diff --git a/examples/vp9_spatial_svc_encoder.c b/examples/vp9_spatial_svc_encoder.c
index 5a60976..af8aa86 100644
--- a/examples/vp9_spatial_svc_encoder.c
+++ b/examples/vp9_spatial_svc_encoder.c
@@ -25,6 +25,7 @@
 #include "../tools_common.h"
 #include "../video_writer.h"
 
+#include "../vpx_ports/vpx_timer.h"
 #include "vpx/svc_context.h"
 #include "vpx/vp8cx.h"
 #include "vpx/vpx_encoder.h"
@@ -564,6 +565,8 @@
   double sum_bitrate2 = 0.0;
   double framerate  = 30.0;
 #endif
+  struct vpx_usec_timer timer;
+  int64_t cx_time = 0;
   memset(&svc_ctx, 0, sizeof(svc_ctx));
   svc_ctx.log_print = 1;
   exec_name = argv[0];
@@ -632,6 +635,9 @@
     vpx_codec_control(&codec, VP8E_SET_CPUUSED, svc_ctx.speed);
   if (svc_ctx.threads)
     vpx_codec_control(&codec, VP9E_SET_TILE_COLUMNS, (svc_ctx.threads >> 1));
+  if (svc_ctx.speed >= 5)
+    vpx_codec_control(&codec, VP9E_SET_AQ_MODE, 3);
+
 
   // Encode frames
   while (!end_of_stream) {
@@ -643,9 +649,12 @@
       end_of_stream = 1;
     }
 
+    vpx_usec_timer_start(&timer);
     res = vpx_svc_encode(&svc_ctx, &codec, (end_of_stream ? NULL : &raw),
                          pts, frame_duration, svc_ctx.speed >= 5 ?
                          VPX_DL_REALTIME : VPX_DL_GOOD_QUALITY);
+    vpx_usec_timer_mark(&timer);
+    cx_time += vpx_usec_timer_elapsed(&timer);
 
     printf("%s", vpx_svc_get_message(&svc_ctx));
     if (res != VPX_CODEC_OK) {
@@ -784,6 +793,10 @@
     }
   }
 #endif
+  printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f \n",
+         frame_cnt,
+         1000 * (float)cx_time / (double)(frame_cnt * 1000000),
+         1000000 * (double)frame_cnt / (double)cx_time);
   vpx_img_free(&raw);
   // display average size, psnr
   printf("%s", vpx_svc_dump_statistics(&svc_ctx));
diff --git a/vp10/common/blockd.h b/vp10/common/blockd.h
index 26703d1..30bf103 100644
--- a/vp10/common/blockd.h
+++ b/vp10/common/blockd.h
@@ -238,7 +238,7 @@
     return TX_4X4;
   } else {
     const BLOCK_SIZE plane_bsize = ss_size_lookup[bsize][xss][yss];
-    return MIN(y_tx_size, max_txsize_lookup[plane_bsize]);
+    return VPXMIN(y_tx_size, max_txsize_lookup[plane_bsize]);
   }
 }
 
diff --git a/vp10/common/common_data.h b/vp10/common/common_data.h
index 8189518..f2ce039 100644
--- a/vp10/common/common_data.h
+++ b/vp10/common/common_data.h
@@ -35,7 +35,7 @@
 static const uint8_t num_8x8_blocks_high_lookup[BLOCK_SIZES] =
   {1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8};
 
-// MIN(3, MIN(b_width_log2(bsize), b_height_log2(bsize)))
+// VPXMIN(3, VPXMIN(b_width_log2(bsize), b_height_log2(bsize)))
 static const uint8_t size_group_lookup[BLOCK_SIZES] =
   {0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 3};
 
diff --git a/vp10/common/loopfilter.c b/vp10/common/loopfilter.c
index c11d46a..f108374 100644
--- a/vp10/common/loopfilter.c
+++ b/vp10/common/loopfilter.c
@@ -1588,7 +1588,7 @@
   if (partial_frame && cm->mi_rows > 8) {
     start_mi_row = cm->mi_rows >> 1;
     start_mi_row &= 0xfffffff8;
-    mi_rows_to_filter = MAX(cm->mi_rows / 8, 8);
+    mi_rows_to_filter = VPXMAX(cm->mi_rows / 8, 8);
   }
   end_mi_row = start_mi_row + mi_rows_to_filter;
   vp10_loop_filter_frame_init(cm, frame_filter_level);
diff --git a/vp10/common/postproc.c b/vp10/common/postproc.c
index 2d5b36a..d273420 100644
--- a/vp10/common/postproc.c
+++ b/vp10/common/postproc.c
@@ -625,7 +625,7 @@
 
 int vp10_post_proc_frame(struct VP10Common *cm,
                         YV12_BUFFER_CONFIG *dest, vp10_ppflags_t *ppflags) {
-  const int q = MIN(105, cm->lf.filter_level * 2);
+  const int q = VPXMIN(105, cm->lf.filter_level * 2);
   const int flags = ppflags->post_proc_flag;
   YV12_BUFFER_CONFIG *const ppbuf = &cm->post_proc_buffer;
   struct postproc_state *const ppstate = &cm->postproc_state;
diff --git a/vp10/common/pred_common.h b/vp10/common/pred_common.h
index 9345c6b..22d2774 100644
--- a/vp10/common/pred_common.h
+++ b/vp10/common/pred_common.h
@@ -24,14 +24,14 @@
   const int mi_offset = mi_row * cm->mi_cols + mi_col;
   const int bw = num_8x8_blocks_wide_lookup[bsize];
   const int bh = num_8x8_blocks_high_lookup[bsize];
-  const int xmis = MIN(cm->mi_cols - mi_col, bw);
-  const int ymis = MIN(cm->mi_rows - mi_row, bh);
+  const int xmis = VPXMIN(cm->mi_cols - mi_col, bw);
+  const int ymis = VPXMIN(cm->mi_rows - mi_row, bh);
   int x, y, segment_id = MAX_SEGMENTS;
 
   for (y = 0; y < ymis; ++y)
     for (x = 0; x < xmis; ++x)
-      segment_id = MIN(segment_id,
-                       segment_ids[mi_offset + y * cm->mi_cols + x]);
+      segment_id =
+          VPXMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]);
 
   assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
   return segment_id;
diff --git a/vp10/common/scan.c b/vp10/common/scan.c
index 7217f6d..ad849af 100644
--- a/vp10/common/scan.c
+++ b/vp10/common/scan.c
@@ -695,13 +695,6 @@
   1023,
 };
 
-const scan_order vp10_default_scan_orders[TX_SIZES] = {
-  {default_scan_4x4,   vp10_default_iscan_4x4,   default_scan_4x4_neighbors},
-  {default_scan_8x8,   vp10_default_iscan_8x8,   default_scan_8x8_neighbors},
-  {default_scan_16x16, vp10_default_iscan_16x16, default_scan_16x16_neighbors},
-  {default_scan_32x32, vp10_default_iscan_32x32, default_scan_32x32_neighbors},
-};
-
 const scan_order vp10_scan_orders[TX_SIZES][TX_TYPES] = {
   {  // TX_4X4
     {default_scan_4x4, vp10_default_iscan_4x4, default_scan_4x4_neighbors},
diff --git a/vp10/common/scan.h b/vp10/common/scan.h
index f5a020f..ada67f4 100644
--- a/vp10/common/scan.h
+++ b/vp10/common/scan.h
@@ -29,7 +29,6 @@
   const int16_t *neighbors;
 } scan_order;
 
-extern const scan_order vp10_default_scan_orders[TX_SIZES];
 extern const scan_order vp10_scan_orders[TX_SIZES][TX_TYPES];
 
 static INLINE int get_coef_context(const int16_t *neighbors,
diff --git a/vp10/common/thread_common.c b/vp10/common/thread_common.c
index 69c6471..e87caab 100644
--- a/vp10/common/thread_common.c
+++ b/vp10/common/thread_common.c
@@ -165,7 +165,7 @@
   // Decoder may allocate more threads than number of tiles based on user's
   // input.
   const int tile_cols = 1 << cm->log2_tile_cols;
-  const int num_workers = MIN(nworkers, tile_cols);
+  const int num_workers = VPXMIN(nworkers, tile_cols);
   int i;
 
   if (!lf_sync->sync_range || sb_rows != lf_sync->rows ||
@@ -229,7 +229,7 @@
   if (partial_frame && cm->mi_rows > 8) {
     start_mi_row = cm->mi_rows >> 1;
     start_mi_row &= 0xfffffff8;
-    mi_rows_to_filter = MAX(cm->mi_rows / 8, 8);
+    mi_rows_to_filter = VPXMAX(cm->mi_rows / 8, 8);
   }
   end_mi_row = start_mi_row + mi_rows_to_filter;
   vp10_loop_filter_frame_init(cm, frame_filter_level);
diff --git a/vp10/common/tile_common.c b/vp10/common/tile_common.c
index 821a5e8..f830e60 100644
--- a/vp10/common/tile_common.c
+++ b/vp10/common/tile_common.c
@@ -18,7 +18,7 @@
 static int get_tile_offset(int idx, int mis, int log2) {
   const int sb_cols = mi_cols_aligned_to_sb(mis) >> MI_BLOCK_SIZE_LOG2;
   const int offset = ((idx * sb_cols) >> log2) << MI_BLOCK_SIZE_LOG2;
-  return MIN(offset, mis);
+  return VPXMIN(offset, mis);
 }
 
 void vp10_tile_set_row(TileInfo *tile, const VP10_COMMON *cm, int row) {
diff --git a/vp10/decoder/decodeframe.c b/vp10/decoder/decodeframe.c
index 6cb2ff5..b17fa1e 100644
--- a/vp10/decoder/decodeframe.c
+++ b/vp10/decoder/decodeframe.c
@@ -624,7 +624,7 @@
     // pixels of each superblock row can be changed by next superblock row.
     if (pbi->frame_parallel_decode)
       vp10_frameworker_wait(pbi->frame_worker_owner, ref_frame_buf,
-                           MAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
+                            VPXMAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
 
     // Skip border extension if block is inside the frame.
     if (x0 < 0 || x0 > frame_width - 1 || x1 < 0 || x1 > frame_width - 1 ||
@@ -652,7 +652,7 @@
      if (pbi->frame_parallel_decode) {
        const int y1 = (y0_16 + (h - 1) * ys) >> SUBPEL_BITS;
        vp10_frameworker_wait(pbi->frame_worker_owner, ref_frame_buf,
-                            MAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
+                             VPXMAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
      }
   }
 #if CONFIG_VP9_HIGHBITDEPTH
@@ -723,8 +723,8 @@
 static INLINE TX_SIZE dec_get_uv_tx_size(const MB_MODE_INFO *mbmi,
                                          int n4_wl, int n4_hl) {
   // get minimum log2 num4x4s dimension
-  const int x = MIN(n4_wl, n4_hl);
-  return MIN(mbmi->tx_size,  x);
+  const int x = VPXMIN(n4_wl, n4_hl);
+  return VPXMIN(mbmi->tx_size,  x);
 }
 
 static INLINE void dec_reset_skip_context(MACROBLOCKD *xd) {
@@ -785,8 +785,8 @@
   const int less8x8 = bsize < BLOCK_8X8;
   const int bw = 1 << (bwl - 1);
   const int bh = 1 << (bhl - 1);
-  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
-  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+  const int x_mis = VPXMIN(bw, cm->mi_cols - mi_col);
+  const int y_mis = VPXMIN(bh, cm->mi_rows - mi_row);
 
   MB_MODE_INFO *mbmi = set_offsets(cm, xd, bsize, mi_row, mi_col,
                                    bw, bh, x_mis, y_mis, bwl, bhl);
@@ -1570,7 +1570,7 @@
   const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
   const int tile_cols = 1 << cm->log2_tile_cols;
   const int tile_rows = 1 << cm->log2_tile_rows;
-  const int num_workers = MIN(pbi->max_threads & ~1, tile_cols);
+  const int num_workers = VPXMIN(pbi->max_threads & ~1, tile_cols);
   TileBuffer tile_buffers[1][1 << 6];
   int n;
   int final_worker = -1;
@@ -1637,7 +1637,7 @@
     int group_start = 0;
     while (group_start < tile_cols) {
       const TileBuffer largest = tile_buffers[0][group_start];
-      const int group_end = MIN(group_start + num_workers, tile_cols) - 1;
+      const int group_end = VPXMIN(group_start + num_workers, tile_cols) - 1;
       memmove(tile_buffers[0] + group_start, tile_buffers[0] + group_start + 1,
               (group_end - group_start) * sizeof(tile_buffers[0][0]));
       tile_buffers[0][group_end] = largest;
@@ -2069,7 +2069,7 @@
   rb->error_handler = error_handler;
   rb->error_handler_data = &pbi->common;
   if (pbi->decrypt_cb) {
-    const int n = (int)MIN(MAX_VP9_HEADER_SIZE, data_end - data);
+    const int n = (int)VPXMIN(MAX_VP9_HEADER_SIZE, data_end - data);
     pbi->decrypt_cb(pbi->decrypt_state, data, clear_data, n);
     rb->bit_buffer = clear_data;
     rb->bit_buffer_end = clear_data + n;
diff --git a/vp10/decoder/decodemv.c b/vp10/decoder/decodemv.c
index 5fbe2ca..cf331ee 100644
--- a/vp10/decoder/decodemv.c
+++ b/vp10/decoder/decodemv.c
@@ -87,7 +87,7 @@
   if (allow_select && tx_mode == TX_MODE_SELECT && bsize >= BLOCK_8X8)
     return read_selected_tx_size(cm, xd, max_tx_size, r);
   else
-    return MIN(max_tx_size, tx_mode_to_biggest_tx_size[tx_mode]);
+    return VPXMIN(max_tx_size, tx_mode_to_biggest_tx_size[tx_mode]);
 }
 
 static int dec_get_segment_id(const VP10_COMMON *cm, const uint8_t *segment_ids,
@@ -96,8 +96,8 @@
 
   for (y = 0; y < y_mis; y++)
     for (x = 0; x < x_mis; x++)
-      segment_id = MIN(segment_id,
-                       segment_ids[mi_offset + y * cm->mi_cols + x]);
+      segment_id =
+          VPXMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]);
 
   assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
   return segment_id;
@@ -156,8 +156,8 @@
   const int bh = xd->plane[0].n4_h >> 1;
 
   // TODO(slavarnway): move x_mis, y_mis into xd ?????
-  const int x_mis = MIN(cm->mi_cols - mi_col, bw);
-  const int y_mis = MIN(cm->mi_rows - mi_row, bh);
+  const int x_mis = VPXMIN(cm->mi_cols - mi_col, bw);
+  const int y_mis = VPXMIN(cm->mi_rows - mi_row, bh);
 
   if (!seg->enabled)
     return 0;  // Default for disabled segmentation
@@ -212,8 +212,8 @@
   const int bh = xd->plane[0].n4_h >> 1;
 
   // TODO(slavarnway): move x_mis, y_mis into xd ?????
-  const int x_mis = MIN(cm->mi_cols - mi_col, bw);
-  const int y_mis = MIN(cm->mi_rows - mi_row, bh);
+  const int x_mis = VPXMIN(cm->mi_cols - mi_col, bw);
+  const int y_mis = VPXMIN(cm->mi_rows - mi_row, bh);
 
   mbmi->segment_id = read_intra_segment_id(cm, mi_offset, x_mis, y_mis, r);
   mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r);
diff --git a/vp10/encoder/aq_complexity.c b/vp10/encoder/aq_complexity.c
index 195732a..7ba879d 100644
--- a/vp10/encoder/aq_complexity.c
+++ b/vp10/encoder/aq_complexity.c
@@ -117,8 +117,8 @@
   const int mi_offset = mi_row * cm->mi_cols + mi_col;
   const int bw = num_8x8_blocks_wide_lookup[BLOCK_64X64];
   const int bh = num_8x8_blocks_high_lookup[BLOCK_64X64];
-  const int xmis = MIN(cm->mi_cols - mi_col, num_8x8_blocks_wide_lookup[bs]);
-  const int ymis = MIN(cm->mi_rows - mi_row, num_8x8_blocks_high_lookup[bs]);
+  const int xmis = VPXMIN(cm->mi_cols - mi_col, num_8x8_blocks_wide_lookup[bs]);
+  const int ymis = VPXMIN(cm->mi_rows - mi_row, num_8x8_blocks_high_lookup[bs]);
   int x, y;
   int i;
   unsigned char segment;
@@ -136,7 +136,7 @@
 
     vpx_clear_system_state();
     low_var_thresh = (cpi->oxcf.pass == 2)
-      ? MAX(cpi->twopass.mb_av_energy, MIN_DEFAULT_LV_THRESH)
+      ? VPXMAX(cpi->twopass.mb_av_energy, MIN_DEFAULT_LV_THRESH)
       : DEFAULT_LV_THRESH;
 
     vp10_setup_src_planes(mb, cpi->Source, mi_row, mi_col);
diff --git a/vp10/encoder/aq_cyclicrefresh.c b/vp10/encoder/aq_cyclicrefresh.c
index 3d6e38b..45181d2 100644
--- a/vp10/encoder/aq_cyclicrefresh.c
+++ b/vp10/encoder/aq_cyclicrefresh.c
@@ -220,8 +220,8 @@
   CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
   const int bw = num_8x8_blocks_wide_lookup[bsize];
   const int bh = num_8x8_blocks_high_lookup[bsize];
-  const int xmis = MIN(cm->mi_cols - mi_col, bw);
-  const int ymis = MIN(cm->mi_rows - mi_row, bh);
+  const int xmis = VPXMIN(cm->mi_cols - mi_col, bw);
+  const int ymis = VPXMIN(cm->mi_rows - mi_row, bh);
   const int block_index = mi_row * cm->mi_cols + mi_col;
   const int refresh_this_block = candidate_refresh_aq(cr, mbmi, rate, dist,
                                                       bsize);
@@ -413,10 +413,10 @@
     assert(mi_col >= 0 && mi_col < cm->mi_cols);
     bl_index = mi_row * cm->mi_cols + mi_col;
     // Loop through all 8x8 blocks in superblock and update map.
-    xmis = MIN(cm->mi_cols - mi_col,
-               num_8x8_blocks_wide_lookup[BLOCK_64X64]);
-    ymis = MIN(cm->mi_rows - mi_row,
-               num_8x8_blocks_high_lookup[BLOCK_64X64]);
+    xmis =
+        VPXMIN(cm->mi_cols - mi_col, num_8x8_blocks_wide_lookup[BLOCK_64X64]);
+    ymis =
+        VPXMIN(cm->mi_rows - mi_row, num_8x8_blocks_high_lookup[BLOCK_64X64]);
     for (y = 0; y < ymis; y++) {
       for (x = 0; x < xmis; x++) {
         const int bl_index2 = bl_index + y * cm->mi_cols + x;
@@ -545,8 +545,9 @@
 
     // Set a more aggressive (higher) q delta for segment BOOST2.
     qindex_delta = compute_deltaq(
-        cpi, cm->base_qindex, MIN(CR_MAX_RATE_TARGET_RATIO,
-        0.1 * cr->rate_boost_fac * cr->rate_ratio_qdelta));
+        cpi, cm->base_qindex,
+        VPXMIN(CR_MAX_RATE_TARGET_RATIO,
+               0.1 * cr->rate_boost_fac * cr->rate_ratio_qdelta));
     cr->qindex_delta[2] = qindex_delta;
     vp10_set_segdata(seg, CR_SEGMENT_ID_BOOST2, SEG_LVL_ALT_Q, qindex_delta);
 
diff --git a/vp10/encoder/bitstream.c b/vp10/encoder/bitstream.c
index b872b4e..11ee0f5 100644
--- a/vp10/encoder/bitstream.c
+++ b/vp10/encoder/bitstream.c
@@ -815,7 +815,7 @@
 static void encode_txfm_probs(VP10_COMMON *cm, vpx_writer *w,
                               FRAME_COUNTS *counts) {
   // Mode
-  vpx_write_literal(w, MIN(cm->tx_mode, ALLOW_32X32), 2);
+  vpx_write_literal(w, VPXMIN(cm->tx_mode, ALLOW_32X32), 2);
   if (cm->tx_mode >= ALLOW_32X32)
     vpx_write_bit(w, cm->tx_mode == TX_MODE_SELECT);
 
diff --git a/vp10/encoder/denoiser.c b/vp10/encoder/denoiser.c
index 3c9d949..271113a 100644
--- a/vp10/encoder/denoiser.c
+++ b/vp10/encoder/denoiser.c
@@ -124,10 +124,10 @@
             adj = adj_val[2];
         }
         if (diff > 0) {
-          avg[c] = MIN(UINT8_MAX, sig[c] + adj);
+          avg[c] = VPXMIN(UINT8_MAX, sig[c] + adj);
           total_adj += adj;
         } else {
-          avg[c] = MAX(0, sig[c] - adj);
+          avg[c] = VPXMAX(0, sig[c] - adj);
           total_adj -= adj;
         }
       }
@@ -164,13 +164,13 @@
         // Diff positive means we made positive adjustment above
         // (in first try/attempt), so now make negative adjustment to bring
         // denoised signal down.
-        avg[c] = MAX(0, avg[c] - adj);
+        avg[c] = VPXMAX(0, avg[c] - adj);
         total_adj -= adj;
       } else {
         // Diff negative means we made negative adjustment above
         // (in first try/attempt), so now make positive adjustment to bring
         // denoised signal up.
-        avg[c] = MIN(UINT8_MAX, avg[c] + adj);
+        avg[c] = VPXMIN(UINT8_MAX, avg[c] + adj);
         total_adj += adj;
       }
     }
diff --git a/vp10/encoder/encodeframe.c b/vp10/encoder/encodeframe.c
index 0277723..1635046 100644
--- a/vp10/encoder/encodeframe.c
+++ b/vp10/encoder/encodeframe.c
@@ -40,7 +40,6 @@
 #include "vp10/encoder/encodemv.h"
 #include "vp10/encoder/ethread.h"
 #include "vp10/encoder/extend.h"
-#include "vp10/encoder/pickmode.h"
 #include "vp10/encoder/rd.h"
 #include "vp10/encoder/rdopt.h"
 #include "vp10/encoder/segmentation.h"
@@ -979,8 +978,8 @@
   const struct segmentation *const seg = &cm->seg;
   const int bw = num_8x8_blocks_wide_lookup[mi->mbmi.sb_type];
   const int bh = num_8x8_blocks_high_lookup[mi->mbmi.sb_type];
-  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
-  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+  const int x_mis = VPXMIN(bw, cm->mi_cols - mi_col);
+  const int y_mis = VPXMIN(bh, cm->mi_rows - mi_row);
   MV_REF *const frame_mvs =
       cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col;
   int w, h;
@@ -1132,8 +1131,8 @@
 
   mbmi->sb_type = bsize;
   mbmi->mode = ZEROMV;
-  mbmi->tx_size = MIN(max_txsize_lookup[bsize],
-                      tx_mode_to_biggest_tx_size[tx_mode]);
+  mbmi->tx_size = VPXMIN(max_txsize_lookup[bsize],
+                         tx_mode_to_biggest_tx_size[tx_mode]);
   mbmi->skip = 1;
   mbmi->uv_mode = DC_PRED;
   mbmi->ref_frame[0] = LAST_FRAME;
@@ -1496,7 +1495,7 @@
                                       int rows_left, int cols_left,
                                       int *bh, int *bw) {
   if (rows_left <= 0 || cols_left <= 0) {
-    return MIN(bsize, BLOCK_8X8);
+    return VPXMIN(bsize, BLOCK_8X8);
   } else {
     for (; bsize > 0; bsize -= 3) {
       *bh = num_8x8_blocks_high_lookup[bsize];
@@ -1561,269 +1560,6 @@
   }
 }
 
-static const struct {
-  int row;
-  int col;
-} coord_lookup[16] = {
-    // 32x32 index = 0
-    {0, 0}, {0, 2}, {2, 0}, {2, 2},
-    // 32x32 index = 1
-    {0, 4}, {0, 6}, {2, 4}, {2, 6},
-    // 32x32 index = 2
-    {4, 0}, {4, 2}, {6, 0}, {6, 2},
-    // 32x32 index = 3
-    {4, 4}, {4, 6}, {6, 4}, {6, 6},
-};
-
-static void set_source_var_based_partition(VP10_COMP *cpi,
-                                           const TileInfo *const tile,
-                                           MACROBLOCK *const x,
-                                           MODE_INFO **mi_8x8,
-                                           int mi_row, int mi_col) {
-  VP10_COMMON *const cm = &cpi->common;
-  const int mis = cm->mi_stride;
-  const int row8x8_remaining = tile->mi_row_end - mi_row;
-  const int col8x8_remaining = tile->mi_col_end - mi_col;
-  MODE_INFO *mi_upper_left = cm->mi + mi_row * mis + mi_col;
-
-  vp10_setup_src_planes(x, cpi->Source, mi_row, mi_col);
-
-  assert((row8x8_remaining > 0) && (col8x8_remaining > 0));
-
-  // In-image SB64
-  if ((col8x8_remaining >= MI_BLOCK_SIZE) &&
-      (row8x8_remaining >= MI_BLOCK_SIZE)) {
-    int i, j;
-    int index;
-    diff d32[4];
-    const int offset = (mi_row >> 1) * cm->mb_cols + (mi_col >> 1);
-    int is_larger_better = 0;
-    int use32x32 = 0;
-    unsigned int thr = cpi->source_var_thresh;
-
-    memset(d32, 0, 4 * sizeof(diff));
-
-    for (i = 0; i < 4; i++) {
-      diff *d16[4];
-
-      for (j = 0; j < 4; j++) {
-        int b_mi_row = coord_lookup[i * 4 + j].row;
-        int b_mi_col = coord_lookup[i * 4 + j].col;
-        int boffset = b_mi_row / 2 * cm->mb_cols +
-                      b_mi_col / 2;
-
-        d16[j] = cpi->source_diff_var + offset + boffset;
-
-        index = b_mi_row * mis + b_mi_col;
-        mi_8x8[index] = mi_upper_left + index;
-        mi_8x8[index]->mbmi.sb_type = BLOCK_16X16;
-
-        // TODO(yunqingwang): If d16[j].var is very large, use 8x8 partition
-        // size to further improve quality.
-      }
-
-      is_larger_better = (d16[0]->var < thr) && (d16[1]->var < thr) &&
-          (d16[2]->var < thr) && (d16[3]->var < thr);
-
-      // Use 32x32 partition
-      if (is_larger_better) {
-        use32x32 += 1;
-
-        for (j = 0; j < 4; j++) {
-          d32[i].sse += d16[j]->sse;
-          d32[i].sum += d16[j]->sum;
-        }
-
-        d32[i].var = d32[i].sse - (((int64_t)d32[i].sum * d32[i].sum) >> 10);
-
-        index = coord_lookup[i*4].row * mis + coord_lookup[i*4].col;
-        mi_8x8[index] = mi_upper_left + index;
-        mi_8x8[index]->mbmi.sb_type = BLOCK_32X32;
-      }
-    }
-
-    if (use32x32 == 4) {
-      thr <<= 1;
-      is_larger_better = (d32[0].var < thr) && (d32[1].var < thr) &&
-          (d32[2].var < thr) && (d32[3].var < thr);
-
-      // Use 64x64 partition
-      if (is_larger_better) {
-        mi_8x8[0] = mi_upper_left;
-        mi_8x8[0]->mbmi.sb_type = BLOCK_64X64;
-      }
-    }
-  } else {   // partial in-image SB64
-    int bh = num_8x8_blocks_high_lookup[BLOCK_16X16];
-    int bw = num_8x8_blocks_wide_lookup[BLOCK_16X16];
-    set_partial_b64x64_partition(mi_upper_left, mis, bh, bw,
-        row8x8_remaining, col8x8_remaining, BLOCK_16X16, mi_8x8);
-  }
-}
-
-static void update_state_rt(VP10_COMP *cpi, ThreadData *td,
-                            PICK_MODE_CONTEXT *ctx,
-                            int mi_row, int mi_col, int bsize) {
-  VP10_COMMON *const cm = &cpi->common;
-  MACROBLOCK *const x = &td->mb;
-  MACROBLOCKD *const xd = &x->e_mbd;
-  MODE_INFO *const mi = xd->mi[0];
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  const struct segmentation *const seg = &cm->seg;
-  const int bw = num_8x8_blocks_wide_lookup[mi->mbmi.sb_type];
-  const int bh = num_8x8_blocks_high_lookup[mi->mbmi.sb_type];
-  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
-  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
-
-  *(xd->mi[0]) = ctx->mic;
-  *(x->mbmi_ext) = ctx->mbmi_ext;
-
-  if (seg->enabled && cpi->oxcf.aq_mode) {
-    // For in frame complexity AQ or variance AQ, copy segment_id from
-    // segmentation_map.
-    if (cpi->oxcf.aq_mode == COMPLEXITY_AQ ||
-        cpi->oxcf.aq_mode == VARIANCE_AQ ) {
-      const uint8_t *const map = seg->update_map ? cpi->segmentation_map
-                                                 : cm->last_frame_seg_map;
-      mbmi->segment_id = get_segment_id(cm, map, bsize, mi_row, mi_col);
-    } else {
-    // Setting segmentation map for cyclic_refresh.
-      vp10_cyclic_refresh_update_segment(cpi, mbmi, mi_row, mi_col, bsize,
-                                        ctx->rate, ctx->dist, x->skip);
-    }
-    vp10_init_plane_quantizers(cpi, x);
-  }
-
-  if (is_inter_block(mbmi)) {
-    vp10_update_mv_count(td);
-    if (cm->interp_filter == SWITCHABLE) {
-      const int pred_ctx = vp10_get_pred_context_switchable_interp(xd);
-      ++td->counts->switchable_interp[pred_ctx][mbmi->interp_filter];
-    }
-
-    if (mbmi->sb_type < BLOCK_8X8) {
-      mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
-      mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
-    }
-  }
-
-  if (cm->use_prev_frame_mvs) {
-    MV_REF *const frame_mvs =
-        cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col;
-    int w, h;
-
-    for (h = 0; h < y_mis; ++h) {
-      MV_REF *const frame_mv = frame_mvs + h * cm->mi_cols;
-      for (w = 0; w < x_mis; ++w) {
-        MV_REF *const mv = frame_mv + w;
-        mv->ref_frame[0] = mi->mbmi.ref_frame[0];
-        mv->ref_frame[1] = mi->mbmi.ref_frame[1];
-        mv->mv[0].as_int = mi->mbmi.mv[0].as_int;
-        mv->mv[1].as_int = mi->mbmi.mv[1].as_int;
-      }
-    }
-  }
-
-  x->skip = ctx->skip;
-  x->skip_txfm[0] = mbmi->segment_id ? 0 : ctx->skip_txfm[0];
-}
-
-static void encode_b_rt(VP10_COMP *cpi, ThreadData *td,
-                        const TileInfo *const tile,
-                        TOKENEXTRA **tp, int mi_row, int mi_col,
-                        int output_enabled, BLOCK_SIZE bsize,
-                        PICK_MODE_CONTEXT *ctx) {
-  MACROBLOCK *const x = &td->mb;
-  set_offsets(cpi, tile, x, mi_row, mi_col, bsize);
-  update_state_rt(cpi, td, ctx, mi_row, mi_col, bsize);
-
-#if CONFIG_VP9_TEMPORAL_DENOISING
-  if (cpi->oxcf.noise_sensitivity > 0 && output_enabled &&
-      cpi->common.frame_type != KEY_FRAME) {
-    vp10_denoiser_denoise(&cpi->denoiser, x, mi_row, mi_col,
-                         MAX(BLOCK_8X8, bsize), ctx);
-  }
-#endif
-
-  encode_superblock(cpi, td, tp, output_enabled, mi_row, mi_col, bsize, ctx);
-  update_stats(&cpi->common, td);
-
-  (*tp)->token = EOSB_TOKEN;
-  (*tp)++;
-}
-
-static void encode_sb_rt(VP10_COMP *cpi, ThreadData *td,
-                         const TileInfo *const tile,
-                         TOKENEXTRA **tp, int mi_row, int mi_col,
-                         int output_enabled, BLOCK_SIZE bsize,
-                         PC_TREE *pc_tree) {
-  VP10_COMMON *const cm = &cpi->common;
-  MACROBLOCK *const x = &td->mb;
-  MACROBLOCKD *const xd = &x->e_mbd;
-
-  const int bsl = b_width_log2_lookup[bsize], hbs = (1 << bsl) / 4;
-  int ctx;
-  PARTITION_TYPE partition;
-  BLOCK_SIZE subsize;
-
-  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
-    return;
-
-  if (bsize >= BLOCK_8X8) {
-    const int idx_str = xd->mi_stride * mi_row + mi_col;
-    MODE_INFO ** mi_8x8 = cm->mi_grid_visible + idx_str;
-    ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
-    subsize = mi_8x8[0]->mbmi.sb_type;
-  } else {
-    ctx = 0;
-    subsize = BLOCK_4X4;
-  }
-
-  partition = partition_lookup[bsl][subsize];
-  if (output_enabled && bsize != BLOCK_4X4)
-    td->counts->partition[ctx][partition]++;
-
-  switch (partition) {
-    case PARTITION_NONE:
-      encode_b_rt(cpi, td, tile, tp, mi_row, mi_col, output_enabled, subsize,
-                  &pc_tree->none);
-      break;
-    case PARTITION_VERT:
-      encode_b_rt(cpi, td, tile, tp, mi_row, mi_col, output_enabled, subsize,
-                  &pc_tree->vertical[0]);
-      if (mi_col + hbs < cm->mi_cols && bsize > BLOCK_8X8) {
-        encode_b_rt(cpi, td, tile, tp, mi_row, mi_col + hbs, output_enabled,
-                    subsize, &pc_tree->vertical[1]);
-      }
-      break;
-    case PARTITION_HORZ:
-      encode_b_rt(cpi, td, tile, tp, mi_row, mi_col, output_enabled, subsize,
-                  &pc_tree->horizontal[0]);
-      if (mi_row + hbs < cm->mi_rows && bsize > BLOCK_8X8) {
-        encode_b_rt(cpi, td, tile, tp, mi_row + hbs, mi_col, output_enabled,
-                    subsize, &pc_tree->horizontal[1]);
-      }
-      break;
-    case PARTITION_SPLIT:
-      subsize = get_subsize(bsize, PARTITION_SPLIT);
-      encode_sb_rt(cpi, td, tile, tp, mi_row, mi_col, output_enabled, subsize,
-                   pc_tree->split[0]);
-      encode_sb_rt(cpi, td, tile, tp, mi_row, mi_col + hbs, output_enabled,
-                   subsize, pc_tree->split[1]);
-      encode_sb_rt(cpi, td, tile, tp, mi_row + hbs, mi_col, output_enabled,
-                   subsize, pc_tree->split[2]);
-      encode_sb_rt(cpi, td, tile, tp, mi_row + hbs, mi_col + hbs,
-                   output_enabled, subsize, pc_tree->split[3]);
-      break;
-    default:
-      assert(0 && "Invalid partition type.");
-      break;
-  }
-
-  if (partition != PARTITION_SPLIT || bsize == BLOCK_8X8)
-    update_partition_context(xd, mi_row, mi_col, subsize, bsize);
-}
-
 static void rd_use_partition(VP10_COMP *cpi,
                              ThreadData *td,
                              TileDataEnc *tile_data,
@@ -2133,8 +1869,8 @@
       MODE_INFO *mi = mi_8x8[index+j];
       BLOCK_SIZE sb_type = mi ? mi->mbmi.sb_type : 0;
       bs_hist[sb_type]++;
-      *min_block_size = MIN(*min_block_size, sb_type);
-      *max_block_size = MAX(*max_block_size, sb_type);
+      *min_block_size = VPXMIN(*min_block_size, sb_type);
+      *max_block_size = VPXMAX(*max_block_size, sb_type);
     }
     index += xd->mi_stride;
   }
@@ -2211,8 +1947,8 @@
   if (vp10_active_edge_sb(cpi, mi_row, mi_col)) {
     min_size = BLOCK_4X4;
   } else {
-    min_size = MIN(cpi->sf.rd_auto_partition_min_limit,
-                   MIN(min_size, max_size));
+    min_size =
+        VPXMIN(cpi->sf.rd_auto_partition_min_limit, VPXMIN(min_size, max_size));
   }
 
   // When use_square_partition_only is true, make sure at least one square
@@ -2248,8 +1984,8 @@
       for (idx = 0; idx < mi_width; ++idx) {
         mi = prev_mi[idy * cm->mi_stride + idx];
         bs = mi ? mi->mbmi.sb_type : bsize;
-        min_size = MIN(min_size, bs);
-        max_size = MAX(max_size, bs);
+        min_size = VPXMIN(min_size, bs);
+        max_size = VPXMAX(max_size, bs);
       }
     }
   }
@@ -2258,8 +1994,8 @@
     for (idy = 0; idy < mi_height; ++idy) {
       mi = xd->mi[idy * cm->mi_stride - 1];
       bs = mi ? mi->mbmi.sb_type : bsize;
-      min_size = MIN(min_size, bs);
-      max_size = MAX(max_size, bs);
+      min_size = VPXMIN(min_size, bs);
+      max_size = VPXMAX(max_size, bs);
     }
   }
 
@@ -2267,8 +2003,8 @@
     for (idx = 0; idx < mi_width; ++idx) {
       mi = xd->mi[idx - cm->mi_stride];
       bs = mi ? mi->mbmi.sb_type : bsize;
-      min_size = MIN(min_size, bs);
-      max_size = MAX(max_size, bs);
+      min_size = VPXMIN(min_size, bs);
+      max_size = VPXMAX(max_size, bs);
     }
   }
 
@@ -2433,9 +2169,9 @@
     int mb_row = mi_row >> 1;
     int mb_col = mi_col >> 1;
     int mb_row_end =
-        MIN(mb_row + num_16x16_blocks_high_lookup[bsize], cm->mb_rows);
+        VPXMIN(mb_row + num_16x16_blocks_high_lookup[bsize], cm->mb_rows);
     int mb_col_end =
-        MIN(mb_col + num_16x16_blocks_wide_lookup[bsize], cm->mb_cols);
+        VPXMIN(mb_col + num_16x16_blocks_wide_lookup[bsize], cm->mb_cols);
     int r, c;
 
     // compute a complexity measure, basically measure inconsistency of motion
@@ -2524,9 +2260,9 @@
           int mb_row = mi_row >> 1;
           int mb_col = mi_col >> 1;
           int mb_row_end =
-              MIN(mb_row + num_16x16_blocks_high_lookup[bsize], cm->mb_rows);
+              VPXMIN(mb_row + num_16x16_blocks_high_lookup[bsize], cm->mb_rows);
           int mb_col_end =
-              MIN(mb_col + num_16x16_blocks_wide_lookup[bsize], cm->mb_cols);
+              VPXMIN(mb_col + num_16x16_blocks_wide_lookup[bsize], cm->mb_cols);
           int r, c;
 
           int skip = 1;
@@ -2893,9 +2629,6 @@
 static TX_MODE select_tx_mode(const VP10_COMP *cpi, MACROBLOCKD *const xd) {
   if (xd->lossless)
     return ONLY_4X4;
-  if (cpi->common.frame_type == KEY_FRAME &&
-      cpi->sf.use_nonrd_pick_mode)
-    return ALLOW_16X16;
   if (cpi->sf.tx_size_search_method == USE_LARGESTALL)
     return ALLOW_32X32;
   else if (cpi->sf.tx_size_search_method == USE_FULL_RD||
@@ -2905,51 +2638,6 @@
     return cpi->common.tx_mode;
 }
 
-static void hybrid_intra_mode_search(VP10_COMP *cpi, MACROBLOCK *const x,
-                                     RD_COST *rd_cost, BLOCK_SIZE bsize,
-                                     PICK_MODE_CONTEXT *ctx) {
-  if (bsize < BLOCK_16X16)
-    vp10_rd_pick_intra_mode_sb(cpi, x, rd_cost, bsize, ctx, INT64_MAX);
-  else
-    vp10_pick_intra_mode(cpi, x, rd_cost, bsize, ctx);
-}
-
-static void nonrd_pick_sb_modes(VP10_COMP *cpi,
-                                TileDataEnc *tile_data, MACROBLOCK *const x,
-                                int mi_row, int mi_col, RD_COST *rd_cost,
-                                BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx) {
-  VP10_COMMON *const cm = &cpi->common;
-  TileInfo *const tile_info = &tile_data->tile_info;
-  MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi;
-  set_offsets(cpi, tile_info, x, mi_row, mi_col, bsize);
-  mbmi = &xd->mi[0]->mbmi;
-  mbmi->sb_type = bsize;
-
-  if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && cm->seg.enabled)
-    if (cyclic_refresh_segment_id_boosted(mbmi->segment_id))
-      x->rdmult = vp10_cyclic_refresh_get_rdmult(cpi->cyclic_refresh);
-
-  if (cm->frame_type == KEY_FRAME)
-    hybrid_intra_mode_search(cpi, x, rd_cost, bsize, ctx);
-  else if (segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP))
-    set_mode_info_seg_skip(x, cm->tx_mode, rd_cost, bsize);
-  else if (bsize >= BLOCK_8X8)
-    vp10_pick_inter_mode(cpi, x, tile_data, mi_row, mi_col,
-                        rd_cost, bsize, ctx);
-  else
-    vp10_pick_inter_mode_sub8x8(cpi, x, mi_row, mi_col,
-                               rd_cost, bsize, ctx);
-
-  duplicate_mode_info_in_sb(cm, xd, mi_row, mi_col, bsize);
-
-  if (rd_cost->rate == INT_MAX)
-    vp10_rd_cost_reset(rd_cost);
-
-  ctx->rate = rd_cost->rate;
-  ctx->dist = rd_cost->dist;
-}
-
 static void fill_mode_info_sb(VP10_COMMON *cm, MACROBLOCK *x,
                               int mi_row, int mi_col,
                               BLOCK_SIZE bsize,
@@ -3027,727 +2715,6 @@
   }
 }
 
