1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/decoder/threading.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,910 @@
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 "vpx_config.h"
1.16 +#include "vp8_rtcd.h"
1.17 +#if !defined(WIN32) && CONFIG_OS_SUPPORT == 1
1.18 +# include <unistd.h>
1.19 +#endif
1.20 +#include "onyxd_int.h"
1.21 +#include "vpx_mem/vpx_mem.h"
1.22 +#include "vp8/common/threading.h"
1.23 +
1.24 +#include "vp8/common/loopfilter.h"
1.25 +#include "vp8/common/extend.h"
1.26 +#include "vpx_ports/vpx_timer.h"
1.27 +#include "detokenize.h"
1.28 +#include "vp8/common/reconintra4x4.h"
1.29 +#include "vp8/common/reconinter.h"
1.30 +#include "vp8/common/setupintrarecon.h"
1.31 +#if CONFIG_ERROR_CONCEALMENT
1.32 +#include "error_concealment.h"
1.33 +#endif
1.34 +
1.35 +#define CALLOC_ARRAY(p, n) CHECK_MEM_ERROR((p), vpx_calloc(sizeof(*(p)), (n)))
1.36 +#define CALLOC_ARRAY_ALIGNED(p, n, algn) do { \
1.37 + CHECK_MEM_ERROR((p), vpx_memalign((algn), sizeof(*(p)) * (n))); \
1.38 + memset((p), 0, (n) * sizeof(*(p))); \
1.39 +} while (0)
1.40 +
1.41 +
1.42 +void vp8_mb_init_dequantizer(VP8D_COMP *pbi, MACROBLOCKD *xd);
1.43 +
1.44 +static void setup_decoding_thread_data(VP8D_COMP *pbi, MACROBLOCKD *xd, MB_ROW_DEC *mbrd, int count)
1.45 +{
1.46 + VP8_COMMON *const pc = & pbi->common;
1.47 + int i;
1.48 +
1.49 + for (i = 0; i < count; i++)
1.50 + {
1.51 + MACROBLOCKD *mbd = &mbrd[i].mbd;
1.52 + mbd->subpixel_predict = xd->subpixel_predict;
1.53 + mbd->subpixel_predict8x4 = xd->subpixel_predict8x4;
1.54 + mbd->subpixel_predict8x8 = xd->subpixel_predict8x8;
1.55 + mbd->subpixel_predict16x16 = xd->subpixel_predict16x16;
1.56 +
1.57 + mbd->mode_info_context = pc->mi + pc->mode_info_stride * (i + 1);
1.58 + mbd->mode_info_stride = pc->mode_info_stride;
1.59 +
1.60 + mbd->frame_type = pc->frame_type;
1.61 + mbd->pre = xd->pre;
1.62 + mbd->dst = xd->dst;
1.63 +
1.64 + mbd->segmentation_enabled = xd->segmentation_enabled;
1.65 + mbd->mb_segement_abs_delta = xd->mb_segement_abs_delta;
1.66 + vpx_memcpy(mbd->segment_feature_data, xd->segment_feature_data, sizeof(xd->segment_feature_data));
1.67 +
1.68 + /*signed char ref_lf_deltas[MAX_REF_LF_DELTAS];*/
1.69 + vpx_memcpy(mbd->ref_lf_deltas, xd->ref_lf_deltas, sizeof(xd->ref_lf_deltas));
1.70 + /*signed char mode_lf_deltas[MAX_MODE_LF_DELTAS];*/
1.71 + vpx_memcpy(mbd->mode_lf_deltas, xd->mode_lf_deltas, sizeof(xd->mode_lf_deltas));
1.72 + /*unsigned char mode_ref_lf_delta_enabled;
1.73 + unsigned char mode_ref_lf_delta_update;*/
1.74 + mbd->mode_ref_lf_delta_enabled = xd->mode_ref_lf_delta_enabled;
1.75 + mbd->mode_ref_lf_delta_update = xd->mode_ref_lf_delta_update;
1.76 +
1.77 + mbd->current_bc = &pbi->mbc[0];
1.78 +
1.79 + vpx_memcpy(mbd->dequant_y1_dc, xd->dequant_y1_dc, sizeof(xd->dequant_y1_dc));
1.80 + vpx_memcpy(mbd->dequant_y1, xd->dequant_y1, sizeof(xd->dequant_y1));
1.81 + vpx_memcpy(mbd->dequant_y2, xd->dequant_y2, sizeof(xd->dequant_y2));
1.82 + vpx_memcpy(mbd->dequant_uv, xd->dequant_uv, sizeof(xd->dequant_uv));
1.83 +
1.84 + mbd->fullpixel_mask = 0xffffffff;
1.85 +
1.86 + if (pc->full_pixel)
1.87 + mbd->fullpixel_mask = 0xfffffff8;
1.88 +
1.89 + }
1.90 +
1.91 + for (i = 0; i < pc->mb_rows; i++)
1.92 + pbi->mt_current_mb_col[i] = -1;
1.93 +}
1.94 +
1.95 +static void mt_decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd,
1.96 + unsigned int mb_idx)
1.97 +{
1.98 + MB_PREDICTION_MODE mode;
1.99 + int i;
1.100 +#if CONFIG_ERROR_CONCEALMENT
1.101 + int corruption_detected = 0;
1.102 +#endif
1.103 +
1.104 + if (xd->mode_info_context->mbmi.mb_skip_coeff)
1.105 + {
1.106 + vp8_reset_mb_tokens_context(xd);
1.107 + }
1.108 + else if (!vp8dx_bool_error(xd->current_bc))
1.109 + {
1.110 + int eobtotal;
1.111 + eobtotal = vp8_decode_mb_tokens(pbi, xd);
1.112 +
1.113 + /* Special case: Force the loopfilter to skip when eobtotal is zero */
1.114 + xd->mode_info_context->mbmi.mb_skip_coeff = (eobtotal==0);
1.115 + }
1.116 +
1.117 + mode = xd->mode_info_context->mbmi.mode;
1.118 +
1.119 + if (xd->segmentation_enabled)
1.120 + vp8_mb_init_dequantizer(pbi, xd);
1.121 +
1.122 +
1.123 +#if CONFIG_ERROR_CONCEALMENT
1.124 +
1.125 + if(pbi->ec_active)
1.126 + {
1.127 + int throw_residual;
1.128 + /* When we have independent partitions we can apply residual even
1.129 + * though other partitions within the frame are corrupt.
