Make the use of pred buffers consistent in MB/SB

Use in-place buffers (dst of MACROBLOCKD) for  macroblock prediction.
This makes the macroblock buffer handling consistent with those of
superblock. Remove predictor buffer MACROBLOCKD.

Change-Id: Id1bcd898961097b1e6230c10f0130753a59fc6df

16 files changed, 212 insertions, 335 deletions

diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 8d67402c8..9525a0e14 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -290,7 +290,6 @@ typedef struct {
 } MODE_INFO;
 
 typedef struct blockd {
-  uint8_t *predictor;
   int16_t *diff;
   int16_t *dequant;
 
@@ -354,7 +353,6 @@ struct mb_plane {
 
 typedef struct macroblockd {
   DECLARE_ALIGNED(16, int16_t,  diff[64*64+32*32*2]);      /* from idct diff */
-  DECLARE_ALIGNED(16, uint8_t,  predictor[384]);  // unused for superblocks
 #if CONFIG_CODE_NONZEROCOUNT
   DECLARE_ALIGNED(16, uint16_t, nzcs[256+64*2]);
 #endif
diff --git a/vp9/common/vp9_mbpitch.c b/vp9/common/vp9_mbpitch.c
index aba950e9a..6ed5f27d9 100644
--- a/vp9/common/vp9_mbpitch.c
+++ b/vp9/common/vp9_mbpitch.c
@@ -78,7 +78,6 @@ void vp9_setup_block_dptrs(MACROBLOCKD *mb) {
       const int to = r * 4 + c;
       const int from = r * 4 * 16 + c * 4;
       blockd[to].diff = &mb->diff[from];
-      blockd[to].predictor = &mb->predictor[from];
     }
   }
 
@@ -87,7 +86,6 @@ void vp9_setup_block_dptrs(MACROBLOCKD *mb) {
       const int to = 16 + r * 2 + c;
       const int from = 256 + r * 4 * 8 + c * 4;
       blockd[to].diff = &mb->diff[from];
-      blockd[to].predictor = &mb->predictor[from];
     }
   }
 
@@ -96,7 +94,6 @@ void vp9_setup_block_dptrs(MACROBLOCKD *mb) {
       const int to = 20 + r * 2 + c;
       const int from = 320 + r * 4 * 8 + c * 4;
       blockd[to].diff = &mb->diff[from];
-      blockd[to].predictor = &mb->predictor[from];
     }
   }
 
diff --git a/vp9/common/vp9_recon.c b/vp9/common/vp9_recon.c
index 0625ccb8e..121776c69 100644
--- a/vp9/common/vp9_recon.c
+++ b/vp9/common/vp9_recon.c
@@ -32,22 +32,22 @@ static INLINE void recon(int rows, int cols,
 
 void vp9_recon_b_c(uint8_t *pred_ptr, int16_t *diff_ptr, uint8_t *dst_ptr,
                    int stride) {
-  recon(4, 4, pred_ptr, 16, diff_ptr, 16, dst_ptr, stride);
+  recon(4, 4, pred_ptr, stride, diff_ptr, 16, dst_ptr, stride);
 }
 
 void vp9_recon_uv_b_c(uint8_t *pred_ptr, int16_t *diff_ptr, uint8_t *dst_ptr,
                       int stride) {
-  recon(4, 4, pred_ptr, 8, diff_ptr, 8, dst_ptr, stride);
+  recon(4, 4, pred_ptr, stride, diff_ptr, 8, dst_ptr, stride);
 }
 
 void vp9_recon4b_c(uint8_t *pred_ptr, int16_t *diff_ptr, uint8_t *dst_ptr,
                    int stride) {
-  recon(4, 16, pred_ptr, 16, diff_ptr, 16, dst_ptr, stride);
+  recon(4, 16, pred_ptr, stride, diff_ptr, 16, dst_ptr, stride);
 }
 
 void vp9_recon2b_c(uint8_t *pred_ptr, int16_t *diff_ptr, uint8_t *dst_ptr,
                    int stride) {
-  recon(4, 8, pred_ptr, 8, diff_ptr, 8, dst_ptr, stride);
+  recon(4, 8, pred_ptr, stride, diff_ptr, 8, dst_ptr, stride);
 }
 
 void vp9_recon_sby_s_c(MACROBLOCKD *mb, uint8_t *dst,
@@ -95,7 +95,8 @@ void vp9_recon_mby_c(MACROBLOCKD *xd) {
   for (i = 0; i < 16; i += 4) {
     BLOCKD *b = &xd->block[i];
 
-    vp9_recon4b(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    vp9_recon4b(*(b->base_dst) + b->dst, b->diff,
+                *(b->base_dst) + b->dst, b->dst_stride);
   }
 }
 
@@ -104,13 +105,13 @@ void vp9_recon_mb_c(MACROBLOCKD *xd) {
 
   for (i = 0; i < 16; i += 4) {
     BLOCKD *b = &xd->block[i];
-
-    vp9_recon4b(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    vp9_recon4b(*(b->base_dst) + b->dst, b->diff,
+                *(b->base_dst) + b->dst, b->dst_stride);
   }
 
   for (i = 16; i < 24; i += 2) {
     BLOCKD *b = &xd->block[i];
-
-    vp9_recon2b(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    vp9_recon2b(*(b->base_dst) + b->dst, b->diff,
+                *(b->base_dst) + b->dst, b->dst_stride);
   }
 }
diff --git a/vp9/common/vp9_reconinter.c b/vp9/common/vp9_reconinter.c
index 79ca0b4e5..71be77df1 100644
--- a/vp9/common/vp9_reconinter.c
+++ b/vp9/common/vp9_reconinter.c
@@ -399,7 +399,7 @@ static void build_2x1_inter_predictor_wh(const BLOCKD *d0, const BLOCKD *d1,
                                          int row, int col) {
   struct scale_factors * scale = &s[which_mv];
 
-  assert(d1->predictor - d0->predictor == block_size);
+  assert(d1->dst - d0->dst == block_size);
   assert(d1->pre == d0->pre + block_size);
 
   scale->set_scaled_offsets(scale, row, col);
@@ -446,11 +446,11 @@ static void build_2x1_inter_predictor(const BLOCKD *d0, const BLOCKD *d1,
                                       int block_size, int stride,
                                       int which_mv, int weight,
                                       const struct subpix_fn_table *subpix,
-                                      int row, int col, int use_dst) {
-  uint8_t *d0_predictor = use_dst ? *(d0->base_dst) + d0->dst : d0->predictor;
-  uint8_t *d1_predictor = use_dst ? *(d1->base_dst) + d1->dst : d1->predictor;
+                                      int row, int col) {
+  uint8_t *d0_predictor = *(d0->base_dst) + d0->dst;
+  uint8_t *d1_predictor = *(d1->base_dst) + d1->dst;
   struct scale_factors * scale = &s[which_mv];
-  stride = use_dst ? d0->dst_stride : stride;
+  stride = d0->dst_stride;
 
   assert(d1_predictor - d0_predictor == block_size);
   assert(d1->pre == d0->pre + block_size);
@@ -1338,8 +1338,7 @@ void vp9_build_inter_predictors_sb(MACROBLOCKD *mb,
 }
 
 static void build_inter4x4_predictors_mb(MACROBLOCKD *xd,
-                                         int mb_row, int mb_col,
-                                         int use_dst) {
+                                         int mb_row, int mb_col) {
   int i;
   MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi;
   BLOCKD *blockd = xd->block;
@@ -1368,8 +1367,7 @@ static void build_inter4x4_predictors_mb(MACROBLOCKD *xd,
 