-static void nonrd_pick_partition(VP10_COMP *cpi, ThreadData *td,
-                                 TileDataEnc *tile_data,
-                                 TOKENEXTRA **tp, int mi_row,
-                                 int mi_col, BLOCK_SIZE bsize, RD_COST *rd_cost,
-                                 int do_recon, int64_t best_rd,
-                                 PC_TREE *pc_tree) {
-  const SPEED_FEATURES *const sf = &cpi->sf;
-  VP10_COMMON *const cm = &cpi->common;
-  TileInfo *const tile_info = &tile_data->tile_info;
-  MACROBLOCK *const x = &td->mb;
-  MACROBLOCKD *const xd = &x->e_mbd;
-  const int ms = num_8x8_blocks_wide_lookup[bsize] / 2;
-  TOKENEXTRA *tp_orig = *tp;
-  PICK_MODE_CONTEXT *ctx = &pc_tree->none;
-  int i;
-  BLOCK_SIZE subsize = bsize;
-  RD_COST this_rdc, sum_rdc, best_rdc;
-  int do_split = bsize >= BLOCK_8X8;
-  int do_rect = 1;
-  // Override skipping rectangular partition operations for edge blocks
-  const int force_horz_split = (mi_row + ms >= cm->mi_rows);
-  const int force_vert_split = (mi_col + ms >= cm->mi_cols);
-  const int xss = x->e_mbd.plane[1].subsampling_x;
-  const int yss = x->e_mbd.plane[1].subsampling_y;
-
-  int partition_none_allowed = !force_horz_split && !force_vert_split;
-  int partition_horz_allowed = !force_vert_split && yss <= xss &&
-                               bsize >= BLOCK_8X8;
-  int partition_vert_allowed = !force_horz_split && xss <= yss &&
-                               bsize >= BLOCK_8X8;
-  (void) *tp_orig;
-
-  assert(num_8x8_blocks_wide_lookup[bsize] ==
-             num_8x8_blocks_high_lookup[bsize]);
-
-  vp10_rd_cost_init(&sum_rdc);
-  vp10_rd_cost_reset(&best_rdc);
-  best_rdc.rdcost = best_rd;
-
-  // Determine partition types in search according to the speed features.
-  // The threshold set here has to be of square block size.
-  if (sf->auto_min_max_partition_size) {
-    partition_none_allowed &= (bsize <= x->max_partition_size &&
-                               bsize >= x->min_partition_size);
-    partition_horz_allowed &= ((bsize <= x->max_partition_size &&
-                                bsize > x->min_partition_size) ||
-                                force_horz_split);
-    partition_vert_allowed &= ((bsize <= x->max_partition_size &&
-                                bsize > x->min_partition_size) ||
-                                force_vert_split);
-    do_split &= bsize > x->min_partition_size;
-  }
-  if (sf->use_square_partition_only) {
-    partition_horz_allowed &= force_horz_split;
-    partition_vert_allowed &= force_vert_split;
-  }
-
-  ctx->pred_pixel_ready = !(partition_vert_allowed ||
-                            partition_horz_allowed ||
-                            do_split);
-
-  // PARTITION_NONE
-  if (partition_none_allowed) {
-    nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col,
-                        &this_rdc, bsize, ctx);
-    ctx->mic.mbmi = xd->mi[0]->mbmi;
-    ctx->mbmi_ext = *x->mbmi_ext;
-    ctx->skip_txfm[0] = x->skip_txfm[0];
-    ctx->skip = x->skip;
-
-    if (this_rdc.rate != INT_MAX) {
-      int pl = partition_plane_context(xd, mi_row, mi_col, bsize);
-      this_rdc.rate += cpi->partition_cost[pl][PARTITION_NONE];
-      this_rdc.rdcost = RDCOST(x->rdmult, x->rddiv,
-                              this_rdc.rate, this_rdc.dist);
-      if (this_rdc.rdcost < best_rdc.rdcost) {
-        int64_t dist_breakout_thr = sf->partition_search_breakout_dist_thr;
-        int64_t rate_breakout_thr = sf->partition_search_breakout_rate_thr;
-
-        dist_breakout_thr >>= 8 - (b_width_log2_lookup[bsize] +
-            b_height_log2_lookup[bsize]);
-
-        rate_breakout_thr *= num_pels_log2_lookup[bsize];
-
-        best_rdc = this_rdc;
-        if (bsize >= BLOCK_8X8)
-          pc_tree->partitioning = PARTITION_NONE;
-
-        if (!x->e_mbd.lossless &&
-            this_rdc.rate < rate_breakout_thr &&
-            this_rdc.dist < dist_breakout_thr) {
-          do_split = 0;
-          do_rect = 0;
-        }
-      }
-    }
-  }
-
-  // store estimated motion vector
-  store_pred_mv(x, ctx);
-
-  // PARTITION_SPLIT
-  if (do_split) {
-    int pl = partition_plane_context(xd, mi_row, mi_col, bsize);
-    sum_rdc.rate += cpi->partition_cost[pl][PARTITION_SPLIT];
-    sum_rdc.rdcost = RDCOST(x->rdmult, x->rddiv, sum_rdc.rate, sum_rdc.dist);
-    subsize = get_subsize(bsize, PARTITION_SPLIT);
-    for (i = 0; i < 4 && sum_rdc.rdcost < best_rdc.rdcost; ++i) {
-      const int x_idx = (i & 1) * ms;
-      const int y_idx = (i >> 1) * ms;
-
-      if (mi_row + y_idx >= cm->mi_rows || mi_col + x_idx >= cm->mi_cols)
-        continue;
-      load_pred_mv(x, ctx);
-      nonrd_pick_partition(cpi, td, tile_data, tp,
-                           mi_row + y_idx, mi_col + x_idx,
-                           subsize, &this_rdc, 0,
-                           best_rdc.rdcost - sum_rdc.rdcost, pc_tree->split[i]);
-
-      if (this_rdc.rate == INT_MAX) {
-        vp10_rd_cost_reset(&sum_rdc);
-      } else {
-        sum_rdc.rate += this_rdc.rate;
-        sum_rdc.dist += this_rdc.dist;
-        sum_rdc.rdcost += this_rdc.rdcost;
-      }
-    }
-
-    if (sum_rdc.rdcost < best_rdc.rdcost) {
-      best_rdc = sum_rdc;
-      pc_tree->partitioning = PARTITION_SPLIT;
-    } else {
-      // skip rectangular partition test when larger block size
-      // gives better rd cost
-      if (sf->less_rectangular_check)
-        do_rect &= !partition_none_allowed;
-    }
-  }
-
-  // PARTITION_HORZ
-  if (partition_horz_allowed && do_rect) {
-    subsize = get_subsize(bsize, PARTITION_HORZ);
-    if (sf->adaptive_motion_search)
-      load_pred_mv(x, ctx);
-    pc_tree->horizontal[0].pred_pixel_ready = 1;
-    nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, &sum_rdc, subsize,
-                        &pc_tree->horizontal[0]);
-
-    pc_tree->horizontal[0].mic.mbmi = xd->mi[0]->mbmi;
-    pc_tree->horizontal[0].mbmi_ext = *x->mbmi_ext;
-    pc_tree->horizontal[0].skip_txfm[0] = x->skip_txfm[0];
-    pc_tree->horizontal[0].skip = x->skip;
-
-    if (sum_rdc.rdcost < best_rdc.rdcost && mi_row + ms < cm->mi_rows) {
-      load_pred_mv(x, ctx);
-      pc_tree->horizontal[1].pred_pixel_ready = 1;
-      nonrd_pick_sb_modes(cpi, tile_data, x, mi_row + ms, mi_col,
-                          &this_rdc, subsize,
-                          &pc_tree->horizontal[1]);
-
-      pc_tree->horizontal[1].mic.mbmi = xd->mi[0]->mbmi;
-      pc_tree->horizontal[1].mbmi_ext = *x->mbmi_ext;
-      pc_tree->horizontal[1].skip_txfm[0] = x->skip_txfm[0];
-      pc_tree->horizontal[1].skip = x->skip;
-
-      if (this_rdc.rate == INT_MAX) {
-        vp10_rd_cost_reset(&sum_rdc);
-      } else {
-        int pl = partition_plane_context(xd, mi_row, mi_col, bsize);
-        this_rdc.rate += cpi->partition_cost[pl][PARTITION_HORZ];
-        sum_rdc.rate += this_rdc.rate;
-        sum_rdc.dist += this_rdc.dist;
-        sum_rdc.rdcost = RDCOST(x->rdmult, x->rddiv,
-                                sum_rdc.rate, sum_rdc.dist);
-      }
-    }
-
-    if (sum_rdc.rdcost < best_rdc.rdcost) {
-      best_rdc = sum_rdc;
-      pc_tree->partitioning = PARTITION_HORZ;
-    } else {
-      pred_pixel_ready_reset(pc_tree, bsize);
-    }
-  }
-
-  // PARTITION_VERT
-  if (partition_vert_allowed && do_rect) {
-    subsize = get_subsize(bsize, PARTITION_VERT);
-    if (sf->adaptive_motion_search)
-      load_pred_mv(x, ctx);
-    pc_tree->vertical[0].pred_pixel_ready = 1;
-    nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, &sum_rdc, subsize,
-                        &pc_tree->vertical[0]);
-    pc_tree->vertical[0].mic.mbmi = xd->mi[0]->mbmi;
-    pc_tree->vertical[0].mbmi_ext = *x->mbmi_ext;
-    pc_tree->vertical[0].skip_txfm[0] = x->skip_txfm[0];
-    pc_tree->vertical[0].skip = x->skip;
-
-    if (sum_rdc.rdcost < best_rdc.rdcost && mi_col + ms < cm->mi_cols) {
-      load_pred_mv(x, ctx);
-      pc_tree->vertical[1].pred_pixel_ready = 1;
-      nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col + ms,
-                          &this_rdc, subsize,
-                          &pc_tree->vertical[1]);
-      pc_tree->vertical[1].mic.mbmi = xd->mi[0]->mbmi;
-      pc_tree->vertical[1].mbmi_ext = *x->mbmi_ext;
-      pc_tree->vertical[1].skip_txfm[0] = x->skip_txfm[0];
-      pc_tree->vertical[1].skip = x->skip;
-
-      if (this_rdc.rate == INT_MAX) {
-        vp10_rd_cost_reset(&sum_rdc);
-      } else {
-        int pl = partition_plane_context(xd, mi_row, mi_col, bsize);
-        sum_rdc.rate += cpi->partition_cost[pl][PARTITION_VERT];
-        sum_rdc.rate += this_rdc.rate;
-        sum_rdc.dist += this_rdc.dist;
-        sum_rdc.rdcost = RDCOST(x->rdmult, x->rddiv,
-                                sum_rdc.rate, sum_rdc.dist);
-      }
-    }
-
-    if (sum_rdc.rdcost < best_rdc.rdcost) {
-      best_rdc = sum_rdc;
-      pc_tree->partitioning = PARTITION_VERT;
-    } else {
-      pred_pixel_ready_reset(pc_tree, bsize);
-    }
-  }
-
-  *rd_cost = best_rdc;
-
-  if (best_rdc.rate == INT_MAX) {
-    vp10_rd_cost_reset(rd_cost);
-    return;
-  }
-
-  // update mode info array
-  fill_mode_info_sb(cm, x, mi_row, mi_col, bsize, pc_tree);
-
-  if (best_rdc.rate < INT_MAX && best_rdc.dist < INT64_MAX && do_recon) {
-    int output_enabled = (bsize == BLOCK_64X64);
-    encode_sb_rt(cpi, td, tile_info, tp, mi_row, mi_col, output_enabled,
-                 bsize, pc_tree);
-  }
-
-  if (bsize == BLOCK_64X64 && do_recon) {
-    assert(tp_orig < *tp);
-    assert(best_rdc.rate < INT_MAX);
-    assert(best_rdc.dist < INT64_MAX);
-  } else {
-    assert(tp_orig == *tp);
-  }
-}
-
-static void nonrd_select_partition(VP10_COMP *cpi,
-                                   ThreadData *td,
-                                   TileDataEnc *tile_data,
-                                   MODE_INFO **mi,
-                                   TOKENEXTRA **tp,
-                                   int mi_row, int mi_col,
-                                   BLOCK_SIZE bsize, int output_enabled,
-                                   RD_COST *rd_cost, PC_TREE *pc_tree) {
-  VP10_COMMON *const cm = &cpi->common;
-  TileInfo *const tile_info = &tile_data->tile_info;
-  MACROBLOCK *const x = &td->mb;
-  MACROBLOCKD *const xd = &x->e_mbd;
-  const int bsl = b_width_log2_lookup[bsize], hbs = (1 << bsl) / 4;
-  const int mis = cm->mi_stride;
-  PARTITION_TYPE partition;
-  BLOCK_SIZE subsize;
-  RD_COST this_rdc;
-
-  vp10_rd_cost_reset(&this_rdc);
-  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
-    return;
-
-  subsize = (bsize >= BLOCK_8X8) ? mi[0]->mbmi.sb_type : BLOCK_4X4;
-  partition = partition_lookup[bsl][subsize];
-
-  if (bsize == BLOCK_32X32 && subsize == BLOCK_32X32) {
-    x->max_partition_size = BLOCK_32X32;
-    x->min_partition_size = BLOCK_16X16;
-    nonrd_pick_partition(cpi, td, tile_data, tp, mi_row, mi_col, bsize,
-                         rd_cost, 0, INT64_MAX, pc_tree);
-  } else if (bsize == BLOCK_32X32 && partition != PARTITION_NONE &&
-             subsize >= BLOCK_16X16) {
-    x->max_partition_size = BLOCK_32X32;
-    x->min_partition_size = BLOCK_8X8;
-    nonrd_pick_partition(cpi, td, tile_data, tp, mi_row, mi_col, bsize,
-                         rd_cost, 0, INT64_MAX, pc_tree);
-  } else if (bsize == BLOCK_16X16 && partition != PARTITION_NONE) {
-    x->max_partition_size = BLOCK_16X16;
-    x->min_partition_size = BLOCK_8X8;
-    nonrd_pick_partition(cpi, td, tile_data, tp, mi_row, mi_col, bsize,
-                         rd_cost, 0, INT64_MAX, pc_tree);
-  } else {
-    switch (partition) {
-      case PARTITION_NONE:
-        pc_tree->none.pred_pixel_ready = 1;
-        nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, rd_cost,
-                            subsize, &pc_tree->none);
-        pc_tree->none.mic.mbmi = xd->mi[0]->mbmi;
-        pc_tree->none.mbmi_ext = *x->mbmi_ext;
-        pc_tree->none.skip_txfm[0] = x->skip_txfm[0];
-        pc_tree->none.skip = x->skip;
-        break;
-      case PARTITION_VERT:
-        pc_tree->vertical[0].pred_pixel_ready = 1;
-        nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, rd_cost,
-                            subsize, &pc_tree->vertical[0]);
-        pc_tree->vertical[0].mic.mbmi = xd->mi[0]->mbmi;
-        pc_tree->vertical[0].mbmi_ext = *x->mbmi_ext;
-        pc_tree->vertical[0].skip_txfm[0] = x->skip_txfm[0];
-        pc_tree->vertical[0].skip = x->skip;
-        if (mi_col + hbs < cm->mi_cols) {
-          pc_tree->vertical[1].pred_pixel_ready = 1;
-          nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col + hbs,
-                              &this_rdc, subsize, &pc_tree->vertical[1]);
-          pc_tree->vertical[1].mic.mbmi = xd->mi[0]->mbmi;
-          pc_tree->vertical[1].mbmi_ext = *x->mbmi_ext;
-          pc_tree->vertical[1].skip_txfm[0] = x->skip_txfm[0];
-          pc_tree->vertical[1].skip = x->skip;
-          if (this_rdc.rate != INT_MAX && this_rdc.dist != INT64_MAX &&
-              rd_cost->rate != INT_MAX && rd_cost->dist != INT64_MAX) {
-            rd_cost->rate += this_rdc.rate;
-            rd_cost->dist += this_rdc.dist;
-          }
-        }
-        break;
-      case PARTITION_HORZ:
-        pc_tree->horizontal[0].pred_pixel_ready = 1;
-        nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, rd_cost,
-                            subsize, &pc_tree->horizontal[0]);
-        pc_tree->horizontal[0].mic.mbmi = xd->mi[0]->mbmi;
-        pc_tree->horizontal[0].mbmi_ext = *x->mbmi_ext;
-        pc_tree->horizontal[0].skip_txfm[0] = x->skip_txfm[0];
-        pc_tree->horizontal[0].skip = x->skip;
-        if (mi_row + hbs < cm->mi_rows) {
-          pc_tree->horizontal[1].pred_pixel_ready = 1;
-          nonrd_pick_sb_modes(cpi, tile_data, x, mi_row + hbs, mi_col,
-                              &this_rdc, subsize, &pc_tree->horizontal[1]);
-          pc_tree->horizontal[1].mic.mbmi = xd->mi[0]->mbmi;
-          pc_tree->horizontal[1].mbmi_ext = *x->mbmi_ext;
-          pc_tree->horizontal[1].skip_txfm[0] = x->skip_txfm[0];
-          pc_tree->horizontal[1].skip = x->skip;
-          if (this_rdc.rate != INT_MAX && this_rdc.dist != INT64_MAX &&
-              rd_cost->rate != INT_MAX && rd_cost->dist != INT64_MAX) {
-            rd_cost->rate += this_rdc.rate;
-            rd_cost->dist += this_rdc.dist;
-          }
-        }
-        break;
-      case PARTITION_SPLIT:
-        subsize = get_subsize(bsize, PARTITION_SPLIT);
-        nonrd_select_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col,
-                               subsize, output_enabled, rd_cost,
-                               pc_tree->split[0]);
-        nonrd_select_partition(cpi, td, tile_data, mi + hbs, tp,
-                               mi_row, mi_col + hbs, subsize, output_enabled,
-                               &this_rdc, pc_tree->split[1]);
-        if (this_rdc.rate != INT_MAX && this_rdc.dist != INT64_MAX &&
-            rd_cost->rate != INT_MAX && rd_cost->dist != INT64_MAX) {
-          rd_cost->rate += this_rdc.rate;
-          rd_cost->dist += this_rdc.dist;
-        }
-        nonrd_select_partition(cpi, td, tile_data, mi + hbs * mis, tp,
-                               mi_row + hbs, mi_col, subsize, output_enabled,
-                               &this_rdc, pc_tree->split[2]);
-        if (this_rdc.rate != INT_MAX && this_rdc.dist != INT64_MAX &&
-            rd_cost->rate != INT_MAX && rd_cost->dist != INT64_MAX) {
-          rd_cost->rate += this_rdc.rate;
-          rd_cost->dist += this_rdc.dist;
-        }
-        nonrd_select_partition(cpi, td, tile_data, mi + hbs * mis + hbs, tp,
-                               mi_row + hbs, mi_col + hbs, subsize,
-                               output_enabled, &this_rdc, pc_tree->split[3]);
-        if (this_rdc.rate != INT_MAX && this_rdc.dist != INT64_MAX &&
-            rd_cost->rate != INT_MAX && rd_cost->dist != INT64_MAX) {
-          rd_cost->rate += this_rdc.rate;
-          rd_cost->dist += this_rdc.dist;
-        }
-        break;
-      default:
-        assert(0 && "Invalid partition type.");
-        break;
-    }
-  }
-
-  if (bsize == BLOCK_64X64 && output_enabled)
-    encode_sb_rt(cpi, td, tile_info, tp, mi_row, mi_col, 1, bsize, pc_tree);
-}
-
-
-static void nonrd_use_partition(VP10_COMP *cpi,
-                                ThreadData *td,
-                                TileDataEnc *tile_data,
-                                MODE_INFO **mi,
-                                TOKENEXTRA **tp,
-                                int mi_row, int mi_col,
-                                BLOCK_SIZE bsize, int output_enabled,
-                                RD_COST *dummy_cost, PC_TREE *pc_tree) {
-  VP10_COMMON *const cm = &cpi->common;
-  TileInfo *tile_info = &tile_data->tile_info;
-  MACROBLOCK *const x = &td->mb;
-  MACROBLOCKD *const xd = &x->e_mbd;
-  const int bsl = b_width_log2_lookup[bsize], hbs = (1 << bsl) / 4;
-  const int mis = cm->mi_stride;
-  PARTITION_TYPE partition;
-  BLOCK_SIZE subsize;
-
-  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
-    return;
-
-  subsize = (bsize >= BLOCK_8X8) ? mi[0]->mbmi.sb_type : BLOCK_4X4;
-  partition = partition_lookup[bsl][subsize];
-
-  if (output_enabled && bsize != BLOCK_4X4) {
-    int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
-    td->counts->partition[ctx][partition]++;
-  }
-
-  switch (partition) {
-    case PARTITION_NONE:
-      pc_tree->none.pred_pixel_ready = 1;
-      nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, dummy_cost,
-                          subsize, &pc_tree->none);
-      pc_tree->none.mic.mbmi = xd->mi[0]->mbmi;
-      pc_tree->none.mbmi_ext = *x->mbmi_ext;
-      pc_tree->none.skip_txfm[0] = x->skip_txfm[0];
-      pc_tree->none.skip = x->skip;
-      encode_b_rt(cpi, td, tile_info, tp, mi_row, mi_col, output_enabled,
-                  subsize, &pc_tree->none);
-      break;
-    case PARTITION_VERT:
-      pc_tree->vertical[0].pred_pixel_ready = 1;
-      nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, dummy_cost,
-                          subsize, &pc_tree->vertical[0]);
-      pc_tree->vertical[0].mic.mbmi = xd->mi[0]->mbmi;
-      pc_tree->vertical[0].mbmi_ext = *x->mbmi_ext;
-      pc_tree->vertical[0].skip_txfm[0] = x->skip_txfm[0];
-      pc_tree->vertical[0].skip = x->skip;
-      encode_b_rt(cpi, td, tile_info, tp, mi_row, mi_col, output_enabled,
-                  subsize, &pc_tree->vertical[0]);
-      if (mi_col + hbs < cm->mi_cols && bsize > BLOCK_8X8) {
-        pc_tree->vertical[1].pred_pixel_ready = 1;
-        nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col + hbs,
-                            dummy_cost, subsize, &pc_tree->vertical[1]);
-        pc_tree->vertical[1].mic.mbmi = xd->mi[0]->mbmi;
-        pc_tree->vertical[1].mbmi_ext = *x->mbmi_ext;
-        pc_tree->vertical[1].skip_txfm[0] = x->skip_txfm[0];
-        pc_tree->vertical[1].skip = x->skip;
-        encode_b_rt(cpi, td, tile_info, tp, mi_row, mi_col + hbs,
-                    output_enabled, subsize, &pc_tree->vertical[1]);
-      }
-      break;
-    case PARTITION_HORZ:
-      pc_tree->horizontal[0].pred_pixel_ready = 1;
-      nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, dummy_cost,
-                          subsize, &pc_tree->horizontal[0]);
-      pc_tree->horizontal[0].mic.mbmi = xd->mi[0]->mbmi;
-      pc_tree->horizontal[0].mbmi_ext = *x->mbmi_ext;
-      pc_tree->horizontal[0].skip_txfm[0] = x->skip_txfm[0];
-      pc_tree->horizontal[0].skip = x->skip;
-      encode_b_rt(cpi, td, tile_info, tp, mi_row, mi_col, output_enabled,
-                  subsize, &pc_tree->horizontal[0]);
-
-      if (mi_row + hbs < cm->mi_rows && bsize > BLOCK_8X8) {
-        pc_tree->horizontal[1].pred_pixel_ready = 1;
-        nonrd_pick_sb_modes(cpi, tile_data, x, mi_row + hbs, mi_col,
-                            dummy_cost, subsize, &pc_tree->horizontal[1]);
-        pc_tree->horizontal[1].mic.mbmi = xd->mi[0]->mbmi;
-        pc_tree->horizontal[1].mbmi_ext = *x->mbmi_ext;
-        pc_tree->horizontal[1].skip_txfm[0] = x->skip_txfm[0];
-        pc_tree->horizontal[1].skip = x->skip;
-        encode_b_rt(cpi, td, tile_info, tp, mi_row + hbs, mi_col,
-                    output_enabled, subsize, &pc_tree->horizontal[1]);
-      }
-      break;
-    case PARTITION_SPLIT:
-      subsize = get_subsize(bsize, PARTITION_SPLIT);
-      if (bsize == BLOCK_8X8) {
-        nonrd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, dummy_cost,
-                            subsize, pc_tree->leaf_split[0]);
-        encode_b_rt(cpi, td, tile_info, tp, mi_row, mi_col,
-                    output_enabled, subsize, pc_tree->leaf_split[0]);
-      } else {
-        nonrd_use_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col,
-                            subsize, output_enabled, dummy_cost,
-                            pc_tree->split[0]);
-        nonrd_use_partition(cpi, td, tile_data, mi + hbs, tp,
-                            mi_row, mi_col + hbs, subsize, output_enabled,
-                            dummy_cost, pc_tree->split[1]);
-        nonrd_use_partition(cpi, td, tile_data, mi + hbs * mis, tp,
-                            mi_row + hbs, mi_col, subsize, output_enabled,
-                            dummy_cost, pc_tree->split[2]);
-        nonrd_use_partition(cpi, td, tile_data, mi + hbs * mis + hbs, tp,
-                            mi_row + hbs, mi_col + hbs, subsize, output_enabled,
-                            dummy_cost, pc_tree->split[3]);
-      }
-      break;
-    default:
-      assert(0 && "Invalid partition type.");
-      break;
-  }
-
-  if (partition != PARTITION_SPLIT || bsize == BLOCK_8X8)
-    update_partition_context(xd, mi_row, mi_col, subsize, bsize);
-}
-
-static void encode_nonrd_sb_row(VP10_COMP *cpi,
-                                ThreadData *td,
-                                TileDataEnc *tile_data,
-                                int mi_row,
-                                TOKENEXTRA **tp) {
-  SPEED_FEATURES *const sf = &cpi->sf;
-  VP10_COMMON *const cm = &cpi->common;
-  TileInfo *const tile_info = &tile_data->tile_info;
-  MACROBLOCK *const x = &td->mb;
-  MACROBLOCKD *const xd = &x->e_mbd;
-  int mi_col;
-
-  // Initialize the left context for the new SB row
-  memset(&xd->left_context, 0, sizeof(xd->left_context));
-  memset(xd->left_seg_context, 0, sizeof(xd->left_seg_context));
-
-  // Code each SB in the row
-  for (mi_col = tile_info->mi_col_start; mi_col < tile_info->mi_col_end;
-       mi_col += MI_BLOCK_SIZE) {
-    const struct segmentation *const seg = &cm->seg;
-    RD_COST dummy_rdc;
-    const int idx_str = cm->mi_stride * mi_row + mi_col;
-    MODE_INFO **mi = cm->mi_grid_visible + idx_str;
-    PARTITION_SEARCH_TYPE partition_search_type = sf->partition_search_type;
-    BLOCK_SIZE bsize = BLOCK_64X64;
-    int seg_skip = 0;
-    x->source_variance = UINT_MAX;
-    vp10_zero(x->pred_mv);
-    vp10_rd_cost_init(&dummy_rdc);
-    x->color_sensitivity[0] = 0;
-    x->color_sensitivity[1] = 0;
-
-    if (seg->enabled) {
-      const uint8_t *const map = seg->update_map ? cpi->segmentation_map
-                                                 : cm->last_frame_seg_map;
-      int segment_id = get_segment_id(cm, map, BLOCK_64X64, mi_row, mi_col);
-      seg_skip = segfeature_active(seg, segment_id, SEG_LVL_SKIP);
-      if (seg_skip) {
-        partition_search_type = FIXED_PARTITION;
-      }
-    }
-
-    // Set the partition type of the 64X64 block
-    switch (partition_search_type) {
-      case VAR_BASED_PARTITION:
-        // TODO(jingning, marpan): The mode decision and encoding process
-        // support both intra and inter sub8x8 block coding for RTC mode.
-        // Tune the thresholds accordingly to use sub8x8 block coding for
-        // coding performance improvement.
-        choose_partitioning(cpi, tile_info, x, mi_row, mi_col);
-        nonrd_use_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col,
-                            BLOCK_64X64, 1, &dummy_rdc, td->pc_root);
-        break;
-      case SOURCE_VAR_BASED_PARTITION:
-        set_source_var_based_partition(cpi, tile_info, x, mi, mi_row, mi_col);
-        nonrd_use_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col,
-                            BLOCK_64X64, 1, &dummy_rdc, td->pc_root);
-        break;
-      case FIXED_PARTITION:
-        if (!seg_skip)
-          bsize = sf->always_this_block_size;
-        set_fixed_partitioning(cpi, tile_info, mi, mi_row, mi_col, bsize);
-        nonrd_use_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col,
-                            BLOCK_64X64, 1, &dummy_rdc, td->pc_root);
-        break;
-      case REFERENCE_PARTITION:
-        set_offsets(cpi, tile_info, x, mi_row, mi_col, BLOCK_64X64);
-        if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && cm->seg.enabled &&
-            xd->mi[0]->mbmi.segment_id) {
-          // Use lower max_partition_size for low resoultions.
-          if (cm->width <= 352 && cm->height <= 288)
-            x->max_partition_size = BLOCK_32X32;
-          else
-            x->max_partition_size = BLOCK_64X64;
-          x->min_partition_size = BLOCK_8X8;
-          nonrd_pick_partition(cpi, td, tile_data, tp, mi_row, mi_col,
-                               BLOCK_64X64, &dummy_rdc, 1,
-                               INT64_MAX, td->pc_root);
-        } else {
-          choose_partitioning(cpi, tile_info, x, mi_row, mi_col);
-          // TODO(marpan): Seems like nonrd_select_partition does not support
-          // 4x4 partition. Since 4x4 is used on key frame, use this switch
-          // for now.
-          if (cm->frame_type == KEY_FRAME)
-            nonrd_use_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col,
-                                BLOCK_64X64, 1, &dummy_rdc, td->pc_root);
-          else
-            nonrd_select_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col,
-                                   BLOCK_64X64, 1, &dummy_rdc, td->pc_root);
-        }
-
-        break;
-      default:
-        assert(0);
-        break;
-    }
-  }
-}
-// end RTC play code
-
-static int set_var_thresh_from_histogram(VP10_COMP *cpi) {
-  const SPEED_FEATURES *const sf = &cpi->sf;
-  const VP10_COMMON *const cm = &cpi->common;
-
-  const uint8_t *src = cpi->Source->y_buffer;
-  const uint8_t *last_src = cpi->Last_Source->y_buffer;
-  const int src_stride = cpi->Source->y_stride;
-  const int last_stride = cpi->Last_Source->y_stride;
-
-  // Pick cutoff threshold
-  const int cutoff = (MIN(cm->width, cm->height) >= 720) ?
-      (cm->MBs * VAR_HIST_LARGE_CUT_OFF / 100) :
-      (cm->MBs * VAR_HIST_SMALL_CUT_OFF / 100);
-  DECLARE_ALIGNED(16, int, hist[VAR_HIST_BINS]);
-  diff *var16 = cpi->source_diff_var;
-
-  int sum = 0;
-  int i, j;
-
-  memset(hist, 0, VAR_HIST_BINS * sizeof(hist[0]));
-
-  for (i = 0; i < cm->mb_rows; i++) {
-    for (j = 0; j < cm->mb_cols; j++) {
-#if CONFIG_VP9_HIGHBITDEPTH
-      if (cm->use_highbitdepth) {
-        switch (cm->bit_depth) {
-          case VPX_BITS_8:
-            vpx_highbd_8_get16x16var(src, src_stride, last_src, last_stride,
-                                   &var16->sse, &var16->sum);
-            break;
-          case VPX_BITS_10:
-            vpx_highbd_10_get16x16var(src, src_stride, last_src, last_stride,
-                                    &var16->sse, &var16->sum);
-            break;
-          case VPX_BITS_12:
-            vpx_highbd_12_get16x16var(src, src_stride, last_src, last_stride,
-                                      &var16->sse, &var16->sum);
-            break;
-          default:
-            assert(0 && "cm->bit_depth should be VPX_BITS_8, VPX_BITS_10"
-                   " or VPX_BITS_12");
-            return -1;
-        }
-      } else {
-        vpx_get16x16var(src, src_stride, last_src, last_stride,
-                        &var16->sse, &var16->sum);
-      }
-#else
-      vpx_get16x16var(src, src_stride, last_src, last_stride,
-                      &var16->sse, &var16->sum);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-      var16->var = var16->sse -
-          (((uint32_t)var16->sum * var16->sum) >> 8);
-
-      if (var16->var >= VAR_HIST_MAX_BG_VAR)
-        hist[VAR_HIST_BINS - 1]++;
-      else
-        hist[var16->var / VAR_HIST_FACTOR]++;
-
-      src += 16;
-      last_src += 16;
-      var16++;
-    }
-
-    src = src - cm->mb_cols * 16 + 16 * src_stride;
-    last_src = last_src - cm->mb_cols * 16 + 16 * last_stride;
-  }
-
-  cpi->source_var_thresh = 0;
-
-  if (hist[VAR_HIST_BINS - 1] < cutoff) {
-    for (i = 0; i < VAR_HIST_BINS - 1; i++) {
-      sum += hist[i];
-
-      if (sum > cutoff) {
-        cpi->source_var_thresh = (i + 1) * VAR_HIST_FACTOR;
-        return 0;
-      }
-    }
-  }
-
-  return sf->search_type_check_frequency;
-}
-
-static void source_var_based_partition_search_method(VP10_COMP *cpi) {
-  VP10_COMMON *const cm = &cpi->common;
-  SPEED_FEATURES *const sf = &cpi->sf;
-
-  if (cm->frame_type == KEY_FRAME) {
-    // For key frame, use SEARCH_PARTITION.
-    sf->partition_search_type = SEARCH_PARTITION;
-  } else if (cm->intra_only) {
-    sf->partition_search_type = FIXED_PARTITION;
-  } else {
-    if (cm->last_width != cm->width || cm->last_height != cm->height) {
-      if (cpi->source_diff_var)
-        vpx_free(cpi->source_diff_var);
-
-      CHECK_MEM_ERROR(cm, cpi->source_diff_var,
-                      vpx_calloc(cm->MBs, sizeof(diff)));
-    }
-
-    if (!cpi->frames_till_next_var_check)
-      cpi->frames_till_next_var_check = set_var_thresh_from_histogram(cpi);
-
-    if (cpi->frames_till_next_var_check > 0) {
-      sf->partition_search_type = FIXED_PARTITION;
-      cpi->frames_till_next_var_check--;
-    }
-  }
-}
-
 static int get_skip_encode_frame(const VP10_COMMON *cm, ThreadData *const td) {
   unsigned int intra_count = 0, inter_count = 0;
   int j;
@@ -3816,10 +2783,7 @@
 
   for (mi_row = tile_info->mi_row_start; mi_row < tile_info->mi_row_end;
        mi_row += MI_BLOCK_SIZE) {
-    if (cpi->sf.use_nonrd_pick_mode)
-      encode_nonrd_sb_row(cpi, td, this_tile, mi_row, &tok);
-    else
-      encode_rd_sb_row(cpi, td, this_tile, mi_row, &tok);
+    encode_rd_sb_row(cpi, td, this_tile, mi_row, &tok);
   }
   cpi->tok_count[tile_row][tile_col] =
       (unsigned int)(tok - cpi->tile_tok[tile_row][tile_col]);
@@ -3910,28 +2874,6 @@
 
   x->quant_fp = cpi->sf.use_quant_fp;
   vp10_zero(x->skip_txfm);
-  if (sf->use_nonrd_pick_mode) {
-    // Initialize internal buffer pointers for rtc coding, where non-RD
-    // mode decision is used and hence no buffer pointer swap needed.
-    int i;
-    struct macroblock_plane *const p = x->plane;
-    struct macroblockd_plane *const pd = xd->plane;
-    PICK_MODE_CONTEXT *ctx = &cpi->td.pc_root->none;
-
-    for (i = 0; i < MAX_MB_PLANE; ++i) {
-      p[i].coeff = ctx->coeff_pbuf[i][0];
-      p[i].qcoeff = ctx->qcoeff_pbuf[i][0];
-      pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][0];
-      p[i].eobs = ctx->eobs_pbuf[i][0];
-    }
-    vp10_zero(x->zcoeff_blk);
-
-    if (cm->frame_type != KEY_FRAME && cpi->rc.frames_since_golden == 0)
-      cpi->ref_frame_flags &= (~VP9_GOLD_FLAG);
-
-    if (sf->partition_search_type == SOURCE_VAR_BASED_PARTITION)
-      source_var_based_partition_search_method(cpi);
-  }
 
   {
     struct vpx_usec_timer emr_timer;
@@ -3945,7 +2887,7 @@
 #endif
 
     // If allowed, encoding tiles in parallel with one thread handling one tile.
-    if (MIN(cpi->oxcf.max_threads, 1 << cm->log2_tile_cols) > 1)
+    if (VPXMIN(cpi->oxcf.max_threads, 1 << cm->log2_tile_cols) > 1)
       vp10_encode_tiles_mt(cpi);
     else
       encode_tiles(cpi);
@@ -4144,7 +3086,7 @@
                    cpi->oxcf.aq_mode != CYCLIC_REFRESH_AQ &&
                    cpi->sf.allow_skip_recode;
 
-  if (!x->skip_recode && !cpi->sf.use_nonrd_pick_mode)
+  if (!x->skip_recode)
     memset(x->skip_txfm, 0, sizeof(x->skip_txfm));
 
   x->skip_optimize = ctx->is_coded;
@@ -4160,10 +3102,10 @@
     int plane;
     mbmi->skip = 1;
     for (plane = 0; plane < MAX_MB_PLANE; ++plane)
-      vp10_encode_intra_block_plane(x, MAX(bsize, BLOCK_8X8), plane);
+      vp10_encode_intra_block_plane(x, VPXMAX(bsize, BLOCK_8X8), plane);
     if (output_enabled)
       sum_intra_stats(td->counts, mi);
-    vp10_tokenize_sb(cpi, td, t, !output_enabled, MAX(bsize, BLOCK_8X8));
+    vp10_tokenize_sb(cpi, td, t, !output_enabled, VPXMAX(bsize, BLOCK_8X8));
   } else {
     int ref;
     const int is_compound = has_second_ref(mbmi);
@@ -4176,12 +3118,14 @@
                            &xd->block_refs[ref]->sf);
     }
     if (!(cpi->sf.reuse_inter_pred_sby && ctx->pred_pixel_ready) || seg_skip)
-      vp10_build_inter_predictors_sby(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
+      vp10_build_inter_predictors_sby(xd, mi_row, mi_col,
+                                      VPXMAX(bsize, BLOCK_8X8));
 
-    vp10_build_inter_predictors_sbuv(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
+    vp10_build_inter_predictors_sbuv(xd, mi_row, mi_col,
+                                     VPXMAX(bsize, BLOCK_8X8));
 
-    vp10_encode_sb(x, MAX(bsize, BLOCK_8X8));
-    vp10_tokenize_sb(cpi, td, t, !output_enabled, MAX(bsize, BLOCK_8X8));
+    vp10_encode_sb(x, VPXMAX(bsize, BLOCK_8X8));
+    vp10_tokenize_sb(cpi, td, t, !output_enabled, VPXMAX(bsize, BLOCK_8X8));
   }
 
   if (output_enabled) {
@@ -4195,8 +3139,8 @@
       TX_SIZE tx_size;
       // The new intra coding scheme requires no change of transform size
       if (is_inter_block(&mi->mbmi)) {
-        tx_size = MIN(tx_mode_to_biggest_tx_size[cm->tx_mode],
-                      max_txsize_lookup[bsize]);
+        tx_size = VPXMIN(tx_mode_to_biggest_tx_size[cm->tx_mode],
+                         max_txsize_lookup[bsize]);
       } else {
         tx_size = (bsize >= BLOCK_8X8) ? mbmi->tx_size : TX_4X4;
       }
diff --git a/vp10/encoder/encodemv.c b/vp10/encoder/encodemv.c
index b67b93e..ecd5a38 100644
--- a/vp10/encoder/encodemv.c
+++ b/vp10/encoder/encodemv.c
@@ -216,8 +216,8 @@
   // If auto_mv_step_size is enabled then keep track of the largest
   // motion vector component used.
   if (cpi->sf.mv.auto_mv_step_size) {
-    unsigned int maxv = MAX(abs(mv->row), abs(mv->col)) >> 3;
-    cpi->max_mv_magnitude = MAX(maxv, cpi->max_mv_magnitude);
+    unsigned int maxv = VPXMAX(abs(mv->row), abs(mv->col)) >> 3;
+    cpi->max_mv_magnitude = VPXMAX(maxv, cpi->max_mv_magnitude);
   }
 }
 
diff --git a/vp10/encoder/encoder.c b/vp10/encoder/encoder.c
index f5cb0a6..33f060f 100644
--- a/vp10/encoder/encoder.c
+++ b/vp10/encoder/encoder.c
@@ -1492,8 +1492,8 @@
 
   // Under a configuration change, where maximum_buffer_size may change,
   // keep buffer level clipped to the maximum allowed buffer size.
-  rc->bits_off_target = MIN(rc->bits_off_target, rc->maximum_buffer_size);
-  rc->buffer_level = MIN(rc->buffer_level, rc->maximum_buffer_size);
+  rc->bits_off_target = VPXMIN(rc->bits_off_target, rc->maximum_buffer_size);
+  rc->buffer_level = VPXMIN(rc->buffer_level, rc->maximum_buffer_size);
 
   // Set up frame rate and related parameters rate control values.
   vp10_new_framerate(cpi, cpi->framerate);
@@ -2615,7 +2615,7 @@
   if (rc->frame_size_selector == UNSCALED &&
       q >= rc->rf_level_maxq[gf_group->rf_level[gf_group->index]]) {
     const int max_size_thresh = (int)(rate_thresh_mult[SCALE_STEP1]
-        * MAX(rc->this_frame_target, rc->avg_frame_bandwidth));
+        * VPXMAX(rc->this_frame_target, rc->avg_frame_bandwidth));
     scale = rc->projected_frame_size > max_size_thresh ? 1 : 0;
   }
   return scale;
@@ -2998,7 +2998,7 @@
 
 static void set_mv_search_params(VP10_COMP *cpi) {
   const VP10_COMMON *const cm = &cpi->common;
-  const unsigned int max_mv_def = MIN(cm->width, cm->height);
+  const unsigned int max_mv_def = VPXMIN(cm->width, cm->height);
 
   // Default based on max resolution.
   cpi->mv_step_param = vp10_init_search_range(max_mv_def);
@@ -3013,8 +3013,8 @@
         // Allow mv_steps to correspond to twice the max mv magnitude found
         // in the previous frame, capped by the default max_mv_magnitude based
         // on resolution.
-        cpi->mv_step_param =
-            vp10_init_search_range(MIN(max_mv_def, 2 * cpi->max_mv_magnitude));
+        cpi->mv_step_param = vp10_init_search_range(
+            VPXMIN(max_mv_def, 2 * cpi->max_mv_magnitude));
       }
       cpi->max_mv_magnitude = 0;
     }
@@ -3341,8 +3341,7 @@
     // to recode.
     if (cpi->sf.recode_loop >= ALLOW_RECODE_KFARFGF) {
       save_coding_context(cpi);
-      if (!cpi->sf.use_nonrd_pick_mode)
-        vp10_pack_bitstream(cpi, dest, size);
+      vp10_pack_bitstream(cpi, dest, size);
 
       rc->projected_frame_size = (int)(*size) << 3;
       restore_coding_context(cpi);
@@ -3387,7 +3386,7 @@
 
           // Adjust Q
           q = (int)((q * high_err_target) / kf_err);
-          q = MIN(q, (q_high + q_low) >> 1);
+          q = VPXMIN(q, (q_high + q_low) >> 1);
         } else if (kf_err < low_err_target &&
                    rc->projected_frame_size >= frame_under_shoot_limit) {
           // The key frame is much better than the previous frame
@@ -3396,7 +3395,7 @@
 
           // Adjust Q
           q = (int)((q * low_err_target) / kf_err);
-          q = MIN(q, (q_high + q_low + 1) >> 1);
+          q = VPXMIN(q, (q_high + q_low + 1) >> 1);
         }
 
         // Clamp Q to upper and lower limits:
@@ -3405,7 +3404,7 @@
         loop = q != last_q;
       } else if (recode_loop_test(
           cpi, frame_over_shoot_limit, frame_under_shoot_limit,
-          q, MAX(q_high, top_index), bottom_index)) {
+          q, VPXMAX(q_high, top_index), bottom_index)) {
         // Is the projected frame size out of range and are we allowed
         // to attempt to recode.
         int last_q = q;
@@ -3447,12 +3446,12 @@
             vp10_rc_update_rate_correction_factors(cpi);
 
             q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
-                                   bottom_index, MAX(q_high, top_index));
+                                   bottom_index, VPXMAX(q_high, top_index));
 
             while (q < q_low && retries < 10) {
               vp10_rc_update_rate_correction_factors(cpi);
               q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
-                                     bottom_index, MAX(q_high, top_index));
+                                     bottom_index, VPXMAX(q_high, top_index));
               retries++;
             }
           }
@@ -4031,8 +4030,8 @@
       // Average this frame's rate into the last second's average
       // frame rate. If we haven't seen 1 second yet, then average
       // over the whole interval seen.
-      const double interval = MIN((double)(source->ts_end
-                                   - cpi->first_time_stamp_ever), 10000000.0);
+      const double interval = VPXMIN(
+          (double)(source->ts_end - cpi->first_time_stamp_ever), 10000000.0);
       double avg_duration = 10000000.0 / cpi->framerate;
       avg_duration *= (interval - avg_duration + this_duration);
       avg_duration /= interval;
@@ -4096,7 +4095,7 @@
   s->stat[U] += u;
   s->stat[V] += v;
   s->stat[ALL] += all;
-  s->worst = MIN(s->worst, all);
+  s->worst = VPXMIN(s->worst, all);
 }
 #endif  // CONFIG_INTERNAL_STATS
 
@@ -4426,7 +4425,7 @@
           frame_ssim2 = vpx_calc_ssim(orig, recon, &weight);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
-          cpi->worst_ssim= MIN(cpi->worst_ssim, frame_ssim2);
+          cpi->worst_ssim= VPXMIN(cpi->worst_ssim, frame_ssim2);
           cpi->summed_quality += frame_ssim2 * weight;
           cpi->summed_weights += weight;
 
@@ -4463,7 +4462,8 @@
               cpi->Source->y_buffer, cpi->Source->y_stride,
               cm->frame_to_show->y_buffer, cm->frame_to_show->y_stride,
               cpi->Source->y_width, cpi->Source->y_height);
-          cpi->worst_blockiness = MAX(cpi->worst_blockiness, frame_blockiness);
+          cpi->worst_blockiness =
+              VPXMAX(cpi->worst_blockiness, frame_blockiness);
           cpi->total_blockiness += frame_blockiness;
         }
       }
@@ -4483,8 +4483,8 @@
           double consistency = vpx_sse_to_psnr(samples, peak,
                                              (double)cpi->total_inconsistency);
           if (consistency > 0.0)
-            cpi->worst_consistency = MIN(cpi->worst_consistency,
-                                         consistency);
+            cpi->worst_consistency =
+                VPXMIN(cpi->worst_consistency, consistency);
           cpi->total_inconsistency += this_inconsistency;
         }
       }
diff --git a/vp10/encoder/ethread.c b/vp10/encoder/ethread.c
index e1d2538..6ba0572 100644
--- a/vp10/encoder/ethread.c
+++ b/vp10/encoder/ethread.c
@@ -67,7 +67,7 @@
   VP10_COMMON *const cm = &cpi->common;
   const int tile_cols = 1 << cm->log2_tile_cols;
   const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
-  const int num_workers = MIN(cpi->oxcf.max_threads, tile_cols);
+  const int num_workers = VPXMIN(cpi->oxcf.max_threads, tile_cols);
   int i;
 
   vp10_init_tile_data(cpi);
@@ -80,7 +80,7 @@
     // resolution.
     if (cpi->use_svc) {
       int max_tile_cols = get_max_tile_cols(cpi);
-      allocated_workers = MIN(cpi->oxcf.max_threads, max_tile_cols);
+      allocated_workers = VPXMIN(cpi->oxcf.max_threads, max_tile_cols);
     }
 
     CHECK_MEM_ERROR(cm, cpi->workers,
@@ -146,23 +146,6 @@
       memcpy(thread_data->td->counts, &cpi->common.counts,
              sizeof(cpi->common.counts));
     }
-
-    // Handle use_nonrd_pick_mode case.
-    if (cpi->sf.use_nonrd_pick_mode) {
-      MACROBLOCK *const x = &thread_data->td->mb;
-      MACROBLOCKD *const xd = &x->e_mbd;
-      struct macroblock_plane *const p = x->plane;
-      struct macroblockd_plane *const pd = xd->plane;
-      PICK_MODE_CONTEXT *ctx = &thread_data->td->pc_root->none;
-      int j;
-
-      for (j = 0; j < MAX_MB_PLANE; ++j) {
-        p[j].coeff = ctx->coeff_pbuf[j][0];
-        p[j].qcoeff = ctx->qcoeff_pbuf[j][0];
-        pd[j].dqcoeff = ctx->dqcoeff_pbuf[j][0];
-        p[j].eobs = ctx->eobs_pbuf[j][0];
-      }
-    }
   }
 
   // Encode a frame
diff --git a/vp10/encoder/extend.c b/vp10/encoder/extend.c
index 1cac5ac..ffd992f 100644
--- a/vp10/encoder/extend.c
+++ b/vp10/encoder/extend.c
@@ -111,10 +111,12 @@
   // Motion estimation may use src block variance with the block size up
   // to 64x64, so the right and bottom need to be extended to 64 multiple
   // or up to 16, whichever is greater.
-  const int er_y = MAX(src->y_width + 16, ALIGN_POWER_OF_TWO(src->y_width, 6))
-      - src->y_crop_width;
-  const int eb_y = MAX(src->y_height + 16, ALIGN_POWER_OF_TWO(src->y_height, 6))
-      - src->y_crop_height;
+  const int er_y =
+      VPXMAX(src->y_width + 16, ALIGN_POWER_OF_TWO(src->y_width, 6)) -
+      src->y_crop_width;
+  const int eb_y =
+      VPXMAX(src->y_height + 16, ALIGN_POWER_OF_TWO(src->y_height, 6)) -
+      src->y_crop_height;
   const int uv_width_subsampling = (src->uv_width != src->y_width);
   const int uv_height_subsampling = (src->uv_height != src->y_height);
   const int et_uv = et_y >> uv_height_subsampling;
diff --git a/vp10/encoder/firstpass.c b/vp10/encoder/firstpass.c
index 328ef4b..a49468e 100644
--- a/vp10/encoder/firstpass.c
+++ b/vp10/encoder/firstpass.c
@@ -383,7 +383,7 @@
 // for first pass test.
 static int get_search_range(const VP10_COMP *cpi) {
   int sr = 0;
-  const int dim = MIN(cpi->initial_width, cpi->initial_height);
+  const int dim = VPXMIN(cpi->initial_width, cpi->initial_height);
 
   while ((dim << sr) < MAX_FULL_PEL_VAL)
     ++sr;
@@ -1026,7 +1026,7 @@
   // Exclude any image dead zone
   if (image_data_start_row > 0) {
     intra_skip_count =
-      MAX(0, intra_skip_count - (image_data_start_row * cm->mb_cols * 2));
+        VPXMAX(0, intra_skip_count - (image_data_start_row * cm->mb_cols * 2));
   }
 
