From a8fbab8697c77dc82013862e0a9c7f82b6402db8 Mon Sep 17 00:00:00 2001
From: Yaowu Xu <yaowu@google.com>
Date: Wed, 30 Nov 2011 16:25:00 -0800
Subject: [PATCH] enabled 8x8 intra prediction modes on inter frames

This commit enabled the usage of 8x8 intra prediction modes on inter
frames. There are a few TODO items related to this: 1)baseline entropy
need be calibrated; 2)cost of UV need to be done more properly rather
than using decision only relying on Y; 3)Threshold for allowing picking
8x8 intra prediction should be lowered to lower than the B_PRED.

Even with all the TODOs, tests showed consistent gain on derf set ~0.1%
(PSNR:0.08% and SSIM:0.14%). It is assumed that 8x8 intra prediction
will help more on large resolution clips, especially with above TODOs
addressed.

Change-Id: I398ada49dfc32575cfab962a569c2885111ae3ba
---
 vp8/common/entropymode.c  |  17 +++++--
 vp8/common/findnearmv.h   |   1 -
 vp8/common/invtrans.c     |   3 ++
 vp8/common/loopfilter.c   |  14 +++++-
 vp8/common/loopfilter.h   |   3 +-
 vp8/decoder/decodemv.c    |  21 ++++++--
 vp8/decoder/decodframe.c  |  19 +++++++-
 vp8/decoder/detokenize.c  |   3 ++
 vp8/decoder/onyxd_if.c    |   3 +-
 vp8/decoder/threading.c   |   3 ++
 vp8/encoder/bitstream.c   |  10 ++++
 vp8/encoder/encodeframe.c |  11 +++--
 vp8/encoder/encodemb.c    |  13 +++++
 vp8/encoder/onyx_if.c     |  48 ++++++++++++++++--
 vp8/encoder/onyx_int.h    |  10 +++-
 vp8/encoder/quantize.c    |   6 +++
 vp8/encoder/rdopt.c       | 100 +++++++++++++++++++++++++++++---------
 vp8/encoder/tokenize.c    |   6 ++-
 18 files changed, 246 insertions(+), 45 deletions(-)

diff --git a/vp8/common/entropymode.c b/vp8/common/entropymode.c
index 45af600ca..8b9934461 100644
--- a/vp8/common/entropymode.c
+++ b/vp8/common/entropymode.c
@@ -29,7 +29,8 @@ const unsigned int kf_y_mode_cts[8][VP8_YMODES] =
 #else
 static const unsigned int kf_y_mode_cts[VP8_YMODES] = { 49, 22, 23, 11, 23, 128};
 #endif
-static const unsigned int y_mode_cts  [VP8_YMODES] = { 8080, 1908, 1582, 1007, 0, 5874};
+/* TODO: calibrate the baseline distribution */
+static const unsigned int y_mode_cts  [VP8_YMODES] = { 8080, 1908, 1582, 1007, 2000, 5874};
 #else
 static const unsigned int kf_y_mode_cts[VP8_YMODES] = { 1607, 915, 812, 811, 5455};
 static const unsigned int y_mode_cts  [VP8_YMODES] = { 8080, 1908, 1582, 1007, 5874};
@@ -168,15 +169,16 @@ const vp8_tree_index vp8_bmode_tree[18] =     /* INTRAMODECONTEXTNODE value */
 
 /* Again, these trees use the same probability indices as their
    explicitly-programmed predecessors. */
-const vp8_tree_index vp8_ymode_tree[8] =
+#if CONFIG_I8X8
+const vp8_tree_index vp8_ymode_tree[10] =
 {
     -DC_PRED, 2,
     4, 6,
     -V_PRED, -H_PRED,
-    -TM_PRED, -B_PRED
+    -TM_PRED, 8,
+    -B_PRED, -I8X8_PRED
 };
 
-#if CONFIG_I8X8
 const vp8_tree_index vp8_kf_ymode_tree[10] =
 {
     -B_PRED, 2,
@@ -193,6 +195,13 @@ const vp8_tree_index vp8_i8x8_mode_tree[6] =
     -H_PRED, -TM_PRED
 };
 #else
+const vp8_tree_index vp8_ymode_tree[8] =
+{
+    -DC_PRED, 2,
+    4, 6,
+    -V_PRED, -H_PRED,
+    -TM_PRED, -B_PRED
+};
 
 const vp8_tree_index vp8_kf_ymode_tree[8] =
 {
diff --git a/vp8/common/findnearmv.h b/vp8/common/findnearmv.h
index 4a5b0f3a4..b07a54b5c 100644
--- a/vp8/common/findnearmv.h
+++ b/vp8/common/findnearmv.h
@@ -145,7 +145,6 @@ static B_PREDICTION_MODE left_block_mode(const MODE_INFO *cur_mb, int b)
                 return B_DC_PRED;
         }
     }
-
     return (cur_mb->bmi + b - 1)->as_mode;
 }
 
