コード例 #1
0
static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h)
{
    MpegEncContext *const s = &h->s;
    const int mb_x    = s->mb_x;
    const int mb_y    = s->mb_y;
    const int mb_xy   = h->mb_xy;
    const int mb_type = s->current_picture.f.mb_type[mb_xy];
    uint8_t *dest[3];
    int linesize;
    int i, j, p;
    int *block_offset = &h->block_offset[0];
    const int transform_bypass = !SIMPLE && (s->qscale == 0 && h->sps.transform_bypass);
    const int plane_count      = (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) ? 3 : 1;

    for (p = 0; p < plane_count; p++) {
        dest[p] = s->current_picture.f.data[p] +
                  ((mb_x << PIXEL_SHIFT) + mb_y * s->linesize) * 16;
        s->vdsp.prefetch(dest[p] + (s->mb_x & 3) * 4 * s->linesize + (64 << PIXEL_SHIFT),
                         s->linesize, 4);
    }

    h->list_counts[mb_xy] = h->list_count;

    if (!SIMPLE && MB_FIELD) {
        linesize     = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
        block_offset = &h->block_offset[48];
        if (mb_y & 1) // FIXME move out of this function?
            for (p = 0; p < 3; p++)
                dest[p] -= s->linesize * 15;
        if (FRAME_MBAFF) {
            int list;
            for (list = 0; list < h->list_count; list++) {
                if (!USES_LIST(mb_type, list))
                    continue;
                if (IS_16X16(mb_type)) {
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
                    fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (s->mb_y & 1), 1);
                } else {
                    for (i = 0; i < 16; i += 4) {
                        int ref = h->ref_cache[list][scan8[i]];
                        if (ref >= 0)
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2,
                                           8, (16 + ref) ^ (s->mb_y & 1), 1);
                    }
                }
            }
        }
    } else {
        linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
    }

    if (!SIMPLE && IS_INTRA_PCM(mb_type)) {
        if (PIXEL_SHIFT) {
            const int bit_depth = h->sps.bit_depth_luma;
            GetBitContext gb;
            init_get_bits(&gb, (uint8_t *)h->mb, 768 * bit_depth);

            for (p = 0; p < plane_count; p++)
                for (i = 0; i < 16; i++) {
                    uint16_t *tmp = (uint16_t *)(dest[p] + i * linesize);
                    for (j = 0; j < 16; j++)
                        tmp[j] = get_bits(&gb, bit_depth);
                }
        } else {
            for (p = 0; p < plane_count; p++)
                for (i = 0; i < 16; i++)
                    memcpy(dest[p] + i * linesize,
                           (uint8_t *)h->mb + p * 256 + i * 16, 16);
        }
    } else {
        if (IS_INTRA(mb_type)) {
            if (h->deblocking_filter)
                xchg_mb_border(h, dest[0], dest[1], dest[2], linesize,
                               linesize, 1, 1, SIMPLE, PIXEL_SHIFT);

            for (p = 0; p < plane_count; p++)
                hl_decode_mb_predict_luma(h, mb_type, 1, SIMPLE,
                                          transform_bypass, PIXEL_SHIFT,
                                          block_offset, linesize, dest[p], p);

            if (h->deblocking_filter)
                xchg_mb_border(h, dest[0], dest[1], dest[2], linesize,
                               linesize, 0, 1, SIMPLE, PIXEL_SHIFT);
        } else {
            FUNC(hl_motion_444)(h, dest[0], dest[1], dest[2],
                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
                      h->h264dsp.weight_h264_pixels_tab,
                      h->h264dsp.biweight_h264_pixels_tab);
        }

        for (p = 0; p < plane_count; p++)
            hl_decode_mb_idct_luma(h, mb_type, 1, SIMPLE, transform_bypass,
                                   PIXEL_SHIFT, block_offset, linesize,
                                   dest[p], p);
    }
    if (h->cbp || IS_INTRA(mb_type)) {
        s->dsp.clear_blocks(h->mb);
        s->dsp.clear_blocks(h->mb + (24 * 16 << PIXEL_SHIFT));
    }
}
コード例 #2
0
static av_noinline void FUNC(hl_decode_mb)(H264Context *h)
{
    MpegEncContext *const s = &h->s;
    const int mb_x    = s->mb_x;
    const int mb_y    = s->mb_y;
    const int mb_xy   = h->mb_xy;
    const int mb_type = s->current_picture.f.mb_type[mb_xy];
    uint8_t *dest_y, *dest_cb, *dest_cr;
    int linesize, uvlinesize /*dct_offset*/;
    int i, j;
    int *block_offset = &h->block_offset[0];
    const int transform_bypass = !SIMPLE && (s->qscale == 0 && h->sps.transform_bypass);
    /* is_h264 should always be true if SVQ3 is disabled. */
    const int is_h264 = !CONFIG_SVQ3_DECODER || SIMPLE || s->codec_id == AV_CODEC_ID_H264;
    void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
    const int block_h   = 16 >> s->chroma_y_shift;
    const int chroma422 = CHROMA422;