   {
@@ -1163,7 +1163,7 @@
 
   // Adjustment based on actual quantizer to power term.
   const double power_term =
-      MIN(vp10_convert_qindex_to_q(q, bit_depth) * 0.01 + pt_low, pt_high);
+      VPXMIN(vp10_convert_qindex_to_q(q, bit_depth) * 0.01 + pt_low, pt_high);
 
   // Calculate correction factor.
   if (power_term < 1.0)
@@ -1192,7 +1192,7 @@
   } else {
     const int num_mbs = (cpi->oxcf.resize_mode != RESIZE_NONE)
                         ? cpi->initial_mbs : cpi->common.MBs;
-    const int active_mbs = MAX(1, num_mbs - (int)(num_mbs * inactive_zone));
+    const int active_mbs = VPXMAX(1, num_mbs - (int)(num_mbs * inactive_zone));
     const double av_err_per_mb = section_err / active_mbs;
     const double speed_term = 1.0 + 0.04 * oxcf->speed;
     const double ediv_size_correction = (double)num_mbs / EDIV_SIZE_FACTOR;
@@ -1225,7 +1225,7 @@
 
     // Restriction on active max q for constrained quality mode.
     if (cpi->oxcf.rc_mode == VPX_CQ)
-      q = MAX(q, oxcf->cq_level);
+      q = VPXMAX(q, oxcf->cq_level);
     return q;
   }
 }
@@ -1235,7 +1235,7 @@
   RATE_CONTROL *const rc = &cpi->rc;
   for (i = INTER_NORMAL; i < RATE_FACTOR_LEVELS; ++i) {
     int qdelta = vp10_frame_type_qdelta(cpi, i, rc->worst_quality);
-    rc->rf_level_maxq[i] = MAX(rc->worst_quality + qdelta, rc->best_quality);
+    rc->rf_level_maxq[i] = VPXMAX(rc->worst_quality + qdelta, rc->best_quality);
   }
 }
 
@@ -1366,12 +1366,12 @@
 
 
   if ((sr_diff > LOW_SR_DIFF_TRHESH)) {
-    sr_diff = MIN(sr_diff, SR_DIFF_MAX);
+    sr_diff = VPXMIN(sr_diff, SR_DIFF_MAX);
     sr_decay = 1.0 - (SR_DIFF_PART * sr_diff) -
                (MOTION_AMP_PART * motion_amplitude_factor) -
                (INTRA_PART * modified_pcnt_intra);
   }
-  return MAX(sr_decay, MIN(DEFAULT_DECAY_LIMIT, modified_pct_inter));
+  return VPXMAX(sr_decay, VPXMIN(DEFAULT_DECAY_LIMIT, modified_pct_inter));
 }
 
 // This function gives an estimate of how badly we believe the prediction
@@ -1381,7 +1381,7 @@
   const double zero_motion_pct = frame->pcnt_inter -
                                  frame->pcnt_motion;
   double sr_decay = get_sr_decay_rate(cpi, frame);
-  return MIN(sr_decay, zero_motion_pct);
+  return VPXMIN(sr_decay, zero_motion_pct);
 }
 
 #define ZM_POWER_FACTOR 0.75
@@ -1393,8 +1393,8 @@
     (0.95 * pow((next_frame->pcnt_inter - next_frame->pcnt_motion),
                 ZM_POWER_FACTOR));
 
-  return MAX(zero_motion_factor,
-             (sr_decay_rate + ((1.0 - sr_decay_rate) * zero_motion_factor)));
+  return VPXMAX(zero_motion_factor,
+                (sr_decay_rate + ((1.0 - sr_decay_rate) * zero_motion_factor)));
 }
 
 // Function to test for a condition where a complex transition is followed
@@ -1485,12 +1485,12 @@
   const double lq =
     vp10_convert_qindex_to_q(cpi->rc.avg_frame_qindex[INTER_FRAME],
                             cpi->common.bit_depth);
-  const double boost_q_correction = MIN((0.5 + (lq * 0.015)), 1.5);
+  const double boost_q_correction = VPXMIN((0.5 + (lq * 0.015)), 1.5);
   int num_mbs = (cpi->oxcf.resize_mode != RESIZE_NONE)
                 ? cpi->initial_mbs : cpi->common.MBs;
 
   // Correct for any inactive region in the image
-  num_mbs = (int)MAX(1, num_mbs * calculate_active_area(cpi, this_frame));
+  num_mbs = (int)VPXMAX(1, num_mbs * calculate_active_area(cpi, this_frame));
 
   // Underlying boost factor is based on inter error ratio.
   frame_boost = (BASELINE_ERR_PER_MB * num_mbs) /
@@ -1506,7 +1506,7 @@
   else
     frame_boost += frame_boost * (this_frame_mv_in_out / 2.0);
 
-  return MIN(frame_boost, max_boost * boost_q_correction);
+  return VPXMIN(frame_boost, max_boost * boost_q_correction);
 }
 
 static int calc_arf_boost(VP10_COMP *cpi, int offset,
@@ -1595,7 +1595,7 @@
   arf_boost = (*f_boost + *b_boost);
   if (arf_boost < ((b_frames + f_frames) * 20))
     arf_boost = ((b_frames + f_frames) * 20);
-  arf_boost = MAX(arf_boost, MIN_ARF_GF_BOOST);
+  arf_boost = VPXMAX(arf_boost, MIN_ARF_GF_BOOST);
 
   return arf_boost;
 }
@@ -1666,7 +1666,8 @@
   }
 
   // Calculate the number of extra bits for use in the boosted frame or frames.
-  return MAX((int)(((int64_t)boost * total_group_bits) / allocation_chunks), 0);
+  return VPXMAX((int)(((int64_t)boost * total_group_bits) / allocation_chunks),
+                0);
 }
 
 // Current limit on maximum number of active arfs in a GF/ARF group.
@@ -1805,7 +1806,7 @@
     gf_group->arf_ref_idx[frame_index] = arf_buffer_indices[arf_idx];
 
     target_frame_size = clamp(target_frame_size, 0,
-                              MIN(max_bits, (int)total_group_bits));
+                              VPXMIN(max_bits, (int)total_group_bits));
 
     gf_group->update_type[frame_index] = LF_UPDATE;
     gf_group->rf_level[frame_index] = INTER_NORMAL;
@@ -1926,7 +1927,7 @@
     int int_lbq =
       (int)(vp10_convert_qindex_to_q(rc->last_boosted_qindex,
                                    cpi->common.bit_depth));
-    active_min_gf_interval = rc->min_gf_interval + MIN(2, int_max_q / 200);
+    active_min_gf_interval = rc->min_gf_interval + VPXMIN(2, int_max_q / 200);
     if (active_min_gf_interval > rc->max_gf_interval)
       active_min_gf_interval = rc->max_gf_interval;
 
@@ -1937,7 +1938,7 @@
       // bits to spare and are better with a smaller interval and smaller boost.
       // At high Q when there are few bits to spare we are better with a longer
       // interval to spread the cost of the GF.
-      active_max_gf_interval = 12 + MIN(4, (int_lbq / 6));
+      active_max_gf_interval = 12 + VPXMIN(4, (int_lbq / 6));
       if (active_max_gf_interval < active_min_gf_interval)
         active_max_gf_interval = active_min_gf_interval;
 
@@ -1982,8 +1983,8 @@
       decay_accumulator = decay_accumulator * loop_decay_rate;
 
       // Monitor for static sections.
-      zero_motion_accumulator =
-        MIN(zero_motion_accumulator, get_zero_motion_factor(cpi, &next_frame));
+      zero_motion_accumulator = VPXMIN(
+          zero_motion_accumulator, get_zero_motion_factor(cpi, &next_frame));
 
       // Break clause to detect very still sections after motion. For example,
       // a static image after a fade or other transition.
@@ -2039,7 +2040,7 @@
       (cpi->multi_arf_allowed && (rc->baseline_gf_interval >= 6) &&
       (zero_motion_accumulator < 0.995)) ? 1 : 0;
   } else {
-    rc->gfu_boost = MAX((int)boost_score, MIN_ARF_GF_BOOST);
+    rc->gfu_boost = VPXMAX((int)boost_score, MIN_ARF_GF_BOOST);
     rc->source_alt_ref_pending = 0;
   }
 
@@ -2094,11 +2095,11 @@
     // rc factor is a weight factor that corrects for local rate control drift.
     double rc_factor = 1.0;
     if (rc->rate_error_estimate > 0) {
-      rc_factor = MAX(RC_FACTOR_MIN,
-                      (double)(100 - rc->rate_error_estimate) / 100.0);
+      rc_factor = VPXMAX(RC_FACTOR_MIN,
+                         (double)(100 - rc->rate_error_estimate) / 100.0);
     } else {
-      rc_factor = MIN(RC_FACTOR_MAX,
-                      (double)(100 - rc->rate_error_estimate) / 100.0);
+      rc_factor = VPXMIN(RC_FACTOR_MAX,
+                         (double)(100 - rc->rate_error_estimate) / 100.0);
     }
     tmp_q =
       get_twopass_worst_quality(cpi, group_av_err,
@@ -2106,7 +2107,7 @@
                                 vbr_group_bits_per_frame,
                                 twopass->kfgroup_inter_fraction * rc_factor);
     twopass->active_worst_quality =
-      MAX(tmp_q, twopass->active_worst_quality >> 1);
+      VPXMAX(tmp_q, twopass->active_worst_quality >> 1);
   }
 #endif
 
@@ -2423,7 +2424,7 @@
   } else {
     twopass->kf_group_bits = 0;
   }
-  twopass->kf_group_bits = MAX(0, twopass->kf_group_bits);
+  twopass->kf_group_bits = VPXMAX(0, twopass->kf_group_bits);
 
   // Reset the first pass file position.
   reset_fpf_position(twopass, start_position);
@@ -2437,9 +2438,8 @@
       break;
 
     // Monitor for static sections.
-    zero_motion_accumulator =
-      MIN(zero_motion_accumulator,
-          get_zero_motion_factor(cpi, &next_frame));
+    zero_motion_accumulator = VPXMIN(
+        zero_motion_accumulator, get_zero_motion_factor(cpi, &next_frame));
 
     // Not all frames in the group are necessarily used in calculating boost.
     if ((i <= rc->max_gf_interval) ||
@@ -2452,7 +2452,7 @@
         const double loop_decay_rate =
           get_prediction_decay_rate(cpi, &next_frame);
         decay_accumulator *= loop_decay_rate;
-        decay_accumulator = MAX(decay_accumulator, MIN_DECAY_FACTOR);
+        decay_accumulator = VPXMAX(decay_accumulator, MIN_DECAY_FACTOR);
         av_decay_accumulator += decay_accumulator;
         ++loop_decay_counter;
       }
@@ -2473,8 +2473,8 @@
 
   // Apply various clamps for min and max boost
   rc->kf_boost = (int)(av_decay_accumulator * boost_score);
-  rc->kf_boost = MAX(rc->kf_boost, (rc->frames_to_key * 3));
-  rc->kf_boost = MAX(rc->kf_boost, MIN_KF_BOOST);
+  rc->kf_boost = VPXMAX(rc->kf_boost, (rc->frames_to_key * 3));
+  rc->kf_boost = VPXMAX(rc->kf_boost, MIN_KF_BOOST);
 
   // Work out how many bits to allocate for the key frame itself.
   kf_bits = calculate_boost_bits((rc->frames_to_key - 1),
@@ -2772,7 +2772,7 @@
   // is designed to prevent extreme behaviour at the end of a clip
   // or group of frames.
   rc->vbr_bits_off_target += rc->base_frame_target - rc->projected_frame_size;
-  twopass->bits_left = MAX(twopass->bits_left - bits_used, 0);
+  twopass->bits_left = VPXMAX(twopass->bits_left - bits_used, 0);
 
   // Calculate the pct rc error.
   if (rc->total_actual_bits) {
@@ -2788,7 +2788,7 @@
     twopass->kf_group_bits -= bits_used;
     twopass->last_kfgroup_zeromotion_pct = twopass->kf_zeromotion_pct;
   }
-  twopass->kf_group_bits = MAX(twopass->kf_group_bits, 0);
+  twopass->kf_group_bits = VPXMAX(twopass->kf_group_bits, 0);
 
   // Increment the gf group index ready for the next frame.
   ++twopass->gf_group.index;
@@ -2838,18 +2838,18 @@
         rc->vbr_bits_off_target_fast +=
           fast_extra_thresh - rc->projected_frame_size;
         rc->vbr_bits_off_target_fast =
-          MIN(rc->vbr_bits_off_target_fast, (4 * rc->avg_frame_bandwidth));
+          VPXMIN(rc->vbr_bits_off_target_fast, (4 * rc->avg_frame_bandwidth));
 
         // Fast adaptation of minQ if necessary to use up the extra bits.
         if (rc->avg_frame_bandwidth) {
           twopass->extend_minq_fast =
             (int)(rc->vbr_bits_off_target_fast * 8 / rc->avg_frame_bandwidth);
         }
-        twopass->extend_minq_fast = MIN(twopass->extend_minq_fast,
-                                        minq_adj_limit - twopass->extend_minq);
+        twopass->extend_minq_fast = VPXMIN(
+            twopass->extend_minq_fast, minq_adj_limit - twopass->extend_minq);
       } else if (rc->vbr_bits_off_target_fast) {
-        twopass->extend_minq_fast = MIN(twopass->extend_minq_fast,
-                                        minq_adj_limit - twopass->extend_minq);
+        twopass->extend_minq_fast = VPXMIN(
+            twopass->extend_minq_fast, minq_adj_limit - twopass->extend_minq);
       } else {
         twopass->extend_minq_fast = 0;
       }
diff --git a/vp10/encoder/mbgraph.c b/vp10/encoder/mbgraph.c
index 6149670..02ba0e5 100644
--- a/vp10/encoder/mbgraph.c
+++ b/vp10/encoder/mbgraph.c
@@ -41,7 +41,7 @@
 
   // Further step/diamond searches as necessary
   int step_param = mv_sf->reduce_first_step_size;
-  step_param = MIN(step_param, MAX_MVSEARCH_STEPS - 2);
+  step_param = VPXMIN(step_param, MAX_MVSEARCH_STEPS - 2);
 
   vp10_set_mv_search_range(x, ref_mv);
 
diff --git a/vp10/encoder/mcomp.c b/vp10/encoder/mcomp.c
index 5720c63..1cfe471 100644
--- a/vp10/encoder/mcomp.c
+++ b/vp10/encoder/mcomp.c
@@ -37,10 +37,10 @@
   int col_max = (mv->col >> 3) + MAX_FULL_PEL_VAL;
   int row_max = (mv->row >> 3) + MAX_FULL_PEL_VAL;
 
-  col_min = MAX(col_min, (MV_LOW >> 3) + 1);
-  row_min = MAX(row_min, (MV_LOW >> 3) + 1);
-  col_max = MIN(col_max, (MV_UPP >> 3) - 1);
-  row_max = MIN(row_max, (MV_UPP >> 3) - 1);
+  col_min = VPXMAX(col_min, (MV_LOW >> 3) + 1);
+  row_min = VPXMAX(row_min, (MV_LOW >> 3) + 1);
+  col_max = VPXMIN(col_max, (MV_UPP >> 3) - 1);
+  row_max = VPXMIN(row_max, (MV_UPP >> 3) - 1);
 
   // Get intersection of UMV window and valid MV window to reduce # of checks
   // in diamond search.
@@ -57,12 +57,12 @@
 int vp10_init_search_range(int size) {
   int sr = 0;
   // Minimum search size no matter what the passed in value.
-  size = MAX(16, size);
+  size = VPXMAX(16, size);
 
   while ((size << sr) < MAX_FULL_PEL_VAL)
     sr++;
 
-  sr = MIN(sr, MAX_MVSEARCH_STEPS - 2);
+  sr = VPXMIN(sr, MAX_MVSEARCH_STEPS - 2);
   return sr;
 }
 
@@ -297,10 +297,10 @@
   int br = bestmv->row * 8;                                                \
   int bc = bestmv->col * 8;                                                \
   int hstep = 4;                                                           \
-  const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);           \
-  const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);           \
-  const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);           \
-  const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);           \
+  const int minc = VPXMAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);        \
+  const int maxc = VPXMIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);        \
+  const int minr = VPXMAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);        \
+  const int maxr = VPXMIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);        \
   int tr = br;                                                             \
   int tc = bc;                                                             \
                                                                            \
@@ -668,10 +668,10 @@
   int bc = bestmv->col * 8;
   int hstep = 4;
   int iter, round = 3 - forced_stop;
-  const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);
-  const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);
-  const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);
-  const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);
+  const int minc = VPXMAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);
+  const int maxc = VPXMIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);
+  const int minr = VPXMAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);
+  const int maxr = VPXMIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);
   int tr = br;
   int tc = bc;
   const MV *search_step = search_step_table;
@@ -1500,9 +1500,9 @@
                         int use_mvcost,
                         const MV *center_mv,
                         MV *best_mv) {
-  return vp10_hex_search(x, ref_mv, MAX(MAX_MVSEARCH_STEPS - 2, search_param),
-                        sad_per_bit, do_init_search, cost_list, vfp, use_mvcost,
-                        center_mv, best_mv);
+  return vp10_hex_search(
+      x, ref_mv, VPXMAX(MAX_MVSEARCH_STEPS - 2, search_param), sad_per_bit,
+      do_init_search, cost_list, vfp, use_mvcost, center_mv, best_mv);
 }
 
 int vp10_fast_dia_search(const MACROBLOCK *x,
@@ -1515,9 +1515,9 @@
                         int use_mvcost,
                         const MV *center_mv,
                         MV *best_mv) {
-  return vp10_bigdia_search(x, ref_mv, MAX(MAX_MVSEARCH_STEPS - 2, search_param),
-                           sad_per_bit, do_init_search, cost_list, vfp,
-                           use_mvcost, center_mv, best_mv);
+  return vp10_bigdia_search(
+      x, ref_mv, VPXMAX(MAX_MVSEARCH_STEPS - 2, search_param), sad_per_bit,
+      do_init_search, cost_list, vfp, use_mvcost, center_mv, best_mv);
 }
 
 #undef CHECK_BETTER
@@ -1547,10 +1547,10 @@
   best_sad = fn_ptr->sdf(what->buf, what->stride,
                          get_buf_from_mv(in_what, ref_mv), in_what->stride) +
                  mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
-  start_row = MAX(-range, x->mv_row_min - ref_mv->row);
-  start_col = MAX(-range, x->mv_col_min - ref_mv->col);
-  end_row = MIN(range, x->mv_row_max - ref_mv->row);
-  end_col = MIN(range, x->mv_col_max - ref_mv->col);
+  start_row = VPXMAX(-range, x->mv_row_min - ref_mv->row);
+  start_col = VPXMAX(-range, x->mv_col_min - ref_mv->col);
+  end_row = VPXMIN(range, x->mv_row_max - ref_mv->row);
+  end_col = VPXMIN(range, x->mv_col_max - ref_mv->col);
 
   for (r = start_row; r <= end_row; ++r) {
     for (c = start_col; c <= end_col; c += 4) {
@@ -2021,10 +2021,10 @@
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct buf_2d *const what = &x->plane[0].src;
   const struct buf_2d *const in_what = &xd->plane[0].pre[0];
-  const int row_min = MAX(ref_mv->row - distance, x->mv_row_min);
-  const int row_max = MIN(ref_mv->row + distance, x->mv_row_max);
-  const int col_min = MAX(ref_mv->col - distance, x->mv_col_min);
-  const int col_max = MIN(ref_mv->col + distance, x->mv_col_max);
+  const int row_min = VPXMAX(ref_mv->row - distance, x->mv_row_min);
+  const int row_max = VPXMIN(ref_mv->row + distance, x->mv_row_max);
+  const int col_min = VPXMAX(ref_mv->col - distance, x->mv_col_min);
+  const int col_max = VPXMIN(ref_mv->col + distance, x->mv_col_max);
   const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
   int best_sad = fn_ptr->sdf(what->buf, what->stride,
       get_buf_from_mv(in_what, ref_mv), in_what->stride) +
@@ -2054,10 +2054,10 @@
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct buf_2d *const what = &x->plane[0].src;
   const struct buf_2d *const in_what = &xd->plane[0].pre[0];
-  const int row_min = MAX(ref_mv->row - distance, x->mv_row_min);
-  const int row_max = MIN(ref_mv->row + distance, x->mv_row_max);
-  const int col_min = MAX(ref_mv->col - distance, x->mv_col_min);
-  const int col_max = MIN(ref_mv->col + distance, x->mv_col_max);
+  const int row_min = VPXMAX(ref_mv->row - distance, x->mv_row_min);
+  const int row_max = VPXMIN(ref_mv->row + distance, x->mv_row_max);
+  const int col_min = VPXMAX(ref_mv->col - distance, x->mv_col_min);
+  const int col_max = VPXMIN(ref_mv->col + distance, x->mv_col_max);
   const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
   unsigned int best_sad = fn_ptr->sdf(what->buf, what->stride,
       get_buf_from_mv(in_what, ref_mv), in_what->stride) +
@@ -2119,10 +2119,10 @@
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct buf_2d *const what = &x->plane[0].src;
   const struct buf_2d *const in_what = &xd->plane[0].pre[0];
-  const int row_min = MAX(ref_mv->row - distance, x->mv_row_min);
-  const int row_max = MIN(ref_mv->row + distance, x->mv_row_max);
-  const int col_min = MAX(ref_mv->col - distance, x->mv_col_min);
-  const int col_max = MIN(ref_mv->col + distance, x->mv_col_max);
+  const int row_min = VPXMAX(ref_mv->row - distance, x->mv_row_min);
+  const int row_max = VPXMIN(ref_mv->row + distance, x->mv_row_max);
+  const int col_min = VPXMAX(ref_mv->col - distance, x->mv_col_min);
+  const int col_max = VPXMIN(ref_mv->col + distance, x->mv_col_max);
   const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
   unsigned int best_sad = fn_ptr->sdf(what->buf, what->stride,
       get_buf_from_mv(in_what, ref_mv), in_what->stride) +
diff --git a/vp10/encoder/picklpf.c b/vp10/encoder/picklpf.c
index 81c428d..ff0f387 100644
--- a/vp10/encoder/picklpf.c
+++ b/vp10/encoder/picklpf.c
@@ -92,8 +92,8 @@
   ss_err[filt_mid] = best_err;
 
   while (filter_step > 0) {
-    const int filt_high = MIN(filt_mid + filter_step, max_filter_level);
-    const int filt_low = MAX(filt_mid - filter_step, min_filter_level);
+    const int filt_high = VPXMIN(filt_mid + filter_step, max_filter_level);
+    const int filt_low = VPXMAX(filt_mid - filter_step, min_filter_level);
 