diff --git a/vp8/common/invtrans.c b/vp8/common/invtrans.c
index d361b654a..c734df42f 100644
--- a/vp8/common/invtrans.c
+++ b/vp8/common/invtrans.c
@@ -77,6 +77,9 @@ void vp8_inverse_transform_mb(const vp8_idct_rtcd_vtable_t *rtcd, MACROBLOCKD *x
     int i;
 
     if (x->mode_info_context->mbmi.mode != B_PRED &&
+#if CONFIG_I8X8
+        x->mode_info_context->mbmi.mode != I8X8_PRED &&
+#endif
         x->mode_info_context->mbmi.mode != SPLITMV)
     {
         /* do 2nd order transform on the dc block */
diff --git a/vp8/common/loopfilter.c b/vp8/common/loopfilter.c
index 98657fcc3..11853310d 100644
--- a/vp8/common/loopfilter.c
+++ b/vp8/common/loopfilter.c
@@ -137,7 +137,9 @@ static void lf_init_lut(loop_filter_info_n *lfi)
     lfi->mode_lf_lut[H_PRED] = 1;
     lfi->mode_lf_lut[TM_PRED] = 1;
     lfi->mode_lf_lut[B_PRED]  = 0;
-
+#if CONFIG_I8X8
+    lfi->mode_lf_lut[I8X8_PRED]=0;
+#endif
     lfi->mode_lf_lut[ZEROMV]  = 1;
     lfi->mode_lf_lut[NEARESTMV] = 2;
     lfi->mode_lf_lut[NEARMV] = 2;
@@ -321,6 +323,9 @@ void vp8_loop_filter_frame
         for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
         {
             int skip_lf = (mode_info_context->mbmi.mode != B_PRED &&
+#if CONFIG_I8X8
+                            mode_info_context->mbmi.mode != I8X8_PRED &&
+#endif
                             mode_info_context->mbmi.mode != SPLITMV &&
                             mode_info_context->mbmi.mb_skip_coeff);
 
@@ -432,6 +437,10 @@ void vp8_loop_filter_frame_yonly
         for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
         {
             int skip_lf = (mode_info_context->mbmi.mode != B_PRED &&
+#if CONFIG_I8X8
+                            mode_info_context->mbmi.mode != I8X8_PRED &&
+#endif
+
                             mode_info_context->mbmi.mode != SPLITMV &&
                             mode_info_context->mbmi.mb_skip_coeff);
 
@@ -569,6 +578,9 @@ void vp8_loop_filter_partial_frame
         for (mb_col = 0; mb_col < mb_cols; mb_col++)
         {
             int skip_lf = (mode_info_context->mbmi.mode != B_PRED &&
+#if CONFIG_I8X8
+                            mode_info_context->mbmi.mode != I8X8_PRED &&
+#endif
                            mode_info_context->mbmi.mode != SPLITMV &&
                            mode_info_context->mbmi.mb_skip_coeff);
 
diff --git a/vp8/common/loopfilter.h b/vp8/common/loopfilter.h
index 9887cf55b..0ae2b9a1b 100644
--- a/vp8/common/loopfilter.h
+++ b/vp8/common/loopfilter.h
@@ -14,6 +14,7 @@
 
 #include "vpx_ports/mem.h"
 #include "vpx_config.h"
+#include "blockd.h"
 
 #define MAX_LOOP_FILTER 63
 
@@ -40,7 +41,7 @@ typedef struct
     DECLARE_ALIGNED(SIMD_WIDTH, unsigned char, hev_thr[4][SIMD_WIDTH]);
     unsigned char lvl[4][4][4];
     unsigned char hev_thr_lut[2][MAX_LOOP_FILTER + 1];
-    unsigned char mode_lf_lut[10];
+    unsigned char mode_lf_lut[MB_MODE_COUNT];
 } loop_filter_info_n;
 
 typedef struct
diff --git a/vp8/decoder/decodemv.c b/vp8/decoder/decodemv.c
index 683f40a5b..68044f4d5 100644
--- a/vp8/decoder/decodemv.c
+++ b/vp8/decoder/decodemv.c
@@ -144,8 +144,6 @@ static void vp8_kfread_modes(VP8D_COMP *pbi,
     {
         int i;
         int mode8x8;
-        //printf("F%3d:%d:%d:",
-                 pbi->common.current_video_frame, mb_row, mb_col);
         for(i=0;i<4;i++)
          {
              int ib = vp8_i8x8_block[i];
@@ -155,8 +153,6 @@ static void vp8_kfread_modes(VP8D_COMP *pbi,
              m->bmi[ib+4].as_mode= mode8x8;
              m->bmi[ib+5].as_mode= mode8x8;
          }
-        //printf("%2d%2d%2d%2d\n", m->bmi[0].as_mode,m->bmi[2].as_mode,
-        //                       m->bmi[8].as_mode,m->bmi[10].as_mode);
    }
     else
 #endif
@@ -724,6 +720,23 @@ static void read_mb_modes_mv(VP8D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi,
             while (++j < 16);
         }
 