         build_2x1_inter_predictor(d0, d1, xd->scale_factor, 8, 16, which_mv,
                                   which_mv ? weight : 0,
-                                  &xd->subpix, mb_row * 16 + y, mb_col * 16,
-                                  use_dst);
+                                  &xd->subpix, mb_row * 16 + y, mb_col * 16);
       }
     }
   } else {
@@ -1386,8 +1384,7 @@ static void build_inter4x4_predictors_mb(MACROBLOCKD *xd,
         build_2x1_inter_predictor(d0, d1, xd->scale_factor, 4, 16, which_mv,
                                   which_mv ? weight : 0,
                                   &xd->subpix,
-                                  mb_row * 16 + y, mb_col * 16 + x,
-                                  use_dst);
+                                  mb_row * 16 + y, mb_col * 16 + x);
       }
     }
   }
@@ -1405,8 +1402,7 @@ static void build_inter4x4_predictors_mb(MACROBLOCKD *xd,
     for (which_mv = 0; which_mv < 1 + use_second_ref; ++which_mv) {
       build_2x1_inter_predictor(d0, d1, xd->scale_factor_uv, 4, 8, which_mv,
                                 which_mv ? weight : 0, &xd->subpix,
-                                mb_row * 8 + y, mb_col * 8 + x,
-                                use_dst);
+                                mb_row * 8 + y, mb_col * 8 + x);
     }
   }
 }
@@ -1493,58 +1489,17 @@ static void build_4x4uvmvs(MACROBLOCKD *xd) {
   }
 }
 
-void vp9_build_inter16x16_predictors_mb(MACROBLOCKD *xd,
-                                        uint8_t *dst_y,
-                                        uint8_t *dst_u,
-                                        uint8_t *dst_v,
-                                        int dst_ystride,
-                                        int dst_uvstride,
-                                        int mb_row,
-                                        int mb_col) {
-  vp9_build_inter16x16_predictors_mby(xd, dst_y, dst_ystride, mb_row, mb_col);
-  vp9_build_inter16x16_predictors_mbuv(xd, dst_u, dst_v, dst_uvstride,
-                                       mb_row, mb_col);
-#if CONFIG_COMP_INTERINTRA_PRED
-  if (xd->mode_info_context->mbmi.second_ref_frame == INTRA_FRAME) {
-    vp9_build_interintra_16x16_predictors_mb(xd, dst_y, dst_u, dst_v,
-                                             dst_ystride, dst_uvstride);
-  }
-#endif
-}
-
 void vp9_build_inter_predictors_mb(MACROBLOCKD *xd,
                                    int mb_row,
                                    int mb_col) {
   if (xd->mode_info_context->mbmi.mode != SPLITMV) {
-    // TODO(jingning): to be replaced with vp9_build_inter_predictors_sb() when
-    // converting buffers from predictors to dst.
-    vp9_build_inter16x16_predictors_mb(xd, xd->predictor,
-                                       &xd->predictor[256],
-                                       &xd->predictor[320], 16, 8,
-                                       mb_row, mb_col);
-
+    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
   } else {
     build_4x4uvmvs(xd);
-    build_inter4x4_predictors_mb(xd, mb_row, mb_col, 0);
+    build_inter4x4_predictors_mb(xd, mb_row, mb_col);
   }
 }
 
-void vp9_build_inter_predictors_mb_s(MACROBLOCKD *xd,
-                                   int mb_row,
-                                   int mb_col) {
-  if (xd->mode_info_context->mbmi.mode != SPLITMV) {
-    vp9_build_inter16x16_predictors_mb(xd, xd->dst.y_buffer,
-                                       xd->dst.u_buffer,
-                                       xd->dst.v_buffer,
-                                       xd->dst.y_stride,
-                                       xd->dst.uv_stride,
-                                       mb_row, mb_col);
-
-  } else {
-    build_4x4uvmvs(xd);
-    build_inter4x4_predictors_mb(xd, mb_row, mb_col, 1);
-  }
-}
 /*encoder only*/
 void vp9_build_inter4x4_predictors_mbuv(MACROBLOCKD *xd,
                                         int mb_row, int mb_col) {
@@ -1593,8 +1548,7 @@ void vp9_build_inter4x4_predictors_mbuv(MACROBLOCKD *xd,
     for (which_mv = 0; which_mv < 1 + use_second_ref; ++which_mv) {
       build_2x1_inter_predictor(d0, d1, xd->scale_factor_uv, 4, 8, which_mv,
                                 which_mv ? weight : 0,
-                                &xd->subpix, mb_row * 8 + y, mb_col * 8 + x,
-                                0);
+                                &xd->subpix, mb_row * 8 + y, mb_col * 8 + x);
     }
   }
 }
diff --git a/vp9/common/vp9_reconinter.h b/vp9/common/vp9_reconinter.h
index 068853d1d..533d30466 100644
--- a/vp9/common/vp9_reconinter.h
+++ b/vp9/common/vp9_reconinter.h
@@ -29,14 +29,20 @@ void vp9_build_inter16x16_predictors_mbuv(MACROBLOCKD *xd,
                                           int mb_row,
                                           int mb_col);
 
-void vp9_build_inter16x16_predictors_mb(MACROBLOCKD *xd,
-                                        uint8_t *dst_y,
-                                        uint8_t *dst_u,
-                                        uint8_t *dst_v,
-                                        int dst_ystride,
-                                        int dst_uvstride,
-                                        int mb_row,
-                                        int mb_col);
+void vp9_build_inter_predictors_sby(MACROBLOCKD *x,
+                                    uint8_t *dst_y,
+                                    int dst_ystride,
+                                    int mb_row,
+                                    int mb_col,
+                                    BLOCK_SIZE_TYPE bsize);
+
+void vp9_build_inter_predictors_sbuv(MACROBLOCKD *x,
+                                     uint8_t *dst_u,
+                                     uint8_t *dst_v,
+                                     int dst_uvstride,
+                                     int mb_row,
+                                     int mb_col,
+                                     BLOCK_SIZE_TYPE bsize);
 
 void vp9_build_inter_predictors_sb(MACROBLOCKD *mb,
                                    int mb_row, int mb_col,
@@ -46,10 +52,6 @@ void vp9_build_inter_predictors_mb(MACROBLOCKD *xd,
                                    int mb_row,
                                    int mb_col);
 
-void vp9_build_inter_predictors_mb_s(MACROBLOCKD *xd,
-                                     int mb_row,
-                                     int mb_col);
-
 void vp9_build_inter4x4_predictors_mbuv(MACROBLOCKD *xd,
                                         int mb_row,
                                         int mb_col);
diff --git a/vp9/common/vp9_reconintra.c b/vp9/common/vp9_reconintra.c
index 632191183..88c3f191e 100644
--- a/vp9/common/vp9_reconintra.c
+++ b/vp9/common/vp9_reconintra.c
@@ -273,7 +273,8 @@ void vp9_recon_intra_mbuv(MACROBLOCKD *xd) {
   int i;
   for (i = 16; i < 24; i += 2) {
     BLOCKD *b = &xd->block[i];
-    vp9_recon2b(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    vp9_recon2b(*(b->base_dst) + b->dst, b->diff,
+                *(b->base_dst) + b->dst, b->dst_stride);
   }
 }
 
@@ -758,40 +759,6 @@ void vp9_build_intra_predictors_sbuv_s(MACROBLOCKD *xd,
                              xd->left_available, xd->right_available);
 }
 