     // Bias against raising loop filter in favor of lowering it.
     int64_t bias = (best_err >> (15 - (filt_mid / 8))) * filter_step;
diff --git a/vp10/encoder/pickmode.c b/vp10/encoder/pickmode.c
deleted file mode 100644
index ef1a060..0000000
--- a/vp10/encoder/pickmode.c
+++ /dev/null
@@ -1,1890 +0,0 @@
-/*
- *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <assert.h>
-#include <limits.h>
-#include <math.h>
-#include <stdio.h>
-
-#include "./vp10_rtcd.h"
-#include "./vpx_dsp_rtcd.h"
-
-#include "vpx_mem/vpx_mem.h"
-#include "vpx_ports/mem.h"
-
-#include "vp10/common/blockd.h"
-#include "vp10/common/common.h"
-#include "vp10/common/mvref_common.h"
-#include "vp10/common/pred_common.h"
-#include "vp10/common/reconinter.h"
-#include "vp10/common/reconintra.h"
-#include "vp10/common/scan.h"
-
-#include "vp10/encoder/cost.h"
-#include "vp10/encoder/encoder.h"
-#include "vp10/encoder/pickmode.h"
-#include "vp10/encoder/ratectrl.h"
-#include "vp10/encoder/rd.h"
-
-typedef struct {
-  uint8_t *data;
-  int stride;
-  int in_use;
-} PRED_BUFFER;
-
-static int mv_refs_rt(const VP10_COMMON *cm, const MACROBLOCK *x,
-                      const MACROBLOCKD *xd,
-                      const TileInfo *const tile,
-                      MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
-                      int_mv *mv_ref_list,
-                      int mi_row, int mi_col) {
-  const int *ref_sign_bias = cm->ref_frame_sign_bias;
-  int i, refmv_count = 0;
-
-  const POSITION *const mv_ref_search = mv_ref_blocks[mi->mbmi.sb_type];
-
-  int different_ref_found = 0;
-  int context_counter = 0;
-  int const_motion = 0;
-
-  // Blank the reference vector list
-  memset(mv_ref_list, 0, sizeof(*mv_ref_list) * MAX_MV_REF_CANDIDATES);
-
-  // The nearest 2 blocks are treated differently
-  // if the size < 8x8 we get the mv from the bmi substructure,
-  // and we also need to keep a mode count.
-  for (i = 0; i < 2; ++i) {
-    const POSITION *const mv_ref = &mv_ref_search[i];
-    if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
-      const MODE_INFO *const candidate_mi = xd->mi[mv_ref->col + mv_ref->row *
-                                                   xd->mi_stride];
-      const MB_MODE_INFO *const candidate = &candidate_mi->mbmi;
-      // Keep counts for entropy encoding.
-      context_counter += mode_2_counter[candidate->mode];
-      different_ref_found = 1;
-
-      if (candidate->ref_frame[0] == ref_frame)
-        ADD_MV_REF_LIST(get_sub_block_mv(candidate_mi, 0, mv_ref->col, -1),
-                        refmv_count, mv_ref_list, Done);
-    }
-  }
-
-  const_motion = 1;
-
-  // Check the rest of the neighbors in much the same way
-  // as before except we don't need to keep track of sub blocks or
-  // mode counts.
-  for (; i < MVREF_NEIGHBOURS && !refmv_count; ++i) {
-    const POSITION *const mv_ref = &mv_ref_search[i];
-    if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
-      const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row *
-                                                    xd->mi_stride]->mbmi;
-      different_ref_found = 1;
-
-      if (candidate->ref_frame[0] == ref_frame)
-        ADD_MV_REF_LIST(candidate->mv[0], refmv_count, mv_ref_list, Done);
-    }
-  }
-
-  // Since we couldn't find 2 mvs from the same reference frame
-  // go back through the neighbors and find motion vectors from
-  // different reference frames.
-  if (different_ref_found && !refmv_count) {
-    for (i = 0; i < MVREF_NEIGHBOURS; ++i) {
-      const POSITION *mv_ref = &mv_ref_search[i];
-      if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
-        const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row
-                                              * xd->mi_stride]->mbmi;
-
-        // If the candidate is INTRA we don't want to consider its mv.
-        IF_DIFF_REF_FRAME_ADD_MV(candidate, ref_frame, ref_sign_bias,
-                                 refmv_count, mv_ref_list, Done);
-      }
-    }
-  }
-
- Done:
-
-  x->mbmi_ext->mode_context[ref_frame] = counter_to_context[context_counter];
-
-  // Clamp vectors
-  for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i)
-    clamp_mv_ref(&mv_ref_list[i].as_mv, xd);
-
-  return const_motion;
-}
-
-static int combined_motion_search(VP10_COMP *cpi, MACROBLOCK *x,
-                                  BLOCK_SIZE bsize, int mi_row, int mi_col,
-                                  int_mv *tmp_mv, int *rate_mv,
-                                  int64_t best_rd_sofar) {
-  MACROBLOCKD *xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
-  struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0, 0}};
-  const int step_param = cpi->sf.mv.fullpel_search_step_param;
-  const int sadpb = x->sadperbit16;
-  MV mvp_full;
-  const int ref = mbmi->ref_frame[0];
-  const MV ref_mv = x->mbmi_ext->ref_mvs[ref][0].as_mv;
-  int dis;
-  int rate_mode;
-  const int tmp_col_min = x->mv_col_min;
-  const int tmp_col_max = x->mv_col_max;
-  const int tmp_row_min = x->mv_row_min;
-  const int tmp_row_max = x->mv_row_max;
-  int rv = 0;
-  int cost_list[5];
-  const YV12_BUFFER_CONFIG *scaled_ref_frame = vp10_get_scaled_ref_frame(cpi,
-                                                                        ref);
-  if (scaled_ref_frame) {
-    int i;
-    // Swap out the reference frame for a version that's been scaled to
-    // match the resolution of the current frame, allowing the existing
-    // motion search code to be used without additional modifications.
-    for (i = 0; i < MAX_MB_PLANE; i++)
-      backup_yv12[i] = xd->plane[i].pre[0];
-    vp10_setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
-  }
-  vp10_set_mv_search_range(x, &ref_mv);
-
-  assert(x->mv_best_ref_index[ref] <= 2);
-  if (x->mv_best_ref_index[ref] < 2)
-    mvp_full = x->mbmi_ext->ref_mvs[ref][x->mv_best_ref_index[ref]].as_mv;
-  else
-    mvp_full = x->pred_mv[ref];
-
-  mvp_full.col >>= 3;
-  mvp_full.row >>= 3;
-
-  vp10_full_pixel_search(cpi, x, bsize, &mvp_full, step_param, sadpb,
-                        cond_cost_list(cpi, cost_list),
-                        &ref_mv, &tmp_mv->as_mv, INT_MAX, 0);
-
-  x->mv_col_min = tmp_col_min;
-  x->mv_col_max = tmp_col_max;
-  x->mv_row_min = tmp_row_min;
-  x->mv_row_max = tmp_row_max;
-
-  // calculate the bit cost on motion vector
-  mvp_full.row = tmp_mv->as_mv.row * 8;
-  mvp_full.col = tmp_mv->as_mv.col * 8;
-
-  *rate_mv = vp10_mv_bit_cost(&mvp_full, &ref_mv,
-                             x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
-
-  rate_mode = cpi->inter_mode_cost[x->mbmi_ext->mode_context[ref]]
-                                  [INTER_OFFSET(NEWMV)];
-  rv = !(RDCOST(x->rdmult, x->rddiv, (*rate_mv + rate_mode), 0) >
-         best_rd_sofar);
-
-  if (rv) {
-    cpi->find_fractional_mv_step(x, &tmp_mv->as_mv, &ref_mv,
-                                 cpi->common.allow_high_precision_mv,
-                                 x->errorperbit,
-                                 &cpi->fn_ptr[bsize],
-                                 cpi->sf.mv.subpel_force_stop,
-                                 cpi->sf.mv.subpel_iters_per_step,
-                                 cond_cost_list(cpi, cost_list),
-                                 x->nmvjointcost, x->mvcost,
-                                 &dis, &x->pred_sse[ref], NULL, 0, 0);
-    *rate_mv = vp10_mv_bit_cost(&tmp_mv->as_mv, &ref_mv,
-                               x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
-  }
-
-  if (scaled_ref_frame) {
-    int i;
-    for (i = 0; i < MAX_MB_PLANE; i++)
-      xd->plane[i].pre[0] = backup_yv12[i];
-  }
-  return rv;
-}
-
-static void block_variance(const uint8_t *src, int src_stride,
-                           const uint8_t *ref, int ref_stride,
-                           int w, int h, unsigned int *sse, int *sum,
-                           int block_size, unsigned int *sse8x8,
-                           int *sum8x8, unsigned int *var8x8) {
-  int i, j, k = 0;
-
-  *sse = 0;
-  *sum = 0;
-
-  for (i = 0; i < h; i += block_size) {
-    for (j = 0; j < w; j += block_size) {
-      vpx_get8x8var(src + src_stride * i + j, src_stride,
-                    ref + ref_stride * i + j, ref_stride,
-                    &sse8x8[k], &sum8x8[k]);
-      *sse += sse8x8[k];
-      *sum += sum8x8[k];
-      var8x8[k] = sse8x8[k] - (((unsigned int)sum8x8[k] * sum8x8[k]) >> 6);
-      k++;
-    }
-  }
-}
-
-static void calculate_variance(int bw, int bh, TX_SIZE tx_size,
-                               unsigned int *sse_i, int *sum_i,
-                               unsigned int *var_o, unsigned int *sse_o,
-                               int *sum_o) {
-  const BLOCK_SIZE unit_size = txsize_to_bsize[tx_size];
-  const int nw = 1 << (bw - b_width_log2_lookup[unit_size]);
-  const int nh = 1 << (bh - b_height_log2_lookup[unit_size]);
-  int i, j, k = 0;
-
-  for (i = 0; i < nh; i += 2) {
-    for (j = 0; j < nw; j += 2) {
-      sse_o[k] = sse_i[i * nw + j] + sse_i[i * nw + j + 1] +
-          sse_i[(i + 1) * nw + j] + sse_i[(i + 1) * nw + j + 1];
-      sum_o[k] = sum_i[i * nw + j] + sum_i[i * nw + j + 1] +
-          sum_i[(i + 1) * nw + j] + sum_i[(i + 1) * nw + j + 1];
-      var_o[k] = sse_o[k] - (((unsigned int)sum_o[k] * sum_o[k]) >>
-          (b_width_log2_lookup[unit_size] +
-              b_height_log2_lookup[unit_size] + 6));
-      k++;
-    }
-  }
-}
-
-static void model_rd_for_sb_y_large(VP10_COMP *cpi, BLOCK_SIZE bsize,
-                                    MACROBLOCK *x, MACROBLOCKD *xd,
-                                    int *out_rate_sum, int64_t *out_dist_sum,
-                                    unsigned int *var_y, unsigned int *sse_y,
-                                    int mi_row, int mi_col, int *early_term) {
-  // 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.
-  unsigned int sse;
-  int rate;
-  int64_t dist;
-  struct macroblock_plane *const p = &x->plane[0];
-  struct macroblockd_plane *const pd = &xd->plane[0];
-  const uint32_t dc_quant = pd->dequant[0];
-  const uint32_t ac_quant = pd->dequant[1];
-  const int64_t dc_thr = dc_quant * dc_quant >> 6;
-  const int64_t ac_thr = ac_quant * ac_quant >> 6;
-  unsigned int var;
-  int sum;
-  int skip_dc = 0;
-
-  const int bw = b_width_log2_lookup[bsize];
-  const int bh = b_height_log2_lookup[bsize];
-  const int num8x8 = 1 << (bw + bh - 2);
-  unsigned int sse8x8[64] = {0};
-  int sum8x8[64] = {0};
-  unsigned int var8x8[64] = {0};
-  TX_SIZE tx_size;
-  int i, k;
-
-  // Calculate variance for whole partition, and also save 8x8 blocks' variance
-  // to be used in following transform skipping test.
-  block_variance(p->src.buf, p->src.stride, pd->dst.buf, pd->dst.stride,
-                 4 << bw, 4 << bh, &sse, &sum, 8, sse8x8, sum8x8, var8x8);
-  var = sse - (((int64_t)sum * sum) >> (bw + bh + 4));
-
-  *var_y = var;
-  *sse_y = sse;
-
-  if (cpi->common.tx_mode == TX_MODE_SELECT) {
-    if (sse > (var << 2))
-      tx_size = MIN(max_txsize_lookup[bsize],
-                    tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
-    else
-      tx_size = TX_8X8;
-
-    if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ &&
-        cyclic_refresh_segment_id_boosted(xd->mi[0]->mbmi.segment_id))
-      tx_size = TX_8X8;
-    else if (tx_size > TX_16X16)
-      tx_size = TX_16X16;
-  } else {
-    tx_size = MIN(max_txsize_lookup[bsize],
-                  tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
-  }
-
-  assert(tx_size >= TX_8X8);
-  xd->mi[0]->mbmi.tx_size = tx_size;
-
-  // Evaluate if the partition block is a skippable block in Y plane.
-  {
-    unsigned int sse16x16[16] = {0};
-    int sum16x16[16] = {0};
-    unsigned int var16x16[16] = {0};
-    const int num16x16 = num8x8 >> 2;
-
-    unsigned int sse32x32[4] = {0};
-    int sum32x32[4] = {0};
-    unsigned int var32x32[4] = {0};
-    const int num32x32 = num8x8 >> 4;
-
-    int ac_test = 1;
-    int dc_test = 1;
-    const int num = (tx_size == TX_8X8) ? num8x8 :
-        ((tx_size == TX_16X16) ? num16x16 : num32x32);
-    const unsigned int *sse_tx = (tx_size == TX_8X8) ? sse8x8 :
-        ((tx_size == TX_16X16) ? sse16x16 : sse32x32);
-    const unsigned int *var_tx = (tx_size == TX_8X8) ? var8x8 :
-        ((tx_size == TX_16X16) ? var16x16 : var32x32);
-
-    // Calculate variance if tx_size > TX_8X8
-    if (tx_size >= TX_16X16)
-      calculate_variance(bw, bh, TX_8X8, sse8x8, sum8x8, var16x16, sse16x16,
-                         sum16x16);
-    if (tx_size == TX_32X32)
-      calculate_variance(bw, bh, TX_16X16, sse16x16, sum16x16, var32x32,
-                         sse32x32, sum32x32);
-
-    // Skipping test
-    x->skip_txfm[0] = SKIP_TXFM_NONE;
-    for (k = 0; k < num; k++)
-      // Check if all ac coefficients can be quantized to zero.
-      if (!(var_tx[k] < ac_thr || var == 0)) {
-        ac_test = 0;
-        break;
-      }
-
-    for (k = 0; k < num; k++)
-      // Check if dc coefficient can be quantized to zero.
-      if (!(sse_tx[k] - var_tx[k] < dc_thr || sse == var)) {
-        dc_test = 0;
-        break;
-      }
-
-    if (ac_test) {
-      x->skip_txfm[0] = SKIP_TXFM_AC_ONLY;
-
-      if (dc_test)
-        x->skip_txfm[0] = SKIP_TXFM_AC_DC;
-    } else if (dc_test) {
-      skip_dc = 1;
-    }
-  }
-
-  if (x->skip_txfm[0] == SKIP_TXFM_AC_DC) {
-    int skip_uv[2] = {0};
-    unsigned int var_uv[2];
-    unsigned int sse_uv[2];
-
-    *out_rate_sum = 0;
-    *out_dist_sum = sse << 4;
-
-    // Transform skipping test in UV planes.
-    for (i = 1; i <= 2; i++) {
-      struct macroblock_plane *const p = &x->plane[i];
-      struct macroblockd_plane *const pd = &xd->plane[i];
-      const TX_SIZE uv_tx_size = get_uv_tx_size(&xd->mi[0]->mbmi, pd);
-      const BLOCK_SIZE unit_size = txsize_to_bsize[uv_tx_size];
-      const BLOCK_SIZE uv_bsize = get_plane_block_size(bsize, pd);
-      const int uv_bw = b_width_log2_lookup[uv_bsize];
-      const int uv_bh = b_height_log2_lookup[uv_bsize];
-      const int sf = (uv_bw - b_width_log2_lookup[unit_size]) +
-          (uv_bh - b_height_log2_lookup[unit_size]);
-      const uint32_t uv_dc_thr = pd->dequant[0] * pd->dequant[0] >> (6 - sf);
-      const uint32_t uv_ac_thr = pd->dequant[1] * pd->dequant[1] >> (6 - sf);
-      int j = i - 1;
-
-      vp10_build_inter_predictors_sbp(xd, mi_row, mi_col, bsize, i);
-      var_uv[j] = cpi->fn_ptr[uv_bsize].vf(p->src.buf, p->src.stride,
-          pd->dst.buf, pd->dst.stride, &sse_uv[j]);
-
-      if ((var_uv[j] < uv_ac_thr || var_uv[j] == 0) &&
-          (sse_uv[j] - var_uv[j] < uv_dc_thr || sse_uv[j] == var_uv[j]))
-        skip_uv[j] = 1;
-      else
-        break;
-    }
-
-    // If the transform in YUV planes are skippable, the mode search checks
-    // fewer inter modes and doesn't check intra modes.
-    if (skip_uv[0] & skip_uv[1]) {
-      *early_term = 1;
-    }
-
-    return;
-  }
-
-  if (!skip_dc) {
-#if CONFIG_VP9_HIGHBITDEPTH
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      vp10_model_rd_from_var_lapndz(sse - var, num_pels_log2_lookup[bsize],
-                                   dc_quant >> (xd->bd - 5), &rate, &dist);
-    } else {
-      vp10_model_rd_from_var_lapndz(sse - var, num_pels_log2_lookup[bsize],
-                                   dc_quant >> 3, &rate, &dist);
-    }
-#else
-    vp10_model_rd_from_var_lapndz(sse - var, num_pels_log2_lookup[bsize],
-                                 dc_quant >> 3, &rate, &dist);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-  }
-
-  if (!skip_dc) {
-    *out_rate_sum = rate >> 1;
-    *out_dist_sum = dist << 3;
-  } else {
-    *out_rate_sum = 0;
-    *out_dist_sum = (sse - var) << 4;
-  }
-
-#if CONFIG_VP9_HIGHBITDEPTH
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-    vp10_model_rd_from_var_lapndz(var, num_pels_log2_lookup[bsize],
-                                 ac_quant >> (xd->bd - 5), &rate, &dist);
-  } else {
-    vp10_model_rd_from_var_lapndz(var, num_pels_log2_lookup[bsize],
-                                 ac_quant >> 3, &rate, &dist);
-  }
-#else
-  vp10_model_rd_from_var_lapndz(var, num_pels_log2_lookup[bsize],
-                               ac_quant >> 3, &rate, &dist);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-  *out_rate_sum += rate;
-  *out_dist_sum += dist << 4;
-}
-
-static void model_rd_for_sb_y(VP10_COMP *cpi, BLOCK_SIZE bsize,
-                              MACROBLOCK *x, MACROBLOCKD *xd,
-                              int *out_rate_sum, int64_t *out_dist_sum,
-                              unsigned int *var_y, unsigned int *sse_y) {
-  // 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.
-  unsigned int sse;
-  int rate;
-  int64_t dist;
-  struct macroblock_plane *const p = &x->plane[0];
-  struct macroblockd_plane *const pd = &xd->plane[0];
-  const int64_t dc_thr = p->quant_thred[0] >> 6;
-  const int64_t ac_thr = p->quant_thred[1] >> 6;
-  const uint32_t dc_quant = pd->dequant[0];
-  const uint32_t ac_quant = pd->dequant[1];
-  unsigned int var = cpi->fn_ptr[bsize].vf(p->src.buf, p->src.stride,
-                                           pd->dst.buf, pd->dst.stride, &sse);
-  int skip_dc = 0;
-
-  *var_y = var;
-  *sse_y = sse;
-
-  if (cpi->common.tx_mode == TX_MODE_SELECT) {
-    if (sse > (var << 2))
-      xd->mi[0]->mbmi.tx_size =
-          MIN(max_txsize_lookup[bsize],
-              tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
-    else
-      xd->mi[0]->mbmi.tx_size = TX_8X8;
-
-    if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ &&
-        cyclic_refresh_segment_id_boosted(xd->mi[0]->mbmi.segment_id))
-      xd->mi[0]->mbmi.tx_size = TX_8X8;
-    else if (xd->mi[0]->mbmi.tx_size > TX_16X16)
-      xd->mi[0]->mbmi.tx_size = TX_16X16;
-  } else {
-    xd->mi[0]->mbmi.tx_size =
-        MIN(max_txsize_lookup[bsize],
-            tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
-  }
-
-  // Evaluate if the partition block is a skippable block in Y plane.
-  {
-    const BLOCK_SIZE unit_size =
-        txsize_to_bsize[xd->mi[0]->mbmi.tx_size];
-    const unsigned int num_blk_log2 =
-        (b_width_log2_lookup[bsize] - b_width_log2_lookup[unit_size]) +
-        (b_height_log2_lookup[bsize] - b_height_log2_lookup[unit_size]);
-    const unsigned int sse_tx = sse >> num_blk_log2;
-    const unsigned int var_tx = var >> num_blk_log2;
-
-    x->skip_txfm[0] = SKIP_TXFM_NONE;
-    // Check if all ac coefficients can be quantized to zero.
-    if (var_tx < ac_thr || var == 0) {
-      x->skip_txfm[0] = SKIP_TXFM_AC_ONLY;
-      // Check if dc coefficient can be quantized to zero.
-      if (sse_tx - var_tx < dc_thr || sse == var)
-        x->skip_txfm[0] = SKIP_TXFM_AC_DC;
-    } else {
-      if (sse_tx - var_tx < dc_thr || sse == var)
-        skip_dc = 1;
-    }
-  }
-
-  if (x->skip_txfm[0] == SKIP_TXFM_AC_DC) {
-    *out_rate_sum = 0;
-    *out_dist_sum = sse << 4;
-    return;
-  }
-
-  if (!skip_dc) {
-#if CONFIG_VP9_HIGHBITDEPTH
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      vp10_model_rd_from_var_lapndz(sse - var, num_pels_log2_lookup[bsize],
-                                   dc_quant >> (xd->bd - 5), &rate, &dist);
-    } else {
-      vp10_model_rd_from_var_lapndz(sse - var, num_pels_log2_lookup[bsize],
-                                   dc_quant >> 3, &rate, &dist);
-    }
-#else
-    vp10_model_rd_from_var_lapndz(sse - var, num_pels_log2_lookup[bsize],
-                                 dc_quant >> 3, &rate, &dist);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-  }
-
-  if (!skip_dc) {
-    *out_rate_sum = rate >> 1;
-    *out_dist_sum = dist << 3;
-  } else {
-    *out_rate_sum = 0;
-    *out_dist_sum = (sse - var) << 4;
-  }
-
-#if CONFIG_VP9_HIGHBITDEPTH
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-    vp10_model_rd_from_var_lapndz(var, num_pels_log2_lookup[bsize],
-                                 ac_quant >> (xd->bd - 5), &rate, &dist);
-  } else {
-    vp10_model_rd_from_var_lapndz(var, num_pels_log2_lookup[bsize],
-                                 ac_quant >> 3, &rate, &dist);
-  }
-#else
-  vp10_model_rd_from_var_lapndz(var, num_pels_log2_lookup[bsize],
-                               ac_quant >> 3, &rate, &dist);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-  *out_rate_sum += rate;
-  *out_dist_sum += dist << 4;
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static void block_yrd(VP10_COMP *cpi, MACROBLOCK *x, int *rate, int64_t *dist,
-                      int *skippable, int64_t *sse, int plane,
-                      BLOCK_SIZE bsize, TX_SIZE tx_size) {
-  MACROBLOCKD *xd = &x->e_mbd;
-  unsigned int var_y, sse_y;
-  (void)plane;
-  (void)tx_size;
-  model_rd_for_sb_y(cpi, bsize, x, xd, rate, dist, &var_y, &sse_y);
-  *sse = INT_MAX;
-  *skippable = 0;
-  return;
-}
-#else
-static void block_yrd(VP10_COMP *cpi, MACROBLOCK *x, int *rate, int64_t *dist,
-                      int *skippable, int64_t *sse, int plane,
-                      BLOCK_SIZE bsize, TX_SIZE tx_size) {
-  MACROBLOCKD *xd = &x->e_mbd;
-  const struct macroblockd_plane *pd = &xd->plane[plane];
-  const struct macroblock_plane *const p = &x->plane[plane];
-  const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];
-  const int num_4x4_h = num_4x4_blocks_high_lookup[bsize];
-  const int step = 1 << (tx_size << 1);
-  const int block_step = (1 << tx_size);
-  int block = 0, r, c;
-  int shift = tx_size == TX_32X32 ? 0 : 2;
-  const int max_blocks_wide = num_4x4_w + (xd->mb_to_right_edge >= 0 ? 0 :
-      xd->mb_to_right_edge >> (5 + pd->subsampling_x));
-  const int max_blocks_high = num_4x4_h + (xd->mb_to_bottom_edge >= 0 ? 0 :
-      xd->mb_to_bottom_edge >> (5 + pd->subsampling_y));
-  int eob_cost = 0;
-
-  (void)cpi;
-  vp10_subtract_plane(x, bsize, plane);
-  *skippable = 1;
-  // Keep track of the row and column of the blocks we use so that we know
-  // if we are in the unrestricted motion border.
-  for (r = 0; r < max_blocks_high; r += block_step) {
-    for (c = 0; c < num_4x4_w; c += block_step) {
-      if (c < max_blocks_wide) {
-        PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
-        TX_TYPE tx_type = get_tx_type(plane_type, xd, block);
-        const scan_order *const scan_order = get_scan(tx_size, tx_type);
-        tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
-        tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
-        tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
-        uint16_t *const eob = &p->eobs[block];
-        const int diff_stride = 4 * num_4x4_blocks_wide_lookup[bsize];
-        const int16_t *src_diff;
-        src_diff = &p->src_diff[(r * diff_stride + c) << 2];
-
-        switch (tx_size) {
-          case TX_32X32:
-            vpx_fdct32x32_rd(src_diff, coeff, diff_stride);
-            vp10_quantize_fp_32x32(coeff, 1024, x->skip_block, p->zbin,
-                                  p->round_fp, p->quant_fp, p->quant_shift,
-                                  qcoeff, dqcoeff, pd->dequant, eob,
-                                  scan_order->scan, scan_order->iscan);
-            break;
-          case TX_16X16:
-            vp10_hadamard_16x16(src_diff, diff_stride, (int16_t *)coeff);
-            vp10_quantize_fp(coeff, 256, x->skip_block, p->zbin, p->round_fp,
-                            p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
-                            pd->dequant, eob,
-                            scan_order->scan, scan_order->iscan);
-            break;
-          case TX_8X8:
-            vp10_hadamard_8x8(src_diff, diff_stride, (int16_t *)coeff);
-            vp10_quantize_fp(coeff, 64, x->skip_block, p->zbin, p->round_fp,
-                            p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
-                            pd->dequant, eob,
-                            scan_order->scan, scan_order->iscan);
-            break;
-          case TX_4X4:
-            x->fwd_txm4x4(src_diff, coeff, diff_stride);
-            vp10_quantize_fp(coeff, 16, x->skip_block, p->zbin, p->round_fp,
-                            p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
-                            pd->dequant, eob,
-                            scan_order->scan, scan_order->iscan);
-            break;
-          default:
-            assert(0);
-            break;
-        }
-        *skippable &= (*eob == 0);
-        eob_cost += 1;
-      }
-      block += step;
-    }
-  }
-
-  if (*skippable && *sse < INT64_MAX) {
-    *rate = 0;
-    *dist = (*sse << 6) >> shift;
-    *sse = *dist;
-    return;
-  }
-
-  block = 0;
-  *rate = 0;
-  *dist = 0;
-  if (*sse < INT64_MAX)
-    *sse = (*sse << 6) >> shift;
-  for (r = 0; r < max_blocks_high; r += block_step) {
-    for (c = 0; c < num_4x4_w; c += block_step) {
-      if (c < max_blocks_wide) {
-        tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
-        tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
-        tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
-        uint16_t *const eob = &p->eobs[block];
-
-        if (*eob == 1)
-          *rate += (int)abs(qcoeff[0]);
-        else if (*eob > 1)
-          *rate += (int)vp10_satd((const int16_t *)qcoeff, step << 4);
-
-        *dist += vp10_block_error_fp(coeff, dqcoeff, step << 4) >> shift;
-      }
-      block += step;
-    }
-  }
-
-  if (*skippable == 0) {
-    *rate <<= 10;
-    *rate += (eob_cost << 8);
-  }
-}
-#endif
-
-static void model_rd_for_sb_uv(VP10_COMP *cpi, BLOCK_SIZE bsize,
-                               MACROBLOCK *x, MACROBLOCKD *xd,
-                               int *out_rate_sum, int64_t *out_dist_sum,
-                               unsigned int *var_y, unsigned int *sse_y) {
-  // 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.
-  unsigned int sse;
-  int rate;
-  int64_t dist;
-  int i;
-
-  *out_rate_sum = 0;
-  *out_dist_sum = 0;
-
-  for (i = 1; i <= 2; ++i) {
-    struct macroblock_plane *const p = &x->plane[i];
-    struct macroblockd_plane *const pd = &xd->plane[i];
-    const uint32_t dc_quant = pd->dequant[0];
-    const uint32_t ac_quant = pd->dequant[1];
-    const BLOCK_SIZE bs = get_plane_block_size(bsize, pd);
-    unsigned int var;
-
-    if (!x->color_sensitivity[i - 1])
-      continue;
-
-    var = cpi->fn_ptr[bs].vf(p->src.buf, p->src.stride,
-                             pd->dst.buf, pd->dst.stride, &sse);
-    *var_y += var;
-    *sse_y += sse;
-
-  #if CONFIG_VP9_HIGHBITDEPTH
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      vp10_model_rd_from_var_lapndz(sse - var, num_pels_log2_lookup[bs],
-                                   dc_quant >> (xd->bd - 5), &rate, &dist);
-    } else {
-      vp10_model_rd_from_var_lapndz(sse - var, num_pels_log2_lookup[bs],
-                                   dc_quant >> 3, &rate, &dist);
-    }
-  #else
-    vp10_model_rd_from_var_lapndz(sse - var, num_pels_log2_lookup[bs],
-                                 dc_quant >> 3, &rate, &dist);
-  #endif  // CONFIG_VP9_HIGHBITDEPTH
-
-    *out_rate_sum += rate >> 1;
-    *out_dist_sum += dist << 3;
-
-  #if CONFIG_VP9_HIGHBITDEPTH
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      vp10_model_rd_from_var_lapndz(var, num_pels_log2_lookup[bs],
-                                   ac_quant >> (xd->bd - 5), &rate, &dist);
-    } else {
-      vp10_model_rd_from_var_lapndz(var, num_pels_log2_lookup[bs],
-                                   ac_quant >> 3, &rate, &dist);
-    }
-  #else
-    vp10_model_rd_from_var_lapndz(var, num_pels_log2_lookup[bs],
-                                 ac_quant >> 3, &rate, &dist);
-  #endif  // CONFIG_VP9_HIGHBITDEPTH
-
-    *out_rate_sum += rate;
-    *out_dist_sum += dist << 4;
-  }
-}
-
-static int get_pred_buffer(PRED_BUFFER *p, int len) {
-  int i;
-
-  for (i = 0; i < len; i++) {
-    if (!p[i].in_use) {
-      p[i].in_use = 1;
-      return i;
-    }
-  }
-  return -1;
-}
-
-static void free_pred_buffer(PRED_BUFFER *p) {
-  if (p != NULL)
-    p->in_use = 0;
-}
-
-static void encode_breakout_test(VP10_COMP *cpi, MACROBLOCK *x,
-                                 BLOCK_SIZE bsize, int mi_row, int mi_col,
-                                 MV_REFERENCE_FRAME ref_frame,
-                                 PREDICTION_MODE this_mode,
-                                 unsigned int var_y, unsigned int sse_y,
-                                 struct buf_2d yv12_mb[][MAX_MB_PLANE],
-                                 int *rate, int64_t *dist) {
-  MACROBLOCKD *xd = &x->e_mbd;
-
-  const BLOCK_SIZE uv_size = get_plane_block_size(bsize, &xd->plane[1]);
-  unsigned int var = var_y, sse = sse_y;
-  // Skipping threshold for ac.
-  unsigned int thresh_ac;
-  // Skipping threshold for dc.
-  unsigned int thresh_dc;
-  if (x->encode_breakout > 0) {
-    // Set a maximum for threshold to avoid big PSNR loss in low bit rate
-    // case. Use extreme low threshold for static frames to limit
-    // skipping.
-    const unsigned int max_thresh = 36000;
-    // The encode_breakout input
-    const unsigned int min_thresh =
-        MIN(((unsigned int)x->encode_breakout << 4), max_thresh);
-#if CONFIG_VP9_HIGHBITDEPTH
-    const int shift = (xd->bd << 1) - 16;
-#endif
-
-    // Calculate threshold according to dequant value.
-    thresh_ac = (xd->plane[0].dequant[1] * xd->plane[0].dequant[1]) >> 3;
-#if CONFIG_VP9_HIGHBITDEPTH
-    if ((xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) && shift > 0) {
-      thresh_ac = ROUND_POWER_OF_TWO(thresh_ac, shift);
-    }
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-    thresh_ac = clamp(thresh_ac, min_thresh, max_thresh);
-
-    // Adjust ac threshold according to partition size.
-    thresh_ac >>=
-        8 - (b_width_log2_lookup[bsize] + b_height_log2_lookup[bsize]);
-
-    thresh_dc = (xd->plane[0].dequant[0] * xd->plane[0].dequant[0] >> 6);
-#if CONFIG_VP9_HIGHBITDEPTH
-    if ((xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) && shift > 0) {
-      thresh_dc = ROUND_POWER_OF_TWO(thresh_dc, shift);
-    }
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-  } else {
-    thresh_ac = 0;
-    thresh_dc = 0;
-  }
-
-  // Y skipping condition checking for ac and dc.
-  if (var <= thresh_ac && (sse - var) <= thresh_dc) {
-    unsigned int sse_u, sse_v;
-    unsigned int var_u, var_v;
-
-    // Skip UV prediction unless breakout is zero (lossless) to save
-    // computation with low impact on the result
-    if (x->encode_breakout == 0) {
-      xd->plane[1].pre[0] = yv12_mb[ref_frame][1];
-      xd->plane[2].pre[0] = yv12_mb[ref_frame][2];
-      vp10_build_inter_predictors_sbuv(xd, mi_row, mi_col, bsize);
-    }
-
-    var_u = cpi->fn_ptr[uv_size].vf(x->plane[1].src.buf,
-                                    x->plane[1].src.stride,
-                                    xd->plane[1].dst.buf,
-                                    xd->plane[1].dst.stride, &sse_u);
-
-    // U skipping condition checking
-    if (((var_u << 2) <= thresh_ac) && (sse_u - var_u <= thresh_dc)) {
-      var_v = cpi->fn_ptr[uv_size].vf(x->plane[2].src.buf,
-                                      x->plane[2].src.stride,
-                                      xd->plane[2].dst.buf,
-                                      xd->plane[2].dst.stride, &sse_v);
-
-      // V skipping condition checking
-      if (((var_v << 2) <= thresh_ac) && (sse_v - var_v <= thresh_dc)) {
-        x->skip = 1;
-
-        // The cost of skip bit needs to be added.
-        *rate = cpi->inter_mode_cost[x->mbmi_ext->mode_context[ref_frame]]
-                                    [INTER_OFFSET(this_mode)];
-
-        // More on this part of rate
-        // rate += vp10_cost_bit(vp10_get_skip_prob(cm, xd), 1);
-
-        // Scaling factor for SSE from spatial domain to frequency
-        // domain is 16. Adjust distortion accordingly.
-        // TODO(yunqingwang): In this function, only y-plane dist is
-        // calculated.
-        *dist = (sse << 4);  // + ((sse_u + sse_v) << 4);
-
-        // *disable_skip = 1;
-      }
-    }
-  }
-}
-
-struct estimate_block_intra_args {
-  VP10_COMP *cpi;
-  MACROBLOCK *x;
-  PREDICTION_MODE mode;
-  int rate;
-  int64_t dist;
-};
-
-static void estimate_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
-                                 TX_SIZE tx_size, void *arg) {
-  struct estimate_block_intra_args* const args = arg;
-  VP10_COMP *const cpi = args->cpi;
-  MACROBLOCK *const x = args->x;
-  MACROBLOCKD *const xd = &x->e_mbd;
-  struct macroblock_plane *const p = &x->plane[0];
-  struct macroblockd_plane *const pd = &xd->plane[0];
-  const BLOCK_SIZE bsize_tx = txsize_to_bsize[tx_size];
-  uint8_t *const src_buf_base = p->src.buf;
-  uint8_t *const dst_buf_base = pd->dst.buf;
-  const int src_stride = p->src.stride;
-  const int dst_stride = pd->dst.stride;
-  int i, j;
-  int rate;
-  int64_t dist;
-  int64_t this_sse = INT64_MAX;
-  int is_skippable;
-
-  txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
-  assert(plane == 0);
-  (void) plane;
-
-  p->src.buf = &src_buf_base[4 * (j * src_stride + i)];
-  pd->dst.buf = &dst_buf_base[4 * (j * dst_stride + i)];
-  // Use source buffer as an approximation for the fully reconstructed buffer.
-  vp10_predict_intra_block(xd, b_width_log2_lookup[plane_bsize],
-                          tx_size, args->mode,
-                          x->skip_encode ? p->src.buf : pd->dst.buf,
-                          x->skip_encode ? src_stride : dst_stride,
-                          pd->dst.buf, dst_stride,
-                          i, j, 0);
-
-  // TODO(jingning): This needs further refactoring.
-  block_yrd(cpi, x, &rate, &dist, &is_skippable, &this_sse, 0,
-            bsize_tx, MIN(tx_size, TX_16X16));
-  x->skip_txfm[0] = is_skippable;
-  rate += vp10_cost_bit(vp10_get_skip_prob(&cpi->common, xd), is_skippable);
-
-  p->src.buf = src_buf_base;
-  pd->dst.buf = dst_buf_base;
-  args->rate += rate;
-  args->dist += dist;
-}
-
-static const THR_MODES mode_idx[MAX_REF_FRAMES - 1][4] = {
-  {THR_DC, THR_V_PRED, THR_H_PRED, THR_TM},
-  {THR_NEARESTMV, THR_NEARMV, THR_ZEROMV, THR_NEWMV},
-  {THR_NEARESTG, THR_NEARG, THR_ZEROG, THR_NEWG},
-};
-
-static const PREDICTION_MODE intra_mode_list[] = {
-  DC_PRED, V_PRED, H_PRED, TM_PRED
-};
-
-static int mode_offset(const PREDICTION_MODE mode) {
-  if (mode >= NEARESTMV) {
-    return INTER_OFFSET(mode);
-  } else {
-    switch (mode) {
-      case DC_PRED:
-        return 0;
-      case V_PRED:
-        return 1;
-      case H_PRED:
-        return 2;
-      case TM_PRED:
-        return 3;
-      default:
-        return -1;
-    }
-  }
-}
-
-static INLINE void update_thresh_freq_fact(VP10_COMP *cpi,
-                                           TileDataEnc *tile_data,
-                                           BLOCK_SIZE bsize,
-                                           MV_REFERENCE_FRAME ref_frame,
-                                           THR_MODES best_mode_idx,
-                                           PREDICTION_MODE mode) {
-  THR_MODES thr_mode_idx = mode_idx[ref_frame][mode_offset(mode)];
-  int *freq_fact = &tile_data->thresh_freq_fact[bsize][thr_mode_idx];
-  if (thr_mode_idx == best_mode_idx)
-    *freq_fact -= (*freq_fact >> 4);
-  else
-    *freq_fact = MIN(*freq_fact + RD_THRESH_INC,
-        cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT);
-}
-
-void vp10_pick_intra_mode(VP10_COMP *cpi, MACROBLOCK *x, RD_COST *rd_cost,
-                         BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx) {
-  MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  RD_COST this_rdc, best_rdc;
-  PREDICTION_MODE this_mode;
-  struct estimate_block_intra_args args = { cpi, x, DC_PRED, 0, 0 };
-  const TX_SIZE intra_tx_size =
-      MIN(max_txsize_lookup[bsize],
-          tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
-  MODE_INFO *const mic = xd->mi[0];
-  int *bmode_costs;
-  const MODE_INFO *above_mi = xd->mi[-xd->mi_stride];
-  const MODE_INFO *left_mi = xd->left_available ? xd->mi[-1] : NULL;
-  const PREDICTION_MODE A = vp10_above_block_mode(mic, above_mi, 0);
-  const PREDICTION_MODE L = vp10_left_block_mode(mic, left_mi, 0);
-  bmode_costs = cpi->y_mode_costs[A][L];
-
-  (void) ctx;
-  vp10_rd_cost_reset(&best_rdc);
-  vp10_rd_cost_reset(&this_rdc);
-
-  mbmi->ref_frame[0] = INTRA_FRAME;
-  mbmi->mv[0].as_int = INVALID_MV;
-  mbmi->uv_mode = DC_PRED;
-  memset(x->skip_txfm, 0, sizeof(x->skip_txfm));
-
-  // Change the limit of this loop to add other intra prediction
-  // mode tests.
-  for (this_mode = DC_PRED; this_mode <= H_PRED; ++this_mode) {
-    args.mode = this_mode;
-    args.rate = 0;
-    args.dist = 0;
-    mbmi->tx_size = intra_tx_size;
-    vp10_foreach_transformed_block_in_plane(xd, bsize, 0,
-                                           estimate_block_intra, &args);
-    this_rdc.rate = args.rate;
-    this_rdc.dist = args.dist;
-    this_rdc.rate += bmode_costs[this_mode];
-    this_rdc.rdcost = RDCOST(x->rdmult, x->rddiv,
-                             this_rdc.rate, this_rdc.dist);
-
-    if (this_rdc.rdcost < best_rdc.rdcost) {
-      best_rdc = this_rdc;
-      mbmi->mode = this_mode;
-    }
-  }
-
-  *rd_cost = best_rdc;
-}
-
-static void init_ref_frame_cost(VP10_COMMON *const cm,
-                                MACROBLOCKD *const xd,
-                                int ref_frame_cost[MAX_REF_FRAMES]) {
-  vpx_prob intra_inter_p = vp10_get_intra_inter_prob(cm, xd);
-  vpx_prob ref_single_p1 = vp10_get_pred_prob_single_ref_p1(cm, xd);
-  vpx_prob ref_single_p2 = vp10_get_pred_prob_single_ref_p2(cm, xd);
-
-  ref_frame_cost[INTRA_FRAME] = vp10_cost_bit(intra_inter_p, 0);
-  ref_frame_cost[LAST_FRAME] = ref_frame_cost[GOLDEN_FRAME] =
-    ref_frame_cost[ALTREF_FRAME] = vp10_cost_bit(intra_inter_p, 1);
-
-  ref_frame_cost[LAST_FRAME] += vp10_cost_bit(ref_single_p1, 0);
-  ref_frame_cost[GOLDEN_FRAME] += vp10_cost_bit(ref_single_p1, 1);
-  ref_frame_cost[ALTREF_FRAME] += vp10_cost_bit(ref_single_p1, 1);
-  ref_frame_cost[GOLDEN_FRAME] += vp10_cost_bit(ref_single_p2, 0);
-  ref_frame_cost[ALTREF_FRAME] += vp10_cost_bit(ref_single_p2, 1);
-}
-
-typedef struct {
-  MV_REFERENCE_FRAME ref_frame;
-  PREDICTION_MODE pred_mode;
-} REF_MODE;
-
-#define RT_INTER_MODES 8
-static const REF_MODE ref_mode_set[RT_INTER_MODES] = {
-    {LAST_FRAME, ZEROMV},
-    {LAST_FRAME, NEARESTMV},
-    {GOLDEN_FRAME, ZEROMV},
-    {LAST_FRAME, NEARMV},
-    {LAST_FRAME, NEWMV},
-    {GOLDEN_FRAME, NEARESTMV},
-    {GOLDEN_FRAME, NEARMV},
-    {GOLDEN_FRAME, NEWMV}
-};
-
-// TODO(jingning) placeholder for inter-frame non-RD mode decision.
-// this needs various further optimizations. to be continued..
-void vp10_pick_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
-                         TileDataEnc *tile_data,
-                         int mi_row, int mi_col, RD_COST *rd_cost,
-                         BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx) {
-  VP10_COMMON *const cm = &cpi->common;
-  SPEED_FEATURES *const sf = &cpi->sf;
-  TileInfo *const tile_info = &tile_data->tile_info;
-  MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  struct macroblockd_plane *const pd = &xd->plane[0];
-  PREDICTION_MODE best_mode = ZEROMV;
-  MV_REFERENCE_FRAME ref_frame, best_ref_frame = LAST_FRAME;
-  MV_REFERENCE_FRAME usable_ref_frame;
-  TX_SIZE best_tx_size = TX_SIZES;
-  INTERP_FILTER best_pred_filter = EIGHTTAP;
-  int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
-  struct buf_2d yv12_mb[4][MAX_MB_PLANE];
-  static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG,
-                                    VP9_ALT_FLAG };
-  RD_COST this_rdc, best_rdc;
-  uint8_t skip_txfm = SKIP_TXFM_NONE, best_mode_skip_txfm = SKIP_TXFM_NONE;
-  // var_y and sse_y are saved to be used in skipping checking
-  unsigned int var_y = UINT_MAX;
-  unsigned int sse_y = UINT_MAX;
-  // Reduce the intra cost penalty for small blocks (<=16x16).
-  const int reduction_fac = (bsize <= BLOCK_16X16) ?
-      ((bsize <= BLOCK_8X8) ? 4 : 2) : 0;
-  const int intra_cost_penalty = vp10_get_intra_cost_penalty(
-      cm->base_qindex, cm->y_dc_delta_q, cm->bit_depth) >> reduction_fac;
-  const int64_t inter_mode_thresh = RDCOST(x->rdmult, x->rddiv,
-                                           intra_cost_penalty, 0);
-  const int *const rd_threshes = cpi->rd.threshes[mbmi->segment_id][bsize];
-  const int *const rd_thresh_freq_fact = tile_data->thresh_freq_fact[bsize];
-  INTERP_FILTER filter_ref;
-  const int bsl = mi_width_log2_lookup[bsize];
-  const int pred_filter_search = cm->interp_filter == SWITCHABLE ?
-      (((mi_row + mi_col) >> bsl) +
-       get_chessboard_index(cm->current_video_frame)) & 0x1 : 0;
-  int const_motion[MAX_REF_FRAMES] = { 0 };
-  const int bh = num_4x4_blocks_high_lookup[bsize] << 2;
-  const int bw = num_4x4_blocks_wide_lookup[bsize] << 2;
-  // For speed 6, the result of interp filter is reused later in actual encoding
-  // process.
-  // tmp[3] points to dst buffer, and the other 3 point to allocated buffers.
-  PRED_BUFFER tmp[4];
-  DECLARE_ALIGNED(16, uint8_t, pred_buf[3 * 64 * 64]);
-#if CONFIG_VP9_HIGHBITDEPTH
-  DECLARE_ALIGNED(16, uint16_t, pred_buf_16[3 * 64 * 64]);
-#endif
-  struct buf_2d orig_dst = pd->dst;
-  PRED_BUFFER *best_pred = NULL;
-  PRED_BUFFER *this_mode_pred = NULL;
-  const int pixels_in_block = bh * bw;
-  int reuse_inter_pred = cpi->sf.reuse_inter_pred_sby && ctx->pred_pixel_ready;
-  int ref_frame_skip_mask = 0;
-  int idx;
-  int best_pred_sad = INT_MAX;
-  int best_early_term = 0;
-  int ref_frame_cost[MAX_REF_FRAMES];
-
-  init_ref_frame_cost(cm, xd, ref_frame_cost);
-
-  if (reuse_inter_pred) {
-    int i;
-    for (i = 0; i < 3; i++) {
-#if CONFIG_VP9_HIGHBITDEPTH
-      if (cm->use_highbitdepth)
-        tmp[i].data = CONVERT_TO_BYTEPTR(&pred_buf_16[pixels_in_block * i]);
-      else
-        tmp[i].data = &pred_buf[pixels_in_block * i];
-#else
-      tmp[i].data = &pred_buf[pixels_in_block * i];
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-      tmp[i].stride = bw;
-      tmp[i].in_use = 0;
-    }
-    tmp[3].data = pd->dst.buf;
-    tmp[3].stride = pd->dst.stride;
-    tmp[3].in_use = 0;
-  }
-
-  x->skip_encode = cpi->sf.skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
-  x->skip = 0;
-
-  if (xd->up_available)
-    filter_ref = xd->mi[-xd->mi_stride]->mbmi.interp_filter;
-  else if (xd->left_available)
-    filter_ref = xd->mi[-1]->mbmi.interp_filter;
-  else
-    filter_ref = cm->interp_filter;
-
-  // initialize mode decisions
-  vp10_rd_cost_reset(&best_rdc);
-  vp10_rd_cost_reset(rd_cost);
-  mbmi->sb_type = bsize;
-  mbmi->ref_frame[0] = NONE;
-  mbmi->ref_frame[1] = NONE;
-  mbmi->tx_size = MIN(max_txsize_lookup[bsize],
-                      tx_mode_to_biggest_tx_size[cm->tx_mode]);
-
-#if CONFIG_VP9_TEMPORAL_DENOISING
-  vp10_denoiser_reset_frame_stats(ctx);
-#endif
-
-  if (cpi->rc.frames_since_golden == 0) {
-    usable_ref_frame = LAST_FRAME;
-  } else {
-    usable_ref_frame = GOLDEN_FRAME;
-  }
-
-  for (ref_frame = LAST_FRAME; ref_frame <= usable_ref_frame; ++ref_frame) {
-    const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame);
-
-    x->pred_mv_sad[ref_frame] = INT_MAX;
-    frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
-    frame_mv[ZEROMV][ref_frame].as_int = 0;
-
-    if ((cpi->ref_frame_flags & flag_list[ref_frame]) && (yv12 != NULL)) {
-      int_mv *const candidates = x->mbmi_ext->ref_mvs[ref_frame];
-      const struct scale_factors *const sf = &cm->frame_refs[ref_frame - 1].sf;
-
-      vp10_setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col,
-                           sf, sf);
-
-      if (cm->use_prev_frame_mvs)
-        vp10_find_mv_refs(cm, xd, xd->mi[0], ref_frame,
-                         candidates, mi_row, mi_col, NULL, NULL,
-                         x->mbmi_ext->mode_context);
-      else
-        const_motion[ref_frame] = mv_refs_rt(cm, x, xd, tile_info,
-                                             xd->mi[0],
-                                             ref_frame, candidates,
-                                             mi_row, mi_col);
-
-      vp10_find_best_ref_mvs(xd, cm->allow_high_precision_mv, candidates,
-                            &frame_mv[NEARESTMV][ref_frame],
-                            &frame_mv[NEARMV][ref_frame]);
-
-      if (!vp10_is_scaled(sf) && bsize >= BLOCK_8X8)
-        vp10_mv_pred(cpi, x, yv12_mb[ref_frame][0].buf, yv12->y_stride,
-                    ref_frame, bsize);
-    } else {
-      ref_frame_skip_mask |= (1 << ref_frame);
-    }
-  }
-
-  for (idx = 0; idx < RT_INTER_MODES; ++idx) {
-    int rate_mv = 0;
-    int mode_rd_thresh;
-    int mode_index;
-    int i;
-    PREDICTION_MODE this_mode = ref_mode_set[idx].pred_mode;
-    int64_t this_sse;
-    int is_skippable;
-    int this_early_term = 0;
-
-    if (!(cpi->sf.inter_mode_mask[bsize] & (1 << this_mode)))
-      continue;
-
-    ref_frame = ref_mode_set[idx].ref_frame;
-    if (!(cpi->ref_frame_flags & flag_list[ref_frame]))
-      continue;
-    if (const_motion[ref_frame] && this_mode == NEARMV)
-      continue;
-
-    i = (ref_frame == LAST_FRAME) ? GOLDEN_FRAME : LAST_FRAME;
-    if ((cpi->ref_frame_flags & flag_list[i]) && sf->reference_masking)
-      if (x->pred_mv_sad[ref_frame] > (x->pred_mv_sad[i] << 1))
-        ref_frame_skip_mask |= (1 << ref_frame);
-    if (ref_frame_skip_mask & (1 << ref_frame))
-      continue;
-
-    // Select prediction reference frames.
-    for (i = 0; i < MAX_MB_PLANE; i++)
-      xd->plane[i].pre[0] = yv12_mb[ref_frame][i];
-
-    mbmi->ref_frame[0] = ref_frame;
-    set_ref_ptrs(cm, xd, ref_frame, NONE);
-
-    mode_index = mode_idx[ref_frame][INTER_OFFSET(this_mode)];
-    mode_rd_thresh = best_mode_skip_txfm ?
-            rd_threshes[mode_index] << 1 : rd_threshes[mode_index];
-    if (rd_less_than_thresh(best_rdc.rdcost, mode_rd_thresh,
-                            rd_thresh_freq_fact[mode_index]))
-      continue;
-
-    if (this_mode == NEWMV) {
-      if (ref_frame > LAST_FRAME) {
-        int tmp_sad;
-        int dis, cost_list[5];
-
-        if (bsize < BLOCK_16X16)
-          continue;
-
-        tmp_sad = vp10_int_pro_motion_estimation(cpi, x, bsize, mi_row, mi_col);
-
-        if (tmp_sad > x->pred_mv_sad[LAST_FRAME])
-          continue;
-        if (tmp_sad + (num_pels_log2_lookup[bsize] << 4) > best_pred_sad)
-          continue;
-
-        frame_mv[NEWMV][ref_frame].as_int = mbmi->mv[0].as_int;
-        rate_mv = vp10_mv_bit_cost(&frame_mv[NEWMV][ref_frame].as_mv,
-          &x->mbmi_ext->ref_mvs[ref_frame][0].as_mv,
-          x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
-        frame_mv[NEWMV][ref_frame].as_mv.row >>= 3;
-        frame_mv[NEWMV][ref_frame].as_mv.col >>= 3;
-
-        cpi->find_fractional_mv_step(x, &frame_mv[NEWMV][ref_frame].as_mv,
-          &x->mbmi_ext->ref_mvs[ref_frame][0].as_mv,
-          cpi->common.allow_high_precision_mv,
-          x->errorperbit,
-          &cpi->fn_ptr[bsize],
-          cpi->sf.mv.subpel_force_stop,
-          cpi->sf.mv.subpel_iters_per_step,
-          cond_cost_list(cpi, cost_list),
-          x->nmvjointcost, x->mvcost, &dis,
-          &x->pred_sse[ref_frame], NULL, 0, 0);
-      } else if (!combined_motion_search(cpi, x, bsize, mi_row, mi_col,
-        &frame_mv[NEWMV][ref_frame], &rate_mv, best_rdc.rdcost)) {
-        continue;
-      }
-    }
-
-    if (this_mode == NEWMV && ref_frame == LAST_FRAME &&
-        frame_mv[NEWMV][LAST_FRAME].as_int != INVALID_MV) {
-      const int pre_stride = xd->plane[0].pre[0].stride;
-      const uint8_t * const pre_buf = xd->plane[0].pre[0].buf +
-          (frame_mv[NEWMV][LAST_FRAME].as_mv.row >> 3) * pre_stride +
-          (frame_mv[NEWMV][LAST_FRAME].as_mv.col >> 3);
-      best_pred_sad = cpi->fn_ptr[bsize].sdf(x->plane[0].src.buf,
-                                   x->plane[0].src.stride,
-                                   pre_buf, pre_stride);
-      x->pred_mv_sad[LAST_FRAME] = best_pred_sad;
-    }
-
-    if (this_mode != NEARESTMV &&
-        frame_mv[this_mode][ref_frame].as_int ==
-            frame_mv[NEARESTMV][ref_frame].as_int)
-      continue;
-
-    mbmi->mode = this_mode;
-    mbmi->mv[0].as_int = frame_mv[this_mode][ref_frame].as_int;
-
-    // Search for the best prediction filter type, when the resulting
-    // motion vector is at sub-pixel accuracy level for luma component, i.e.,
-    // the last three bits are all zeros.
-    if (reuse_inter_pred) {
-      if (!this_mode_pred) {
-        this_mode_pred = &tmp[3];
-      } else {
-        this_mode_pred = &tmp[get_pred_buffer(tmp, 3)];
-        pd->dst.buf = this_mode_pred->data;
-        pd->dst.stride = bw;
-      }
-    }
-
-    if ((this_mode == NEWMV || filter_ref == SWITCHABLE) && pred_filter_search
-        && (ref_frame == LAST_FRAME)
-        && (((mbmi->mv[0].as_mv.row | mbmi->mv[0].as_mv.col) & 0x07) != 0)) {
-      int pf_rate[3];
-      int64_t pf_dist[3];
-      unsigned int pf_var[3];
-      unsigned int pf_sse[3];
-      TX_SIZE pf_tx_size[3];
-      int64_t best_cost = INT64_MAX;
-      INTERP_FILTER best_filter = SWITCHABLE, filter;
-      PRED_BUFFER *current_pred = this_mode_pred;
-
-      for (filter = EIGHTTAP; filter <= EIGHTTAP_SMOOTH; ++filter) {
-        int64_t cost;
-        mbmi->interp_filter = filter;
-        vp10_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
-        model_rd_for_sb_y(cpi, bsize, x, xd, &pf_rate[filter], &pf_dist[filter],
-                          &pf_var[filter], &pf_sse[filter]);
-        pf_rate[filter] += vp10_get_switchable_rate(cpi, xd);
-        cost = RDCOST(x->rdmult, x->rddiv, pf_rate[filter], pf_dist[filter]);
-        pf_tx_size[filter] = mbmi->tx_size;
-        if (cost < best_cost) {
-          best_filter = filter;
-          best_cost = cost;
-          skip_txfm = x->skip_txfm[0];
-
-          if (reuse_inter_pred) {
-            if (this_mode_pred != current_pred) {
-              free_pred_buffer(this_mode_pred);
-              this_mode_pred = current_pred;
-            }
-
-            if (filter < EIGHTTAP_SHARP) {
-              current_pred = &tmp[get_pred_buffer(tmp, 3)];
-              pd->dst.buf = current_pred->data;
-              pd->dst.stride = bw;
-            }
-          }
-        }
-      }
-
-      if (reuse_inter_pred && this_mode_pred != current_pred)
-        free_pred_buffer(current_pred);
-
-      mbmi->interp_filter = best_filter;
-      mbmi->tx_size = pf_tx_size[best_filter];
-      this_rdc.rate = pf_rate[best_filter];
-      this_rdc.dist = pf_dist[best_filter];
-      var_y = pf_var[best_filter];
-      sse_y = pf_sse[best_filter];
-      x->skip_txfm[0] = skip_txfm;
-      if (reuse_inter_pred) {
-        pd->dst.buf = this_mode_pred->data;
-        pd->dst.stride = this_mode_pred->stride;
-      }
-    } else {
-      mbmi->interp_filter = (filter_ref == SWITCHABLE) ? EIGHTTAP : filter_ref;
-      vp10_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
-
-      // For large partition blocks, extra testing is done.
-      if (bsize > BLOCK_32X32 &&
-        !cyclic_refresh_segment_id_boosted(xd->mi[0]->mbmi.segment_id) &&
-        cm->base_qindex) {
-        model_rd_for_sb_y_large(cpi, bsize, x, xd, &this_rdc.rate,
-                                &this_rdc.dist, &var_y, &sse_y, mi_row, mi_col,
-                                &this_early_term);
-      } else {
-        model_rd_for_sb_y(cpi, bsize, x, xd, &this_rdc.rate, &this_rdc.dist,
-                          &var_y, &sse_y);
-      }
-    }
-
-    if (!this_early_term) {
-      this_sse = (int64_t)sse_y;
-      block_yrd(cpi, x, &this_rdc.rate, &this_rdc.dist, &is_skippable,
-                &this_sse, 0, bsize, MIN(mbmi->tx_size, TX_16X16));
-      x->skip_txfm[0] = is_skippable;
-      if (is_skippable) {
-        this_rdc.rate = vp10_cost_bit(vp10_get_skip_prob(cm, xd), 1);
-      } else {
-        if (RDCOST(x->rdmult, x->rddiv, this_rdc.rate, this_rdc.dist) <
-            RDCOST(x->rdmult, x->rddiv, 0, this_sse)) {
-          this_rdc.rate += vp10_cost_bit(vp10_get_skip_prob(cm, xd), 0);
-        } else {
-          this_rdc.rate = vp10_cost_bit(vp10_get_skip_prob(cm, xd), 1);
-          this_rdc.dist = this_sse;
-          x->skip_txfm[0] = SKIP_TXFM_AC_DC;
-        }
-      }
-
-      if (cm->interp_filter == SWITCHABLE) {
-        if ((mbmi->mv[0].as_mv.row | mbmi->mv[0].as_mv.col) & 0x07)
-          this_rdc.rate += vp10_get_switchable_rate(cpi, xd);
-      }
-    } else {
-      this_rdc.rate += cm->interp_filter == SWITCHABLE ?
-          vp10_get_switchable_rate(cpi, xd) : 0;
-      this_rdc.rate += vp10_cost_bit(vp10_get_skip_prob(cm, xd), 1);
-    }
-
-    if (x->color_sensitivity[0] || x->color_sensitivity[1]) {
-      int uv_rate = 0;
-      int64_t uv_dist = 0;
-      if (x->color_sensitivity[0])
-        vp10_build_inter_predictors_sbp(xd, mi_row, mi_col, bsize, 1);
-      if (x->color_sensitivity[1])
-        vp10_build_inter_predictors_sbp(xd, mi_row, mi_col, bsize, 2);
-      model_rd_for_sb_uv(cpi, bsize, x, xd, &uv_rate, &uv_dist,
-                         &var_y, &sse_y);
-      this_rdc.rate += uv_rate;
-      this_rdc.dist += uv_dist;
-    }
-
-    this_rdc.rate += rate_mv;
-    this_rdc.rate +=
-        cpi->inter_mode_cost[x->mbmi_ext->mode_context[ref_frame]][INTER_OFFSET(
-            this_mode)];
-    this_rdc.rate += ref_frame_cost[ref_frame];
-    this_rdc.rdcost = RDCOST(x->rdmult, x->rddiv, this_rdc.rate, this_rdc.dist);
-
-    // Skipping checking: test to see if this block can be reconstructed by
-    // prediction only.
-    if (cpi->allow_encode_breakout) {
-      encode_breakout_test(cpi, x, bsize, mi_row, mi_col, ref_frame, this_mode,
-                           var_y, sse_y, yv12_mb, &this_rdc.rate,
-                           &this_rdc.dist);
-      if (x->skip) {
-        this_rdc.rate += rate_mv;
-        this_rdc.rdcost = RDCOST(x->rdmult, x->rddiv, this_rdc.rate,
-                                 this_rdc.dist);
-      }
-    }
-
-#if CONFIG_VP9_TEMPORAL_DENOISING
-    if (cpi->oxcf.noise_sensitivity > 0)
-      vp10_denoiser_update_frame_stats(mbmi, sse_y, this_mode, ctx);
-#else
-    (void)ctx;
-#endif
-
-    if (this_rdc.rdcost < best_rdc.rdcost || x->skip) {
-      best_rdc = this_rdc;
-      best_mode = this_mode;
-      best_pred_filter = mbmi->interp_filter;
-      best_tx_size = mbmi->tx_size;
-      best_ref_frame = ref_frame;
-      best_mode_skip_txfm = x->skip_txfm[0];
-      best_early_term = this_early_term;
-
-      if (reuse_inter_pred) {
-        free_pred_buffer(best_pred);
-        best_pred = this_mode_pred;
-      }
-    } else {
-      if (reuse_inter_pred)
-        free_pred_buffer(this_mode_pred);
-    }
-
-    if (x->skip)
-      break;
-
-    // If early termination flag is 1 and at least 2 modes are checked,
-    // the mode search is terminated.
-    if (best_early_term && idx > 0) {
-      x->skip = 1;
-      break;
-    }
-  }
-
-  mbmi->mode          = best_mode;
-  mbmi->interp_filter = best_pred_filter;
-  mbmi->tx_size       = best_tx_size;
-  mbmi->ref_frame[0]  = best_ref_frame;
-  mbmi->mv[0].as_int  = frame_mv[best_mode][best_ref_frame].as_int;
-  xd->mi[0]->bmi[0].as_mv[0].as_int = mbmi->mv[0].as_int;
-  x->skip_txfm[0] = best_mode_skip_txfm;
-
-  // Perform intra prediction search, if the best SAD is above a certain
-  // threshold.
-  if (best_rdc.rdcost == INT64_MAX ||
-      (!x->skip && best_rdc.rdcost > inter_mode_thresh &&
-       bsize <= cpi->sf.max_intra_bsize)) {
-    struct estimate_block_intra_args args = { cpi, x, DC_PRED, 0, 0 };
-    const TX_SIZE intra_tx_size =
-        MIN(max_txsize_lookup[bsize],
-            tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
-    int i;
-    TX_SIZE best_intra_tx_size = TX_SIZES;
-
-    if (reuse_inter_pred && best_pred != NULL) {
-      if (best_pred->data == orig_dst.buf) {
-        this_mode_pred = &tmp[get_pred_buffer(tmp, 3)];
-#if CONFIG_VP9_HIGHBITDEPTH
-        if (cm->use_highbitdepth)
-          vpx_highbd_convolve_copy(best_pred->data, best_pred->stride,
-                                   this_mode_pred->data, this_mode_pred->stride,
-                                   NULL, 0, NULL, 0, bw, bh, xd->bd);
-        else
-          vpx_convolve_copy(best_pred->data, best_pred->stride,
-                          this_mode_pred->data, this_mode_pred->stride,
-                          NULL, 0, NULL, 0, bw, bh);
-#else
-        vpx_convolve_copy(best_pred->data, best_pred->stride,
-                          this_mode_pred->data, this_mode_pred->stride,
-                          NULL, 0, NULL, 0, bw, bh);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-        best_pred = this_mode_pred;
-      }
-    }
-    pd->dst = orig_dst;
-
-    for (i = 0; i < 4; ++i) {
-      const PREDICTION_MODE this_mode = intra_mode_list[i];
-      THR_MODES mode_index = mode_idx[INTRA_FRAME][mode_offset(this_mode)];
-      int mode_rd_thresh = rd_threshes[mode_index];
-
-      if (!((1 << this_mode) & cpi->sf.intra_y_mode_bsize_mask[bsize]))
-        continue;
-
-      if (rd_less_than_thresh(best_rdc.rdcost, mode_rd_thresh,
-                              rd_thresh_freq_fact[mode_index]))
-        continue;
-
-      mbmi->mode = this_mode;
-      mbmi->ref_frame[0] = INTRA_FRAME;
-      args.mode = this_mode;
-      args.rate = 0;
-      args.dist = 0;
-      mbmi->tx_size = intra_tx_size;
-      vp10_foreach_transformed_block_in_plane(xd, bsize, 0,
-                                             estimate_block_intra, &args);
-      this_rdc.rate = args.rate;
-      this_rdc.dist = args.dist;
-      this_rdc.rate += cpi->mbmode_cost[this_mode];
-      this_rdc.rate += ref_frame_cost[INTRA_FRAME];
-      this_rdc.rate += intra_cost_penalty;
-      this_rdc.rdcost = RDCOST(x->rdmult, x->rddiv,
-                               this_rdc.rate, this_rdc.dist);
-
-      if (this_rdc.rdcost < best_rdc.rdcost) {
-        best_rdc = this_rdc;
-        best_mode = this_mode;
-        best_intra_tx_size = mbmi->tx_size;
-        best_ref_frame = INTRA_FRAME;
-        mbmi->uv_mode = this_mode;
-        mbmi->mv[0].as_int = INVALID_MV;
-        best_mode_skip_txfm = x->skip_txfm[0];
-      }
-    }
-
-    // Reset mb_mode_info to the best inter mode.
-    if (best_ref_frame != INTRA_FRAME) {
-      mbmi->tx_size = best_tx_size;
-    } else {
-      mbmi->tx_size = best_intra_tx_size;
-    }
-  }
-
-  pd->dst = orig_dst;
-  mbmi->mode = best_mode;
-  mbmi->ref_frame[0] = best_ref_frame;
-  x->skip_txfm[0] = best_mode_skip_txfm;
-
-  if (reuse_inter_pred && best_pred != NULL) {
-    if (best_pred->data != orig_dst.buf && is_inter_mode(mbmi->mode)) {
-#if CONFIG_VP9_HIGHBITDEPTH
-      if (cm->use_highbitdepth)
-        vpx_highbd_convolve_copy(best_pred->data, best_pred->stride,
-                                 pd->dst.buf, pd->dst.stride, NULL, 0,
-                                 NULL, 0, bw, bh, xd->bd);
-      else
-        vpx_convolve_copy(best_pred->data, best_pred->stride,
-                          pd->dst.buf, pd->dst.stride, NULL, 0,
-                          NULL, 0, bw, bh);
-#else
-      vpx_convolve_copy(best_pred->data, best_pred->stride,
-                        pd->dst.buf, pd->dst.stride, NULL, 0,
-                        NULL, 0, bw, bh);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-    }
-  }
-
-  if (cpi->sf.adaptive_rd_thresh) {
-    THR_MODES best_mode_idx = mode_idx[best_ref_frame][mode_offset(mbmi->mode)];
-
-    if (best_ref_frame == INTRA_FRAME) {
-      // Only consider the modes that are included in the intra_mode_list.
-      int intra_modes = sizeof(intra_mode_list)/sizeof(PREDICTION_MODE);
-      int i;
-
-      // TODO(yunqingwang): Check intra mode mask and only update freq_fact
-      // for those valid modes.
-      for (i = 0; i < intra_modes; i++) {
-        update_thresh_freq_fact(cpi, tile_data, bsize, INTRA_FRAME,
-                                best_mode_idx, intra_mode_list[i]);
-      }
-    } else {
-      for (ref_frame = LAST_FRAME; ref_frame <= GOLDEN_FRAME; ++ref_frame) {
-        PREDICTION_MODE this_mode;
-        if (best_ref_frame != ref_frame) continue;
-        for (this_mode = NEARESTMV; this_mode <= NEWMV; ++this_mode) {
-          update_thresh_freq_fact(cpi, tile_data, bsize, ref_frame,
-                                  best_mode_idx, this_mode);
-        }
-      }
-    }
-  }
-
-  *rd_cost = best_rdc;
-}
-
-void vp10_pick_inter_mode_sub8x8(VP10_COMP *cpi, MACROBLOCK *x,
-                                int mi_row, int mi_col, RD_COST *rd_cost,
-                                BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx) {
-  VP10_COMMON *const cm = &cpi->common;
-  SPEED_FEATURES *const sf = &cpi->sf;
-  MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
-  MB_MODE_INFO_EXT *const mbmi_ext = x->mbmi_ext;
-  const struct segmentation *const seg = &cm->seg;
-  MV_REFERENCE_FRAME ref_frame, second_ref_frame = NONE;
-  MV_REFERENCE_FRAME best_ref_frame = NONE;
-  unsigned char segment_id = mbmi->segment_id;
-  struct buf_2d yv12_mb[4][MAX_MB_PLANE];
-  static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG,
-                                    VP9_ALT_FLAG };
-  int64_t best_rd = INT64_MAX;
-  b_mode_info bsi[MAX_REF_FRAMES][4];
-  int ref_frame_skip_mask = 0;
-  const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
-  const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
-  int idx, idy;
-
-  x->skip_encode = sf->skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
-  ctx->pred_pixel_ready = 0;
-
-  for (ref_frame = LAST_FRAME; ref_frame <= GOLDEN_FRAME; ++ref_frame) {
-    const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame);
-    int_mv dummy_mv[2];
-    x->pred_mv_sad[ref_frame] = INT_MAX;
-
-    if ((cpi->ref_frame_flags & flag_list[ref_frame]) && (yv12 != NULL)) {
-      int_mv *const candidates = mbmi_ext->ref_mvs[ref_frame];
-      const struct scale_factors *const sf =
-                             &cm->frame_refs[ref_frame - 1].sf;
-      vp10_setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col,
-                           sf, sf);
-      vp10_find_mv_refs(cm, xd, xd->mi[0], ref_frame,
-                       candidates, mi_row, mi_col, NULL, NULL,
-                       mbmi_ext->mode_context);
-
-      vp10_find_best_ref_mvs(xd, cm->allow_high_precision_mv, candidates,
-                            &dummy_mv[0], &dummy_mv[1]);
-    } else {
-      ref_frame_skip_mask |= (1 << ref_frame);
-    }
-  }
-
-  mbmi->sb_type = bsize;
-  mbmi->tx_size = TX_4X4;
-  mbmi->uv_mode = DC_PRED;
-  mbmi->ref_frame[0] = LAST_FRAME;
-  mbmi->ref_frame[1] = NONE;
-  mbmi->interp_filter = cm->interp_filter == SWITCHABLE ? EIGHTTAP
-                                                        : cm->interp_filter;
-
-  for (ref_frame = LAST_FRAME; ref_frame <= GOLDEN_FRAME; ++ref_frame) {
-    int64_t this_rd = 0;
-    int plane;
-
-    if (ref_frame_skip_mask & (1 << ref_frame))
-      continue;
-
-    // TODO(jingning, agrange): Scaling reference frame not supported for
-    // sub8x8 blocks. Is this supported now?
-    if (ref_frame > INTRA_FRAME &&
-        vp10_is_scaled(&cm->frame_refs[ref_frame - 1].sf))
-      continue;
-
-    // If the segment reference frame feature is enabled....
-    // then do nothing if the current ref frame is not allowed..
-    if (segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME) &&
-        get_segdata(seg, segment_id, SEG_LVL_REF_FRAME) != (int)ref_frame)
-      continue;
-
-    mbmi->ref_frame[0] = ref_frame;
-    x->skip = 0;
-    set_ref_ptrs(cm, xd, ref_frame, second_ref_frame);
-
-    // Select prediction reference frames.
-    for (plane = 0; plane < MAX_MB_PLANE; plane++)
-      xd->plane[plane].pre[0] = yv12_mb[ref_frame][plane];
-
-    for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
-      for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
-        int_mv b_mv[MB_MODE_COUNT];
-        int64_t b_best_rd = INT64_MAX;
-        const int i = idy * 2 + idx;
-        PREDICTION_MODE this_mode;
-        RD_COST this_rdc;
-        unsigned int var_y, sse_y;
-
-        struct macroblock_plane *p = &x->plane[0];
-        struct macroblockd_plane *pd = &xd->plane[0];
-
-        const struct buf_2d orig_src = p->src;
-        const struct buf_2d orig_dst = pd->dst;
-        struct buf_2d orig_pre[2];
-        memcpy(orig_pre, xd->plane[0].pre, sizeof(orig_pre));
-
-        // set buffer pointers for sub8x8 motion search.
-        p->src.buf =
-            &p->src.buf[vp10_raster_block_offset(BLOCK_8X8, i, p->src.stride)];
-        pd->dst.buf =
-            &pd->dst.buf[vp10_raster_block_offset(BLOCK_8X8, i, pd->dst.stride)];
-        pd->pre[0].buf =
-            &pd->pre[0].buf[vp10_raster_block_offset(BLOCK_8X8,
-                                                    i, pd->pre[0].stride)];
-
-        b_mv[ZEROMV].as_int = 0;
-        b_mv[NEWMV].as_int = INVALID_MV;
-        vp10_append_sub8x8_mvs_for_idx(cm, xd, i, 0, mi_row, mi_col,
-                                      &b_mv[NEARESTMV],
-                                      &b_mv[NEARMV],
-                                      mbmi_ext->mode_context);
-
-        for (this_mode = NEARESTMV; this_mode <= NEWMV; ++this_mode) {
-          int b_rate = 0;
-          xd->mi[0]->bmi[i].as_mv[0].as_int = b_mv[this_mode].as_int;
-
-          if (this_mode == NEWMV) {
-            const int step_param = cpi->sf.mv.fullpel_search_step_param;
-            MV mvp_full;
-            MV tmp_mv;
-            int cost_list[5];
-            const int tmp_col_min = x->mv_col_min;
-            const int tmp_col_max = x->mv_col_max;
-            const int tmp_row_min = x->mv_row_min;
-            const int tmp_row_max = x->mv_row_max;
-            int dummy_dist;
-
-            if (i == 0) {
-              mvp_full.row = b_mv[NEARESTMV].as_mv.row >> 3;
-              mvp_full.col = b_mv[NEARESTMV].as_mv.col >> 3;
-            } else {
-              mvp_full.row = xd->mi[0]->bmi[0].as_mv[0].as_mv.row >> 3;
-              mvp_full.col = xd->mi[0]->bmi[0].as_mv[0].as_mv.col >> 3;
-            }
-
-            vp10_set_mv_search_range(x, &mbmi_ext->ref_mvs[0]->as_mv);
-
-            vp10_full_pixel_search(
-                cpi, x, bsize, &mvp_full, step_param, x->sadperbit4,
-                cond_cost_list(cpi, cost_list),
-                &mbmi_ext->ref_mvs[ref_frame][0].as_mv, &tmp_mv,
-                INT_MAX, 0);
-
-            x->mv_col_min = tmp_col_min;
-            x->mv_col_max = tmp_col_max;
-            x->mv_row_min = tmp_row_min;
-            x->mv_row_max = tmp_row_max;
-
-            // calculate the bit cost on motion vector
-            mvp_full.row = tmp_mv.row * 8;
-            mvp_full.col = tmp_mv.col * 8;
-
-            b_rate += vp10_mv_bit_cost(&mvp_full,
-                                      &mbmi_ext->ref_mvs[ref_frame][0].as_mv,
-                                      x->nmvjointcost, x->mvcost,
-                                      MV_COST_WEIGHT);
-
-            b_rate += cpi->inter_mode_cost[x->mbmi_ext->mode_context[ref_frame]]
-                                          [INTER_OFFSET(NEWMV)];
-            if (RDCOST(x->rdmult, x->rddiv, b_rate, 0) > b_best_rd)
-              continue;
-
-            cpi->find_fractional_mv_step(x, &tmp_mv,
-                                         &mbmi_ext->ref_mvs[ref_frame][0].as_mv,
-                                         cpi->common.allow_high_precision_mv,
-                                         x->errorperbit,
-                                         &cpi->fn_ptr[bsize],
-                                         cpi->sf.mv.subpel_force_stop,
-                                         cpi->sf.mv.subpel_iters_per_step,
-                                         cond_cost_list(cpi, cost_list),
-                                         x->nmvjointcost, x->mvcost,
-                                         &dummy_dist,
-                                         &x->pred_sse[ref_frame], NULL, 0, 0);
-
-            xd->mi[0]->bmi[i].as_mv[0].as_mv = tmp_mv;
-          } else {
-            b_rate += cpi->inter_mode_cost[x->mbmi_ext->mode_context[ref_frame]]
-                                          [INTER_OFFSET(this_mode)];
-          }
-
-#if CONFIG_VP9_HIGHBITDEPTH
-          if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-            vp10_highbd_build_inter_predictor(pd->pre[0].buf, pd->pre[0].stride,
-                                    pd->dst.buf, pd->dst.stride,
-                                    &xd->mi[0]->bmi[i].as_mv[0].as_mv,
-                                    &xd->block_refs[0]->sf,
-                                    4 * num_4x4_blocks_wide,
-                                    4 * num_4x4_blocks_high, 0,
-                                    vp10_filter_kernels[mbmi->interp_filter],
-                                    MV_PRECISION_Q3,
-                                    mi_col * MI_SIZE + 4 * (i & 0x01),
-                                    mi_row * MI_SIZE + 4 * (i >> 1), xd->bd);
-          } else {
-#endif
-            vp10_build_inter_predictor(pd->pre[0].buf, pd->pre[0].stride,
-                                     pd->dst.buf, pd->dst.stride,
-                                     &xd->mi[0]->bmi[i].as_mv[0].as_mv,
-                                     &xd->block_refs[0]->sf,
-                                     4 * num_4x4_blocks_wide,
-                                     4 * num_4x4_blocks_high, 0,
-                                     vp10_filter_kernels[mbmi->interp_filter],
-                                     MV_PRECISION_Q3,
-                                     mi_col * MI_SIZE + 4 * (i & 0x01),
-                                     mi_row * MI_SIZE + 4 * (i >> 1));
-
-#if CONFIG_VP9_HIGHBITDEPTH
-          }
-#endif
-
-          model_rd_for_sb_y(cpi, bsize, x, xd, &this_rdc.rate, &this_rdc.dist,
-                            &var_y, &sse_y);
-
-          this_rdc.rate += b_rate;
-          this_rdc.rdcost = RDCOST(x->rdmult, x->rddiv,
-                                   this_rdc.rate, this_rdc.dist);
-          if (this_rdc.rdcost < b_best_rd) {
-            b_best_rd = this_rdc.rdcost;
-            bsi[ref_frame][i].as_mode = this_mode;
-            bsi[ref_frame][i].as_mv[0].as_mv = xd->mi[0]->bmi[i].as_mv[0].as_mv;
-          }
-        }  // mode search
-
-        // restore source and prediction buffer pointers.
-        p->src = orig_src;
-        pd->pre[0] = orig_pre[0];
-        pd->dst = orig_dst;
-        this_rd += b_best_rd;
-
-        xd->mi[0]->bmi[i] = bsi[ref_frame][i];
-        if (num_4x4_blocks_wide > 1)
-          xd->mi[0]->bmi[i + 1] = xd->mi[0]->bmi[i];
-        if (num_4x4_blocks_high > 1)
-          xd->mi[0]->bmi[i + 2] = xd->mi[0]->bmi[i];
-      }
-    }  // loop through sub8x8 blocks
-
-    if (this_rd < best_rd) {
-      best_rd = this_rd;
-      best_ref_frame = ref_frame;
-    }
-  }  // reference frames
-
-  mbmi->tx_size = TX_4X4;
-  mbmi->ref_frame[0] = best_ref_frame;
-  for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
-    for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
-      const int block = idy * 2 + idx;
-      xd->mi[0]->bmi[block] = bsi[best_ref_frame][block];
-      if (num_4x4_blocks_wide > 1)
-        xd->mi[0]->bmi[block + 1] = bsi[best_ref_frame][block];
-      if (num_4x4_blocks_high > 1)
-        xd->mi[0]->bmi[block + 2] = bsi[best_ref_frame][block];
-    }
-  }
-  mbmi->mode = xd->mi[0]->bmi[3].as_mode;
-  ctx->mic = *(xd->mi[0]);
-  ctx->mbmi_ext = *x->mbmi_ext;
-  ctx->skip_txfm[0] = SKIP_TXFM_NONE;
-  ctx->skip = 0;
-  // Dummy assignment for speed -5. No effect in speed -6.
-  rd_cost->rdcost = best_rd;
-}
diff --git a/vp10/encoder/pickmode.h b/vp10/encoder/pickmode.h
deleted file mode 100644
index b596f53..0000000
--- a/vp10/encoder/pickmode.h
+++ /dev/null
@@ -1,38 +0,0 @@
-/*
- *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-#ifndef VP10_ENCODER_PICKMODE_H_
-#define VP10_ENCODER_PICKMODE_H_
-
-#include "vp10/encoder/encoder.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-void vp10_pick_intra_mode(VP10_COMP *cpi, MACROBLOCK *x, RD_COST *rd_cost,
-                         BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx);
-
-void vp10_pick_inter_mode(VP10_COMP *cpi, MACROBLOCK *x,
-                         TileDataEnc *tile_data,
-                         int mi_row, int mi_col, RD_COST *rd_cost,
-                         BLOCK_SIZE bsize,
-                         PICK_MODE_CONTEXT *ctx);
-
-void vp10_pick_inter_mode_sub8x8(VP10_COMP *cpi, MACROBLOCK *x,
-                                int mi_row, int mi_col, RD_COST *rd_cost,
-                                BLOCK_SIZE bsize,
-                                PICK_MODE_CONTEXT *ctx);
-
-#ifdef __cplusplus
-}  // extern "C"
-#endif
-
-#endif  // VP10_ENCODER_PICKMODE_H_
diff --git a/vp10/encoder/ratectrl.c b/vp10/encoder/ratectrl.c
index 0cd0bff..477a3f5 100644
--- a/vp10/encoder/ratectrl.c
+++ b/vp10/encoder/ratectrl.c
@@ -106,8 +106,7 @@
 static int get_minq_index(double maxq, double x3, double x2, double x1,
                           vpx_bit_depth_t bit_depth) {
   int i;
-  const double minqtarget = MIN(((x3 * maxq + x2) * maxq + x1) * maxq,
-                                maxq);
+  const double minqtarget = VPXMIN(((x3 * maxq + x2) * maxq + x1) * maxq, maxq);
 