+#if CONFIG_I8X8
+        if(mbmi->mode == I8X8_PRED)
+        {
+            int i;
+            int mode8x8;
+            for(i=0;i<4;i++)
+            {
+                int ib = vp8_i8x8_block[i];
+                mode8x8 = vp8_read_i8x8_mode(bc, pbi->common.i8x8_mode_prob);
+                mi->bmi[ib+0].as_mode= mode8x8;
+                mi->bmi[ib+1].as_mode= mode8x8;
+                mi->bmi[ib+4].as_mode= mode8x8;
+                mi->bmi[ib+5].as_mode= mode8x8;
+            }
+        }
+        else
+#endif
         mbmi->uv_mode = (MB_PREDICTION_MODE)vp8_read_uv_mode(bc, pbi->common.fc.uv_mode_prob);
     }
 
diff --git a/vp8/decoder/decodframe.c b/vp8/decoder/decodframe.c
index a3421ef89..5e8925286 100644
--- a/vp8/decoder/decodframe.c
+++ b/vp8/decoder/decodframe.c
@@ -43,6 +43,7 @@
 #include <assert.h>
 #include <stdio.h>
 
+
 #ifdef DEC_DEBUG
 int dec_debug = 0;
 #endif
@@ -134,6 +135,7 @@ static void skip_recon_mb(VP8D_COMP *pbi, MACROBLOCKD *xd)
         vp8_build_inter16x16_predictors_mb(xd, xd->dst.y_buffer,
                                            xd->dst.u_buffer, xd->dst.v_buffer,
                                            xd->dst.y_stride, xd->dst.uv_stride);
+    }
 #ifdef DEC_DEBUG
         if (dec_debug) {
           int i, j;
@@ -144,7 +146,7 @@ static void skip_recon_mb(VP8D_COMP *pbi, MACROBLOCKD *xd)
           }
         }
 #endif
-    }
+
 }
 
 static void clamp_mv_to_umv_border(MV *mv, const MACROBLOCKD *xd)
@@ -271,6 +273,19 @@ static void decode_macroblock(VP8D_COMP *pbi, MACROBLOCKD *xd,
         return;
     }
 
+#ifdef DEC_DEBUG
+        if (dec_debug) {
+          int i, j;
+          printf("Generating predictors\n");
+          for (i=0;i<16;i++) {
+            for (j=0;j<16;j++) printf("%3d ", xd->dst.y_buffer[i*xd->dst.y_stride+j]);
+            printf("\n");
+          }
+        }
+#endif
+
+
+
     if (xd->segmentation_enabled)
         mb_init_dequantizer(pbi, xd);
 
@@ -597,7 +612,7 @@ decode_mb_row(VP8D_COMP *pbi, VP8_COMMON *pc, int mb_row, MACROBLOCKD *xd)
         }
 
 #ifdef DEC_DEBUG
-        dec_debug = (pc->current_video_frame==0 && mb_row==1 && mb_col==11);
+        dec_debug = (pc->current_video_frame==1 && mb_row==4 && mb_col==0);
 #endif
         decode_macroblock(pbi, xd, mb_row * pc->mb_cols  + mb_col);
 
diff --git a/vp8/decoder/detokenize.c b/vp8/decoder/detokenize.c
index f2a5db8f1..ef57b4f4e 100644
--- a/vp8/decoder/detokenize.c
+++ b/vp8/decoder/detokenize.c
@@ -90,6 +90,9 @@ void vp8_reset_mb_tokens_context(MACROBLOCKD *x)
 {
     /* Clear entropy contexts for Y2 blocks */
     if (x->mode_info_context->mbmi.mode != B_PRED &&
+#if CONFIG_I8X8
+        x->mode_info_context->mbmi.mode != I8X8_PRED &&
+#endif
         x->mode_info_context->mbmi.mode != SPLITMV)
     {
         vpx_memset(x->above_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
diff --git a/vp8/decoder/onyxd_if.c b/vp8/decoder/onyxd_if.c
index f103bd788..a45d692d0 100644
--- a/vp8/decoder/onyxd_if.c
+++ b/vp8/decoder/onyxd_if.c
@@ -80,7 +80,6 @@ void vp8_recon_write_yuv_frame(char *name, YV12_BUFFER_CONFIG *s)
     fclose(yuv_file);
 }
 #endif
-
 #if WRITE_RECON_BUFFER
 void write_dx_frame_to_file(YV12_BUFFER_CONFIG *frame, int this_frame)
 {
@@ -581,6 +580,8 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign
 #if WRITE_RECON_BUFFER
         if(cm->show_frame)
             write_dx_frame_to_file(cm->frame_to_show, cm->current_video_frame);
+        else
+            write_dx_frame_to_file(cm->frame_to_show, cm->current_video_frame+1000);
 #endif
 