1.130 + */
1.131 + throw_residual = (!pbi->independent_partitions &&
1.132 + pbi->frame_corrupt_residual);
1.133 + throw_residual = (throw_residual || vp8dx_bool_error(xd->current_bc));
1.134 +
1.135 + if ((mb_idx >= pbi->mvs_corrupt_from_mb || throw_residual))
1.136 + {
1.137 + /* MB with corrupt residuals or corrupt mode/motion vectors.
1.138 + * Better to use the predictor as reconstruction.
1.139 + */
1.140 + pbi->frame_corrupt_residual = 1;
1.141 + vpx_memset(xd->qcoeff, 0, sizeof(xd->qcoeff));
1.142 + vp8_conceal_corrupt_mb(xd);
1.143 +
1.144 +
1.145 + corruption_detected = 1;
1.146 +
1.147 + /* force idct to be skipped for B_PRED and use the
1.148 + * prediction only for reconstruction
1.149 + * */
1.150 + vpx_memset(xd->eobs, 0, 25);
1.151 + }
1.152 + }
1.153 +#endif
1.154 +
1.155 + /* do prediction */
1.156 + if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME)
1.157 + {
1.158 + vp8_build_intra_predictors_mbuv_s(xd,
1.159 + xd->recon_above[1],
1.160 + xd->recon_above[2],
1.161 + xd->recon_left[1],
1.162 + xd->recon_left[2],
1.163 + xd->recon_left_stride[1],
1.164 + xd->dst.u_buffer, xd->dst.v_buffer,
1.165 + xd->dst.uv_stride);
1.166 +
1.167 + if (mode != B_PRED)
1.168 + {
1.169 + vp8_build_intra_predictors_mby_s(xd,
1.170 + xd->recon_above[0],
1.171 + xd->recon_left[0],
1.172 + xd->recon_left_stride[0],
1.173 + xd->dst.y_buffer,
1.174 + xd->dst.y_stride);
1.175 + }
1.176 + else
1.177 + {
1.178 + short *DQC = xd->dequant_y1;
1.179 + int dst_stride = xd->dst.y_stride;
1.180 +
1.181 + /* clear out residual eob info */
1.182 + if(xd->mode_info_context->mbmi.mb_skip_coeff)
1.183 + vpx_memset(xd->eobs, 0, 25);
1.184 +
1.185 + intra_prediction_down_copy(xd, xd->recon_above[0] + 16);
1.186 +
1.187 + for (i = 0; i < 16; i++)
1.188 + {
1.189 + BLOCKD *b = &xd->block[i];
1.190 + unsigned char *dst = xd->dst.y_buffer + b->offset;
1.191 + B_PREDICTION_MODE b_mode =
1.192 + xd->mode_info_context->bmi[i].as_mode;
1.193 + unsigned char *Above;
1.194 + unsigned char *yleft;
1.195 + int left_stride;
1.196 + unsigned char top_left;
1.197 +
1.198 + /*Caution: For some b_mode, it needs 8 pixels (4 above + 4 above-right).*/
1.199 + if (i < 4 && pbi->common.filter_level)
1.200 + Above = xd->recon_above[0] + b->offset;
1.201 + else
1.202 + Above = dst - dst_stride;
1.203 +
1.204 + if (i%4==0 && pbi->common.filter_level)
1.205 + {
1.206 + yleft = xd->recon_left[0] + i;
1.207 + left_stride = 1;
1.208 + }
1.209 + else
1.210 + {
1.211 + yleft = dst - 1;
1.212 + left_stride = dst_stride;
1.213 + }
1.214 +
1.215 + if ((i==4 || i==8 || i==12) && pbi->common.filter_level)
1.216 + top_left = *(xd->recon_left[0] + i - 1);
1.217 + else
1.218 + top_left = Above[-1];
1.219 +
1.220 + vp8_intra4x4_predict(Above, yleft, left_stride,
1.221 + b_mode, dst, dst_stride, top_left);
1.222 +
1.223 + if (xd->eobs[i] )
1.224 + {
1.225 + if (xd->eobs[i] > 1)
1.226 + {
1.227 + vp8_dequant_idct_add(b->qcoeff, DQC, dst, dst_stride);
1.228 + }
1.229 + else
1.230 + {
1.231 + vp8_dc_only_idct_add(b->qcoeff[0] * DQC[0],
1.232 + dst, dst_stride, dst, dst_stride);
1.233 + vpx_memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0]));
1.234 + }
1.235 + }
1.236 + }
1.237 + }
1.238 + }
1.239 + else
1.240 + {
1.241 + vp8_build_inter_predictors_mb(xd);
1.242 + }
1.243 +
1.244 +
1.245 +#if CONFIG_ERROR_CONCEALMENT
1.246 + if (corruption_detected)
1.247 + {
