encodeframe.c

                    {
                        if (cpi->cyclic_refresh_map[map_index+mb_col] == 1)
                            cpi->cyclic_refresh_map[map_index+mb_col] = 0;
                    }
                    else
                        cpi->cyclic_refresh_map[map_index+mb_col] = 1;
                }
            }
        }

        cpi->tplist[mb_row].stop = *tp;

        // Increment pointer into gf usage flags structure.
        x->gf_active_ptr++;

        // Increment the activity mask pointers.
        x->mb_activity_ptr++;

        // adjust to the next column of macroblocks
        x->src.y_buffer += 16;
        x->src.u_buffer += 8;
        x->src.v_buffer += 8;

        recon_yoffset += 16;
        recon_uvoffset += 8;

        // skip to next mb
        xd->mode_info_context++;

        xd->prev_mode_info_context++;
        assert((xd->prev_mode_info_context - cpi->common.prev_mip)
            ==(xd->mode_info_context - cpi->common.mip));
        x->partition_info++;

        xd->above_context++;
#if CONFIG_MULTITHREAD
        if (cpi->b_multi_threaded != 0)
        {
            cpi->mt_current_mb_col[mb_row] = mb_col;
        }
#endif
    }

    //extend the recon for intra prediction
    vp8_extend_mb_row(
        &cm->yv12_fb[dst_fb_idx],
        xd->dst.y_buffer + 16,
        xd->dst.u_buffer + 8,
        xd->dst.v_buffer + 8);

    // this is to account for the border
    xd->prev_mode_info_context++;
    xd->mode_info_context++;
    x->partition_info++;

#if CONFIG_MULTITHREAD
    if ((cpi->b_multi_threaded != 0) && (mb_row == cm->mb_rows - 1))
    {
        sem_post(&cpi->h_event_end_encoding); /* signal frame encoding end */
    }
#endif


//#if CONFIG_SEGFEATURES
// debug output
#if DBG_PRNT_SEGMAP
    {
        FILE *statsfile;
        statsfile = fopen("segmap2.stt", "a");
        fprintf(statsfile, "\n" );
        fclose(statsfile);
    }
#endif
}
#endif /* CONFIG_SUPERBLOCKS */

void init_encode_frame_mb_context(VP8_COMP *cpi)
{
    MACROBLOCK *const x = & cpi->mb;
    VP8_COMMON *const cm = & cpi->common;
    MACROBLOCKD *const xd = & x->e_mbd;

    // GF active flags data structure
    x->gf_active_ptr = (signed char *)cpi->gf_active_flags;

    // Activity map pointer
    x->mb_activity_ptr = cpi->mb_activity_map;

    x->vector_range = 32;

    x->act_zbin_adj = 0;

    x->partition_info = x->pi;

    xd->mode_info_context = cm->mi;
    xd->mode_info_stride = cm->mode_info_stride;
    xd->prev_mode_info_context = cm->prev_mi;

    xd->frame_type = cm->frame_type;

    xd->frames_since_golden = cm->frames_since_golden;
    xd->frames_till_alt_ref_frame = cm->frames_till_alt_ref_frame;

    // reset intra mode contexts
    if (cm->frame_type == KEY_FRAME)
        vp8_init_mbmode_probs(cm);

    // Copy data over into macro block data sturctures.
    x->src = * cpi->Source;
    xd->pre = cm->yv12_fb[cm->lst_fb_idx];
    xd->dst = cm->yv12_fb[cm->new_fb_idx];

    // set up frame for intra coded blocks
    vp8_setup_intra_recon(&cm->yv12_fb[cm->new_fb_idx]);

    vp8_build_block_offsets(x);

    vp8_setup_block_dptrs(&x->e_mbd);

    vp8_setup_block_ptrs(x);

    xd->mode_info_context->mbmi.mode = DC_PRED;
    xd->mode_info_context->mbmi.uv_mode = DC_PRED;

    xd->left_context = &cm->left_context;

    vp8_zero(cpi->count_mb_ref_frame_usage)
    vp8_zero(cpi->ymode_count)
    vp8_zero(cpi->uv_mode_count)

    x->mvc = cm->fc.mvc;

    vpx_memset(cm->above_context, 0,
               sizeof(ENTROPY_CONTEXT_PLANES) * cm->mb_cols);


//#if CONFIG_COMPRED
// TODO... this will all need changing for new reference frame coding model
// in addition... ref_frame_cost should not be in the MACROBLOCKD structure as
// it is only referenced in the encoder.
//#endif

    xd->ref_frame_cost[INTRA_FRAME]   = vp8_cost_zero(cm->prob_intra_coded);