         if(cm->filter_level)
diff --git a/vp8/decoder/threading.c b/vp8/decoder/threading.c
index e6d31008d..09dffe269 100644
--- a/vp8/decoder/threading.c
+++ b/vp8/decoder/threading.c
@@ -894,6 +894,9 @@ void vp8mt_decode_mb_rows( VP8D_COMP *pbi, MACROBLOCKD *xd)
                 if (pbi->common.filter_level)
                 {
                     int skip_lf = (xd->mode_info_context->mbmi.mode != B_PRED &&
+#if CONFIG_I8X8
+                                    xd->mode_info_context->mbmi.mode != I8X8_PRED &&
+#endif
                                     xd->mode_info_context->mbmi.mode != SPLITMV &&
                                     xd->mode_info_context->mbmi.mb_skip_coeff);
 
diff --git a/vp8/encoder/bitstream.c b/vp8/encoder/bitstream.c
index 07077208a..6171492c6 100644
--- a/vp8/encoder/bitstream.c
+++ b/vp8/encoder/bitstream.c
@@ -1104,6 +1104,16 @@ static void pack_inter_mode_mvs(VP8_COMP *const cpi)
                         write_bmode(w, m->bmi[j].as_mode, pc->fc.bmode_prob);
                     while (++j < 16);
                 }
+#if CONFIG_I8X8
+                if(mode == I8X8_PRED)
+                {
+                    write_i8x8_mode(w, m->bmi[0].as_mode, pc->i8x8_mode_prob);
+                    write_i8x8_mode(w, m->bmi[2].as_mode, pc->i8x8_mode_prob);
+                    write_i8x8_mode(w, m->bmi[8].as_mode, pc->i8x8_mode_prob);
+                    write_i8x8_mode(w, m->bmi[10].as_mode, pc->i8x8_mode_prob);
+                }
+                else
+#endif
 
                 write_uv_mode(w, mi->uv_mode, pc->fc.uv_mode_prob);
             }
diff --git a/vp8/encoder/encodeframe.c b/vp8/encoder/encodeframe.c
index 22d53302a..a6a09247a 100644
--- a/vp8/encoder/encodeframe.c
+++ b/vp8/encoder/encodeframe.c
@@ -617,8 +617,7 @@ void encode_mb_row(VP8_COMP *cpi,
     for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
     {
 #ifdef ENC_DEBUG
-        //enc_debug = (cpi->count==29 && mb_row==5 && mb_col==0);
-        enc_debug = (cpi->count==4 && mb_row==17 && mb_col==13);
+        enc_debug = (cpi->common.current_video_frame ==1 && mb_row==4 && mb_col==0);
         mb_col_debug=mb_col;
         mb_row_debug=mb_row;
 #endif
@@ -1553,12 +1552,18 @@ int vp8cx_encode_inter_macroblock
             vp8_encode_intra16x16mbuv(IF_RTCD(&cpi->rtcd), x);
             vp8_encode_intra4x4mby(IF_RTCD(&cpi->rtcd), x);
         }
+#if CONFIG_I8X8
+        else if(xd->mode_info_context->mbmi.mode == I8X8_PRED)
+        {
+            vp8_encode_intra8x8mby(IF_RTCD(&cpi->rtcd), x);
+            vp8_encode_intra8x8mbuv(IF_RTCD(&cpi->rtcd), x);
+        }
+#endif
         else
         {
             vp8_encode_intra16x16mbuv(IF_RTCD(&cpi->rtcd), x);
             vp8_encode_intra16x16mby(IF_RTCD(&cpi->rtcd), x);
         }
-
         sum_intra_stats(cpi, x);
     }
     else
diff --git a/vp8/encoder/encodemb.c b/vp8/encoder/encodemb.c
index 483ec768f..2178e5bae 100644
--- a/vp8/encoder/encodemb.c
+++ b/vp8/encoder/encodemb.c
@@ -719,6 +719,9 @@ static void optimize_mb(MACROBLOCK *x, const VP8_ENCODER_RTCD *rtcd)
     tl = (ENTROPY_CONTEXT *)&t_left;
 
     has_2nd_order = (x->e_mbd.mode_info_context->mbmi.mode != B_PRED
+#if CONFIG_I8X8
+        &&x->e_mbd.mode_info_context->mbmi.mode != I8X8_PRED
+#endif
         && x->e_mbd.mode_info_context->mbmi.mode != SPLITMV);
     type = has_2nd_order ? PLANE_TYPE_Y_NO_DC : PLANE_TYPE_Y_WITH_DC;
 
