diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index 6427f7f..3c98f7d 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -27,16 +27,16 @@
 typedef struct {
   MODE_INFO mic;
   uint8_t *zcoeff_blk;
-  int16_t *coeff[MAX_MB_PLANE][2];
-  int16_t *qcoeff[MAX_MB_PLANE][2];
-  int16_t *dqcoeff[MAX_MB_PLANE][2];
-  uint16_t *eobs[MAX_MB_PLANE][2];
+  int16_t *coeff[MAX_MB_PLANE][3];
+  int16_t *qcoeff[MAX_MB_PLANE][3];
+  int16_t *dqcoeff[MAX_MB_PLANE][3];
+  uint16_t *eobs[MAX_MB_PLANE][3];
 
   // dual buffer pointers, 0: in use, 1: best in store
-  int16_t *coeff_pbuf[MAX_MB_PLANE][2];
-  int16_t *qcoeff_pbuf[MAX_MB_PLANE][2];
-  int16_t *dqcoeff_pbuf[MAX_MB_PLANE][2];
-  uint16_t *eobs_pbuf[MAX_MB_PLANE][2];
+  int16_t *coeff_pbuf[MAX_MB_PLANE][3];
+  int16_t *qcoeff_pbuf[MAX_MB_PLANE][3];
+  int16_t *dqcoeff_pbuf[MAX_MB_PLANE][3];
+  uint16_t *eobs_pbuf[MAX_MB_PLANE][3];
 
   int is_coded;
   int num_4x4_blk;
@@ -94,6 +94,7 @@
   MACROBLOCKD e_mbd;
   int skip_block;
   int select_txfm_size;
+  int skip_recode;
   int skip_optimize;
   int q_index;
 
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index e9b68cc..3e75f3b 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -377,6 +377,7 @@
   const int mis = cm->mode_info_stride;
   const int mi_width = num_8x8_blocks_wide_lookup[bsize];
   const int mi_height = num_8x8_blocks_high_lookup[bsize];
+  int max_plane;
 
   assert(mi->mbmi.mode < MB_MODE_COUNT);
   assert(mi->mbmi.ref_frame[0] < MAX_REF_FRAMES);
@@ -385,13 +386,21 @@
 
   *mi_addr = *mi;
 
-  for (i = 0; i < MAX_MB_PLANE; ++i) {
+  max_plane = is_inter_block(mbmi) ? MAX_MB_PLANE : 1;
+  for (i = 0; i < max_plane; ++i) {
     p[i].coeff = ctx->coeff_pbuf[i][1];
     pd[i].qcoeff = ctx->qcoeff_pbuf[i][1];
     pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1];
     pd[i].eobs = ctx->eobs_pbuf[i][1];
   }
 
+  for (i = max_plane; i < MAX_MB_PLANE; ++i) {
+    p[i].coeff = ctx->coeff_pbuf[i][2];
+    pd[i].qcoeff = ctx->qcoeff_pbuf[i][2];
+    pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][2];
+    pd[i].eobs = ctx->eobs_pbuf[i][2];
+  }
+
   // Restore the coding context of the MB to that that was in place
   // when the mode was picked for it
   for (y = 0; y < mi_height; y++)
@@ -619,6 +628,7 @@
     pd[i].eobs = ctx->eobs_pbuf[i][0];
   }
   ctx->is_coded = 0;
+  x->skip_recode = 0;
 
   // Set to zero to make sure we do not use the previous encoded frame stats
   xd->mi_8x8[0]->mbmi.skip_coeff = 0;
@@ -2406,6 +2416,7 @@
   const int mis = cm->mode_info_stride;
   const int mi_width = num_8x8_blocks_wide_lookup[bsize];
   const int mi_height = num_8x8_blocks_high_lookup[bsize];
+  x->skip_recode = !x->select_txfm_size && mbmi->sb_type >= BLOCK_8X8;
   x->skip_optimize = ctx->is_coded;
   ctx->is_coded = 1;
   x->use_lp32x32fdct = cpi->sf.use_lp32x32fdct;
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index b709848..ca46e85 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -431,19 +431,18 @@
   // 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) {
-    int i, k;
+    int i, j;
     pd->eobs[block] = 0;
-    txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &k);
+    txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
     ctx->ta[plane][i] = 0;
-    ctx->tl[plane][k] = 0;
+    ctx->tl[plane][j] = 0;
     return;
   }
 