    dest_y  = s->current_picture.f.data[0] + ((mb_x << PIXEL_SHIFT)     + mb_y * s->linesize)  * 16;
    dest_cb = s->current_picture.f.data[1] +  (mb_x << PIXEL_SHIFT) * 8 + mb_y * s->uvlinesize * block_h;
    dest_cr = s->current_picture.f.data[2] +  (mb_x << PIXEL_SHIFT) * 8 + mb_y * s->uvlinesize * block_h;

    s->vdsp.prefetch(dest_y  + (s->mb_x & 3) * 4 * s->linesize   + (64 << PIXEL_SHIFT), s->linesize,       4);
    s->vdsp.prefetch(dest_cb + (s->mb_x & 7)     * s->uvlinesize + (64 << PIXEL_SHIFT), dest_cr - dest_cb, 2);

    h->list_counts[mb_xy] = h->list_count;

    if (!SIMPLE && MB_FIELD) {
        linesize     = h->mb_linesize = s->linesize * 2;
        uvlinesize   = h->mb_uvlinesize = s->uvlinesize * 2;
        block_offset = &h->block_offset[48];
        if (mb_y & 1) { // FIXME move out of this function?
            dest_y  -= s->linesize * 15;
            dest_cb -= s->uvlinesize * (block_h - 1);
            dest_cr -= s->uvlinesize * (block_h - 1);
        }
        if (FRAME_MBAFF) {
            int list;
            for (list = 0; list < h->list_count; list++) {
                if (!USES_LIST(mb_type, list))
                    continue;
                if (IS_16X16(mb_type)) {
                    int8_t *ref = &h->ref_cache[list][scan8[0]];
                    fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (s->mb_y & 1), 1);
                } else {
                    for (i = 0; i < 16; i += 4) {
                        int ref = h->ref_cache[list][scan8[i]];
                        if (ref >= 0)
                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2,
                                           8, (16 + ref) ^ (s->mb_y & 1), 1);
                    }
                }
            }
        }
    } else {
        linesize   = h->mb_linesize   = s->linesize;
        uvlinesize = h->mb_uvlinesize = s->uvlinesize;
        // dct_offset = s->linesize * 16;
    }

    if (!SIMPLE && IS_INTRA_PCM(mb_type)) {
        const int bit_depth = h->sps.bit_depth_luma;
        if (PIXEL_SHIFT) {
            int j;
            GetBitContext gb;
            init_get_bits(&gb, (uint8_t *)h->mb,
                          ff_h264_mb_sizes[h->sps.chroma_format_idc] * bit_depth);