-// TODO(jingning): merge mby and mbuv into the above sby and sbmu functions
-void vp9_build_intra_predictors_mby(MACROBLOCKD *xd) {
-  vp9_build_intra_predictors(xd->dst.y_buffer, xd->dst.y_stride,
-                             xd->predictor, 16,
-                             xd->mode_info_context->mbmi.mode,
-                             16, 16,
-                             xd->up_available, xd->left_available,
-                             xd->right_available);
-}
-
-void vp9_build_intra_predictors_mbuv_internal(MACROBLOCKD *xd,
-                                              uint8_t *upred_ptr,
-                                              uint8_t *vpred_ptr,
-                                              int uv_stride,
-                                              int mode, int bsize) {
-  vp9_build_intra_predictors(xd->dst.u_buffer, xd->dst.uv_stride,
-                             upred_ptr, uv_stride, mode,
-                             bsize, bsize,
-                             xd->up_available, xd->left_available,
-                             xd->right_available);
-  vp9_build_intra_predictors(xd->dst.v_buffer, xd->dst.uv_stride,
-                             vpred_ptr, uv_stride, mode,
-                             bsize, bsize,
-                             xd->up_available, xd->left_available,
-                             xd->right_available);
-}
-
-void vp9_build_intra_predictors_mbuv(MACROBLOCKD *xd) {
-  vp9_build_intra_predictors_mbuv_internal(xd, &xd->predictor[256],
-                                           &xd->predictor[320], 8,
-                                           xd->mode_info_context->mbmi.uv_mode,
-                                           8);
-}
-
 void vp9_intra8x8_predict(MACROBLOCKD *xd,
                           BLOCKD *b,
                           int mode,
diff --git a/vp9/common/vp9_rtcd_defs.sh b/vp9/common/vp9_rtcd_defs.sh
index ae5b7fbc4..f9f2395f3 100644
--- a/vp9/common/vp9_rtcd_defs.sh
+++ b/vp9/common/vp9_rtcd_defs.sh
@@ -68,11 +68,15 @@ specialize vp9_recon_b
 prototype void vp9_recon_uv_b "uint8_t *pred_ptr, int16_t *diff_ptr, uint8_t *dst_ptr, int stride"
 specialize vp9_recon_uv_b
 
+# TODO(jingning): The prototype functions in c are modified to enable block-size configurable
+# operations. Need to change the sse2 accrodingly.
 prototype void vp9_recon2b "uint8_t *pred_ptr, int16_t *diff_ptr, uint8_t *dst_ptr, int stride"
-specialize vp9_recon2b sse2
+specialize vp9_recon2b
+# specialize vp9_recon2b sse2
 
 prototype void vp9_recon4b "uint8_t *pred_ptr, int16_t *diff_ptr, uint8_t *dst_ptr, int stride"
-specialize vp9_recon4b sse2
+specialize vp9_recon4b
+# specialize vp9_recon4b sse2
 
 prototype void vp9_recon_mb "struct macroblockd *x"
 specialize vp9_recon_mb
@@ -86,17 +90,14 @@ specialize vp9_recon_sby_s
 prototype void vp9_recon_sbuv_s "struct macroblockd *x, uint8_t *udst, uint8_t *vdst, enum BLOCK_SIZE_TYPE bsize"
 specialize void vp9_recon_sbuv_s
 
+prototype void vp9_build_intra_predictors "uint8_t *src, int src_stride, uint8_t *pred, int y_stride, int mode, int bw, int bh, int up_available, int left_available, int right_available"
+specialize void vp9_build_intra_predictors
+
 prototype void vp9_build_intra_predictors_sby_s "struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize"
-specialize vp9_build_intra_predictors_sby_s;
+specialize vp9_build_intra_predictors_sby_s
 
 prototype void vp9_build_intra_predictors_sbuv_s "struct macroblockd *x, enum BLOCK_SIZE_TYPE bsize"
-specialize vp9_build_intra_predictors_sbuv_s;
-
-prototype void vp9_build_intra_predictors_mby "struct macroblockd *x"
-specialize vp9_build_intra_predictors_mby;
-
-prototype void vp9_build_intra_predictors_mbuv "struct macroblockd *x"
-specialize vp9_build_intra_predictors_mbuv;
+specialize vp9_build_intra_predictors_sbuv_s
 
 prototype void vp9_intra4x4_predict "struct macroblockd *xd, struct blockd *x, int b_mode, uint8_t *predictor, int pre_stride"
 specialize vp9_intra4x4_predict;
@@ -620,16 +621,10 @@ specialize vp9_block_error mmx sse2
 vp9_block_error_sse2=vp9_block_error_xmm
 
 prototype void vp9_subtract_b "struct block *be, struct blockd *bd, int pitch"
-specialize vp9_subtract_b mmx sse2
-
-prototype void vp9_subtract_b "struct block *be, struct blockd *bd, int pitch"
-specialize vp9_subtract_b mmx sse2
-
-prototype void vp9_subtract_mby "int16_t *diff, uint8_t *src, uint8_t *pred, int stride"
-specialize vp9_subtract_mby mmx sse2
-
-prototype void vp9_subtract_mbuv "int16_t *diff, uint8_t *usrc, uint8_t *vsrc, uint8_t *pred, int stride"
-specialize vp9_subtract_mbuv mmx sse2
+# TODO(jingning): The prototype function in c has been changed to remove
+# the use of predictor buffer in MACROBLOCKD. Need to modify the mmx and sse2
+# versions accordingly.
+specialize vp9_subtract_b
 
 #
 # Structured Similarity (SSIM)
diff --git a/vp9/common/x86/vp9_recon_wrapper_sse2.c b/vp9/common/x86/vp9_recon_wrapper_sse2.c
index bb7baf8a0..12d2f970c 100644
--- a/vp9/common/x86/vp9_recon_wrapper_sse2.c
+++ b/vp9/common/x86/vp9_recon_wrapper_sse2.c
@@ -73,15 +73,15 @@ static void build_intra_predictors_mbuv_x86(MACROBLOCKD *xd,
 }
 
 void vp9_build_intra_predictors_mbuv_sse2(MACROBLOCKD *xd) {
-  build_intra_predictors_mbuv_x86(xd, &xd->predictor[256],
-                                  &xd->predictor[320], 8,
+  build_intra_predictors_mbuv_x86(xd, xd->dst.u_buffer,
+                                  xd->dst.v_buffer, xd->dst.uv_stride,
                                   vp9_intra_pred_uv_tm_sse2,
                                   vp9_intra_pred_uv_ho_mmx2);
 }
 
 void vp9_build_intra_predictors_mbuv_ssse3(MACROBLOCKD *xd) {
-  build_intra_predictors_mbuv_x86(xd, &xd->predictor[256],
-                                  &xd->predictor[320], 8,
+  build_intra_predictors_mbuv_x86(xd, xd->dst.u_buffer,
+                                  xd->dst.v_buffer, xd->dst.uv_stride,
                                   vp9_intra_pred_uv_tm_ssse3,
                                   vp9_intra_pred_uv_ho_ssse3);
 }
diff --git a/vp9/decoder/vp9_decodframe.c b/vp9/decoder/vp9_decodframe.c
index d3b18d765..20d4f19c0 100644
--- a/vp9/decoder/vp9_decodframe.c
+++ b/vp9/decoder/vp9_decodframe.c
@@ -645,7 +645,7 @@ static void decode_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
            xd->mode_info_context->mbmi.mode, tx_size,
            xd->mode_info_context->mbmi.interp_filter);
 #endif
-    vp9_build_inter_predictors_mb_s(xd, mb_row, mb_col);
+    vp9_build_inter_predictors_mb(xd, mb_row, mb_col);
   }
 
   if (xd->mode_info_context->mbmi.mb_skip_coeff) {
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 6f0e8c7f4..8db2796db 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -1405,9 +1405,9 @@ static void encode_frame_internal(VP9_COMP *cpi) {
   MACROBLOCKD *const xd = &x->e_mbd;
   int totalrate;
 
-//   fprintf(stderr, "encode_frame_internal frame %d (%d) type %d\n",
-//            cpi->common.current_video_frame, cpi->common.show_frame,
-//            cm->frame_type);
+//  fprintf(stderr, "encode_frame_internal frame %d (%d) type %d\n",
+//           cpi->common.current_video_frame, cpi->common.show_frame,
+//           cm->frame_type);
 