@@ -768,6 +771,9 @@ void vp8_optimize_mby(MACROBLOCK *x, const VP8_ENCODER_RTCD *rtcd)
     tl = (ENTROPY_CONTEXT *)&t_left;
 
     has_2nd_order = (x->e_mbd.mode_info_context->mbmi.mode != B_PRED
+#if CONFIG_I8X8
+        &&x->e_mbd.mode_info_context->mbmi.mode != I8X8_PRED
+#endif
         && x->e_mbd.mode_info_context->mbmi.mode != SPLITMV);
     type = has_2nd_order ? PLANE_TYPE_Y_NO_DC : PLANE_TYPE_Y_WITH_DC;
 
@@ -1082,6 +1088,10 @@ void optimize_mb_8x8(MACROBLOCK *x, const VP8_ENCODER_RTCD *rtcd)
     tl = (ENTROPY_CONTEXT *)&t_left;
 
     has_2nd_order = (x->e_mbd.mode_info_context->mbmi.mode != B_PRED
+#if CONFIG_I8X8
+        &&x->e_mbd.mode_info_context->mbmi.mode != I8X8_PRED
+#endif
+
         && x->e_mbd.mode_info_context->mbmi.mode != SPLITMV);
     type = has_2nd_order ? 0 : 3;
 
@@ -1180,6 +1190,9 @@ void vp8_optimize_mby_8x8(MACROBLOCK *x, const VP8_ENCODER_RTCD *rtcd)
     tl = (ENTROPY_CONTEXT *)&t_left;
 
     has_2nd_order = (x->e_mbd.mode_info_context->mbmi.mode != B_PRED
+#if CONFIG_I8X8
+        && x->e_mbd.mode_info_context->mbmi.mode != I8X8_PRED
+#endif
         && x->e_mbd.mode_info_context->mbmi.mode != SPLITMV);
     type = has_2nd_order ? 0 : 3;
 
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index 265c2dd4c..867c4bdec 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -810,6 +810,9 @@ void vp8_set_speed_features(VP8_COMP *cpi)
         sf->thresh_mult[THR_V_PRED   ] = 1000;
         sf->thresh_mult[THR_H_PRED   ] = 1000;
         sf->thresh_mult[THR_B_PRED   ] = 2000;
+#if CONFIG_I8X8
+        sf->thresh_mult[THR_I8X8_PRED] = 2000;
+#endif
         sf->thresh_mult[THR_TM       ] = 1000;
 
         sf->thresh_mult[THR_NEWMV    ] = 1000;
@@ -833,6 +836,9 @@ void vp8_set_speed_features(VP8_COMP *cpi)
         sf->thresh_mult[THR_V_PRED   ] = 1000;
         sf->thresh_mult[THR_H_PRED   ] = 1000;
         sf->thresh_mult[THR_B_PRED   ] = 2500;
+#if CONFIG_I8X8
+        sf->thresh_mult[THR_I8X8_PRED] = 2500;
+#endif
         sf->thresh_mult[THR_TM       ] = 1000;
 
         sf->thresh_mult[THR_NEARESTG ] = 1000;
@@ -902,7 +908,9 @@ void vp8_set_speed_features(VP8_COMP *cpi)
             sf->thresh_mult[THR_V_PRED   ] = 1500;
             sf->thresh_mult[THR_H_PRED   ] = 1500;
             sf->thresh_mult[THR_B_PRED   ] = 5000;
-
+#if CONFIG_I8X8
+            sf->thresh_mult[THR_I8X8_PRED] = 5000;
+#endif
             if (cpi->ref_frame_flags & VP8_LAST_FLAG)
             {
                 sf->thresh_mult[THR_NEWMV    ] = 2000;
@@ -938,6 +946,9 @@ void vp8_set_speed_features(VP8_COMP *cpi)
             sf->thresh_mult[THR_V_PRED   ] = 2000;
             sf->thresh_mult[THR_H_PRED   ] = 2000;
             sf->thresh_mult[THR_B_PRED   ] = 7500;
+#if CONFIG_I8X8
+            sf->thresh_mult[THR_I8X8_PRED] = 7500;
+#endif
 
             if (cpi->ref_frame_flags & VP8_LAST_FLAG)
             {
@@ -981,6 +992,10 @@ void vp8_set_speed_features(VP8_COMP *cpi)
             cpi->mode_check_freq[THR_V_PRED] = 0;
             cpi->mode_check_freq[THR_H_PRED] = 0;
             cpi->mode_check_freq[THR_B_PRED] = 0;
+#if CONFIG_I8X8
+            cpi->mode_check_freq[THR_I8X8_PRED]   = 0;
+#endif
+
             cpi->mode_check_freq[THR_NEARG] = 0;
             cpi->mode_check_freq[THR_NEWG] = 0;
             cpi->mode_check_freq[THR_NEARA] = 0;
@@ -999,6 +1014,10 @@ void vp8_set_speed_features(VP8_COMP *cpi)
             cpi->mode_check_freq[THR_V_PRED] = 2;
             cpi->mode_check_freq[THR_H_PRED] = 2;
             cpi->mode_check_freq[THR_B_PRED] = 2;
+#if CONFIG_I8X8
+            cpi->mode_check_freq[THR_I8X8_PRED]=2;
+#endif
+
 