            for (i = 0; i < 16; i++) {
                uint16_t *tmp_y = (uint16_t *)(dest_y + i * linesize);
                for (j = 0; j < 16; j++)
                    tmp_y[j] = get_bits(&gb, bit_depth);
            }
            if (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
                if (!h->sps.chroma_format_idc) {
                    for (i = 0; i < block_h; i++) {
                        uint16_t *tmp_cb = (uint16_t *)(dest_cb + i * uvlinesize);
                        uint16_t *tmp_cr = (uint16_t *)(dest_cr + i * uvlinesize);
                        for (j = 0; j < 8; j++) {
                            tmp_cb[j] = tmp_cr[j] = 1 << (bit_depth - 1);
                        }
                    }
                } else {
                    for (i = 0; i < block_h; i++) {
                        uint16_t *tmp_cb = (uint16_t *)(dest_cb + i * uvlinesize);
                        for (j = 0; j < 8; j++)
                            tmp_cb[j] = get_bits(&gb, bit_depth);
                    }
                    for (i = 0; i < block_h; i++) {
                        uint16_t *tmp_cr = (uint16_t *)(dest_cr + i * uvlinesize);
                        for (j = 0; j < 8; j++)
                            tmp_cr[j] = get_bits(&gb, bit_depth);
                    }
                }
            }
        } else {
            for (i = 0; i < 16; i++)
                memcpy(dest_y + i * linesize, (uint8_t *)h->mb + i * 16, 16);
            if (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
                if (!h->sps.chroma_format_idc) {
                    for (i = 0; i < 8; i++) {
                        memset(dest_cb + i*uvlinesize, 1 << (bit_depth - 1), 8);
                        memset(dest_cr + i*uvlinesize, 1 << (bit_depth - 1), 8);
                    }
                } else {
                    uint8_t *src_cb = (uint8_t *)h->mb + 256;
                    uint8_t *src_cr = (uint8_t *)h->mb + 256 + block_h * 8;
                    for (i = 0; i < block_h; i++) {
                        memcpy(dest_cb + i * uvlinesize, src_cb + i * 8, 8);
                        memcpy(dest_cr + i * uvlinesize, src_cr + i * 8, 8);
                    }
                }
            }
        }
    } else {
        if (IS_INTRA(mb_type)) {
            if (h->deblocking_filter)
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
                               uvlinesize, 1, 0, SIMPLE, PIXEL_SHIFT);

            if (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
                h->hpc.pred8x8[h->chroma_pred_mode](dest_cb, uvlinesize);
                h->hpc.pred8x8[h->chroma_pred_mode](dest_cr, uvlinesize);
            }

            hl_decode_mb_predict_luma(h, mb_type, is_h264, SIMPLE,
                                      transform_bypass, PIXEL_SHIFT,
                                      block_offset, linesize, dest_y, 0);

            if (h->deblocking_filter)
                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
                               uvlinesize, 0, 0, SIMPLE, PIXEL_SHIFT);
        } else if (is_h264) {
            if (chroma422) {
                FUNC(hl_motion_422)(h, dest_y, dest_cb, dest_cr,
                              s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
                              s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
                              h->h264dsp.weight_h264_pixels_tab,
                              h->h264dsp.biweight_h264_pixels_tab);
            } else {
                FUNC(hl_motion_420)(h, dest_y, dest_cb, dest_cr,
                              s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
                              s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
                              h->h264dsp.weight_h264_pixels_tab,
                              h->h264dsp.biweight_h264_pixels_tab);
            }
        }

        hl_decode_mb_idct_luma(h, mb_type, is_h264, SIMPLE, transform_bypass,
                               PIXEL_SHIFT, block_offset, linesize, dest_y, 0);

        if ((SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) &&
            (h->cbp & 0x30)) {
            uint8_t *dest[2] = { dest_cb, dest_cr };
            if (transform_bypass) {
                if (IS_INTRA(mb_type) && h->sps.profile_idc == 244 &&
                    (h->chroma_pred_mode == VERT_PRED8x8 ||
                     h->chroma_pred_mode == HOR_PRED8x8)) {
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0],
                                                            block_offset + 16,
                                                            h->mb + (16 * 16 * 1 << PIXEL_SHIFT),
                                                            uvlinesize);
                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1],
                                                            block_offset + 32,
                                                            h->mb + (16 * 16 * 2 << PIXEL_SHIFT),
                                                            uvlinesize);
                } else {
                    idct_add = s->dsp.add_pixels4;
                    for (j = 1; j < 3; j++) {
                        for (i = j * 16; i < j * 16 + 4; i++)
                            if (h->non_zero_count_cache[scan8[i]] ||
                                dctcoef_get(h->mb, PIXEL_SHIFT, i * 16))
                                idct_add(dest[j - 1] + block_offset[i],
                                         h->mb + (i * 16 << PIXEL_SHIFT),
                                         uvlinesize);
                        if (chroma422) {
                            for (i = j * 16 + 4; i < j * 16 + 8; i++)
                                if (h->non_zero_count_cache[scan8[i + 4]] ||
                                    dctcoef_get(h->mb, PIXEL_SHIFT, i * 16))
                                    idct_add(dest[j - 1] + block_offset[i + 4],
                                             h->mb + (i * 16 << PIXEL_SHIFT),
                                             uvlinesize);
                        }
                    }
                }
            } else {
                if (is_h264) {
                    int qp[2];
                    if (chroma422) {
                        qp[0] = h->chroma_qp[0] + 3;
                        qp[1] = h->chroma_qp[1] + 3;
                    } else {
                        qp[0] = h->chroma_qp[0];
                        qp[1] = h->chroma_qp[1];
                    }
                    if (h->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 0]])
                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16 * 16 * 1 << PIXEL_SHIFT),
                                                               h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][qp[0]][0]);
                    if (h->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 1]])
                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16 * 16 * 2 << PIXEL_SHIFT),
                                                               h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][qp[1]][0]);
                    h->h264dsp.h264_idct_add8(dest, block_offset,
                                              h->mb, uvlinesize,
                                              h->non_zero_count_cache);
                } else if (CONFIG_SVQ3_DECODER) {
                    h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16 * 16 * 1,
                                                           h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][h->chroma_qp[0]][0]);
                    h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16 * 16 * 2,
                                                           h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][h->chroma_qp[1]][0]);
                    for (j = 1; j < 3; j++) {
                        for (i = j * 16; i < j * 16 + 4; i++)
                            if (h->non_zero_count_cache[scan8[i]] || h->mb[i * 16]) {
                                uint8_t *const ptr = dest[j - 1] + block_offset[i];
                                ff_svq3_add_idct_c(ptr, h->mb + i * 16,
                                                   uvlinesize,
                                                   ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
                            }
                    }
                }
            }
        }
    }
    if (h->cbp || IS_INTRA(mb_type)) {
        s->dsp.clear_blocks(h->mb);
        s->dsp.clear_blocks(h->mb + (24 * 16 << PIXEL_SHIFT));
    }
}
コード例 #3
0
__tcsm1_main int main() {

