Merge remote branch 'internal/upstream-experimental' into HEAD

Conflicts:
	vp8/decoder/detokenize.c
	vp8/decoder/onyxd_if.c
	vp8/vp8_common.mk

Change-Id: Ifca1108186a8bc715da86a44021ee2fa5550b5b8

25 files changed, 232 insertions, 220 deletions

diff --git a/vp8/common/asm_com_offsets.c b/vp8/common/asm_com_offsets.c
index e167d2654..e135a4d78 100644
--- a/vp8/common/asm_com_offsets.c
+++ b/vp8/common/asm_com_offsets.c
@@ -9,6 +9,8 @@
  */
 
 
+#include "vpx_config.h"
+#include "vpx/vpx_codec.h"
 #include "vpx_ports/asm_offsets.h"
 #include "vpx_scale/yv12config.h"
 
@@ -25,8 +27,14 @@ DEFINE(yv12_buffer_config_y_buffer,             offsetof(YV12_BUFFER_CONFIG, y_b
 DEFINE(yv12_buffer_config_u_buffer,             offsetof(YV12_BUFFER_CONFIG, u_buffer));
 DEFINE(yv12_buffer_config_v_buffer,             offsetof(YV12_BUFFER_CONFIG, v_buffer));
 DEFINE(yv12_buffer_config_border,               offsetof(YV12_BUFFER_CONFIG, border));
+DEFINE(VP8BORDERINPIXELS_VAL,                   VP8BORDERINPIXELS);
 
 END
 
 /* add asserts for any offset that is not supported by assembly code */
 /* add asserts for any size that is not supported by assembly code */
+
+#if HAVE_ARMV7
+/* vp8_yv12_extend_frame_borders_neon makes several assumptions based on this */
+ct_assert(VP8BORDERINPIXELS_VAL, VP8BORDERINPIXELS == 32)
+#endif
diff --git a/vp8/common/onyx.h b/vp8/common/onyx.h
index 545798ac7..015b4c4d4 100644
--- a/vp8/common/onyx.h
+++ b/vp8/common/onyx.h
@@ -211,8 +211,8 @@ extern "C"
 
 // receive a frames worth of data caller can assume that a copy of this frame is made
 // and not just a copy of the pointer..
-    int vp8_receive_raw_frame(VP8_PTR comp, unsigned int frame_flags, YV12_BUFFER_CONFIG *sd, INT64 time_stamp, INT64 end_time_stamp);
-    int vp8_get_compressed_data(VP8_PTR comp, unsigned int *frame_flags, unsigned long *size, unsigned char *dest, INT64 *time_stamp, INT64 *time_end, int flush);
+    int vp8_receive_raw_frame(VP8_PTR comp, unsigned int frame_flags, YV12_BUFFER_CONFIG *sd, int64_t time_stamp, int64_t end_time_stamp);
+    int vp8_get_compressed_data(VP8_PTR comp, unsigned int *frame_flags, unsigned long *size, unsigned char *dest, int64_t *time_stamp, int64_t *time_end, int flush);
     int vp8_get_preview_raw_frame(VP8_PTR comp, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t *flags);
 
     int vp8_use_as_reference(VP8_PTR comp, int ref_frame_flags);
diff --git a/vp8/common/onyxc_int.h b/vp8/common/onyxc_int.h
index 77e98450c..f84bd6e4a 100644
--- a/vp8/common/onyxc_int.h
+++ b/vp8/common/onyxc_int.h
@@ -19,7 +19,9 @@
 #include "entropy.h"
 #include "idct.h"
 #include "recon.h"
+#if CONFIG_POSTPROC
 #include "postproc.h"
+#endif
 
 /*#ifdef PACKET_TESTING*/
 #include "header.h"
@@ -78,7 +80,9 @@ typedef struct VP8_COMMON_RTCD
     vp8_recon_rtcd_vtable_t       recon;
     vp8_subpix_rtcd_vtable_t      subpix;
     vp8_loopfilter_rtcd_vtable_t  loopfilter;
+#if CONFIG_POSTPROC
     vp8_postproc_rtcd_vtable_t    postproc;
+#endif
     int                           flags;
 #else
     int unused;
@@ -205,7 +209,9 @@ typedef struct VP8Common
 #if CONFIG_MULTITHREAD
     int processor_core_count;
 #endif
+#if CONFIG_POSTPROC
     struct postproc_state  postproc_state;
+#endif
 } VP8_COMMON;
 
 #endif
diff --git a/vp8/common/onyxd.h b/vp8/common/onyxd.h
index 2dfdc66da..d3e5c2fa7 100644
--- a/vp8/common/onyxd.h
+++ b/vp8/common/onyxd.h
@@ -54,8 +54,8 @@ extern "C"
 
     int vp8dx_get_setting(VP8D_PTR comp, VP8D_SETTING oxst);
 
-    int vp8dx_receive_compressed_data(VP8D_PTR comp, unsigned long size, const unsigned char *dest, INT64 time_stamp);
-    int vp8dx_get_raw_frame(VP8D_PTR comp, YV12_BUFFER_CONFIG *sd, INT64 *time_stamp, INT64 *time_end_stamp, vp8_ppflags_t *flags);
+    int vp8dx_receive_compressed_data(VP8D_PTR comp, unsigned long size, const unsigned char *dest, int64_t time_stamp);
+    int vp8dx_get_raw_frame(VP8D_PTR comp, YV12_BUFFER_CONFIG *sd, int64_t *time_stamp, int64_t *time_end_stamp, vp8_ppflags_t *flags);
 
     vpx_codec_err_t vp8dx_get_reference(VP8D_PTR comp, VP8_REFFRAME ref_frame_flag, YV12_BUFFER_CONFIG *sd);
     vpx_codec_err_t vp8dx_set_reference(VP8D_PTR comp, VP8_REFFRAME ref_frame_flag, YV12_BUFFER_CONFIG *sd);
diff --git a/vp8/decoder/arm/arm_dsystemdependent.c b/vp8/decoder/arm/arm_dsystemdependent.c
index 79e2e1ba9..6ce471217 100644
--- a/vp8/decoder/arm/arm_dsystemdependent.c
+++ b/vp8/decoder/arm/arm_dsystemdependent.c
@@ -13,7 +13,6 @@
 #include "vpx_ports/arm.h"
 #include "vp8/common/blockd.h"
 #include "vp8/common/pragmas.h"
-#include "vp8/common/postproc.h"
 #include "vp8/decoder/dequantize.h"
 #include "vp8/decoder/onyxd_int.h"
 
diff --git a/vp8/decoder/detokenize.c b/vp8/decoder/detokenize.c
index 025f9dfe4..f8a6e7eb2 100644
--- a/vp8/decoder/detokenize.c
+++ b/vp8/decoder/detokenize.c
@@ -83,7 +83,6 @@ static const unsigned char cat6_prob[12] =
 { 129, 130, 133, 140, 153, 177, 196, 230, 243, 254, 254, 0 };
 #endif
 
