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 283 matching lines @@
 }
 
 static INLINE void fdct32x32(int rd_transform,
                              const int16_t *src, int16_t *dst, int src_stride) {
   if (rd_transform)
     vp9_fdct32x32_rd(src, dst, src_stride);
   else
     vp9_fdct32x32(src, dst, src_stride);
 }
 
+void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block,
+                        BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
+  MACROBLOCKD *const xd = &x->e_mbd;
+  const struct macroblock_plane *const p = &x->plane[plane];
+  const struct macroblockd_plane *const pd = &xd->plane[plane];
+  int16_t *const coeff = BLOCK_OFFSET(p->coeff, block);
+  int16_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
+  int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
+  uint16_t *const eob = &p->eobs[block];
+  const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
+  int i, j;
+  const int16_t *src_diff;
+
+  txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
+  src_diff = &p->src_diff[4 * (j * diff_stride + i)];
+
+  switch (tx_size) {
+    case TX_32X32:
+      vp9_fdct32x32_1(src_diff, coeff, diff_stride);
+      vp9_quantize_dc_32x32(coeff, x->skip_block, p->round,
+                            p->quant_fp[0], qcoeff, dqcoeff,
+                            pd->dequant[0], eob);
+      break;
+    case TX_16X16:
+      vp9_fdct16x16_1(src_diff, coeff, diff_stride);
+      vp9_quantize_dc(coeff, x->skip_block, p->round,
+                     p->quant_fp[0], qcoeff, dqcoeff,
+                     pd->dequant[0], eob);
+      break;
+    case TX_8X8:
+      vp9_fdct8x8_1(src_diff, coeff, diff_stride);
+      vp9_quantize_dc(coeff, x->skip_block, p->round,
+                      p->quant_fp[0], qcoeff, dqcoeff,
+                      pd->dequant[0], eob);
+      break;
+    case TX_4X4:
+      x->fwd_txm4x4(src_diff, coeff, diff_stride);
+      vp9_quantize_dc(coeff, x->skip_block, p->round,
+                      p->quant_fp[0], qcoeff, dqcoeff,
+                      pd->dequant[0], eob);
+      break;
+    default:
+      assert(0);
+  }
+}
+
 void vp9_xform_quant(MACROBLOCK *x, int plane, int block,
                      BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
   MACROBLOCKD *const xd = &x->e_mbd;
   const struct macroblock_plane *const p = &x->plane[plane];
   const struct macroblockd_plane *const pd = &xd->plane[plane];
   const scan_order *const scan_order = &vp9_default_scan_orders[tx_size];
   int16_t *const coeff = BLOCK_OFFSET(p->coeff, block);
   int16_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
   int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   uint16_t *const eob = &p->eobs[block];
@@ skipping 55 matching lines @@
   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;
     *a = *l = 0;
     return;
   }
 
-  if (!x->skip_recode)
-    vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
+  if (x->skip_txfm == 0) {
+    // full forward transform and quantization
+    if (!x->skip_recode)
+      vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
+  } else if (x->skip_txfm == 2) {
+    // fast path forward transform and quantization
+    vp9_xform_quant_fp(x, plane, block, plane_bsize, tx_size);
+  } else {
+    // skip forward transform
+    p->eobs[block] = 0;
+    *a = *l = 0;
+    return;
+  }
 
   if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
     const int ctx = combine_entropy_contexts(*a, *l);
     *a = *l = optimize_b(x, plane, block, tx_size, ctx) > 0;
   } else {
     *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]);
+      x->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]);
+    x->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;
@@ skipping 125 matching lines @@
                        p->quant_shift, qcoeff, dqcoeff,
                        pd->dequant, p->zbin_extra, eob, scan_order->scan,
                        scan_order->iscan);
       }
 
       if (!x->skip_encode && *eob) {
         if (tx_type == DCT_DCT)
           // 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, dst_stride, *eob);
+          x->itxm_add(dqcoeff, dst, dst_stride, *eob);
         else
           vp9_iht4x4_16_add(dqcoeff, dst, dst_stride, tx_type);
       }
       break;
     default:
       assert(0);
   }
   if (*eob)
     *(args->skip) = 0;
 }
 
 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[0]->mbmi.skip};
 
   vp9_foreach_transformed_block_in_plane(xd, bsize, plane, encode_block_intra,
                                          &arg);
 }