1.248 + return;
1.249 + }
1.250 +#endif
1.251 +
1.252 + if(!xd->mode_info_context->mbmi.mb_skip_coeff)
1.253 + {
1.254 + /* dequantization and idct */
1.255 + if (mode != B_PRED)
1.256 + {
1.257 + short *DQC = xd->dequant_y1;
1.258 +
1.259 + if (mode != SPLITMV)
1.260 + {
1.261 + BLOCKD *b = &xd->block[24];
1.262 +
1.263 + /* do 2nd order transform on the dc block */
1.264 + if (xd->eobs[24] > 1)
1.265 + {
1.266 + vp8_dequantize_b(b, xd->dequant_y2);
1.267 +
1.268 + vp8_short_inv_walsh4x4(&b->dqcoeff[0],
1.269 + xd->qcoeff);
1.270 + vpx_memset(b->qcoeff, 0, 16 * sizeof(b->qcoeff[0]));
1.271 + }
1.272 + else
1.273 + {
1.274 + b->dqcoeff[0] = b->qcoeff[0] * xd->dequant_y2[0];
1.275 + vp8_short_inv_walsh4x4_1(&b->dqcoeff[0],
1.276 + xd->qcoeff);
1.277 + vpx_memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0]));
1.278 + }
1.279 +
1.280 + /* override the dc dequant constant in order to preserve the
1.281 + * dc components
1.282 + */
1.283 + DQC = xd->dequant_y1_dc;
1.284 + }
1.285 +
1.286 + vp8_dequant_idct_add_y_block
1.287 + (xd->qcoeff, DQC,
1.288 + xd->dst.y_buffer,
1.289 + xd->dst.y_stride, xd->eobs);
1.290 + }
1.291 +
1.292 + vp8_dequant_idct_add_uv_block
1.293 + (xd->qcoeff+16*16, xd->dequant_uv,
1.294 + xd->dst.u_buffer, xd->dst.v_buffer,
1.295 + xd->dst.uv_stride, xd->eobs+16);
1.296 + }
1.297 +}
1.298 +
1.299 +static void mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd, int start_mb_row)
1.300 +{
1.301 + volatile const int *last_row_current_mb_col;
1.302 + volatile int *current_mb_col;
1.303 + int mb_row;
1.304 + VP8_COMMON *pc = &pbi->common;
1.305 + const int nsync = pbi->sync_range;
1.306 + const int first_row_no_sync_above = pc->mb_cols + nsync;
1.307 + int num_part = 1 << pbi->common.multi_token_partition;
1.308 + int last_mb_row = start_mb_row;
1.309 +
1.310 + YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
1.311 + YV12_BUFFER_CONFIG *yv12_fb_lst = pbi->dec_fb_ref[LAST_FRAME];
1.312 +
1.313 + int recon_y_stride = yv12_fb_new->y_stride;
1.314 + int recon_uv_stride = yv12_fb_new->uv_stride;
1.315 +
1.316 + unsigned char *ref_buffer[MAX_REF_FRAMES][3];
1.317 + unsigned char *dst_buffer[3];
1.318 + int i;
1.319 + int ref_fb_corrupted[MAX_REF_FRAMES];
1.320 +
1.321 + ref_fb_corrupted[INTRA_FRAME] = 0;
1.322 +
1.323 + for(i = 1; i < MAX_REF_FRAMES; i++)
1.324 + {
1.325 + YV12_BUFFER_CONFIG *this_fb = pbi->dec_fb_ref[i];
1.326 +
1.327 + ref_buffer[i][0] = this_fb->y_buffer;
1.328 + ref_buffer[i][1] = this_fb->u_buffer;
1.329 + ref_buffer[i][2] = this_fb->v_buffer;
1.330 +
1.331 + ref_fb_corrupted[i] = this_fb->corrupted;
1.332 + }
1.333 +
1.334 + dst_buffer[0] = yv12_fb_new->y_buffer;
1.335 + dst_buffer[1] = yv12_fb_new->u_buffer;
1.336 + dst_buffer[2] = yv12_fb_new->v_buffer;
1.337 +
1.338 + xd->up_available = (start_mb_row != 0);
1.339 +
1.340 + for (mb_row = start_mb_row; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1))
1.341 + {
1.342 + int recon_yoffset, recon_uvoffset;
1.343 + int mb_col;
1.344 + int filter_level;
1.345 + loop_filter_info_n *lfi_n = &pc->lf_info;
1.346 +
1.347 + /* save last row processed by this thread */
1.348 + last_mb_row = mb_row;
1.349 + /* select bool coder for current partition */
1.350 + xd->current_bc = &pbi->mbc[mb_row%num_part];
1.351 +
1.352 + if (mb_row > 0)
1.353 + last_row_current_mb_col = &pbi->mt_current_mb_col[mb_row -1];
1.354 + else