-  if (x->select_txfm_size || xd->mi_8x8[0]->mbmi.sb_type < BLOCK_8X8)
+  if (!x->skip_recode)
     vp9_xform_quant(plane, block, plane_bsize, tx_size, arg);
 
-  if (x->optimize && (x->select_txfm_size ||
-      xd->mi_8x8[0]->mbmi.sb_type < BLOCK_8X8|| !x->skip_optimize)) {
+  if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
     vp9_optimize_b(plane, block, plane_bsize, tx_size, x, ctx);
   } else {
     int i, k;
@@ -514,10 +513,10 @@
   struct optimize_ctx ctx;
   struct encode_b_args arg = {x, &ctx};
 
-  if (x->select_txfm_size || xd->mi_8x8[0]->mbmi.sb_type < BLOCK_8X8)
+  if (!x->skip_recode)
     vp9_subtract_sb(x, bsize);
 
-  if (x->optimize) {
+  if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
     int i;
     for (i = 0; i < MAX_MB_PLANE; ++i)
       optimize_init_b(i, bsize, &arg);
@@ -562,19 +561,22 @@
       xoff = 32 * (block & twmask);
       yoff = 32 * (block >> twl);
       dst = pd->dst.buf + yoff * pd->dst.stride + xoff;
-      src = p->src.buf + yoff * p->src.stride + xoff;
-      src_diff = p->src_diff + 4 * bw * yoff + xoff;
       vp9_predict_intra_block(xd, block, bwl, TX_32X32, mode,
                               dst, pd->dst.stride, dst, pd->dst.stride);
-      vp9_subtract_block(32, 32, src_diff, bw * 4,
-                         src, p->src.stride, dst, pd->dst.stride);
-      if (x->use_lp32x32fdct)
-        vp9_fdct32x32_rd(src_diff, coeff, bw * 4);
-      else
-        vp9_fdct32x32(src_diff, coeff, bw * 4);
-      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, iscan);
+
+      if (!x->skip_recode) {
+        src = p->src.buf + yoff * p->src.stride + xoff;
+        src_diff = p->src_diff + 4 * bw * yoff + xoff;
+        vp9_subtract_block(32, 32, src_diff, bw * 4,
+                           src, p->src.stride, dst, pd->dst.stride);
+        if (x->use_lp32x32fdct)
+          vp9_fdct32x32_rd(src_diff, coeff, bw * 4);
+        else
+          vp9_fdct32x32(src_diff, coeff, bw * 4);
+        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, iscan);
+      }
       if (!x->skip_encode && *eob)
         vp9_idct32x32_add(dqcoeff, dst, pd->dst.stride, *eob);
       break;
@@ -587,16 +589,18 @@
       xoff = 16 * (block & twmask);
       yoff = 16 * (block >> twl);
       dst = pd->dst.buf + yoff * pd->dst.stride + xoff;
-      src = p->src.buf + yoff * p->src.stride + xoff;
-      src_diff = p->src_diff + 4 * bw * yoff + xoff;
       vp9_predict_intra_block(xd, block, bwl, TX_16X16, mode,
                               dst, pd->dst.stride, dst, pd->dst.stride);
-      vp9_subtract_block(16, 16, src_diff, bw * 4,
-                         src, p->src.stride, dst, pd->dst.stride);
-      vp9_fht16x16(tx_type, src_diff, coeff, bw * 4);
-      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, iscan);
+      if (!x->skip_recode) {
+        src = p->src.buf + yoff * p->src.stride + xoff;
+        src_diff = p->src_diff + 4 * bw * yoff + xoff;
+        vp9_subtract_block(16, 16, src_diff, bw * 4,
+                           src, p->src.stride, dst, pd->dst.stride);
+        vp9_fht16x16(tx_type, src_diff, coeff, bw * 4);
