Merge "Move src_diff to per-plane MACROBLOCK data" into experimental

8 files changed, 146 insertions, 194 deletions

diff --git a/vp9/encoder/vp9_asm_enc_offsets.c b/vp9/encoder/vp9_asm_enc_offsets.c
index 1a770dcf7..db9a4fc43 100644
--- a/vp9/encoder/vp9_asm_enc_offsets.c
+++ b/vp9/encoder/vp9_asm_enc_offsets.c
@@ -20,7 +20,6 @@
 BEGIN
 
 /* regular quantize */
-DEFINE(vp9_block_coeff,                         offsetof(BLOCK, coeff));
 DEFINE(vp9_block_zbin,                          offsetof(BLOCK, zbin));
 DEFINE(vp9_block_round,                         offsetof(BLOCK, round));
 DEFINE(vp9_block_quant,                         offsetof(BLOCK, quant));
diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index 1aa84338f..709c29ede 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -25,7 +25,6 @@ typedef struct {
 
 typedef struct block {
   // 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries
-  int16_t *src_diff;
   int16_t *coeff;
 
   // 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries
@@ -89,9 +88,13 @@ typedef struct {
   unsigned int frames_with_high_error;
 } PICK_MODE_CONTEXT;
 
+struct macroblock_plane {
+  DECLARE_ALIGNED(16, int16_t, src_diff[64*64]);
+};
+
 typedef struct macroblock MACROBLOCK;
 struct macroblock {
-  DECLARE_ALIGNED(16, int16_t, src_diff[64*64+32*32*2]);
+  struct macroblock_plane plane[MAX_MB_PLANE];
   DECLARE_ALIGNED(16, int16_t, coeff[64*64+32*32*2]);
   // 16 Y blocks, 4 U blocks, 4 V blocks,
   BLOCK block[24];
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 1978f19f6..646d2c4c6 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -1734,25 +1734,8 @@ void vp9_encode_frame(VP9_COMP *cpi) {
 }
 
 void vp9_setup_block_ptrs(MACROBLOCK *x) {
-  int r, c;
   int i;
 
-  for (r = 0; r < 4; r++) {
-    for (c = 0; c < 4; c++)
-      x->block[r * 4 + c].src_diff = x->src_diff + r * 4 * 16 + c * 4;
-  }
-
-  for (r = 0; r < 2; r++) {
-    for (c = 0; c < 2; c++)
-      x->block[16 + r * 2 + c].src_diff = x->src_diff + 256 + r * 4 * 8 + c * 4;
-  }
-
-
-  for (r = 0; r < 2; r++) {
-    for (c = 0; c < 2; c++)
-      x->block[20 + r * 2 + c].src_diff = x->src_diff + 320 + r * 4 * 8 + c * 4;
-  }
-
   for (i = 0; i < 24; i++)
     x->block[i].coeff = x->coeff + i * 16;
 }
@@ -2100,14 +2083,6 @@ static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t,
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCK *const x = &cpi->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
-  const uint8_t *src = x->src.y_buffer;
-  uint8_t *dst = xd->plane[0].dst.buf;
-  const uint8_t *usrc = x->src.u_buffer;
-  uint8_t *udst = xd->plane[1].dst.buf;
-  const uint8_t *vsrc = x->src.v_buffer;
-  uint8_t *vdst = xd->plane[2].dst.buf;
-  int src_y_stride = x->src.y_stride, dst_y_stride = xd->plane[0].dst.stride;
-  int src_uv_stride = x->src.uv_stride, dst_uv_stride = xd->plane[1].dst.stride;
   int n;
   MODE_INFO *mi = x->e_mbd.mode_info_context;
   unsigned int segment_id = mi->mbmi.segment_id;
@@ -2187,10 +2162,7 @@ static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t,
   }
 
   if (!x->skip) {
-    vp9_subtract_sby_s_c(x->src_diff, src, src_y_stride, dst, dst_y_stride,
-                         bsize);
-    vp9_subtract_sbuv_s_c(x->src_diff, usrc, vsrc, src_uv_stride,
-                          udst, vdst, dst_uv_stride, bsize);
+    vp9_subtract_sb(x, bsize);
 
     switch (xd->mode_info_context->mbmi.txfm_size) {
       case TX_32X32:
diff --git a/vp9/encoder/vp9_encodeintra.c b/vp9/encoder/vp9_encodeintra.c
index 5c6559bdb..32a92b8e7 100644
--- a/vp9/encoder/vp9_encodeintra.c
+++ b/vp9/encoder/vp9_encodeintra.c
@@ -37,7 +37,7 @@ int vp9_encode_intra(VP9_COMP *cpi, MACROBLOCK *x, int use_16x16_pred) {
     }
   }
 
