vp9_encodeframe.c

      for (i = 0; i < 384; i++) {
        printf("%3d ", xd->predictor[i]);
        if (i % 16 == 15) printf("\n");
      }
      printf("\n");
      printf("src_diff\n");
      for (i = 0; i < 384; i++) {
        printf("%3d ", x->src_diff[i]);
        if (i % 16 == 15) printf("\n");
      }
      printf("\n");
      printf("diff\n");
      for (i = 0; i < 384; i++) {
        printf("%3d ", xd->block[0].diff[i]);
        if (i % 16 == 15) printf("\n");
      }
      printf("\n");
      printf("final y\n");
      for (i = 0; i < 16; i++) {
        for (j = 0; j < 16; j++)
          printf("%3d ", xd->dst.y_buffer[i * xd->dst.y_stride + j]);
        printf("\n");
      }
      printf("\n");
      printf("final u\n");
      for (i = 0; i < 8; i++) {
        for (j = 0; j < 8; j++)
          printf("%3d ", xd->dst.u_buffer[i * xd->dst.uv_stride + j]);
        printf("\n");
      }
      printf("\n");
      printf("final v\n");
      for (i = 0; i < 8; i++) {
        for (j = 0; j < 8; j++)
          printf("%3d ", xd->dst.v_buffer[i * xd->dst.uv_stride + j]);
        printf("\n");
      }
      fflush(stdout);
    }
#endif

#if CONFIG_CODE_NONZEROCOUNT
    gather_nzcs_mb16(cm, xd);
#endif
    vp9_tokenize_mb(cpi, xd, t, !output_enabled);

  } else {
    // FIXME(rbultje): not tile-aware (mi - 1)
    int mb_skip_context = cpi->common.mb_no_coeff_skip ?
      (mi - 1)->mbmi.mb_skip_coeff + (mi - mis)->mbmi.mb_skip_coeff : 0;

    if (cm->mb_no_coeff_skip) {
      mbmi->mb_skip_coeff = 1;
      if (output_enabled)
        cpi->skip_true_count[mb_skip_context]++;
      vp9_reset_sb_tokens_context(xd, BLOCK_SIZE_MB16X16);
    } else {
      vp9_stuff_mb(cpi, xd, t, !output_enabled);
      mbmi->mb_skip_coeff = 0;
      if (output_enabled)
        cpi->skip_false_count[mb_skip_context]++;
    }
  }

  if (output_enabled) {
    int segment_id = mbmi->segment_id;
    if (cpi->common.txfm_mode == TX_MODE_SELECT &&
        !((cpi->common.mb_no_coeff_skip && mbmi->mb_skip_coeff) ||
          (vp9_segfeature_active(&x->e_mbd, segment_id, SEG_LVL_SKIP)))) {
      assert(mbmi->txfm_size <= TX_16X16);
      if (mbmi->mode != I4X4_PRED && mbmi->mode != I8X8_PRED &&
          mbmi->mode != SPLITMV) {
        cpi->txfm_count_16x16p[mbmi->txfm_size]++;
      } else if (mbmi->mode == I8X8_PRED ||
                 (mbmi->mode == SPLITMV &&
                  mbmi->partitioning != PARTITIONING_4X4)) {
        cpi->txfm_count_8x8p[mbmi->txfm_size]++;
      }
    } else if (mbmi->mode != I4X4_PRED && mbmi->mode != I8X8_PRED &&
        mbmi->mode != SPLITMV && cpi->common.txfm_mode >= ALLOW_16X16) {
      mbmi->txfm_size = TX_16X16;
    } else if (mbmi->mode != I4X4_PRED &&
               !(mbmi->mode == SPLITMV &&
                 mbmi->partitioning == PARTITIONING_4X4) &&
               cpi->common.txfm_mode >= ALLOW_8X8) {
      mbmi->txfm_size = TX_8X8;
    } else {
      mbmi->txfm_size = TX_4X4;
    }
  }
}