+        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, iscan);
+      }
       if (!x->skip_encode && *eob)
         vp9_iht16x16_add(tx_type, dqcoeff, dst, pd->dst.stride, *eob);
       break;
@@ -609,16 +613,18 @@
       xoff = 8 * (block & twmask);
       yoff = 8 * (block >> twl);
       dst = pd->dst.buf + yoff * pd->dst.stride + xoff;
-      src = p->src.buf + yoff * p->src.stride + xoff;
-      src_diff = p->src_diff + 4 * bw * yoff + xoff;
       vp9_predict_intra_block(xd, block, bwl, TX_8X8, mode,
                               dst, pd->dst.stride, dst, pd->dst.stride);
-      vp9_subtract_block(8, 8, src_diff, bw * 4,
-                         src, p->src.stride, dst, pd->dst.stride);
-      vp9_fht8x8(tx_type, src_diff, coeff, bw * 4);
-      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, iscan);
+      if (!x->skip_recode) {
+        src = p->src.buf + yoff * p->src.stride + xoff;
+        src_diff = p->src_diff + 4 * bw * yoff + xoff;
+        vp9_subtract_block(8, 8, src_diff, bw * 4,
+                           src, p->src.stride, dst, pd->dst.stride);
+        vp9_fht8x8(tx_type, src_diff, coeff, bw * 4);
+        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, iscan);
+      }
       if (!x->skip_encode && *eob)
         vp9_iht8x8_add(tx_type, dqcoeff, dst, pd->dst.stride, *eob);
       break;
@@ -634,19 +640,23 @@
       xoff = 4 * (block & twmask);
       yoff = 4 * (block >> twl);
       dst = pd->dst.buf + yoff * pd->dst.stride + xoff;
-      src = p->src.buf + yoff * p->src.stride + xoff;
-      src_diff = p->src_diff + 4 * bw * yoff + xoff;
       vp9_predict_intra_block(xd, block, bwl, TX_4X4, mode,
                               dst, pd->dst.stride, dst, pd->dst.stride);
-      vp9_subtract_block(4, 4, src_diff, bw * 4,
-                         src, p->src.stride, dst, pd->dst.stride);
-      if (tx_type != DCT_DCT)
-        vp9_short_fht4x4(src_diff, coeff, bw * 4, tx_type);
-      else
-        x->fwd_txm4x4(src_diff, coeff, bw * 4);
-      vp9_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round, p->quant,
-                     p->quant_shift, qcoeff, dqcoeff,
-                     pd->dequant, p->zbin_extra, eob, scan, iscan);
+
+      if (!x->skip_recode) {
+        src = p->src.buf + yoff * p->src.stride + xoff;
+        src_diff = p->src_diff + 4 * bw * yoff + xoff;
+        vp9_subtract_block(4, 4, src_diff, bw * 4,
+                           src, p->src.stride, dst, pd->dst.stride);
+        if (tx_type != DCT_DCT)
+          vp9_short_fht4x4(src_diff, coeff, bw * 4, tx_type);
+        else
+          x->fwd_txm4x4(src_diff, coeff, bw * 4);
+        vp9_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round, p->quant,
+                       p->quant_shift, qcoeff, dqcoeff,
+                       pd->dequant, p->zbin_extra, eob, scan, 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
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index b5428df..974c300 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -535,6 +535,7 @@
     pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1];
     pd[i].eobs = ctx->eobs_pbuf[i][1];
   }
+  x->skip_recode = 0;
 