1.355 + last_row_current_mb_col = &first_row_no_sync_above;
1.356 +
1.357 + current_mb_col = &pbi->mt_current_mb_col[mb_row];
1.358 +
1.359 + recon_yoffset = mb_row * recon_y_stride * 16;
1.360 + recon_uvoffset = mb_row * recon_uv_stride * 8;
1.361 +
1.362 + /* reset contexts */
1.363 + xd->above_context = pc->above_context;
1.364 + vpx_memset(xd->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
1.365 +
1.366 + xd->left_available = 0;
1.367 +
1.368 + xd->mb_to_top_edge = -((mb_row * 16)) << 3;
1.369 + xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3;
1.370 +
1.371 + if (pbi->common.filter_level)
1.372 + {
1.373 + xd->recon_above[0] = pbi->mt_yabove_row[mb_row] + 0*16 +32;
1.374 + xd->recon_above[1] = pbi->mt_uabove_row[mb_row] + 0*8 +16;
1.375 + xd->recon_above[2] = pbi->mt_vabove_row[mb_row] + 0*8 +16;
1.376 +
1.377 + xd->recon_left[0] = pbi->mt_yleft_col[mb_row];
1.378 + xd->recon_left[1] = pbi->mt_uleft_col[mb_row];
1.379 + xd->recon_left[2] = pbi->mt_vleft_col[mb_row];
1.380 +
1.381 + /* TODO: move to outside row loop */
1.382 + xd->recon_left_stride[0] = 1;
1.383 + xd->recon_left_stride[1] = 1;
1.384 + }
1.385 + else
1.386 + {
1.387 + xd->recon_above[0] = dst_buffer[0] + recon_yoffset;
1.388 + xd->recon_above[1] = dst_buffer[1] + recon_uvoffset;
1.389 + xd->recon_above[2] = dst_buffer[2] + recon_uvoffset;
1.390 +
1.391 + xd->recon_left[0] = xd->recon_above[0] - 1;
1.392 + xd->recon_left[1] = xd->recon_above[1] - 1;
1.393 + xd->recon_left[2] = xd->recon_above[2] - 1;
1.394 +
1.395 + xd->recon_above[0] -= xd->dst.y_stride;
1.396 + xd->recon_above[1] -= xd->dst.uv_stride;
1.397 + xd->recon_above[2] -= xd->dst.uv_stride;
1.398 +
1.399 + /* TODO: move to outside row loop */
1.400 + xd->recon_left_stride[0] = xd->dst.y_stride;
1.401 + xd->recon_left_stride[1] = xd->dst.uv_stride;
1.402 +
1.403 + setup_intra_recon_left(xd->recon_left[0], xd->recon_left[1],
1.404 + xd->recon_left[2], xd->dst.y_stride,
1.405 + xd->dst.uv_stride);
1.406 + }
1.407 +
1.408 + for (mb_col = 0; mb_col < pc->mb_cols; mb_col++)
1.409 + {
1.410 + *current_mb_col = mb_col - 1;
1.411 +
1.412 + if ((mb_col & (nsync - 1)) == 0)
1.413 + {
1.414 + while (mb_col > (*last_row_current_mb_col - nsync))
1.415 + {
1.416 + x86_pause_hint();
1.417 + thread_sleep(0);
1.418 + }
1.419 + }
1.420 +
1.421 + /* Distance of MB to the various image edges.
1.422 + * These are specified to 8th pel as they are always
1.423 + * compared to values that are in 1/8th pel units.
1.424 + */
1.425 + xd->mb_to_left_edge = -((mb_col * 16) << 3);
1.426 + xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3;
1.427 +
1.428 + #if CONFIG_ERROR_CONCEALMENT
1.429 + {
1.430 + int corrupt_residual =
1.431 + (!pbi->independent_partitions &&
1.432 + pbi->frame_corrupt_residual) ||
1.433 + vp8dx_bool_error(xd->current_bc);
1.434 + if (pbi->ec_active &&
1.435 + (xd->mode_info_context->mbmi.ref_frame ==
1.436 + INTRA_FRAME) &&
1.437 + corrupt_residual)
1.438 + {
1.439 + /* We have an intra block with corrupt
1.440 + * coefficients, better to conceal with an inter
1.441 + * block.
1.442 + * Interpolate MVs from neighboring MBs
1.443 + *
1.444 + * Note that for the first mb with corrupt
1.445 + * residual in a frame, we might not discover
1.446 + * that before decoding the residual. That
1.447 + * happens after this check, and therefore no
1.448 + * inter concealment will be done.
