michael@0: /*
michael@0:  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0:  *
michael@0:  *  Use of this source code is governed by a BSD-style license
michael@0:  *  that can be found in the LICENSE file in the root of the source
michael@0:  *  tree. An additional intellectual property rights grant can be found
michael@0:  *  in the file PATENTS.  All contributing project authors may
michael@0:  *  be found in the AUTHORS file in the root of the source tree.
michael@0:  */
michael@0: 
michael@0: 
michael@0: #include "vpx_config.h"
michael@0: #include "vp8_rtcd.h"
michael@0: #if !defined(WIN32) && CONFIG_OS_SUPPORT == 1
michael@0: # include <unistd.h>
michael@0: #endif
michael@0: #include "onyxd_int.h"
michael@0: #include "vpx_mem/vpx_mem.h"
michael@0: #include "vp8/common/threading.h"
michael@0: 
michael@0: #include "vp8/common/loopfilter.h"
michael@0: #include "vp8/common/extend.h"
michael@0: #include "vpx_ports/vpx_timer.h"
michael@0: #include "detokenize.h"
michael@0: #include "vp8/common/reconintra4x4.h"
michael@0: #include "vp8/common/reconinter.h"
michael@0: #include "vp8/common/setupintrarecon.h"
michael@0: #if CONFIG_ERROR_CONCEALMENT
michael@0: #include "error_concealment.h"
michael@0: #endif
michael@0: 
michael@0: #define CALLOC_ARRAY(p, n) CHECK_MEM_ERROR((p), vpx_calloc(sizeof(*(p)), (n)))
michael@0: #define CALLOC_ARRAY_ALIGNED(p, n, algn) do {                      \
michael@0:   CHECK_MEM_ERROR((p), vpx_memalign((algn), sizeof(*(p)) * (n)));  \
michael@0:   memset((p), 0, (n) * sizeof(*(p)));                              \
michael@0: } while (0)
michael@0: 
michael@0: 
michael@0: void vp8_mb_init_dequantizer(VP8D_COMP *pbi, MACROBLOCKD *xd);
michael@0: 
michael@0: static void setup_decoding_thread_data(VP8D_COMP *pbi, MACROBLOCKD *xd, MB_ROW_DEC *mbrd, int count)
michael@0: {
michael@0:     VP8_COMMON *const pc = & pbi->common;
michael@0:     int i;
michael@0: 
michael@0:     for (i = 0; i < count; i++)
michael@0:     {
michael@0:         MACROBLOCKD *mbd = &mbrd[i].mbd;
michael@0:         mbd->subpixel_predict        = xd->subpixel_predict;
michael@0:         mbd->subpixel_predict8x4     = xd->subpixel_predict8x4;
michael@0:         mbd->subpixel_predict8x8     = xd->subpixel_predict8x8;
michael@0:         mbd->subpixel_predict16x16   = xd->subpixel_predict16x16;
michael@0: 
michael@0:         mbd->mode_info_context = pc->mi   + pc->mode_info_stride * (i + 1);
michael@0:         mbd->mode_info_stride  = pc->mode_info_stride;
michael@0: 
michael@0:         mbd->frame_type = pc->frame_type;
michael@0:         mbd->pre = xd->pre;
michael@0:         mbd->dst = xd->dst;
michael@0: 
michael@0:         mbd->segmentation_enabled    = xd->segmentation_enabled;
michael@0:         mbd->mb_segement_abs_delta     = xd->mb_segement_abs_delta;
michael@0:         vpx_memcpy(mbd->segment_feature_data, xd->segment_feature_data, sizeof(xd->segment_feature_data));
michael@0: 
michael@0:         /*signed char ref_lf_deltas[MAX_REF_LF_DELTAS];*/
michael@0:         vpx_memcpy(mbd->ref_lf_deltas, xd->ref_lf_deltas, sizeof(xd->ref_lf_deltas));
michael@0:         /*signed char mode_lf_deltas[MAX_MODE_LF_DELTAS];*/
michael@0:         vpx_memcpy(mbd->mode_lf_deltas, xd->mode_lf_deltas, sizeof(xd->mode_lf_deltas));
michael@0:         /*unsigned char mode_ref_lf_delta_enabled;
michael@0:         unsigned char mode_ref_lf_delta_update;*/
michael@0:         mbd->mode_ref_lf_delta_enabled    = xd->mode_ref_lf_delta_enabled;
michael@0:         mbd->mode_ref_lf_delta_update    = xd->mode_ref_lf_delta_update;
michael@0: 
michael@0:         mbd->current_bc = &pbi->mbc[0];
michael@0: 
michael@0:         vpx_memcpy(mbd->dequant_y1_dc, xd->dequant_y1_dc, sizeof(xd->dequant_y1_dc));
michael@0:         vpx_memcpy(mbd->dequant_y1, xd->dequant_y1, sizeof(xd->dequant_y1));
michael@0:         vpx_memcpy(mbd->dequant_y2, xd->dequant_y2, sizeof(xd->dequant_y2));
michael@0:         vpx_memcpy(mbd->dequant_uv, xd->dequant_uv, sizeof(xd->dequant_uv));
michael@0: 
michael@0:         mbd->fullpixel_mask = 0xffffffff;
michael@0: 
michael@0:         if (pc->full_pixel)
michael@0:             mbd->fullpixel_mask = 0xfffffff8;
michael@0: 
michael@0:     }
michael@0: 
michael@0:     for (i = 0; i < pc->mb_rows; i++)
michael@0:         pbi->mt_current_mb_col[i] = -1;
michael@0: }
michael@0: 
michael@0: static void mt_decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd,
michael@0:                                  unsigned int mb_idx)
michael@0: {
michael@0:     MB_PREDICTION_MODE mode;
michael@0:     int i;
michael@0: #if CONFIG_ERROR_CONCEALMENT
michael@0:     int corruption_detected = 0;
michael@0: #endif
michael@0: 
michael@0:     if (xd->mode_info_context->mbmi.mb_skip_coeff)
michael@0:     {
michael@0:         vp8_reset_mb_tokens_context(xd);
michael@0:     }
michael@0:     else if (!vp8dx_bool_error(xd->current_bc))
michael@0:     {
michael@0:         int eobtotal;
michael@0:         eobtotal = vp8_decode_mb_tokens(pbi, xd);
michael@0: 
michael@0:         /* Special case:  Force the loopfilter to skip when eobtotal is zero */
michael@0:         xd->mode_info_context->mbmi.mb_skip_coeff = (eobtotal==0);
michael@0:     }
michael@0: 
michael@0:     mode = xd->mode_info_context->mbmi.mode;
michael@0: 
michael@0:     if (xd->segmentation_enabled)
michael@0:         vp8_mb_init_dequantizer(pbi, xd);
michael@0: 
michael@0: 
michael@0: #if CONFIG_ERROR_CONCEALMENT
michael@0: 
michael@0:     if(pbi->ec_active)
michael@0:     {
michael@0:         int throw_residual;
michael@0:         /* When we have independent partitions we can apply residual even
michael@0:          * though other partitions within the frame are corrupt.