-  return vp9_get_mb_ss(x->src_diff);
+  return vp9_get_mb_ss(x->plane[0].src_diff);
 }
 
 static void encode_intra4x4block(MACROBLOCK *x, int ib) {
@@ -45,6 +45,9 @@ static void encode_intra4x4block(MACROBLOCK *x, int ib) {
   BLOCK *be = &x->block[ib];
   MACROBLOCKD * const xd = &x->e_mbd;
   TX_TYPE tx_type;
+  int16_t* const src_diff =
+      raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+                                x->plane[0].src_diff);
 
   assert(ib < 16);
 
@@ -54,16 +57,18 @@ static void encode_intra4x4block(MACROBLOCK *x, int ib) {
 
   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);
+  vp9_subtract_block(4, 4, src_diff, 16,
+                     *(be->base_src) + be->src, be->src_stride,
+                     *(b->base_dst) + b->dst, b->dst_stride);
 
   tx_type = get_tx_type_4x4(&x->e_mbd, ib);
   if (tx_type != DCT_DCT) {
-    vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
+    vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
     vp9_ht_quantize_b_4x4(x, ib, tx_type);
     vp9_short_iht4x4(BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16),
                      b->diff, 16, tx_type);
   } else {
-    x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+    x->fwd_txm4x4(src_diff, be->coeff, 32);
     x->quantize_b_4x4(x, ib, 16);
     vp9_inverse_transform_b_4x4(&x->e_mbd, xd->plane[0].eobs[ib],
                                 BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16),
@@ -86,10 +91,7 @@ void vp9_encode_intra16x16mby(VP9_COMMON *const cm, MACROBLOCK *x) {
   TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size;
 
   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->plane[0].dst.buf, xd->plane[0].dst.stride,
-                       BLOCK_SIZE_MB16X16);
+  vp9_subtract_sby(x, BLOCK_SIZE_MB16X16);
 
   switch (tx_size) {
     case TX_16X16:
@@ -123,11 +125,7 @@ void vp9_encode_intra16x16mbuv(VP9_COMMON *const cm, MACROBLOCK *x) {
   TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size;
 
   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->plane[1].dst.buf, xd->plane[2].dst.buf,
-                        xd->plane[1].dst.stride,
-                        BLOCK_SIZE_MB16X16);
+  vp9_subtract_sbuv(x, BLOCK_SIZE_MB16X16);
 
   switch (tx_size) {
     case TX_4X4:
@@ -153,6 +151,9 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
   MACROBLOCKD *xd = &x->e_mbd;
   BLOCKD *b = &xd->block[ib];
   BLOCK *be = &x->block[ib];
+  int16_t* const src_diff =
+      raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+                                x->plane[0].src_diff);
   const int iblock[4] = {0, 1, 4, 5};
   int i;
   TX_TYPE tx_type;
@@ -160,7 +161,9 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
   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);
+  vp9_subtract_block(8, 8, src_diff, 16,
+                     *(be->base_src) + be->src, be->src_stride,
+                     *(b->base_dst) + b->dst, b->dst_stride);
 
   if (xd->mode_info_context->mbmi.txfm_size == TX_8X8) {
     int idx = (ib & 0x02) ? (ib + 2) : ib;
@@ -169,12 +172,12 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
     assert(idx < 16);
     tx_type = get_tx_type_8x8(xd, ib);
     if (tx_type != DCT_DCT) {
-      vp9_short_fht8x8(be->src_diff, (x->block + idx)->coeff, 16, tx_type);
+      vp9_short_fht8x8(src_diff, (x->block + idx)->coeff, 16, tx_type);
       x->quantize_b_8x8(x, idx, tx_type, 16);
       vp9_short_iht8x8(dqcoeff, xd->block[ib].diff,
                             16, tx_type);
     } else {
-      x->fwd_txm8x8(be->src_diff, (x->block + idx)->coeff, 32);
+      x->fwd_txm8x8(src_diff, (x->block + idx)->coeff, 32);
       x->quantize_b_8x8(x, idx, DCT_DCT, 16);
       vp9_short_idct8x8(dqcoeff, xd->block[ib].diff, 32);
     }
@@ -182,18 +185,21 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
     for (i = 0; i < 4; i++) {
       int idx = ib + iblock[i];
       int16_t * const dqcoeff = BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16);
+      int16_t* const src_diff =
+          raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, idx,
+                                    x->plane[0].src_diff);
 