1.449 + */
1.450 + vp8_interpolate_motion(xd,
1.451 + mb_row, mb_col,
1.452 + pc->mb_rows, pc->mb_cols,
1.453 + pc->mode_info_stride);
1.454 + }
1.455 + }
1.456 + #endif
1.457 +
1.458 +
1.459 + xd->dst.y_buffer = dst_buffer[0] + recon_yoffset;
1.460 + xd->dst.u_buffer = dst_buffer[1] + recon_uvoffset;
1.461 + xd->dst.v_buffer = dst_buffer[2] + recon_uvoffset;
1.462 +
1.463 + xd->pre.y_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][0] + recon_yoffset;
1.464 + xd->pre.u_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][1] + recon_uvoffset;
1.465 + xd->pre.v_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][2] + recon_uvoffset;
1.466 +
1.467 + /* propagate errors from reference frames */
1.468 + xd->corrupted |= ref_fb_corrupted[xd->mode_info_context->mbmi.ref_frame];
1.469 +
1.470 + mt_decode_macroblock(pbi, xd, 0);
1.471 +
1.472 + xd->left_available = 1;
1.473 +
1.474 + /* check if the boolean decoder has suffered an error */
1.475 + xd->corrupted |= vp8dx_bool_error(xd->current_bc);
1.476 +
1.477 + xd->recon_above[0] += 16;
1.478 + xd->recon_above[1] += 8;
1.479 + xd->recon_above[2] += 8;
1.480 +
1.481 + if (!pbi->common.filter_level)
1.482 + {
1.483 + xd->recon_left[0] += 16;
1.484 + xd->recon_left[1] += 8;
1.485 + xd->recon_left[2] += 8;
1.486 + }
1.487 +
1.488 + if (pbi->common.filter_level)
1.489 + {
1.490 + int skip_lf = (xd->mode_info_context->mbmi.mode != B_PRED &&
1.491 + xd->mode_info_context->mbmi.mode != SPLITMV &&
1.492 + xd->mode_info_context->mbmi.mb_skip_coeff);
1.493 +
1.494 + const int mode_index = lfi_n->mode_lf_lut[xd->mode_info_context->mbmi.mode];
1.495 + const int seg = xd->mode_info_context->mbmi.segment_id;
1.496 + const int ref_frame = xd->mode_info_context->mbmi.ref_frame;
1.497 +
1.498 + filter_level = lfi_n->lvl[seg][ref_frame][mode_index];
1.499 +
1.500 + if( mb_row != pc->mb_rows-1 )
1.501 + {
1.502 + /* Save decoded MB last row data for next-row decoding */
1.503 + vpx_memcpy((pbi->mt_yabove_row[mb_row + 1] + 32 + mb_col*16), (xd->dst.y_buffer + 15 * recon_y_stride), 16);
1.504 + vpx_memcpy((pbi->mt_uabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.u_buffer + 7 * recon_uv_stride), 8);
1.505 + vpx_memcpy((pbi->mt_vabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.v_buffer + 7 * recon_uv_stride), 8);
1.506 + }
1.507 +
1.508 + /* save left_col for next MB decoding */
1.509 + if(mb_col != pc->mb_cols-1)
1.510 + {
1.511 + MODE_INFO *next = xd->mode_info_context +1;
1.512 +
1.513 + if (next->mbmi.ref_frame == INTRA_FRAME)
1.514 + {
1.515 + for (i = 0; i < 16; i++)
1.516 + pbi->mt_yleft_col[mb_row][i] = xd->dst.y_buffer [i* recon_y_stride + 15];
1.517 + for (i = 0; i < 8; i++)
1.518 + {
1.519 + pbi->mt_uleft_col[mb_row][i] = xd->dst.u_buffer [i* recon_uv_stride + 7];
1.520 + pbi->mt_vleft_col[mb_row][i] = xd->dst.v_buffer [i* recon_uv_stride + 7];
1.521 + }
1.522 + }
1.523 + }
1.524 +
1.525 + /* loopfilter on this macroblock. */
1.526 + if (filter_level)
1.527 + {
1.528 + if(pc->filter_type == NORMAL_LOOPFILTER)
1.529 + {
1.530 + loop_filter_info lfi;
1.531 + FRAME_TYPE frame_type = pc->frame_type;
1.532 + const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level];
1.533 + lfi.mblim = lfi_n->mblim[filter_level];
1.534 + lfi.blim = lfi_n->blim[filter_level];
1.535 + lfi.lim = lfi_n->lim[filter_level];
1.536 + lfi.hev_thr = lfi_n->hev_thr[hev_index];
1.537 +
1.538 + if (mb_col > 0)
1.539 + vp8_loop_filter_mbv
1.540 + (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
1.541 +
1.542 + if (!skip_lf)
1.543 + vp8_loop_filter_bv
1.544 + (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
1.545 +
1.546 + /* don't apply across umv border */
1.547 + if (mb_row > 0)
1.548 + vp8_loop_filter_mbh
1.549 + (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
1.550 +
1.551 + if (!skip_lf)
1.552 + vp8_loop_filter_bh
1.553 + (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