 
   // Initialise the MV cost table to the defaults
diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c
index 8a98343..66919de 100644
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -1452,7 +1452,7 @@
   CHECK_MEM_ERROR(cm, ctx->zcoeff_blk,
                   vpx_calloc(num_4x4_blk, sizeof(uint8_t)));
   for (i = 0; i < MAX_MB_PLANE; ++i) {
-    for (k = 0; k < 2; ++k) {
+    for (k = 0; k < 3; ++k) {
       CHECK_MEM_ERROR(cm, ctx->coeff[i][k],
                       vpx_memalign(16, num_pix * sizeof(int16_t)));
       CHECK_MEM_ERROR(cm, ctx->qcoeff[i][k],
@@ -1474,7 +1474,7 @@
   vpx_free(ctx->zcoeff_blk);
   ctx->zcoeff_blk = 0;
   for (i = 0; i < MAX_MB_PLANE; ++i) {
-    for (k = 0; k < 2; ++k) {
+    for (k = 0; k < 3; ++k) {
       vpx_free(ctx->coeff[i][k]);
       ctx->coeff[i][k] = 0;
       vpx_free(ctx->qcoeff[i][k]);
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 695a2e2..78cb06b 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -246,7 +246,8 @@
 
   vp9_set_speed_features(cpi);
 
-  cpi->mb.select_txfm_size = cpi->sf.tx_size_search_method == USE_LARGESTALL ?
+  cpi->mb.select_txfm_size = (cpi->sf.tx_size_search_method == USE_LARGESTALL &&
+                              cm->frame_type != KEY_FRAME) ?
                              0 : 1;
 
   set_block_thresholds(cpi);
@@ -1329,6 +1330,7 @@
 }
 
 static int64_t rd_pick_intra_sbuv_mode(VP9_COMP *cpi, MACROBLOCK *x,
+                                       PICK_MODE_CONTEXT *ctx,
                                        int *rate, int *rate_tokenonly,
                                        int64_t *distortion, int *skippable,
                                        BLOCK_SIZE bsize) {
@@ -1364,6 +1366,27 @@
       *rate_tokenonly = this_rate_tokenonly;
       *distortion     = this_distortion;
       *skippable      = s;
+      if (!x->select_txfm_size) {
+        int i;
+        struct macroblock_plane *const p = x->plane;
+        struct macroblockd_plane *const pd = x->e_mbd.plane;
+        for (i = 1; i < MAX_MB_PLANE; ++i) {
+          p[i].coeff    = ctx->coeff_pbuf[i][2];
+          pd[i].qcoeff  = ctx->qcoeff_pbuf[i][2];
+          pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][2];
+          pd[i].eobs    = ctx->eobs_pbuf[i][2];
+
+          ctx->coeff_pbuf[i][2]   = ctx->coeff_pbuf[i][0];
+          ctx->qcoeff_pbuf[i][2]  = ctx->qcoeff_pbuf[i][0];
+          ctx->dqcoeff_pbuf[i][2] = ctx->dqcoeff_pbuf[i][0];
+          ctx->eobs_pbuf[i][2]    = ctx->eobs_pbuf[i][0];
+
+          ctx->coeff_pbuf[i][0]   = p[i].coeff;
+          ctx->qcoeff_pbuf[i][0]  = pd[i].qcoeff;
+          ctx->dqcoeff_pbuf[i][0] = pd[i].dqcoeff;
+          ctx->eobs_pbuf[i][0]    = pd[i].eobs;
+        }
+      }
     }
   }
 
@@ -1389,8 +1412,9 @@
   return this_rd;
 }
 
-static void choose_intra_uv_mode(VP9_COMP *cpi, BLOCK_SIZE bsize,
-                                 int *rate_uv, int *rate_uv_tokenonly,
+static void choose_intra_uv_mode(VP9_COMP *cpi, PICK_MODE_CONTEXT *ctx,
+                                 BLOCK_SIZE bsize, int *rate_uv,
+                                 int *rate_uv_tokenonly,
                                  int64_t *dist_uv, int *skip_uv,
                                  MB_PREDICTION_MODE *mode_uv) {
   MACROBLOCK *const x = &cpi->mb;
@@ -1403,7 +1427,7 @@
   // Else do a proper rd search for each possible transform size that may
   // be considered in the main rd loop.
   } else {
-    rd_pick_intra_sbuv_mode(cpi, x,
+    rd_pick_intra_sbuv_mode(cpi, x, ctx,
                             rate_uv, rate_uv_tokenonly, dist_uv, skip_uv,
                             bsize < BLOCK_8X8 ? BLOCK_8X8 : bsize);
   }
@@ -3033,12 +3057,13 @@
   return this_rd;  // if 0, this will be re-calculated by caller
 }
 