   // Special case handling to deal with the step from q2.0
   // down to lossless mode represented by q 1.0.
@@ -192,15 +191,15 @@
                            vpx_bit_depth_t bit_depth) {
   const int bpm = (int)(vp10_rc_bits_per_mb(frame_type, q, correction_factor,
                                            bit_depth));
-  return MAX(FRAME_OVERHEAD_BITS,
-             (int)((uint64_t)bpm * mbs) >> BPER_MB_NORMBITS);
+  return VPXMAX(FRAME_OVERHEAD_BITS,
+                (int)((uint64_t)bpm * mbs) >> BPER_MB_NORMBITS);
 }
 
 int vp10_rc_clamp_pframe_target_size(const VP10_COMP *const cpi, int target) {
   const RATE_CONTROL *rc = &cpi->rc;
   const VP10EncoderConfig *oxcf = &cpi->oxcf;
-  const int min_frame_target = MAX(rc->min_frame_bandwidth,
-                                   rc->avg_frame_bandwidth >> 5);
+  const int min_frame_target = VPXMAX(rc->min_frame_bandwidth,
+                                      rc->avg_frame_bandwidth >> 5);
   if (target < min_frame_target)
     target = min_frame_target;
   if (cpi->refresh_golden_frame && rc->is_src_frame_alt_ref) {
@@ -216,7 +215,7 @@
   if (oxcf->rc_max_inter_bitrate_pct) {
     const int max_rate = rc->avg_frame_bandwidth *
                          oxcf->rc_max_inter_bitrate_pct / 100;
-    target = MIN(target, max_rate);
+    target = VPXMIN(target, max_rate);
   }
   return target;
 }
@@ -227,7 +226,7 @@
   if (oxcf->rc_max_intra_bitrate_pct) {
     const int max_rate = rc->avg_frame_bandwidth *
                              oxcf->rc_max_intra_bitrate_pct / 100;
-    target = MIN(target, max_rate);
+    target = VPXMIN(target, max_rate);
   }
   if (target > rc->max_frame_bandwidth)
     target = rc->max_frame_bandwidth;
@@ -250,7 +249,8 @@
     lrc->bits_off_target += bits_off_for_this_layer;
 
     // Clip buffer level to maximum buffer size for the layer.
-    lrc->bits_off_target = MIN(lrc->bits_off_target, lrc->maximum_buffer_size);
+    lrc->bits_off_target =
+        VPXMIN(lrc->bits_off_target, lrc->maximum_buffer_size);
     lrc->buffer_level = lrc->bits_off_target;
   }
 }
@@ -268,7 +268,7 @@
   }
 
   // Clip the buffer level to the maximum specified buffer size.
-  rc->bits_off_target = MIN(rc->bits_off_target, rc->maximum_buffer_size);
+  rc->bits_off_target = VPXMIN(rc->bits_off_target, rc->maximum_buffer_size);
   rc->buffer_level = rc->bits_off_target;
 
   if (is_one_pass_cbr_svc(cpi)) {
@@ -287,8 +287,8 @@
   if (factor <= factor_safe)
     return default_interval;
   else
-    return MAX(default_interval,
-               (int)(MIN_GF_INTERVAL * factor / factor_safe + 0.5));
+    return VPXMAX(default_interval,
+                  (int)(MIN_GF_INTERVAL * factor / factor_safe + 0.5));
   // Note this logic makes:
   // 4K24: 5
   // 4K30: 6
@@ -296,9 +296,9 @@
 }
 
 int vp10_rc_get_default_max_gf_interval(double framerate, int min_gf_interval) {
-  int interval = MIN(MAX_GF_INTERVAL, (int)(framerate * 0.75));
+  int interval = VPXMIN(MAX_GF_INTERVAL, (int)(framerate * 0.75));
   interval += (interval & 0x01);  // Round to even value
-  return MAX(interval, min_gf_interval);
+  return VPXMAX(interval, min_gf_interval);
 }
 
 void vp10_rc_init(const VP10EncoderConfig *oxcf, int pass, RATE_CONTROL *rc) {
@@ -478,7 +478,7 @@
   // More heavily damped adjustment used if we have been oscillating either side
   // of target.
   adjustment_limit = 0.25 +
-      0.5 * MIN(1, fabs(log10(0.01 * correction_factor)));
+      0.5 * VPXMIN(1, fabs(log10(0.01 * correction_factor)));
 
   cpi->rc.q_2_frame = cpi->rc.q_1_frame;
   cpi->rc.q_1_frame = cm->base_qindex;
@@ -558,8 +558,8 @@
   if (cpi->oxcf.rc_mode == VPX_CBR &&
       (cpi->rc.rc_1_frame * cpi->rc.rc_2_frame == -1) &&
       cpi->rc.q_1_frame != cpi->rc.q_2_frame) {
-    q = clamp(q, MIN(cpi->rc.q_1_frame, cpi->rc.q_2_frame),
-              MAX(cpi->rc.q_1_frame, cpi->rc.q_2_frame));
+    q = clamp(q, VPXMIN(cpi->rc.q_1_frame, cpi->rc.q_2_frame),
+              VPXMAX(cpi->rc.q_1_frame, cpi->rc.q_2_frame));
   }
   return q;
 }
@@ -617,7 +617,7 @@
                                              : rc->last_q[INTER_FRAME] * 2;
     }
   }
-  return MIN(active_worst_quality, rc->worst_quality);
+  return VPXMIN(active_worst_quality, rc->worst_quality);
 }
 
 // Adjust active_worst_quality level based on buffer level.
@@ -643,10 +643,10 @@
   // So for first few frames following key, the qp of that key frame is weighted
   // into the active_worst_quality setting.
   ambient_qp = (cm->current_video_frame < 5) ?
-      MIN(rc->avg_frame_qindex[INTER_FRAME], rc->avg_frame_qindex[KEY_FRAME]) :
-      rc->avg_frame_qindex[INTER_FRAME];
-  active_worst_quality = MIN(rc->worst_quality,
-                             ambient_qp * 5 / 4);
+                   VPXMIN(rc->avg_frame_qindex[INTER_FRAME],
+                          rc->avg_frame_qindex[KEY_FRAME]) :
+                   rc->avg_frame_qindex[INTER_FRAME];
+  active_worst_quality = VPXMIN(rc->worst_quality, ambient_qp * 5 / 4);
   if (rc->buffer_level > rc->optimal_buffer_level) {
     // Adjust down.
     // Maximum limit for down adjustment, ~30%.
@@ -699,7 +699,7 @@
       int delta_qindex = vp10_compute_qdelta(rc, last_boosted_q,
                                             (last_boosted_q * 0.75),
                                             cm->bit_depth);
-      active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
+      active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
     } else if (cm->current_video_frame > 0) {
       // not first frame of one pass and kf_boost is set
       double q_adj_factor = 1.0;
@@ -833,7 +833,7 @@
       int delta_qindex = vp10_compute_qdelta(rc, last_boosted_q,
                                             last_boosted_q * 0.75,
                                             cm->bit_depth);
-      active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
+      active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
     } else {
       // not first frame of one pass and kf_boost is set
       double q_adj_factor = 1.0;
@@ -1002,21 +1002,21 @@
       int qindex;
 
       if (cpi->twopass.last_kfgroup_zeromotion_pct >= STATIC_MOTION_THRESH) {
-        qindex = MIN(rc->last_kf_qindex, rc->last_boosted_qindex);
+        qindex = VPXMIN(rc->last_kf_qindex, rc->last_boosted_qindex);
         active_best_quality = qindex;
         last_boosted_q = vp10_convert_qindex_to_q(qindex, cm->bit_depth);
         delta_qindex = vp10_compute_qdelta(rc, last_boosted_q,
                                               last_boosted_q * 1.25,
                                               cm->bit_depth);
-        active_worst_quality = MIN(qindex + delta_qindex, active_worst_quality);
-
+        active_worst_quality =
+            VPXMIN(qindex + delta_qindex, active_worst_quality);
       } else {
         qindex = rc->last_boosted_qindex;
         last_boosted_q = vp10_convert_qindex_to_q(qindex, cm->bit_depth);
         delta_qindex = vp10_compute_qdelta(rc, last_boosted_q,
                                               last_boosted_q * 0.75,
                                               cm->bit_depth);
-        active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
+        active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
       }
     } else {
       // Not forced keyframe.
@@ -1116,8 +1116,8 @@
       (cpi->twopass.last_kfgroup_zeromotion_pct < STATIC_MOTION_THRESH)) {
     int qdelta = vp10_frame_type_qdelta(cpi, gf_group->rf_level[gf_group->index],
                                        active_worst_quality);
-    active_worst_quality = MAX(active_worst_quality + qdelta,
-                               active_best_quality);
+    active_worst_quality = VPXMAX(active_worst_quality + qdelta,
+                                  active_best_quality);
   }
 #endif
 
@@ -1126,7 +1126,8 @@
     int qdelta = vp10_compute_qdelta_by_rate(rc, cm->frame_type,
                                             active_best_quality, 2.0,
                                             cm->bit_depth);
-    active_best_quality = MAX(active_best_quality + qdelta, rc->best_quality);
+    active_best_quality =
+        VPXMAX(active_best_quality + qdelta, rc->best_quality);
   }
 
   active_best_quality = clamp(active_best_quality,
@@ -1141,7 +1142,7 @@
              rc->this_key_frame_forced) {
     // If static since last kf use better of last boosted and last kf q.
     if (cpi->twopass.last_kfgroup_zeromotion_pct >= STATIC_MOTION_THRESH) {
-      q = MIN(rc->last_kf_qindex, rc->last_boosted_qindex);
+      q = VPXMIN(rc->last_kf_qindex, rc->last_boosted_qindex);
     } else {
       q = rc->last_boosted_qindex;
     }
@@ -1180,15 +1181,7 @@
   } else {
     q = rc_pick_q_and_bounds_two_pass(cpi, bottom_index, top_index);
   }
-  if (cpi->sf.use_nonrd_pick_mode) {
-    if (cpi->sf.force_frame_boost == 1)
-      q -= cpi->sf.max_delta_qindex;
 
-    if (q < *bottom_index)
-      *bottom_index = q;
-    else if (q > *top_index)
-      *top_index = q;
-  }
   return q;
 }
 
@@ -1203,9 +1196,9 @@
     // For very small rate targets where the fractional adjustment
     // may be tiny make sure there is at least a minimum range.
     const int tolerance = (cpi->sf.recode_tolerance * frame_target) / 100;
-    *frame_under_shoot_limit = MAX(frame_target - tolerance - 200, 0);
-    *frame_over_shoot_limit = MIN(frame_target + tolerance + 200,
-                                  cpi->rc.max_frame_bandwidth);
+    *frame_under_shoot_limit = VPXMAX(frame_target - tolerance - 200, 0);
+    *frame_over_shoot_limit = VPXMIN(frame_target + tolerance + 200,
+                                     cpi->rc.max_frame_bandwidth);
   }
 }
 
@@ -1444,7 +1437,8 @@
   const SVC *const svc = &cpi->svc;
   const int64_t diff = rc->optimal_buffer_level - rc->buffer_level;
   const int64_t one_pct_bits = 1 + rc->optimal_buffer_level / 100;
-  int min_frame_target = MAX(rc->avg_frame_bandwidth >> 4, FRAME_OVERHEAD_BITS);
+  int min_frame_target =
+      VPXMAX(rc->avg_frame_bandwidth >> 4, FRAME_OVERHEAD_BITS);
   int target;
 
   if (oxcf->gf_cbr_boost_pct) {
@@ -1466,23 +1460,24 @@
             svc->temporal_layer_id, svc->number_temporal_layers);
     const LAYER_CONTEXT *lc = &svc->layer_context[layer];
     target = lc->avg_frame_size;
-    min_frame_target = MAX(lc->avg_frame_size >> 4, FRAME_OVERHEAD_BITS);
+    min_frame_target = VPXMAX(lc->avg_frame_size >> 4, FRAME_OVERHEAD_BITS);
   }
   if (diff > 0) {
     // Lower the target bandwidth for this frame.
-    const int pct_low = (int)MIN(diff / one_pct_bits, oxcf->under_shoot_pct);
+    const int pct_low = (int)VPXMIN(diff / one_pct_bits, oxcf->under_shoot_pct);
     target -= (target * pct_low) / 200;
   } else if (diff < 0) {
     // Increase the target bandwidth for this frame.
-    const int pct_high = (int)MIN(-diff / one_pct_bits, oxcf->over_shoot_pct);
+    const int pct_high =
+        (int)VPXMIN(-diff / one_pct_bits, oxcf->over_shoot_pct);
     target += (target * pct_high) / 200;
   }
   if (oxcf->rc_max_inter_bitrate_pct) {
     const int max_rate = rc->avg_frame_bandwidth *
                          oxcf->rc_max_inter_bitrate_pct / 100;
-    target = MIN(target, max_rate);
+    target = VPXMIN(target, max_rate);
   }
-  return MAX(min_frame_target, target);
+  return VPXMAX(min_frame_target, target);
 }
 
 static int calc_iframe_target_size_one_pass_cbr(const VP10_COMP *cpi) {
@@ -1504,7 +1499,7 @@
       const LAYER_CONTEXT *lc = &svc->layer_context[layer];
       framerate = lc->framerate;
     }
-    kf_boost = MAX(kf_boost, (int)(2 * framerate - 16));
+    kf_boost = VPXMAX(kf_boost, (int)(2 * framerate - 16));
     if (rc->frames_since_key <  framerate / 2) {
       kf_boost = (int)(kf_boost * rc->frames_since_key /
                        (framerate / 2));
@@ -1712,7 +1707,7 @@
     rc->max_gf_interval = rc->static_scene_max_gf_interval;
 
   // Clamp min to max
-  rc->min_gf_interval = MIN(rc->min_gf_interval, rc->max_gf_interval);
+  rc->min_gf_interval = VPXMIN(rc->min_gf_interval, rc->max_gf_interval);
 }
 
 void vp10_rc_update_framerate(VP10_COMP *cpi) {
@@ -1725,7 +1720,8 @@
   rc->min_frame_bandwidth = (int)(rc->avg_frame_bandwidth *
                                 oxcf->two_pass_vbrmin_section / 100);
 
-  rc->min_frame_bandwidth = MAX(rc->min_frame_bandwidth, FRAME_OVERHEAD_BITS);
+  rc->min_frame_bandwidth =
+      VPXMAX(rc->min_frame_bandwidth, FRAME_OVERHEAD_BITS);
 
   // A maximum bitrate for a frame is defined.
   // The baseline for this aligns with HW implementations that
@@ -1736,8 +1732,8 @@
   // specifies lossless encode.
   vbr_max_bits = (int)(((int64_t)rc->avg_frame_bandwidth *
                      oxcf->two_pass_vbrmax_section) / 100);
-  rc->max_frame_bandwidth = MAX(MAX((cm->MBs * MAX_MB_RATE), MAXRATE_1080P),
-                                    vbr_max_bits);
+  rc->max_frame_bandwidth =
+      VPXMAX(VPXMAX((cm->MBs * MAX_MB_RATE), MAXRATE_1080P), vbr_max_bits);
 
   vp10_rc_set_gf_interval_range(cpi, rc);
 }
@@ -1775,12 +1771,12 @@
   // Dont do it for kf,arf,gf or overlay frames.
   if (!frame_is_kf_gf_arf(cpi) && !rc->is_src_frame_alt_ref &&
       rc->vbr_bits_off_target_fast) {
-    int one_frame_bits = MAX(rc->avg_frame_bandwidth, *this_frame_target);
+    int one_frame_bits = VPXMAX(rc->avg_frame_bandwidth, *this_frame_target);
     int fast_extra_bits;
-    fast_extra_bits =
-      (int)MIN(rc->vbr_bits_off_target_fast, one_frame_bits);
-    fast_extra_bits = (int)MIN(fast_extra_bits,
-      MAX(one_frame_bits / 8, rc->vbr_bits_off_target_fast / 8));
+    fast_extra_bits = (int)VPXMIN(rc->vbr_bits_off_target_fast, one_frame_bits);
+    fast_extra_bits = (int)VPXMIN(
+        fast_extra_bits,
+        VPXMAX(one_frame_bits / 8, rc->vbr_bits_off_target_fast / 8));
     *this_frame_target += (int)fast_extra_bits;
     rc->vbr_bits_off_target_fast -= fast_extra_bits;
   }
diff --git a/vp10/encoder/rd.c b/vp10/encoder/rd.c
index dad1d2a..00e7a94 100644
--- a/vp10/encoder/rd.c
+++ b/vp10/encoder/rd.c
@@ -172,7 +172,7 @@
   if (cpi->oxcf.pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
     const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
     const FRAME_UPDATE_TYPE frame_type = gf_group->update_type[gf_group->index];
-    const int boost_index = MIN(15, (cpi->rc.gfu_boost / 100));
+    const int boost_index = VPXMIN(15, (cpi->rc.gfu_boost / 100));
 
     rdmult = (rdmult * rd_frame_type_factor[frame_type]) >> 7;
     rdmult += ((rdmult * rd_boost_factor[boost_index]) >> 7);
@@ -204,7 +204,7 @@
   q = vp10_dc_quant(qindex, 0, VPX_BITS_8) / 4.0;
 #endif  // CONFIG_VP9_HIGHBITDEPTH
   // TODO(debargha): Adjust the function below.
-  return MAX((int)(pow(q, RD_THRESH_POW) * 5.12), 8);
+  return VPXMAX((int)(pow(q, RD_THRESH_POW) * 5.12), 8);
 }
 
 void vp10_initialize_me_consts(VP10_COMP *cpi, MACROBLOCK *x, int qindex) {
@@ -285,8 +285,7 @@
   set_block_thresholds(cm, rd);
   set_partition_probs(cm, xd);
 
-  if (!cpi->sf.use_nonrd_pick_mode || cm->frame_type == KEY_FRAME)
-    fill_token_costs(x->token_costs, cm->fc->coef_probs);
+  fill_token_costs(x->token_costs, cm->fc->coef_probs);
 
   if (cpi->sf.partition_search_type != VAR_BASED_PARTITION ||
       cm->frame_type == KEY_FRAME) {
@@ -295,20 +294,17 @@
                       vp10_partition_tree);
   }
 
-  if (!cpi->sf.use_nonrd_pick_mode || (cm->current_video_frame & 0x07) == 1 ||
-      cm->frame_type == KEY_FRAME) {
-    fill_mode_costs(cpi);
+  fill_mode_costs(cpi);
 
-    if (!frame_is_intra_only(cm)) {
-      vp10_build_nmv_cost_table(x->nmvjointcost,
-                               cm->allow_high_precision_mv ? x->nmvcost_hp
-                                                           : x->nmvcost,
-                               &cm->fc->nmvc, cm->allow_high_precision_mv);
+  if (!frame_is_intra_only(cm)) {
+    vp10_build_nmv_cost_table(x->nmvjointcost,
+                             cm->allow_high_precision_mv ? x->nmvcost_hp
+                                                         : x->nmvcost,
+                             &cm->fc->nmvc, cm->allow_high_precision_mv);
 
-      for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
-        vp10_cost_tokens((int *)cpi->inter_mode_cost[i],
-                        cm->fc->inter_mode_probs[i], vp10_inter_mode_tree);
-    }
+    for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
+      vp10_cost_tokens((int *)cpi->inter_mode_cost[i],
+                      cm->fc->inter_mode_probs[i], vp10_inter_mode_tree);
   }
 }
 
@@ -404,7 +400,7 @@
     static const uint32_t MAX_XSQ_Q10 = 245727;
     const uint64_t xsq_q10_64 =
         (((uint64_t)qstep * qstep << (n_log2 + 10)) + (var >> 1)) / var;
-    const int xsq_q10 = (int)MIN(xsq_q10_64, MAX_XSQ_Q10);
+    const int xsq_q10 = (int)VPXMIN(xsq_q10_64, MAX_XSQ_Q10);
     model_rd_norm(xsq_q10, &r_q10, &d_q10);
     *rate = ((r_q10 << n_log2) + 2) >> 2;
     *dist = (var * (int64_t)d_q10 + 512) >> 10;
@@ -485,7 +481,7 @@
       continue;
     fp_row = (this_mv->row + 3 + (this_mv->row >= 0)) >> 3;
     fp_col = (this_mv->col + 3 + (this_mv->col >= 0)) >> 3;
-    max_mv = MAX(max_mv, MAX(abs(this_mv->row), abs(this_mv->col)) >> 3);
+    max_mv = VPXMAX(max_mv, VPXMAX(abs(this_mv->row), abs(this_mv->col)) >> 3);
 
     if (fp_row ==0 && fp_col == 0 && zero_seen)
       continue;
@@ -630,16 +626,15 @@
     const int top_mode = bsize < BLOCK_8X8 ? MAX_REFS : MAX_MODES;
     int mode;
     for (mode = 0; mode < top_mode; ++mode) {
-      const BLOCK_SIZE min_size = MAX(bsize - 1, BLOCK_4X4);
-      const BLOCK_SIZE max_size = MIN(bsize + 2, BLOCK_64X64);
+      const BLOCK_SIZE min_size = VPXMAX(bsize - 1, BLOCK_4X4);
+      const BLOCK_SIZE max_size = VPXMIN(bsize + 2, BLOCK_64X64);
       BLOCK_SIZE bs;
       for (bs = min_size; bs <= max_size; ++bs) {
         int *const fact = &factor_buf[bs][mode];
         if (mode == best_mode_index) {
           *fact -= (*fact >> 4);
         } else {
-          *fact = MIN(*fact + RD_THRESH_INC,
-                      rd_thresh * RD_THRESH_MAX_FACT);
+          *fact = VPXMIN(*fact + RD_THRESH_INC, rd_thresh * RD_THRESH_MAX_FACT);
         }
       }
     }
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index 14efe0e..90a7aa3 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -192,8 +192,8 @@
     const int64_t ac_thr = p->quant_thred[1] >> shift;
     // The low thresholds are used to measure if the prediction errors are
     // low enough so that we can skip the mode search.
-    const int64_t low_dc_thr = MIN(50, dc_thr >> 2);
-    const int64_t low_ac_thr = MIN(80, ac_thr >> 2);
+    const int64_t low_dc_thr = VPXMIN(50, dc_thr >> 2);
+    const int64_t low_ac_thr = VPXMIN(80, ac_thr >> 2);
     int bw = 1 << (b_width_log2_lookup[bs] - b_width_log2_lookup[unit_size]);
     int bh = 1 << (b_height_log2_lookup[bs] - b_width_log2_lookup[unit_size]);
     int idx, idy;
@@ -505,7 +505,7 @@
         if (tx_size != TX_32X32)
           dc_correct >>= 2;
 
-        dist = MAX(0, sse - dc_correct);
+        dist = VPXMAX(0, sse - dc_correct);
       }
     } else {
       // SKIP_TXFM_AC_DC
@@ -531,7 +531,7 @@
   rd2 = RDCOST(x->rdmult, x->rddiv, 0, sse);
 
   // TODO(jingning): temporarily enabled only for luma component
-  rd = MIN(rd1, rd2);
+  rd = VPXMIN(rd1, rd2);
   if (plane == 0)
     x->zcoeff_blk[tx_size][block] = !x->plane[plane].eobs[block] ||
                                     (rd1 > rd2 && !xd->lossless);
@@ -599,7 +599,7 @@
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
 
-  mbmi->tx_size = MIN(max_tx_size, largest_tx_size);
+  mbmi->tx_size = VPXMIN(max_tx_size, largest_tx_size);
 
   txfm_rd_in_plane(x, rate, distortion, skip,
                    sse, ref_best_rd, 0, bs,
@@ -639,8 +639,8 @@
     start_tx = max_tx_size;
     end_tx = 0;
   } else {
-    TX_SIZE chosen_tx_size = MIN(max_tx_size,
-                                 tx_mode_to_biggest_tx_size[cm->tx_mode]);
+    TX_SIZE chosen_tx_size = VPXMIN(max_tx_size,
+                                    tx_mode_to_biggest_tx_size[cm->tx_mode]);
     start_tx = chosen_tx_size;
     end_tx = chosen_tx_size;
   }
@@ -1060,16 +1060,6 @@
   memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
   /* Y Search for intra prediction mode */
   for (mode = DC_PRED; mode <= TM_PRED; mode++) {
-
-    if (cpi->sf.use_nonrd_pick_mode) {
-      // These speed features are turned on in hybrid non-RD and RD mode
-      // for key frame coding in the context of real-time setting.
-      if (conditional_skipintra(mode, mode_selected))
-          continue;
-      if (*skippable)
-        break;
-    }
-
     mic->mbmi.mode = mode;
 
     super_block_yrd(cpi, x, &this_rate_tokenonly, &this_distortion,
@@ -1399,7 +1389,7 @@
                               cpi->sf.use_fast_coef_costing);
       rd1 = RDCOST(x->rdmult, x->rddiv, thisrate, thisdistortion >> 2);
       rd2 = RDCOST(x->rdmult, x->rddiv, 0, thissse >> 2);
-      rd = MIN(rd1, rd2);
+      rd = VPXMIN(rd1, rd2);
       if (rd >= best_yrd)
         return INT64_MAX;
     }
@@ -1818,7 +1808,8 @@
           if (i == 0)
             max_mv = x->max_mv_context[mbmi->ref_frame[0]];
           else
-            max_mv = MAX(abs(bsi->mvp.as_mv.row), abs(bsi->mvp.as_mv.col)) >> 3;
+            max_mv =
+                VPXMAX(abs(bsi->mvp.as_mv.row), abs(bsi->mvp.as_mv.col)) >> 3;
 
           if (cpi->sf.mv.auto_mv_step_size && cm->show_frame) {
             // Take wtd average of the step_params based on the last frame's
@@ -1836,7 +1827,7 @@
           if (cpi->sf.adaptive_motion_search) {
             mvp_full.row = x->pred_mv[mbmi->ref_frame[0]].row >> 3;
             mvp_full.col = x->pred_mv[mbmi->ref_frame[0]].col >> 3;
-            step_param = MAX(step_param, 8);
+            step_param = VPXMAX(step_param, 8);
           }
 
           // adjust src pointer for this block
@@ -2241,7 +2232,7 @@
   vp10_set_mv_search_range(x, &ref_mv);
 
   // Work out the size of the first step in the mv step search.
-  // 0 here is maximum length first step. 1 is MAX >> 1 etc.
+  // 0 here is maximum length first step. 1 is VPXMAX >> 1 etc.
   if (cpi->sf.mv.auto_mv_step_size && cm->show_frame) {
     // Take wtd average of the step_params based on the last frame's
     // max mv magnitude and that based on the best ref mvs of the current
@@ -2253,9 +2244,10 @@
   }
 
   if (cpi->sf.adaptive_motion_search && bsize < BLOCK_64X64) {
-    int boffset = 2 * (b_width_log2_lookup[BLOCK_64X64] -
-          MIN(b_height_log2_lookup[bsize], b_width_log2_lookup[bsize]));
-    step_param = MAX(step_param, boffset);
+    int boffset =
+        2 * (b_width_log2_lookup[BLOCK_64X64] -
+             VPXMIN(b_height_log2_lookup[bsize], b_width_log2_lookup[bsize]));
+    step_param = VPXMAX(step_param, boffset);
   }
 
   if (cpi->sf.adaptive_motion_search) {
@@ -2476,7 +2468,7 @@
       // motion field, where the distortion gain for a single block may not
       // be enough to overcome the cost of a new mv.
       if (discount_newmv_test(cpi, this_mode, tmp_mv, mode_mv, refs[0])) {
-        *rate2 += MAX((rate_mv / NEW_MV_DISCOUNT_FACTOR), 1);
+        *rate2 += VPXMAX((rate_mv / NEW_MV_DISCOUNT_FACTOR), 1);
       } else {
         *rate2 += rate_mv;
       }
@@ -2513,10 +2505,10 @@
   // initiation of a motion field.
   if (discount_newmv_test(cpi, this_mode, frame_mv[refs[0]],
                           mode_mv, refs[0])) {
-    *rate2 += MIN(cost_mv_ref(cpi, this_mode,
-                              mbmi_ext->mode_context[refs[0]]),
-                  cost_mv_ref(cpi, NEARESTMV,
-                              mbmi_ext->mode_context[refs[0]]));
+    *rate2 += VPXMIN(cost_mv_ref(cpi, this_mode,
+                                 mbmi_ext->mode_context[refs[0]]),
+                     cost_mv_ref(cpi, NEARESTMV,
+                                 mbmi_ext->mode_context[refs[0]]));
   } else {
     *rate2 += cost_mv_ref(cpi, this_mode, mbmi_ext->mode_context[refs[0]]);
   }
@@ -2558,10 +2550,10 @@
           rd = RDCOST(x->rdmult, x->rddiv, tmp_rate_sum, tmp_dist_sum);
           filter_cache[i] = rd;
           filter_cache[SWITCHABLE_FILTERS] =
-              MIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
+              VPXMIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
           if (cm->interp_filter == SWITCHABLE)
             rd += rs_rd;
-          *mask_filter = MAX(*mask_filter, rd);
+          *mask_filter = VPXMAX(*mask_filter, rd);
         } else {
           int rate_sum = 0;
           int64_t dist_sum = 0;
@@ -2591,10 +2583,10 @@
           rd = RDCOST(x->rdmult, x->rddiv, rate_sum, dist_sum);
           filter_cache[i] = rd;
           filter_cache[SWITCHABLE_FILTERS] =
-              MIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
+              VPXMIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
           if (cm->interp_filter == SWITCHABLE)
             rd += rs_rd;
-          *mask_filter = MAX(*mask_filter, rd);
+          *mask_filter = VPXMAX(*mask_filter, rd);
 
           if (i == 0 && intpel_mv) {
             tmp_rate_sum = rate_sum;
@@ -2705,7 +2697,7 @@
     *distortion += distortion_y;
 
     rdcosty = RDCOST(x->rdmult, x->rddiv, *rate2, *distortion);
-    rdcosty = MIN(rdcosty, RDCOST(x->rdmult, x->rddiv, 0, *psse));
+    rdcosty = VPXMIN(rdcosty, RDCOST(x->rdmult, x->rddiv, 0, *psse));
 
     if (!super_block_uvrd(cpi, x, rate_uv, &distortion_uv, &skippable_uv,
                           &sseuv, bsize, ref_best_rd - rdcosty)) {
@@ -2770,7 +2762,7 @@
                                        pd[1].subsampling_x,
                                        pd[1].subsampling_y);
   rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv, &rate_uv_tokenonly,
-                          &dist_uv, &uv_skip, MAX(BLOCK_8X8, bsize),
+                          &dist_uv, &uv_skip, VPXMAX(BLOCK_8X8, bsize),
                           max_uv_tx_size);
 
   if (y_skip && uv_skip) {
@@ -2837,12 +2829,12 @@
   // to a predictor with a low spatial complexity compared to the source.
   if ((source_variance > LOW_VAR_THRESH) && (ref_frame == INTRA_FRAME) &&
       (source_variance > recon_variance)) {
-    var_factor = MIN(absvar_diff, MIN(VLOW_ADJ_MAX, var_error));
+    var_factor = VPXMIN(absvar_diff, VPXMIN(VLOW_ADJ_MAX, var_error));
   // A second possible case of interest is where the source variance
   // is very low and we wish to discourage false texture or motion trails.
   } else if ((source_variance < (LOW_VAR_THRESH >> 1)) &&
              (recon_variance > source_variance)) {
-    var_factor = MIN(absvar_diff, MIN(VHIGH_ADJ_MAX, var_error));
+    var_factor = VPXMIN(absvar_diff, VPXMIN(VHIGH_ADJ_MAX, var_error));
   }
   *this_rd += (*this_rd * var_factor) / 100;
 }
@@ -2872,7 +2864,7 @@
     top_edge += (int)(twopass->this_frame_stats.inactive_zone_rows * 2);
 
     bottom_edge -= (int)(twopass->this_frame_stats.inactive_zone_rows * 2);
-    bottom_edge = MAX(top_edge, bottom_edge);
+    bottom_edge = VPXMAX(top_edge, bottom_edge);
   }
 
   if (((top_edge >= mi_row) && (top_edge < (mi_row + mi_step))) ||
@@ -2899,7 +2891,7 @@
     left_edge += (int)(twopass->this_frame_stats.inactive_zone_cols * 2);
 
     right_edge -= (int)(twopass->this_frame_stats.inactive_zone_cols * 2);
-    right_edge = MAX(left_edge, right_edge);
+    right_edge = VPXMAX(left_edge, right_edge);
   }
 
   if (((left_edge >= mi_col) && (left_edge < (mi_col + mi_step))) ||
@@ -3146,7 +3138,7 @@
     }
 
     if ((ref_frame_skip_mask[0] & (1 << ref_frame)) &&
-        (ref_frame_skip_mask[1] & (1 << MAX(0, second_ref_frame))))
+        (ref_frame_skip_mask[1] & (1 << VPXMAX(0, second_ref_frame))))
       continue;
 
     if (mode_skip_mask[ref_frame] & (1 << this_mode))
@@ -3160,10 +3152,10 @@
       continue;
 
     if (sf->motion_field_mode_search) {
-      const int mi_width  = MIN(num_8x8_blocks_wide_lookup[bsize],
-                                tile_info->mi_col_end - mi_col);
-      const int mi_height = MIN(num_8x8_blocks_high_lookup[bsize],
-                                tile_info->mi_row_end - mi_row);
+      const int mi_width  = VPXMIN(num_8x8_blocks_wide_lookup[bsize],
+                                   tile_info->mi_col_end - mi_col);
+      const int mi_height = VPXMIN(num_8x8_blocks_high_lookup[bsize],
+                                   tile_info->mi_row_end - mi_row);
       const int bsl = mi_width_log2_lookup[bsize];
       int cb_partition_search_ctrl = (((mi_row + mi_col) >> bsl)
           + get_chessboard_index(cm->current_video_frame)) & 0x1;
@@ -3381,9 +3373,9 @@
 
     if (!disable_skip && ref_frame == INTRA_FRAME) {
       for (i = 0; i < REFERENCE_MODES; ++i)
-        best_pred_rd[i] = MIN(best_pred_rd[i], this_rd);
+        best_pred_rd[i] = VPXMIN(best_pred_rd[i], this_rd);
       for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++)
-        best_filter_rd[i] = MIN(best_filter_rd[i], this_rd);
+        best_filter_rd[i] = VPXMIN(best_filter_rd[i], this_rd);
     }
 