michael@0:          */
michael@0:         throw_residual = (!pbi->independent_partitions &&
michael@0:                           pbi->frame_corrupt_residual);
michael@0:         throw_residual = (throw_residual || vp8dx_bool_error(xd->current_bc));
michael@0: 
michael@0:         if ((mb_idx >= pbi->mvs_corrupt_from_mb || throw_residual))
michael@0:         {
michael@0:             /* MB with corrupt residuals or corrupt mode/motion vectors.
michael@0:              * Better to use the predictor as reconstruction.
michael@0:              */
michael@0:             pbi->frame_corrupt_residual = 1;
michael@0:             vpx_memset(xd->qcoeff, 0, sizeof(xd->qcoeff));
michael@0:             vp8_conceal_corrupt_mb(xd);
michael@0: 
michael@0: 
michael@0:             corruption_detected = 1;
michael@0: 
michael@0:             /* force idct to be skipped for B_PRED and use the
michael@0:              * prediction only for reconstruction
michael@0:              * */
michael@0:             vpx_memset(xd->eobs, 0, 25);
michael@0:         }
michael@0:     }
michael@0: #endif
michael@0: 
michael@0:     /* do prediction */
michael@0:     if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME)
michael@0:     {
michael@0:         vp8_build_intra_predictors_mbuv_s(xd,
michael@0:                                           xd->recon_above[1],
michael@0:                                           xd->recon_above[2],
michael@0:                                           xd->recon_left[1],
michael@0:                                           xd->recon_left[2],
michael@0:                                           xd->recon_left_stride[1],
michael@0:                                           xd->dst.u_buffer, xd->dst.v_buffer,
michael@0:                                           xd->dst.uv_stride);
michael@0: 
michael@0:         if (mode != B_PRED)
michael@0:         {
michael@0:             vp8_build_intra_predictors_mby_s(xd,
michael@0:                                                  xd->recon_above[0],
michael@0:                                                  xd->recon_left[0],
michael@0:                                                  xd->recon_left_stride[0],
michael@0:                                                  xd->dst.y_buffer,
michael@0:                                                  xd->dst.y_stride);
michael@0:         }
michael@0:         else
michael@0:         {
michael@0:             short *DQC = xd->dequant_y1;
michael@0:             int dst_stride = xd->dst.y_stride;
michael@0: 
michael@0:             /* clear out residual eob info */
michael@0:             if(xd->mode_info_context->mbmi.mb_skip_coeff)
michael@0:                 vpx_memset(xd->eobs, 0, 25);
michael@0: 
michael@0:             intra_prediction_down_copy(xd, xd->recon_above[0] + 16);
michael@0: 
michael@0:             for (i = 0; i < 16; i++)
michael@0:             {
michael@0:                 BLOCKD *b = &xd->block[i];
michael@0:                 unsigned char *dst = xd->dst.y_buffer + b->offset;
michael@0:                 B_PREDICTION_MODE b_mode =
michael@0:                     xd->mode_info_context->bmi[i].as_mode;
michael@0:                 unsigned char *Above;
michael@0:                 unsigned char *yleft;
michael@0:                 int left_stride;
michael@0:                 unsigned char top_left;
michael@0: 
michael@0:                 /*Caution: For some b_mode, it needs 8 pixels (4 above + 4 above-right).*/
michael@0:                 if (i < 4 && pbi->common.filter_level)
michael@0:                     Above = xd->recon_above[0] + b->offset;
michael@0:                 else
michael@0:                     Above = dst - dst_stride;
michael@0: 
michael@0:                 if (i%4==0 && pbi->common.filter_level)
michael@0:                 {
michael@0:                     yleft = xd->recon_left[0] + i;
michael@0:                     left_stride = 1;
michael@0:                 }
michael@0:                 else
michael@0:                 {
michael@0:                     yleft = dst - 1;
michael@0:                     left_stride = dst_stride;
michael@0:                 }
michael@0: 
michael@0:                 if ((i==4 || i==8 || i==12) && pbi->common.filter_level)
michael@0:                     top_left = *(xd->recon_left[0] + i - 1);
michael@0:                 else
michael@0:                     top_left = Above[-1];
michael@0: 
michael@0:                 vp8_intra4x4_predict(Above, yleft, left_stride,
michael@0:                                      b_mode, dst, dst_stride, top_left);
michael@0: 
michael@0:                 if (xd->eobs[i] )
michael@0:                 {
michael@0:                     if (xd->eobs[i] > 1)
michael@0:                     {
michael@0:                         vp8_dequant_idct_add(b->qcoeff, DQC, dst, dst_stride);
michael@0:                     }
michael@0:                     else
michael@0:                     {
michael@0:                         vp8_dc_only_idct_add(b->qcoeff[0] * DQC[0],