-static void swap_block_ptr(MACROBLOCK *x, PICK_MODE_CONTEXT *ctx) {
-  int i;
+static void swap_block_ptr(MACROBLOCK *x, PICK_MODE_CONTEXT *ctx,
+                           int max_plane) {
   struct macroblock_plane *const p = x->plane;
   struct macroblockd_plane *const pd = x->e_mbd.plane;
+  int i;
 
-  for (i = 0; i < MAX_MB_PLANE; ++i) {
+  for (i = 0; i < max_plane; ++i) {
     p[i].coeff    = ctx->coeff_pbuf[i][1];
     pd[i].qcoeff  = ctx->qcoeff_pbuf[i][1];
     pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1];
@@ -3075,7 +3100,7 @@
       *returnrate = INT_MAX;
       return;
     }
-    rd_pick_intra_sbuv_mode(cpi, x, &rate_uv, &rate_uv_tokenonly,
+    rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv, &rate_uv_tokenonly,
                             &dist_uv, &uv_skip, bsize);
   } else {
     y_skip = 0;
@@ -3084,7 +3109,7 @@
       *returnrate = INT_MAX;
       return;
     }
-    rd_pick_intra_sbuv_mode(cpi, x, &rate_uv, &rate_uv_tokenonly,
+    rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv, &rate_uv_tokenonly,
                             &dist_uv, &uv_skip, BLOCK_8X8);
   }
 
@@ -3450,7 +3475,7 @@
 
       uv_tx = MIN(mbmi->tx_size, max_uv_txsize_lookup[bsize]);
       if (rate_uv_intra[uv_tx] == INT_MAX) {
-        choose_intra_uv_mode(cpi, bsize, &rate_uv_intra[uv_tx],
+        choose_intra_uv_mode(cpi, ctx, bsize, &rate_uv_intra[uv_tx],
                              &rate_uv_tokenonly[uv_tx],
                              &dist_uv[uv_tx], &skip_uv[uv_tx],
                              &mode_uv[uv_tx]);
@@ -3584,6 +3609,7 @@
 
     // Did this mode help.. i.e. is it the new best mode
     if (this_rd < best_rd || x->skip) {
+      int max_plane = MAX_MB_PLANE;
       if (!mode_excluded) {
         // Note index of best mode so far
         best_mode_index = mode_index;
@@ -3591,6 +3617,7 @@
         if (ref_frame == INTRA_FRAME) {
           /* required for left and above block mv */
           mbmi->mv[0].as_int = 0;
+          max_plane = 1;
         }
 
         *returnrate = rate2;
@@ -3599,7 +3626,7 @@
         best_mbmode = *mbmi;
         best_skip2 = this_skip2;
         if (!x->select_txfm_size)
-          swap_block_ptr(x, ctx);
+          swap_block_ptr(x, ctx, max_plane);
         vpx_memcpy(ctx->zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
                    sizeof(uint8_t) * ctx->num_4x4_blk);
 
@@ -3706,7 +3733,7 @@
     // Do Intra UV best rd mode selection if best mode choice above was intra.
     if (vp9_mode_order[best_mode_index].ref_frame == INTRA_FRAME) {
       TX_SIZE uv_tx_size = get_uv_tx_size(mbmi);
-      rd_pick_intra_sbuv_mode(cpi, x, &rate_uv_intra[uv_tx_size],
+      rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv_intra[uv_tx_size],
                               &rate_uv_tokenonly[uv_tx_size],
                               &dist_uv[uv_tx_size],
                               &skip_uv[uv_tx_size],
@@ -4075,7 +4102,7 @@
       distortion2 += distortion_y;
 