     // Did this mode help.. i.e. is it the new best mode
@@ -3482,7 +3474,7 @@
             adj_rd = filter_cache[i] - ref;
 
           adj_rd += this_rd;
-          best_filter_rd[i] = MIN(best_filter_rd[i], adj_rd);
+          best_filter_rd[i] = VPXMIN(best_filter_rd[i], adj_rd);
         }
       }
     }
@@ -3825,7 +3817,7 @@
     }
 
     if ((ref_frame_skip_mask[0] & (1 << ref_frame)) &&
-        (ref_frame_skip_mask[1] & (1 << MAX(0, second_ref_frame))))
+        (ref_frame_skip_mask[1] & (1 << VPXMAX(0, second_ref_frame))))
       continue;
 
     // Test best rd so far against threshold for trying this mode.
@@ -3980,12 +3972,11 @@
             rs_rd = RDCOST(x->rdmult, x->rddiv, rs, 0);
             filter_cache[switchable_filter_index] = tmp_rd;
             filter_cache[SWITCHABLE_FILTERS] =
-                MIN(filter_cache[SWITCHABLE_FILTERS],
-                    tmp_rd + rs_rd);
+                VPXMIN(filter_cache[SWITCHABLE_FILTERS], tmp_rd + rs_rd);
             if (cm->interp_filter == SWITCHABLE)
               tmp_rd += rs_rd;
 
-            mask_filter = MAX(mask_filter, tmp_rd);
+            mask_filter = VPXMAX(mask_filter, tmp_rd);
 
             newbest = (tmp_rd < tmp_best_rd);
             if (newbest) {
@@ -4063,8 +4054,8 @@
       compmode_cost = vp10_cost_bit(comp_mode_p, comp_pred);
 
       tmp_best_rdu = best_rd -
-          MIN(RDCOST(x->rdmult, x->rddiv, rate2, distortion2),
-              RDCOST(x->rdmult, x->rddiv, 0, total_sse));
+          VPXMIN(RDCOST(x->rdmult, x->rddiv, rate2, distortion2),
+                 RDCOST(x->rdmult, x->rddiv, 0, total_sse));
 
       if (tmp_best_rdu > 0) {
         // If even the 'Y' rd value of split is higher than best so far
@@ -4124,9 +4115,9 @@
 
     if (!disable_skip && ref_frame == INTRA_FRAME) {
       for (i = 0; i < REFERENCE_MODES; ++i)
-        best_pred_rd[i] = MIN(best_pred_rd[i], this_rd);
+        best_pred_rd[i] = VPXMIN(best_pred_rd[i], this_rd);
       for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++)
-        best_filter_rd[i] = MIN(best_filter_rd[i], this_rd);
+        best_filter_rd[i] = VPXMIN(best_filter_rd[i], this_rd);
     }
 
     // Did this mode help.. i.e. is it the new best mode
@@ -4225,7 +4216,7 @@
           adj_rd = filter_cache[i] - ref;
 
         adj_rd += this_rd;
-        best_filter_rd[i] = MIN(best_filter_rd[i], adj_rd);
+        best_filter_rd[i] = VPXMIN(best_filter_rd[i], adj_rd);
       }
     }
 
diff --git a/vp10/encoder/speed_features.c b/vp10/encoder/speed_features.c
index 0f7f603..3fb56dd 100644
--- a/vp10/encoder/speed_features.c
+++ b/vp10/encoder/speed_features.c
@@ -49,7 +49,7 @@
   VP10_COMMON *const cm = &cpi->common;
 
   if (speed >= 1) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
       sf->partition_search_breakout_dist_thr = (1 << 23);
@@ -60,7 +60,7 @@
   }
 
   if (speed >= 2) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
       sf->adaptive_pred_interp_filter = 0;
@@ -75,7 +75,7 @@
   }
 
   if (speed >= 3) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = DISABLE_ALL_SPLIT;
       sf->schedule_mode_search = cm->base_qindex < 220 ? 1 : 0;
       sf->partition_search_breakout_dist_thr = (1 << 25);
@@ -99,7 +99,7 @@
   }
 
   if (speed >= 4) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->partition_search_breakout_dist_thr = (1 << 26);
     } else {
       sf->partition_search_breakout_dist_thr = (1 << 24);
@@ -215,7 +215,7 @@
   VP10_COMMON *const cm = &cpi->common;
 
   if (speed >= 1) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
     } else {
@@ -224,7 +224,7 @@
   }
 
   if (speed >= 2) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
     } else {
@@ -233,7 +233,7 @@
   }
 
   if (speed >= 5) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->partition_search_breakout_dist_thr = (1 << 25);
     } else {
       sf->partition_search_breakout_dist_thr = (1 << 23);
@@ -241,7 +241,7 @@
   }
 
   if (speed >= 7) {
-    sf->encode_breakout_thresh = (MIN(cm->width, cm->height) >= 720) ?
+    sf->encode_breakout_thresh = (VPXMIN(cm->width, cm->height) >= 720) ?
         800 : 300;
   }
 }
@@ -348,7 +348,6 @@
         (frames_since_key % (sf->last_partitioning_redo_frequency << 1) == 1);
     sf->max_delta_qindex = is_keyframe ? 20 : 15;
     sf->partition_search_type = REFERENCE_PARTITION;
-    sf->use_nonrd_pick_mode = 1;
     sf->allow_skip_recode = 0;
     sf->inter_mode_mask[BLOCK_32X32] = INTER_NEAREST_NEW_ZERO;
     sf->inter_mode_mask[BLOCK_32X64] = INTER_NEAREST_NEW_ZERO;
@@ -384,7 +383,6 @@
     // Adaptively switch between SOURCE_VAR_BASED_PARTITION and FIXED_PARTITION.
     sf->partition_search_type = VAR_BASED_PARTITION;
     // Turn on this to use non-RD key frame coding mode.
-    sf->use_nonrd_pick_mode = 1;
     sf->mv.search_method = NSTEP;
     sf->mv.reduce_first_step_size = 1;
     sf->skip_encode_sb = 0;
@@ -493,7 +491,6 @@
   sf->use_fast_coef_costing = 0;
   sf->mode_skip_start = MAX_MODES;  // Mode index at which mode skip mask set
   sf->schedule_mode_search = 0;
-  sf->use_nonrd_pick_mode = 0;
   for (i = 0; i < BLOCK_SIZES; ++i)
     sf->inter_mode_mask[i] = INTER_ALL;
   sf->max_intra_bsize = BLOCK_64X64;
diff --git a/vp10/encoder/speed_features.h b/vp10/encoder/speed_features.h
index 70c11e9..08a1893 100644
--- a/vp10/encoder/speed_features.h
+++ b/vp10/encoder/speed_features.h
@@ -141,7 +141,7 @@
 } INTERP_FILTER_MASK;
 
 typedef enum {
-  // Search partitions using RD/NONRD criterion
+  // Search partitions using RD criterion
   SEARCH_PARTITION,
 
   // Always use a fixed size partition
@@ -363,9 +363,6 @@
   // by only looking at counts from 1/2 the bands.
   FAST_COEFF_UPDATE use_fast_coef_updates;
 
-  // This flag controls the use of non-RD mode decision.
-  int use_nonrd_pick_mode;
-
   // A binary mask indicating if NEARESTMV, NEARMV, ZEROMV, NEWMV
   // modes are used in order from LSB to MSB for each BLOCK_SIZE.
   int inter_mode_mask[BLOCK_SIZES];
diff --git a/vp10/encoder/svc_layercontext.c b/vp10/encoder/svc_layercontext.c
index 6c73256..49d46ee 100644
--- a/vp10/encoder/svc_layercontext.c
+++ b/vp10/encoder/svc_layercontext.c
@@ -141,8 +141,8 @@
         lrc->maximum_buffer_size =
             (int64_t)(rc->maximum_buffer_size * bitrate_alloc);
         lrc->bits_off_target =
-            MIN(lrc->bits_off_target, lrc->maximum_buffer_size);
-        lrc->buffer_level = MIN(lrc->buffer_level, lrc->maximum_buffer_size);
+            VPXMIN(lrc->bits_off_target, lrc->maximum_buffer_size);
+        lrc->buffer_level = VPXMIN(lrc->buffer_level, lrc->maximum_buffer_size);
         lc->framerate = cpi->framerate / oxcf->ts_rate_decimator[tl];
         lrc->avg_frame_bandwidth = (int)(lc->target_bandwidth / lc->framerate);
         lrc->max_frame_bandwidth = rc->max_frame_bandwidth;
@@ -173,9 +173,9 @@
           (int64_t)(rc->optimal_buffer_level * bitrate_alloc);
       lrc->maximum_buffer_size =
           (int64_t)(rc->maximum_buffer_size * bitrate_alloc);
-      lrc->bits_off_target = MIN(lrc->bits_off_target,
-                                 lrc->maximum_buffer_size);
-      lrc->buffer_level = MIN(lrc->buffer_level, lrc->maximum_buffer_size);
+      lrc->bits_off_target = VPXMIN(lrc->bits_off_target,
+                                    lrc->maximum_buffer_size);
+      lrc->buffer_level = VPXMIN(lrc->buffer_level, lrc->maximum_buffer_size);
       // Update framerate-related quantities.
       if (svc->number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR) {
         lc->framerate = cpi->framerate / oxcf->ts_rate_decimator[layer];
diff --git a/vp10/encoder/temporal_filter.c b/vp10/encoder/temporal_filter.c
index 1341dc4..dba7891 100644
--- a/vp10/encoder/temporal_filter.c
+++ b/vp10/encoder/temporal_filter.c
@@ -242,7 +242,7 @@
   xd->plane[0].pre[0].stride = stride;
 
   step_param = mv_sf->reduce_first_step_size;
-  step_param = MIN(step_param, MAX_MVSEARCH_STEPS - 2);
+  step_param = VPXMIN(step_param, MAX_MVSEARCH_STEPS - 2);
 
   // Ignore mv costing by sending NULL pointer instead of cost arrays
   vp10_hex_search(x, &best_ref_mv1_full, step_param, sadpb, 1,
diff --git a/vp10/vp10_dx_iface.c b/vp10/vp10_dx_iface.c
index 1848103..0ff64df 100644
--- a/vp10/vp10_dx_iface.c
+++ b/vp10/vp10_dx_iface.c
@@ -183,7 +183,7 @@
   si->w = si->h = 0;
 
   if (decrypt_cb) {
-    data_sz = MIN(sizeof(clear_buffer), data_sz);
+    data_sz = VPXMIN(sizeof(clear_buffer), data_sz);
     decrypt_cb(decrypt_state, data, clear_buffer, data_sz);
     data = clear_buffer;
   }
diff --git a/vp10/vp10cx.mk b/vp10/vp10cx.mk
index 283b705..b1990b3 100644
--- a/vp10/vp10cx.mk
+++ b/vp10/vp10cx.mk
@@ -48,7 +48,6 @@
 VP10_CX_SRCS-yes += encoder/ratectrl.h
 VP10_CX_SRCS-yes += encoder/rd.h
 VP10_CX_SRCS-yes += encoder/rdopt.h
-VP10_CX_SRCS-yes += encoder/pickmode.h
 VP10_CX_SRCS-yes += encoder/svc_layercontext.h
 VP10_CX_SRCS-yes += encoder/tokenize.h
 VP10_CX_SRCS-yes += encoder/treewriter.h
@@ -60,7 +59,6 @@
 VP10_CX_SRCS-yes += encoder/ratectrl.c
 VP10_CX_SRCS-yes += encoder/rd.c
 VP10_CX_SRCS-yes += encoder/rdopt.c
-VP10_CX_SRCS-yes += encoder/pickmode.c
 VP10_CX_SRCS-yes += encoder/segmentation.c
 VP10_CX_SRCS-yes += encoder/segmentation.h
 VP10_CX_SRCS-yes += encoder/speed_features.c
diff --git a/vp8/decoder/dboolhuff.h b/vp8/decoder/dboolhuff.h
index 51c5adc..cc9eaaf 100644
--- a/vp8/decoder/dboolhuff.h
+++ b/vp8/decoder/dboolhuff.h
@@ -15,7 +15,7 @@
 #include <stddef.h>
 #include <limits.h>
 
-#include "vpx_config.h"
+#include "./vpx_config.h"
 #include "vpx_ports/mem.h"
 #include "vpx/vp8dx.h"
 #include "vpx/vpx_integer.h"
@@ -95,7 +95,7 @@
     return bit;
 }
 
-static int vp8_decode_value(BOOL_DECODER *br, int bits)
+static INLINE int vp8_decode_value(BOOL_DECODER *br, int bits)
 {
     int z = 0;
     int bit;
@@ -108,7 +108,7 @@
     return z;
 }
 
-static int vp8dx_bool_error(BOOL_DECODER *br)
+static INLINE int vp8dx_bool_error(BOOL_DECODER *br)
 {
     /* Check if we have reached the end of the buffer.
      *
diff --git a/vp8/decoder/treereader.h b/vp8/decoder/treereader.h
index 35ee696..f7d23c3 100644
--- a/vp8/decoder/treereader.h
+++ b/vp8/decoder/treereader.h
@@ -12,6 +12,7 @@
 #ifndef VP8_DECODER_TREEREADER_H_
 #define VP8_DECODER_TREEREADER_H_
 
+#include "./vpx_config.h"
 #include "vp8/common/treecoder.h"
 #include "dboolhuff.h"
 
@@ -28,7 +29,7 @@
 
 /* Intent of tree data structure is to make decoding trivial. */
 
-static int vp8_treed_read(
+static INLINE int vp8_treed_read(
     vp8_reader *const r,        /* !!! must return a 0 or 1 !!! */
     vp8_tree t,
     const vp8_prob *const p
diff --git a/vp8/vp8_cx_iface.c b/vp8/vp8_cx_iface.c
index fe88cd4..960c2f4 100644
--- a/vp8/vp8_cx_iface.c
+++ b/vp8/vp8_cx_iface.c
@@ -879,7 +879,8 @@
     }
     ctx->control_frame_flags = 0;
 
-    res = set_reference_and_update(ctx, flags);
+    if (!res)
+        res = set_reference_and_update(ctx, flags);
 
     /* Handle fixed keyframe intervals */
     if (ctx->cfg.kf_mode == VPX_KF_AUTO
diff --git a/vp8/vp8_dx_iface.c b/vp8/vp8_dx_iface.c
index 72e4770..1af3aa6 100644
--- a/vp8/vp8_dx_iface.c
+++ b/vp8/vp8_dx_iface.c
@@ -42,8 +42,6 @@
 } mem_seg_id_t;
 #define NELEMENTS(x) ((int)(sizeof(x)/sizeof(x[0])))
 
-static unsigned long vp8_priv_sz(const vpx_codec_dec_cfg_t *si, vpx_codec_flags_t);
-
 struct vpx_codec_alg_priv
 {
     vpx_codec_priv_t        base;
@@ -68,18 +66,6 @@
     FRAGMENT_DATA           fragments;
 };
 
-static unsigned long vp8_priv_sz(const vpx_codec_dec_cfg_t *si, vpx_codec_flags_t flags)
-{
-    /* Although this declaration is constant, we can't use it in the requested
-     * segments list because we want to define the requested segments list
-     * before defining the private type (so that the number of memory maps is
-     * known)
-     */
-    (void)si;
-    (void)flags;
-    return sizeof(vpx_codec_alg_priv_t);
-}
-
 static void vp8_init_ctx(vpx_codec_ctx_t *ctx)
 {
     vpx_codec_alg_priv_t *priv =
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index d776b44..b757c01 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -235,7 +235,7 @@
     return TX_4X4;
   } else {
     const BLOCK_SIZE plane_bsize = ss_size_lookup[bsize][xss][yss];
-    return MIN(y_tx_size, max_txsize_lookup[plane_bsize]);
+    return VPXMIN(y_tx_size, max_txsize_lookup[plane_bsize]);
   }
 }
 
diff --git a/vp9/common/vp9_common_data.c b/vp9/common/vp9_common_data.c
index 0bf7cbc..ca7f4ad 100644
--- a/vp9/common/vp9_common_data.c
+++ b/vp9/common/vp9_common_data.c
@@ -27,7 +27,7 @@
 const uint8_t num_8x8_blocks_high_lookup[BLOCK_SIZES] =
   {1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8};
 
-// MIN(3, MIN(b_width_log2(bsize), b_height_log2(bsize)))
+// VPXMIN(3, VPXMIN(b_width_log2(bsize), b_height_log2(bsize)))
 const uint8_t size_group_lookup[BLOCK_SIZES] =
   {0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 3};
 
diff --git a/vp9/common/vp9_loopfilter.c b/vp9/common/vp9_loopfilter.c
index 0915918..08d55c6 100644
--- a/vp9/common/vp9_loopfilter.c
+++ b/vp9/common/vp9_loopfilter.c
@@ -1588,7 +1588,7 @@
   if (partial_frame && cm->mi_rows > 8) {
     start_mi_row = cm->mi_rows >> 1;
     start_mi_row &= 0xfffffff8;
-    mi_rows_to_filter = MAX(cm->mi_rows / 8, 8);
+    mi_rows_to_filter = VPXMAX(cm->mi_rows / 8, 8);
   }
   end_mi_row = start_mi_row + mi_rows_to_filter;
   vp9_loop_filter_frame_init(cm, frame_filter_level);
diff --git a/vp9/common/vp9_postproc.c b/vp9/common/vp9_postproc.c
index 71ab861..6c87cee 100644
--- a/vp9/common/vp9_postproc.c
+++ b/vp9/common/vp9_postproc.c
@@ -625,7 +625,7 @@
 
 int vp9_post_proc_frame(struct VP9Common *cm,
                         YV12_BUFFER_CONFIG *dest, vp9_ppflags_t *ppflags) {
-  const int q = MIN(105, cm->lf.filter_level * 2);
+  const int q = VPXMIN(105, cm->lf.filter_level * 2);
   const int flags = ppflags->post_proc_flag;
   YV12_BUFFER_CONFIG *const ppbuf = &cm->post_proc_buffer;
   struct postproc_state *const ppstate = &cm->postproc_state;
diff --git a/vp9/common/vp9_pred_common.h b/vp9/common/vp9_pred_common.h
index 67b95db..d56eacf 100644
--- a/vp9/common/vp9_pred_common.h
+++ b/vp9/common/vp9_pred_common.h
@@ -24,14 +24,14 @@
   const int mi_offset = mi_row * cm->mi_cols + mi_col;
   const int bw = num_8x8_blocks_wide_lookup[bsize];
   const int bh = num_8x8_blocks_high_lookup[bsize];
-  const int xmis = MIN(cm->mi_cols - mi_col, bw);
-  const int ymis = MIN(cm->mi_rows - mi_row, bh);
+  const int xmis = VPXMIN(cm->mi_cols - mi_col, bw);
+  const int ymis = VPXMIN(cm->mi_rows - mi_row, bh);
   int x, y, segment_id = MAX_SEGMENTS;
 
   for (y = 0; y < ymis; ++y)
     for (x = 0; x < xmis; ++x)
-      segment_id = MIN(segment_id,
-                       segment_ids[mi_offset + y * cm->mi_cols + x]);
+      segment_id =
+          VPXMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]);
 
   assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
   return segment_id;
diff --git a/vp9/common/vp9_thread_common.c b/vp9/common/vp9_thread_common.c
index 6b11c93..2e6285a 100644
--- a/vp9/common/vp9_thread_common.c
+++ b/vp9/common/vp9_thread_common.c
@@ -165,7 +165,7 @@
   // Decoder may allocate more threads than number of tiles based on user's
   // input.
   const int tile_cols = 1 << cm->log2_tile_cols;
-  const int num_workers = MIN(nworkers, tile_cols);
+  const int num_workers = VPXMIN(nworkers, tile_cols);
   int i;
 
   if (!lf_sync->sync_range || sb_rows != lf_sync->rows ||
@@ -229,7 +229,7 @@
   if (partial_frame && cm->mi_rows > 8) {
     start_mi_row = cm->mi_rows >> 1;
     start_mi_row &= 0xfffffff8;
-    mi_rows_to_filter = MAX(cm->mi_rows / 8, 8);
+    mi_rows_to_filter = VPXMAX(cm->mi_rows / 8, 8);
   }
   end_mi_row = start_mi_row + mi_rows_to_filter;
   vp9_loop_filter_frame_init(cm, frame_filter_level);
diff --git a/vp9/common/vp9_tile_common.c b/vp9/common/vp9_tile_common.c
index 7a20e0a..f276412 100644
--- a/vp9/common/vp9_tile_common.c
+++ b/vp9/common/vp9_tile_common.c
@@ -18,7 +18,7 @@
 static int get_tile_offset(int idx, int mis, int log2) {
   const int sb_cols = mi_cols_aligned_to_sb(mis) >> MI_BLOCK_SIZE_LOG2;
   const int offset = ((idx * sb_cols) >> log2) << MI_BLOCK_SIZE_LOG2;
-  return MIN(offset, mis);
+  return VPXMIN(offset, mis);
 }
 
 void vp9_tile_set_row(TileInfo *tile, const VP9_COMMON *cm, int row) {
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index fb7b3b8..f9f991d 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -658,7 +658,7 @@
     // pixels of each superblock row can be changed by next superblock row.
     if (pbi->frame_parallel_decode)
       vp9_frameworker_wait(pbi->frame_worker_owner, ref_frame_buf,
-                           MAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
+                           VPXMAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
 
     // Skip border extension if block is inside the frame.
     if (x0 < 0 || x0 > frame_width - 1 || x1 < 0 || x1 > frame_width - 1 ||
@@ -686,7 +686,7 @@
      if (pbi->frame_parallel_decode) {
        const int y1 = (y0_16 + (h - 1) * ys) >> SUBPEL_BITS;
        vp9_frameworker_wait(pbi->frame_worker_owner, ref_frame_buf,
-                            MAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
+                            VPXMAX(0, (y1 + 7)) << (plane == 0 ? 0 : 1));
      }
   }
 #if CONFIG_VP9_HIGHBITDEPTH
@@ -757,8 +757,8 @@
 static INLINE TX_SIZE dec_get_uv_tx_size(const MB_MODE_INFO *mbmi,
                                          int n4_wl, int n4_hl) {
   // get minimum log2 num4x4s dimension
-  const int x = MIN(n4_wl, n4_hl);
-  return MIN(mbmi->tx_size,  x);
+  const int x = VPXMIN(n4_wl, n4_hl);
+  return VPXMIN(mbmi->tx_size,  x);
 }
 
 static INLINE void dec_reset_skip_context(MACROBLOCKD *xd) {
@@ -819,8 +819,8 @@
   const int less8x8 = bsize < BLOCK_8X8;
   const int bw = 1 << (bwl - 1);
   const int bh = 1 << (bhl - 1);
-  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
-  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+  const int x_mis = VPXMIN(bw, cm->mi_cols - mi_col);
+  const int y_mis = VPXMIN(bh, cm->mi_rows - mi_row);
 
   MB_MODE_INFO *mbmi = set_offsets(cm, xd, bsize, mi_row, mi_col,
                                    bw, bh, x_mis, y_mis, bwl, bhl);
@@ -1603,7 +1603,7 @@
   const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
   const int tile_cols = 1 << cm->log2_tile_cols;
   const int tile_rows = 1 << cm->log2_tile_rows;
-  const int num_workers = MIN(pbi->max_threads & ~1, tile_cols);
+  const int num_workers = VPXMIN(pbi->max_threads & ~1, tile_cols);
   TileBuffer tile_buffers[1][1 << 6];
   int n;
   int final_worker = -1;
@@ -1670,7 +1670,7 @@
     int group_start = 0;
     while (group_start < tile_cols) {
       const TileBuffer largest = tile_buffers[0][group_start];
-      const int group_end = MIN(group_start + num_workers, tile_cols) - 1;
+      const int group_end = VPXMIN(group_start + num_workers, tile_cols) - 1;
       memmove(tile_buffers[0] + group_start, tile_buffers[0] + group_start + 1,
               (group_end - group_start) * sizeof(tile_buffers[0][0]));
       tile_buffers[0][group_end] = largest;
@@ -2102,7 +2102,7 @@
   rb->error_handler = error_handler;
   rb->error_handler_data = &pbi->common;
   if (pbi->decrypt_cb) {
-    const int n = (int)MIN(MAX_VP9_HEADER_SIZE, data_end - data);
+    const int n = (int)VPXMIN(MAX_VP9_HEADER_SIZE, data_end - data);
     pbi->decrypt_cb(pbi->decrypt_state, data, clear_data, n);
     rb->bit_buffer = clear_data;
     rb->bit_buffer_end = clear_data + n;
diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c
index 33818a9..9db88a4 100644
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -87,7 +87,7 @@
   if (allow_select && tx_mode == TX_MODE_SELECT && bsize >= BLOCK_8X8)
     return read_selected_tx_size(cm, xd, max_tx_size, r);
   else
-    return MIN(max_tx_size, tx_mode_to_biggest_tx_size[tx_mode]);
+    return VPXMIN(max_tx_size, tx_mode_to_biggest_tx_size[tx_mode]);
 }
 
 static int dec_get_segment_id(const VP9_COMMON *cm, const uint8_t *segment_ids,
@@ -96,8 +96,8 @@
 
   for (y = 0; y < y_mis; y++)
     for (x = 0; x < x_mis; x++)
-      segment_id = MIN(segment_id,
-                       segment_ids[mi_offset + y * cm->mi_cols + x]);
+      segment_id =
+          VPXMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]);
 
   assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
   return segment_id;
@@ -156,8 +156,8 @@
   const int bh = xd->plane[0].n4_h >> 1;
 
   // TODO(slavarnway): move x_mis, y_mis into xd ?????
-  const int x_mis = MIN(cm->mi_cols - mi_col, bw);
-  const int y_mis = MIN(cm->mi_rows - mi_row, bh);
+  const int x_mis = VPXMIN(cm->mi_cols - mi_col, bw);
+  const int y_mis = VPXMIN(cm->mi_rows - mi_row, bh);
 
   if (!seg->enabled)
     return 0;  // Default for disabled segmentation
@@ -212,8 +212,8 @@
   const int bh = xd->plane[0].n4_h >> 1;
 
   // TODO(slavarnway): move x_mis, y_mis into xd ?????
-  const int x_mis = MIN(cm->mi_cols - mi_col, bw);
-  const int y_mis = MIN(cm->mi_rows - mi_row, bh);
+  const int x_mis = VPXMIN(cm->mi_cols - mi_col, bw);
+  const int y_mis = VPXMIN(cm->mi_rows - mi_row, bh);
 
   mbmi->segment_id = read_intra_segment_id(cm, mi_offset, x_mis, y_mis, r);
   mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r);
diff --git a/vp9/encoder/vp9_aq_complexity.c b/vp9/encoder/vp9_aq_complexity.c
index 15f227f..93ea270 100644
--- a/vp9/encoder/vp9_aq_complexity.c
+++ b/vp9/encoder/vp9_aq_complexity.c
@@ -117,8 +117,8 @@
   const int mi_offset = mi_row * cm->mi_cols + mi_col;
   const int bw = num_8x8_blocks_wide_lookup[BLOCK_64X64];
   const int bh = num_8x8_blocks_high_lookup[BLOCK_64X64];
-  const int xmis = MIN(cm->mi_cols - mi_col, num_8x8_blocks_wide_lookup[bs]);
-  const int ymis = MIN(cm->mi_rows - mi_row, num_8x8_blocks_high_lookup[bs]);
+  const int xmis = VPXMIN(cm->mi_cols - mi_col, num_8x8_blocks_wide_lookup[bs]);
+  const int ymis = VPXMIN(cm->mi_rows - mi_row, num_8x8_blocks_high_lookup[bs]);
   int x, y;
   int i;
   unsigned char segment;
@@ -136,7 +136,7 @@
 
     vpx_clear_system_state();
     low_var_thresh = (cpi->oxcf.pass == 2)
-      ? MAX(cpi->twopass.mb_av_energy, MIN_DEFAULT_LV_THRESH)
+      ? VPXMAX(cpi->twopass.mb_av_energy, MIN_DEFAULT_LV_THRESH)
       : DEFAULT_LV_THRESH;
 
     vp9_setup_src_planes(mb, cpi->Source, mi_row, mi_col);
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.c b/vp9/encoder/vp9_aq_cyclicrefresh.c
index e6b3686..813c339 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.c
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.c
@@ -223,8 +223,8 @@
   CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
   const int bw = num_8x8_blocks_wide_lookup[bsize];
   const int bh = num_8x8_blocks_high_lookup[bsize];
-  const int xmis = MIN(cm->mi_cols - mi_col, bw);
-  const int ymis = MIN(cm->mi_rows - mi_row, bh);
+  const int xmis = VPXMIN(cm->mi_cols - mi_col, bw);
+  const int ymis = VPXMIN(cm->mi_rows - mi_row, bh);
   const int block_index = mi_row * cm->mi_cols + mi_col;
   const int refresh_this_block = candidate_refresh_aq(cr, mbmi, rate, dist,
                                                       bsize);
@@ -416,10 +416,10 @@
     assert(mi_col >= 0 && mi_col < cm->mi_cols);
     bl_index = mi_row * cm->mi_cols + mi_col;
     // Loop through all 8x8 blocks in superblock and update map.
-    xmis = MIN(cm->mi_cols - mi_col,
-               num_8x8_blocks_wide_lookup[BLOCK_64X64]);
-    ymis = MIN(cm->mi_rows - mi_row,
-               num_8x8_blocks_high_lookup[BLOCK_64X64]);
+    xmis =
+        VPXMIN(cm->mi_cols - mi_col, num_8x8_blocks_wide_lookup[BLOCK_64X64]);
+    ymis =
+        VPXMIN(cm->mi_rows - mi_row, num_8x8_blocks_high_lookup[BLOCK_64X64]);
     for (y = 0; y < ymis; y++) {
       for (x = 0; x < xmis; x++) {
         const int bl_index2 = bl_index + y * cm->mi_cols + x;
@@ -551,8 +551,9 @@
 
     // Set a more aggressive (higher) q delta for segment BOOST2.
     qindex_delta = compute_deltaq(
-        cpi, cm->base_qindex, MIN(CR_MAX_RATE_TARGET_RATIO,
-        0.1 * cr->rate_boost_fac * cr->rate_ratio_qdelta));
+        cpi, cm->base_qindex,
+        VPXMIN(CR_MAX_RATE_TARGET_RATIO,
+               0.1 * cr->rate_boost_fac * cr->rate_ratio_qdelta));
     cr->qindex_delta[2] = qindex_delta;
     vp9_set_segdata(seg, CR_SEGMENT_ID_BOOST2, SEG_LVL_ALT_Q, qindex_delta);
 
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index d0de095..502fcd5 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -815,7 +815,7 @@
 static void encode_txfm_probs(VP9_COMMON *cm, vpx_writer *w,
                               FRAME_COUNTS *counts) {
   // Mode
-  vpx_write_literal(w, MIN(cm->tx_mode, ALLOW_32X32), 2);
+  vpx_write_literal(w, VPXMIN(cm->tx_mode, ALLOW_32X32), 2);
   if (cm->tx_mode >= ALLOW_32X32)
     vpx_write_bit(w, cm->tx_mode == TX_MODE_SELECT);
 
diff --git a/vp9/encoder/vp9_denoiser.c b/vp9/encoder/vp9_denoiser.c
index 5f99285..bf6c533 100644
--- a/vp9/encoder/vp9_denoiser.c
+++ b/vp9/encoder/vp9_denoiser.c
@@ -120,10 +120,10 @@
             adj = adj_val[2];
         }
         if (diff > 0) {
-          avg[c] = MIN(UINT8_MAX, sig[c] + adj);
+          avg[c] = VPXMIN(UINT8_MAX, sig[c] + adj);
           total_adj += adj;
         } else {
-          avg[c] = MAX(0, sig[c] - adj);
+          avg[c] = VPXMAX(0, sig[c] - adj);
           total_adj -= adj;
         }
       }
@@ -160,13 +160,13 @@
         // Diff positive means we made positive adjustment above
         // (in first try/attempt), so now make negative adjustment to bring
         // denoised signal down.
-        avg[c] = MAX(0, avg[c] - adj);
+        avg[c] = VPXMAX(0, avg[c] - adj);
         total_adj -= adj;
       } else {
         // Diff negative means we made negative adjustment above
         // (in first try/attempt), so now make positive adjustment to bring
         // denoised signal up.
-        avg[c] = MIN(UINT8_MAX, avg[c] + adj);
+        avg[c] = VPXMIN(UINT8_MAX, avg[c] + adj);
         total_adj += adj;
       }
     }
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 295a751..d02d48c 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -979,8 +979,8 @@
   const struct segmentation *const seg = &cm->seg;
   const int bw = num_8x8_blocks_wide_lookup[mi->mbmi.sb_type];
   const int bh = num_8x8_blocks_high_lookup[mi->mbmi.sb_type];
-  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
-  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+  const int x_mis = VPXMIN(bw, cm->mi_cols - mi_col);
+  const int y_mis = VPXMIN(bh, cm->mi_rows - mi_row);
   MV_REF *const frame_mvs =
       cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col;
   int w, h;
@@ -1132,8 +1132,8 @@
 
   mbmi->sb_type = bsize;
   mbmi->mode = ZEROMV;
-  mbmi->tx_size = MIN(max_txsize_lookup[bsize],
-                      tx_mode_to_biggest_tx_size[tx_mode]);
+  mbmi->tx_size =
+      VPXMIN(max_txsize_lookup[bsize], tx_mode_to_biggest_tx_size[tx_mode]);
   mbmi->skip = 1;
   mbmi->uv_mode = DC_PRED;
   mbmi->ref_frame[0] = LAST_FRAME;
@@ -1496,7 +1496,7 @@
                                       int rows_left, int cols_left,
                                       int *bh, int *bw) {
   if (rows_left <= 0 || cols_left <= 0) {
-    return MIN(bsize, BLOCK_8X8);
+    return VPXMIN(bsize, BLOCK_8X8);
   } else {
     for (; bsize > 0; bsize -= 3) {
       *bh = num_8x8_blocks_high_lookup[bsize];
@@ -1672,8 +1672,8 @@
   const struct segmentation *const seg = &cm->seg;
   const int bw = num_8x8_blocks_wide_lookup[mi->mbmi.sb_type];
   const int bh = num_8x8_blocks_high_lookup[mi->mbmi.sb_type];
-  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
-  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
+  const int x_mis = VPXMIN(bw, cm->mi_cols - mi_col);
+  const int y_mis = VPXMIN(bh, cm->mi_rows - mi_row);
 
   *(xd->mi[0]) = ctx->mic;
   *(x->mbmi_ext) = ctx->mbmi_ext;
@@ -1741,7 +1741,7 @@
   if (cpi->oxcf.noise_sensitivity > 0 && output_enabled &&
       cpi->common.frame_type != KEY_FRAME) {
     vp9_denoiser_denoise(&cpi->denoiser, x, mi_row, mi_col,
-                         MAX(BLOCK_8X8, bsize), ctx);
+                         VPXMAX(BLOCK_8X8, bsize), ctx);
   }
 #endif
 
@@ -2133,8 +2133,8 @@
       MODE_INFO *mi = mi_8x8[index+j];
       BLOCK_SIZE sb_type = mi ? mi->mbmi.sb_type : 0;
       bs_hist[sb_type]++;
-      *min_block_size = MIN(*min_block_size, sb_type);
-      *max_block_size = MAX(*max_block_size, sb_type);
+      *min_block_size = VPXMIN(*min_block_size, sb_type);
+      *max_block_size = VPXMAX(*max_block_size, sb_type);
     }
     index += xd->mi_stride;
   }
@@ -2211,8 +2211,8 @@
   if (vp9_active_edge_sb(cpi, mi_row, mi_col)) {
     min_size = BLOCK_4X4;
   } else {
-    min_size = MIN(cpi->sf.rd_auto_partition_min_limit,
-                   MIN(min_size, max_size));
+    min_size =
+        VPXMIN(cpi->sf.rd_auto_partition_min_limit, VPXMIN(min_size, max_size));
   }
 
   // When use_square_partition_only is true, make sure at least one square
@@ -2248,8 +2248,8 @@
       for (idx = 0; idx < mi_width; ++idx) {
         mi = prev_mi[idy * cm->mi_stride + idx];
         bs = mi ? mi->mbmi.sb_type : bsize;
-        min_size = MIN(min_size, bs);
-        max_size = MAX(max_size, bs);
+        min_size = VPXMIN(min_size, bs);
+        max_size = VPXMAX(max_size, bs);
       }
     }
   }
@@ -2258,8 +2258,8 @@
     for (idy = 0; idy < mi_height; ++idy) {
       mi = xd->mi[idy * cm->mi_stride - 1];
       bs = mi ? mi->mbmi.sb_type : bsize;
-      min_size = MIN(min_size, bs);
-      max_size = MAX(max_size, bs);
+      min_size = VPXMIN(min_size, bs);
+      max_size = VPXMAX(max_size, bs);
     }
   }
 
@@ -2267,8 +2267,8 @@
     for (idx = 0; idx < mi_width; ++idx) {
       mi = xd->mi[idx - cm->mi_stride];
       bs = mi ? mi->mbmi.sb_type : bsize;
-      min_size = MIN(min_size, bs);
-      max_size = MAX(max_size, bs);
+      min_size = VPXMIN(min_size, bs);
+      max_size = VPXMAX(max_size, bs);
     }
   }
 
@@ -2433,9 +2433,9 @@
     int mb_row = mi_row >> 1;
     int mb_col = mi_col >> 1;
     int mb_row_end =
-        MIN(mb_row + num_16x16_blocks_high_lookup[bsize], cm->mb_rows);
+        VPXMIN(mb_row + num_16x16_blocks_high_lookup[bsize], cm->mb_rows);
     int mb_col_end =
-        MIN(mb_col + num_16x16_blocks_wide_lookup[bsize], cm->mb_cols);
+        VPXMIN(mb_col + num_16x16_blocks_wide_lookup[bsize], cm->mb_cols);
     int r, c;
 
     // compute a complexity measure, basically measure inconsistency of motion
@@ -2524,9 +2524,9 @@
           int mb_row = mi_row >> 1;
           int mb_col = mi_col >> 1;
           int mb_row_end =
-              MIN(mb_row + num_16x16_blocks_high_lookup[bsize], cm->mb_rows);
+              VPXMIN(mb_row + num_16x16_blocks_high_lookup[bsize], cm->mb_rows);
           int mb_col_end =
-              MIN(mb_col + num_16x16_blocks_wide_lookup[bsize], cm->mb_cols);
+              VPXMIN(mb_col + num_16x16_blocks_wide_lookup[bsize], cm->mb_cols);
           int r, c;
 
           int skip = 1;
@@ -3646,7 +3646,7 @@
   const int last_stride = cpi->Last_Source->y_stride;
 
   // Pick cutoff threshold
-  const int cutoff = (MIN(cm->width, cm->height) >= 720) ?
+  const int cutoff = (VPXMIN(cm->width, cm->height) >= 720) ?
       (cm->MBs * VAR_HIST_LARGE_CUT_OFF / 100) :
       (cm->MBs * VAR_HIST_SMALL_CUT_OFF / 100);
   DECLARE_ALIGNED(16, int, hist[VAR_HIST_BINS]);
@@ -3947,7 +3947,7 @@
 #endif
 
     // If allowed, encoding tiles in parallel with one thread handling one tile.
-    if (MIN(cpi->oxcf.max_threads, 1 << cm->log2_tile_cols) > 1)
+    if (VPXMIN(cpi->oxcf.max_threads, 1 << cm->log2_tile_cols) > 1)
       vp9_encode_tiles_mt(cpi);
     else
       encode_tiles(cpi);
@@ -4162,10 +4162,10 @@
     int plane;
     mbmi->skip = 1;
     for (plane = 0; plane < MAX_MB_PLANE; ++plane)
-      vp9_encode_intra_block_plane(x, MAX(bsize, BLOCK_8X8), plane);
+      vp9_encode_intra_block_plane(x, VPXMAX(bsize, BLOCK_8X8), plane);
     if (output_enabled)
       sum_intra_stats(td->counts, mi);
-    vp9_tokenize_sb(cpi, td, t, !output_enabled, MAX(bsize, BLOCK_8X8));
+    vp9_tokenize_sb(cpi, td, t, !output_enabled, VPXMAX(bsize, BLOCK_8X8));
   } else {
     int ref;
     const int is_compound = has_second_ref(mbmi);
@@ -4178,12 +4178,14 @@
                            &xd->block_refs[ref]->sf);
     }
     if (!(cpi->sf.reuse_inter_pred_sby && ctx->pred_pixel_ready) || seg_skip)
-      vp9_build_inter_predictors_sby(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
+      vp9_build_inter_predictors_sby(xd, mi_row, mi_col,
+                                     VPXMAX(bsize, BLOCK_8X8));
 
-    vp9_build_inter_predictors_sbuv(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
+    vp9_build_inter_predictors_sbuv(xd, mi_row, mi_col,
+                                    VPXMAX(bsize, BLOCK_8X8));
 
-    vp9_encode_sb(x, MAX(bsize, BLOCK_8X8));
-    vp9_tokenize_sb(cpi, td, t, !output_enabled, MAX(bsize, BLOCK_8X8));
+    vp9_encode_sb(x, VPXMAX(bsize, BLOCK_8X8));
+    vp9_tokenize_sb(cpi, td, t, !output_enabled, VPXMAX(bsize, BLOCK_8X8));
   }
 
   if (output_enabled) {
@@ -4197,8 +4199,8 @@
       TX_SIZE tx_size;
       // The new intra coding scheme requires no change of transform size
       if (is_inter_block(&mi->mbmi)) {
-        tx_size = MIN(tx_mode_to_biggest_tx_size[cm->tx_mode],
-                      max_txsize_lookup[bsize]);
+        tx_size = VPXMIN(tx_mode_to_biggest_tx_size[cm->tx_mode],
+                         max_txsize_lookup[bsize]);
       } else {
         tx_size = (bsize >= BLOCK_8X8) ? mbmi->tx_size : TX_4X4;
       }
diff --git a/vp9/encoder/vp9_encodemv.c b/vp9/encoder/vp9_encodemv.c
index 7848c93..a4dee73 100644
--- a/vp9/encoder/vp9_encodemv.c
+++ b/vp9/encoder/vp9_encodemv.c
@@ -216,8 +216,8 @@
   // If auto_mv_step_size is enabled then keep track of the largest
   // motion vector component used.
   if (cpi->sf.mv.auto_mv_step_size) {
-    unsigned int maxv = MAX(abs(mv->row), abs(mv->col)) >> 3;
-    cpi->max_mv_magnitude = MAX(maxv, cpi->max_mv_magnitude);
+    unsigned int maxv = VPXMAX(abs(mv->row), abs(mv->col)) >> 3;
+    cpi->max_mv_magnitude = VPXMAX(maxv, cpi->max_mv_magnitude);
   }
 }
 
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 4654d63..d9ecb9a 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -1490,8 +1490,8 @@
 
   // Under a configuration change, where maximum_buffer_size may change,
   // keep buffer level clipped to the maximum allowed buffer size.
-  rc->bits_off_target = MIN(rc->bits_off_target, rc->maximum_buffer_size);
-  rc->buffer_level = MIN(rc->buffer_level, rc->maximum_buffer_size);
+  rc->bits_off_target = VPXMIN(rc->bits_off_target, rc->maximum_buffer_size);
+  rc->buffer_level = VPXMIN(rc->buffer_level, rc->maximum_buffer_size);
 
   // Set up frame rate and related parameters rate control values.
   vp9_new_framerate(cpi, cpi->framerate);
@@ -2615,7 +2615,7 @@
   if (rc->frame_size_selector == UNSCALED &&
       q >= rc->rf_level_maxq[gf_group->rf_level[gf_group->index]]) {
     const int max_size_thresh = (int)(rate_thresh_mult[SCALE_STEP1]
-        * MAX(rc->this_frame_target, rc->avg_frame_bandwidth));
+        * VPXMAX(rc->this_frame_target, rc->avg_frame_bandwidth));
     scale = rc->projected_frame_size > max_size_thresh ? 1 : 0;
   }
   return scale;
@@ -2998,7 +2998,7 @@
 
 static void set_mv_search_params(VP9_COMP *cpi) {
   const VP9_COMMON *const cm = &cpi->common;
-  const unsigned int max_mv_def = MIN(cm->width, cm->height);
+  const unsigned int max_mv_def = VPXMIN(cm->width, cm->height);
 