  int c;
  S32I2M(xr16,3);
  c = i_la(_gp); 
 
  int fifo_rp = 0;
  unsigned int XCHGtmp;
  H264_Frame_GlbARGs *dFRM = (H264_Frame_GlbARGs *)TCSM1_FRM_ARGS;
  H264_MB_DecARGs *dMB_aux = (H264_MB_DecARGs *)TCSM1_MBARGS_BUF;
  H264_MB_DecARGs *dMB2 = (H264_MB_DecARGs *)TCSM1_MBARGS_BUF2;
  H264_AUX_T *AUX_T = (H264_AUX_T *)TCSM1_AUX_T;
  MDMA_DesNode *MDMA1_TRAN = (MDMA_DesNode *)TCSM1_GP1_TRAN_CHAIN;
  MDMA_DesNode *MDMA1_ARG = (MDMA_DesNode *)TCSM1_GP1_ARG_CHAIN;
  H264_MB_DecARGs *dMBsrc;

  fifo_wp = (int *)TCSM1_FIFO_WP;
  dMBsrc = (H264_MB_DecARGs *)(dFRM->dMB_baseaddr_aux);
  int mb_num;
  int mb_start;
  int total_mbs;    

  mb_start = dFRM->first_mb_in_frame;
  total_mbs = dFRM->mb_width * dFRM->mb_height;

  int i;

  uint16_t *intra_pred4x4_top_ptr[2];
  intra_pred4x4_top_ptr[0]=intra_pred4x4_top[0];
  intra_pred4x4_top_ptr[1]=intra_pred4x4_top[1];
  uint16_t *intra_pred4x4_left_ptr[2];
  intra_pred4x4_left_ptr[0]=intra_pred4x4_left[0];
  intra_pred4x4_left_ptr[1]=intra_pred4x4_left[1];

  uint8_t mb_x_d1, mb_y_d1;
  uint8_t mb_x_d2, mb_y_d2;
  mb_x_d1 = mb_y_d1 = mb_x_d2 = mb_y_d2 =0;

  AUX_PMON_CREAT(mc);
  AUX_PMON_CREAT(idct);
  AUX_PMON_CREAT(dblk);
  AUX_PMON_CREAT(intra);
  AUX_PMON_CREAT(mdma);
  AUX_PMON_CREAT(sync);

  do{
  }while(fifo_rp == *fifo_wp);
  fifo_rp++;