1.554 + }
1.555 + else
1.556 + {
1.557 + if (mb_col > 0)
1.558 + vp8_loop_filter_simple_mbv
1.559 + (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]);
1.560 +
1.561 + if (!skip_lf)
1.562 + vp8_loop_filter_simple_bv
1.563 + (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]);
1.564 +
1.565 + /* don't apply across umv border */
1.566 + if (mb_row > 0)
1.567 + vp8_loop_filter_simple_mbh
1.568 + (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]);
1.569 +
1.570 + if (!skip_lf)
1.571 + vp8_loop_filter_simple_bh
1.572 + (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]);
1.573 + }
1.574 + }
1.575 +
1.576 + }
1.577 +
1.578 + recon_yoffset += 16;
1.579 + recon_uvoffset += 8;
1.580 +
1.581 + ++xd->mode_info_context; /* next mb */
1.582 +
1.583 + xd->above_context++;
1.584 + }
1.585 +
1.586 + /* adjust to the next row of mbs */
1.587 + if (pbi->common.filter_level)
1.588 + {
1.589 + if(mb_row != pc->mb_rows-1)
1.590 + {
1.591 + int lasty = yv12_fb_lst->y_width + VP8BORDERINPIXELS;
1.592 + int lastuv = (yv12_fb_lst->y_width>>1) + (VP8BORDERINPIXELS>>1);
1.593 +
1.594 + for (i = 0; i < 4; i++)
1.595 + {
1.596 + pbi->mt_yabove_row[mb_row +1][lasty + i] = pbi->mt_yabove_row[mb_row +1][lasty -1];
1.597 + pbi->mt_uabove_row[mb_row +1][lastuv + i] = pbi->mt_uabove_row[mb_row +1][lastuv -1];
1.598 + pbi->mt_vabove_row[mb_row +1][lastuv + i] = pbi->mt_vabove_row[mb_row +1][lastuv -1];
1.599 + }
1.600 + }
1.601 + }
1.602 + else
1.603 + vp8_extend_mb_row(yv12_fb_new, xd->dst.y_buffer + 16,
1.604 + xd->dst.u_buffer + 8, xd->dst.v_buffer + 8);
1.605 +
1.606 + /* last MB of row is ready just after extension is done */
1.607 + *current_mb_col = mb_col + nsync;
1.608 +
1.609 + ++xd->mode_info_context; /* skip prediction column */
1.610 + xd->up_available = 1;
1.611 +
1.612 + /* since we have multithread */
1.613 + xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count;
1.614 + }
1.615 +
1.616 + /* signal end of frame decoding if this thread processed the last mb_row */
1.617 + if (last_mb_row == (pc->mb_rows - 1))
1.618 + sem_post(&pbi->h_event_end_decoding);
1.619 +
1.620 +}
1.621 +
1.622 +
1.623 +static THREAD_FUNCTION thread_decoding_proc(void *p_data)
1.624 +{
1.625 + int ithread = ((DECODETHREAD_DATA *)p_data)->ithread;
1.626 + VP8D_COMP *pbi = (VP8D_COMP *)(((DECODETHREAD_DATA *)p_data)->ptr1);
1.627 + MB_ROW_DEC *mbrd = (MB_ROW_DEC *)(((DECODETHREAD_DATA *)p_data)->ptr2);
1.628 + ENTROPY_CONTEXT_PLANES mb_row_left_context;
1.629 +
1.630 + while (1)
1.631 + {
1.632 + if (pbi->b_multithreaded_rd == 0)
1.633 + break;
1.634 +
1.635 + if (sem_wait(&pbi->h_event_start_decoding[ithread]) == 0)
1.636 + {
1.637 + if (pbi->b_multithreaded_rd == 0)
1.638 + break;
1.639 + else
1.640 + {
1.641 + MACROBLOCKD *xd = &mbrd->mbd;
1.642 + xd->left_context = &mb_row_left_context;
1.643 +
1.644 + mt_decode_mb_rows(pbi, xd, ithread+1);
1.645 + }
1.646 + }
1.647 + }
1.648 +
1.649 + return 0 ;
1.650 +}
1.651 +
1.652 +
1.653 +void vp8_decoder_create_threads(VP8D_COMP *pbi)
1.654 +{
1.655 + int core_count = 0;
1.656 + unsigned int ithread;
1.657 +
1.658 + pbi->b_multithreaded_rd = 0;
1.659 + pbi->allocated_decoding_thread_count = 0;
1.660 +
1.661 + /* limit decoding threads to the max number of token partitions */
1.662 + core_count = (pbi->max_threads > 8) ? 8 : pbi->max_threads;
1.663 +
1.664 + /* limit decoding threads to the available cores */
1.665 + if (core_count > pbi->common.processor_core_count)
1.666 + core_count = pbi->common.processor_core_count;
1.667 +
1.668 + if (core_count > 1)
1.669 + {
1.670 + pbi->b_multithreaded_rd = 1;
1.671 + pbi->decoding_thread_count = core_count - 1;
1.672 +
1.673 + CALLOC_ARRAY(pbi->h_decoding_thread, pbi->decoding_thread_count);
1.674 + CALLOC_ARRAY(pbi->h_event_start_decoding, pbi->decoding_thread_count);
1.675 + CALLOC_ARRAY_ALIGNED(pbi->mb_row_di, pbi->decoding_thread_count, 32);