   // Default based on max resolution.
   cpi->mv_step_param = vp9_init_search_range(max_mv_def);
@@ -3013,8 +3013,8 @@
         // Allow mv_steps to correspond to twice the max mv magnitude found
         // in the previous frame, capped by the default max_mv_magnitude based
         // on resolution.
-        cpi->mv_step_param =
-            vp9_init_search_range(MIN(max_mv_def, 2 * cpi->max_mv_magnitude));
+        cpi->mv_step_param = vp9_init_search_range(
+            VPXMIN(max_mv_def, 2 * cpi->max_mv_magnitude));
       }
       cpi->max_mv_magnitude = 0;
     }
@@ -3414,7 +3414,7 @@
 
           // Adjust Q
           q = (int)((q * high_err_target) / kf_err);
-          q = MIN(q, (q_high + q_low) >> 1);
+          q = VPXMIN(q, (q_high + q_low) >> 1);
         } else if (kf_err < low_err_target &&
                    rc->projected_frame_size >= frame_under_shoot_limit) {
           // The key frame is much better than the previous frame
@@ -3423,7 +3423,7 @@
 
           // Adjust Q
           q = (int)((q * low_err_target) / kf_err);
-          q = MIN(q, (q_high + q_low + 1) >> 1);
+          q = VPXMIN(q, (q_high + q_low + 1) >> 1);
         }
 
         // Clamp Q to upper and lower limits:
@@ -3432,7 +3432,7 @@
         loop = q != last_q;
       } else if (recode_loop_test(
           cpi, frame_over_shoot_limit, frame_under_shoot_limit,
-          q, MAX(q_high, top_index), bottom_index)) {
+          q, VPXMAX(q_high, top_index), bottom_index)) {
         // Is the projected frame size out of range and are we allowed
         // to attempt to recode.
         int last_q = q;
@@ -3474,12 +3474,12 @@
             vp9_rc_update_rate_correction_factors(cpi);
 
             q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
-                                   bottom_index, MAX(q_high, top_index));
+                                  bottom_index, VPXMAX(q_high, top_index));
 
             while (q < q_low && retries < 10) {
               vp9_rc_update_rate_correction_factors(cpi);
               q = vp9_rc_regulate_q(cpi, rc->this_frame_target,
-                                     bottom_index, MAX(q_high, top_index));
+                                    bottom_index, VPXMAX(q_high, top_index));
               retries++;
             }
           }
@@ -4053,8 +4053,8 @@
       // Average this frame's rate into the last second's average
       // frame rate. If we haven't seen 1 second yet, then average
       // over the whole interval seen.
-      const double interval = MIN((double)(source->ts_end
-                                   - cpi->first_time_stamp_ever), 10000000.0);
+      const double interval = VPXMIN(
+          (double)(source->ts_end - cpi->first_time_stamp_ever), 10000000.0);
       double avg_duration = 10000000.0 / cpi->framerate;
       avg_duration *= (interval - avg_duration + this_duration);
       avg_duration /= interval;
@@ -4118,7 +4118,7 @@
   s->stat[U] += u;
   s->stat[V] += v;
   s->stat[ALL] += all;
-  s->worst = MIN(s->worst, all);
+  s->worst = VPXMIN(s->worst, all);
 }
 #endif  // CONFIG_INTERNAL_STATS
 
@@ -4448,7 +4448,7 @@
           frame_ssim2 = vpx_calc_ssim(orig, recon, &weight);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
-          cpi->worst_ssim= MIN(cpi->worst_ssim, frame_ssim2);
+          cpi->worst_ssim = VPXMIN(cpi->worst_ssim, frame_ssim2);
           cpi->summed_quality += frame_ssim2 * weight;
           cpi->summed_weights += weight;
 
@@ -4485,7 +4485,8 @@
               cpi->Source->y_buffer, cpi->Source->y_stride,
               cm->frame_to_show->y_buffer, cm->frame_to_show->y_stride,
               cpi->Source->y_width, cpi->Source->y_height);
-          cpi->worst_blockiness = MAX(cpi->worst_blockiness, frame_blockiness);
+          cpi->worst_blockiness =
+              VPXMAX(cpi->worst_blockiness, frame_blockiness);
           cpi->total_blockiness += frame_blockiness;
         }
       }
@@ -4505,8 +4506,8 @@
           double consistency = vpx_sse_to_psnr(samples, peak,
                                              (double)cpi->total_inconsistency);
           if (consistency > 0.0)
-            cpi->worst_consistency = MIN(cpi->worst_consistency,
-                                         consistency);
+            cpi->worst_consistency =
+                VPXMIN(cpi->worst_consistency, consistency);
           cpi->total_inconsistency += this_inconsistency;
         }
       }
diff --git a/vp9/encoder/vp9_ethread.c b/vp9/encoder/vp9_ethread.c
index 00025b7..adb3fd8 100644
--- a/vp9/encoder/vp9_ethread.c
+++ b/vp9/encoder/vp9_ethread.c
@@ -67,7 +67,7 @@
   VP9_COMMON *const cm = &cpi->common;
   const int tile_cols = 1 << cm->log2_tile_cols;
   const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
-  const int num_workers = MIN(cpi->oxcf.max_threads, tile_cols);
+  const int num_workers = VPXMIN(cpi->oxcf.max_threads, tile_cols);
   int i;
 
   vp9_init_tile_data(cpi);
@@ -80,7 +80,7 @@
     // resolution.
     if (cpi->use_svc) {
       int max_tile_cols = get_max_tile_cols(cpi);
-      allocated_workers = MIN(cpi->oxcf.max_threads, max_tile_cols);
+      allocated_workers = VPXMIN(cpi->oxcf.max_threads, max_tile_cols);
     }
 
     CHECK_MEM_ERROR(cm, cpi->workers,
diff --git a/vp9/encoder/vp9_extend.c b/vp9/encoder/vp9_extend.c
index 0c304dc..4a47f87 100644
--- a/vp9/encoder/vp9_extend.c
+++ b/vp9/encoder/vp9_extend.c
@@ -111,10 +111,12 @@
   // Motion estimation may use src block variance with the block size up
   // to 64x64, so the right and bottom need to be extended to 64 multiple
   // or up to 16, whichever is greater.
-  const int er_y = MAX(src->y_width + 16, ALIGN_POWER_OF_TWO(src->y_width, 6))
-      - src->y_crop_width;
-  const int eb_y = MAX(src->y_height + 16, ALIGN_POWER_OF_TWO(src->y_height, 6))
-      - src->y_crop_height;
+  const int er_y =
+      VPXMAX(src->y_width + 16, ALIGN_POWER_OF_TWO(src->y_width, 6)) -
+      src->y_crop_width;
+  const int eb_y =
+      VPXMAX(src->y_height + 16, ALIGN_POWER_OF_TWO(src->y_height, 6)) -
+      src->y_crop_height;
   const int uv_width_subsampling = (src->uv_width != src->y_width);
   const int uv_height_subsampling = (src->uv_height != src->y_height);
   const int et_uv = et_y >> uv_height_subsampling;
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index e0c5966..65bf4b4 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -381,7 +381,7 @@
 // for first pass test.
 static int get_search_range(const VP9_COMP *cpi) {
   int sr = 0;
-  const int dim = MIN(cpi->initial_width, cpi->initial_height);
+  const int dim = VPXMIN(cpi->initial_width, cpi->initial_height);
 
   while ((dim << sr) < MAX_FULL_PEL_VAL)
     ++sr;
@@ -1024,7 +1024,7 @@
   // Exclude any image dead zone
   if (image_data_start_row > 0) {
     intra_skip_count =
-      MAX(0, intra_skip_count - (image_data_start_row * cm->mb_cols * 2));
+        VPXMAX(0, intra_skip_count - (image_data_start_row * cm->mb_cols * 2));
   }
 
   {
@@ -1161,7 +1161,7 @@
 
   // Adjustment based on actual quantizer to power term.
   const double power_term =
-      MIN(vp9_convert_qindex_to_q(q, bit_depth) * 0.01 + pt_low, pt_high);
+      VPXMIN(vp9_convert_qindex_to_q(q, bit_depth) * 0.01 + pt_low, pt_high);
 
   // Calculate correction factor.
   if (power_term < 1.0)
@@ -1190,7 +1190,7 @@
   } else {
     const int num_mbs = (cpi->oxcf.resize_mode != RESIZE_NONE)
                         ? cpi->initial_mbs : cpi->common.MBs;
-    const int active_mbs = MAX(1, num_mbs - (int)(num_mbs * inactive_zone));
+    const int active_mbs = VPXMAX(1, num_mbs - (int)(num_mbs * inactive_zone));
     const double av_err_per_mb = section_err / active_mbs;
     const double speed_term = 1.0 + 0.04 * oxcf->speed;
     const double ediv_size_correction = (double)num_mbs / EDIV_SIZE_FACTOR;
@@ -1223,7 +1223,7 @@
 
     // Restriction on active max q for constrained quality mode.
     if (cpi->oxcf.rc_mode == VPX_CQ)
-      q = MAX(q, oxcf->cq_level);
+      q = VPXMAX(q, oxcf->cq_level);
     return q;
   }
 }
@@ -1233,7 +1233,7 @@
   RATE_CONTROL *const rc = &cpi->rc;
   for (i = INTER_NORMAL; i < RATE_FACTOR_LEVELS; ++i) {
     int qdelta = vp9_frame_type_qdelta(cpi, i, rc->worst_quality);
-    rc->rf_level_maxq[i] = MAX(rc->worst_quality + qdelta, rc->best_quality);
+    rc->rf_level_maxq[i] = VPXMAX(rc->worst_quality + qdelta, rc->best_quality);
   }
 }
 
@@ -1364,12 +1364,12 @@
 
 
   if ((sr_diff > LOW_SR_DIFF_TRHESH)) {
-    sr_diff = MIN(sr_diff, SR_DIFF_MAX);
+    sr_diff = VPXMIN(sr_diff, SR_DIFF_MAX);
     sr_decay = 1.0 - (SR_DIFF_PART * sr_diff) -
                (MOTION_AMP_PART * motion_amplitude_factor) -
                (INTRA_PART * modified_pcnt_intra);
   }
-  return MAX(sr_decay, MIN(DEFAULT_DECAY_LIMIT, modified_pct_inter));
+  return VPXMAX(sr_decay, VPXMIN(DEFAULT_DECAY_LIMIT, modified_pct_inter));
 }
 
 // This function gives an estimate of how badly we believe the prediction
@@ -1379,7 +1379,7 @@
   const double zero_motion_pct = frame->pcnt_inter -
                                  frame->pcnt_motion;
   double sr_decay = get_sr_decay_rate(cpi, frame);
-  return MIN(sr_decay, zero_motion_pct);
+  return VPXMIN(sr_decay, zero_motion_pct);
 }
 
 #define ZM_POWER_FACTOR 0.75
@@ -1391,8 +1391,8 @@
     (0.95 * pow((next_frame->pcnt_inter - next_frame->pcnt_motion),
                 ZM_POWER_FACTOR));
 
-  return MAX(zero_motion_factor,
-             (sr_decay_rate + ((1.0 - sr_decay_rate) * zero_motion_factor)));
+  return VPXMAX(zero_motion_factor,
+                (sr_decay_rate + ((1.0 - sr_decay_rate) * zero_motion_factor)));
 }
 
 // Function to test for a condition where a complex transition is followed
@@ -1483,12 +1483,12 @@
   const double lq =
     vp9_convert_qindex_to_q(cpi->rc.avg_frame_qindex[INTER_FRAME],
                             cpi->common.bit_depth);
-  const double boost_q_correction = MIN((0.5 + (lq * 0.015)), 1.5);
+  const double boost_q_correction = VPXMIN((0.5 + (lq * 0.015)), 1.5);
   int num_mbs = (cpi->oxcf.resize_mode != RESIZE_NONE)
                 ? cpi->initial_mbs : cpi->common.MBs;
 
   // Correct for any inactive region in the image
-  num_mbs = (int)MAX(1, num_mbs * calculate_active_area(cpi, this_frame));
+  num_mbs = (int)VPXMAX(1, num_mbs * calculate_active_area(cpi, this_frame));
 
   // Underlying boost factor is based on inter error ratio.
   frame_boost = (BASELINE_ERR_PER_MB * num_mbs) /
@@ -1504,7 +1504,7 @@
   else
     frame_boost += frame_boost * (this_frame_mv_in_out / 2.0);
 
-  return MIN(frame_boost, max_boost * boost_q_correction);
+  return VPXMIN(frame_boost, max_boost * boost_q_correction);
 }
 
 static int calc_arf_boost(VP9_COMP *cpi, int offset,
@@ -1593,7 +1593,7 @@
   arf_boost = (*f_boost + *b_boost);
   if (arf_boost < ((b_frames + f_frames) * 20))
     arf_boost = ((b_frames + f_frames) * 20);
-  arf_boost = MAX(arf_boost, MIN_ARF_GF_BOOST);
+  arf_boost = VPXMAX(arf_boost, MIN_ARF_GF_BOOST);
 
   return arf_boost;
 }
@@ -1664,7 +1664,8 @@
   }
 
   // Calculate the number of extra bits for use in the boosted frame or frames.
-  return MAX((int)(((int64_t)boost * total_group_bits) / allocation_chunks), 0);
+  return VPXMAX((int)(((int64_t)boost * total_group_bits) / allocation_chunks),
+                0);
 }
 
 // Current limit on maximum number of active arfs in a GF/ARF group.
@@ -1803,7 +1804,7 @@
     gf_group->arf_ref_idx[frame_index] = arf_buffer_indices[arf_idx];
 
     target_frame_size = clamp(target_frame_size, 0,
-                              MIN(max_bits, (int)total_group_bits));
+                              VPXMIN(max_bits, (int)total_group_bits));
 
     gf_group->update_type[frame_index] = LF_UPDATE;
     gf_group->rf_level[frame_index] = INTER_NORMAL;
@@ -1924,7 +1925,7 @@
     int int_lbq =
       (int)(vp9_convert_qindex_to_q(rc->last_boosted_qindex,
                                    cpi->common.bit_depth));
-    active_min_gf_interval = rc->min_gf_interval + MIN(2, int_max_q / 200);
+    active_min_gf_interval = rc->min_gf_interval + VPXMIN(2, int_max_q / 200);
     if (active_min_gf_interval > rc->max_gf_interval)
       active_min_gf_interval = rc->max_gf_interval;
 
@@ -1935,7 +1936,7 @@
       // bits to spare and are better with a smaller interval and smaller boost.
       // At high Q when there are few bits to spare we are better with a longer
       // interval to spread the cost of the GF.
-      active_max_gf_interval = 12 + MIN(4, (int_lbq / 6));
+      active_max_gf_interval = 12 + VPXMIN(4, (int_lbq / 6));
       if (active_max_gf_interval < active_min_gf_interval)
         active_max_gf_interval = active_min_gf_interval;
 
@@ -1980,8 +1981,8 @@
       decay_accumulator = decay_accumulator * loop_decay_rate;
 
       // Monitor for static sections.
-      zero_motion_accumulator =
-        MIN(zero_motion_accumulator, get_zero_motion_factor(cpi, &next_frame));
+      zero_motion_accumulator = VPXMIN(
+          zero_motion_accumulator, get_zero_motion_factor(cpi, &next_frame));
 
       // Break clause to detect very still sections after motion. For example,
       // a static image after a fade or other transition.
@@ -2037,7 +2038,7 @@
       (cpi->multi_arf_allowed && (rc->baseline_gf_interval >= 6) &&
       (zero_motion_accumulator < 0.995)) ? 1 : 0;
   } else {
-    rc->gfu_boost = MAX((int)boost_score, MIN_ARF_GF_BOOST);
+    rc->gfu_boost = VPXMAX((int)boost_score, MIN_ARF_GF_BOOST);
     rc->source_alt_ref_pending = 0;
   }
 
@@ -2092,11 +2093,11 @@
     // rc factor is a weight factor that corrects for local rate control drift.
     double rc_factor = 1.0;
     if (rc->rate_error_estimate > 0) {
-      rc_factor = MAX(RC_FACTOR_MIN,
-                      (double)(100 - rc->rate_error_estimate) / 100.0);
+      rc_factor = VPXMAX(RC_FACTOR_MIN,
+                         (double)(100 - rc->rate_error_estimate) / 100.0);
     } else {
-      rc_factor = MIN(RC_FACTOR_MAX,
-                      (double)(100 - rc->rate_error_estimate) / 100.0);
+      rc_factor = VPXMIN(RC_FACTOR_MAX,
+                         (double)(100 - rc->rate_error_estimate) / 100.0);
     }
     tmp_q =
       get_twopass_worst_quality(cpi, group_av_err,
@@ -2104,7 +2105,7 @@
                                 vbr_group_bits_per_frame,
                                 twopass->kfgroup_inter_fraction * rc_factor);
     twopass->active_worst_quality =
-      MAX(tmp_q, twopass->active_worst_quality >> 1);
+        VPXMAX(tmp_q, twopass->active_worst_quality >> 1);
   }
 #endif
 
@@ -2421,7 +2422,7 @@
   } else {
     twopass->kf_group_bits = 0;
   }
-  twopass->kf_group_bits = MAX(0, twopass->kf_group_bits);
+  twopass->kf_group_bits = VPXMAX(0, twopass->kf_group_bits);
 
   // Reset the first pass file position.
   reset_fpf_position(twopass, start_position);
@@ -2435,9 +2436,8 @@
       break;
 
     // Monitor for static sections.
-    zero_motion_accumulator =
-      MIN(zero_motion_accumulator,
-          get_zero_motion_factor(cpi, &next_frame));
+    zero_motion_accumulator = VPXMIN(
+        zero_motion_accumulator, get_zero_motion_factor(cpi, &next_frame));
 
     // Not all frames in the group are necessarily used in calculating boost.
     if ((i <= rc->max_gf_interval) ||
@@ -2450,7 +2450,7 @@
         const double loop_decay_rate =
           get_prediction_decay_rate(cpi, &next_frame);
         decay_accumulator *= loop_decay_rate;
-        decay_accumulator = MAX(decay_accumulator, MIN_DECAY_FACTOR);
+        decay_accumulator = VPXMAX(decay_accumulator, MIN_DECAY_FACTOR);
         av_decay_accumulator += decay_accumulator;
         ++loop_decay_counter;
       }
@@ -2471,8 +2471,8 @@
 
   // Apply various clamps for min and max boost
   rc->kf_boost = (int)(av_decay_accumulator * boost_score);
-  rc->kf_boost = MAX(rc->kf_boost, (rc->frames_to_key * 3));
-  rc->kf_boost = MAX(rc->kf_boost, MIN_KF_BOOST);
+  rc->kf_boost = VPXMAX(rc->kf_boost, (rc->frames_to_key * 3));
+  rc->kf_boost = VPXMAX(rc->kf_boost, MIN_KF_BOOST);
 
   // Work out how many bits to allocate for the key frame itself.
   kf_bits = calculate_boost_bits((rc->frames_to_key - 1),
@@ -2770,7 +2770,7 @@
   // is designed to prevent extreme behaviour at the end of a clip
   // or group of frames.
   rc->vbr_bits_off_target += rc->base_frame_target - rc->projected_frame_size;
-  twopass->bits_left = MAX(twopass->bits_left - bits_used, 0);
+  twopass->bits_left = VPXMAX(twopass->bits_left - bits_used, 0);
 
   // Calculate the pct rc error.
   if (rc->total_actual_bits) {
@@ -2786,7 +2786,7 @@
     twopass->kf_group_bits -= bits_used;
     twopass->last_kfgroup_zeromotion_pct = twopass->kf_zeromotion_pct;
   }
-  twopass->kf_group_bits = MAX(twopass->kf_group_bits, 0);
+  twopass->kf_group_bits = VPXMAX(twopass->kf_group_bits, 0);
 
   // Increment the gf group index ready for the next frame.
   ++twopass->gf_group.index;
@@ -2836,18 +2836,18 @@
         rc->vbr_bits_off_target_fast +=
           fast_extra_thresh - rc->projected_frame_size;
         rc->vbr_bits_off_target_fast =
-          MIN(rc->vbr_bits_off_target_fast, (4 * rc->avg_frame_bandwidth));
+          VPXMIN(rc->vbr_bits_off_target_fast, (4 * rc->avg_frame_bandwidth));
 
         // Fast adaptation of minQ if necessary to use up the extra bits.
         if (rc->avg_frame_bandwidth) {
           twopass->extend_minq_fast =
             (int)(rc->vbr_bits_off_target_fast * 8 / rc->avg_frame_bandwidth);
         }
-        twopass->extend_minq_fast = MIN(twopass->extend_minq_fast,
-                                        minq_adj_limit - twopass->extend_minq);
+        twopass->extend_minq_fast = VPXMIN(
+            twopass->extend_minq_fast, minq_adj_limit - twopass->extend_minq);
       } else if (rc->vbr_bits_off_target_fast) {
-        twopass->extend_minq_fast = MIN(twopass->extend_minq_fast,
-                                        minq_adj_limit - twopass->extend_minq);
+        twopass->extend_minq_fast = VPXMIN(
+            twopass->extend_minq_fast, minq_adj_limit - twopass->extend_minq);
       } else {
         twopass->extend_minq_fast = 0;
       }
diff --git a/vp9/encoder/vp9_mbgraph.c b/vp9/encoder/vp9_mbgraph.c
index d59f315..1784666 100644
--- a/vp9/encoder/vp9_mbgraph.c
+++ b/vp9/encoder/vp9_mbgraph.c
@@ -41,7 +41,7 @@
 
   // Further step/diamond searches as necessary
   int step_param = mv_sf->reduce_first_step_size;
-  step_param = MIN(step_param, MAX_MVSEARCH_STEPS - 2);
+  step_param = VPXMIN(step_param, MAX_MVSEARCH_STEPS - 2);
 
   vp9_set_mv_search_range(x, ref_mv);
 
diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index aa3e51c..6a02481 100644
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -37,10 +37,10 @@
   int col_max = (mv->col >> 3) + MAX_FULL_PEL_VAL;
   int row_max = (mv->row >> 3) + MAX_FULL_PEL_VAL;
 
-  col_min = MAX(col_min, (MV_LOW >> 3) + 1);
-  row_min = MAX(row_min, (MV_LOW >> 3) + 1);
-  col_max = MIN(col_max, (MV_UPP >> 3) - 1);
-  row_max = MIN(row_max, (MV_UPP >> 3) - 1);
+  col_min = VPXMAX(col_min, (MV_LOW >> 3) + 1);
+  row_min = VPXMAX(row_min, (MV_LOW >> 3) + 1);
+  col_max = VPXMIN(col_max, (MV_UPP >> 3) - 1);
+  row_max = VPXMIN(row_max, (MV_UPP >> 3) - 1);
 
   // Get intersection of UMV window and valid MV window to reduce # of checks
   // in diamond search.
@@ -57,12 +57,12 @@
 int vp9_init_search_range(int size) {
   int sr = 0;
   // Minimum search size no matter what the passed in value.
-  size = MAX(16, size);
+  size = VPXMAX(16, size);
 
   while ((size << sr) < MAX_FULL_PEL_VAL)
     sr++;
 
-  sr = MIN(sr, MAX_MVSEARCH_STEPS - 2);
+  sr = VPXMIN(sr, MAX_MVSEARCH_STEPS - 2);
   return sr;
 }
 
@@ -297,10 +297,10 @@
   int br = bestmv->row * 8;                                                \
   int bc = bestmv->col * 8;                                                \
   int hstep = 4;                                                           \
-  const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);           \
-  const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);           \
-  const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);           \
-  const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);           \
+  const int minc = VPXMAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);        \
+  const int maxc = VPXMIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);        \
+  const int minr = VPXMAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);        \
+  const int maxr = VPXMIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);        \
   int tr = br;                                                             \
   int tc = bc;                                                             \
                                                                            \
@@ -668,10 +668,10 @@
   int bc = bestmv->col * 8;
   int hstep = 4;
   int iter, round = 3 - forced_stop;
-  const int minc = MAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);
-  const int maxc = MIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);
-  const int minr = MAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);
-  const int maxr = MIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);
+  const int minc = VPXMAX(x->mv_col_min * 8, ref_mv->col - MV_MAX);
+  const int maxc = VPXMIN(x->mv_col_max * 8, ref_mv->col + MV_MAX);
+  const int minr = VPXMAX(x->mv_row_min * 8, ref_mv->row - MV_MAX);
+  const int maxr = VPXMIN(x->mv_row_max * 8, ref_mv->row + MV_MAX);
   int tr = br;
   int tc = bc;
   const MV *search_step = search_step_table;
@@ -1500,7 +1500,7 @@
                         int use_mvcost,
                         const MV *center_mv,
                         MV *best_mv) {
-  return vp9_hex_search(x, ref_mv, MAX(MAX_MVSEARCH_STEPS - 2, search_param),
+  return vp9_hex_search(x, ref_mv, VPXMAX(MAX_MVSEARCH_STEPS - 2, search_param),
                         sad_per_bit, do_init_search, cost_list, vfp, use_mvcost,
                         center_mv, best_mv);
 }
@@ -1515,9 +1515,9 @@
                         int use_mvcost,
                         const MV *center_mv,
                         MV *best_mv) {
-  return vp9_bigdia_search(x, ref_mv, MAX(MAX_MVSEARCH_STEPS - 2, search_param),
-                           sad_per_bit, do_init_search, cost_list, vfp,
-                           use_mvcost, center_mv, best_mv);
+  return vp9_bigdia_search(
+      x, ref_mv, VPXMAX(MAX_MVSEARCH_STEPS - 2, search_param), sad_per_bit,
+      do_init_search, cost_list, vfp, use_mvcost, center_mv, best_mv);
 }
 
 #undef CHECK_BETTER
@@ -1547,10 +1547,10 @@
   best_sad = fn_ptr->sdf(what->buf, what->stride,
                          get_buf_from_mv(in_what, ref_mv), in_what->stride) +
                  mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
-  start_row = MAX(-range, x->mv_row_min - ref_mv->row);
-  start_col = MAX(-range, x->mv_col_min - ref_mv->col);
-  end_row = MIN(range, x->mv_row_max - ref_mv->row);
-  end_col = MIN(range, x->mv_col_max - ref_mv->col);
+  start_row = VPXMAX(-range, x->mv_row_min - ref_mv->row);
+  start_col = VPXMAX(-range, x->mv_col_min - ref_mv->col);
+  end_row = VPXMIN(range, x->mv_row_max - ref_mv->row);
+  end_col = VPXMIN(range, x->mv_col_max - ref_mv->col);
 
   for (r = start_row; r <= end_row; ++r) {
     for (c = start_col; c <= end_col; c += 4) {
@@ -2021,10 +2021,10 @@
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct buf_2d *const what = &x->plane[0].src;
   const struct buf_2d *const in_what = &xd->plane[0].pre[0];
-  const int row_min = MAX(ref_mv->row - distance, x->mv_row_min);
-  const int row_max = MIN(ref_mv->row + distance, x->mv_row_max);
-  const int col_min = MAX(ref_mv->col - distance, x->mv_col_min);
-  const int col_max = MIN(ref_mv->col + distance, x->mv_col_max);
+  const int row_min = VPXMAX(ref_mv->row - distance, x->mv_row_min);
+  const int row_max = VPXMIN(ref_mv->row + distance, x->mv_row_max);
+  const int col_min = VPXMAX(ref_mv->col - distance, x->mv_col_min);
+  const int col_max = VPXMIN(ref_mv->col + distance, x->mv_col_max);
   const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
   int best_sad = fn_ptr->sdf(what->buf, what->stride,
       get_buf_from_mv(in_what, ref_mv), in_what->stride) +
@@ -2054,10 +2054,10 @@
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct buf_2d *const what = &x->plane[0].src;
   const struct buf_2d *const in_what = &xd->plane[0].pre[0];
-  const int row_min = MAX(ref_mv->row - distance, x->mv_row_min);
-  const int row_max = MIN(ref_mv->row + distance, x->mv_row_max);
-  const int col_min = MAX(ref_mv->col - distance, x->mv_col_min);
-  const int col_max = MIN(ref_mv->col + distance, x->mv_col_max);
+  const int row_min = VPXMAX(ref_mv->row - distance, x->mv_row_min);
+  const int row_max = VPXMIN(ref_mv->row + distance, x->mv_row_max);
+  const int col_min = VPXMAX(ref_mv->col - distance, x->mv_col_min);
+  const int col_max = VPXMIN(ref_mv->col + distance, x->mv_col_max);
   const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
   unsigned int best_sad = fn_ptr->sdf(what->buf, what->stride,
       get_buf_from_mv(in_what, ref_mv), in_what->stride) +
@@ -2119,10 +2119,10 @@
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct buf_2d *const what = &x->plane[0].src;
   const struct buf_2d *const in_what = &xd->plane[0].pre[0];
-  const int row_min = MAX(ref_mv->row - distance, x->mv_row_min);
-  const int row_max = MIN(ref_mv->row + distance, x->mv_row_max);
-  const int col_min = MAX(ref_mv->col - distance, x->mv_col_min);
-  const int col_max = MIN(ref_mv->col + distance, x->mv_col_max);
+  const int row_min = VPXMAX(ref_mv->row - distance, x->mv_row_min);
+  const int row_max = VPXMIN(ref_mv->row + distance, x->mv_row_max);
+  const int col_min = VPXMAX(ref_mv->col - distance, x->mv_col_min);
+  const int col_max = VPXMIN(ref_mv->col + distance, x->mv_col_max);
   const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
   unsigned int best_sad = fn_ptr->sdf(what->buf, what->stride,
       get_buf_from_mv(in_what, ref_mv), in_what->stride) +
diff --git a/vp9/encoder/vp9_picklpf.c b/vp9/encoder/vp9_picklpf.c
index 8e19103..1b1068e 100644
--- a/vp9/encoder/vp9_picklpf.c
+++ b/vp9/encoder/vp9_picklpf.c
@@ -92,8 +92,8 @@
   ss_err[filt_mid] = best_err;
 
   while (filter_step > 0) {
-    const int filt_high = MIN(filt_mid + filter_step, max_filter_level);
-    const int filt_low = MAX(filt_mid - filter_step, min_filter_level);
+    const int filt_high = VPXMIN(filt_mid + filter_step, max_filter_level);
+    const int filt_low = VPXMAX(filt_mid - filter_step, min_filter_level);
 
     // Bias against raising loop filter in favor of lowering it.
     int64_t bias = (best_err >> (15 - (filt_mid / 8))) * filter_step;
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index cc018fc..a065f59 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -293,8 +293,8 @@
 
   if (cpi->common.tx_mode == TX_MODE_SELECT) {
     if (sse > (var << 2))
-      tx_size = MIN(max_txsize_lookup[bsize],
-                    tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+      tx_size = VPXMIN(max_txsize_lookup[bsize],
+                       tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
     else
       tx_size = TX_8X8;
 
@@ -304,8 +304,8 @@
     else if (tx_size > TX_16X16)
       tx_size = TX_16X16;
   } else {
-    tx_size = MIN(max_txsize_lookup[bsize],
-                  tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+    tx_size = VPXMIN(max_txsize_lookup[bsize],
+                     tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
   }
 
   assert(tx_size >= TX_8X8);
@@ -475,8 +475,8 @@
   if (cpi->common.tx_mode == TX_MODE_SELECT) {
     if (sse > (var << 2))
       xd->mi[0]->mbmi.tx_size =
-          MIN(max_txsize_lookup[bsize],
-              tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+          VPXMIN(max_txsize_lookup[bsize],
+                 tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
     else
       xd->mi[0]->mbmi.tx_size = TX_8X8;
 
@@ -487,8 +487,8 @@
       xd->mi[0]->mbmi.tx_size = TX_16X16;
   } else {
     xd->mi[0]->mbmi.tx_size =
-        MIN(max_txsize_lookup[bsize],
-            tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+        VPXMIN(max_txsize_lookup[bsize],
+               tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
   }
 
   // Evaluate if the partition block is a skippable block in Y plane.
@@ -791,7 +791,7 @@
     const unsigned int max_thresh = 36000;
     // The encode_breakout input
     const unsigned int min_thresh =
-        MIN(((unsigned int)x->encode_breakout << 4), max_thresh);
+        VPXMIN(((unsigned int)x->encode_breakout << 4), max_thresh);
 #if CONFIG_VP9_HIGHBITDEPTH
     const int shift = (xd->bd << 1) - 16;
 #endif
@@ -911,7 +911,7 @@
 
   // TODO(jingning): This needs further refactoring.
   block_yrd(cpi, x, &rate, &dist, &is_skippable, &this_sse, 0,
-            bsize_tx, MIN(tx_size, TX_16X16));
+            bsize_tx, VPXMIN(tx_size, TX_16X16));
   x->skip_txfm[0] = is_skippable;
   rate += vp9_cost_bit(vp9_get_skip_prob(&cpi->common, xd), is_skippable);
 
@@ -961,8 +961,8 @@
   if (thr_mode_idx == best_mode_idx)
     *freq_fact -= (*freq_fact >> 4);
   else
-    *freq_fact = MIN(*freq_fact + RD_THRESH_INC,
-        cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT);
+    *freq_fact = VPXMIN(*freq_fact + RD_THRESH_INC,
+                        cpi->sf.adaptive_rd_thresh * RD_THRESH_MAX_FACT);
 }
 
 void vp9_pick_intra_mode(VP9_COMP *cpi, MACROBLOCK *x, RD_COST *rd_cost,
@@ -973,8 +973,8 @@
   PREDICTION_MODE this_mode;
   struct estimate_block_intra_args args = { cpi, x, DC_PRED, 0, 0 };
   const TX_SIZE intra_tx_size =
-      MIN(max_txsize_lookup[bsize],
-          tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+      VPXMIN(max_txsize_lookup[bsize],
+             tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
   MODE_INFO *const mic = xd->mi[0];
   int *bmode_costs;
   const MODE_INFO *above_mi = xd->mi[-xd->mi_stride];
@@ -1160,8 +1160,8 @@
   mbmi->sb_type = bsize;
   mbmi->ref_frame[0] = NONE;
   mbmi->ref_frame[1] = NONE;
-  mbmi->tx_size = MIN(max_txsize_lookup[bsize],
-                      tx_mode_to_biggest_tx_size[cm->tx_mode]);
+  mbmi->tx_size = VPXMIN(max_txsize_lookup[bsize],
+                         tx_mode_to_biggest_tx_size[cm->tx_mode]);
 
 #if CONFIG_VP9_TEMPORAL_DENOISING
   vp9_denoiser_reset_frame_stats(ctx);
@@ -1414,7 +1414,7 @@
     if (!this_early_term) {
       this_sse = (int64_t)sse_y;
       block_yrd(cpi, x, &this_rdc.rate, &this_rdc.dist, &is_skippable,
-                &this_sse, 0, bsize, MIN(mbmi->tx_size, TX_16X16));
+                &this_sse, 0, bsize, VPXMIN(mbmi->tx_size, TX_16X16));
       x->skip_txfm[0] = is_skippable;
       if (is_skippable) {
         this_rdc.rate = vp9_cost_bit(vp9_get_skip_prob(cm, xd), 1);
@@ -1523,8 +1523,8 @@
        bsize <= cpi->sf.max_intra_bsize)) {
     struct estimate_block_intra_args args = { cpi, x, DC_PRED, 0, 0 };
     const TX_SIZE intra_tx_size =
-        MIN(max_txsize_lookup[bsize],
-            tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
+        VPXMIN(max_txsize_lookup[bsize],
+               tx_mode_to_biggest_tx_size[cpi->common.tx_mode]);
     int i;
     TX_SIZE best_intra_tx_size = TX_SIZES;
 
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index 4ba3406..fdc55f7 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -106,8 +106,8 @@
 static int get_minq_index(double maxq, double x3, double x2, double x1,
                           vpx_bit_depth_t bit_depth) {
   int i;
-  const double minqtarget = MIN(((x3 * maxq + x2) * maxq + x1) * maxq,
-                                maxq);
+  const double minqtarget = VPXMIN(((x3 * maxq + x2) * maxq + x1) * maxq,
+                                   maxq);
 
   // Special case handling to deal with the step from q2.0
   // down to lossless mode represented by q 1.0.
@@ -192,15 +192,15 @@
                            vpx_bit_depth_t bit_depth) {
   const int bpm = (int)(vp9_rc_bits_per_mb(frame_type, q, correction_factor,
                                            bit_depth));
-  return MAX(FRAME_OVERHEAD_BITS,
-             (int)((uint64_t)bpm * mbs) >> BPER_MB_NORMBITS);
+  return VPXMAX(FRAME_OVERHEAD_BITS,
+                (int)((uint64_t)bpm * mbs) >> BPER_MB_NORMBITS);
 }
 
 int vp9_rc_clamp_pframe_target_size(const VP9_COMP *const cpi, int target) {
   const RATE_CONTROL *rc = &cpi->rc;
   const VP9EncoderConfig *oxcf = &cpi->oxcf;
-  const int min_frame_target = MAX(rc->min_frame_bandwidth,
-                                   rc->avg_frame_bandwidth >> 5);
+  const int min_frame_target = VPXMAX(rc->min_frame_bandwidth,
+                                      rc->avg_frame_bandwidth >> 5);
   if (target < min_frame_target)
     target = min_frame_target;
   if (cpi->refresh_golden_frame && rc->is_src_frame_alt_ref) {
@@ -216,7 +216,7 @@
   if (oxcf->rc_max_inter_bitrate_pct) {
     const int max_rate = rc->avg_frame_bandwidth *
                          oxcf->rc_max_inter_bitrate_pct / 100;
-    target = MIN(target, max_rate);
+    target = VPXMIN(target, max_rate);
   }
   return target;
 }
@@ -227,7 +227,7 @@
   if (oxcf->rc_max_intra_bitrate_pct) {
     const int max_rate = rc->avg_frame_bandwidth *
                              oxcf->rc_max_intra_bitrate_pct / 100;
-    target = MIN(target, max_rate);
+    target = VPXMIN(target, max_rate);
   }
   if (target > rc->max_frame_bandwidth)
     target = rc->max_frame_bandwidth;
@@ -250,7 +250,8 @@
     lrc->bits_off_target += bits_off_for_this_layer;
 
     // Clip buffer level to maximum buffer size for the layer.
-    lrc->bits_off_target = MIN(lrc->bits_off_target, lrc->maximum_buffer_size);
+    lrc->bits_off_target =
+        VPXMIN(lrc->bits_off_target, lrc->maximum_buffer_size);
     lrc->buffer_level = lrc->bits_off_target;
   }
 }
@@ -268,7 +269,7 @@
   }
 
   // Clip the buffer level to the maximum specified buffer size.
-  rc->bits_off_target = MIN(rc->bits_off_target, rc->maximum_buffer_size);
+  rc->bits_off_target = VPXMIN(rc->bits_off_target, rc->maximum_buffer_size);
   rc->buffer_level = rc->bits_off_target;
 
   if (is_one_pass_cbr_svc(cpi)) {
@@ -287,8 +288,8 @@
   if (factor <= factor_safe)
     return default_interval;
   else
-    return MAX(default_interval,
-               (int)(MIN_GF_INTERVAL * factor / factor_safe + 0.5));
+    return VPXMAX(default_interval,
+                  (int)(MIN_GF_INTERVAL * factor / factor_safe + 0.5));
   // Note this logic makes:
   // 4K24: 5
   // 4K30: 6
@@ -296,9 +297,9 @@
 }
 
 int vp9_rc_get_default_max_gf_interval(double framerate, int min_gf_interval) {
-  int interval = MIN(MAX_GF_INTERVAL, (int)(framerate * 0.75));
+  int interval = VPXMIN(MAX_GF_INTERVAL, (int)(framerate * 0.75));
   interval += (interval & 0x01);  // Round to even value
-  return MAX(interval, min_gf_interval);
+  return VPXMAX(interval, min_gf_interval);
 }
 
 void vp9_rc_init(const VP9EncoderConfig *oxcf, int pass, RATE_CONTROL *rc) {
@@ -478,7 +479,7 @@
   // More heavily damped adjustment used if we have been oscillating either side
   // of target.
   adjustment_limit = 0.25 +
-      0.5 * MIN(1, fabs(log10(0.01 * correction_factor)));
+      0.5 * VPXMIN(1, fabs(log10(0.01 * correction_factor)));
 
   cpi->rc.q_2_frame = cpi->rc.q_1_frame;
   cpi->rc.q_1_frame = cm->base_qindex;
@@ -558,8 +559,8 @@
   if (cpi->oxcf.rc_mode == VPX_CBR &&
       (cpi->rc.rc_1_frame * cpi->rc.rc_2_frame == -1) &&
       cpi->rc.q_1_frame != cpi->rc.q_2_frame) {
-    q = clamp(q, MIN(cpi->rc.q_1_frame, cpi->rc.q_2_frame),
-              MAX(cpi->rc.q_1_frame, cpi->rc.q_2_frame));
+    q = clamp(q, VPXMIN(cpi->rc.q_1_frame, cpi->rc.q_2_frame),
+              VPXMAX(cpi->rc.q_1_frame, cpi->rc.q_2_frame));
   }
   return q;
 }
@@ -617,7 +618,7 @@
                                              : rc->last_q[INTER_FRAME] * 2;
     }
   }
-  return MIN(active_worst_quality, rc->worst_quality);
+  return VPXMIN(active_worst_quality, rc->worst_quality);
 }
 
 // Adjust active_worst_quality level based on buffer level.
@@ -644,10 +645,10 @@
   // So for first few frames following key, the qp of that key frame is weighted
   // into the active_worst_quality setting.
   ambient_qp = (cm->current_video_frame < num_frames_weight_key) ?
-      MIN(rc->avg_frame_qindex[INTER_FRAME], rc->avg_frame_qindex[KEY_FRAME]) :
-      rc->avg_frame_qindex[INTER_FRAME];
-  active_worst_quality = MIN(rc->worst_quality,
-                             ambient_qp * 5 / 4);
+                   VPXMIN(rc->avg_frame_qindex[INTER_FRAME],
+                          rc->avg_frame_qindex[KEY_FRAME]) :
+                   rc->avg_frame_qindex[INTER_FRAME];
+  active_worst_quality = VPXMIN(rc->worst_quality, ambient_qp * 5 / 4);
   if (rc->buffer_level > rc->optimal_buffer_level) {
     // Adjust down.
     // Maximum limit for down adjustment, ~30%.
@@ -700,7 +701,7 @@
       int delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
                                             (last_boosted_q * 0.75),
                                             cm->bit_depth);
-      active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
+      active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
     } else if (cm->current_video_frame > 0) {
       // not first frame of one pass and kf_boost is set
       double q_adj_factor = 1.0;
@@ -833,7 +834,7 @@
       int delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
                                             last_boosted_q * 0.75,
                                             cm->bit_depth);
-      active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
+      active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
     } else {
       // not first frame of one pass and kf_boost is set
       double q_adj_factor = 1.0;
@@ -1002,21 +1003,21 @@
       int qindex;
 
       if (cpi->twopass.last_kfgroup_zeromotion_pct >= STATIC_MOTION_THRESH) {
-        qindex = MIN(rc->last_kf_qindex, rc->last_boosted_qindex);
+        qindex = VPXMIN(rc->last_kf_qindex, rc->last_boosted_qindex);
         active_best_quality = qindex;
         last_boosted_q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
         delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
                                               last_boosted_q * 1.25,
                                               cm->bit_depth);
-        active_worst_quality = MIN(qindex + delta_qindex, active_worst_quality);
-
+        active_worst_quality =
+            VPXMIN(qindex + delta_qindex, active_worst_quality);
       } else {
         qindex = rc->last_boosted_qindex;
         last_boosted_q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
         delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
                                               last_boosted_q * 0.75,
                                               cm->bit_depth);
-        active_best_quality = MAX(qindex + delta_qindex, rc->best_quality);
+        active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
       }
     } else {
       // Not forced keyframe.
@@ -1116,8 +1117,8 @@
       (cpi->twopass.last_kfgroup_zeromotion_pct < STATIC_MOTION_THRESH)) {
     int qdelta = vp9_frame_type_qdelta(cpi, gf_group->rf_level[gf_group->index],
                                        active_worst_quality);
-    active_worst_quality = MAX(active_worst_quality + qdelta,
-                               active_best_quality);
+    active_worst_quality = VPXMAX(active_worst_quality + qdelta,
+                                  active_best_quality);
   }
 #endif
 
@@ -1126,7 +1127,8 @@
     int qdelta = vp9_compute_qdelta_by_rate(rc, cm->frame_type,
                                             active_best_quality, 2.0,
                                             cm->bit_depth);
-    active_best_quality = MAX(active_best_quality + qdelta, rc->best_quality);
+    active_best_quality =
+        VPXMAX(active_best_quality + qdelta, rc->best_quality);
   }
 
   active_best_quality = clamp(active_best_quality,
@@ -1141,7 +1143,7 @@
              rc->this_key_frame_forced) {
     // If static since last kf use better of last boosted and last kf q.
     if (cpi->twopass.last_kfgroup_zeromotion_pct >= STATIC_MOTION_THRESH) {
-      q = MIN(rc->last_kf_qindex, rc->last_boosted_qindex);
+      q = VPXMIN(rc->last_kf_qindex, rc->last_boosted_qindex);
     } else {
       q = rc->last_boosted_qindex;
     }
@@ -1203,9 +1205,9 @@
     // For very small rate targets where the fractional adjustment
     // may be tiny make sure there is at least a minimum range.
     const int tolerance = (cpi->sf.recode_tolerance * frame_target) / 100;
-    *frame_under_shoot_limit = MAX(frame_target - tolerance - 200, 0);
-    *frame_over_shoot_limit = MIN(frame_target + tolerance + 200,
-                                  cpi->rc.max_frame_bandwidth);
+    *frame_under_shoot_limit = VPXMAX(frame_target - tolerance - 200, 0);
+    *frame_over_shoot_limit = VPXMIN(frame_target + tolerance + 200,
+                                     cpi->rc.max_frame_bandwidth);
   }
 }
 
