libvpx: Pull from upstream

source/libvpx/vp9/encoder/vp9_encodemb.c

 /*
  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 89 matching lines @@
                                      uint8_t *token_cache) {
   int bak = token_cache[scan[idx]], pt;
   token_cache[scan[idx]] = vp9_pt_energy_class[token];
   pt = get_coef_context(nb, token_cache, idx + 1);
   token_cache[scan[idx]] = bak;
   return pt;
 }
 
 static void optimize_b(int plane, int block, BLOCK_SIZE plane_bsize,
                        TX_SIZE tx_size, MACROBLOCK *mb,
-                       struct optimize_ctx *ctx) {
+                       ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l) {
   MACROBLOCKD *const xd = &mb->e_mbd;
   struct macroblock_plane *p = &mb->plane[plane];
   struct macroblockd_plane *pd = &xd->plane[plane];
-  const int ref = is_inter_block(&xd->mi_8x8[0]->mbmi);
+  const int ref = is_inter_block(&xd->mi[0]->mbmi);
   vp9_token_state tokens[1025][2];
   unsigned best_index[1025][2];
   const int16_t *coeff = BLOCK_OFFSET(mb->plane[plane].coeff, block);
   int16_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
   int16_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   int eob = p->eobs[block], final_eob, sz = 0;
   const int i0 = 0;
   int rc, x, next, i;
   int64_t rdmult, rddiv, rd_cost0, rd_cost1;
   int rate0, rate1, error0, error1, t0, t1;
   int best, band, pt;
   PLANE_TYPE type = pd->plane_type;
   int err_mult = plane_rd_mult[type];
   const int default_eob = 16 << (tx_size << 1);
   const int mul = 1 + (tx_size == TX_32X32);
   uint8_t token_cache[1024];
   const int16_t *dequant_ptr = pd->dequant;
   const uint8_t *const band_translate = get_band_translate(tx_size);
   const scan_order *so = get_scan(xd, tx_size, type, block);
   const int16_t *scan = so->scan;
   const int16_t *nb = so->neighbors;
-  ENTROPY_CONTEXT *a, *l;
-  int tx_x, tx_y;
-  txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &tx_x, &tx_y);
-  a = &ctx->ta[plane][tx_x];
-  l = &ctx->tl[plane][tx_y];
 
   assert((!type && !plane) || (type && plane));
   assert(eob <= default_eob);
 
   /* Now set up a Viterbi trellis to evaluate alternative roundings. */
   rdmult = mb->rdmult * err_mult;
-  if (!is_inter_block(&mb->e_mbd.mi_8x8[0]->mbmi))
+  if (!is_inter_block(&mb->e_mbd.mi[0]->mbmi))
     rdmult = (rdmult * 9) >> 4;
   rddiv = mb->rddiv;
   /* Initialize the sentinel node of the trellis. */
   tokens[eob][0].rate = 0;
   tokens[eob][0].error = 0;
   tokens[eob][0].next = default_eob;
   tokens[eob][0].token = EOB_TOKEN;
   tokens[eob][0].qc = 0;
   *(tokens[eob] + 1) = *(tokens[eob] + 0);
   next = eob;
@@ skipping 215 matching lines @@
                          TX_SIZE tx_size, void *arg) {
   struct encode_b_args *const args = arg;
   MACROBLOCK *const x = args->x;
   MACROBLOCKD *const xd = &x->e_mbd;
   struct optimize_ctx *const ctx = args->ctx;
   struct macroblock_plane *const p = &x->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
   int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   int i, j;
   uint8_t *dst;
+  ENTROPY_CONTEXT *a, *l;
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
   dst = &pd->dst.buf[4 * j * pd->dst.stride + 4 * i];
+  a = &ctx->ta[plane][i];
+  l = &ctx->tl[plane][j];
 
   // TODO(jingning): per transformed block zero forcing only enabled for
   // luma component. will integrate chroma components as well.
   if (x->zcoeff_blk[tx_size][block] && plane == 0) {
     p->eobs[block] = 0;
-    ctx->ta[plane][i] = 0;
-    ctx->tl[plane][j] = 0;
+    *a = *l = 0;
     return;
   }
 
   if (!x->skip_recode)
     vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
 
   if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
-    optimize_b(plane, block, plane_bsize, tx_size, x, ctx);
+    optimize_b(plane, block, plane_bsize, tx_size, x, a, l);
   } else {
-    ctx->ta[plane][i] = p->eobs[block] > 0;
-    ctx->tl[plane][j] = p->eobs[block] > 0;
+    *a = *l = p->eobs[block] > 0;
   }
 