1.676 + CALLOC_ARRAY(pbi->de_thread_data, pbi->decoding_thread_count);
1.677 +
1.678 + for (ithread = 0; ithread < pbi->decoding_thread_count; ithread++)
1.679 + {
1.680 + sem_init(&pbi->h_event_start_decoding[ithread], 0, 0);
1.681 +
1.682 + vp8_setup_block_dptrs(&pbi->mb_row_di[ithread].mbd);
1.683 +
1.684 + pbi->de_thread_data[ithread].ithread = ithread;
1.685 + pbi->de_thread_data[ithread].ptr1 = (void *)pbi;
1.686 + pbi->de_thread_data[ithread].ptr2 = (void *) &pbi->mb_row_di[ithread];
1.687 +
1.688 + pthread_create(&pbi->h_decoding_thread[ithread], 0, thread_decoding_proc, (&pbi->de_thread_data[ithread]));
1.689 + }
1.690 +
1.691 + sem_init(&pbi->h_event_end_decoding, 0, 0);
1.692 +
1.693 + pbi->allocated_decoding_thread_count = pbi->decoding_thread_count;
1.694 + }
1.695 +}
1.696 +
1.697 +
1.698 +void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows)
1.699 +{
1.700 + int i;
1.701 +
1.702 + if (pbi->b_multithreaded_rd)
1.703 + {
1.704 + vpx_free(pbi->mt_current_mb_col);
1.705 + pbi->mt_current_mb_col = NULL ;
1.706 +
1.707 + /* Free above_row buffers. */
1.708 + if (pbi->mt_yabove_row)
1.709 + {
1.710 + for (i=0; i< mb_rows; i++)
1.711 + {
1.712 + vpx_free(pbi->mt_yabove_row[i]);
1.713 + pbi->mt_yabove_row[i] = NULL ;
1.714 + }
1.715 + vpx_free(pbi->mt_yabove_row);
1.716 + pbi->mt_yabove_row = NULL ;
1.717 + }
1.718 +
1.719 + if (pbi->mt_uabove_row)
1.720 + {
1.721 + for (i=0; i< mb_rows; i++)
1.722 + {
1.723 + vpx_free(pbi->mt_uabove_row[i]);
1.724 + pbi->mt_uabove_row[i] = NULL ;
1.725 + }
1.726 + vpx_free(pbi->mt_uabove_row);
1.727 + pbi->mt_uabove_row = NULL ;
1.728 + }
1.729 +
1.730 + if (pbi->mt_vabove_row)
1.731 + {
1.732 + for (i=0; i< mb_rows; i++)
1.733 + {
1.734 + vpx_free(pbi->mt_vabove_row[i]);
1.735 + pbi->mt_vabove_row[i] = NULL ;
1.736 + }
1.737 + vpx_free(pbi->mt_vabove_row);
1.738 + pbi->mt_vabove_row = NULL ;
1.739 + }
1.740 +
1.741 + /* Free left_col buffers. */
1.742 + if (pbi->mt_yleft_col)
1.743 + {
1.744 + for (i=0; i< mb_rows; i++)
1.745 + {
1.746 + vpx_free(pbi->mt_yleft_col[i]);
1.747 + pbi->mt_yleft_col[i] = NULL ;
1.748 + }
1.749 + vpx_free(pbi->mt_yleft_col);
1.750 + pbi->mt_yleft_col = NULL ;
1.751 + }
1.752 +
1.753 + if (pbi->mt_uleft_col)
1.754 + {
1.755 + for (i=0; i< mb_rows; i++)
1.756 + {
1.757 + vpx_free(pbi->mt_uleft_col[i]);
1.758 + pbi->mt_uleft_col[i] = NULL ;
1.759 + }
1.760 + vpx_free(pbi->mt_uleft_col);
1.761 + pbi->mt_uleft_col = NULL ;
1.762 + }
1.763 +
1.764 + if (pbi->mt_vleft_col)
1.765 + {
1.766 + for (i=0; i< mb_rows; i++)
1.767 + {
1.768 + vpx_free(pbi->mt_vleft_col[i]);
1.769 + pbi->mt_vleft_col[i] = NULL ;
1.770 + }
1.771 + vpx_free(pbi->mt_vleft_col);
1.772 + pbi->mt_vleft_col = NULL ;
1.773 + }
1.774 + }
1.775 +}
1.776 +
1.777 +
1.778 +void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows)
1.779 +{
1.780 + VP8_COMMON *const pc = & pbi->common;
1.781 + int i;
1.782 + int uv_width;
1.783 +
1.784 + if (pbi->b_multithreaded_rd)
1.785 + {
1.786 + vp8mt_de_alloc_temp_buffers(pbi, prev_mb_rows);
1.787 +
1.788 + /* our internal buffers are always multiples of 16 */
1.789 + if ((width & 0xf) != 0)
1.790 + width += 16 - (width & 0xf);
1.791 +
1.792 + if (width < 640) pbi->sync_range = 1;
1.793 + else if (width <= 1280) pbi->sync_range = 8;
1.794 + else if (width <= 2560) pbi->sync_range =16;
1.795 + else pbi->sync_range = 32;
1.796 +
1.797 + uv_width = width >>1;
1.798 +
1.799 + /* Allocate an int for each mb row. */
1.800 + CALLOC_ARRAY(pbi->mt_current_mb_col, pc->mb_rows);
1.801 +
1.802 + /* Allocate memory for above_row buffers. */
1.803 + CALLOC_ARRAY(pbi->mt_yabove_row, pc->mb_rows);
1.804 + for (i = 0; i < pc->mb_rows; i++)
1.805 + CHECK_MEM_ERROR(pbi->mt_yabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (width + (VP8BORDERINPIXELS<<1))));
1.806 +
1.807 + CALLOC_ARRAY(pbi->mt_uabove_row, pc->mb_rows);