    // Special case treatment when GF and ARF are not sensible options for reference
    if (cpi->ref_frame_flags == VP8_LAST_FLAG)
    {
        xd->ref_frame_cost[LAST_FRAME]    = vp8_cost_one(cm->prob_intra_coded)
                                        + vp8_cost_zero(255);
        xd->ref_frame_cost[GOLDEN_FRAME]  = vp8_cost_one(cm->prob_intra_coded)
                                        + vp8_cost_one(255)
                                        + vp8_cost_zero(128);
        xd->ref_frame_cost[ALTREF_FRAME]  = vp8_cost_one(cm->prob_intra_coded)
                                        + vp8_cost_one(255)
                                        + vp8_cost_one(128);
    }
    else
    {
        xd->ref_frame_cost[LAST_FRAME]    = vp8_cost_one(cm->prob_intra_coded)
                                        + vp8_cost_zero(cm->prob_last_coded);
        xd->ref_frame_cost[GOLDEN_FRAME]  = vp8_cost_one(cm->prob_intra_coded)
                                        + vp8_cost_one(cm->prob_last_coded)
                                        + vp8_cost_zero(cm->prob_gf_coded);
        xd->ref_frame_cost[ALTREF_FRAME]  = vp8_cost_one(cm->prob_intra_coded)
                                        + vp8_cost_one(cm->prob_last_coded)
                                        + vp8_cost_one(cm->prob_gf_coded);
    }

    xd->fullpixel_mask = 0xffffffff;
    if(cm->full_pixel)
        xd->fullpixel_mask = 0xfffffff8;
}

static void encode_frame_internal(VP8_COMP *cpi)
{
    int mb_row;
    MACROBLOCK *const x = & cpi->mb;
    VP8_COMMON *const cm = & cpi->common;
    MACROBLOCKD *const xd = & x->e_mbd;

    TOKENEXTRA *tp = cpi->tok;
    int totalrate;

#if CONFIG_COMPRED
    // Compute a modified set of reference frame probabilities to use when
    // prediction fails. These are based on the current genreal estimates for
    // this frame which may be updated with each itteration of the recode loop.
    compute_mod_refprobs( cm );
#endif

//#if CONFIG_SEGFEATURES
// debug output
#if DBG_PRNT_SEGMAP
    {
        FILE *statsfile;
        statsfile = fopen("segmap2.stt", "a");
        fprintf(statsfile, "\n" );
        fclose(statsfile);
    }
#endif

    totalrate = 0;

    if (cpi->compressor_speed == 2)
    {
        if (cpi->oxcf.cpu_used < 0)
            cpi->Speed = -(cpi->oxcf.cpu_used);
        else
            vp8_auto_select_speed(cpi);
    }

    // Functions setup for all frame types so we can use MC in AltRef
    if (cm->mcomp_filter_type == SIXTAP)
    {
        xd->subpixel_predict        = SUBPIX_INVOKE(
                                        &cpi->common.rtcd.subpix, sixtap4x4);
        xd->subpixel_predict8x4     = SUBPIX_INVOKE(
                                        &cpi->common.rtcd.subpix, sixtap8x4);
        xd->subpixel_predict8x8     = SUBPIX_INVOKE(
                                        &cpi->common.rtcd.subpix, sixtap8x8);
        xd->subpixel_predict16x16   = SUBPIX_INVOKE(
                                        &cpi->common.rtcd.subpix, sixtap16x16);
#if CONFIG_DUALPRED
        xd->subpixel_predict_avg8x8 = SUBPIX_INVOKE(
                                        &cpi->common.rtcd.subpix, sixtap_avg8x8);
        xd->subpixel_predict_avg16x16 = SUBPIX_INVOKE(
                                        &cpi->common.rtcd.subpix, sixtap_avg16x16);
#endif /* CONFIG_DUALPRED */
    }
    else
    {
        xd->subpixel_predict        = SUBPIX_INVOKE(
                                        &cpi->common.rtcd.subpix, bilinear4x4);
        xd->subpixel_predict8x4     = SUBPIX_INVOKE(
                                        &cpi->common.rtcd.subpix, bilinear8x4);
        xd->subpixel_predict8x8     = SUBPIX_INVOKE(
                                        &cpi->common.rtcd.subpix, bilinear8x8);
        xd->subpixel_predict16x16   = SUBPIX_INVOKE(
                                      &cpi->common.rtcd.subpix, bilinear16x16);
#if CONFIG_DUALPRED
        xd->subpixel_predict_avg8x8 = SUBPIX_INVOKE(
                                      &cpi->common.rtcd.subpix, bilinear_avg8x8);
        xd->subpixel_predict_avg16x16 = SUBPIX_INVOKE(
                                      &cpi->common.rtcd.subpix, bilinear_avg16x16);
#endif /* CONFIG_DUALPRED */
    }