-
 void vp8_reset_mb_tokens_context(MACROBLOCKD *x)
 {
     /* Clear entropy contexts for Y2 blocks */
diff --git a/vp8/decoder/onyxd_if.c b/vp8/decoder/onyxd_if.c
index 7c84b6efb..71f323282 100644
--- a/vp8/decoder/onyxd_if.c
+++ b/vp8/decoder/onyxd_if.c
@@ -244,8 +244,8 @@ vpx_codec_err_t vp8dx_set_reference(VP8D_PTR ptr, VP8_REFFRAME ref_frame_flag, Y
 
 /*For ARM NEON, d8-d15 are callee-saved registers, and need to be saved by us.*/
 #if HAVE_ARMV7
-extern void vp8_push_neon(INT64 *store);
-extern void vp8_pop_neon(INT64 *store);
+extern void vp8_push_neon(int64_t *store);
+extern void vp8_pop_neon(int64_t *store);
 #endif
 
 static int get_free_fb (VP8_COMMON *cm)
@@ -344,10 +344,10 @@ static void vp8_print_yuv_rec_mb(VP8_COMMON *cm, int mb_row, int mb_col)
 }
 */
 
-int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsigned char *source, INT64 time_stamp)
+int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsigned char *source, int64_t time_stamp)
 {
 #if HAVE_ARMV7
-    INT64 dx_store_reg[8];
+    int64_t dx_store_reg[8];
 #endif
     VP8D_COMP *pbi = (VP8D_COMP *) ptr;
     VP8_COMMON *cm = &pbi->common;
@@ -554,9 +554,9 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign
 #if 0
     {
         int i;
-        INT64 earliest_time = pbi->dr[0].time_stamp;
-        INT64 latest_time = pbi->dr[0].time_stamp;
-        INT64 time_diff = 0;
+        int64_t earliest_time = pbi->dr[0].time_stamp;
+        int64_t latest_time = pbi->dr[0].time_stamp;
+        int64_t time_diff = 0;
         int bytes = 0;
 
         pbi->dr[pbi->common.current_video_frame&0xf].size = pbi->bc.pos + pbi->bc2.pos + 4;;
@@ -596,7 +596,7 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign
     pbi->common.error.setjmp = 0;
     return retcode;
 }
-int vp8dx_get_raw_frame(VP8D_PTR ptr, YV12_BUFFER_CONFIG *sd, INT64 *time_stamp, INT64 *time_end_stamp, vp8_ppflags_t *flags)
+int vp8dx_get_raw_frame(VP8D_PTR ptr, YV12_BUFFER_CONFIG *sd, int64_t *time_stamp, int64_t *time_end_stamp, vp8_ppflags_t *flags)
 {
     int ret = -1;
     VP8D_COMP *pbi = (VP8D_COMP *) ptr;
diff --git a/vp8/decoder/onyxd_int.h b/vp8/decoder/onyxd_int.h
index bb05ea826..a7d8b5a35 100644
--- a/vp8/decoder/onyxd_int.h
+++ b/vp8/decoder/onyxd_int.h
@@ -40,7 +40,7 @@ typedef struct
 
 typedef struct
 {
-    INT64 time_stamp;
+    int64_t time_stamp;
     int size;
 } DATARATE;
 
@@ -121,7 +121,7 @@ typedef struct VP8Decompressor
 #endif
 
     vp8_reader *mbc;
-    INT64 last_time_stamp;
+    int64_t last_time_stamp;
     int   ready_for_new_data;
 
     DATARATE dr[16];
diff --git a/vp8/encoder/encodeframe.c b/vp8/encoder/encodeframe.c
index ba5409754..1c4a93636 100644
--- a/vp8/encoder/encodeframe.c
+++ b/vp8/encoder/encodeframe.c
@@ -336,7 +336,7 @@ static unsigned int mb_activity_measure( VP8_COMP *cpi, MACROBLOCK *x,
 
 // Calculate an "average" mb activity value for the frame
 #define ACT_MEDIAN 0
-static void calc_av_activity( VP8_COMP *cpi, INT64 activity_sum )
+static void calc_av_activity( VP8_COMP *cpi, int64_t activity_sum )
 {
 #if ACT_MEDIAN
     // Find median: Simple n^2 algorithm for experimentation
@@ -404,9 +404,9 @@ static void calc_activity_index( VP8_COMP *cpi, MACROBLOCK *x )
     VP8_COMMON *const cm = & cpi->common;
     int mb_row, mb_col;
 
-    INT64 act;
-    INT64 a;
-    INT64 b;
+    int64_t act;
+    int64_t a;
+    int64_t b;
 
 #if OUTPUT_NORM_ACT_STATS
     FILE *f = fopen("norm_act.stt", "a");
@@ -470,7 +470,7 @@ static void build_activity_map( VP8_COMP *cpi )
 
     int mb_row, mb_col;
     unsigned int mb_activity;
-    INT64 activity_sum = 0;
+    int64_t activity_sum = 0;
 
     // for each macroblock row in image
     for (mb_row = 0; mb_row < cm->mb_rows; mb_row++)
@@ -537,15 +537,15 @@ void vp8_activity_masking(VP8_COMP *cpi, MACROBLOCK *x)
     x->errorperbit = x->rdmult * 100 /(110 * x->rddiv);
     x->errorperbit += (x->errorperbit==0);
 #else
-    INT64 a;
-    INT64 b;
-    INT64 act = *(x->mb_activity_ptr);
+    int64_t a;
+    int64_t b;
+    int64_t act = *(x->mb_activity_ptr);
 
     // Apply the masking to the RD multiplier.
     a = act + (2*cpi->activity_avg);
     b = (2*act) + cpi->activity_avg;
 
-    x->rdmult = (unsigned int)(((INT64)x->rdmult*b + (a>>1))/a);
+    x->rdmult = (unsigned int)(((int64_t)x->rdmult*b + (a>>1))/a);
     x->errorperbit = x->rdmult * 100 /(110 * x->rddiv);
     x->errorperbit += (x->errorperbit==0);
 #endif
@@ -1444,18 +1444,18 @@ static void adjust_act_zbin( VP8_COMP *cpi, MACROBLOCK *x )
 #if USE_ACT_INDEX
     x->act_zbin_adj = *(x->mb_activity_ptr);
 #else
-    INT64 a;
-    INT64 b;
-    INT64 act = *(x->mb_activity_ptr);
+    int64_t a;
+    int64_t b;
+    int64_t act = *(x->mb_activity_ptr);
 
     // Apply the masking to the RD multiplier.
     a = act + 4*cpi->activity_avg;
     b = 4*act + cpi->activity_avg;
 
     if ( act > cpi->activity_avg )
-        x->act_zbin_adj = (int)(((INT64)b + (a>>1))/a) - 1;
+        x->act_zbin_adj = (int)(((int64_t)b + (a>>1))/a) - 1;
     else
-        x->act_zbin_adj = 1 - (int)(((INT64)a + (b>>1))/b);
+        x->act_zbin_adj = 1 - (int)(((int64_t)a + (b>>1))/b);
 #endif
 }
 
diff --git a/vp8/encoder/firstpass.c b/vp8/encoder/firstpass.c
index 5422365f8..b063e45af 100644
--- a/vp8/encoder/firstpass.c
+++ b/vp8/encoder/firstpass.c
@@ -530,8 +530,8 @@ void vp8_first_pass(VP8_COMP *cpi)
     YV12_BUFFER_CONFIG *gld_yv12 = &cm->yv12_fb[cm->gld_fb_idx];
     int recon_y_stride = lst_yv12->y_stride;
     int recon_uv_stride = lst_yv12->uv_stride;
-    long long intra_error = 0;
-    long long coded_error = 0;
+    int64_t intra_error = 0;
+    int64_t coded_error = 0;
 
     int sum_mvr = 0, sum_mvc = 0;
     int sum_mvr_abs = 0, sum_mvc_abs = 0;
@@ -620,7 +620,7 @@ void vp8_first_pass(VP8_COMP *cpi)
             this_error += intrapenalty;
 