@@ -1458,7 +1460,8 @@
   const SVC *const svc = &cpi->svc;
   const int64_t diff = rc->optimal_buffer_level - rc->buffer_level;
   const int64_t one_pct_bits = 1 + rc->optimal_buffer_level / 100;
-  int min_frame_target = MAX(rc->avg_frame_bandwidth >> 4, FRAME_OVERHEAD_BITS);
+  int min_frame_target =
+      VPXMAX(rc->avg_frame_bandwidth >> 4, FRAME_OVERHEAD_BITS);
   int target;
 
   if (oxcf->gf_cbr_boost_pct) {
@@ -1480,23 +1483,24 @@
             svc->temporal_layer_id, svc->number_temporal_layers);
     const LAYER_CONTEXT *lc = &svc->layer_context[layer];
     target = lc->avg_frame_size;
-    min_frame_target = MAX(lc->avg_frame_size >> 4, FRAME_OVERHEAD_BITS);
+    min_frame_target = VPXMAX(lc->avg_frame_size >> 4, FRAME_OVERHEAD_BITS);
   }
   if (diff > 0) {
     // Lower the target bandwidth for this frame.
-    const int pct_low = (int)MIN(diff / one_pct_bits, oxcf->under_shoot_pct);
+    const int pct_low = (int)VPXMIN(diff / one_pct_bits, oxcf->under_shoot_pct);
     target -= (target * pct_low) / 200;
   } else if (diff < 0) {
     // Increase the target bandwidth for this frame.
-    const int pct_high = (int)MIN(-diff / one_pct_bits, oxcf->over_shoot_pct);
+    const int pct_high =
+        (int)VPXMIN(-diff / one_pct_bits, oxcf->over_shoot_pct);
     target += (target * pct_high) / 200;
   }
   if (oxcf->rc_max_inter_bitrate_pct) {
     const int max_rate = rc->avg_frame_bandwidth *
                          oxcf->rc_max_inter_bitrate_pct / 100;
-    target = MIN(target, max_rate);
+    target = VPXMIN(target, max_rate);
   }
-  return MAX(min_frame_target, target);
+  return VPXMAX(min_frame_target, target);
 }
 
 static int calc_iframe_target_size_one_pass_cbr(const VP9_COMP *cpi) {
@@ -1518,7 +1522,7 @@
       const LAYER_CONTEXT *lc = &svc->layer_context[layer];
       framerate = lc->framerate;
     }
-    kf_boost = MAX(kf_boost, (int)(2 * framerate - 16));
+    kf_boost = VPXMAX(kf_boost, (int)(2 * framerate - 16));
     if (rc->frames_since_key <  framerate / 2) {
       kf_boost = (int)(kf_boost * rc->frames_since_key /
                        (framerate / 2));
@@ -1726,7 +1730,7 @@
     rc->max_gf_interval = rc->static_scene_max_gf_interval;
 
   // Clamp min to max
-  rc->min_gf_interval = MIN(rc->min_gf_interval, rc->max_gf_interval);
+  rc->min_gf_interval = VPXMIN(rc->min_gf_interval, rc->max_gf_interval);
 }
 
 void vp9_rc_update_framerate(VP9_COMP *cpi) {
@@ -1739,7 +1743,8 @@
   rc->min_frame_bandwidth = (int)(rc->avg_frame_bandwidth *
                                 oxcf->two_pass_vbrmin_section / 100);
 
-  rc->min_frame_bandwidth = MAX(rc->min_frame_bandwidth, FRAME_OVERHEAD_BITS);
+  rc->min_frame_bandwidth =
+      VPXMAX(rc->min_frame_bandwidth, FRAME_OVERHEAD_BITS);
 
   // A maximum bitrate for a frame is defined.
   // The baseline for this aligns with HW implementations that
@@ -1750,8 +1755,8 @@
   // specifies lossless encode.
   vbr_max_bits = (int)(((int64_t)rc->avg_frame_bandwidth *
                      oxcf->two_pass_vbrmax_section) / 100);
-  rc->max_frame_bandwidth = MAX(MAX((cm->MBs * MAX_MB_RATE), MAXRATE_1080P),
-                                    vbr_max_bits);
+  rc->max_frame_bandwidth =
+      VPXMAX(VPXMAX((cm->MBs * MAX_MB_RATE), MAXRATE_1080P), vbr_max_bits);
 
   vp9_rc_set_gf_interval_range(cpi, rc);
 }
@@ -1789,12 +1794,12 @@
   // Dont do it for kf,arf,gf or overlay frames.
   if (!frame_is_kf_gf_arf(cpi) && !rc->is_src_frame_alt_ref &&
       rc->vbr_bits_off_target_fast) {
-    int one_frame_bits = MAX(rc->avg_frame_bandwidth, *this_frame_target);
+    int one_frame_bits = VPXMAX(rc->avg_frame_bandwidth, *this_frame_target);
     int fast_extra_bits;
-    fast_extra_bits =
-      (int)MIN(rc->vbr_bits_off_target_fast, one_frame_bits);
-    fast_extra_bits = (int)MIN(fast_extra_bits,
-      MAX(one_frame_bits / 8, rc->vbr_bits_off_target_fast / 8));
+    fast_extra_bits = (int)VPXMIN(rc->vbr_bits_off_target_fast, one_frame_bits);
+    fast_extra_bits = (int)VPXMIN(
+        fast_extra_bits,
+        VPXMAX(one_frame_bits / 8, rc->vbr_bits_off_target_fast / 8));
     *this_frame_target += (int)fast_extra_bits;
     rc->vbr_bits_off_target_fast -= fast_extra_bits;
   }
@@ -1948,7 +1953,7 @@
     // between current and the previous frame value(s). Use a minimum threshold
     // for cases where there is small change from content that is completely
     // static.
-    if (avg_sad > MAX(4000, (rc->avg_source_sad << 3)) &&
+    if (avg_sad > VPXMAX(4000, (rc->avg_source_sad << 3)) &&
         rc->frames_since_key > 1)
       rc->high_source_sad = 1;
     else
diff --git a/vp9/encoder/vp9_rd.c b/vp9/encoder/vp9_rd.c
index 2f2f7c1..fdee153 100644
--- a/vp9/encoder/vp9_rd.c
+++ b/vp9/encoder/vp9_rd.c
@@ -172,7 +172,7 @@
   if (cpi->oxcf.pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
     const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
     const FRAME_UPDATE_TYPE frame_type = gf_group->update_type[gf_group->index];
-    const int boost_index = MIN(15, (cpi->rc.gfu_boost / 100));
+    const int boost_index = VPXMIN(15, (cpi->rc.gfu_boost / 100));
 
     rdmult = (rdmult * rd_frame_type_factor[frame_type]) >> 7;
     rdmult += ((rdmult * rd_boost_factor[boost_index]) >> 7);
@@ -204,7 +204,7 @@
   q = vp9_dc_quant(qindex, 0, VPX_BITS_8) / 4.0;
 #endif  // CONFIG_VP9_HIGHBITDEPTH
   // TODO(debargha): Adjust the function below.
-  return MAX((int)(pow(q, RD_THRESH_POW) * 5.12), 8);
+  return VPXMAX((int)(pow(q, RD_THRESH_POW) * 5.12), 8);
 }
 
 void vp9_initialize_me_consts(VP9_COMP *cpi, MACROBLOCK *x, int qindex) {
@@ -404,7 +404,7 @@
     static const uint32_t MAX_XSQ_Q10 = 245727;
     const uint64_t xsq_q10_64 =
         (((uint64_t)qstep * qstep << (n_log2 + 10)) + (var >> 1)) / var;
-    const int xsq_q10 = (int)MIN(xsq_q10_64, MAX_XSQ_Q10);
+    const int xsq_q10 = (int)VPXMIN(xsq_q10_64, MAX_XSQ_Q10);
     model_rd_norm(xsq_q10, &r_q10, &d_q10);
     *rate = ((r_q10 << n_log2) + 2) >> 2;
     *dist = (var * (int64_t)d_q10 + 512) >> 10;
@@ -485,7 +485,7 @@
       continue;
     fp_row = (this_mv->row + 3 + (this_mv->row >= 0)) >> 3;
     fp_col = (this_mv->col + 3 + (this_mv->col >= 0)) >> 3;
-    max_mv = MAX(max_mv, MAX(abs(this_mv->row), abs(this_mv->col)) >> 3);
+    max_mv = VPXMAX(max_mv, VPXMAX(abs(this_mv->row), abs(this_mv->col)) >> 3);
 
     if (fp_row ==0 && fp_col == 0 && zero_seen)
       continue;
@@ -629,16 +629,15 @@
     const int top_mode = bsize < BLOCK_8X8 ? MAX_REFS : MAX_MODES;
     int mode;
     for (mode = 0; mode < top_mode; ++mode) {
-      const BLOCK_SIZE min_size = MAX(bsize - 1, BLOCK_4X4);
-      const BLOCK_SIZE max_size = MIN(bsize + 2, BLOCK_64X64);
+      const BLOCK_SIZE min_size = VPXMAX(bsize - 1, BLOCK_4X4);
+      const BLOCK_SIZE max_size = VPXMIN(bsize + 2, BLOCK_64X64);
       BLOCK_SIZE bs;
       for (bs = min_size; bs <= max_size; ++bs) {
         int *const fact = &factor_buf[bs][mode];
         if (mode == best_mode_index) {
           *fact -= (*fact >> 4);
         } else {
-          *fact = MIN(*fact + RD_THRESH_INC,
-                      rd_thresh * RD_THRESH_MAX_FACT);
+          *fact = VPXMIN(*fact + RD_THRESH_INC, rd_thresh * RD_THRESH_MAX_FACT);
         }
       }
     }
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 96c6474..038d1e1 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -192,8 +192,8 @@
     const int64_t ac_thr = p->quant_thred[1] >> shift;
     // The low thresholds are used to measure if the prediction errors are
     // low enough so that we can skip the mode search.
-    const int64_t low_dc_thr = MIN(50, dc_thr >> 2);
-    const int64_t low_ac_thr = MIN(80, ac_thr >> 2);
+    const int64_t low_dc_thr = VPXMIN(50, dc_thr >> 2);
+    const int64_t low_ac_thr = VPXMIN(80, ac_thr >> 2);
     int bw = 1 << (b_width_log2_lookup[bs] - b_width_log2_lookup[unit_size]);
     int bh = 1 << (b_height_log2_lookup[bs] - b_width_log2_lookup[unit_size]);
     int idx, idy;
@@ -505,7 +505,7 @@
         if (tx_size != TX_32X32)
           dc_correct >>= 2;
 
-        dist = MAX(0, sse - dc_correct);
+        dist = VPXMAX(0, sse - dc_correct);
       }
     } else {
       // SKIP_TXFM_AC_DC
@@ -531,7 +531,7 @@
   rd2 = RDCOST(x->rdmult, x->rddiv, 0, sse);
 
   // TODO(jingning): temporarily enabled only for luma component
-  rd = MIN(rd1, rd2);
+  rd = VPXMIN(rd1, rd2);
   if (plane == 0)
     x->zcoeff_blk[tx_size][block] = !x->plane[plane].eobs[block] ||
                                     (rd1 > rd2 && !xd->lossless);
@@ -597,7 +597,7 @@
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
 
-  mbmi->tx_size = MIN(max_tx_size, largest_tx_size);
+  mbmi->tx_size = VPXMIN(max_tx_size, largest_tx_size);
 
   txfm_rd_in_plane(x, rate, distortion, skip,
                    sse, ref_best_rd, 0, bs,
@@ -637,8 +637,8 @@
     start_tx = max_tx_size;
     end_tx = 0;
   } else {
-    TX_SIZE chosen_tx_size = MIN(max_tx_size,
-                                 tx_mode_to_biggest_tx_size[cm->tx_mode]);
+    TX_SIZE chosen_tx_size = VPXMIN(max_tx_size,
+                                    tx_mode_to_biggest_tx_size[cm->tx_mode]);
     start_tx = chosen_tx_size;
     end_tx = chosen_tx_size;
   }
@@ -1389,7 +1389,7 @@
                               cpi->sf.use_fast_coef_costing);
       rd1 = RDCOST(x->rdmult, x->rddiv, thisrate, thisdistortion >> 2);
       rd2 = RDCOST(x->rdmult, x->rddiv, 0, thissse >> 2);
-      rd = MIN(rd1, rd2);
+      rd = VPXMIN(rd1, rd2);
       if (rd >= best_yrd)
         return INT64_MAX;
     }
@@ -1808,7 +1808,8 @@
           if (i == 0)
             max_mv = x->max_mv_context[mbmi->ref_frame[0]];
           else
-            max_mv = MAX(abs(bsi->mvp.as_mv.row), abs(bsi->mvp.as_mv.col)) >> 3;
+            max_mv =
+                VPXMAX(abs(bsi->mvp.as_mv.row), abs(bsi->mvp.as_mv.col)) >> 3;
 
           if (cpi->sf.mv.auto_mv_step_size && cm->show_frame) {
             // Take wtd average of the step_params based on the last frame's
@@ -1826,7 +1827,7 @@
           if (cpi->sf.adaptive_motion_search) {
             mvp_full.row = x->pred_mv[mbmi->ref_frame[0]].row >> 3;
             mvp_full.col = x->pred_mv[mbmi->ref_frame[0]].col >> 3;
-            step_param = MAX(step_param, 8);
+            step_param = VPXMAX(step_param, 8);
           }
 
           // adjust src pointer for this block
@@ -2231,7 +2232,7 @@
   vp9_set_mv_search_range(x, &ref_mv);
 
   // Work out the size of the first step in the mv step search.
-  // 0 here is maximum length first step. 1 is MAX >> 1 etc.
+  // 0 here is maximum length first step. 1 is VPXMAX >> 1 etc.
   if (cpi->sf.mv.auto_mv_step_size && cm->show_frame) {
     // Take wtd average of the step_params based on the last frame's
     // max mv magnitude and that based on the best ref mvs of the current
@@ -2243,9 +2244,10 @@
   }
 
   if (cpi->sf.adaptive_motion_search && bsize < BLOCK_64X64) {
-    int boffset = 2 * (b_width_log2_lookup[BLOCK_64X64] -
-          MIN(b_height_log2_lookup[bsize], b_width_log2_lookup[bsize]));
-    step_param = MAX(step_param, boffset);
+    int boffset =
+        2 * (b_width_log2_lookup[BLOCK_64X64] -
+             VPXMIN(b_height_log2_lookup[bsize], b_width_log2_lookup[bsize]));
+    step_param = VPXMAX(step_param, boffset);
   }
 
   if (cpi->sf.adaptive_motion_search) {
@@ -2466,7 +2468,7 @@
       // motion field, where the distortion gain for a single block may not
       // be enough to overcome the cost of a new mv.
       if (discount_newmv_test(cpi, this_mode, tmp_mv, mode_mv, refs[0])) {
-        *rate2 += MAX((rate_mv / NEW_MV_DISCOUNT_FACTOR), 1);
+        *rate2 += VPXMAX((rate_mv / NEW_MV_DISCOUNT_FACTOR), 1);
       } else {
         *rate2 += rate_mv;
       }
@@ -2502,10 +2504,10 @@
   // initiation of a motion field.
   if (discount_newmv_test(cpi, this_mode, frame_mv[refs[0]],
                           mode_mv, refs[0])) {
-    *rate2 += MIN(cost_mv_ref(cpi, this_mode,
-                              mbmi_ext->mode_context[refs[0]]),
-                  cost_mv_ref(cpi, NEARESTMV,
-                              mbmi_ext->mode_context[refs[0]]));
+    *rate2 += VPXMIN(cost_mv_ref(cpi, this_mode,
+                                 mbmi_ext->mode_context[refs[0]]),
+                     cost_mv_ref(cpi, NEARESTMV,
+                                 mbmi_ext->mode_context[refs[0]]));
   } else {
     *rate2 += cost_mv_ref(cpi, this_mode, mbmi_ext->mode_context[refs[0]]);
   }
@@ -2547,10 +2549,10 @@
           rd = RDCOST(x->rdmult, x->rddiv, tmp_rate_sum, tmp_dist_sum);
           filter_cache[i] = rd;
           filter_cache[SWITCHABLE_FILTERS] =
-              MIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
+              VPXMIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
           if (cm->interp_filter == SWITCHABLE)
             rd += rs_rd;
-          *mask_filter = MAX(*mask_filter, rd);
+          *mask_filter = VPXMAX(*mask_filter, rd);
         } else {
           int rate_sum = 0;
           int64_t dist_sum = 0;
@@ -2580,10 +2582,10 @@
           rd = RDCOST(x->rdmult, x->rddiv, rate_sum, dist_sum);
           filter_cache[i] = rd;
           filter_cache[SWITCHABLE_FILTERS] =
-              MIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
+              VPXMIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
           if (cm->interp_filter == SWITCHABLE)
             rd += rs_rd;
-          *mask_filter = MAX(*mask_filter, rd);
+          *mask_filter = VPXMAX(*mask_filter, rd);
 
           if (i == 0 && intpel_mv) {
             tmp_rate_sum = rate_sum;
@@ -2694,7 +2696,7 @@
     *distortion += distortion_y;
 
     rdcosty = RDCOST(x->rdmult, x->rddiv, *rate2, *distortion);
-    rdcosty = MIN(rdcosty, RDCOST(x->rdmult, x->rddiv, 0, *psse));
+    rdcosty = VPXMIN(rdcosty, RDCOST(x->rdmult, x->rddiv, 0, *psse));
 
     if (!super_block_uvrd(cpi, x, rate_uv, &distortion_uv, &skippable_uv,
                           &sseuv, bsize, ref_best_rd - rdcosty)) {
@@ -2759,7 +2761,7 @@
                                        pd[1].subsampling_x,
                                        pd[1].subsampling_y);
   rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv, &rate_uv_tokenonly,
-                          &dist_uv, &uv_skip, MAX(BLOCK_8X8, bsize),
+                          &dist_uv, &uv_skip, VPXMAX(BLOCK_8X8, bsize),
                           max_uv_tx_size);
 
   if (y_skip && uv_skip) {
@@ -2826,12 +2828,12 @@
   // to a predictor with a low spatial complexity compared to the source.
   if ((source_variance > LOW_VAR_THRESH) && (ref_frame == INTRA_FRAME) &&
       (source_variance > recon_variance)) {
-    var_factor = MIN(absvar_diff, MIN(VLOW_ADJ_MAX, var_error));
+    var_factor = VPXMIN(absvar_diff, VPXMIN(VLOW_ADJ_MAX, var_error));
   // A second possible case of interest is where the source variance
   // is very low and we wish to discourage false texture or motion trails.
   } else if ((source_variance < (LOW_VAR_THRESH >> 1)) &&
              (recon_variance > source_variance)) {
-    var_factor = MIN(absvar_diff, MIN(VHIGH_ADJ_MAX, var_error));
+    var_factor = VPXMIN(absvar_diff, VPXMIN(VHIGH_ADJ_MAX, var_error));
   }
   *this_rd += (*this_rd * var_factor) / 100;
 }
@@ -2861,7 +2863,7 @@
     top_edge += (int)(twopass->this_frame_stats.inactive_zone_rows * 2);
 
     bottom_edge -= (int)(twopass->this_frame_stats.inactive_zone_rows * 2);
-    bottom_edge = MAX(top_edge, bottom_edge);
+    bottom_edge = VPXMAX(top_edge, bottom_edge);
   }
 
   if (((top_edge >= mi_row) && (top_edge < (mi_row + mi_step))) ||
@@ -2888,7 +2890,7 @@
     left_edge += (int)(twopass->this_frame_stats.inactive_zone_cols * 2);
 
     right_edge -= (int)(twopass->this_frame_stats.inactive_zone_cols * 2);
-    right_edge = MAX(left_edge, right_edge);
+    right_edge = VPXMAX(left_edge, right_edge);
   }
 
   if (((left_edge >= mi_col) && (left_edge < (mi_col + mi_step))) ||
@@ -3135,7 +3137,7 @@
     }
 
     if ((ref_frame_skip_mask[0] & (1 << ref_frame)) &&
-        (ref_frame_skip_mask[1] & (1 << MAX(0, second_ref_frame))))
+        (ref_frame_skip_mask[1] & (1 << VPXMAX(0, second_ref_frame))))
       continue;
 
     if (mode_skip_mask[ref_frame] & (1 << this_mode))
@@ -3149,10 +3151,10 @@
       continue;
 
     if (sf->motion_field_mode_search) {
-      const int mi_width  = MIN(num_8x8_blocks_wide_lookup[bsize],
-                                tile_info->mi_col_end - mi_col);
-      const int mi_height = MIN(num_8x8_blocks_high_lookup[bsize],
-                                tile_info->mi_row_end - mi_row);
+      const int mi_width  = VPXMIN(num_8x8_blocks_wide_lookup[bsize],
+                                   tile_info->mi_col_end - mi_col);
+      const int mi_height = VPXMIN(num_8x8_blocks_high_lookup[bsize],
+                                   tile_info->mi_row_end - mi_row);
       const int bsl = mi_width_log2_lookup[bsize];
       int cb_partition_search_ctrl = (((mi_row + mi_col) >> bsl)
           + get_chessboard_index(cm->current_video_frame)) & 0x1;
@@ -3370,9 +3372,9 @@
 
     if (!disable_skip && ref_frame == INTRA_FRAME) {
       for (i = 0; i < REFERENCE_MODES; ++i)
-        best_pred_rd[i] = MIN(best_pred_rd[i], this_rd);
+        best_pred_rd[i] = VPXMIN(best_pred_rd[i], this_rd);
       for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++)
-        best_filter_rd[i] = MIN(best_filter_rd[i], this_rd);
+        best_filter_rd[i] = VPXMIN(best_filter_rd[i], this_rd);
     }
 
     // Did this mode help.. i.e. is it the new best mode
@@ -3471,7 +3473,7 @@
             adj_rd = filter_cache[i] - ref;
 
           adj_rd += this_rd;
-          best_filter_rd[i] = MIN(best_filter_rd[i], adj_rd);
+          best_filter_rd[i] = VPXMIN(best_filter_rd[i], adj_rd);
         }
       }
     }
@@ -3814,7 +3816,7 @@
     }
 
     if ((ref_frame_skip_mask[0] & (1 << ref_frame)) &&
-        (ref_frame_skip_mask[1] & (1 << MAX(0, second_ref_frame))))
+        (ref_frame_skip_mask[1] & (1 << VPXMAX(0, second_ref_frame))))
       continue;
 
     // Test best rd so far against threshold for trying this mode.
@@ -3969,12 +3971,11 @@
             rs_rd = RDCOST(x->rdmult, x->rddiv, rs, 0);
             filter_cache[switchable_filter_index] = tmp_rd;
             filter_cache[SWITCHABLE_FILTERS] =
-                MIN(filter_cache[SWITCHABLE_FILTERS],
-                    tmp_rd + rs_rd);
+                VPXMIN(filter_cache[SWITCHABLE_FILTERS], tmp_rd + rs_rd);
             if (cm->interp_filter == SWITCHABLE)
               tmp_rd += rs_rd;
 
-            mask_filter = MAX(mask_filter, tmp_rd);
+            mask_filter = VPXMAX(mask_filter, tmp_rd);
 
             newbest = (tmp_rd < tmp_best_rd);
             if (newbest) {
@@ -4051,9 +4052,9 @@
 
       compmode_cost = vp9_cost_bit(comp_mode_p, comp_pred);
 
-      tmp_best_rdu = best_rd -
-          MIN(RDCOST(x->rdmult, x->rddiv, rate2, distortion2),
-              RDCOST(x->rdmult, x->rddiv, 0, total_sse));
+      tmp_best_rdu =
+          best_rd - VPXMIN(RDCOST(x->rdmult, x->rddiv, rate2, distortion2),
+                           RDCOST(x->rdmult, x->rddiv, 0, total_sse));
 
       if (tmp_best_rdu > 0) {
         // If even the 'Y' rd value of split is higher than best so far
@@ -4113,9 +4114,9 @@
 
     if (!disable_skip && ref_frame == INTRA_FRAME) {
       for (i = 0; i < REFERENCE_MODES; ++i)
-        best_pred_rd[i] = MIN(best_pred_rd[i], this_rd);
+        best_pred_rd[i] = VPXMIN(best_pred_rd[i], this_rd);
       for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++)
-        best_filter_rd[i] = MIN(best_filter_rd[i], this_rd);
+        best_filter_rd[i] = VPXMIN(best_filter_rd[i], this_rd);
     }
 
     // Did this mode help.. i.e. is it the new best mode
@@ -4214,7 +4215,7 @@
           adj_rd = filter_cache[i] - ref;
 
         adj_rd += this_rd;
-        best_filter_rd[i] = MIN(best_filter_rd[i], adj_rd);
+        best_filter_rd[i] = VPXMIN(best_filter_rd[i], adj_rd);
       }
     }
 
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index 5e72c4c..8530f98 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -49,7 +49,7 @@
   VP9_COMMON *const cm = &cpi->common;
 
   if (speed >= 1) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
       sf->partition_search_breakout_dist_thr = (1 << 23);
@@ -60,7 +60,7 @@
   }
 
   if (speed >= 2) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
       sf->adaptive_pred_interp_filter = 0;
@@ -75,7 +75,7 @@
   }
 
   if (speed >= 3) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = DISABLE_ALL_SPLIT;
       sf->schedule_mode_search = cm->base_qindex < 220 ? 1 : 0;
       sf->partition_search_breakout_dist_thr = (1 << 25);
@@ -99,7 +99,7 @@
   }
 
   if (speed >= 4) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->partition_search_breakout_dist_thr = (1 << 26);
     } else {
       sf->partition_search_breakout_dist_thr = (1 << 24);
@@ -215,7 +215,7 @@
   VP9_COMMON *const cm = &cpi->common;
 
   if (speed >= 1) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
     } else {
@@ -224,7 +224,7 @@
   }
 
   if (speed >= 2) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
                                               : DISABLE_ALL_INTER_SPLIT;
     } else {
@@ -233,7 +233,7 @@
   }
 
   if (speed >= 5) {
-    if (MIN(cm->width, cm->height) >= 720) {
+    if (VPXMIN(cm->width, cm->height) >= 720) {
       sf->partition_search_breakout_dist_thr = (1 << 25);
     } else {
       sf->partition_search_breakout_dist_thr = (1 << 23);
@@ -241,7 +241,7 @@
   }
 
   if (speed >= 7) {
-    sf->encode_breakout_thresh = (MIN(cm->width, cm->height) >= 720) ?
+    sf->encode_breakout_thresh = (VPXMIN(cm->width, cm->height) >= 720) ?
         800 : 300;
   }
 }
diff --git a/vp9/encoder/vp9_svc_layercontext.c b/vp9/encoder/vp9_svc_layercontext.c
index e69404a..b619840 100644
--- a/vp9/encoder/vp9_svc_layercontext.c
+++ b/vp9/encoder/vp9_svc_layercontext.c
@@ -113,8 +113,6 @@
 
   if (svc->temporal_layering_mode != VP9E_TEMPORAL_LAYERING_MODE_NOLAYERING) {
     for (sl = 0; sl < oxcf->ss_number_layers; ++sl) {
-      spatial_layer_target = 0;
-
       for (tl = 0; tl < oxcf->ts_number_layers; ++tl) {
         layer = LAYER_IDS_TO_IDX(sl, tl, oxcf->ts_number_layers);
         svc->layer_context[layer].target_bandwidth =
@@ -141,8 +139,8 @@
         lrc->maximum_buffer_size =
             (int64_t)(rc->maximum_buffer_size * bitrate_alloc);
         lrc->bits_off_target =
-            MIN(lrc->bits_off_target, lrc->maximum_buffer_size);
-        lrc->buffer_level = MIN(lrc->buffer_level, lrc->maximum_buffer_size);
+            VPXMIN(lrc->bits_off_target, lrc->maximum_buffer_size);
+        lrc->buffer_level = VPXMIN(lrc->buffer_level, lrc->maximum_buffer_size);
         lc->framerate = cpi->framerate / oxcf->ts_rate_decimator[tl];
         lrc->avg_frame_bandwidth = (int)(lc->target_bandwidth / lc->framerate);
         lrc->max_frame_bandwidth = rc->max_frame_bandwidth;
@@ -173,9 +171,9 @@
           (int64_t)(rc->optimal_buffer_level * bitrate_alloc);
       lrc->maximum_buffer_size =
           (int64_t)(rc->maximum_buffer_size * bitrate_alloc);
-      lrc->bits_off_target = MIN(lrc->bits_off_target,
-                                 lrc->maximum_buffer_size);
-      lrc->buffer_level = MIN(lrc->buffer_level, lrc->maximum_buffer_size);
+      lrc->bits_off_target = VPXMIN(lrc->bits_off_target,
+                                    lrc->maximum_buffer_size);
+      lrc->buffer_level = VPXMIN(lrc->buffer_level, lrc->maximum_buffer_size);
       // Update framerate-related quantities.
       if (svc->number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR) {
         lc->framerate = cpi->framerate / oxcf->ts_rate_decimator[layer];
diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index 439eac6..9c1629e 100644
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -242,7 +242,7 @@
   xd->plane[0].pre[0].stride = stride;
 
   step_param = mv_sf->reduce_first_step_size;
-  step_param = MIN(step_param, MAX_MVSEARCH_STEPS - 2);
+  step_param = VPXMIN(step_param, MAX_MVSEARCH_STEPS - 2);
 
   // Ignore mv costing by sending NULL pointer instead of cost arrays
   vp9_hex_search(x, &best_ref_mv1_full, step_param, sadpb, 1,
diff --git a/vp9/vp9_dx_iface.c b/vp9/vp9_dx_iface.c
index 96ede3c..8119df8 100644
--- a/vp9/vp9_dx_iface.c
+++ b/vp9/vp9_dx_iface.c
@@ -18,6 +18,7 @@
 #include "vpx/vp8dx.h"
 #include "vpx/vpx_decoder.h"
 #include "vpx_dsp/bitreader_buffer.h"
+#include "vpx_dsp/vpx_dsp_common.h"
 #include "vpx_util/vpx_thread.h"
 
 #include "vp9/common/vp9_alloccommon.h"
@@ -183,7 +184,7 @@
   si->w = si->h = 0;
 
   if (decrypt_cb) {
-    data_sz = MIN(sizeof(clear_buffer), data_sz);
+    data_sz = VPXMIN(sizeof(clear_buffer), data_sz);
     decrypt_cb(decrypt_state, data, clear_buffer, data_sz);
     data = clear_buffer;
   }
diff --git a/vpx_dsp/bitreader.c b/vpx_dsp/bitreader.c
index 4420fad..ba87022 100644
--- a/vpx_dsp/bitreader.c
+++ b/vpx_dsp/bitreader.c
@@ -48,7 +48,7 @@
   int shift = BD_VALUE_SIZE - CHAR_BIT - (count + CHAR_BIT);
 
   if (r->decrypt_cb) {
-    size_t n = MIN(sizeof(r->clear_buffer), bytes_left);
+    size_t n = VPXMIN(sizeof(r->clear_buffer), bytes_left);
     r->decrypt_cb(r->decrypt_state, buffer, r->clear_buffer, (int)n);
     buffer = r->clear_buffer;
     buffer_start = r->clear_buffer;
diff --git a/vpx_dsp/prob.h b/vpx_dsp/prob.h
index 729f90a..c3cb103 100644
--- a/vpx_dsp/prob.h
+++ b/vpx_dsp/prob.h
@@ -65,7 +65,7 @@
                                    unsigned int count_sat,
                                    unsigned int max_update_factor) {
   const vpx_prob prob = get_binary_prob(ct[0], ct[1]);
-  const unsigned int count = MIN(ct[0] + ct[1], count_sat);
+  const unsigned int count = VPXMIN(ct[0] + ct[1], count_sat);
   const unsigned int factor = max_update_factor * count / count_sat;
   return weighted_prob(pre_prob, prob, factor);
 }
@@ -82,7 +82,7 @@
   if (den == 0) {
     return pre_prob;
   } else {
-    const unsigned int count = MIN(den, MODE_MV_COUNT_SAT);
+    const unsigned int count = VPXMIN(den, MODE_MV_COUNT_SAT);
     const unsigned int factor = count_to_update_factor[count];
     const vpx_prob prob =
         clip_prob(((int64_t)(ct[0]) * 256 + (den >> 1)) / den);
diff --git a/vpx_dsp/vpx_dsp.mk b/vpx_dsp/vpx_dsp.mk
index 1959c4d..61670d9 100644
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -308,6 +308,8 @@
 DSP_SRCS-$(HAVE_MMX)    += x86/variance_impl_mmx.asm
 DSP_SRCS-$(HAVE_SSE)    += x86/variance_sse2.c
 DSP_SRCS-$(HAVE_SSE2)   += x86/variance_sse2.c  # Contains SSE2 and SSSE3
+DSP_SRCS-$(HAVE_SSE2)   += x86/halfpix_variance_sse2.c
+DSP_SRCS-$(HAVE_SSE2)   += x86/halfpix_variance_impl_sse2.asm
 DSP_SRCS-$(HAVE_AVX2)   += x86/variance_avx2.c
 DSP_SRCS-$(HAVE_AVX2)   += x86/variance_impl_avx2.c
 
diff --git a/vpx_dsp/vpx_dsp_common.h b/vpx_dsp/vpx_dsp_common.h
index ccb8189..a83339e 100644
--- a/vpx_dsp/vpx_dsp_common.h
+++ b/vpx_dsp/vpx_dsp_common.h
@@ -19,8 +19,8 @@
 extern "C" {
 #endif
 
-#define MIN(x, y) (((x) < (y)) ? (x) : (y))
-#define MAX(x, y) (((x) > (y)) ? (x) : (y))
+#define VPXMIN(x, y) (((x) < (y)) ? (x) : (y))
+#define VPXMAX(x, y) (((x) > (y)) ? (x) : (y))
 
 #if CONFIG_VP9_HIGHBITDEPTH
 // Note:
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 1e56d53..128dd8b 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -1373,13 +1373,13 @@
 # Specialty Subpixel
 #
 add_proto qw/uint32_t vpx_variance_halfpixvar16x16_h/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, uint32_t *sse";
-  specialize qw/vpx_variance_halfpixvar16x16_h mmx media/;
+  specialize qw/vpx_variance_halfpixvar16x16_h mmx sse2 media/;
 
 add_proto qw/uint32_t vpx_variance_halfpixvar16x16_v/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, uint32_t *sse";
-  specialize qw/vpx_variance_halfpixvar16x16_v mmx media/;
+  specialize qw/vpx_variance_halfpixvar16x16_v mmx sse2 media/;
 
 add_proto qw/uint32_t vpx_variance_halfpixvar16x16_hv/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, uint32_t *sse";
-  specialize qw/vpx_variance_halfpixvar16x16_hv mmx media/;
+  specialize qw/vpx_variance_halfpixvar16x16_hv mmx sse2 media/;
 
 if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   add_proto qw/unsigned int vpx_highbd_12_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
diff --git a/vpx_dsp/x86/halfpix_variance_impl_sse2.asm b/vpx_dsp/x86/halfpix_variance_impl_sse2.asm
new file mode 100644
index 0000000..cc26bb6
--- /dev/null
+++ b/vpx_dsp/x86/halfpix_variance_impl_sse2.asm
@@ -0,0 +1,346 @@
+;
+;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+;  Use of this source code is governed by a BSD-style license
+;  that can be found in the LICENSE file in the root of the source
+;  tree. An additional intellectual property rights grant can be found
+;  in the file PATENTS.  All contributing project authors may
+;  be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vpx_half_horiz_vert_variance16x_h_sse2(unsigned char *ref,
+;                                            int ref_stride,
+;                                            unsigned char *src,
+;                                            int src_stride,
+;                                            unsigned int height,
+;                                            int *sum,
+;                                            unsigned int *sumsquared)
+global sym(vpx_half_horiz_vert_variance16x_h_sse2) PRIVATE
+sym(vpx_half_horiz_vert_variance16x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0) ;ref
+
+        mov             rdi,            arg(2) ;src
+        movsxd          rcx,            dword ptr arg(4) ;height
+        movsxd          rax,            dword ptr arg(1) ;ref_stride
+        movsxd          rdx,            dword ptr arg(3)    ;src_stride
+
+        pxor            xmm0,           xmm0                ;
+
+        movdqu          xmm5,           XMMWORD PTR [rsi]
+        movdqu          xmm3,           XMMWORD PTR [rsi+1]
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3) horizontal line 1
+
+        lea             rsi,            [rsi + rax]
+
+vpx_half_horiz_vert_variance16x_h_1:
+        movdqu          xmm1,           XMMWORD PTR [rsi]     ;
+        movdqu          xmm2,           XMMWORD PTR [rsi+1]   ;
+        pavgb           xmm1,           xmm2                ;  xmm1 = avg(xmm1,xmm3) horizontal line i+1
+
+        pavgb           xmm5,           xmm1                ;  xmm = vertical average of the above
+
+        movdqa          xmm4,           xmm5
+        punpcklbw       xmm5,           xmm0                ;  xmm5 = words of above
+        punpckhbw       xmm4,           xmm0
+
+        movq            xmm3,           QWORD PTR [rdi]     ;  xmm3 = d0,d1,d2..d7
+        punpcklbw       xmm3,           xmm0                ;  xmm3 = words of above
+        psubw           xmm5,           xmm3                ;  xmm5 -= xmm3
+
+        movq            xmm3,           QWORD PTR [rdi+8]
+        punpcklbw       xmm3,           xmm0
+        psubw           xmm4,           xmm3
+
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        paddw           xmm6,           xmm4
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        pmaddwd         xmm4,           xmm4
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+        paddd           xmm7,           xmm4
+
+        movdqa          xmm5,           xmm1                ;  save xmm1 for use on the next row
+
+        lea             rsi,            [rsi + rax]
+        lea             rdi,            [rdi + rdx]
+
+        sub             rcx,            1                   ;
+        jnz             vpx_half_horiz_vert_variance16x_h_1     ;
+
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(5) ;[Sum]
+        mov         rdi,            arg(6) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vpx_half_vert_variance16x_h_sse2(unsigned char *ref,
+;                                      int ref_stride,
+;                                      unsigned char *src,
+;                                      int src_stride,
+;                                      unsigned int height,
+;                                      int *sum,
+;                                      unsigned int *sumsquared)
+global sym(vpx_half_vert_variance16x_h_sse2) PRIVATE
+sym(vpx_half_vert_variance16x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0)              ;ref
+
+        mov             rdi,            arg(2)              ;src
+        movsxd          rcx,            dword ptr arg(4)    ;height
+        movsxd          rax,            dword ptr arg(1)    ;ref_stride
+        movsxd          rdx,            dword ptr arg(3)    ;src_stride
+
+        movdqu          xmm5,           XMMWORD PTR [rsi]
+        lea             rsi,            [rsi + rax          ]
+        pxor            xmm0,           xmm0
+
+vpx_half_vert_variance16x_h_1:
+        movdqu          xmm3,           XMMWORD PTR [rsi]
+
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3)
+        movdqa          xmm4,           xmm5
+        punpcklbw       xmm5,           xmm0
+        punpckhbw       xmm4,           xmm0
+
+        movq            xmm2,           QWORD PTR [rdi]
+        punpcklbw       xmm2,           xmm0
+        psubw           xmm5,           xmm2
+        movq            xmm2,           QWORD PTR [rdi+8]
+        punpcklbw       xmm2,           xmm0
+        psubw           xmm4,           xmm2
+
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        paddw           xmm6,           xmm4
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        pmaddwd         xmm4,           xmm4
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+        paddd           xmm7,           xmm4
+
+        movdqa          xmm5,           xmm3
+
+        lea             rsi,            [rsi + rax]
+        lea             rdi,            [rdi + rdx]
+
+        sub             rcx,            1
+        jnz             vpx_half_vert_variance16x_h_1
+
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(5) ;[Sum]
+        mov         rdi,            arg(6) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+
+;void vpx_half_horiz_variance16x_h_sse2(unsigned char *ref,
+;                                       int ref_stride
+;                                       unsigned char *src,
+;                                       int src_stride,
+;                                       unsigned int height,
+;                                       int *sum,
+;                                       unsigned int *sumsquared)
+global sym(vpx_half_horiz_variance16x_h_sse2) PRIVATE
+sym(vpx_half_horiz_variance16x_h_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    SAVE_XMM 7
+    GET_GOT     rbx
+    push rsi
+    push rdi
+    ; end prolog
+
+        pxor            xmm6,           xmm6                ;  error accumulator
+        pxor            xmm7,           xmm7                ;  sse eaccumulator
+        mov             rsi,            arg(0) ;ref
+
+        mov             rdi,            arg(2) ;src
+        movsxd          rcx,            dword ptr arg(4) ;height
+        movsxd          rax,            dword ptr arg(1) ;ref_stride
+        movsxd          rdx,            dword ptr arg(3)    ;src_stride
+
+        pxor            xmm0,           xmm0                ;
+
+vpx_half_horiz_variance16x_h_1:
+        movdqu          xmm5,           XMMWORD PTR [rsi]     ;  xmm5 = s0,s1,s2..s15
+        movdqu          xmm3,           XMMWORD PTR [rsi+1]   ;  xmm3 = s1,s2,s3..s16
+
+        pavgb           xmm5,           xmm3                ;  xmm5 = avg(xmm1,xmm3)
+        movdqa          xmm1,           xmm5
+        punpcklbw       xmm5,           xmm0                ;  xmm5 = words of above
+        punpckhbw       xmm1,           xmm0
+
+        movq            xmm3,           QWORD PTR [rdi]     ;  xmm3 = d0,d1,d2..d7
+        punpcklbw       xmm3,           xmm0                ;  xmm3 = words of above
+        movq            xmm2,           QWORD PTR [rdi+8]
+        punpcklbw       xmm2,           xmm0
+
+        psubw           xmm5,           xmm3                ;  xmm5 -= xmm3
+        psubw           xmm1,           xmm2
+        paddw           xmm6,           xmm5                ;  xmm6 += accumulated column differences
+        paddw           xmm6,           xmm1
+        pmaddwd         xmm5,           xmm5                ;  xmm5 *= xmm5
+        pmaddwd         xmm1,           xmm1
+        paddd           xmm7,           xmm5                ;  xmm7 += accumulated square column differences
+        paddd           xmm7,           xmm1
+
+        lea             rsi,            [rsi + rax]
+        lea             rdi,            [rdi + rdx]
+
+        sub             rcx,            1                   ;
+        jnz             vpx_half_horiz_variance16x_h_1        ;
+
+        pxor        xmm1,           xmm1
+        pxor        xmm5,           xmm5
+
+        punpcklwd   xmm0,           xmm6
+        punpckhwd   xmm1,           xmm6
+        psrad       xmm0,           16
+        psrad       xmm1,           16
+        paddd       xmm0,           xmm1
+        movdqa      xmm1,           xmm0
+
+        movdqa      xmm6,           xmm7
+        punpckldq   xmm6,           xmm5
+        punpckhdq   xmm7,           xmm5
+        paddd       xmm6,           xmm7
+
+        punpckldq   xmm0,           xmm5
+        punpckhdq   xmm1,           xmm5
+        paddd       xmm0,           xmm1
+
+        movdqa      xmm7,           xmm6
+        movdqa      xmm1,           xmm0
+
+        psrldq      xmm7,           8
+        psrldq      xmm1,           8
+
+        paddd       xmm6,           xmm7
+        paddd       xmm0,           xmm1
+
+        mov         rsi,            arg(5) ;[Sum]
+        mov         rdi,            arg(6) ;[SSE]
+
+        movd        [rsi],       xmm0
+        movd        [rdi],       xmm6
+
+    ; begin epilog
+    pop rdi
+    pop rsi
+    RESTORE_GOT
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+SECTION_RODATA
+;    short xmm_bi_rd[8] = { 64, 64, 64, 64,64, 64, 64, 64};
+align 16
+xmm_bi_rd:
+    times 8 dw 64
+align 16
+vpx_bilinear_filters_sse2:
+    dw 128, 128, 128, 128, 128, 128, 128, 128,  0,  0,  0,  0,  0,  0,  0,  0
+    dw 112, 112, 112, 112, 112, 112, 112, 112, 16, 16, 16, 16, 16, 16, 16, 16
+    dw 96, 96, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32
+    dw 80, 80, 80, 80, 80, 80, 80, 80, 48, 48, 48, 48, 48, 48, 48, 48
+    dw 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64
+    dw 48, 48, 48, 48, 48, 48, 48, 48, 80, 80, 80, 80, 80, 80, 80, 80
+    dw 32, 32, 32, 32, 32, 32, 32, 32, 96, 96, 96, 96, 96, 96, 96, 96
+    dw 16, 16, 16, 16, 16, 16, 16, 16, 112, 112, 112, 112, 112, 112, 112, 112
diff --git a/vpx_dsp/x86/halfpix_variance_sse2.c b/vpx_dsp/x86/halfpix_variance_sse2.c
new file mode 100644
index 0000000..5782155
--- /dev/null
+++ b/vpx_dsp/x86/halfpix_variance_sse2.c
@@ -0,0 +1,74 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+
+void vpx_half_horiz_vert_variance16x_h_sse2(const unsigned char *ref,
+                                            int ref_stride,
+                                            const unsigned char *src,
+                                            int src_stride,
+                                            unsigned int height,
+                                            int *sum,
+                                            unsigned int *sumsquared);
+void vpx_half_horiz_variance16x_h_sse2(const unsigned char *ref, int ref_stride,
+                                       const unsigned char *src, int src_stride,
+                                       unsigned int height, int *sum,
+                                       unsigned int *sumsquared);
+void vpx_half_vert_variance16x_h_sse2(const unsigned char *ref, int ref_stride,
+                                      const unsigned char *src, int src_stride,
+                                      unsigned int height, int *sum,
+                                      unsigned int *sumsquared);
+
+uint32_t vpx_variance_halfpixvar16x16_h_sse2(const unsigned char *src,
+                                             int src_stride,
+                                             const unsigned char *dst,
+                                             int dst_stride,
+                                             uint32_t *sse) {
+  int xsum0;
+  unsigned int xxsum0;
+
+  vpx_half_horiz_variance16x_h_sse2(src, src_stride, dst, dst_stride, 16,
+                                    &xsum0, &xxsum0);
+
+  *sse = xxsum0;
+  return (xxsum0 - (((uint32_t)xsum0 * xsum0) >> 8));
+}
+
+uint32_t vpx_variance_halfpixvar16x16_v_sse2(const unsigned char *src,
+                                             int src_stride,
+                                             const unsigned char *dst,
+                                             int dst_stride,
+                                             uint32_t *sse) {
+  int xsum0;
+  unsigned int xxsum0;
+  vpx_half_vert_variance16x_h_sse2(src, src_stride, dst, dst_stride, 16,
+                                   &xsum0, &xxsum0);
+
+  *sse = xxsum0;
+  return (xxsum0 - (((uint32_t)xsum0 * xsum0) >> 8));
+}
+
+
+uint32_t vpx_variance_halfpixvar16x16_hv_sse2(const unsigned char *src,
+                                              int src_stride,
+                                              const unsigned char *dst,
+                                              int dst_stride,
+                                              uint32_t *sse) {
+  int xsum0;
+  unsigned int xxsum0;
+
+  vpx_half_horiz_vert_variance16x_h_sse2(src, src_stride, dst, dst_stride, 16,
+                                         &xsum0, &xxsum0);
+
+  *sse = xxsum0;
+  return (xxsum0 - (((uint32_t)xsum0 * xsum0) >> 8));
+}