    // Reset frame count of inter 0,0 motion vector usage.
    cpi->inter_zz_count = 0;

    cpi->prediction_error = 0;
    cpi->intra_error = 0;
    cpi->skip_true_count = 0;
    cpi->skip_false_count = 0;

#if 0
    // Experimental code
    cpi->frame_distortion = 0;
    cpi->last_mb_distortion = 0;
#endif

    xd->mode_info_context = cm->mi;

    xd->prev_mode_info_context = cm->prev_mi;

    vp8_zero(cpi->MVcount);
    vp8_zero(cpi->coef_counts);

    vp8cx_frame_init_quantizer(cpi);

    vp8_initialize_rd_consts(cpi, cm->base_qindex + cm->y1dc_delta_q);
    vp8cx_initialize_me_consts(cpi, cm->base_qindex);

    if(cpi->oxcf.tuning == VP8_TUNE_SSIM)
    {
        // Initialize encode frame context.
        init_encode_frame_mb_context(cpi);

        // Build a frame level activity map
        build_activity_map(cpi);
    }

    // re-initencode frame context.
    init_encode_frame_mb_context(cpi);
#if CONFIG_DUALPRED
    cpi->rd_single_diff = cpi->rd_dual_diff = cpi->rd_hybrid_diff = 0;
    cpi->single_pred_count[0] = cpi->single_pred_count[1] = cpi->single_pred_count[2] = 0;
    cpi->dual_pred_count[0]   = cpi->dual_pred_count[1]   = cpi->dual_pred_count[2]   = 0;
#endif /* CONFIG_DUALPRED */

    {
        struct vpx_usec_timer  emr_timer;
        vpx_usec_timer_start(&emr_timer);

#if CONFIG_MULTITHREAD
        if (cpi->b_multi_threaded)
        {
            int i;

            vp8cx_init_mbrthread_data(cpi, x, cpi->mb_row_ei, 1,  cpi->encoding_thread_count);

            for (i = 0; i < cm->mb_rows; i++)
                cpi->mt_current_mb_col[i] = -1;

            for (i = 0; i < cpi->encoding_thread_count; i++)
            {
                sem_post(&cpi->h_event_start_encoding[i]);
            }

            for (mb_row = 0; mb_row < cm->mb_rows; mb_row += (cpi->encoding_thread_count + 1))
            {
                //vp8_zero(cm->left_context)

                tp = cpi->tok + mb_row * (cm->mb_cols * 16 * 24);

                encode_mb_row(cpi, cm, mb_row, x, xd, &tp, &totalrate);

                // adjust to the next row of mbs
                x->src.y_buffer += 16 * x->src.y_stride * (cpi->encoding_thread_count + 1) - 16 * cm->mb_cols;
                x->src.u_buffer +=  8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
                x->src.v_buffer +=  8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;

                xd->mode_info_context += xd->mode_info_stride
                                        * cpi->encoding_thread_count;
                xd->prev_mode_info_context += xd->mode_info_stride
                                            * cpi->encoding_thread_count;

                x->partition_info  += xd->mode_info_stride * cpi->encoding_thread_count;
                x->gf_active_ptr   += cm->mb_cols * cpi->encoding_thread_count;

            }

            sem_wait(&cpi->h_event_end_encoding); /* wait for other threads to finish */

            cpi->tok_count = 0;

            for (mb_row = 0; mb_row < cm->mb_rows; mb_row ++)
            {
                cpi->tok_count += cpi->tplist[mb_row].stop - cpi->tplist[mb_row].start;
            }

            for (i = 0; i < cpi->encoding_thread_count; i++)
            {
                totalrate += cpi->mb_row_ei[i].totalrate;
            }

        }
        else
#endif
        {
#if CONFIG_SUPERBLOCKS
            // for each superblock row in the image
            for (mb_row = 0; mb_row < cm->mb_rows; mb_row+=2)
             {
                int offset = cm->mb_cols - 1 + (cm->mb_cols & 0x1);

                encode_sb_row(cpi, cm, mb_row, x, xd, &tp, &totalrate);

                // adjust to the next row of SBs
                x->src.y_buffer += 16 * x->src.y_stride - 16 * offset;
                x->src.u_buffer += 8 * x->src.uv_stride - 8 * offset;
                x->src.v_buffer += 8 * x->src.uv_stride - 8 * offset;
            }
#else
            // for each macroblock row in the image
            for (mb_row = 0; mb_row < cm->mb_rows; mb_row++)
            {
                encode_mb_row(cpi, cm, mb_row, x, xd, &tp, &totalrate);