             if (cpi->ref_frame_flags & VP8_GOLD_FLAG)
             {
@@ -1044,6 +1063,9 @@ void vp8_set_speed_features(VP8_COMP *cpi)
         sf->thresh_mult[THR_V_PRED   ] = 1000;
         sf->thresh_mult[THR_H_PRED   ] = 1000;
         sf->thresh_mult[THR_B_PRED   ] = 2500;
+#if CONFIG_I8X8
+            sf->thresh_mult[THR_I8X8_PRED] = 2500;
+#endif
         sf->thresh_mult[THR_NEARESTG ] = 1000;
         sf->thresh_mult[THR_ZEROG    ] = 1000;
         sf->thresh_mult[THR_NEARG    ] = 1000;
@@ -1069,6 +1091,9 @@ void vp8_set_speed_features(VP8_COMP *cpi)
             sf->thresh_mult[THR_V_PRED   ] = 2000;
             sf->thresh_mult[THR_H_PRED   ] = 2000;
             sf->thresh_mult[THR_B_PRED   ] = 5000;
+#if CONFIG_I8X8
+            sf->thresh_mult[THR_I8X8_PRED] = 5000;
+#endif
 
             if (cpi->ref_frame_flags & VP8_LAST_FLAG)
             {
@@ -1115,7 +1140,9 @@ void vp8_set_speed_features(VP8_COMP *cpi)
             sf->thresh_mult[THR_V_PRED   ] = 2000;
             sf->thresh_mult[THR_H_PRED   ] = 2000;
             sf->thresh_mult[THR_B_PRED   ] = 5000;
-
+#if CONFIG_I8X8
+            sf->thresh_mult[THR_I8X8_PRED] = 5000;
+#endif
             if (cpi->ref_frame_flags & VP8_LAST_FLAG)
             {
                 sf->thresh_mult[THR_NEWMV    ] = 2000;
@@ -1149,6 +1176,10 @@ void vp8_set_speed_features(VP8_COMP *cpi)
             cpi->mode_check_freq[THR_V_PRED] = 2;
             cpi->mode_check_freq[THR_H_PRED] = 2;
             cpi->mode_check_freq[THR_B_PRED] = 2;
+#if CONFIG_I8X8
+            cpi->mode_check_freq[THR_I8X8_PRED]=2;
+#endif
+
 
             if (cpi->ref_frame_flags & VP8_GOLD_FLAG)
             {
@@ -1187,7 +1218,9 @@ void vp8_set_speed_features(VP8_COMP *cpi)
             cpi->mode_check_freq[THR_V_PRED] = 4;
             cpi->mode_check_freq[THR_H_PRED] = 4;
             cpi->mode_check_freq[THR_B_PRED] = 4;
-
+#if CONFIG_I8X8
+            cpi->mode_check_freq[THR_I8X8_PRED]=4;
+#endif
             if (cpi->ref_frame_flags & VP8_GOLD_FLAG)
             {
                 cpi->mode_check_freq[THR_NEARG] = 2;
@@ -1202,6 +1235,10 @@ void vp8_set_speed_features(VP8_COMP *cpi)
 
             sf->thresh_mult[THR_TM       ] = 2000;
             sf->thresh_mult[THR_B_PRED   ] = 5000;
+#if CONFIG_I8X8
+            sf->thresh_mult[THR_I8X8_PRED] = 5000;
+#endif
+
 
             if (cpi->ref_frame_flags & VP8_GOLD_FLAG)
             {
@@ -1224,6 +1261,9 @@ void vp8_set_speed_features(VP8_COMP *cpi)
         {
             // Disable split MB intra prediction mode
             sf->thresh_mult[THR_B_PRED] = INT_MAX;
+#if CONFIG_I8X8
+            sf->thresh_mult[THR_I8X8_PRED] = INT_MAX;
+#endif
         }
 
         if (Speed > 6)
@@ -4473,6 +4513,8 @@ static void encode_frame_to_data_rate
 #if WRITE_RECON_BUFFER
     if(cm->show_frame)
         write_cx_frame_to_file(cm->frame_to_show, cm->current_video_frame);
+    else
+        write_cx_frame_to_file(cm->frame_to_show, cm->current_video_frame+1000);
 #endif
 