michael@0:                                              dst, dst_stride, dst, dst_stride);
michael@0:                         vpx_memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0]));
michael@0:                     }
michael@0:                 }
michael@0:             }
michael@0:         }
michael@0:     }
michael@0:     else
michael@0:     {
michael@0:         vp8_build_inter_predictors_mb(xd);
michael@0:     }
michael@0: 
michael@0: 
michael@0: #if CONFIG_ERROR_CONCEALMENT
michael@0:     if (corruption_detected)
michael@0:     {
michael@0:         return;
michael@0:     }
michael@0: #endif
michael@0: 
michael@0:     if(!xd->mode_info_context->mbmi.mb_skip_coeff)
michael@0:     {
michael@0:         /* dequantization and idct */
michael@0:         if (mode != B_PRED)
michael@0:         {
michael@0:             short *DQC = xd->dequant_y1;
michael@0: 
michael@0:             if (mode != SPLITMV)
michael@0:             {
michael@0:                 BLOCKD *b = &xd->block[24];
michael@0: 
michael@0:                 /* do 2nd order transform on the dc block */
michael@0:                 if (xd->eobs[24] > 1)
michael@0:                 {
michael@0:                     vp8_dequantize_b(b, xd->dequant_y2);
michael@0: 
michael@0:                     vp8_short_inv_walsh4x4(&b->dqcoeff[0],
michael@0:                         xd->qcoeff);
michael@0:                     vpx_memset(b->qcoeff, 0, 16 * sizeof(b->qcoeff[0]));
michael@0:                 }
michael@0:                 else
michael@0:                 {
michael@0:                     b->dqcoeff[0] = b->qcoeff[0] * xd->dequant_y2[0];
michael@0:                     vp8_short_inv_walsh4x4_1(&b->dqcoeff[0],
michael@0:                         xd->qcoeff);
michael@0:                     vpx_memset(b->qcoeff, 0, 2 * sizeof(b->qcoeff[0]));
michael@0:                 }
michael@0: 
michael@0:                 /* override the dc dequant constant in order to preserve the
michael@0:                  * dc components
michael@0:                  */
michael@0:                 DQC = xd->dequant_y1_dc;
michael@0:             }
michael@0: 
michael@0:             vp8_dequant_idct_add_y_block
michael@0:                             (xd->qcoeff, DQC,
michael@0:                              xd->dst.y_buffer,
michael@0:                              xd->dst.y_stride, xd->eobs);
michael@0:         }
michael@0: 
michael@0:         vp8_dequant_idct_add_uv_block
michael@0:                         (xd->qcoeff+16*16, xd->dequant_uv,
michael@0:                          xd->dst.u_buffer, xd->dst.v_buffer,
michael@0:                          xd->dst.uv_stride, xd->eobs+16);
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void mt_decode_mb_rows(VP8D_COMP *pbi, MACROBLOCKD *xd, int start_mb_row)
michael@0: {
michael@0:     volatile const int *last_row_current_mb_col;
michael@0:     volatile int *current_mb_col;
michael@0:     int mb_row;
michael@0:     VP8_COMMON *pc = &pbi->common;
michael@0:     const int nsync = pbi->sync_range;
michael@0:     const int first_row_no_sync_above = pc->mb_cols + nsync;
michael@0:     int num_part = 1 << pbi->common.multi_token_partition;
michael@0:     int last_mb_row = start_mb_row;
michael@0: 
michael@0:     YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
michael@0:     YV12_BUFFER_CONFIG *yv12_fb_lst = pbi->dec_fb_ref[LAST_FRAME];
michael@0: 
michael@0:     int recon_y_stride = yv12_fb_new->y_stride;
michael@0:     int recon_uv_stride = yv12_fb_new->uv_stride;
michael@0: 
michael@0:     unsigned char *ref_buffer[MAX_REF_FRAMES][3];
michael@0:     unsigned char *dst_buffer[3];
michael@0:     int i;
michael@0:     int ref_fb_corrupted[MAX_REF_FRAMES];
michael@0: 
michael@0:     ref_fb_corrupted[INTRA_FRAME] = 0;
michael@0: 
michael@0:     for(i = 1; i < MAX_REF_FRAMES; i++)
michael@0:     {
michael@0:         YV12_BUFFER_CONFIG *this_fb = pbi->dec_fb_ref[i];
michael@0: 
michael@0:         ref_buffer[i][0] = this_fb->y_buffer;
michael@0:         ref_buffer[i][1] = this_fb->u_buffer;
michael@0:         ref_buffer[i][2] = this_fb->v_buffer;
michael@0: 
michael@0:         ref_fb_corrupted[i] = this_fb->corrupted;
michael@0:     }
michael@0: 
michael@0:     dst_buffer[0] = yv12_fb_new->y_buffer;
michael@0:     dst_buffer[1] = yv12_fb_new->u_buffer;
michael@0:     dst_buffer[2] = yv12_fb_new->v_buffer;
michael@0: 
michael@0:     xd->up_available = (start_mb_row != 0);
michael@0: 
michael@0:     for (mb_row = start_mb_row; mb_row < pc->mb_rows; mb_row += (pbi->decoding_thread_count + 1))
michael@0:     {
michael@0:        int recon_yoffset, recon_uvoffset;
michael@0:        int mb_col;
michael@0:        int filter_level;
michael@0:        loop_filter_info_n *lfi_n = &pc->lf_info;
michael@0: 
michael@0:        /* save last row processed by this thread */
michael@0:        last_mb_row = mb_row;
michael@0:        /* select bool coder for current partition */