                // adjust to the next row of MBs
                x->src.y_buffer += 16 * x->src.y_stride - 16 * cm->mb_cols;
                x->src.u_buffer += 8 * x->src.uv_stride - 8 * cm->mb_cols;
                x->src.v_buffer += 8 * x->src.uv_stride - 8 * cm->mb_cols;
            }
#endif  // CONFIG_SUPERBLOCKS

            cpi->tok_count = tp - cpi->tok;

        }

        vpx_usec_timer_mark(&emr_timer);
        cpi->time_encode_mb_row += vpx_usec_timer_elapsed(&emr_timer);

    }

    // 256 rate units to the bit
    cpi->projected_frame_size = totalrate >> 8;   // projected_frame_size in units of BYTES

    // Make a note of the percentage MBs coded Intra.
    if (cm->frame_type == KEY_FRAME)
    {
        cpi->this_frame_percent_intra = 100;
    }
    else
    {
        int tot_modes;

        tot_modes = cpi->count_mb_ref_frame_usage[INTRA_FRAME]
                    + cpi->count_mb_ref_frame_usage[LAST_FRAME]
                    + cpi->count_mb_ref_frame_usage[GOLDEN_FRAME]
                    + cpi->count_mb_ref_frame_usage[ALTREF_FRAME];

        if (tot_modes)
            cpi->this_frame_percent_intra = cpi->count_mb_ref_frame_usage[INTRA_FRAME] * 100 / tot_modes;

    }

#if 0
    {
        int cnt = 0;
        int flag[2] = {0, 0};

        for (cnt = 0; cnt < MVPcount; cnt++)
        {
            if (cm->fc.pre_mvc[0][cnt] != cm->fc.mvc[0][cnt])
            {
                flag[0] = 1;
                vpx_memcpy(cm->fc.pre_mvc[0], cm->fc.mvc[0], MVPcount);
                break;
            }
        }

        for (cnt = 0; cnt < MVPcount; cnt++)
        {
            if (cm->fc.pre_mvc[1][cnt] != cm->fc.mvc[1][cnt])
            {
                flag[1] = 1;
                vpx_memcpy(cm->fc.pre_mvc[1], cm->fc.mvc[1], MVPcount);
                break;
            }
        }

        if (flag[0] || flag[1])
            vp8_build_component_cost_table(cpi->mb.mvcost, (const MV_CONTEXT *) cm->fc.mvc, flag);
    }
#endif

#if 0
    // Keep record of the total distortion this time around for future use
    cpi->last_frame_distortion = cpi->frame_distortion;
#endif

}

void vp8_encode_frame(VP8_COMP *cpi)
{
#if CONFIG_DUALPRED
    if (cpi->sf.RD)
    {
        int frame_type, pred_type;
        int redo = 0;

        /*
         * This code does a single RD pass over the whole frame assuming
         * either dual, single or hybrid prediction as per whatever has
         * worked best for that type of frame in the past.
         * It also predicts whether another coding mode would have worked
         * better that this coding mode. If that is the case, it remembers
         * that for subsequent frames. If the difference is above a certain
         * threshold, it will actually re-encode the current frame using
         * that different coding mode.
         */
        if (cpi->common.frame_type == KEY_FRAME)
            frame_type = 0;
        else if (cpi->is_src_frame_alt_ref && cpi->common.refresh_golden_frame)
            frame_type = 3;
        else if (cpi->common.refresh_golden_frame || cpi->common.refresh_alt_ref_frame)
            frame_type = 1;
        else
            frame_type = 2;

        if (cpi->rd_prediction_type_threshes[frame_type][1] >
                cpi->rd_prediction_type_threshes[frame_type][0] &&
            cpi->rd_prediction_type_threshes[frame_type][1] >
                cpi->rd_prediction_type_threshes[frame_type][2])
            pred_type = DUAL_PREDICTION_ONLY;
        else if (cpi->rd_prediction_type_threshes[frame_type][0] >
                    cpi->rd_prediction_type_threshes[frame_type][1] &&
                 cpi->rd_prediction_type_threshes[frame_type][0] >
                    cpi->rd_prediction_type_threshes[frame_type][2])
            pred_type = SINGLE_PREDICTION_ONLY;
        else
            pred_type = HYBRID_PREDICTION;

        cpi->common.dual_pred_mode = pred_type;
        encode_frame_internal(cpi);

        cpi->rd_single_diff /= cpi->common.MBs;
        cpi->rd_prediction_type_threshes[frame_type][0] += cpi->rd_single_diff;
        cpi->rd_prediction_type_threshes[frame_type][0] >>= 1;
        cpi->rd_dual_diff   /= cpi->common.MBs;
        cpi->rd_prediction_type_threshes[frame_type][1] += cpi->rd_dual_diff;
        cpi->rd_prediction_type_threshes[frame_type][1] >>= 1;
        cpi->rd_hybrid_diff /= cpi->common.MBs;
        cpi->rd_prediction_type_threshes[frame_type][2] += cpi->rd_hybrid_diff;
        cpi->rd_prediction_type_threshes[frame_type][2] >>= 1;