             // Cumulative intra error total
-            intra_error += (long long)this_error;
+            intra_error += (int64_t)this_error;
 
             // Set up limit values for motion vectors to prevent them extending outside the UMV borders
             x->mv_col_min = -((mb_col * 16) + (VP8BORDERINPIXELS - 16));
@@ -757,7 +757,7 @@ void vp8_first_pass(VP8_COMP *cpi)
                 }
             }
 
-            coded_error += (long long)this_error;
+            coded_error += (int64_t)this_error;
 
             // adjust to the next column of macroblocks
             x->src.y_buffer += 16;
@@ -1219,8 +1219,8 @@ void vp8_init_second_pass(VP8_COMP *cpi)
     cpi->twopass.total_coded_error_left = cpi->twopass.total_stats->coded_error;
     cpi->twopass.start_tot_err_left = cpi->twopass.total_error_left;
 
-    //cpi->twopass.bits_left = (long long)(cpi->twopass.total_stats->count * cpi->oxcf.target_bandwidth / DOUBLE_DIVIDE_CHECK((double)cpi->oxcf.frame_rate));
-    //cpi->twopass.bits_left -= (long long)(cpi->twopass.total_stats->count * two_pass_min_rate / DOUBLE_DIVIDE_CHECK((double)cpi->oxcf.frame_rate));
+    //cpi->twopass.bits_left = (int64_t)(cpi->twopass.total_stats->count * cpi->oxcf.target_bandwidth / DOUBLE_DIVIDE_CHECK((double)cpi->oxcf.frame_rate));
+    //cpi->twopass.bits_left -= (int64_t)(cpi->twopass.total_stats->count * two_pass_min_rate / DOUBLE_DIVIDE_CHECK((double)cpi->oxcf.frame_rate));
 
     // each frame can have a different duration, as the frame rate in the source
     // isn't guaranteed to be constant.   The frame rate prior to the first frame
@@ -1230,8 +1230,8 @@ void vp8_init_second_pass(VP8_COMP *cpi)
     vp8_new_frame_rate(cpi, 10000000.0 * cpi->twopass.total_stats->count / cpi->twopass.total_stats->duration);
 
     cpi->output_frame_rate = cpi->oxcf.frame_rate;
-    cpi->twopass.bits_left = (long long)(cpi->twopass.total_stats->duration * cpi->oxcf.target_bandwidth / 10000000.0) ;
-    cpi->twopass.bits_left -= (long long)(cpi->twopass.total_stats->duration * two_pass_min_rate / 10000000.0);
+    cpi->twopass.bits_left = (int64_t)(cpi->twopass.total_stats->duration * cpi->oxcf.target_bandwidth / 10000000.0) ;
+    cpi->twopass.bits_left -= (int64_t)(cpi->twopass.total_stats->duration * two_pass_min_rate / 10000000.0);
     cpi->twopass.clip_bits_total = cpi->twopass.bits_left;
 
     // Calculate a minimum intra value to be used in determining the IIratio
@@ -2671,16 +2671,16 @@ static void find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame)
         int max_bits = frame_max_bits(cpi);
 
         // Maximum bits for the kf group
-        long long max_grp_bits;
+        int64_t max_grp_bits;
 
         // Default allocation based on bits left and relative
         // complexity of the section
-        cpi->twopass.kf_group_bits = (long long)( cpi->twopass.bits_left *
+        cpi->twopass.kf_group_bits = (int64_t)( cpi->twopass.bits_left *
                                           ( kf_group_err /
                                             cpi->twopass.modified_error_left ));
 
         // Clip based on maximum per frame rate defined by the user.
-        max_grp_bits = (long long)max_bits * (long long)cpi->twopass.frames_to_key;
+        max_grp_bits = (int64_t)max_bits * (int64_t)cpi->twopass.frames_to_key;
         if (cpi->twopass.kf_group_bits > max_grp_bits)
             cpi->twopass.kf_group_bits = max_grp_bits;
 
@@ -2697,19 +2697,19 @@ static void find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame)
                 int high_water_mark = (opt_buffer_lvl +
                                        cpi->oxcf.maximum_buffer_size) >> 1;
 
-                long long av_group_bits;
+                int64_t av_group_bits;
 
                 // Av bits per frame * number of frames
-                av_group_bits = (long long)cpi->av_per_frame_bandwidth *
-                                (long long)cpi->twopass.frames_to_key;
+                av_group_bits = (int64_t)cpi->av_per_frame_bandwidth *
+                                (int64_t)cpi->twopass.frames_to_key;
 
                 // We are at or above the maximum.
                 if (cpi->buffer_level >= high_water_mark)
                 {
-                    long long min_group_bits;
+                    int64_t min_group_bits;
 
                     min_group_bits = av_group_bits +
-                                     (long long)(buffer_lvl -
+                                     (int64_t)(buffer_lvl -
                                                  high_water_mark);
 
                     if (cpi->twopass.kf_group_bits < min_group_bits)
@@ -2718,11 +2718,11 @@ static void find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame)
                 // We are above optimal but below the maximum
                 else if (cpi->twopass.kf_group_bits < av_group_bits)
                 {
-                    long long bits_below_av = av_group_bits -
+                    int64_t bits_below_av = av_group_bits -
                                               cpi->twopass.kf_group_bits;
 
                     cpi->twopass.kf_group_bits +=
-                       (long long)((double)bits_below_av *
+                       (int64_t)((double)bits_below_av *
                                    (double)(buffer_lvl - opt_buffer_lvl) /
                                    (double)(high_water_mark - opt_buffer_lvl));
                 }
@@ -3043,8 +3043,8 @@ static void find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame)
         }
         else
         {
-            long long clip_bits = (long long)(cpi->twopass.total_stats->count * cpi->oxcf.target_bandwidth / DOUBLE_DIVIDE_CHECK((double)cpi->oxcf.frame_rate));
-            long long over_spend = cpi->oxcf.starting_buffer_level - cpi->buffer_level;
+            int64_t clip_bits = (int64_t)(cpi->twopass.total_stats->count * cpi->oxcf.target_bandwidth / DOUBLE_DIVIDE_CHECK((double)cpi->oxcf.frame_rate));
+            int64_t over_spend = cpi->oxcf.starting_buffer_level - cpi->buffer_level;
 
             if ((last_kf_resampled && (kf_q > cpi->worst_quality)) ||                                               // If triggered last time the threshold for triggering again is reduced
                 ((kf_q > cpi->worst_quality) &&                                                                  // Projected Q higher than allowed and ...
diff --git a/vp8/encoder/mcomp.c b/vp8/encoder/mcomp.c
index 58b524f82..aebcfce70 100644
--- a/vp8/encoder/mcomp.c
+++ b/vp8/encoder/mcomp.c
@@ -15,6 +15,7 @@
 #include <stdio.h>
 #include <limits.h>
 #include <math.h>
+#include "vp8/common/findnearmv.h"
 