       assert(idx < 16);
       b = &xd->block[ib + iblock[i]];
       be = &x->block[ib + iblock[i]];
       tx_type = get_tx_type_4x4(xd, ib + iblock[i]);
       if (tx_type != DCT_DCT) {
-        vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
+        vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
         vp9_ht_quantize_b_4x4(x, ib + iblock[i], tx_type);
         vp9_short_iht4x4(dqcoeff, b->diff, 16, tx_type);
       } else if (!(i & 1) &&
                  get_tx_type_4x4(xd, ib + iblock[i] + 1) == DCT_DCT) {
-        x->fwd_txm8x4(be->src_diff, be->coeff, 32);
+        x->fwd_txm8x4(src_diff, be->coeff, 32);
         x->quantize_b_4x4_pair(x, ib + iblock[i], ib + iblock[i] + 1, 16);
         vp9_inverse_transform_b_4x4(xd, xd->plane[0].eobs[ib + iblock[i]],
                                     dqcoeff, b->diff, 32);
@@ -201,7 +207,7 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
                                     dqcoeff + 16, (b + 1)->diff, 32);
         i++;
       } else {
-        x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+        x->fwd_txm4x4(src_diff, be->coeff, 32);
         x->quantize_b_4x4(x, ib + iblock[i], 16);
         vp9_inverse_transform_b_4x4(xd, xd->plane[0].eobs[ib + iblock[i]],
                                     dqcoeff, b->diff, 32);
@@ -231,14 +237,20 @@ static void encode_intra_uv4x4(MACROBLOCK *x, int ib, int mode) {
   int16_t * const dqcoeff = MB_SUBBLOCK_FIELD(xd, dqcoeff, ib);
   const int plane = ib < 20 ? 1 : 2;
   const int block = ib < 20 ? ib - 16 : ib - 20;
+  int16_t* const src_diff =
+      raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, plane, block,
+                                x->plane[plane].src_diff);
 
   assert(ib >= 16 && ib < 24);
   vp9_intra_uv4x4_predict(&x->e_mbd, b, mode,
                           *(b->base_dst) + b->dst, b->dst_stride);
 
-  vp9_subtract_b(be, b, 8);
+  assert(xd->plane[1].subsampling_x == 1);
+  vp9_subtract_block(4, 4, src_diff, 8,
+                     *(be->base_src) + be->src, be->src_stride,
+                     *(b->base_dst) + b->dst, b->dst_stride);
 
-  x->fwd_txm4x4(be->src_diff, be->coeff, 16);
+  x->fwd_txm4x4(src_diff, be->coeff, 16);
   x->quantize_b_4x4(x, ib, 16);
   vp9_inverse_transform_b_4x4(&x->e_mbd, xd->plane[plane].eobs[block],
                               dqcoeff, b->diff, 16);
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index 8121f8639..e1e0df641 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -20,102 +20,54 @@
 #include "vp9/common/vp9_systemdependent.h"
 #include "vp9_rtcd.h"
 
-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->base_dst) + bd->dst;
-  int src_stride = be->src_stride;
-  int dst_stride = bd->dst_stride;
-
+void vp9_subtract_block(int rows, int cols,
+                        int16_t *diff_ptr, int diff_stride,
+                        const uint8_t *src_ptr, int src_stride,
+                        const uint8_t *pred_ptr, int pred_stride) {
   int r, c;
 
-  for (r = 0; r < 4; r++) {
-    for (c = 0; c < 4; c++)
+  for (r = 0; r < rows; r++) {
+    for (c = 0; c < cols; c++)
       diff_ptr[c] = src_ptr[c] - pred_ptr[c];
 
-    diff_ptr += pitch;
-    pred_ptr += dst_stride;
+    diff_ptr += diff_stride;
+    pred_ptr += pred_stride;
     src_ptr  += src_stride;
   }
 }
 
-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->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++) {
-    for (c = 0; c < 8; c++)
-      diff_ptr[c] = src_ptr[c] - pred_ptr[c];
+static void subtract_plane(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize, int plane) {
+  const MACROBLOCKD * const xd = &x->e_mbd;
+  const int bw = 4 << (b_width_log2(bsize) - xd->plane[plane].subsampling_x);
+  const int bh = 4 << (b_height_log2(bsize) - xd->plane[plane].subsampling_y);
+  const uint8_t *src = plane == 0 ? x->src.y_buffer :
+                       plane == 1 ? x->src.u_buffer : x->src.v_buffer;
+  const int src_stride = plane == 0 ? x->src.y_stride : x->src.uv_stride;
 
-    diff_ptr += pitch;
-    pred_ptr += dst_stride;
-    src_ptr  += src_stride;
-  }
+  assert(plane < 3);
+  vp9_subtract_block(bh, bw,
+                     x->plane[plane].src_diff, bw, src, src_stride,
+                     xd->plane[plane].dst.buf, xd->plane[plane].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) {
-  const int bh = 16 << mb_height_log2(bsize), bw = 16 << mb_width_log2(bsize);
-  int r, c;
-
-  for (r = 0; r < bh; r++) {
-    for (c = 0; c < bw; c++)
-      diff[c] = src[c] - pred[c];
-
-    diff += bw;
-    pred += dst_stride;
-    src  += src_stride;
-  }
+void vp9_subtract_sby(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  subtract_plane(x, bsize, 0);
 }
 