michael@0:        xd->current_bc =  &pbi->mbc[mb_row%num_part];
michael@0: 
michael@0:        if (mb_row > 0)
michael@0:            last_row_current_mb_col = &pbi->mt_current_mb_col[mb_row -1];
michael@0:        else
michael@0:            last_row_current_mb_col = &first_row_no_sync_above;
michael@0: 
michael@0:        current_mb_col = &pbi->mt_current_mb_col[mb_row];
michael@0: 
michael@0:        recon_yoffset = mb_row * recon_y_stride * 16;
michael@0:        recon_uvoffset = mb_row * recon_uv_stride * 8;
michael@0: 
michael@0:        /* reset contexts */
michael@0:        xd->above_context = pc->above_context;
michael@0:        vpx_memset(xd->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
michael@0: 
michael@0:        xd->left_available = 0;
michael@0: 
michael@0:        xd->mb_to_top_edge = -((mb_row * 16)) << 3;
michael@0:        xd->mb_to_bottom_edge = ((pc->mb_rows - 1 - mb_row) * 16) << 3;
michael@0: 
michael@0:        if (pbi->common.filter_level)
michael@0:        {
michael@0:           xd->recon_above[0] = pbi->mt_yabove_row[mb_row] + 0*16 +32;
michael@0:           xd->recon_above[1] = pbi->mt_uabove_row[mb_row] + 0*8 +16;
michael@0:           xd->recon_above[2] = pbi->mt_vabove_row[mb_row] + 0*8 +16;
michael@0: 
michael@0:           xd->recon_left[0] = pbi->mt_yleft_col[mb_row];
michael@0:           xd->recon_left[1] = pbi->mt_uleft_col[mb_row];
michael@0:           xd->recon_left[2] = pbi->mt_vleft_col[mb_row];
michael@0: 
michael@0:           /* TODO: move to outside row loop */
michael@0:           xd->recon_left_stride[0] = 1;
michael@0:           xd->recon_left_stride[1] = 1;
michael@0:        }
michael@0:        else
michael@0:        {
michael@0:           xd->recon_above[0] = dst_buffer[0] + recon_yoffset;
michael@0:           xd->recon_above[1] = dst_buffer[1] + recon_uvoffset;
michael@0:           xd->recon_above[2] = dst_buffer[2] + recon_uvoffset;
michael@0: 
michael@0:           xd->recon_left[0] = xd->recon_above[0] - 1;
michael@0:           xd->recon_left[1] = xd->recon_above[1] - 1;
michael@0:           xd->recon_left[2] = xd->recon_above[2] - 1;
michael@0: 
michael@0:           xd->recon_above[0] -= xd->dst.y_stride;
michael@0:           xd->recon_above[1] -= xd->dst.uv_stride;
michael@0:           xd->recon_above[2] -= xd->dst.uv_stride;
michael@0: 
michael@0:           /* TODO: move to outside row loop */
michael@0:           xd->recon_left_stride[0] = xd->dst.y_stride;
michael@0:           xd->recon_left_stride[1] = xd->dst.uv_stride;
michael@0: 
michael@0:           setup_intra_recon_left(xd->recon_left[0], xd->recon_left[1],
michael@0:                                  xd->recon_left[2], xd->dst.y_stride,
michael@0:                                  xd->dst.uv_stride);
michael@0:        }
michael@0: 
michael@0:        for (mb_col = 0; mb_col < pc->mb_cols; mb_col++)
michael@0:        {
michael@0:            *current_mb_col = mb_col - 1;
michael@0: 
michael@0:            if ((mb_col & (nsync - 1)) == 0)
michael@0:            {
michael@0:                while (mb_col > (*last_row_current_mb_col - nsync))
michael@0:                {
michael@0:                    x86_pause_hint();
michael@0:                    thread_sleep(0);
michael@0:                }
michael@0:            }
michael@0: 
michael@0:            /* Distance of MB to the various image edges.
michael@0:             * These are specified to 8th pel as they are always
michael@0:             * compared to values that are in 1/8th pel units.
michael@0:             */
michael@0:            xd->mb_to_left_edge = -((mb_col * 16) << 3);
michael@0:            xd->mb_to_right_edge = ((pc->mb_cols - 1 - mb_col) * 16) << 3;
michael@0: 
michael@0:     #if CONFIG_ERROR_CONCEALMENT
michael@0:            {
michael@0:                int corrupt_residual =
michael@0:                            (!pbi->independent_partitions &&
michael@0:                            pbi->frame_corrupt_residual) ||
michael@0:                            vp8dx_bool_error(xd->current_bc);
michael@0:                if (pbi->ec_active &&
michael@0:                    (xd->mode_info_context->mbmi.ref_frame ==
michael@0:                                                     INTRA_FRAME) &&
michael@0:                    corrupt_residual)
michael@0:                {
michael@0:                    /* We have an intra block with corrupt
michael@0:                     * coefficients, better to conceal with an inter
michael@0:                     * block.
michael@0:                     * Interpolate MVs from neighboring MBs
michael@0:                     *
michael@0:                     * Note that for the first mb with corrupt
michael@0:                     * residual in a frame, we might not discover
michael@0:                     * that before decoding the residual. That
michael@0:                     * happens after this check, and therefore no
michael@0:                     * inter concealment will be done.