   // Compute a modified set of reference frame probabilities to use when
   // prediction fails. These are based on the current general estimates for
@@ -2230,15 +2230,8 @@ static void encode_macroblock(VP9_COMP *cpi, TOKENEXTRA **t,
 
     if (!x->skip) {
       vp9_encode_inter16x16(cm, x, mb_row, mb_col);
-
     } else {
-      vp9_build_inter16x16_predictors_mb(xd,
-                                         xd->dst.y_buffer,
-                                         xd->dst.u_buffer,
-                                         xd->dst.v_buffer,
-                                         xd->dst.y_stride,
-                                         xd->dst.uv_stride,
-                                         mb_row, mb_col);
+      vp9_build_inter_predictors_sb(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
 #if CONFIG_COMP_INTERINTRA_PRED
       if (xd->mode_info_context->mbmi.second_ref_frame == INTRA_FRAME) {
         vp9_build_interintra_16x16_predictors_mb(xd,
diff --git a/vp9/encoder/vp9_encodeintra.c b/vp9/encoder/vp9_encodeintra.c
index 355867ba7..bccd22bf9 100644
--- a/vp9/encoder/vp9_encodeintra.c
+++ b/vp9/encoder/vp9_encodeintra.c
@@ -52,7 +52,8 @@ static void encode_intra4x4block(MACROBLOCK *x, int ib) {
   b->bmi.as_mode.context = vp9_find_bpred_context(&x->e_mbd, b);
 #endif
 
-  vp9_intra4x4_predict(&x->e_mbd, b, b->bmi.as_mode.first, b->predictor, 16);
+  vp9_intra4x4_predict(&x->e_mbd, b, b->bmi.as_mode.first,
+                       *(b->base_dst) + b->dst, b->dst_stride);
   vp9_subtract_b(be, b, 16);
 
   tx_type = get_tx_type_4x4(&x->e_mbd, ib);
@@ -69,7 +70,8 @@ static void encode_intra4x4block(MACROBLOCK *x, int ib) {
                                 b->diff, 32);
   }
 
-  vp9_recon_b(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+  vp9_recon_b(*(b->base_dst) + b->dst, b->diff,
+              *(b->base_dst) + b->dst, b->dst_stride);
 }
 
 void vp9_encode_intra4x4mby(MACROBLOCK *mb) {
@@ -81,12 +83,13 @@ void vp9_encode_intra4x4mby(MACROBLOCK *mb) {
 
 void vp9_encode_intra16x16mby(VP9_COMMON *const cm, MACROBLOCK *x) {
   MACROBLOCKD *xd = &x->e_mbd;
-  BLOCK *b = &x->block[0];
   TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size;
 
-  vp9_build_intra_predictors_mby(xd);
-
-  vp9_subtract_mby(x->src_diff, *(b->base_src), xd->predictor, b->src_stride);
+  vp9_build_intra_predictors_sby_s(xd, BLOCK_SIZE_MB16X16);
+  vp9_subtract_sby_s_c(x->src_diff,
+                       x->src.y_buffer, x->src.y_stride,
+                       xd->dst.y_buffer, xd->dst.y_stride,
+                       BLOCK_SIZE_MB16X16);
 
   switch (tx_size) {
     case TX_16X16:
@@ -119,10 +122,11 @@ void vp9_encode_intra16x16mbuv(VP9_COMMON *const cm, MACROBLOCK *x) {
   MACROBLOCKD *xd = &x->e_mbd;
   TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size;
 
-  vp9_build_intra_predictors_mbuv(xd);
-
-  vp9_subtract_mbuv(x->src_diff, x->src.u_buffer, x->src.v_buffer,
-                    xd->predictor, x->src.uv_stride);
+  vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
+  vp9_subtract_sbuv_s_c(x->src_diff,
+                        x->src.u_buffer, x->src.v_buffer, x->src.uv_stride,
+                        xd->dst.u_buffer, xd->dst.v_buffer, xd->dst.uv_stride,
+                        BLOCK_SIZE_MB16X16);
 
   switch (tx_size) {
     case TX_4X4:
@@ -152,7 +156,8 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
   int i;
   TX_TYPE tx_type;
 
-  vp9_intra8x8_predict(xd, b, b->bmi.as_mode.first, b->predictor, 16);
+  vp9_intra8x8_predict(xd, b, b->bmi.as_mode.first,
+                       *(b->base_dst) + b->dst, b->dst_stride);
   // generate residual blocks
   vp9_subtract_4b_c(be, b, 16);
 
@@ -206,7 +211,7 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
   // reconstruct submacroblock
   for (i = 0; i < 4; i++) {
     b = &xd->block[ib + iblock[i]];
-    vp9_recon_b_c(b->predictor, b->diff, *(b->base_dst) + b->dst,
+    vp9_recon_b_c(*(b->base_dst) + b->dst, b->diff, *(b->base_dst) + b->dst,
                   b->dst_stride);
   }
 }
@@ -227,7 +232,8 @@ static void encode_intra_uv4x4(MACROBLOCK *x, int ib, int mode) {
   const int block = ib < 20 ? ib - 16 : ib - 20;
 
   assert(ib >= 16 && ib < 24);
-  vp9_intra_uv4x4_predict(&x->e_mbd, b, mode, b->predictor, 8);
+  vp9_intra_uv4x4_predict(&x->e_mbd, b, mode,
+                          *(b->base_dst) + b->dst, b->dst_stride);
 
   vp9_subtract_b(be, b, 8);
 
@@ -236,7 +242,7 @@ static void encode_intra_uv4x4(MACROBLOCK *x, int ib, int mode) {
   vp9_inverse_transform_b_4x4(&x->e_mbd, xd->plane[plane].eobs[block],
                               dqcoeff, b->diff, 16);
 
-  vp9_recon_uv_b_c(b->predictor, b->diff, *(b->base_dst) + b->dst,
+  vp9_recon_uv_b_c(*(b->base_dst) + b->dst, b->diff, *(b->base_dst) + b->dst,
                    b->dst_stride);
 }
 
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index 6ecaaa509..e786532a1 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -23,8 +23,9 @@
 void vp9_subtract_b_c(BLOCK *be, BLOCKD *bd, int pitch) {
   uint8_t *src_ptr = (*(be->base_src) + be->src);
   int16_t *diff_ptr = be->src_diff;
-  uint8_t *pred_ptr = bd->predictor;
+  uint8_t *pred_ptr = *(bd->base_dst) + bd->dst;
   int src_stride = be->src_stride;
+  int dst_stride = bd->dst_stride;
 
   int r, c;
 
@@ -33,7 +34,7 @@ void vp9_subtract_b_c(BLOCK *be, BLOCKD *bd, int pitch) {
       diff_ptr[c] = src_ptr[c] - pred_ptr[c];
 
     diff_ptr += pitch;
-    pred_ptr += pitch;
+    pred_ptr += dst_stride;
     src_ptr  += src_stride;
   }
 }
@@ -41,8 +42,9 @@ void vp9_subtract_b_c(BLOCK *be, BLOCKD *bd, int pitch) {
 void vp9_subtract_4b_c(BLOCK *be, BLOCKD *bd, int pitch) {
   uint8_t *src_ptr = (*(be->base_src) + be->src);
   int16_t *diff_ptr = be->src_diff;
-  uint8_t *pred_ptr = bd->predictor;
+  uint8_t *pred_ptr = *(bd->base_dst) + bd->dst;
   int src_stride = be->src_stride;
+  int dst_stride = bd->dst_stride;
   int r, c;
 
   for (r = 0; r < 8; r++) {
@@ -50,7 +52,7 @@ void vp9_subtract_4b_c(BLOCK *be, BLOCKD *bd, int pitch) {
       diff_ptr[c] = src_ptr[c] - pred_ptr[c];
 
     diff_ptr += pitch;
-    pred_ptr += pitch;
+    pred_ptr += dst_stride;
     src_ptr  += src_stride;
   }
 }
@@ -102,25 +104,15 @@ void vp9_subtract_sbuv_s_c(int16_t *diff, const uint8_t *usrc,
   }
 }
 