 #ifdef ENTROPY_STATS
 static int mv_ref_ct [31] [4] [2];
@@ -342,12 +343,26 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     int bestmse = INT_MAX;
     int_mv startmv;
     int_mv this_mv;
-    unsigned char *y = *(d->base_pre) + d->pre + (bestmv->as_mv.row) * d->pre_stride + bestmv->as_mv.col;
     unsigned char *z = (*(b->base_src) + b->src);
     int left, right, up, down, diag;
     unsigned int sse;
     int whichdir ;
     int thismse;
+    int y_stride;
+
+#if ARCH_X86 || ARCH_X86_64
+    MACROBLOCKD *xd = &x->e_mbd;
+    unsigned char *y0 = *(d->base_pre) + d->pre + (bestmv->as_mv.row) * d->pre_stride + bestmv->as_mv.col;
+    unsigned char *y;
+
+    y_stride = 32;
+    /* Copy 18 rows x 32 cols area to intermediate buffer before searching. */
+     vfp->copymem(y0 - 1 - d->pre_stride, d->pre_stride, xd->y_buf, y_stride, 18);
+     y = xd->y_buf + y_stride + 1;
+#else
+     unsigned char *y = *(d->base_pre) + d->pre + (bestmv->as_mv.row) * d->pre_stride + bestmv->as_mv.col;
+     y_stride = d->pre_stride;
+#endif
 
     // central mv
     bestmv->as_mv.row <<= 3;
@@ -355,14 +370,14 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     startmv = *bestmv;
 
     // calculate central point error
-    bestmse = vfp->vf(y, d->pre_stride, z, b->src_stride, sse1);
+    bestmse = vfp->vf(y, y_stride, z, b->src_stride, sse1);
     *distortion = bestmse;
     bestmse += mv_err_cost(bestmv, ref_mv, mvcost, error_per_bit);
 
     // go left then right and check error
     this_mv.as_mv.row = startmv.as_mv.row;
     this_mv.as_mv.col = ((startmv.as_mv.col - 8) | 4);
-    thismse = vfp->svf_halfpix_h(y - 1, d->pre_stride, z, b->src_stride, &sse);
+    thismse = vfp->svf_halfpix_h(y - 1, y_stride, z, b->src_stride, &sse);
     left = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (left < bestmse)
@@ -374,7 +389,7 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     }
 
     this_mv.as_mv.col += 8;
-    thismse = vfp->svf_halfpix_h(y, d->pre_stride, z, b->src_stride, &sse);
+    thismse = vfp->svf_halfpix_h(y, y_stride, z, b->src_stride, &sse);
     right = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (right < bestmse)
@@ -388,7 +403,7 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     // go up then down and check error
     this_mv.as_mv.col = startmv.as_mv.col;
     this_mv.as_mv.row = ((startmv.as_mv.row - 8) | 4);
-    thismse =  vfp->svf_halfpix_v(y - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
+    thismse =  vfp->svf_halfpix_v(y - y_stride, y_stride, z, b->src_stride, &sse);
     up = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (up < bestmse)
@@ -400,7 +415,7 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     }
 
     this_mv.as_mv.row += 8;
-    thismse = vfp->svf_halfpix_v(y, d->pre_stride, z, b->src_stride, &sse);
+    thismse = vfp->svf_halfpix_v(y, y_stride, z, b->src_stride, &sse);
     down = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (down < bestmse)
@@ -423,23 +438,23 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     case 0:
         this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
         this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
-        thismse = vfp->svf_halfpix_hv(y - 1 - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
+        thismse = vfp->svf_halfpix_hv(y - 1 - y_stride, y_stride, z, b->src_stride, &sse);
         break;
     case 1:
         this_mv.as_mv.col += 4;
         this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
-        thismse = vfp->svf_halfpix_hv(y - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
+        thismse = vfp->svf_halfpix_hv(y - y_stride, y_stride, z, b->src_stride, &sse);
         break;
     case 2:
         this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
         this_mv.as_mv.row += 4;
-        thismse = vfp->svf_halfpix_hv(y - 1, d->pre_stride, z, b->src_stride, &sse);
+        thismse = vfp->svf_halfpix_hv(y - 1, y_stride, z, b->src_stride, &sse);
         break;
     case 3:
     default:
         this_mv.as_mv.col += 4;
         this_mv.as_mv.row += 4;
-        thismse = vfp->svf_halfpix_hv(y, d->pre_stride, z, b->src_stride, &sse);
+        thismse = vfp->svf_halfpix_hv(y, y_stride, z, b->src_stride, &sse);
         break;
     }
 
@@ -458,7 +473,7 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
 
     // time to check quarter pels.
     if (bestmv->as_mv.row < startmv.as_mv.row)
-        y -= d->pre_stride;
+        y -= y_stride;
 
     if (bestmv->as_mv.col < startmv.as_mv.col)
         y--;
@@ -473,12 +488,12 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     if (startmv.as_mv.col & 7)
     {
         this_mv.as_mv.col = startmv.as_mv.col - 2;
-        thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+        thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
     }
     else
     {
         this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
-        thismse = vfp->svf(y - 1, d->pre_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+        thismse = vfp->svf(y - 1, y_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
     }
 
     left = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
@@ -492,7 +507,7 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     }
 
     this_mv.as_mv.col += 4;
-    thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+    thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
     right = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (right < bestmse)
@@ -509,12 +524,12 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     if (startmv.as_mv.row & 7)
     {
         this_mv.as_mv.row = startmv.as_mv.row - 2;
-        thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+        thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
     }
     else
     {
         this_mv.as_mv.row = (startmv.as_mv.row - 8) | 6;
-        thismse = vfp->svf(y - d->pre_stride, d->pre_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
+        thismse = vfp->svf(y - y_stride, y_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
     }
 
     up = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
@@ -528,7 +543,7 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     }
 
     this_mv.as_mv.row += 4;
-    thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+    thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
     down = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (down < bestmse)
@@ -558,12 +573,12 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
             if (startmv.as_mv.col & 7)
             {
                 this_mv.as_mv.col -= 2;
-                thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+                thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
             }
             else
             {
                 this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
-                thismse = vfp->svf(y - 1, d->pre_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);;
+                thismse = vfp->svf(y - 1, y_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);;
             }
         }
         else
@@ -573,12 +588,12 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
             if (startmv.as_mv.col & 7)
             {
                 this_mv.as_mv.col -= 2;
-                thismse = vfp->svf(y - d->pre_stride, d->pre_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
+                thismse = vfp->svf(y - y_stride, y_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
             }
             else
             {
                 this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
-                thismse = vfp->svf(y - d->pre_stride - 1, d->pre_stride, 6, 6, z, b->src_stride, &sse);
+                thismse = vfp->svf(y - y_stride - 1, y_stride, 6, 6, z, b->src_stride, &sse);
             }
         }
 