  MDMA1_ARG->TSA = get_phy_addr_aux((uint32_t)dMBsrc);
  MDMA1_ARG->TDA = (uint32_t)dMB_aux;
  MDMA1_ARG->STRD = MDMA_STRD(64, 64);
  MDMA1_ARG->UNIT = MDMA_UNIT(1,64,(sizeof(H264_MB_DecARGs)));
  SET_MDMA1_DHA((uint32_t)TCSM1_GP1_ARG_CHAIN);
  MDMA1_RUN();
  dMBsrc++;
  POLLING_MDMA1_END();

  char bakup_x0 = dFRM->mb_width-1;
  char bakup_x1 = 0;
  int mb_type_last= 0;

  volatile unsigned int *infar = (unsigned int *)TCSM1_H264_DBLK_INFAR;
  
  uint8_t non_zero_count_cache_luma_last[16];

  for ( mb_num = mb_start; (mb_num < total_mbs+2); mb_num ++ ) {

    AUX_PMON_ON(sync);

    do{
    }while(fifo_rp == *fifo_wp);
    fifo_rp++;

    int gp1_tran_start = 0;
    int gp1_tran_len   = sizeof(H264_MB_DecARGs);
    if (dMB_aux->next_mb_no_weight_flag) {
      gp1_tran_start = (53 << 2);
      gp1_tran_len -= (53 << 2);
    }
    if (dMB_aux->next_mb_skip_flag) {
      gp1_tran_len -= (192 << 2);
    }

    AUX_PMON_OFF(sync);

    AUX_PMON_ON(mdma);
    AUX_PMON_OFF(mdma);

    AUX_PMON_ON(intra);

    MDMA1_ARG->TSA = (get_phy_addr_aux((uint32_t)dMBsrc) + gp1_tran_start);
    MDMA1_ARG->TDA = ((uint32_t)dMB2 + gp1_tran_start);
    MDMA1_ARG->STRD = MDMA_STRD(64, 64);
    MDMA1_ARG->UNIT = MDMA_UNIT(1,64,gp1_tran_len);
    dMBsrc++;

    POLLING_MDMA1_END(); //ensure curr dblked MB has been trans out

    SET_MDMA1_DHA((uint32_t)TCSM1_GP1_ARG_CHAIN);
    MDMA1_RUN();

    int mb_x= dMB_aux->mb_x;
    int mb_y= dMB_aux->mb_y;
    int mb_type= dMB_aux->mb_type;

    if (dMB_aux->curr_mb_skip_flag)
      for(i=0; i<24; i++)
	dMB_aux->mb[i*16] = 0;
 
    AUX_T->mc_des_dirty=0;

    if(IS_INTRA(mb_type_last)){
      // chroma predict
      Intra_pred_chroma(dMB_aux->chroma_pred_mode, AUX_T->h264_urecon[1], 
			AUX_T->h264_urecon[0] + MB_CHROM_WIDTH, TCSM1_BOTTOM_U_13PIX+4);
      Intra_pred_chroma(dMB_aux->chroma_pred_mode, AUX_T->h264_vrecon[1], 
			AUX_T->h264_vrecon[0] + MB_CHROM_WIDTH, TCSM1_BOTTOM_V_13PIX+4);

      // luma predict
      if(IS_INTRA4x4(mb_type_last)){
	if(IS_8x8DCT(mb_type_last)){
	  for(i=0; i<16; i+=4){
	    uint8_t * src_left = (uint32_t)(intra_pred4x4_left_ptr[1][i]) | 0xF4000000;
	    uint8_t * src_top = (uint32_t)(intra_pred4x4_top_ptr[1][i]) | 0xF4000000;
	    uint8_t * src_topleft;
	    if ( i==8 )
	      src_topleft = src_left - RECON_BUF_STRIDE - 1;
	    else src_topleft = src_top - 1;

	    uint8_t * ptr= AUX_T->h264_yrecon[1] + MB_offset_recon[i];
	    int dir= dMB_aux->intra4x4_pred_mode_cache[ i ];
	    int nnz = non_zero_count_cache_luma_last[i];
	    Intra_pred_luma_8x8l(dir,ptr,(dMB_aux->topleft_samples_available<<i)&0x8000,
				 (dMB_aux->topright_samples_available<<i)&0x4000, src_left, src_top, src_topleft, RECON_BUF_STRIDE);
	    if(nnz){
	      if(nnz == 1 && dMB_aux->mb[i*16])
		ff_h264_idct8_dc_add_c(ptr, dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	      else
		ff_h264_idct8_add_c(ptr, dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	    }
	  }
	}else {
	  for(i=0; i<16; i++){
	    uint8_t * src_left = (uint32_t)(intra_pred4x4_left_ptr[1][i]) | 0xf4000000;
	    uint8_t * src_top = (uint32_t)(intra_pred4x4_top_ptr[1][i]) | 0xf4000000;