-void vp9_subtract_mby_c(int16_t *diff, uint8_t *src,
-                        uint8_t *pred, int stride) {
-  vp9_subtract_sby_s_c(diff, src, stride, pred, 16, BLOCK_SIZE_MB16X16);
-}
-
-void vp9_subtract_mbuv_c(int16_t *diff, uint8_t *usrc,
-                         uint8_t *vsrc, uint8_t *pred, int stride) {
-  uint8_t *upred = pred + 256;
-  uint8_t *vpred = pred + 320;
-
-  vp9_subtract_sbuv_s_c(diff, usrc, vsrc, stride, upred, vpred, 8,
-                        BLOCK_SIZE_MB16X16);
-}
-
 static void subtract_mb(MACROBLOCK *x) {
-  vp9_subtract_mby(x->src_diff, x->src.y_buffer, x->e_mbd.predictor,
-                   x->src.y_stride);
-  vp9_subtract_mbuv(x->src_diff, x->src.u_buffer, x->src.v_buffer,
-                    x->e_mbd.predictor, x->src.uv_stride);
+  MACROBLOCKD *xd = &x->e_mbd;
+  vp9_subtract_sby_s_c(x->src_diff, x->src.y_buffer, x->src.y_stride,
+                       xd->dst.y_buffer, xd->dst.y_stride,
+                       BLOCK_SIZE_MB16X16);
+  vp9_subtract_sbuv_s_c(x->src_diff, x->src.u_buffer, x->src.v_buffer,
+                        x->src.uv_stride,
+                        xd->dst.u_buffer, xd->dst.v_buffer, xd->dst.uv_stride,
+                        BLOCK_SIZE_MB16X16);
 }
 
 void vp9_transform_sby_32x32(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
@@ -920,11 +912,12 @@ void vp9_encode_inter16x16(VP9_COMMON *const cm, MACROBLOCK *x,
 /* this function is used by first pass only */
 void vp9_encode_inter16x16y(MACROBLOCK *x, int mb_row, int mb_col) {
   MACROBLOCKD *xd = &x->e_mbd;
-  BLOCK *b = &x->block[0];
-
-  vp9_build_inter16x16_predictors_mby(xd, xd->predictor, 16, mb_row, mb_col);
 
-  vp9_subtract_mby(x->src_diff, *(b->base_src), xd->predictor, b->src_stride);
+  vp9_build_inter_predictors_sby(xd, xd->dst.y_buffer, xd->dst.y_stride,
+                                 mb_row, mb_col, BLOCK_SIZE_MB16X16);
+  vp9_subtract_sby_s_c(x->src_diff, x->src.y_buffer, x->src.y_stride,
+                       xd->dst.y_buffer, xd->dst.y_stride,
+                       BLOCK_SIZE_MB16X16);
 
   vp9_transform_sby_4x4(x, BLOCK_SIZE_MB16X16);
   vp9_quantize_sby_4x4(x, BLOCK_SIZE_MB16X16);
diff --git a/vp9/encoder/vp9_encodemb.h b/vp9/encoder/vp9_encodemb.h
index 76fb0f7cd..3c0d760a1 100644
--- a/vp9/encoder/vp9_encodemb.h
+++ b/vp9/encoder/vp9_encodemb.h
@@ -58,13 +58,6 @@ void vp9_fidct_mb(VP9_COMMON *const cm, MACROBLOCK *x);
 
 void vp9_subtract_4b_c(BLOCK *be, BLOCKD *bd, int pitch);
 
-void vp9_subtract_mbuv_s_c(int16_t *diff, const uint8_t *usrc,
-                           const uint8_t *vsrc, int src_stride,
-                           const uint8_t *upred,
-                           const uint8_t *vpred, int dst_stride);
-void vp9_subtract_mby_s_c(int16_t *diff, const uint8_t *src,
-                          int src_stride, const uint8_t *pred,
-                          int dst_stride);
 void vp9_subtract_sby_s_c(int16_t *diff, const uint8_t *src, int src_stride,
                           const uint8_t *pred, int dst_stride,
                           BLOCK_SIZE_TYPE bsize);
diff --git a/vp9/encoder/vp9_mbgraph.c b/vp9/encoder/vp9_mbgraph.c
index 715d68377..e9da395bc 100644
--- a/vp9/encoder/vp9_mbgraph.c
+++ b/vp9/encoder/vp9_mbgraph.c
@@ -71,9 +71,10 @@ static unsigned int do_16x16_motion_iteration(VP9_COMP *cpi,
   }
 
   vp9_set_mbmode_and_mvs(x, NEWMV, dst_mv);
-  vp9_build_inter16x16_predictors_mby(xd, xd->predictor, 16, mb_row, mb_col);
-  best_err = vp9_sad16x16(xd->dst.y_buffer, xd->dst.y_stride,
-                          xd->predictor, 16, INT_MAX);
+  vp9_build_inter_predictors_sby(xd, xd->dst.y_buffer, xd->dst.y_stride,
+                                 mb_row, mb_col, BLOCK_SIZE_MB16X16);
+  best_err = vp9_sad16x16(x->src.y_buffer, x->src.y_stride,
+                          xd->dst.y_buffer, xd->dst.y_stride, INT_MAX);
 
   /* restore UMV window */
   x->mv_col_min = tmp_col_min;
@@ -105,21 +106,19 @@ static int do_16x16_motion_search
     BLOCKD *d = &xd->block[n];
     BLOCK *b  = &x->block[n];
 
-    b->base_src   = &buf->y_buffer;
-    b->src_stride = buf->y_stride;
-    b->src        = buf->y_stride * (n & 12) + (n & 3) * 4 + buf_mb_y_offset;
+    b->base_src   = &x->src.y_buffer;
+    b->src_stride = x->src.y_stride;
+    b->src        = x->src.y_stride * (n & 12) + (n & 3) * 4;
 
-    d->base_pre   = &ref->y_buffer;
-    d->pre_stride = ref->y_stride;
-    d->pre        = ref->y_stride * (n & 12) + (n & 3) * 4 + mb_y_offset;
+    d->base_pre   = &xd->pre.y_buffer;
+    d->pre_stride = xd->pre.y_stride;
+    d->pre        = xd->pre.y_stride * (n & 12) + (n & 3) * 4;
   }
 
   // Try zero MV first
   // FIXME should really use something like near/nearest MV and/or MV prediction
-  xd->pre.y_buffer = ref->y_buffer + mb_y_offset;
-  xd->pre.y_stride = ref->y_stride;
-  err = vp9_sad16x16(ref->y_buffer + mb_y_offset, ref->y_stride,
-                     xd->dst.y_buffer, xd->dst.y_stride, INT_MAX);
+  err = vp9_sad16x16(x->src.y_buffer, x->src.y_stride,
+                     xd->pre.y_buffer, xd->pre.y_stride, INT_MAX);
   dst_mv->as_int = 0;
 
   // Test last reference frame using the previous best mv as the
@@ -159,27 +158,11 @@ static int do_16x16_zerozero_search
   MACROBLOCK   *const x  = &cpi->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
   unsigned int err;
-  int n;
-
-  for (n = 0; n < 16; n++) {
-    BLOCKD *d = &xd->block[n];
-    BLOCK *b  = &x->block[n];
-
-    b->base_src   = &buf->y_buffer;
-    b->src_stride = buf->y_stride;
-    b->src        = buf->y_stride * (n & 12) + (n & 3) * 4 + buf_mb_y_offset;
-
-    d->base_pre   = &ref->y_buffer;
-    d->pre_stride = ref->y_stride;
-    d->pre        = ref->y_stride * (n & 12) + (n & 3) * 4 + mb_y_offset;
-  }
 