-void vp9_subtract_sbuv_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,
-                           BLOCK_SIZE_TYPE bsize) {
-  const int bhl = mb_height_log2(bsize), bwl = mb_width_log2(bsize);
-  const int uoff = (16 * 16) << (bhl + bwl), voff = (uoff * 5) >> 2;
-  const int bw = 8 << bwl, bh = 8 << bhl;
-  int16_t *udiff = diff + uoff;
-  int16_t *vdiff = diff + voff;
-  int r, c;
-
-  for (r = 0; r < bh; r++) {
-    for (c = 0; c < bw; c++)
-      udiff[c] = usrc[c] - upred[c];
-
-    udiff += bw;
-    upred += dst_stride;
-    usrc  += src_stride;
-  }
+void vp9_subtract_sbuv(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  int i;
 
-  for (r = 0; r < bh; r++) {
-    for (c = 0; c < bw; c++)
-      vdiff[c] = vsrc[c] - vpred[c];
-
-    vdiff += bw;
-    vpred += dst_stride;
-    vsrc  += src_stride;
-  }
+  for (i = 1; i < MAX_MB_PLANE; i++)
+    subtract_plane(x, bsize, i);
 }
 
-static void subtract_mb(MACROBLOCK *x) {
-  MACROBLOCKD *xd = &x->e_mbd;
-  vp9_subtract_sby_s_c(x->src_diff, x->src.y_buffer, x->src.y_stride,
-                       xd->plane[0].dst.buf, xd->plane[0].dst.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->plane[1].dst.buf, xd->plane[2].dst.buf,
-                        xd->plane[1].dst.stride,
-                        BLOCK_SIZE_MB16X16);
+void vp9_subtract_sb(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
+  vp9_subtract_sby(x, bsize);
+  vp9_subtract_sbuv(x, bsize);
 }
 
+
 void vp9_transform_sby_32x32(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
   const int bwl = mb_width_log2(bsize) - 1, bw = 1 << bwl;
   const int bh = 1 << (mb_height_log2(bsize) - 1);
@@ -125,7 +77,7 @@ void vp9_transform_sby_32x32(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
   for (n = 0; n < bw * bh; n++) {
     const int x_idx = n & (bw - 1), y_idx = n >> bwl;
 
-    vp9_short_fdct32x32(x->src_diff + y_idx * stride * 32 + x_idx * 32,
+    vp9_short_fdct32x32(x->plane[0].src_diff + y_idx * stride * 32 + x_idx * 32,
                         x->coeff + n * 1024, stride * 2);
   }
 }
@@ -143,10 +95,11 @@ void vp9_transform_sby_16x16(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
                                               (y_idx * bstride + x_idx) * 4);
 
     if (tx_type != DCT_DCT) {
-      vp9_short_fht16x16(x->src_diff + y_idx * stride * 16 + x_idx * 16,
+      vp9_short_fht16x16(x->plane[0].src_diff +
+                             y_idx * stride * 16 + x_idx * 16,
                          x->coeff + n * 256, stride, tx_type);
     } else {
-      x->fwd_txm16x16(x->src_diff + y_idx * stride * 16 + x_idx * 16,
+      x->fwd_txm16x16(x->plane[0].src_diff + y_idx * stride * 16 + x_idx * 16,
                       x->coeff + n * 256, stride * 2);
     }
   }
@@ -164,10 +117,10 @@ void vp9_transform_sby_8x8(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
     const TX_TYPE tx_type = get_tx_type_8x8(xd, (y_idx * bstride + x_idx) * 2);
 
     if (tx_type != DCT_DCT) {
-      vp9_short_fht8x8(x->src_diff + y_idx * stride * 8 + x_idx * 8,
+      vp9_short_fht8x8(x->plane[0].src_diff + y_idx * stride * 8 + x_idx * 8,
                        x->coeff + n * 64, stride, tx_type);
     } else {
-      x->fwd_txm8x8(x->src_diff + y_idx * stride * 8 + x_idx * 8,
+      x->fwd_txm8x8(x->plane[0].src_diff + y_idx * stride * 8 + x_idx * 8,
                     x->coeff + n * 64, stride * 2);
     }
   }
@@ -185,10 +138,10 @@ void vp9_transform_sby_4x4(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
     const TX_TYPE tx_type = get_tx_type_4x4(xd, n);
 