	    uint8_t * src_topleft;
	    if ( (i==2) || (i==8) || (i==10) )
	      src_topleft = src_left - RECON_BUF_STRIDE;
	    else src_topleft = src_top;

	    uint8_t * ptr= AUX_T->h264_yrecon[1] + MB_offset_recon[i];
	    uint8_t *topright;
	    int dir= dMB_aux->intra4x4_pred_mode_cache[ i ];
	    int nnz, tr;

	    if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
	      int topright_avail= (dMB_aux->topright_samples_available<<i)&0x8000;
	      if(!topright_avail){
		tr= src_top[3]*0x01010101;
		topright= (uint8_t*) &tr;
	      }else{
		topright= src_top + 4;
	      }
	    }
	    Intra_pred_luma_4x4(dir, ptr, src_left, topright, src_top, src_topleft);

	    nnz = non_zero_count_cache_luma_last[i];
	    if(nnz){
	      if(nnz == 1 && dMB_aux->mb[i*16])
		ff_h264_idct_dc_add_c(ptr, dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	      else
		ff_h264_idct_add_c(ptr, dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	    }

	  }

	}
      }else{
	Intra_pred_luma_16x16(dMB_aux->intra16x16_pred_mode, AUX_T->h264_yrecon[1], 
			      AUX_T->h264_yrecon[0] + MB_LUMA_WIDTH, TCSM1_BOTTOM_Y_25PIX+4);
	luma_dc_dequant_idct_c(dMB_aux->mb, dMB_aux->dequant4_coeff[0]);
	for(i=0; i<16; i++){
	  if(non_zero_count_cache_luma_last[i])
	    ff_h264_idct_add_c(AUX_T->h264_yrecon[1] + MB_offset_recon[i], 
			       dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	  else if(dMB_aux->mb[i*16])
	    ff_h264_idct_dc_add_c(AUX_T->h264_yrecon[1] + MB_offset_recon[i],
				  dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	}  
      }
    }

    AUX_PMON_OFF(intra);

    if(IS_INTER(mb_type)){
      hl_motion_hw(dFRM, dMB_aux, AUX_T);
    }

    AUX_PMON_ON(mc);
    MC_POLLING_END();
    AUX_PMON_OFF(mc);

    AUX_PMON_ON(dblk);
    while ((*infar)!= TCSM1_H264_DBLK_INFDA) {}
    AUX_PMON_OFF(dblk);

    if(AUX_T->mc_des_dirty){
      H264_MC_DesNode *h264_mc = (H264_MC_DesNode *)(AUX_T->h264_mc_des_ptr[0]);
      h264_mc[AUX_T->mc_des_dirty - 1].VNodeHead = H264_VHEAD_UNLINK;
      SET_MC_DHA((uint32_t)h264_mc);
      CLEAR_MC_TTEND();
      SET_MC_DCS();
    }

    if(IS_INTRA(mb_type)){
      uint32_t * bakup_src = AUX_T->BackupMBbottom_Y[bakup_x1]-4;
      uint32_t * bakup_dst = TCSM1_BOTTOM_Y_25PIX;
      bakup_dst[0] = bakup_src[0];
      bakup_dst[1] = bakup_src[1];
      bakup_dst[2] = bakup_src[2];
      bakup_dst[3] = bakup_src[3];
      bakup_dst[4] = bakup_src[4];
      bakup_dst[5] = bakup_src[5];
      bakup_dst[6] = bakup_src[6];

      bakup_src = AUX_T->BackupMBbottom_U[bakup_x1]-4;
      bakup_dst = TCSM1_BOTTOM_U_13PIX;
      bakup_dst[0] = bakup_src[0];
      bakup_dst[1] = bakup_src[1];
      bakup_dst[2] = bakup_src[2];
      bakup_dst[3] = bakup_src[3];

      bakup_src = AUX_T->BackupMBbottom_V[bakup_x1]-4;
      bakup_dst = TCSM1_BOTTOM_V_13PIX;
      bakup_dst[0] = bakup_src[0];
      bakup_dst[1] = bakup_src[1];
      bakup_dst[2] = bakup_src[2];
      bakup_dst[3] = bakup_src[3];
    }

    if(IS_INTER(mb_type_last))
      hl_motion_hw_next(dMB_aux,AUX_T,mb_type_last);
   