 #if CONFIG_MULTITHREAD
diff --git a/vp8/encoder/onyx_int.h b/vp8/encoder/onyx_int.h
index 622f25068..160f22d05 100644
--- a/vp8/encoder/onyx_int.h
+++ b/vp8/encoder/onyx_int.h
@@ -42,8 +42,11 @@
 #define AF_THRESH   25
 #define AF_THRESH2  100
 #define ARF_DECAY_THRESH 12
+#if CONFIG_I8X8
+#define MAX_MODES 21
+#else
 #define MAX_MODES 20
-
+#endif
 #define MIN_THRESHMULT  32
 #define MAX_THRESHMULT  512
 
@@ -74,7 +77,7 @@ typedef struct
     int uv_modes[VP8_UV_MODES];
     int i8x8_modes[VP8_I8X8_MODES];
     int b_modes[10];
-    int inter_y_modes[10];
+    int inter_y_modes[MB_MODE_COUNT];
     int inter_uv_modes[VP8_UV_MODES];
     int inter_b_modes[10];
 #endif
@@ -186,6 +189,9 @@ typedef enum
     THR_SPLITA         = 18,
 
     THR_B_PRED         = 19,
+#if CONFIG_I8X8
+    THR_I8X8_PRED      = 20,
+#endif
 }
 THR_MODES;
 
diff --git a/vp8/encoder/quantize.c b/vp8/encoder/quantize.c
index 2e2a8dcbf..78f7dd4c6 100644
--- a/vp8/encoder/quantize.c
+++ b/vp8/encoder/quantize.c
@@ -286,6 +286,9 @@ void vp8_quantize_mby_c(MACROBLOCK *x)
 {
     int i;
     int has_2nd_order = (x->e_mbd.mode_info_context->mbmi.mode != B_PRED
+#if CONFIG_I8X8
+        && x->e_mbd.mode_info_context->mbmi.mode != I8X8_PRED
+#endif
         && x->e_mbd.mode_info_context->mbmi.mode != SPLITMV);
 
     for (i = 0; i < 16; i++)
@@ -299,6 +302,9 @@ void vp8_quantize_mb_c(MACROBLOCK *x)
 {
     int i;
     int has_2nd_order=(x->e_mbd.mode_info_context->mbmi.mode != B_PRED
+#if CONFIG_I8X8
+        && x->e_mbd.mode_info_context->mbmi.mode != I8X8_PRED
+#endif
         && x->e_mbd.mode_info_context->mbmi.mode != SPLITMV);
 
     for (i = 0; i < 24+has_2nd_order; i++)
diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c
index ff0018be5..00a395845 100644
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -103,6 +103,9 @@ const MB_PREDICTION_MODE vp8_mode_order[MAX_MODES] =
     SPLITMV,
 
     B_PRED,
+#if CONFIG_I8X8
+    I8X8_PRED,
+#endif
 };
 
 const MV_REFERENCE_FRAME vp8_ref_frame_order[MAX_MODES] =
@@ -135,6 +138,9 @@ const MV_REFERENCE_FRAME vp8_ref_frame_order[MAX_MODES] =
     ALTREF_FRAME,
 
     INTRA_FRAME,
+#if CONFIG_I8X8
+    INTRA_FRAME,
+#endif
 };
 
 static void fill_token_costs(
@@ -1941,6 +1947,29 @@ static void rd_update_mvcount(VP8_COMP *cpi, MACROBLOCK *x, int_mv *best_ref_mv)
     }
 }
 
+#if CONFIG_I8X8
+static void set_i8x8_block_modes(MACROBLOCK *x, int *modes)
+{
+    int i;
+    MACROBLOCKD *xd = &x->e_mbd;
+    for(i=0;i<4;i++)
+    {
+        int ib = vp8_i8x8_block[i];
+        x->e_mbd.mode_info_context->bmi[ib+0].as_mode= modes[i];
+        x->e_mbd.mode_info_context->bmi[ib+1].as_mode= modes[i];
+        x->e_mbd.mode_info_context->bmi[ib+4].as_mode= modes[i];
+        x->e_mbd.mode_info_context->bmi[ib+5].as_mode= modes[i];
+    }
+
+    for (i = 0; i < 16; i++)
+    {
+        xd->block[i].bmi = xd->mode_info_context->bmi[i];
+    }
+}
+#endif
+
+
+
 void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int recon_uvoffset, int *returnrate, int *returndistortion, int *returnintra)
 {
     BLOCK *b = &x->block[0];
@@ -1955,6 +1984,9 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int
     MB_PREDICTION_MODE this_mode;
     int num00;
     int best_mode_index = 0;
+#if CONFIG_I8X8
+    int mode8x8[4];
+#endif
     unsigned char segment_id = xd->mode_info_context->mbmi.segment_id;
 