       if (rate_uv_intra[TX_4X4] == INT_MAX) {
-        choose_intra_uv_mode(cpi, bsize, &rate_uv_intra[TX_4X4],
+        choose_intra_uv_mode(cpi, ctx, bsize, &rate_uv_intra[TX_4X4],
                              &rate_uv_tokenonly[TX_4X4],
                              &dist_uv[TX_4X4], &skip_uv[TX_4X4],
                              &mode_uv[TX_4X4]);
@@ -4329,12 +4356,14 @@
     // Did this mode help.. i.e. is it the new best mode
     if (this_rd < best_rd || x->skip) {
       if (!mode_excluded) {
+        int max_plane = MAX_MB_PLANE;
         // Note index of best mode so far
         best_mode_index = mode_index;
 
         if (ref_frame == INTRA_FRAME) {
           /* required for left and above block mv */
           mbmi->mv[0].as_int = 0;
+          max_plane = 1;
         }
 
         *returnrate = rate2;
@@ -4345,7 +4374,7 @@
         best_mbmode = *mbmi;
         best_skip2 = this_skip2;
         if (!x->select_txfm_size)
-          swap_block_ptr(x, ctx);
+          swap_block_ptr(x, ctx, max_plane);
         vpx_memcpy(ctx->zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
                    sizeof(uint8_t) * ctx->num_4x4_blk);
 
@@ -4452,7 +4481,7 @@
     // Do Intra UV best rd mode selection if best mode choice above was intra.
     if (vp9_ref_order[best_mode_index].ref_frame == INTRA_FRAME) {
       TX_SIZE uv_tx_size = get_uv_tx_size(mbmi);
-      rd_pick_intra_sbuv_mode(cpi, x, &rate_uv_intra[uv_tx_size],
+      rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv_intra[uv_tx_size],
                               &rate_uv_tokenonly[uv_tx_size],
                               &dist_uv[uv_tx_size],
                               &skip_uv[uv_tx_size],
diff --git a/vp9/encoder/x86/vp9_dct_sse2.c b/vp9/encoder/x86/vp9_dct_sse2.c
index dc11501..fefca66 100644
--- a/vp9/encoder/x86/vp9_dct_sse2.c
+++ b/vp9/encoder/x86/vp9_dct_sse2.c
@@ -206,12 +206,12 @@
   const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
   __m128i u[8], v[8];
   __m128i in7 = _mm_add_epi16(in[0], in[1]);
-  in7 = _mm_sub_epi16(in7, in[3]);
 
   u[0] = _mm_unpacklo_epi16(in[0], in[1]);
   u[1] = _mm_unpacklo_epi16(in[2], in[3]);
   u[2] = _mm_unpacklo_epi16(in7, kZero);
   u[3] = _mm_unpacklo_epi16(in[2], kZero);
+  u[4] = _mm_unpacklo_epi16(in[3], kZero);
 
   v[0] = _mm_madd_epi16(u[0], k__sinpi_p01_p02);  // s0 + s2
   v[1] = _mm_madd_epi16(u[1], k__sinpi_p03_p04);  // s4 + s5
@@ -219,9 +219,10 @@
   v[3] = _mm_madd_epi16(u[0], k__sinpi_p04_m01);  // s1 - s3
   v[4] = _mm_madd_epi16(u[1], k__sinpi_m03_p02);  // -s4 + s6
   v[5] = _mm_madd_epi16(u[3], k__sinpi_p03_p03);  // s4
+  v[6] = _mm_madd_epi16(u[4], k__sinpi_p03_p03);
 
   u[0] = _mm_add_epi32(v[0], v[1]);
-  u[1] = v[2];
+  u[1] = _mm_sub_epi32(v[2], v[6]);
   u[2] = _mm_add_epi32(v[3], v[4]);
   u[3] = _mm_sub_epi32(u[2], u[0]);
   u[4] = _mm_slli_epi32(v[5], 2);