    *(uint32_t*)&AUX_T->sub_mb_type[0] = *(uint32_t*)&dMB_aux->sub_mb_type[0];
    *(uint32_t*)&AUX_T->sub_mb_type[2] = *(uint32_t*)&dMB_aux->sub_mb_type[2];

    AUX_PMON_ON(idct);
    AUX_PMON_OFF(idct);

    if (IS_INTRA_PCM(mb_type)) {
      unsigned int x, y;
      for(i=0; i<24; i++) {
	for (y=0; y<4; y++) {
	  for (x=0; x<4; x++) {
	    *(AUX_T->h264_yrecon[0] + MB_offset_recon[i] + y*RECON_BUF_STRIDE + x) = dMB_aux->mb[i*16+y*4+x];
	  }
	}
      }
    } 

    {
      int linesize, uvlinesize;
      uint8_t *dest_y_d2, *dest_u_d2, *dest_v_d2;

      linesize = dFRM->linesize;
      uvlinesize = dFRM->uvlinesize;

      dest_y_d2 = dFRM->current_picture.y_ptr + (mb_y_d2 * 16* linesize  ) + mb_x_d2 * 16;
      dest_u_d2 = dFRM->current_picture.u_ptr + (mb_y_d2 * 8 * uvlinesize) + mb_x_d2 * 8;
      dest_v_d2 = dFRM->current_picture.v_ptr + (mb_y_d2 * 8 * uvlinesize) + mb_x_d2 * 8;

      //move dblked MB out
      MDMA1_TRAN[0].TSA = AUX_T->h264_ydblk[1]-4;
      MDMA1_TRAN[0].TDA = get_phy_addr_aux((uint32_t)dest_y_d2-4);

      MDMA1_TRAN[1].TSA = AUX_T->h264_ydblk[1]-4+DBLK_U_OFST;
      MDMA1_TRAN[1].TDA = get_phy_addr_aux((uint32_t)dest_u_d2-4);
      //MDMA1_TRAN[1].UNIT = MDMA_UNIT(1,12,96);

      MDMA1_TRAN[2].TSA = AUX_T->h264_ydblk[1]-4+DBLK_V_OFST;
      MDMA1_TRAN[2].TDA = get_phy_addr_aux((uint32_t)dest_v_d2-4);

      POLLING_MDMA1_END(); //ensure next MB's args has been received

      SET_MDMA1_DHA((uint32_t)MDMA1_TRAN);
      MDMA1_RUN();
    }

    //---------idct fo inter---------------
    if(IS_INTER(mb_type_last)){
      void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
      void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
      int di;
      if(IS_8x8DCT(mb_type_last)){
        idct_dc_add = ff_h264_idct8_dc_add_c;
        idct_add = ff_h264_idct8_add_c;
	di = 4;
      }else{
        idct_dc_add = ff_h264_idct_dc_add_c;
        idct_add = ff_h264_idct_add_c;
	di = 1;
      }

      for(i=0; i<16; i+=di){
	int nnz = non_zero_count_cache_luma_last[i];
	if(nnz){
	  if(nnz==1 && dMB_aux->mb[i*16])
	    idct_dc_add(AUX_T->h264_yrecon[1] + MB_offset_recon[i], 
			dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	  else
	    idct_add(AUX_T->h264_yrecon[1] + MB_offset_recon[i], 
		     dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	}
      }
    }

    //------------- chroma idct------------
    if(mb_type_last){
      chroma_dc_dequant_idct_c(dMB_aux->mb + 16*16, dMB_aux->dequant4_coeff[1]);
      chroma_dc_dequant_idct_c(dMB_aux->mb + 16*16+4*16, dMB_aux->dequant4_coeff[2]);
      for(i=16; i<16+8; i++){
	if(dMB_aux->non_zero_count_cache_chroma[ i - 16 ]) {
	  ff_h264_idct_add_c(AUX_T->h264_yrecon[1] + MB_offset_recon[i], 
			     dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	} else if (dMB_aux->mb[i*16]) {
	  ff_h264_idct_dc_add_c(AUX_T->h264_yrecon[1] + MB_offset_recon[i], 
				dMB_aux->mb + i*16, RECON_BUF_STRIDE);
	}
      }
    }

    if(!(mb_x==0 && mb_y==0)){
      SET_DHA_DBLK((unsigned int)AUX_T->h264_dblk_des_ptr[1]);
      *infar = 0;
      SET_DCS_DBLK(0x1);
    }

    filter_mb_dblk(dFRM, dMB_aux, AUX_T);