@@ -589,12 +604,12 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
         if (startmv.as_mv.row & 7)
         {
             this_mv.as_mv.row -= 2;
-            thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+            thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
         }
         else
         {
             this_mv.as_mv.row = (startmv.as_mv.row - 8) | 6;
-            thismse = vfp->svf(y - d->pre_stride, d->pre_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
+            thismse = vfp->svf(y - y_stride, y_stride, this_mv.as_mv.col & 7, 6, z, b->src_stride, &sse);
         }
 
         break;
@@ -604,19 +619,19 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
         if (startmv.as_mv.col & 7)
         {
             this_mv.as_mv.col -= 2;
-            thismse = vfp->svf(y, d->pre_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+            thismse = vfp->svf(y, y_stride, this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
         }
         else
         {
             this_mv.as_mv.col = (startmv.as_mv.col - 8) | 6;
-            thismse = vfp->svf(y - 1, d->pre_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+            thismse = vfp->svf(y - 1, y_stride, 6, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
         }
 
         break;
     case 3:
         this_mv.as_mv.col += 2;
         this_mv.as_mv.row += 2;
-        thismse = vfp->svf(y, d->pre_stride,  this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
+        thismse = vfp->svf(y, y_stride,  this_mv.as_mv.col & 7, this_mv.as_mv.row & 7, z, b->src_stride, &sse);
         break;
     }
 
@@ -633,7 +648,7 @@ int vp8_find_best_sub_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
     return bestmse;
 }
 
-int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
+int vp8_find_best_half_pixel_step(MACROBLOCK *x, BLOCK *b, BLOCKD *d,
                                   int_mv *bestmv, int_mv *ref_mv,
                                   int error_per_bit,
                                   const vp8_variance_fn_ptr_t *vfp,
@@ -643,11 +658,26 @@ int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
     int bestmse = INT_MAX;
     int_mv startmv;
     int_mv this_mv;
-    unsigned char *y = *(d->base_pre) + d->pre + (bestmv->as_mv.row) * d->pre_stride + bestmv->as_mv.col;
     unsigned char *z = (*(b->base_src) + b->src);
     int left, right, up, down, diag;
     unsigned int sse;
+    int whichdir ;
     int thismse;
+    int y_stride;
+
+#if ARCH_X86 || ARCH_X86_64
+    MACROBLOCKD *xd = &x->e_mbd;
+    unsigned char *y0 = *(d->base_pre) + d->pre + (bestmv->as_mv.row) * d->pre_stride + bestmv->as_mv.col;
+    unsigned char *y;
+
+    y_stride = 32;
+    /* Copy 18 rows x 32 cols area to intermediate buffer before searching. */
+    vfp->copymem(y0 - 1 - d->pre_stride, d->pre_stride, xd->y_buf, y_stride, 18);
+    y = xd->y_buf + y_stride + 1;
+#else
+    unsigned char *y = *(d->base_pre) + d->pre + (bestmv->as_mv.row) * d->pre_stride + bestmv->as_mv.col;
+    y_stride = d->pre_stride;
+#endif
 
     // central mv
     bestmv->as_mv.row <<= 3;
@@ -655,14 +685,14 @@ int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
     startmv = *bestmv;
 
     // calculate central point error
-    bestmse = vfp->vf(y, d->pre_stride, z, b->src_stride, sse1);
+    bestmse = vfp->vf(y, y_stride, z, b->src_stride, sse1);
     *distortion = bestmse;
     bestmse += mv_err_cost(bestmv, ref_mv, mvcost, error_per_bit);
 
     // go left then right and check error
     this_mv.as_mv.row = startmv.as_mv.row;
     this_mv.as_mv.col = ((startmv.as_mv.col - 8) | 4);
-    thismse = vfp->svf_halfpix_h(y - 1, d->pre_stride, z, b->src_stride, &sse);
+    thismse = vfp->svf_halfpix_h(y - 1, y_stride, z, b->src_stride, &sse);
     left = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (left < bestmse)
@@ -674,7 +704,7 @@ int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
     }
 
     this_mv.as_mv.col += 8;
-    thismse = vfp->svf_halfpix_h(y, d->pre_stride, z, b->src_stride, &sse);
+    thismse = vfp->svf_halfpix_h(y, y_stride, z, b->src_stride, &sse);
     right = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (right < bestmse)
@@ -688,7 +718,7 @@ int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
     // go up then down and check error
     this_mv.as_mv.col = startmv.as_mv.col;
     this_mv.as_mv.row = ((startmv.as_mv.row - 8) | 4);
-    thismse = vfp->svf_halfpix_v(y - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
+    thismse = vfp->svf_halfpix_v(y - y_stride, y_stride, z, b->src_stride, &sse);
     up = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (up < bestmse)
@@ -700,7 +730,7 @@ int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
     }
 
     this_mv.as_mv.row += 8;
-    thismse = vfp->svf_halfpix_v(y, d->pre_stride, z, b->src_stride, &sse);
+    thismse = vfp->svf_halfpix_v(y, y_stride, z, b->src_stride, &sse);
     down = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (down < bestmse)
@@ -711,8 +741,6 @@ int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
         *sse1 = sse;
     }
 
-    // somewhat strangely not doing all the diagonals for half pel is slower than doing them.
-#if 0
     // now check 1 more diagonal -
     whichdir = (left < right ? 0 : 1) + (up < down ? 0 : 2);
     this_mv = startmv;
@@ -720,76 +748,28 @@ int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
     switch (whichdir)
     {
     case 0:
-        this_mv.col = (this_mv.col - 8) | 4;
-        this_mv.row = (this_mv.row - 8) | 4;
-        diag = vfp->svf(y - 1 - d->pre_stride, d->pre_stride, 4, 4, z, b->src_stride, &sse);
+        this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
+        this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
+        thismse = vfp->svf_halfpix_hv(y - 1 - y_stride, y_stride, z, b->src_stride, &sse);
         break;
     case 1:
-        this_mv.col += 4;
-        this_mv.row = (this_mv.row - 8) | 4;
-        diag = vfp->svf(y - d->pre_stride, d->pre_stride, 4, 4, z, b->src_stride, &sse);
+        this_mv.as_mv.col += 4;
+        this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
+        thismse = vfp->svf_halfpix_hv(y - y_stride, y_stride, z, b->src_stride, &sse);
         break;
     case 2:
-        this_mv.col = (this_mv.col - 8) | 4;
-        this_mv.row += 4;
-        diag = vfp->svf(y - 1, d->pre_stride, 4, 4, z, b->src_stride, &sse);
+        this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
+        this_mv.as_mv.row += 4;
+        thismse = vfp->svf_halfpix_hv(y - 1, y_stride, z, b->src_stride, &sse);
         break;
     case 3:
-        this_mv.col += 4;
-        this_mv.row += 4;
-        diag = vfp->svf(y, d->pre_stride, 4, 4, z, b->src_stride, &sse);
+    default:
+        this_mv.as_mv.col += 4;
+        this_mv.as_mv.row += 4;
+        thismse = vfp->svf_halfpix_hv(y, y_stride, z, b->src_stride, &sse);
         break;
     }
 