static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t,
                              int output_enabled, int mb_row, int mb_col,
                              BLOCK_SIZE_TYPE bsize) {
  VP9_COMMON *const cm = &cpi->common;
  MACROBLOCK *const x = &cpi->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  const uint8_t *src = x->src.y_buffer;
  uint8_t *dst = xd->dst.y_buffer;
  const uint8_t *usrc = x->src.u_buffer;
  uint8_t *udst = xd->dst.u_buffer;
  const uint8_t *vsrc = x->src.v_buffer;
  uint8_t *vdst = xd->dst.v_buffer;
  int src_y_stride = x->src.y_stride, dst_y_stride = xd->dst.y_stride;
  int src_uv_stride = x->src.uv_stride, dst_uv_stride = xd->dst.uv_stride;
  int n;
  MODE_INFO *mi = x->e_mbd.mode_info_context;
  unsigned int segment_id = mi->mbmi.segment_id;
  const int mis = cm->mode_info_stride;
  const int bwl = mb_width_log2(bsize);
  const int bw = 1 << bwl, bh = 1 << mb_height_log2(bsize);

  if (cm->frame_type == KEY_FRAME) {
    if (cpi->oxcf.tuning == VP8_TUNE_SSIM) {
      adjust_act_zbin(cpi, x);
      vp9_update_zbin_extra(cpi, x);
    }
  } else {
    vp9_setup_interp_filters(xd, xd->mode_info_context->mbmi.interp_filter, cm);

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

    // Experimental code. Special case for gf and arf zeromv modes.
    // Increase zbin size to suppress 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 = SPLIT_MV_ZBIN_BOOST;
        } else {
          cpi->zbin_mode_boost = MV_ZBIN_BOOST;
        }
      } else {
        cpi->zbin_mode_boost = INTRA_ZBIN_BOOST;
      }
    }

    vp9_update_zbin_extra(cpi, x);
  }

  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    vp9_build_intra_predictors_sby_s(&x->e_mbd, bsize);
    vp9_build_intra_predictors_sbuv_s(&x->e_mbd, bsize);
    if (output_enabled)
      sum_intra_stats(cpi, x);
  } else {
    int ref_fb_idx;

    assert(cm->frame_type != KEY_FRAME);

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

    setup_pred_block(&xd->pre,
                     &cpi->common.yv12_fb[ref_fb_idx],
                     mb_row, mb_col,
                     &xd->scale_factor[0], &xd->scale_factor_uv[0]);

    if (xd->mode_info_context->mbmi.second_ref_frame > 0) {
      int second_ref_fb_idx;

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

      setup_pred_block(&xd->second_pre,
                       &cpi->common.yv12_fb[second_ref_fb_idx],
                       mb_row, mb_col,
                       &xd->scale_factor[1], &xd->scale_factor_uv[1]);
    }

    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
  }

  if (!x->skip) {
    vp9_subtract_sby_s_c(x->src_diff, src, src_y_stride, dst, dst_y_stride,
                         bsize);
    vp9_subtract_sbuv_s_c(x->src_diff, usrc, vsrc, src_uv_stride,
                          udst, vdst, dst_uv_stride, bsize);