        /* FIXME make "100" (the threshold at which to re-encode the
         * current frame) a commandline option. */
        if (cpi->common.dual_pred_mode == SINGLE_PREDICTION_ONLY &&
            (cpi->rd_dual_diff >= 100 || cpi->rd_hybrid_diff >= 100))
        {
            redo = 1;
            cpi->common.dual_pred_mode = cpi->rd_dual_diff > cpi->rd_hybrid_diff ?
                        DUAL_PREDICTION_ONLY : HYBRID_PREDICTION;
        }
        else if (cpi->common.dual_pred_mode == DUAL_PREDICTION_ONLY &&
                 (cpi->rd_single_diff >= 100 || cpi->rd_hybrid_diff >= 100))
        {
            redo = 1;
            cpi->common.dual_pred_mode = cpi->rd_single_diff > cpi->rd_hybrid_diff ?
                        SINGLE_PREDICTION_ONLY : HYBRID_PREDICTION;
        }
        else if (cpi->common.dual_pred_mode == HYBRID_PREDICTION)
        {
            if (cpi->dual_pred_count[0] == 0 &&
                cpi->dual_pred_count[1] == 0 &&
                cpi->dual_pred_count[2] == 0)
            {
                cpi->common.dual_pred_mode = SINGLE_PREDICTION_ONLY;
            }
            else if (cpi->single_pred_count[0] == 0 &&
                     cpi->single_pred_count[1] == 0 &&
                     cpi->single_pred_count[2] == 0)
            {
                cpi->common.dual_pred_mode = DUAL_PREDICTION_ONLY;
            }
            else if (cpi->rd_single_diff >= 100 || cpi->rd_dual_diff >= 100)
            {
                redo = 1;
                cpi->common.dual_pred_mode = cpi->rd_single_diff > cpi->rd_dual_diff ?
                            SINGLE_PREDICTION_ONLY : DUAL_PREDICTION_ONLY;
            }
        }


        if (redo)
        {
            encode_frame_internal(cpi);

            if (cpi->common.dual_pred_mode == HYBRID_PREDICTION)
            {
                if (cpi->dual_pred_count[0] == 0 &&
                    cpi->dual_pred_count[1] == 0 &&
                    cpi->dual_pred_count[2] == 0)
                {
                    cpi->common.dual_pred_mode = SINGLE_PREDICTION_ONLY;
                }
                else if (cpi->single_pred_count[0] == 0 &&
                         cpi->single_pred_count[1] == 0 &&
                         cpi->single_pred_count[2] == 0)
                {
                    cpi->common.dual_pred_mode = DUAL_PREDICTION_ONLY;
                }
            }
        }
    }
    else
#endif /* CONFIG_DUALPRED */
    {
        encode_frame_internal(cpi);
    }

}

void vp8_setup_block_ptrs(MACROBLOCK *x)
{
    int r, c;
    int i;

    for (r = 0; r < 4; r++)
    {
        for (c = 0; c < 4; c++)
        {
            x->block[r*4+c].src_diff = x->src_diff + r * 4 * 16 + c * 4;
        }
    }

    for (r = 0; r < 2; r++)
    {
        for (c = 0; c < 2; c++)
        {
            x->block[16 + r*2+c].src_diff = x->src_diff + 256 + r * 4 * 8 + c * 4;
        }
    }


    for (r = 0; r < 2; r++)
    {
        for (c = 0; c < 2; c++)
        {
            x->block[20 + r*2+c].src_diff = x->src_diff + 320 + r * 4 * 8 + c * 4;
        }
    }

    x->block[24].src_diff = x->src_diff + 384;


    for (i = 0; i < 25; i++)
    {
        x->block[i].coeff = x->coeff + i * 16;
    }
}

void vp8_build_block_offsets(MACROBLOCK *x)
{
    int block = 0;
    int br, bc;

    vp8_build_block_doffsets(&x->e_mbd);

    // y blocks
    x->thismb_ptr = &x->thismb[0];
    for (br = 0; br < 4; br++)
    {
        for (bc = 0; bc < 4; bc++)
        {
            BLOCK *this_block = &x->block[block];
            //this_block->base_src = &x->src.y_buffer;
            //this_block->src_stride = x->src.y_stride;
            //this_block->src = 4 * br * this_block->src_stride + 4 * bc;
            this_block->base_src = &x->thismb_ptr;
            this_block->src_stride = 16;
            this_block->src = 4 * br * 16 + 4 * bc;
            ++block;
        }
    }