michael@0:                     */
michael@0:                    vp8_interpolate_motion(xd,
michael@0:                                           mb_row, mb_col,
michael@0:                                           pc->mb_rows, pc->mb_cols,
michael@0:                                           pc->mode_info_stride);
michael@0:                }
michael@0:            }
michael@0:     #endif
michael@0: 
michael@0: 
michael@0:            xd->dst.y_buffer = dst_buffer[0] + recon_yoffset;
michael@0:            xd->dst.u_buffer = dst_buffer[1] + recon_uvoffset;
michael@0:            xd->dst.v_buffer = dst_buffer[2] + recon_uvoffset;
michael@0: 
michael@0:            xd->pre.y_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][0] + recon_yoffset;
michael@0:            xd->pre.u_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][1] + recon_uvoffset;
michael@0:            xd->pre.v_buffer = ref_buffer[xd->mode_info_context->mbmi.ref_frame][2] + recon_uvoffset;
michael@0: 
michael@0:            /* propagate errors from reference frames */
michael@0:            xd->corrupted |= ref_fb_corrupted[xd->mode_info_context->mbmi.ref_frame];
michael@0: 
michael@0:            mt_decode_macroblock(pbi, xd, 0);
michael@0: 
michael@0:            xd->left_available = 1;
michael@0: 
michael@0:            /* check if the boolean decoder has suffered an error */
michael@0:            xd->corrupted |= vp8dx_bool_error(xd->current_bc);
michael@0: 
michael@0:            xd->recon_above[0] += 16;
michael@0:            xd->recon_above[1] += 8;
michael@0:            xd->recon_above[2] += 8;
michael@0: 
michael@0:            if (!pbi->common.filter_level)
michael@0:            {
michael@0:               xd->recon_left[0] += 16;
michael@0:               xd->recon_left[1] += 8;
michael@0:               xd->recon_left[2] += 8;
michael@0:            }
michael@0: 
michael@0:            if (pbi->common.filter_level)
michael@0:            {
michael@0:                int skip_lf = (xd->mode_info_context->mbmi.mode != B_PRED &&
michael@0:                                xd->mode_info_context->mbmi.mode != SPLITMV &&
michael@0:                                xd->mode_info_context->mbmi.mb_skip_coeff);
michael@0: 
michael@0:                const int mode_index = lfi_n->mode_lf_lut[xd->mode_info_context->mbmi.mode];
michael@0:                const int seg = xd->mode_info_context->mbmi.segment_id;
michael@0:                const int ref_frame = xd->mode_info_context->mbmi.ref_frame;
michael@0: 
michael@0:                filter_level = lfi_n->lvl[seg][ref_frame][mode_index];
michael@0: 
michael@0:                if( mb_row != pc->mb_rows-1 )
michael@0:                {
michael@0:                    /* Save decoded MB last row data for next-row decoding */
michael@0:                    vpx_memcpy((pbi->mt_yabove_row[mb_row + 1] + 32 + mb_col*16), (xd->dst.y_buffer + 15 * recon_y_stride), 16);
michael@0:                    vpx_memcpy((pbi->mt_uabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.u_buffer + 7 * recon_uv_stride), 8);
michael@0:                    vpx_memcpy((pbi->mt_vabove_row[mb_row + 1] + 16 + mb_col*8), (xd->dst.v_buffer + 7 * recon_uv_stride), 8);
michael@0:                }
michael@0: 
michael@0:                /* save left_col for next MB decoding */
michael@0:                if(mb_col != pc->mb_cols-1)
michael@0:                {
michael@0:                    MODE_INFO *next = xd->mode_info_context +1;
michael@0: 
michael@0:                    if (next->mbmi.ref_frame == INTRA_FRAME)
michael@0:                    {
michael@0:                        for (i = 0; i < 16; i++)
michael@0:                            pbi->mt_yleft_col[mb_row][i] = xd->dst.y_buffer [i* recon_y_stride + 15];
michael@0:                        for (i = 0; i < 8; i++)
michael@0:                        {
michael@0:                            pbi->mt_uleft_col[mb_row][i] = xd->dst.u_buffer [i* recon_uv_stride + 7];
michael@0:                            pbi->mt_vleft_col[mb_row][i] = xd->dst.v_buffer [i* recon_uv_stride + 7];
michael@0:                        }
michael@0:                    }
michael@0:                }
michael@0: 
michael@0:                /* loopfilter on this macroblock. */
michael@0:                if (filter_level)
michael@0:                {
michael@0:                    if(pc->filter_type == NORMAL_LOOPFILTER)
michael@0:                    {
michael@0:                        loop_filter_info lfi;
michael@0:                        FRAME_TYPE frame_type = pc->frame_type;
michael@0:                        const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level];
michael@0:                        lfi.mblim = lfi_n->mblim[filter_level];
michael@0:                        lfi.blim = lfi_n->blim[filter_level];
michael@0:                        lfi.lim = lfi_n->lim[filter_level];
michael@0:                        lfi.hev_thr = lfi_n->hev_thr[hev_index];
michael@0: 
michael@0:                        if (mb_col > 0)
michael@0:                            vp8_loop_filter_mbv
michael@0:                            (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
michael@0: 
michael@0:                        if (!skip_lf)
michael@0:                            vp8_loop_filter_bv
michael@0:                            (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
michael@0: 