    switch (xd->mode_info_context->mbmi.txfm_size) {
      case TX_32X32:
        vp9_transform_sby_32x32(x, bsize);
        vp9_quantize_sby_32x32(x, bsize);
        if (bsize == BLOCK_SIZE_SB64X64) {
          vp9_transform_sbuv_32x32(x, bsize);
          vp9_quantize_sbuv_32x32(x, bsize);
        } else {
          vp9_transform_sbuv_16x16(x, bsize);
          vp9_quantize_sbuv_16x16(x, bsize);
        }
        if (x->optimize) {
          vp9_optimize_sby_32x32(cm, x, bsize);
          if (bsize == BLOCK_SIZE_SB64X64)
            vp9_optimize_sbuv_32x32(cm, x, bsize);
          else
            vp9_optimize_sbuv_16x16(cm, x, bsize);
        }
        vp9_inverse_transform_sby_32x32(xd, bsize);
        if (bsize == BLOCK_SIZE_SB64X64)
          vp9_inverse_transform_sbuv_32x32(xd, bsize);
        else
          vp9_inverse_transform_sbuv_16x16(xd, bsize);
        break;
      case TX_16X16:
        vp9_transform_sby_16x16(x, bsize);
        vp9_quantize_sby_16x16(x, bsize);
        if (bsize >= BLOCK_SIZE_SB32X32) {
          vp9_transform_sbuv_16x16(x, bsize);
          vp9_quantize_sbuv_16x16(x, bsize);
        } else {
          vp9_transform_sbuv_8x8(x, bsize);
          vp9_quantize_sbuv_8x8(x, bsize);
        }
        if (x->optimize) {
          vp9_optimize_sby_16x16(cm, x, bsize);
          if (bsize >= BLOCK_SIZE_SB32X32)
            vp9_optimize_sbuv_16x16(cm, x, bsize);
          else
            vp9_optimize_sbuv_8x8(cm, x, bsize);
        }
        vp9_inverse_transform_sby_16x16(xd, bsize);
        if (bsize >= BLOCK_SIZE_SB32X32)
          vp9_inverse_transform_sbuv_16x16(xd, bsize);
        else
          vp9_inverse_transform_sbuv_8x8(xd, bsize);
        break;
      case TX_8X8:
        vp9_transform_sby_8x8(x, bsize);
        vp9_transform_sbuv_8x8(x, bsize);
        vp9_quantize_sby_8x8(x, bsize);
        vp9_quantize_sbuv_8x8(x, bsize);
        if (x->optimize) {
          vp9_optimize_sby_8x8(cm, x, bsize);
          vp9_optimize_sbuv_8x8(cm, x, bsize);
        }
        vp9_inverse_transform_sby_8x8(xd, bsize);
        vp9_inverse_transform_sbuv_8x8(xd, bsize);
        break;
      case TX_4X4:
        vp9_transform_sby_4x4(x, bsize);
        vp9_transform_sbuv_4x4(x, bsize);
        vp9_quantize_sby_4x4(x, bsize);
        vp9_quantize_sbuv_4x4(x, bsize);
        if (x->optimize) {
          vp9_optimize_sby_4x4(cm, x, bsize);
          vp9_optimize_sbuv_4x4(cm, x, bsize);
        }
        vp9_inverse_transform_sby_4x4(xd, bsize);
        vp9_inverse_transform_sbuv_4x4(xd, bsize);
        break;
      default: assert(0);
    }
    vp9_recon_sby_s_c(xd, dst, bsize);
    vp9_recon_sbuv_s_c(&x->e_mbd, udst, vdst, bsize);
#if CONFIG_CODE_NONZEROCOUNT
    if (bsize == BLOCK_SIZE_SB32X32) {
      gather_nzcs_sb32(cm, &x->e_mbd);
    } else {
      gather_nzcs_sb64(cm, &x->e_mbd);
    }
#endif
    vp9_tokenize_sb(cpi, &x->e_mbd, t, !output_enabled, bsize);
  } else {
    // FIXME(rbultje): not tile-aware (mi - 1)
    int mb_skip_context = cpi->common.mb_no_coeff_skip ?
        (mi - 1)->mbmi.mb_skip_coeff + (mi - mis)->mbmi.mb_skip_coeff : 0;

    xd->mode_info_context->mbmi.mb_skip_coeff = 1;
    if (cm->mb_no_coeff_skip) {
      if (output_enabled)
        cpi->skip_true_count[mb_skip_context]++;
      vp9_reset_sb_tokens_context(xd, bsize);
    } else {
      vp9_stuff_sb(cpi, xd, t, !output_enabled, bsize);
      if (output_enabled)
        cpi->skip_false_count[mb_skip_context]++;
    }
  }

  // copy skip flag on all mb_mode_info contexts in this SB
  // if this was a skip at this txfm size
  for (n = 1; n < bw * bh; n++) {
    const int x_idx = n & (bw - 1), y_idx = n >> bwl;
    if (mb_col + x_idx < cm->mb_cols && mb_row + y_idx < cm->mb_rows)
      mi[x_idx + y_idx * mis].mbmi.mb_skip_coeff = mi->mbmi.mb_skip_coeff;
  }

  if (output_enabled) {
    if (cm->txfm_mode == TX_MODE_SELECT &&
        !((cm->mb_no_coeff_skip && mi->mbmi.mb_skip_coeff) ||
          (vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP)))) {
      if (bsize >= BLOCK_SIZE_SB32X32) {
        cpi->txfm_count_32x32p[mi->mbmi.txfm_size]++;
      } else {
        cpi->txfm_count_16x16p[mi->mbmi.txfm_size]++;
      }
    } else {
      int x, y;
      TX_SIZE sz = (cm->txfm_mode == TX_MODE_SELECT) ? TX_32X32 : cm->txfm_mode;

      if (sz == TX_32X32 && bsize < BLOCK_SIZE_SB32X32)
        sz = TX_16X16;

      for (y = 0; y < bh; y++) {
        for (x = 0; x < bw; x++) {
          if (mb_col + x < cm->mb_cols && mb_row + y < cm->mb_rows) {
            mi[mis * y + x].mbmi.txfm_size = sz;
          }
        }
      }
    }
  }
}