-    diag += mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
-
-    if (diag < bestmse)
-    {
-        *bestmv = this_mv;
-        bestmse = diag;
-    }
-
-#else
-    this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
-    this_mv.as_mv.row = (this_mv.as_mv.row - 8) | 4;
-    thismse = vfp->svf_halfpix_hv(y - 1 - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
-    diag = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
-
-    if (diag < bestmse)
-    {
-        *bestmv = this_mv;
-        bestmse = diag;
-        *distortion = thismse;
-        *sse1 = sse;
-    }
-
-    this_mv.as_mv.col += 8;
-    thismse = vfp->svf_halfpix_hv(y - d->pre_stride, d->pre_stride, z, b->src_stride, &sse);
-    diag = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
-
-    if (diag < bestmse)
-    {
-        *bestmv = this_mv;
-        bestmse = diag;
-        *distortion = thismse;
-        *sse1 = sse;
-    }
-
-    this_mv.as_mv.col = (this_mv.as_mv.col - 8) | 4;
-    this_mv.as_mv.row = startmv.as_mv.row + 4;
-    thismse = vfp->svf_halfpix_hv(y - 1, d->pre_stride, z, b->src_stride, &sse);
-    diag = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
-
-    if (diag < bestmse)
-    {
-        *bestmv = this_mv;
-        bestmse = diag;
-        *distortion = thismse;
-        *sse1 = sse;
-    }
-
-    this_mv.as_mv.col += 8;
-    thismse = vfp->svf_halfpix_hv(y, d->pre_stride, z, b->src_stride, &sse);
     diag = thismse + mv_err_cost(&this_mv, ref_mv, mvcost, error_per_bit);
 
     if (diag < bestmse)
@@ -800,7 +780,6 @@ int vp8_find_best_half_pixel_step(MACROBLOCK *mb, BLOCK *b, BLOCKD *d,
         *sse1 = sse;
     }
 
-#endif
     return bestmse;
 }
 
@@ -866,7 +845,7 @@ int vp8_hex_search
     unsigned char *what = (*(b->base_src) + b->src);
     int what_stride = b->src_stride;
     int in_what_stride = d->pre_stride;
-    int br = ref_mv->as_mv.row, bc = ref_mv->as_mv.col;
+    int br, bc;
     int_mv this_mv;
     unsigned int bestsad = 0x7fffffff;
     unsigned int thissad;
@@ -880,6 +859,11 @@ int vp8_hex_search
     fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
     fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
 
+    // adjust ref_mv to make sure it is within MV range
+    vp8_clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+    br = ref_mv->as_mv.row;
+    bc = ref_mv->as_mv.col;
+
     // Work out the start point for the search
     base_offset = (unsigned char *)(*(d->base_pre) + d->pre);
     this_offset = base_offset + (br * (d->pre_stride)) + bc;
@@ -1043,8 +1027,8 @@ int vp8_diamond_search_sad
     int best_site = 0;
     int last_site = 0;
 
-    int ref_row = ref_mv->as_mv.row;
-    int ref_col = ref_mv->as_mv.col;
+    int ref_row;
+    int ref_col;
     int this_row_offset;
     int this_col_offset;
     search_site *ss;
@@ -1057,8 +1041,10 @@ int vp8_diamond_search_sad
     fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
     fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
 
+    vp8_clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+    ref_row = ref_mv->as_mv.row;
+    ref_col = ref_mv->as_mv.col;
     *num00 = 0;
-
     best_mv->as_mv.row = ref_row;
     best_mv->as_mv.col = ref_col;
 
@@ -1162,8 +1148,8 @@ int vp8_diamond_search_sadx4
     int best_site = 0;
     int last_site = 0;
 
-    int ref_row = ref_mv->as_mv.row;
-    int ref_col = ref_mv->as_mv.col;
+    int ref_row;
+    int ref_col;
     int this_row_offset;
     int this_col_offset;
     search_site *ss;
@@ -1176,6 +1162,9 @@ int vp8_diamond_search_sadx4
     fcenter_mv.as_mv.row = center_mv->as_mv.row >> 3;
     fcenter_mv.as_mv.col = center_mv->as_mv.col >> 3;
 
+    vp8_clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+    ref_row = ref_mv->as_mv.row;
+    ref_col = ref_mv->as_mv.col;
     *num00 = 0;
     best_mv->as_mv.row = ref_row;
     best_mv->as_mv.col = ref_col;
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index b9537d1ba..f862fd0e4 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -9,6 +9,7 @@
  */
 
 
+#include "vpx_config.h"
 #include "vp8/common/onyxc_int.h"
 #include "onyx_int.h"
 #include "vp8/common/systemdependent.h"
@@ -24,7 +25,9 @@
 #include "segmentation.h"
 #include "vp8/common/g_common.h"
 #include "vpx_scale/yv12extend.h"
+#if CONFIG_POSTPROC
 #include "vp8/common/postproc.h"
+#endif
 #include "vpx_mem/vpx_mem.h"
 #include "vp8/common/swapyv12buffer.h"
 #include "vp8/common/threading.h"
@@ -2729,6 +2732,8 @@ static void scale_and_extend_source(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi)
         cpi->Source = &cpi->scaled_source;
 #endif
     }
+    else
+        cpi->Source = sd;
 }
 
 
@@ -3340,7 +3345,7 @@ void loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
 
 static void update_buffer_level(VP8_COMP *cpi)
 {
-    long long tmp;
+    int64_t tmp;
 
     /* Update the buffered average bitrate.
      *
@@ -3410,7 +3415,7 @@ static void update_buffer_level(VP8_COMP *cpi)
          */
         if(cpi->total_actual_bits > cpi->oxcf.starting_buffer_level)
         {
-            tmp = (long long)cpi->buffered_av_per_frame_bandwidth
+            tmp = (int64_t)cpi->buffered_av_per_frame_bandwidth
                   * cpi->oxcf.maximum_buffer_size
                   / cpi->av_per_frame_bandwidth;
             cpi->buffer_level = cpi->oxcf.maximum_buffer_size
@@ -3428,7 +3433,7 @@ static void update_buffer_level(VP8_COMP *cpi)
          */
         if(cpi->total_actual_bits > cpi->oxcf.starting_buffer_level)
         {
-            long long decayed_overshoot;
+            int64_t decayed_overshoot;
 
             decayed_overshoot = cpi->accumulated_overshoot;
             decayed_overshoot *= (cpi->oxcf.maximum_buffer_size
@@ -4796,22 +4801,22 @@ static void Pass2Encode(VP8_COMP *cpi, unsigned long *size, unsigned char *dest,
     {
         double two_pass_min_rate = (double)(cpi->oxcf.target_bandwidth
             *cpi->oxcf.two_pass_vbrmin_section / 100);
-        cpi->twopass.bits_left += (long long)(two_pass_min_rate / cpi->oxcf.frame_rate);
+        cpi->twopass.bits_left += (int64_t)(two_pass_min_rate / cpi->oxcf.frame_rate);
     }
 }
 #endif
 
 //For ARM NEON, d8-d15 are callee-saved registers, and need to be saved by us.
 #if HAVE_ARMV7
-extern void vp8_push_neon(INT64 *store);
-extern void vp8_pop_neon(INT64 *store);
+extern void vp8_push_neon(int64_t *store);
+extern void vp8_pop_neon(int64_t *store);
 #endif
 