   // Try zero MV first
   // FIXME should really use something like near/nearest MV and/or MV prediction
-  xd->pre.y_buffer = ref->y_buffer + mb_y_offset;
-  xd->pre.y_stride = ref->y_stride;
-  err = vp9_sad16x16(ref->y_buffer + mb_y_offset, ref->y_stride,
-                     xd->dst.y_buffer, xd->dst.y_stride, INT_MAX);
+  err = vp9_sad16x16(x->src.y_buffer, x->src.y_stride,
+                     xd->pre.y_buffer, xd->pre.y_stride, INT_MAX);
 
   dst_mv->as_int = 0;
 
@@ -201,11 +184,19 @@ static int find_best_16x16_intra
   // we're intentionally not doing 4x4, we just want a rough estimate
   for (mode = DC_PRED; mode <= TM_PRED; mode++) {
     unsigned int err;
+    const int bwl = b_width_log2(BLOCK_SIZE_MB16X16),  bw = 4 << bwl;
+    const int bhl = b_height_log2(BLOCK_SIZE_MB16X16), bh = 4 << bhl;
 
     xd->mode_info_context->mbmi.mode = mode;
-    vp9_build_intra_predictors_mby(xd);
-    err = vp9_sad16x16(xd->predictor, 16, buf->y_buffer + mb_y_offset,
-                       buf->y_stride, best_err);
+    vp9_build_intra_predictors(x->src.y_buffer, x->src.y_stride,
+                               xd->dst.y_buffer, xd->dst.y_stride,
+                               xd->mode_info_context->mbmi.mode,
+                               bw, bh,
+                               xd->up_available, xd->left_available,
+                               xd->right_available);
+    err = vp9_sad16x16(x->src.y_buffer, x->src.y_stride,
+                       xd->dst.y_buffer, xd->dst.y_stride, best_err);
+
     // find best
     if (err < best_err) {
       best_err  = err;
@@ -237,23 +228,32 @@ static void update_mbgraph_mb_stats
   MACROBLOCK   *const x  = &cpi->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
   int intra_error;
+  VP9_COMMON *cm = &cpi->common;
 
   // FIXME in practice we're completely ignoring chroma here
-  xd->dst.y_buffer = buf->y_buffer + mb_y_offset;
+  x->src.y_buffer = buf->y_buffer + mb_y_offset;
+  x->src.y_stride = buf->y_stride;
+
+  xd->dst.y_buffer = cm->yv12_fb[cm->new_fb_idx].y_buffer + mb_y_offset;
+  xd->dst.y_stride = cm->yv12_fb[cm->new_fb_idx].y_stride;
 
   // do intra 16x16 prediction
-  intra_error = find_best_16x16_intra(cpi, buf, mb_y_offset, &stats->ref[INTRA_FRAME].m.mode);
+  intra_error = find_best_16x16_intra(cpi, buf, mb_y_offset,
+                                      &stats->ref[INTRA_FRAME].m.mode);
   if (intra_error <= 0)
     intra_error = 1;
   stats->ref[INTRA_FRAME].err = intra_error;
 
   // Golden frame MV search, if it exists and is different than last frame
   if (golden_ref) {
-    int g_motion_error = do_16x16_motion_search(cpi, prev_golden_ref_mv,
-                                                &stats->ref[GOLDEN_FRAME].m.mv,
-                                                buf, mb_y_offset,
-                                                golden_ref, gld_y_offset,
-                                                mb_row, mb_col);
+    int g_motion_error;
+    xd->pre.y_buffer = golden_ref->y_buffer + mb_y_offset;
+    xd->pre.y_stride = golden_ref->y_stride;
+    g_motion_error = do_16x16_motion_search(cpi, prev_golden_ref_mv,
+                                            &stats->ref[GOLDEN_FRAME].m.mv,
+                                            buf, mb_y_offset,
+                                            golden_ref, gld_y_offset,
+                                            mb_row, mb_col);
     stats->ref[GOLDEN_FRAME].err = g_motion_error;
   } else {
     stats->ref[GOLDEN_FRAME].err = INT_MAX;
@@ -262,16 +262,13 @@ static void update_mbgraph_mb_stats
 
   // Alt-ref frame MV search, if it exists and is different than last/golden frame
   if (alt_ref) {
-    // int a_motion_error = do_16x16_motion_search(cpi, prev_alt_ref_mv,
-    //                                            &stats->ref[ALTREF_FRAME].m.mv,
-    //                                            buf, mb_y_offset,
-    //                                            alt_ref, arf_y_offset);
-
-    int a_motion_error =
-      do_16x16_zerozero_search(cpi,
-                               &stats->ref[ALTREF_FRAME].m.mv,
-                               buf, mb_y_offset,
-                               alt_ref, arf_y_offset);
+    int a_motion_error;
+    xd->pre.y_buffer = alt_ref->y_buffer + mb_y_offset;
+    xd->pre.y_stride = alt_ref->y_stride;
+    a_motion_error = do_16x16_zerozero_search(cpi,
+                                              &stats->ref[ALTREF_FRAME].m.mv,
+                                              buf, mb_y_offset,
+                                              alt_ref, arf_y_offset);
 
     stats->ref[ALTREF_FRAME].err = a_motion_error;
   } else {
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 2f29b1dc3..73c126912 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -638,15 +638,6 @@ static void choose_txfm_size_from_rd(VP9_COMP *cpi, MACROBLOCK *x,
                                  rd[TX_4X4][1] : rd[TX_8X8][1];
 }
 
-static void copy_predictor(uint8_t *dst, const uint8_t *predictor) {
-  const unsigned int *p = (const unsigned int *)predictor;
-  unsigned int *d = (unsigned int *)dst;
-  d[0] = p[0];
-  d[4] = p[4];
-  d[8] = p[8];
-  d[12] = p[12];
-}
-
 static int vp9_sb_block_error_c(int16_t *coeff, int16_t *dqcoeff,
                                 int block_size, int shift) {
   int i;
@@ -849,13 +840,7 @@ static void super_block_yrd(VP9_COMP *cpi,
   uint8_t *src = x->src.y_buffer, *dst = xd->dst.y_buffer;
   int src_y_stride = x->src.y_stride, dst_y_stride = xd->dst.y_stride;
 
-  // FIXME(rbultje): mb code still predicts into xd->predictor
-  if (bs == BLOCK_SIZE_MB16X16) {
-    vp9_subtract_mby(x->src_diff, src, xd->predictor, src_y_stride);
-  } else {
-    vp9_subtract_sby_s_c(x->src_diff, src, src_y_stride, dst, dst_y_stride,
-                         bs);
-  }
+  vp9_subtract_sby_s_c(x->src_diff, src, src_y_stride, dst, dst_y_stride, bs);
 
   if (bs >= BLOCK_SIZE_SB32X32)
     super_block_yrd_32x32(cm, x, &r[TX_32X32][0], &d[TX_32X32], &s[TX_32X32],
@@ -892,7 +877,6 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
    * a temp buffer that meets the stride requirements, but we are only
    * interested in the left 4x4 block
    * */
-  DECLARE_ALIGNED_ARRAY(16, uint8_t, best_predictor, 16 * 4);
   DECLARE_ALIGNED_ARRAY(16, int16_t, best_dqcoeff, 16);
 
   assert(ib < 16);
@@ -922,7 +906,7 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
     rate = bmode_costs[mode];
 #endif
 