michael@0:                        /* don't apply across umv border */
michael@0:                        if (mb_row > 0)
michael@0:                            vp8_loop_filter_mbh
michael@0:                            (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer, recon_y_stride, recon_uv_stride, &lfi);
michael@0: 
michael@0:                        if (!skip_lf)
michael@0:                            vp8_loop_filter_bh
michael@0:                            (xd->dst.y_buffer, xd->dst.u_buffer, xd->dst.v_buffer,  recon_y_stride, recon_uv_stride, &lfi);
michael@0:                    }
michael@0:                    else
michael@0:                    {
michael@0:                        if (mb_col > 0)
michael@0:                            vp8_loop_filter_simple_mbv
michael@0:                            (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]);
michael@0: 
michael@0:                        if (!skip_lf)
michael@0:                            vp8_loop_filter_simple_bv
michael@0:                            (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]);
michael@0: 
michael@0:                        /* don't apply across umv border */
michael@0:                        if (mb_row > 0)
michael@0:                            vp8_loop_filter_simple_mbh
michael@0:                            (xd->dst.y_buffer, recon_y_stride, lfi_n->mblim[filter_level]);
michael@0: 
michael@0:                        if (!skip_lf)
michael@0:                            vp8_loop_filter_simple_bh
michael@0:                            (xd->dst.y_buffer, recon_y_stride, lfi_n->blim[filter_level]);
michael@0:                    }
michael@0:                }
michael@0: 
michael@0:            }
michael@0: 
michael@0:            recon_yoffset += 16;
michael@0:            recon_uvoffset += 8;
michael@0: 
michael@0:            ++xd->mode_info_context;  /* next mb */
michael@0: 
michael@0:            xd->above_context++;
michael@0:        }
michael@0: 
michael@0:        /* adjust to the next row of mbs */
michael@0:        if (pbi->common.filter_level)
michael@0:        {
michael@0:            if(mb_row != pc->mb_rows-1)
michael@0:            {
michael@0:                int lasty = yv12_fb_lst->y_width + VP8BORDERINPIXELS;
michael@0:                int lastuv = (yv12_fb_lst->y_width>>1) + (VP8BORDERINPIXELS>>1);
michael@0: 
michael@0:                for (i = 0; i < 4; i++)
michael@0:                {
michael@0:                    pbi->mt_yabove_row[mb_row +1][lasty + i] = pbi->mt_yabove_row[mb_row +1][lasty -1];
michael@0:                    pbi->mt_uabove_row[mb_row +1][lastuv + i] = pbi->mt_uabove_row[mb_row +1][lastuv -1];
michael@0:                    pbi->mt_vabove_row[mb_row +1][lastuv + i] = pbi->mt_vabove_row[mb_row +1][lastuv -1];
michael@0:                }
michael@0:            }
michael@0:        }
michael@0:        else
michael@0:            vp8_extend_mb_row(yv12_fb_new, xd->dst.y_buffer + 16,
michael@0:                              xd->dst.u_buffer + 8, xd->dst.v_buffer + 8);
michael@0: 
michael@0:        /* last MB of row is ready just after extension is done */
michael@0:        *current_mb_col = mb_col + nsync;
michael@0: 
michael@0:        ++xd->mode_info_context;      /* skip prediction column */
michael@0:        xd->up_available = 1;
michael@0: 
michael@0:        /* since we have multithread */
michael@0:        xd->mode_info_context += xd->mode_info_stride * pbi->decoding_thread_count;
michael@0:     }
michael@0: 
michael@0:     /* signal end of frame decoding if this thread processed the last mb_row */
michael@0:     if (last_mb_row == (pc->mb_rows - 1))
michael@0:         sem_post(&pbi->h_event_end_decoding);
michael@0: 
michael@0: }
michael@0: 
michael@0: 
michael@0: static THREAD_FUNCTION thread_decoding_proc(void *p_data)
michael@0: {
michael@0:     int ithread = ((DECODETHREAD_DATA *)p_data)->ithread;
michael@0:     VP8D_COMP *pbi = (VP8D_COMP *)(((DECODETHREAD_DATA *)p_data)->ptr1);
michael@0:     MB_ROW_DEC *mbrd = (MB_ROW_DEC *)(((DECODETHREAD_DATA *)p_data)->ptr2);
michael@0:     ENTROPY_CONTEXT_PLANES mb_row_left_context;
michael@0: 
michael@0:     while (1)
michael@0:     {
michael@0:         if (pbi->b_multithreaded_rd == 0)
michael@0:             break;
michael@0: 
michael@0:         if (sem_wait(&pbi->h_event_start_decoding[ithread]) == 0)
michael@0:         {
michael@0:             if (pbi->b_multithreaded_rd == 0)
michael@0:                 break;
michael@0:             else
michael@0:             {
michael@0:                 MACROBLOCKD *xd = &mbrd->mbd;
michael@0:                 xd->left_context = &mb_row_left_context;
michael@0: 
michael@0:                 mt_decode_mb_rows(pbi, xd, ithread+1);
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     return 0 ;
michael@0: }
michael@0: 
michael@0: 
michael@0: void vp8_decoder_create_threads(VP8D_COMP *pbi)
michael@0: {
michael@0:     int core_count = 0;
michael@0:     unsigned int ithread;
michael@0: 
michael@0:     pbi->b_multithreaded_rd = 0;
michael@0:     pbi->allocated_decoding_thread_count = 0;
michael@0: 
michael@0:     /* limit decoding threads to the max number of token partitions */
michael@0:     core_count = (pbi->max_threads > 8) ? 8 : pbi->max_threads;
michael@0: 
michael@0:     /* limit decoding threads to the available cores */
michael@0:     if (core_count > pbi->common.processor_core_count)