   if (p->eobs[block])
     *(args->skip) = 0;
 
   if (x->skip_encode || p->eobs[block] == 0)
     return;
 
   switch (tx_size) {
     case TX_32X32:
       vp9_idct32x32_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
       break;
     case TX_16X16:
       vp9_idct16x16_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
       break;
     case TX_8X8:
       vp9_idct8x8_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
       break;
     case TX_4X4:
       // this is like vp9_short_idct4x4 but has a special case around eob<=1
       // which is significant (not just an optimization) for the lossless
       // case.
       xd->itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
       break;
     default:
       assert(0 && "Invalid transform size");
   }
 }
+
 static void encode_block_pass1(int plane, int block, BLOCK_SIZE plane_bsize,
                                TX_SIZE tx_size, void *arg) {
   MACROBLOCK *const x = (MACROBLOCK *)arg;
   MACROBLOCKD *const xd = &x->e_mbd;
   struct macroblock_plane *const p = &x->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
   int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   int i, j;
   uint8_t *dst;
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
   dst = &pd->dst.buf[4 * j * pd->dst.stride + 4 * i];
 
   vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
 
   if (p->eobs[block] > 0)
     xd->itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
 }
 
 void vp9_encode_sby_pass1(MACROBLOCK *x, BLOCK_SIZE bsize) {
   vp9_subtract_plane(x, bsize, 0);
   vp9_foreach_transformed_block_in_plane(&x->e_mbd, bsize, 0,
                                          encode_block_pass1, x);
 }
 
 void vp9_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize) {
   MACROBLOCKD *const xd = &x->e_mbd;
   struct optimize_ctx ctx;
-  MB_MODE_INFO *mbmi = &xd->mi_8x8[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
   struct encode_b_args arg = {x, &ctx, &mbmi->skip};
   int plane;
 
   for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
     if (!x->skip_recode)
       vp9_subtract_plane(x, bsize, plane);
 
     if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
       const struct macroblockd_plane* const pd = &xd->plane[plane];
       const TX_SIZE tx_size = plane ? get_uv_tx_size(mbmi) : mbmi->tx_size;
       vp9_get_entropy_contexts(bsize, tx_size, pd,
                                ctx.ta[plane], ctx.tl[plane]);
     }
 
     vp9_foreach_transformed_block_in_plane(xd, bsize, plane, encode_block,
                                            &arg);
   }
 }
 
 static void encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
                                TX_SIZE tx_size, void *arg) {
   struct encode_b_args* const args = arg;
   MACROBLOCK *const x = args->x;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi_8x8[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
   struct macroblock_plane *const p = &x->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
   int16_t *coeff = BLOCK_OFFSET(p->coeff, block);
   int16_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
   int16_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   const scan_order *scan_order;
   TX_TYPE tx_type;
   MB_PREDICTION_MODE mode;
   const int bwl = b_width_log2(plane_bsize);
   const int diff_stride = 4 * (1 << bwl);
   uint8_t *src, *dst;
   int16_t *src_diff;
   uint16_t *eob = &p->eobs[block];
   const int src_stride = p->src.stride;
   const int dst_stride = pd->dst.stride;
   int i, j;
   txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
   dst = &pd->dst.buf[4 * (j * dst_stride + i)];
   src = &p->src.buf[4 * (j * src_stride + i)];
   src_diff = &p->src_diff[4 * (j * diff_stride + i)];
 