     if (tx_type != DCT_DCT) {
-      vp9_short_fht4x4(x->src_diff + y_idx * stride * 4 + x_idx * 4,
+      vp9_short_fht4x4(x->plane[0].src_diff + y_idx * stride * 4 + x_idx * 4,
                        x->coeff + n * 16, stride, tx_type);
     } else {
-      x->fwd_txm4x4(x->src_diff + y_idx * stride * 4 + x_idx * 4,
+      x->fwd_txm4x4(x->plane[0].src_diff + y_idx * stride * 4 + x_idx * 4,
                     x->coeff + n * 16, stride * 2);
     }
   }
@@ -197,9 +150,9 @@ void vp9_transform_sby_4x4(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
 void vp9_transform_sbuv_32x32(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
   assert(bsize == BLOCK_SIZE_SB64X64);
   vp9_clear_system_state();
-  vp9_short_fdct32x32(x->src_diff + 4096,
+  vp9_short_fdct32x32(x->plane[1].src_diff,
                       x->coeff + 4096, 64);
-  vp9_short_fdct32x32(x->src_diff + 4096 + 1024,
+  vp9_short_fdct32x32(x->plane[2].src_diff,
                       x->coeff + 4096 + 1024, 64);
 }
 
@@ -214,9 +167,9 @@ void vp9_transform_sbuv_16x16(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
   for (n = 0; n < bw * bh; n++) {
     const int x_idx = n & (bw - 1), y_idx = n >> (bwl - 1);
 
-    x->fwd_txm16x16(x->src_diff + uoff + y_idx * stride * 16 + x_idx * 16,
+    x->fwd_txm16x16(x->plane[1].src_diff + y_idx * stride * 16 + x_idx * 16,
                     x->coeff + uoff + n * 256, stride * 2);
-    x->fwd_txm16x16(x->src_diff + voff + y_idx * stride * 16 + x_idx * 16,
+    x->fwd_txm16x16(x->plane[2].src_diff + y_idx * stride * 16 + x_idx * 16,
                     x->coeff + voff + n * 256, stride * 2);
   }
 }
@@ -232,9 +185,9 @@ void vp9_transform_sbuv_8x8(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
   for (n = 0; n < bw * bh; n++) {
     const int x_idx = n & (bw - 1), y_idx = n >> (bwl - 1);
 
-    x->fwd_txm8x8(x->src_diff + uoff + y_idx * stride * 8 + x_idx * 8,
+    x->fwd_txm8x8(x->plane[1].src_diff + y_idx * stride * 8 + x_idx * 8,
                   x->coeff + uoff + n * 64, stride * 2);
-    x->fwd_txm8x8(x->src_diff + voff + y_idx * stride * 8 + x_idx * 8,
+    x->fwd_txm8x8(x->plane[2].src_diff + y_idx * stride * 8 + x_idx * 8,
                   x->coeff + voff + n * 64, stride * 2);
   }
 }
@@ -250,9 +203,9 @@ void vp9_transform_sbuv_4x4(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize) {
   for (n = 0; n < bw * bh; n++) {
     const int x_idx = n & (bw - 1), y_idx = n >> (bwl - 1);
 
-    x->fwd_txm4x4(x->src_diff + uoff + y_idx * stride * 4 + x_idx * 4,
+    x->fwd_txm4x4(x->plane[1].src_diff + y_idx * stride * 4 + x_idx * 4,
                   x->coeff + uoff + n * 16, stride * 2);
-    x->fwd_txm4x4(x->src_diff + voff + y_idx * stride * 4 + x_idx * 4,
+    x->fwd_txm4x4(x->plane[2].src_diff + y_idx * stride * 4 + x_idx * 4,
                   x->coeff + voff + n * 16, stride * 2);
   }
 }
@@ -826,7 +779,7 @@ void vp9_encode_inter16x16(VP9_COMMON *const cm, MACROBLOCK *x,
   MACROBLOCKD *const xd = &x->e_mbd;
 
   vp9_build_inter_predictors_sb(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
-  subtract_mb(x);
+  vp9_subtract_sb(x, BLOCK_SIZE_MB16X16);
   vp9_fidct_mb(cm, x);
   vp9_recon_sb(xd, BLOCK_SIZE_MB16X16);
 }
@@ -836,9 +789,7 @@ void vp9_encode_inter16x16y(MACROBLOCK *x, int mb_row, int mb_col) {
   MACROBLOCKD *xd = &x->e_mbd;
 
   vp9_build_inter_predictors_sby(xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
-  vp9_subtract_sby_s_c(x->src_diff, x->src.y_buffer, x->src.y_stride,
-                       xd->plane[0].dst.buf, xd->plane[0].dst.stride,
-                       BLOCK_SIZE_MB16X16);
+  vp9_subtract_sby(x, 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 3c0d760a1..da134a86b 100644
--- a/vp9/encoder/vp9_encodemb.h
+++ b/vp9/encoder/vp9_encodemb.h
@@ -56,15 +56,12 @@ void vp9_optimize_sbuv_4x4(VP9_COMMON *const cm, MACROBLOCK *x,
 
 void vp9_fidct_mb(VP9_COMMON *const cm, MACROBLOCK *x);
 