michael@0:         core_count = pbi->common.processor_core_count;
michael@0: 
michael@0:     if (core_count > 1)
michael@0:     {
michael@0:         pbi->b_multithreaded_rd = 1;
michael@0:         pbi->decoding_thread_count = core_count - 1;
michael@0: 
michael@0:         CALLOC_ARRAY(pbi->h_decoding_thread, pbi->decoding_thread_count);
michael@0:         CALLOC_ARRAY(pbi->h_event_start_decoding, pbi->decoding_thread_count);
michael@0:         CALLOC_ARRAY_ALIGNED(pbi->mb_row_di, pbi->decoding_thread_count, 32);
michael@0:         CALLOC_ARRAY(pbi->de_thread_data, pbi->decoding_thread_count);
michael@0: 
michael@0:         for (ithread = 0; ithread < pbi->decoding_thread_count; ithread++)
michael@0:         {
michael@0:             sem_init(&pbi->h_event_start_decoding[ithread], 0, 0);
michael@0: 
michael@0:             vp8_setup_block_dptrs(&pbi->mb_row_di[ithread].mbd);
michael@0: 
michael@0:             pbi->de_thread_data[ithread].ithread  = ithread;
michael@0:             pbi->de_thread_data[ithread].ptr1     = (void *)pbi;
michael@0:             pbi->de_thread_data[ithread].ptr2     = (void *) &pbi->mb_row_di[ithread];
michael@0: 
michael@0:             pthread_create(&pbi->h_decoding_thread[ithread], 0, thread_decoding_proc, (&pbi->de_thread_data[ithread]));
michael@0:         }
michael@0: 
michael@0:         sem_init(&pbi->h_event_end_decoding, 0, 0);
michael@0: 
michael@0:         pbi->allocated_decoding_thread_count = pbi->decoding_thread_count;
michael@0:     }
michael@0: }
michael@0: 
michael@0: 
michael@0: void vp8mt_de_alloc_temp_buffers(VP8D_COMP *pbi, int mb_rows)
michael@0: {
michael@0:     int i;
michael@0: 
michael@0:     if (pbi->b_multithreaded_rd)
michael@0:     {
michael@0:             vpx_free(pbi->mt_current_mb_col);
michael@0:             pbi->mt_current_mb_col = NULL ;
michael@0: 
michael@0:         /* Free above_row buffers. */
michael@0:         if (pbi->mt_yabove_row)
michael@0:         {
michael@0:             for (i=0; i< mb_rows; i++)
michael@0:             {
michael@0:                     vpx_free(pbi->mt_yabove_row[i]);
michael@0:                     pbi->mt_yabove_row[i] = NULL ;
michael@0:             }
michael@0:             vpx_free(pbi->mt_yabove_row);
michael@0:             pbi->mt_yabove_row = NULL ;
michael@0:         }
michael@0: 
michael@0:         if (pbi->mt_uabove_row)
michael@0:         {
michael@0:             for (i=0; i< mb_rows; i++)
michael@0:             {
michael@0:                     vpx_free(pbi->mt_uabove_row[i]);
michael@0:                     pbi->mt_uabove_row[i] = NULL ;
michael@0:             }
michael@0:             vpx_free(pbi->mt_uabove_row);
michael@0:             pbi->mt_uabove_row = NULL ;
michael@0:         }
michael@0: 
michael@0:         if (pbi->mt_vabove_row)
michael@0:         {
michael@0:             for (i=0; i< mb_rows; i++)
michael@0:             {
michael@0:                     vpx_free(pbi->mt_vabove_row[i]);
michael@0:                     pbi->mt_vabove_row[i] = NULL ;
michael@0:             }
michael@0:             vpx_free(pbi->mt_vabove_row);
michael@0:             pbi->mt_vabove_row = NULL ;
michael@0:         }
michael@0: 
michael@0:         /* Free left_col buffers. */
michael@0:         if (pbi->mt_yleft_col)
michael@0:         {
michael@0:             for (i=0; i< mb_rows; i++)
michael@0:             {
michael@0:                     vpx_free(pbi->mt_yleft_col[i]);
michael@0:                     pbi->mt_yleft_col[i] = NULL ;
michael@0:             }
michael@0:             vpx_free(pbi->mt_yleft_col);
michael@0:             pbi->mt_yleft_col = NULL ;
michael@0:         }
michael@0: 
michael@0:         if (pbi->mt_uleft_col)
michael@0:         {
michael@0:             for (i=0; i< mb_rows; i++)
michael@0:             {
michael@0:                     vpx_free(pbi->mt_uleft_col[i]);
michael@0:                     pbi->mt_uleft_col[i] = NULL ;
michael@0:             }
michael@0:             vpx_free(pbi->mt_uleft_col);
michael@0:             pbi->mt_uleft_col = NULL ;
michael@0:         }
michael@0: 
michael@0:         if (pbi->mt_vleft_col)
michael@0:         {
michael@0:             for (i=0; i< mb_rows; i++)
michael@0:             {
michael@0:                     vpx_free(pbi->mt_vleft_col[i]);
michael@0:                     pbi->mt_vleft_col[i] = NULL ;
michael@0:             }
michael@0:             vpx_free(pbi->mt_vleft_col);
michael@0:             pbi->mt_vleft_col = NULL ;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: 
michael@0: void vp8mt_alloc_temp_buffers(VP8D_COMP *pbi, int width, int prev_mb_rows)
michael@0: {
michael@0:     VP8_COMMON *const pc = & pbi->common;
michael@0:     int i;
michael@0:     int uv_width;
michael@0: 
michael@0:     if (pbi->b_multithreaded_rd)
michael@0:     {
michael@0:         vp8mt_de_alloc_temp_buffers(pbi, prev_mb_rows);
michael@0: 
michael@0:         /* our internal buffers are always multiples of 16 */
michael@0:         if ((width & 0xf) != 0)
michael@0:             width += 16 - (width & 0xf);
michael@0: 
michael@0:         if (width < 640) pbi->sync_range = 1;
michael@0:         else if (width <= 1280) pbi->sync_range = 8;
michael@0:         else if (width <= 2560) pbi->sync_range =16;
michael@0:         else pbi->sync_range = 32;
michael@0: 
michael@0:         uv_width = width >>1;
michael@0: 
michael@0:         /* Allocate an int for each mb row. */