     int i;
@@ -2182,6 +2214,37 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int
             }
         }
         break;
+#if CONFIG_I8X8
+        case I8X8_PRED:
+        {
+            int tmp_rd;
+            tmp_rd = rd_pick_intra8x8mby_modes(cpi,
+                            x, &rate, &rate_y, &distortion, best_yrd);
+            rate2 += rate;
+            distortion2 += distortion;
+
+            mode8x8[0]= x->e_mbd.mode_info_context->bmi[0].as_mode;
+            mode8x8[1]= x->e_mbd.mode_info_context->bmi[2].as_mode;
+            mode8x8[2]= x->e_mbd.mode_info_context->bmi[8].as_mode;
+            mode8x8[3]= x->e_mbd.mode_info_context->bmi[10].as_mode;
+
+            /* TODO: uv rate maybe over-estimated here since there is UV intra
+                     mode coded in I8X8_PRED prediction */
+            if(tmp_rd < best_yrd)
+            {
+                rate2 += uv_intra_rate;
+                rate_uv = uv_intra_rate_tokenonly;
+                distortion2 += uv_intra_distortion;
+                distortion_uv = uv_intra_distortion;
+            }
+            else
+            {
+                this_rd = INT_MAX;
+                disable_skip = 1;
+            }
+        }
+        break;
+#endif
 
         case SPLITMV:
         {
@@ -2552,7 +2615,11 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int
             vpx_memcpy(&best_mbmode, &x->e_mbd.mode_info_context->mbmi, sizeof(MB_MODE_INFO));
             vpx_memcpy(&best_partition, x->partition_info, sizeof(PARTITION_INFO));
 
-            if ((this_mode == B_PRED) || (this_mode == SPLITMV))
+            if ((this_mode == B_PRED)
+#if CONFIG_I8X8
+                ||(this_mode == I8X8_PRED)
+#endif
+                || (this_mode == SPLITMV))
                 for (i = 0; i < 16; i++)
                 {
                     best_bmodes[i] = x->e_mbd.block[i].bmi;
@@ -2637,9 +2704,19 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int
     if (best_mbmode.mode == B_PRED)
     {
         for (i = 0; i < 16; i++)
+        {
             xd->mode_info_context->bmi[i].as_mode = best_bmodes[i].as_mode;
+            xd->block[i].bmi.as_mode = xd->mode_info_context->bmi[i].as_mode;
+        }
     }
 
+#if CONFIG_I8X8
+    if (best_mbmode.mode == I8X8_PRED)
+    {
+        set_i8x8_block_modes(x, mode8x8);
+    }
+#endif
+
     if (best_mbmode.mode == SPLITMV)
     {
         for (i = 0; i < 16; i++)
@@ -2657,27 +2734,6 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int
 
 }
 
-#if CONFIG_I8X8
-static void set_i8x8_block_modes(MACROBLOCK *x, int *modes)
-{
-    int i;
-    MACROBLOCKD *xd = &x->e_mbd;
-    for(i=0;i<4;i++)
-    {
-        int ib = vp8_i8x8_block[i];
-        x->e_mbd.mode_info_context->bmi[ib+0].as_mode= modes[i];
-        x->e_mbd.mode_info_context->bmi[ib+1].as_mode= modes[i];
-        x->e_mbd.mode_info_context->bmi[ib+4].as_mode= modes[i];
-        x->e_mbd.mode_info_context->bmi[ib+5].as_mode= modes[i];
-    }
-
-    for (i = 0; i < 16; i++)
-    {
-        xd->block[i].bmi = xd->mode_info_context->bmi[i];
-    }
-}
-#endif
-
 void vp8_rd_pick_intra_mode(VP8_COMP *cpi, MACROBLOCK *x, int *rate_)
 {
     MACROBLOCKD *xd = &x->e_mbd;
diff --git a/vp8/encoder/tokenize.c b/vp8/encoder/tokenize.c
index 304bb4f98..0f8525e20 100644
--- a/vp8/encoder/tokenize.c
+++ b/vp8/encoder/tokenize.c
@@ -1028,7 +1028,11 @@ void vp8_stuff_mb(VP8_COMP *cpi, MACROBLOCKD *x, TOKENEXTRA **t)
 void vp8_fix_contexts(MACROBLOCKD *x)
 {
     /* Clear entropy contexts for Y2 blocks */
-    if (x->mode_info_context->mbmi.mode != B_PRED && x->mode_info_context->mbmi.mode != SPLITMV)
+    if (x->mode_info_context->mbmi.mode != B_PRED
+#if CONFIG_I8X8
+        && x->mode_info_context->mbmi.mode != I8X8_PRED
+#endif
+        && x->mode_info_context->mbmi.mode != SPLITMV)
     {
         vpx_memset(x->above_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));
         vpx_memset(x->left_context, 0, sizeof(ENTROPY_CONTEXT_PLANES));