    // u blocks
    for (br = 0; br < 2; br++)
    {
        for (bc = 0; bc < 2; bc++)
        {
            BLOCK *this_block = &x->block[block];
            this_block->base_src = &x->src.u_buffer;
            this_block->src_stride = x->src.uv_stride;
            this_block->src = 4 * br * this_block->src_stride + 4 * bc;
            ++block;
        }
    }

    // v blocks
    for (br = 0; br < 2; br++)
    {
        for (bc = 0; bc < 2; bc++)
        {
            BLOCK *this_block = &x->block[block];
            this_block->base_src = &x->src.v_buffer;
            this_block->src_stride = x->src.uv_stride;
            this_block->src = 4 * br * this_block->src_stride + 4 * bc;
            ++block;
        }
    }
}

static void sum_intra_stats(VP8_COMP *cpi, MACROBLOCK *x)
{
    const MACROBLOCKD *xd = & x->e_mbd;
    const MB_PREDICTION_MODE m = xd->mode_info_context->mbmi.mode;
    const MB_PREDICTION_MODE uvm = xd->mode_info_context->mbmi.uv_mode;

#ifdef MODE_STATS
    const int is_key = cpi->common.frame_type == KEY_FRAME;

    ++ (is_key ? uv_modes : inter_uv_modes)[uvm];
    ++ uv_modes_y[m][uvm];

    if (m == B_PRED)
    {
        unsigned int *const bct = is_key ? b_modes : inter_b_modes;

        int b = 0;

        do
        {
            ++ bct[xd->block[b].bmi.as_mode];
        }
        while (++b < 16);
    }

    if(m==I8X8_PRED)
    {
        i8x8_modes[xd->block[0].bmi.as_mode]++;
        i8x8_modes[xd->block[2].bmi.as_mode]++;
        i8x8_modes[xd->block[8].bmi.as_mode]++;
        i8x8_modes[xd->block[10].bmi.as_mode]++;
    }
#endif

    ++cpi->ymode_count[m];
    ++cpi->uv_mode_count[uvm];

}

// Experimental stub function to create a per MB zbin adjustment based on
// some previously calculated measure of MB activity.
static void adjust_act_zbin( VP8_COMP *cpi, MACROBLOCK *x )
{
#if USE_ACT_INDEX
    x->act_zbin_adj = *(x->mb_activity_ptr);
#else
    int64_t a;
    int64_t b;
    int64_t act = *(x->mb_activity_ptr);

    // Apply the masking to the RD multiplier.
    a = act + 4*cpi->activity_avg;
    b = 4*act + cpi->activity_avg;

    if ( act > cpi->activity_avg )
        x->act_zbin_adj = (int)(((int64_t)b + (a>>1))/a) - 1;
    else
        x->act_zbin_adj = 1 - (int)(((int64_t)a + (b>>1))/b);
#endif
}

int vp8cx_encode_intra_macro_block(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t)
{
    int rate;

    if (cpi->sf.RD && cpi->compressor_speed != 2)
        vp8_rd_pick_intra_mode(cpi, x, &rate);
    else
        vp8_pick_intra_mode(cpi, x, &rate);

    if(cpi->oxcf.tuning == VP8_TUNE_SSIM)
    {
        adjust_act_zbin( cpi, x );
        vp8_update_zbin_extra(cpi, x);
    }

    if(x->e_mbd.mode_info_context->mbmi.mode == I8X8_PRED)
    {
        vp8_encode_intra8x8mby(IF_RTCD(&cpi->rtcd), x);
        vp8_encode_intra8x8mbuv(IF_RTCD(&cpi->rtcd), x);
    }
    else if (x->e_mbd.mode_info_context->mbmi.mode == B_PRED)
        vp8_encode_intra4x4mby(IF_RTCD(&cpi->rtcd), x);
    else
        vp8_encode_intra16x16mby(IF_RTCD(&cpi->rtcd), x);

    if(x->e_mbd.mode_info_context->mbmi.mode != I8X8_PRED)
        vp8_encode_intra16x16mbuv(IF_RTCD(&cpi->rtcd), x);
    sum_intra_stats(cpi, x);
    vp8_tokenize_mb(cpi, &x->e_mbd, t);
#if CONFIG_T8X8
        if ( get_seg_tx_type(&x->e_mbd,
                             x->e_mbd.mode_info_context->mbmi.segment_id)
             == TX_8X8 )
        {
            cpi->t8x8_count++;
        }
        else
            cpi->t4x4_count++;
#endif
    return rate;
}
#ifdef SPEEDSTATS
extern int cnt_pm;
#endif

extern void vp8_fix_contexts(MACROBLOCKD *x);