-void vp9_subtract_4b_c(BLOCK *be, BLOCKD *bd, int pitch);
-
-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);
-void vp9_subtract_sbuv_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,
-                           BLOCK_SIZE_TYPE bsize);
+void vp9_subtract_block(int rows, int cols,
+                        int16_t *diff_ptr, int diff_stride,
+                        const uint8_t *src_ptr, int src_stride,
+                        const uint8_t *pred_ptr, int pred_stride);
+void vp9_subtract_sby(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_subtract_sbuv(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize);
+void vp9_subtract_sb(MACROBLOCK *xd, BLOCK_SIZE_TYPE bsize);
 
 #endif  // VP9_ENCODER_VP9_ENCODEMB_H_
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index cbbdc6750..69d2ac7ba 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -827,12 +827,9 @@ static void super_block_yrd(VP9_COMP *cpi,
                             int *skip, BLOCK_SIZE_TYPE bs,
                             int64_t txfm_cache[NB_TXFM_MODES]) {
   VP9_COMMON *const cm = &cpi->common;
-  MACROBLOCKD *const xd = &x->e_mbd;
   int r[TX_SIZE_MAX_SB][2], d[TX_SIZE_MAX_SB], s[TX_SIZE_MAX_SB];
-  uint8_t *src = x->src.y_buffer, *dst = xd->plane[0].dst.buf;
-  int src_y_stride = x->src.y_stride, dst_y_stride = xd->plane[0].dst.stride;
 
-  vp9_subtract_sby_s_c(x->src_diff, src, src_y_stride, dst, dst_y_stride, bs);
+  vp9_subtract_sby(x, bs);
 
   if (bs >= BLOCK_SIZE_SB32X32)
     super_block_yrd_32x32(cm, x, &r[TX_32X32][0], &d[TX_32X32], &s[TX_32X32],
@@ -859,7 +856,9 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
   VP9_COMMON *const cm = &cpi->common;
   BLOCK *be = x->block + ib;
   BLOCKD *b = xd->block + ib;
-
+  int16_t* const src_diff =
+      raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+                                x->plane[0].src_diff);
   ENTROPY_CONTEXT ta = *a, tempa = *a;
   ENTROPY_CONTEXT tl = *l, templ = *l;
   TX_TYPE tx_type = DCT_DCT;
@@ -899,15 +898,17 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
 #endif
 
     vp9_intra4x4_predict(xd, b, mode, *(b->base_dst) + b->dst, b->dst_stride);
-    vp9_subtract_b(be, b, 16);
+    vp9_subtract_block(4, 4, src_diff, 16,
+                       *(be->base_src) + be->src, be->src_stride,
+                       *(b->base_dst) + b->dst, b->dst_stride);
 
     b->bmi.as_mode.first = mode;
     tx_type = get_tx_type_4x4(xd, be - x->block);
     if (tx_type != DCT_DCT) {
-      vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
+      vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
       vp9_ht_quantize_b_4x4(x, be - x->block, tx_type);
     } else {
-      x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+      x->fwd_txm4x4(src_diff, be->coeff, 32);
       x->quantize_b_4x4(x, be - x->block, 16);
     }
 
@@ -1089,10 +1090,12 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
   ENTROPY_CONTEXT_PLANES ta, tl;
   ENTROPY_CONTEXT *ta0, *ta1, besta0 = 0, besta1 = 0;
   ENTROPY_CONTEXT *tl0, *tl1, bestl0 = 0, bestl1 = 0;
-
   // perform transformation of dimension 8x8
   // note the input and output index mapping
   int idx = (ib & 0x02) ? (ib + 2) : ib;
+  int16_t* const src_diff =
+      raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+                                x->plane[0].src_diff);
 
   assert(ib < 16);
   for (mode = DC_PRED; mode <= TM_PRED; mode++) {
@@ -1105,14 +1108,16 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
 
     vp9_intra8x8_predict(xd, b, mode, *(b->base_dst) + b->dst, b->dst_stride);
 