1.808 + for (i = 0; i < pc->mb_rows; i++)
1.809 + CHECK_MEM_ERROR(pbi->mt_uabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS)));
1.810 +
1.811 + CALLOC_ARRAY(pbi->mt_vabove_row, pc->mb_rows);
1.812 + for (i = 0; i < pc->mb_rows; i++)
1.813 + CHECK_MEM_ERROR(pbi->mt_vabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS)));
1.814 +
1.815 + /* Allocate memory for left_col buffers. */
1.816 + CALLOC_ARRAY(pbi->mt_yleft_col, pc->mb_rows);
1.817 + for (i = 0; i < pc->mb_rows; i++)
1.818 + CHECK_MEM_ERROR(pbi->mt_yleft_col[i], vpx_calloc(sizeof(unsigned char) * 16, 1));
1.819 +
1.820 + CALLOC_ARRAY(pbi->mt_uleft_col, pc->mb_rows);
1.821 + for (i = 0; i < pc->mb_rows; i++)
1.822 + CHECK_MEM_ERROR(pbi->mt_uleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1));
1.823 +
1.824 + CALLOC_ARRAY(pbi->mt_vleft_col, pc->mb_rows);
1.825 + for (i = 0; i < pc->mb_rows; i++)
1.826 + CHECK_MEM_ERROR(pbi->mt_vleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1));
1.827 + }
1.828 +}
1.829 +
1.830 +
1.831 +void vp8_decoder_remove_threads(VP8D_COMP *pbi)
1.832 +{
1.833 + /* shutdown MB Decoding thread; */
1.834 + if (pbi->b_multithreaded_rd)
1.835 + {
1.836 + int i;
1.837 +
1.838 + pbi->b_multithreaded_rd = 0;
1.839 +
1.840 + /* allow all threads to exit */
1.841 + for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
1.842 + {
1.843 + sem_post(&pbi->h_event_start_decoding[i]);
1.844 + pthread_join(pbi->h_decoding_thread[i], NULL);
1.845 + }
1.846 +
1.847 + for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
1.848 + {
1.849 + sem_destroy(&pbi->h_event_start_decoding[i]);
1.850 + }
1.851 +
1.852 + sem_destroy(&pbi->h_event_end_decoding);
1.853 +
1.854 + vpx_free(pbi->h_decoding_thread);
1.855 + pbi->h_decoding_thread = NULL;
1.856 +
1.857 + vpx_free(pbi->h_event_start_decoding);
1.858 + pbi->h_event_start_decoding = NULL;
1.859 +
1.860 + vpx_free(pbi->mb_row_di);
1.861 + pbi->mb_row_di = NULL ;
1.862 +
1.863 + vpx_free(pbi->de_thread_data);
1.864 + pbi->de_thread_data = NULL;
1.865 + }
1.866 +}
1.867 +
1.868 +void vp8mt_decode_mb_rows( VP8D_COMP *pbi, MACROBLOCKD *xd)
1.869 +{
1.870 + VP8_COMMON *pc = &pbi->common;
1.871 + unsigned int i;
1.872 + int j;
1.873 +
1.874 + int filter_level = pc->filter_level;
1.875 + YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
1.876 +
1.877 + if (filter_level)
1.878 + {
1.879 + /* Set above_row buffer to 127 for decoding first MB row */
1.880 + vpx_memset(pbi->mt_yabove_row[0] + VP8BORDERINPIXELS-1, 127, yv12_fb_new->y_width + 5);
1.881 + vpx_memset(pbi->mt_uabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5);
1.882 + vpx_memset(pbi->mt_vabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5);
1.883 +
1.884 + for (j=1; j<pc->mb_rows; j++)
1.885 + {
1.886 + vpx_memset(pbi->mt_yabove_row[j] + VP8BORDERINPIXELS-1, (unsigned char)129, 1);
1.887 + vpx_memset(pbi->mt_uabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1);
1.888 + vpx_memset(pbi->mt_vabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1);
1.889 + }
1.890 +
1.891 + /* Set left_col to 129 initially */
1.892 + for (j=0; j<pc->mb_rows; j++)
1.893 + {
1.894 + vpx_memset(pbi->mt_yleft_col[j], (unsigned char)129, 16);
1.895 + vpx_memset(pbi->mt_uleft_col[j], (unsigned char)129, 8);
1.896 + vpx_memset(pbi->mt_vleft_col[j], (unsigned char)129, 8);
1.897 + }
1.898 +
1.899 + /* Initialize the loop filter for this frame. */
1.900 + vp8_loop_filter_frame_init(pc, &pbi->mb, filter_level);
1.901 + }
1.902 + else
1.903 + vp8_setup_intra_recon_top_line(yv12_fb_new);
1.904 +
1.905 + setup_decoding_thread_data(pbi, xd, pbi->mb_row_di, pbi->decoding_thread_count);
1.906 +
1.907 + for (i = 0; i < pbi->decoding_thread_count; i++)
1.908 + sem_post(&pbi->h_event_start_decoding[i]);
1.909 +
1.910 + mt_decode_mb_rows(pbi, xd, 0);
1.911 +
1.912 + sem_wait(&pbi->h_event_end_decoding); /* add back for each frame */
1.913 +}