int vp8cx_encode_inter_macroblock
(
    VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t,
    int recon_yoffset, int recon_uvoffset
)
{
    VP8_COMMON *cm = &cpi->common;
    MACROBLOCKD *const xd = &x->e_mbd;
    int intra_error = 0;
    int rate;
    int distortion;
    unsigned char *segment_id = &xd->mode_info_context->mbmi.segment_id;
    int seg_ref_active;
#if CONFIG_COMPRED
     unsigned char ref_pred_flag;
#endif

    x->skip = 0;

    if (xd->segmentation_enabled)
        x->encode_breakout = cpi->segment_encode_breakout[*segment_id];
    else
        x->encode_breakout = cpi->oxcf.encode_breakout;

    if (cpi->sf.RD)
    {
        int zbin_mode_boost_enabled = cpi->zbin_mode_boost_enabled;
        int single, dual, hybrid;

        /* Are we using the fast quantizer for the mode selection? */
        if(cpi->sf.use_fastquant_for_pick)
        {
            cpi->mb.quantize_b      = QUANTIZE_INVOKE(&cpi->rtcd.quantize,
                                                      fastquantb);
            cpi->mb.quantize_b_pair = QUANTIZE_INVOKE(&cpi->rtcd.quantize,
                                                      fastquantb_pair);

            /* the fast quantizer does not use zbin_extra, so
             * do not recalculate */
            cpi->zbin_mode_boost_enabled = 0;
        }
        vp8_rd_pick_inter_mode(cpi, x, recon_yoffset, recon_uvoffset, &rate,
                               &distortion, &intra_error, &single, &dual, &hybrid);
#if CONFIG_DUALPRED
        cpi->rd_single_diff += single;
        cpi->rd_dual_diff   += dual;
        cpi->rd_hybrid_diff += hybrid;
        if (x->e_mbd.mode_info_context->mbmi.ref_frame &&
            x->e_mbd.mode_info_context->mbmi.mode != SPLITMV)
        {
            MB_MODE_INFO *t = &x->e_mbd.mode_info_context[-cpi->common.mode_info_stride].mbmi;
            MB_MODE_INFO *l = &x->e_mbd.mode_info_context[-1].mbmi;
            int cnt = (t->second_ref_frame != INTRA_FRAME) + (l->second_ref_frame != INTRA_FRAME);
            if (x->e_mbd.mode_info_context->mbmi.second_ref_frame == INTRA_FRAME)
                cpi->single_pred_count[cnt]++;
            else
                cpi->dual_pred_count[cnt]++;
        }
#endif /* CONFIG_DUALPRED */

        /* switch back to the regular quantizer for the encode */
        if (cpi->sf.improved_quant)
        {
            cpi->mb.quantize_b      = QUANTIZE_INVOKE(&cpi->rtcd.quantize,
                                                      quantb);
            cpi->mb.quantize_b_pair = QUANTIZE_INVOKE(&cpi->rtcd.quantize,
                                                      quantb_pair);
        }

        /* restore cpi->zbin_mode_boost_enabled */
        cpi->zbin_mode_boost_enabled = zbin_mode_boost_enabled;

    }
    else
        vp8_pick_inter_mode(cpi, x, recon_yoffset, recon_uvoffset, &rate,
                            &distortion, &intra_error);

    cpi->prediction_error += distortion;
    cpi->intra_error += intra_error;

    if(cpi->oxcf.tuning == VP8_TUNE_SSIM)
    {
        // Adjust the zbin based on this MB rate.
        adjust_act_zbin( cpi, x );
    }

#if 0
    // Experimental RD code
    cpi->frame_distortion += distortion;
    cpi->last_mb_distortion = distortion;
#endif

    // MB level adjutment to quantizer setup
    if (xd->segmentation_enabled)
    {
        // If cyclic update enabled
        if (cpi->cyclic_refresh_mode_enabled)
        {
            // Clear segment_id back to 0 if not coded (last frame 0,0)
            if ( (*segment_id == 1) &&
                 ( (xd->mode_info_context->mbmi.ref_frame != LAST_FRAME) ||
                   (xd->mode_info_context->mbmi.mode != ZEROMV) ) )
            {
                *segment_id = 0;