-    vp9_subtract_4b_c(be, b, 16);
+    vp9_subtract_block(8, 8, src_diff, 16,
+                       *(be->base_src) + be->src, be->src_stride,
+                       *(b->base_dst) + b->dst, b->dst_stride);
 
     if (xd->mode_info_context->mbmi.txfm_size == TX_8X8) {
       TX_TYPE tx_type = get_tx_type_8x8(xd, ib);
       if (tx_type != DCT_DCT)
-        vp9_short_fht8x8(be->src_diff, (x->block + idx)->coeff, 16, tx_type);
+        vp9_short_fht8x8(src_diff, (x->block + idx)->coeff, 16, tx_type);
       else
-        x->fwd_txm8x8(be->src_diff, (x->block + idx)->coeff, 32);
+        x->fwd_txm8x8(src_diff, (x->block + idx)->coeff, 32);
       x->quantize_b_8x8(x, idx, tx_type, 16);
 
       // compute quantization mse of 8x8 block
@@ -1144,20 +1149,24 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
       distortion = 0;
       rate_t = 0;
       for (i = 0; i < 4; ++i) {
+        int16_t* const src_diff =
+            raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16,
+                                      0, ib + iblock[i],
+                                      x->plane[0].src_diff);
         int do_two = 0;
         b = &xd->block[ib + iblock[i]];
         be = &x->block[ib + iblock[i]];
         tx_type = get_tx_type_4x4(xd, ib + iblock[i]);
         if (tx_type != DCT_DCT) {
-          vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
+          vp9_short_fht4x4(src_diff, be->coeff, 16, tx_type);
           vp9_ht_quantize_b_4x4(x, ib + iblock[i], tx_type);
         } else if (!(i & 1) &&
                    get_tx_type_4x4(xd, ib + iblock[i] + 1) == DCT_DCT) {
-          x->fwd_txm8x4(be->src_diff, be->coeff, 32);
+          x->fwd_txm8x4(src_diff, be->coeff, 32);
           x->quantize_b_4x4_pair(x, ib + iblock[i], ib + iblock[i] + 1, 16);
           do_two = 1;
         } else {
-          x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+          x->fwd_txm4x4(src_diff, be->coeff, 32);
           x->quantize_b_4x4(x, ib + iblock[i], 16);
         }
         distortion += vp9_block_error_c(be->coeff,
@@ -1513,12 +1522,8 @@ static void super_block_uvrd(VP9_COMMON *const cm, MACROBLOCK *x,
                              BLOCK_SIZE_TYPE bsize) {
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = &xd->mode_info_context->mbmi;
-  uint8_t *usrc = x->src.u_buffer, *udst = xd->plane[1].dst.buf;
-  uint8_t *vsrc = x->src.v_buffer, *vdst = xd->plane[2].dst.buf;
-  int src_uv_stride = x->src.uv_stride, dst_uv_stride = xd->plane[1].dst.stride;
 
-  vp9_subtract_sbuv_s_c(x->src_diff, usrc, vsrc, src_uv_stride,
-                        udst, vdst, dst_uv_stride, bsize);
+  vp9_subtract_sbuv(x, bsize);
 
   if (mbmi->txfm_size >= TX_32X32 && bsize >= BLOCK_SIZE_SB64X64) {
     super_block_uvrd_32x32(cm, x, rate, distortion, skippable, bsize);
@@ -1720,6 +1725,9 @@ static int64_t encode_inter_mb_segment(VP9_COMMON *const cm,
     if (labels[i] == which_label) {
       BLOCKD *bd = &x->e_mbd.block[i];
       BLOCK *be = &x->block[i];
+      int16_t* const src_diff =
+          raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, i,
+                                    x->plane[0].src_diff);
       int thisdistortion;
 
       vp9_build_inter_predictor(*(bd->base_pre) + bd->pre,
@@ -1741,8 +1749,10 @@ static int64_t encode_inter_mb_segment(VP9_COMMON *const cm,
             &xd->subpix);
       }
 
-      vp9_subtract_b(be, bd, 16);
-      x->fwd_txm4x4(be->src_diff, be->coeff, 32);
+      vp9_subtract_block(4, 4, src_diff, 16,
+                         *(be->base_src) + be->src, be->src_stride,
+                         *(bd->base_dst) + bd->dst, bd->dst_stride);
+      x->fwd_txm4x4(src_diff, be->coeff, 32);
       x->quantize_b_4x4(x, i, 16);
       thisdistortion = vp9_block_error(be->coeff,
           BLOCK_OFFSET(xd->plane[0].dqcoeff, i, 16), 16);
@@ -1790,6 +1800,9 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
       const int idx = (ib & 8) + ((ib & 2) << 1);
       BLOCKD *bd = &xd->block[ib];
       BLOCK *be = &x->block[ib], *be2 = &x->block[idx];
+      int16_t* const src_diff =
+          raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16, 0, ib,
+                                    x->plane[0].src_diff);
       int thisdistortion;
 
       assert(idx < 16);
@@ -1806,11 +1819,13 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
             which_mv, &xd->subpix);
       }
 