michael@0:         CALLOC_ARRAY(pbi->mt_current_mb_col, pc->mb_rows);
michael@0: 
michael@0:         /* Allocate memory for above_row buffers. */
michael@0:         CALLOC_ARRAY(pbi->mt_yabove_row, pc->mb_rows);
michael@0:         for (i = 0; i < pc->mb_rows; i++)
michael@0:             CHECK_MEM_ERROR(pbi->mt_yabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (width + (VP8BORDERINPIXELS<<1))));
michael@0: 
michael@0:         CALLOC_ARRAY(pbi->mt_uabove_row, pc->mb_rows);
michael@0:         for (i = 0; i < pc->mb_rows; i++)
michael@0:             CHECK_MEM_ERROR(pbi->mt_uabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS)));
michael@0: 
michael@0:         CALLOC_ARRAY(pbi->mt_vabove_row, pc->mb_rows);
michael@0:         for (i = 0; i < pc->mb_rows; i++)
michael@0:             CHECK_MEM_ERROR(pbi->mt_vabove_row[i], vpx_memalign(16,sizeof(unsigned char) * (uv_width + VP8BORDERINPIXELS)));
michael@0: 
michael@0:         /* Allocate memory for left_col buffers. */
michael@0:         CALLOC_ARRAY(pbi->mt_yleft_col, pc->mb_rows);
michael@0:         for (i = 0; i < pc->mb_rows; i++)
michael@0:             CHECK_MEM_ERROR(pbi->mt_yleft_col[i], vpx_calloc(sizeof(unsigned char) * 16, 1));
michael@0: 
michael@0:         CALLOC_ARRAY(pbi->mt_uleft_col, pc->mb_rows);
michael@0:         for (i = 0; i < pc->mb_rows; i++)
michael@0:             CHECK_MEM_ERROR(pbi->mt_uleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1));
michael@0: 
michael@0:         CALLOC_ARRAY(pbi->mt_vleft_col, pc->mb_rows);
michael@0:         for (i = 0; i < pc->mb_rows; i++)
michael@0:             CHECK_MEM_ERROR(pbi->mt_vleft_col[i], vpx_calloc(sizeof(unsigned char) * 8, 1));
michael@0:     }
michael@0: }
michael@0: 
michael@0: 
michael@0: void vp8_decoder_remove_threads(VP8D_COMP *pbi)
michael@0: {
michael@0:     /* shutdown MB Decoding thread; */
michael@0:     if (pbi->b_multithreaded_rd)
michael@0:     {
michael@0:         int i;
michael@0: 
michael@0:         pbi->b_multithreaded_rd = 0;
michael@0: 
michael@0:         /* allow all threads to exit */
michael@0:         for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
michael@0:         {
michael@0:             sem_post(&pbi->h_event_start_decoding[i]);
michael@0:             pthread_join(pbi->h_decoding_thread[i], NULL);
michael@0:         }
michael@0: 
michael@0:         for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
michael@0:         {
michael@0:             sem_destroy(&pbi->h_event_start_decoding[i]);
michael@0:         }
michael@0: 
michael@0:         sem_destroy(&pbi->h_event_end_decoding);
michael@0: 
michael@0:             vpx_free(pbi->h_decoding_thread);
michael@0:             pbi->h_decoding_thread = NULL;
michael@0: 
michael@0:             vpx_free(pbi->h_event_start_decoding);
michael@0:             pbi->h_event_start_decoding = NULL;
michael@0: 
michael@0:             vpx_free(pbi->mb_row_di);
michael@0:             pbi->mb_row_di = NULL ;
michael@0: 
michael@0:             vpx_free(pbi->de_thread_data);
michael@0:             pbi->de_thread_data = NULL;
michael@0:     }
michael@0: }
michael@0: 
michael@0: void vp8mt_decode_mb_rows( VP8D_COMP *pbi, MACROBLOCKD *xd)
michael@0: {
michael@0:     VP8_COMMON *pc = &pbi->common;
michael@0:     unsigned int i;
michael@0:     int j;
michael@0: 
michael@0:     int filter_level = pc->filter_level;
michael@0:     YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
michael@0: 
michael@0:     if (filter_level)
michael@0:     {
michael@0:         /* Set above_row buffer to 127 for decoding first MB row */
michael@0:         vpx_memset(pbi->mt_yabove_row[0] + VP8BORDERINPIXELS-1, 127, yv12_fb_new->y_width + 5);
michael@0:         vpx_memset(pbi->mt_uabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5);
michael@0:         vpx_memset(pbi->mt_vabove_row[0] + (VP8BORDERINPIXELS>>1)-1, 127, (yv12_fb_new->y_width>>1) +5);
michael@0: 
michael@0:         for (j=1; j<pc->mb_rows; j++)
michael@0:         {
michael@0:             vpx_memset(pbi->mt_yabove_row[j] + VP8BORDERINPIXELS-1, (unsigned char)129, 1);
michael@0:             vpx_memset(pbi->mt_uabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1);
michael@0:             vpx_memset(pbi->mt_vabove_row[j] + (VP8BORDERINPIXELS>>1)-1, (unsigned char)129, 1);
michael@0:         }
michael@0: 
michael@0:         /* Set left_col to 129 initially */
michael@0:         for (j=0; j<pc->mb_rows; j++)
michael@0:         {
michael@0:             vpx_memset(pbi->mt_yleft_col[j], (unsigned char)129, 16);
michael@0:             vpx_memset(pbi->mt_uleft_col[j], (unsigned char)129, 8);
michael@0:             vpx_memset(pbi->mt_vleft_col[j], (unsigned char)129, 8);
michael@0:         }
michael@0: 
michael@0:         /* Initialize the loop filter for this frame. */
michael@0:         vp8_loop_filter_frame_init(pc, &pbi->mb, filter_level);
michael@0:     }
michael@0:     else
michael@0:         vp8_setup_intra_recon_top_line(yv12_fb_new);
michael@0: 
michael@0:     setup_decoding_thread_data(pbi, xd, pbi->mb_row_di, pbi->decoding_thread_count);
michael@0: 
michael@0:     for (i = 0; i < pbi->decoding_thread_count; i++)
michael@0:         sem_post(&pbi->h_event_start_decoding[i]);
michael@0: 
michael@0:     mt_decode_mb_rows(pbi, xd, 0);
michael@0: 
michael@0:     sem_wait(&pbi->h_event_end_decoding);   /* add back for each frame */
michael@0: }