                /* segment_id changed, so update */
                vp8cx_mb_init_quantizer(cpi, x);
            }
        }
//#if CONFIG_SEGFEATURES
        else
        {
            //segfeature_test_function(cpi, xd);
#if DBG_PRNT_SEGMAP
            // Debug output
            {
                FILE *statsfile;
                statsfile = fopen("segmap2.stt", "a");

                fprintf(statsfile, "%2d%2d%2d   ",
                    *segment_id,
                    xd->mode_info_context->mbmi.ref_frame,
                    xd->mode_info_context->mbmi.mode );

                fclose(statsfile);
            }
#endif
        }
    }

    {
        // Experimental code. Special case for gf and arf zeromv modes.
        // Increase zbin size to supress noise
        cpi->zbin_mode_boost = 0;
        if (cpi->zbin_mode_boost_enabled)
        {
            if ( xd->mode_info_context->mbmi.ref_frame != INTRA_FRAME )
            {
                if (xd->mode_info_context->mbmi.mode == ZEROMV)
                {
                    if (xd->mode_info_context->mbmi.ref_frame != LAST_FRAME)
                        cpi->zbin_mode_boost = GF_ZEROMV_ZBIN_BOOST;
                    else
                        cpi->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST;
                }
                else if (xd->mode_info_context->mbmi.mode == SPLITMV)
                    cpi->zbin_mode_boost = 0;
                else
                    cpi->zbin_mode_boost = MV_ZBIN_BOOST;
            }
        }

        /* The fast quantizer doesn't use zbin_extra, only do so with
         * the regular quantizer. */
        if (cpi->sf.improved_quant)
            vp8_update_zbin_extra(cpi, x);
    }

//#if CONFIG_SEGFEATURES
    seg_ref_active = segfeature_active( xd, *segment_id, SEG_LVL_REF_FRAME );

#if CONFIG_COMPRED
    // SET VARIOUS PREDICTION FLAGS

    // Did the chosen reference frame match its predicted value.
    ref_pred_flag = ( (xd->mode_info_context->mbmi.ref_frame ==
                           get_pred_ref( cm, xd )) );
    set_pred_flag( xd, PRED_REF, ref_pred_flag );

#endif

    // If we have just a single reference frame coded for a segment then
    // exclude from the reference frame counts used to work out
    // probabilities. NOTE: At the moment we dont support custom trees
    // for the reference frame coding for each segment but this is a
    // possible future action.
    if ( !seg_ref_active ||
         ( ( check_segref( xd, *segment_id, INTRA_FRAME ) +
             check_segref( xd, *segment_id, LAST_FRAME ) +
             check_segref( xd, *segment_id, GOLDEN_FRAME ) +
             check_segref( xd, *segment_id, ALTREF_FRAME ) ) > 1 ) )
    {
// TODO this may not be a good idea as it makes sample size small and means
// the predictor functions cannot use data about most likely value only most
// likely unpredicted value.
//#if CONFIG_COMPRED
//        // Only update count for incorrectly predicted cases
//        if ( !ref_pred_flag )
//#endif
        {
            cpi->count_mb_ref_frame_usage
                [xd->mode_info_context->mbmi.ref_frame]++;
        }
    }

    if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME)
    {
        if (xd->mode_info_context->mbmi.mode == B_PRED)
        {
            vp8_encode_intra16x16mbuv(IF_RTCD(&cpi->rtcd), x);
            vp8_encode_intra4x4mby(IF_RTCD(&cpi->rtcd), x);
        }
        else if(xd->mode_info_context->mbmi.mode == I8X8_PRED)
        {
            vp8_encode_intra8x8mby(IF_RTCD(&cpi->rtcd), x);
            vp8_encode_intra8x8mbuv(IF_RTCD(&cpi->rtcd), x);
        }
        else
        {
            vp8_encode_intra16x16mbuv(IF_RTCD(&cpi->rtcd), x);
            vp8_encode_intra16x16mby(IF_RTCD(&cpi->rtcd), x);
        }
        sum_intra_stats(cpi, x);
    }
    else
    {
        int ref_fb_idx;

        if (xd->mode_info_context->mbmi.ref_frame == LAST_FRAME)
            ref_fb_idx = cpi->common.lst_fb_idx;
        else if (xd->mode_info_context->mbmi.ref_frame == GOLDEN_FRAME)
            ref_fb_idx = cpi->common.gld_fb_idx;
        else
            ref_fb_idx = cpi->common.alt_fb_idx;

        xd->pre.y_buffer = cpi->common.yv12_fb[ref_fb_idx].y_buffer + recon_yoffset;
        xd->pre.u_buffer = cpi->common.yv12_fb[ref_fb_idx].u_buffer + recon_uvoffset;
        xd->pre.v_buffer = cpi->common.yv12_fb[ref_fb_idx].v_buffer + recon_uvoffset;

#if CONFIG_DUALPRED
        if (xd->mode_info_context->mbmi.second_ref_frame) {
            int second_ref_fb_idx;

            if (xd->mode_info_context->mbmi.second_ref_frame == LAST_FRAME)
                second_ref_fb_idx = cpi->common.lst_fb_idx;
            else if (xd->mode_info_context->mbmi.second_ref_frame == GOLDEN_FRAME)
                second_ref_fb_idx = cpi->common.gld_fb_idx;