 
-int vp8_receive_raw_frame(VP8_PTR ptr, unsigned int frame_flags, YV12_BUFFER_CONFIG *sd, INT64 time_stamp, INT64 end_time)
+int vp8_receive_raw_frame(VP8_PTR ptr, unsigned int frame_flags, YV12_BUFFER_CONFIG *sd, int64_t time_stamp, int64_t end_time)
 {
 #if HAVE_ARMV7
-    INT64 store_reg[8];
+    int64_t store_reg[8];
 #endif
     VP8_COMP              *cpi = (VP8_COMP *) ptr;
     VP8_COMMON            *cm = &cpi->common;
@@ -4862,10 +4867,10 @@ static int frame_is_reference(const VP8_COMP *cpi)
 }
 
 
-int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned long *size, unsigned char *dest, INT64 *time_stamp, INT64 *time_end, int flush)
+int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned long *size, unsigned char *dest, int64_t *time_stamp, int64_t *time_end, int flush)
 {
 #if HAVE_ARMV7
-    INT64 store_reg[8];
+    int64_t store_reg[8];
 #endif
     VP8_COMP *cpi = (VP8_COMP *) ptr;
     VP8_COMMON *cm = &cpi->common;
@@ -4972,7 +4977,7 @@ int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned lon
     // adjust frame rates based on timestamps given
     if (!cm->refresh_alt_ref_frame)
     {
-        long long this_duration;
+        int64_t this_duration;
         int step = 0;
 
         if (cpi->source->ts_start == cpi->first_time_stamp_ever)
@@ -4982,7 +4987,7 @@ int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned lon
         }
         else
         {
-            long long last_duration;
+            int64_t last_duration;
 
             this_duration = cpi->source->ts_end - cpi->last_end_time_stamp_seen;
             last_duration = cpi->last_end_time_stamp_seen
@@ -5158,7 +5163,7 @@ int vp8_get_compressed_data(VP8_PTR ptr, unsigned int *frame_flags, unsigned lon
                 int y_samples = orig->y_height * orig->y_width ;
                 int uv_samples = orig->uv_height * orig->uv_width ;
                 int t_samples = y_samples + 2 * uv_samples;
-                long long sq_error;
+                int64_t sq_error;
 
                 ye = calc_plane_error(orig->y_buffer, orig->y_stride,
                   recon->y_buffer, recon->y_stride, orig->y_width, orig->y_height,
diff --git a/vp8/encoder/onyx_int.h b/vp8/encoder/onyx_int.h
index a27822da6..460da21d4 100644
--- a/vp8/encoder/onyx_int.h
+++ b/vp8/encoder/onyx_int.h
@@ -321,10 +321,10 @@ typedef struct VP8_COMP
     CODING_CONTEXT coding_context;
 
     // Rate targetting variables
-    long long prediction_error;
-    long long last_prediction_error;
-    long long intra_error;
-    long long last_intra_error;
+    int64_t prediction_error;
+    int64_t last_prediction_error;
+    int64_t intra_error;
+    int64_t last_intra_error;
 
     int this_frame_target;
     int projected_frame_size;
@@ -347,7 +347,7 @@ typedef struct VP8_COMP
     int baseline_gf_interval;
     int active_arnr_frames;           // <= cpi->oxcf.arnr_max_frames
 
-    INT64 key_frame_count;
+    int64_t key_frame_count;
     int prior_key_frame_distance[KEY_FRAME_CONTEXT];
     int per_frame_bandwidth;          // Current section per frame bandwidth target
     int av_per_frame_bandwidth;        // Average frame size target for clip
@@ -358,9 +358,9 @@ typedef struct VP8_COMP
 
     int inter_frame_target;
     double output_frame_rate;
-    long long last_time_stamp_seen;
-    long long last_end_time_stamp_seen;
-    long long first_time_stamp_ever;
+    int64_t last_time_stamp_seen;
+    int64_t last_end_time_stamp_seen;
+    int64_t first_time_stamp_ever;
 
     int ni_av_qi;
     int ni_tot_qi;
@@ -371,7 +371,7 @@ typedef struct VP8_COMP
     int zbin_mode_boost;
     int zbin_mode_boost_enabled;
 
-    INT64 total_byte_count;
+    int64_t total_byte_count;
 
     int buffered_mode;
 
@@ -384,7 +384,7 @@ typedef struct VP8_COMP
     int long_rolling_target_bits;
     int long_rolling_actual_bits;
 
-    long long total_actual_bits;
+    int64_t total_actual_bits;
     int total_target_vs_actual;        // debug stats
 
     int worst_quality;
@@ -541,8 +541,8 @@ typedef struct VP8_COMP
         FIRSTPASS_STATS *this_frame_stats;
         FIRSTPASS_STATS *stats_in, *stats_in_end, *stats_in_start;
         int first_pass_done;
-        long long bits_left;
-        long long clip_bits_total;
+        int64_t bits_left;
+        int64_t clip_bits_total;
         double avg_iiratio;
         double modified_error_total;
         double modified_error_used;
@@ -562,10 +562,10 @@ typedef struct VP8_COMP
         int gf_group_error_left;           // Remaining error from uncoded frames in a gf group. Two pass use only
 
         // Projected total bits available for a key frame group of frames
-        long long kf_group_bits;
+        int64_t kf_group_bits;
 
         // Error score of frames still to be coded in kf group
-        long long kf_group_error_left;
+        int64_t kf_group_error_left;
 
         int gf_group_bits;                // Projected Bits available for a group of frames including 1 GF or ARF
         int gf_bits;                     // Bits for the golden frame or ARF - 2 pass only
diff --git a/vp8/encoder/pickinter.c b/vp8/encoder/pickinter.c
index 725e44e62..9cdaf7d53 100644
--- a/vp8/encoder/pickinter.c
+++ b/vp8/encoder/pickinter.c
@@ -669,8 +669,6 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
                 mvp_full.as_mv.col = mvp.as_mv.col>>3;
                 mvp_full.as_mv.row = mvp.as_mv.row>>3;
 
-                /* adjust mvp to make sure it is within MV range */
-                vp8_clamp_mv(&mvp_full, col_min, col_max, row_min, row_max);
             }else
             {
                 mvp.as_int = best_ref_mv.as_int;
diff --git a/vp8/encoder/picklpf.c b/vp8/encoder/picklpf.c
index 49de62d7d..beefe8d8e 100644
--- a/vp8/encoder/picklpf.c
+++ b/vp8/encoder/picklpf.c
@@ -161,7 +161,7 @@ void vp8cx_pick_filter_level_fast(YV12_BUFFER_CONFIG *sd, VP8_COMP *cpi)
     if (cm->sharpness_level != cm->last_sharpness_level)
     {
         vp8_loop_filter_update_sharpness(&cm->lf_info, cm->sharpness_level);
-        cm->last_sharpness_level = cm->last_sharpness_level;
+        cm->last_sharpness_level = cm->sharpness_level;
     }
 