-    vp9_intra4x4_predict(xd, b, mode, b->predictor, 16);
+    vp9_intra4x4_predict(xd, b, mode, *(b->base_dst) + b->dst, b->dst_stride);
     vp9_subtract_b(be, b, 16);
 
     b->bmi.as_mode.first = mode;
@@ -956,7 +940,6 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
       best_tx_type = tx_type;
       *a = tempa;
       *l = templ;
-      copy_predictor(best_predictor, b->predictor);
       vpx_memcpy(best_dqcoeff, BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16), 32);
     }
   }
@@ -968,7 +951,10 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
   else
     xd->inv_txm4x4(best_dqcoeff, b->diff, 32);
 
-  vp9_recon_b(best_predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+  vp9_intra4x4_predict(xd, b, *best_mode,
+                       *(b->base_dst) + b->dst, b->dst_stride);
+  vp9_recon_b(*(b->base_dst) + b->dst, b->diff,
+              *(b->base_dst) + b->dst, b->dst_stride);
 
   return best_rd;
 }
@@ -1063,11 +1049,7 @@ static int64_t rd_pick_intra_sby_mode(VP9_COMP *cpi, MACROBLOCK *x,
     int64_t local_txfm_cache[NB_TXFM_MODES];
 
     x->e_mbd.mode_info_context->mbmi.mode = mode;
-    if (bsize == BLOCK_SIZE_MB16X16) {
-      vp9_build_intra_predictors_mby(&x->e_mbd);
-    } else {
-      vp9_build_intra_predictors_sby_s(&x->e_mbd, bsize);
-    }
+    vp9_build_intra_predictors_sby_s(&x->e_mbd, bsize);
 
     super_block_yrd(cpi, x, &this_rate_tokenonly, &this_distortion, &s,
                     bsize, local_txfm_cache);
@@ -1129,7 +1111,7 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
     rate = mode_costs[mode];
     b->bmi.as_mode.first = mode;
 
-    vp9_intra8x8_predict(xd, b, mode, b->predictor, 16);
+    vp9_intra8x8_predict(xd, b, mode, *(b->base_dst) + b->dst, b->dst_stride);
 
     vp9_subtract_4b_c(be, b, 16);
 
@@ -1543,14 +1525,8 @@ static void super_block_uvrd(VP9_COMMON *const cm, MACROBLOCK *x,
   uint8_t *vsrc = x->src.v_buffer, *vdst = xd->dst.v_buffer;
   int src_uv_stride = x->src.uv_stride, dst_uv_stride = xd->dst.uv_stride;
 
-  // FIXME(rbultje): mb code still predicts into xd->predictor
-  if (bsize == BLOCK_SIZE_MB16X16) {
-    vp9_subtract_mbuv(x->src_diff, usrc, vsrc, xd->predictor,
-                      x->src.uv_stride);
-  } else {
-    vp9_subtract_sbuv_s_c(x->src_diff, usrc, vsrc, src_uv_stride,
-                          udst, vdst, dst_uv_stride, bsize);
-  }
+  vp9_subtract_sbuv_s_c(x->src_diff, usrc, vsrc, src_uv_stride,
+                        udst, vdst, dst_uv_stride, bsize);
 
   if (mbmi->txfm_size >= TX_32X32 && bsize >= BLOCK_SIZE_SB64X64) {
     super_block_uvrd_32x32(cm, x, rate, distortion, skippable, bsize);
@@ -1576,10 +1552,7 @@ static int64_t rd_pick_intra_sbuv_mode(VP9_COMP *cpi, MACROBLOCK *x,
 
   for (mode = DC_PRED; mode <= TM_PRED; mode++) {
     x->e_mbd.mode_info_context->mbmi.uv_mode = mode;
-    if (bsize == BLOCK_SIZE_MB16X16)
-      vp9_build_intra_predictors_mbuv(&x->e_mbd);
-    else
-      vp9_build_intra_predictors_sbuv_s(&x->e_mbd, bsize);
+    vp9_build_intra_predictors_sbuv_s(&x->e_mbd, bsize);
 
     super_block_uvrd(&cpi->common, x, &this_rate_tokenonly,
                      &this_distortion, &s, bsize);
@@ -1759,7 +1732,8 @@ static int64_t encode_inter_mb_segment(VP9_COMMON *const cm,
 
       vp9_build_inter_predictor(*(bd->base_pre) + bd->pre,
                                 bd->pre_stride,
-                                bd->predictor, 16,
+                                *(bd->base_dst) + bd->dst,
+                                bd->dst_stride,
                                 &bd->bmi.as_mv[0],
                                 &xd->scale_factor[0],
                                 4, 4, 0 /* no avg */, &xd->subpix);
@@ -1769,7 +1743,8 @@ static int64_t encode_inter_mb_segment(VP9_COMMON *const cm,
       // weighting for splitmv modes is turned on.
       if (xd->mode_info_context->mbmi.second_ref_frame > 0) {
         vp9_build_inter_predictor(
-            *(bd->base_second_pre) + bd->pre, bd->pre_stride, bd->predictor, 16,
+            *(bd->base_second_pre) + bd->pre, bd->pre_stride,
+            *(bd->base_dst) + bd->dst, bd->dst_stride,
             &bd->bmi.as_mv[1], &xd->scale_factor[1], 4, 4,
             1 << (2 * CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT) /* avg */,
             &xd->subpix);
@@ -1834,7 +1809,8 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
         // implicit-compoundinter-weight experiment when implicit
         // weighting for splitmv modes is turned on.
         vp9_build_inter_predictor(
-            *base_pre + bd->pre, bd->pre_stride, bd->predictor, 16,
+            *base_pre + bd->pre, bd->pre_stride,
+            *(bd->base_dst) + bd->dst, bd->dst_stride,
             &bd->bmi.as_mv[which_mv], &xd->scale_factor[which_mv], 8, 8,
             which_mv << (2 * CONFIG_IMPLICIT_COMPOUNDINTER_WEIGHT),
             &xd->subpix);
@@ -3144,23 +3120,20 @@ static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
         unsigned int sse, var;
         int tmp_rate_y, tmp_rate_u, tmp_rate_v;
         int tmp_dist_y, tmp_dist_u, tmp_dist_v;
-        vp9_build_inter16x16_predictors_mb(xd, xd->predictor,
-                                           xd->predictor + 256,
-                                           xd->predictor + 320,
-                                           16, 8, mb_row, mb_col);
+        vp9_build_inter_predictors_sb(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
         var = vp9_variance16x16(*(b->base_src), b->src_stride,
-                                xd->predictor, 16, &sse);
+                                xd->dst.y_buffer, xd->dst.y_stride, &sse);
         // Note our transform coeffs are 8 times an orthogonal transform.
         // Hence quantizer step is also 8 times. To get effective quantizer
         // we need to divide by 8 before sending to modeling function.
         model_rd_from_var_lapndz(var, 16 * 16, xd->block[0].dequant[1] >> 3,
                                  &tmp_rate_y, &tmp_dist_y);
         var = vp9_variance8x8(x->src.u_buffer, x->src.uv_stride,
-                              &xd->predictor[256], 8, &sse);
+                              xd->dst.u_buffer, xd->dst.uv_stride, &sse);
         model_rd_from_var_lapndz(var, 8 * 8, xd->block[16].dequant[1] >> 3,
                                  &tmp_rate_u, &tmp_dist_u);
         var = vp9_variance8x8(x->src.v_buffer, x->src.uv_stride,
-                              &xd->predictor[320], 8, &sse);
+                              xd->dst.v_buffer, xd->dst.uv_stride, &sse);
         model_rd_from_var_lapndz(var, 8 * 8, xd->block[20].dequant[1] >> 3,
                                  &tmp_rate_v, &tmp_dist_v);
         rd = RDCOST(x->rdmult, x->rddiv,
@@ -3184,9 +3157,19 @@ static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
       if ((cm->mcomp_filter_type == SWITCHABLE && newbest) ||
           (cm->mcomp_filter_type != SWITCHABLE &&
            cm->mcomp_filter_type == mbmi->interp_filter)) {
-        vpx_memcpy(tmp_ybuf, xd->predictor, sizeof(unsigned char) * 256);
-        vpx_memcpy(tmp_ubuf, xd->predictor + 256, sizeof(unsigned char) * 64);
-        vpx_memcpy(tmp_vbuf, xd->predictor + 320, sizeof(unsigned char) * 64);
+        int i;
+        for (i = 0; i < 16 * bh; ++i)
+          vpx_memcpy(tmp_ybuf + i * 16 * bw,
+                     xd->dst.y_buffer + i * xd->dst.y_stride,
+                     sizeof(unsigned char) * 16 * bw);
+        for (i = 0; i < 8 * bh; ++i)
+          vpx_memcpy(tmp_ubuf + i * 8 * bw,
+                     xd->dst.u_buffer + i * xd->dst.uv_stride,
+                     sizeof(unsigned char) * 8 * bw);
+        for (i = 0; i < 8 * bh; ++i)
+          vpx_memcpy(tmp_vbuf + i * 8 * bw,
+                     xd->dst.v_buffer + i * xd->dst.uv_stride,
+                     sizeof(unsigned char) * 8 * bw);
         pred_exists = 1;
       }
       interpolating_intpel_seen |=
@@ -3203,32 +3186,19 @@ static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
 