-  // if (x->optimize)
-  //   optimize_b(plane, block, plane_bsize, tx_size, x, args->ctx);
-
   switch (tx_size) {
     case TX_32X32:
       scan_order = &vp9_default_scan_orders[TX_32X32];
       mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
       vp9_predict_intra_block(xd, block >> 6, bwl, TX_32X32, mode,
                               x->skip_encode ? src : dst,
                               x->skip_encode ? src_stride : dst_stride,
                               dst, dst_stride, i, j, plane);
       if (!x->skip_recode) {
         vp9_subtract_block(32, 32, src_diff, diff_stride,
                            src, src_stride, dst, dst_stride);
         fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
         vp9_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin, p->round,
                              p->quant, p->quant_shift, qcoeff, dqcoeff,
                              pd->dequant, p->zbin_extra, eob, scan_order->scan,
                              scan_order->iscan);
       }
       if (!x->skip_encode && *eob)
         vp9_idct32x32_add(dqcoeff, dst, dst_stride, *eob);
       break;
     case TX_16X16:
-      tx_type = get_tx_type_16x16(pd->plane_type, xd);
+      tx_type = get_tx_type(pd->plane_type, xd);
       scan_order = &vp9_scan_orders[TX_16X16][tx_type];
       mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
       vp9_predict_intra_block(xd, block >> 4, bwl, TX_16X16, mode,
                               x->skip_encode ? src : dst,
                               x->skip_encode ? src_stride : dst_stride,
                               dst, dst_stride, i, j, plane);
       if (!x->skip_recode) {
         vp9_subtract_block(16, 16, src_diff, diff_stride,
                            src, src_stride, dst, dst_stride);
         vp9_fht16x16(src_diff, coeff, diff_stride, tx_type);
         vp9_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
                        p->quant, p->quant_shift, qcoeff, dqcoeff,
                        pd->dequant, p->zbin_extra, eob, scan_order->scan,
                        scan_order->iscan);
       }
       if (!x->skip_encode && *eob)
         vp9_iht16x16_add(tx_type, dqcoeff, dst, dst_stride, *eob);
       break;
     case TX_8X8:
-      tx_type = get_tx_type_8x8(pd->plane_type, xd);
+      tx_type = get_tx_type(pd->plane_type, xd);
       scan_order = &vp9_scan_orders[TX_8X8][tx_type];
       mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
       vp9_predict_intra_block(xd, block >> 2, bwl, TX_8X8, mode,
                               x->skip_encode ? src : dst,
                               x->skip_encode ? src_stride : dst_stride,
                               dst, dst_stride, i, j, plane);
       if (!x->skip_recode) {
         vp9_subtract_block(8, 8, src_diff, diff_stride,
                            src, src_stride, dst, dst_stride);
         vp9_fht8x8(src_diff, coeff, diff_stride, tx_type);
         vp9_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round, p->quant,
                        p->quant_shift, qcoeff, dqcoeff,
                        pd->dequant, p->zbin_extra, eob, scan_order->scan,
                        scan_order->iscan);
       }
       if (!x->skip_encode && *eob)
         vp9_iht8x8_add(tx_type, dqcoeff, dst, dst_stride, *eob);
       break;
     case TX_4X4:
       tx_type = get_tx_type_4x4(pd->plane_type, xd, block);
       scan_order = &vp9_scan_orders[TX_4X4][tx_type];
-      mode = plane == 0 ? get_y_mode(xd->mi_8x8[0], block) : mbmi->uv_mode;
+      mode = plane == 0 ? get_y_mode(xd->mi[0], block) : mbmi->uv_mode;
       vp9_predict_intra_block(xd, block, bwl, TX_4X4, mode,
                               x->skip_encode ? src : dst,
                               x->skip_encode ? src_stride : dst_stride,
                               dst, dst_stride, i, j, plane);
 
       if (!x->skip_recode) {
         vp9_subtract_block(4, 4, src_diff, diff_stride,
                            src, src_stride, dst, dst_stride);
         if (tx_type != DCT_DCT)
           vp9_fht4x4(src_diff, coeff, diff_stride, tx_type);
@@ skipping 25 matching lines @@
 void vp9_encode_block_intra(MACROBLOCK *x, int plane, int block,
                             BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
                             unsigned char *skip) {
   struct encode_b_args arg = {x, NULL, skip};
   encode_block_intra(plane, block, plane_bsize, tx_size, &arg);
 }
 
 
 void vp9_encode_intra_block_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
   const MACROBLOCKD *const xd = &x->e_mbd;
-  struct encode_b_args arg = {x, NULL, &xd->mi_8x8[0]->mbmi.skip};
+  struct encode_b_args arg = {x, NULL, &xd->mi[0]->mbmi.skip};
 
   vp9_foreach_transformed_block_in_plane(xd, bsize, plane, encode_block_intra,
                                          &arg);
 }
 
 int vp9_encode_intra(MACROBLOCK *x, int use_16x16_pred) {
-  MB_MODE_INFO * mbmi = &x->e_mbd.mi_8x8[0]->mbmi;
+  MB_MODE_INFO * mbmi = &x->e_mbd.mi[0]->mbmi;
   x->skip_encode = 0;
   mbmi->mode = DC_PRED;
   mbmi->ref_frame[0] = INTRA_FRAME;
   mbmi->tx_size = use_16x16_pred ? (mbmi->sb_type >= BLOCK_16X16 ? TX_16X16
                                                                  : TX_8X8)
                                    : TX_4X4;
   vp9_encode_intra_block_plane(x, mbmi->sb_type, 0);
   return vp9_get_mb_ss(x->plane[0].src_diff);
 }