-      vp9_subtract_4b_c(be, bd, 16);
+      vp9_subtract_block(8, 8, src_diff, 16,
+                         *(be->base_src) + be->src, be->src_stride,
+                         *(bd->base_dst) + bd->dst, bd->dst_stride);
 
       if (xd->mode_info_context->mbmi.txfm_size == TX_4X4) {
         if (otherrd) {
-          x->fwd_txm8x8(be->src_diff, be2->coeff, 32);
+          x->fwd_txm8x8(src_diff, be2->coeff, 32);
           x->quantize_b_8x8(x, idx, DCT_DCT, 16);
           thisdistortion = vp9_block_error_c(be2->coeff,
               BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64);
@@ -1823,9 +1838,13 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
           xd->mode_info_context->mbmi.txfm_size = TX_4X4;
         }
         for (j = 0; j < 4; j += 2) {
+          int16_t* const src_diff =
+              raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16,
+                                        0, ib + iblock[j],
+                                        x->plane[0].src_diff);
           bd = &xd->block[ib + iblock[j]];
           be = &x->block[ib + iblock[j]];
-          x->fwd_txm8x4(be->src_diff, be->coeff, 32);
+          x->fwd_txm8x4(src_diff, be->coeff, 32);
           x->quantize_b_4x4_pair(x, ib + iblock[j], ib + iblock[j] + 1, 16);
           thisdistortion = vp9_block_error_c(be->coeff,
               BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[j], 16), 32);
@@ -1846,7 +1865,11 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
         if (otherrd) {
           for (j = 0; j < 4; j += 2) {
             BLOCK *be = &x->block[ib + iblock[j]];
-            x->fwd_txm8x4(be->src_diff, be->coeff, 32);
+            int16_t* const src_diff =
+                raster_block_offset_int16(xd, BLOCK_SIZE_MB16X16,
+                                          0, ib + iblock[j],
+                                          x->plane[0].src_diff);
+            x->fwd_txm8x4(src_diff, be->coeff, 32);
             x->quantize_b_4x4_pair(x, ib + iblock[j], ib + iblock[j] + 1, 16);
             thisdistortion = vp9_block_error_c(be->coeff,
                 BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[j], 16), 32);
@@ -1866,7 +1889,7 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm,
             xd->mode_info_context->mbmi.txfm_size = TX_8X8;
           }
         }
-        x->fwd_txm8x8(be->src_diff, be2->coeff, 32);
+        x->fwd_txm8x8(src_diff, be2->coeff, 32);
         x->quantize_b_8x8(x, idx, DCT_DCT, 16);
         thisdistortion = vp9_block_error_c(be2->coeff,
             BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64);
@@ -3768,12 +3791,7 @@ static void rd_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
         vp9_build_inter_predictors_sbuv(&x->e_mbd, mb_row, mb_col,
                                         BLOCK_SIZE_MB16X16);
 
-        vp9_subtract_sbuv_s_c(x->src_diff,
-                              x->src.u_buffer,
-                              x->src.v_buffer, x->src.uv_stride,
-                              xd->plane[1].dst.buf,
-                              xd->plane[2].dst.buf, xd->plane[1].dst.stride,
-                              BLOCK_SIZE_MB16X16);
+        vp9_subtract_sbuv(x, BLOCK_SIZE_MB16X16);
 
         super_block_uvrd_4x4(cm, x, &rate_uv, &distortion_uv,
                              &uv_skippable, BLOCK_SIZE_MB16X16);
diff --git a/vp9/encoder/x86/vp9_x86_csystemdependent.c b/vp9/encoder/x86/vp9_x86_csystemdependent.c
index 04383fcb4..6016e14eb 100644
--- a/vp9/encoder/x86/vp9_x86_csystemdependent.c
+++ b/vp9/encoder/x86/vp9_x86_csystemdependent.c
@@ -17,7 +17,7 @@
 
 // TODO(jimbankoski) Consider rewriting the c to take the same values rather
 // than going through these pointer conversions
-#if HAVE_MMX
+#if 0 && HAVE_MMX
 void vp9_short_fdct8x4_mmx(short *input, short *output, int pitch) {
   vp9_short_fdct4x4_mmx(input,   output,    pitch);
   vp9_short_fdct4x4_mmx(input + 4, output + 16, pitch);
@@ -38,7 +38,7 @@ void vp9_subtract_b_mmx(BLOCK *be, BLOCKD *bd, int pitch) {
 
 #endif
 
-#if HAVE_SSE2
+#if 0 && HAVE_SSE2
 void vp9_subtract_b_sse2_impl(unsigned char *z,  int src_stride,
                               short *diff, unsigned char *predictor,
                               int pitch);