
1.2630 +
1.2631 + error4x4 = rd_pick_intra4x4mby_modes(x, &rate4x4, &rate4x4_tokenonly,
1.2632 + &dist4x4, error16x16);
1.2633 +
1.2634 + if (error4x4 < error16x16)
1.2635 + {
1.2636 + x->e_mbd.mode_info_context->mbmi.mode = B_PRED;
1.2637 + rate += rate4x4;
1.2638 + }
1.2639 + else
1.2640 + {
1.2641 + rate += rate16x16;
1.2642 + }
1.2643 +
1.2644 + *rate_ = rate;
1.2645 +}