    {
      uint32_t * bakup_src = AUX_T->h264_yrecon[1] + 15*RECON_BUF_STRIDE;
      uint32_t * bakup_dst = AUX_T->BackupMBbottom_Y[bakup_x0];
      bakup_dst[0] = bakup_src[0];
      bakup_dst[1] = bakup_src[1];
      bakup_dst[2] = bakup_src[2];
      bakup_dst[3] = bakup_src[3];

      bakup_src = AUX_T->h264_urecon[1] + 7*RECON_BUF_STRIDE;
      bakup_dst = AUX_T->BackupMBbottom_U[bakup_x0];
      bakup_dst[0] = bakup_src[0];
      bakup_dst[1] = bakup_src[1];

      bakup_src = AUX_T->h264_vrecon[1] + 7*RECON_BUF_STRIDE;
      bakup_dst = AUX_T->BackupMBbottom_V[bakup_x0];
      bakup_dst[0] = bakup_src[0];
      bakup_dst[1] = bakup_src[1];
    }

    mb_x_d2 = mb_x_d1;
    mb_y_d2 = mb_y_d1;
    mb_x_d1 = mb_x;
    mb_y_d1 = mb_y;

    mb_type_last=mb_type;

    for(i=0;i<16;i++)
      non_zero_count_cache_luma_last[i]=dMB_aux->non_zero_count_cache_luma[ scan5[i] ];

    // abnormal exit
    if (dMB_aux->deblocking_filter & 0x8) break;

    bakup_x0=(bakup_x0==((dFRM->mb_width)-1))?0:(bakup_x0+1);//hit second line's tail
    bakup_x1=(bakup_x1==((dFRM->mb_width)-1))?0:(bakup_x1+1);
    XCHG2(AUX_T->h264_yrecon[0],AUX_T->h264_yrecon[1],XCHGtmp);
    XCHG2(AUX_T->h264_urecon[0],AUX_T->h264_urecon[1],XCHGtmp);
    XCHG2(AUX_T->h264_vrecon[0],AUX_T->h264_vrecon[1],XCHGtmp);
    XCHG2(AUX_T->h264_ydblk[0],AUX_T->h264_ydblk[1],XCHGtmp);
    XCHG2(AUX_T->h264_dblk_des_ptr[0],AUX_T->h264_dblk_des_ptr[1],XCHGtmp);
    XCHG2(AUX_T->h264_mc_des_ptr[0],AUX_T->h264_mc_des_ptr[1],XCHGtmp);
    XCHG2(intra_pred4x4_top_ptr[0],intra_pred4x4_top_ptr[1],XCHGtmp);
    XCHG2(intra_pred4x4_left_ptr[0],intra_pred4x4_left_ptr[1],XCHGtmp);
    XCHG2(dMB_aux,dMB2,XCHGtmp);

  }

  AUX_PMON_TRAN(mc,PMON_MC_BUF);
  AUX_PMON_TRAN(idct,PMON_IDCT_BUF);
  AUX_PMON_TRAN(dblk,PMON_DBLK_BUF);
  AUX_PMON_TRAN(intra,PMON_INTRA_BUF);
  AUX_PMON_TRAN(mdma,PMON_MDMA_BUF);
  AUX_PMON_TRAN(sync,PMON_SYNC_BUF);

  /*
    task_done: aux-cpu task done flag, only write by aux-cpu, only read by main-cpu
  */
  MDMA1_ARG->TSA = (TCSM1_TASK_DONE);
  MDMA1_ARG->TDA = (TCSM0_TASK_DONE);
  MDMA1_ARG->STRD = MDMA_STRD(4,4);
  MDMA1_ARG->UNIT = MDMA_UNIT(1,4,4);
  SET_MDMA1_DHA((uint32_t)TCSM1_GP1_ARG_CHAIN);
  MDMA1_RUN();

  i_nop;  
  i_nop;    
  i_nop;      
  i_nop;  
  __asm__ __volatile__ ("wait");

}