   if (pred_exists) {
     // FIXME(rbultje): mb code still predicts into xd->predictor
-    if (bsize != BLOCK_SIZE_MB16X16) {
-      for (i = 0; i < bh * 16; ++i)
-        vpx_memcpy(xd->dst.y_buffer + i * xd->dst.y_stride,
-                   tmp_ybuf + i * bw * 16, sizeof(unsigned char) * bw * 16);
-      for (i = 0; i < bh * 8; ++i)
-        vpx_memcpy(xd->dst.u_buffer + i * xd->dst.uv_stride,
-                   tmp_ubuf + i * bw * 8, sizeof(unsigned char) * bw * 8);
-      for (i = 0; i < bh * 8; ++i)
-        vpx_memcpy(xd->dst.v_buffer + i * xd->dst.uv_stride,
-                   tmp_vbuf + i * bw * 8, sizeof(unsigned char) * bw * 8);
-    } else {
-      vpx_memcpy(xd->predictor, tmp_ybuf, sizeof(unsigned char) * 256);
-      vpx_memcpy(xd->predictor + 256, tmp_ubuf, sizeof(unsigned char) * 64);
-      vpx_memcpy(xd->predictor + 320, tmp_vbuf, sizeof(unsigned char) * 64);
-    }
+    for (i = 0; i < bh * 16; ++i)
+      vpx_memcpy(xd->dst.y_buffer + i * xd->dst.y_stride,
+                 tmp_ybuf + i * bw * 16, sizeof(unsigned char) * bw * 16);
+    for (i = 0; i < bh * 8; ++i)
+      vpx_memcpy(xd->dst.u_buffer + i * xd->dst.uv_stride,
+                 tmp_ubuf + i * bw * 8, sizeof(unsigned char) * bw * 8);
+    for (i = 0; i < bh * 8; ++i)
+      vpx_memcpy(xd->dst.v_buffer + i * xd->dst.uv_stride,
+                 tmp_vbuf + i * bw * 8, sizeof(unsigned char) * bw * 8);
   } else {
     // Handles the special case when a filter that is not in the
     // switchable list (ex. bilinear, 6-tap) is indicated at the frame level
-    if (bsize > BLOCK_SIZE_MB16X16) {
-      vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
-    } else {
-      vp9_build_inter16x16_predictors_mb(xd, xd->predictor,
-                                         xd->predictor + 256,
-                                         xd->predictor + 320,
-                                         16, 8, mb_row, mb_col);
-    }
+    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
   }
 
   if (cpi->common.mcomp_filter_type == SWITCHABLE) {
@@ -3253,7 +3223,7 @@ static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
                                        &sse);
     } else {
       var = vp9_variance16x16(*(b->base_src), b->src_stride,
-                              xd->predictor, 16, &sse);
+                              xd->dst.y_buffer, xd->dst.y_stride, &sse);
     }
 
     if ((int)sse < threshold) {
@@ -3278,9 +3248,9 @@ static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
         } else {
           unsigned int sse2u, sse2v;
           var = vp9_variance8x8(x->src.u_buffer, x->src.uv_stride,
-                                xd->predictor + 256, 8, &sse2u);
+                                xd->dst.u_buffer, xd->dst.uv_stride, &sse2u);
           var = vp9_variance8x8(x->src.v_buffer, x->src.uv_stride,
-                                xd->predictor + 320, 8, &sse2v);
+                                xd->dst.v_buffer, xd->dst.uv_stride, &sse2v);
           sse2 = sse2u + sse2v;
         }
 
@@ -3614,7 +3584,8 @@ static void rd_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
         case TM_PRED:
           mbmi->ref_frame = INTRA_FRAME;
           // FIXME compound intra prediction
-          vp9_build_intra_predictors_mby(&x->e_mbd);
+          vp9_build_intra_predictors_sby_s(&x->e_mbd, BLOCK_SIZE_MB16X16);
+          // vp9_build_intra_predictors_mby(&x->e_mbd);
           super_block_yrd(cpi, x, &rate_y, &distortion, &skippable,
                           BLOCK_SIZE_MB16X16, txfm_cache);
           rate2 += rate_y;
@@ -3790,8 +3761,14 @@ static void rd_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
         int uv_skippable;
 
         vp9_build_inter4x4_predictors_mbuv(&x->e_mbd, mb_row, mb_col);
-        vp9_subtract_mbuv(x->src_diff, x->src.u_buffer, x->src.v_buffer,
-                          x->e_mbd.predictor, x->src.uv_stride);
+
+        vp9_subtract_sbuv_s_c(x->src_diff,
+                              x->src.u_buffer,
+                              x->src.v_buffer, x->src.uv_stride,
+                              xd->dst.u_buffer,
+                              xd->dst.v_buffer, xd->dst.uv_stride,
+                              BLOCK_SIZE_MB16X16);
+
         super_block_uvrd_4x4(cm, x, &rate_uv, &distortion_uv,
                              &uv_skippable, BLOCK_SIZE_MB16X16);
         rate2 += rate_uv;
diff --git a/vp9/encoder/x86/vp9_x86_csystemdependent.c b/vp9/encoder/x86/vp9_x86_csystemdependent.c
index 310f0d99d..04383fcb4 100644
--- a/vp9/encoder/x86/vp9_x86_csystemdependent.c
+++ b/vp9/encoder/x86/vp9_x86_csystemdependent.c
@@ -30,7 +30,9 @@ void vp9_subtract_b_mmx(BLOCK *be, BLOCKD *bd, int pitch) {
   unsigned char *z = *(be->base_src) + be->src;
   unsigned int  src_stride = be->src_stride;
   short *diff = &be->src_diff[0];
-  unsigned char *predictor = &bd->predictor[0];
+  unsigned char *predictor = *(bd->base_dst) + bd->dst;
+  // TODO(jingning): The prototype function in c has been changed. Need to
+  // modify the mmx and sse versions.
   vp9_subtract_b_mmx_impl(z, src_stride, diff, predictor, pitch);
 }
 
@@ -44,7 +46,9 @@ void vp9_subtract_b_sse2(BLOCK *be, BLOCKD *bd, int pitch) {
   unsigned char *z = *(be->base_src) + be->src;
   unsigned int  src_stride = be->src_stride;
   short *diff = &be->src_diff[0];
-  unsigned char *predictor = &bd->predictor[0];
+  unsigned char *predictor = *(bd->base_dst) + bd->dst;
+  // TODO(jingning): The prototype function in c has been changed. Need to
+  // modify the mmx and sse versions.
   vp9_subtract_b_sse2_impl(z, src_stride, diff, predictor, pitch);
 }