     // Start the search at the previous frame filter level unless it is now out of range.
diff --git a/vp8/encoder/ratectrl.c b/vp8/encoder/ratectrl.c
index dea690b13..69066e5ab 100644
--- a/vp8/encoder/ratectrl.c
+++ b/vp8/encoder/ratectrl.c
@@ -897,9 +897,9 @@ static void calc_pframe_target_size(VP8_COMP *cpi)
                         if ( critical_buffer_level >
                              (cpi->oxcf.optimal_buffer_level >> 2) )
                         {
-                            INT64 qadjustment_range =
+                            int64_t qadjustment_range =
                                       cpi->worst_quality - cpi->ni_av_qi;
-                            INT64 above_base =
+                            int64_t above_base =
                                       (critical_buffer_level -
                                        (cpi->oxcf.optimal_buffer_level >> 2));
 
@@ -1163,7 +1163,7 @@ static void calc_pframe_target_size(VP8_COMP *cpi)
     if (cpi->pass==0
         && cpi->common.refresh_golden_frame
         && cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) {
-        long long adjust;
+        int64_t adjust;
 
         /*
         frames_in_buffer = cpi->oxcf.maximum_buffer_size
diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c
index 3d77bb35a..952977094 100644
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -728,7 +728,7 @@ static int rd_pick_intra4x4mby_modes(VP8_COMP *cpi, MACROBLOCK *mb, int *Rate,
     int cost = mb->mbmode_cost [xd->frame_type] [B_PRED];
     int distortion = 0;
     int tot_rate_y = 0;
-    long long total_rd = 0;
+    int64_t total_rd = 0;
     ENTROPY_CONTEXT_PLANES t_above, t_left;
     ENTROPY_CONTEXT *ta;
     ENTROPY_CONTEXT *tl;
@@ -770,11 +770,11 @@ static int rd_pick_intra4x4mby_modes(VP8_COMP *cpi, MACROBLOCK *mb, int *Rate,
 
         mic->bmi[i].as_mode = best_mode;
 
-        if(total_rd >= (long long)best_rd)
+        if(total_rd >= (int64_t)best_rd)
             break;
     }
 
-    if(total_rd >= (long long)best_rd)
+    if(total_rd >= (int64_t)best_rd)
         return INT_MAX;
 
     *Rate = cost;
@@ -1244,6 +1244,9 @@ static void rd_check_segment(VP8_COMP *cpi, MACROBLOCK *x,
                     // Should we do a full search (best quality only)
                     if ((cpi->compressor_speed == 0) && (bestsme >> sseshift) > 4000)
                     {
+                        /* Check if mvp_full is within the range. */
+                        vp8_clamp_mv(&mvp_full, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
+
                         thissme = cpi->full_search_sad(x, c, e, &mvp_full,
                                                        sadpb, 16, v_fn_ptr,
                                                        x->mvcost, bsi->ref_mv);
@@ -2079,9 +2082,6 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int
             mvp_full.as_mv.col = mvp.as_mv.col>>3;
             mvp_full.as_mv.row = mvp.as_mv.row>>3;
 
-            /* adjust mvp to make sure it is within MV range */
-            vp8_clamp_mv(&mvp_full, col_min, col_max, row_min, row_max);
-
             // Get intersection of UMV window and valid MV window to reduce # of checks in diamond search.
             if (x->mv_col_min < col_min )
                 x->mv_col_min = col_min;
diff --git a/vp8/encoder/ssim.c b/vp8/encoder/ssim.c
index b8701e813..fea756f7b 100644
--- a/vp8/encoder/ssim.c
+++ b/vp8/encoder/ssim.c
@@ -73,8 +73,8 @@ void ssim_parms_8x8_c
      }
 }
 
-const static long long cc1 =  26634; // (64^2*(.01*255)^2
-const static long long cc2 = 239708; // (64^2*(.03*255)^2
+const static int64_t cc1 =  26634; // (64^2*(.01*255)^2
+const static int64_t cc2 = 239708; // (64^2*(.03*255)^2
 
 static double similarity
 (
@@ -86,19 +86,19 @@ static double similarity
     int count
 )
 {
-    long long ssim_n, ssim_d;
-    long long c1, c2;
+    int64_t ssim_n, ssim_d;
+    int64_t c1, c2;
 
     //scale the constants by number of pixels
     c1 = (cc1*count*count)>>12;
     c2 = (cc2*count*count)>>12;
 
-    ssim_n = (2*sum_s*sum_r+ c1)*((long long) 2*count*sum_sxr-
-          (long long) 2*sum_s*sum_r+c2);
+    ssim_n = (2*sum_s*sum_r+ c1)*((int64_t) 2*count*sum_sxr-
+          (int64_t) 2*sum_s*sum_r+c2);
 
     ssim_d = (sum_s*sum_s +sum_r*sum_r+c1)*
-        ((long long)count*sum_sq_s-(long long)sum_s*sum_s +
-        (long long)count*sum_sq_r-(long long) sum_r*sum_r +c2) ;
+        ((int64_t)count*sum_sq_s-(int64_t)sum_s*sum_s +
+        (int64_t)count*sum_sq_r-(int64_t) sum_r*sum_r +c2) ;
 
     return ssim_n * 1.0 / ssim_d;
 }
@@ -124,11 +124,11 @@ long dssim(unsigned char *s,int sp, unsigned char *r,int rp,
            const vp8_variance_rtcd_vtable_t *rtcd)
 {
     unsigned long sum_s=0,sum_r=0,sum_sq_s=0,sum_sq_r=0,sum_sxr=0;
-    long long ssim3;
-    long long ssim_n1,ssim_n2;
-    long long ssim_d1,ssim_d2;
-    long long ssim_t1,ssim_t2;
-    long long c1, c2;
+    int64_t ssim3;
+    int64_t ssim_n1,ssim_n2;
+    int64_t ssim_d1,ssim_d2;
+    int64_t ssim_t1,ssim_t2;
+    int64_t c1, c2;
 
     // normalize by 256/64
     c1 = cc1*16;
@@ -137,12 +137,12 @@ long dssim(unsigned char *s,int sp, unsigned char *r,int rp,
     rtcd->ssimpf(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
     ssim_n1 = (2*sum_s*sum_r+ c1);
 
-    ssim_n2 =((long long) 2*256*sum_sxr-(long long) 2*sum_s*sum_r+c2);
+    ssim_n2 =((int64_t) 2*256*sum_sxr-(int64_t) 2*sum_s*sum_r+c2);
 
-    ssim_d1 =((long long)sum_s*sum_s +(long long)sum_r*sum_r+c1);
+    ssim_d1 =((int64_t)sum_s*sum_s +(int64_t)sum_r*sum_r+c1);
 
-    ssim_d2 = (256 * (long long) sum_sq_s-(long long) sum_s*sum_s +
-                    (long long) 256*sum_sq_r-(long long) sum_r*sum_r +c2) ;
+    ssim_d2 = (256 * (int64_t) sum_sq_s-(int64_t) sum_s*sum_s +
+                    (int64_t) 256*sum_sq_r-(int64_t) sum_r*sum_r +c2) ;
 
     ssim_t1 = 256 - 256 * ssim_n1 / ssim_d1;
     ssim_t2 = 256 - 256 * ssim_n2 / ssim_d2;
diff --git a/vp8/encoder/temporal_filter.c b/vp8/encoder/temporal_filter.c
index 19913a9b1..b9ade1c6c 100644
--- a/vp8/encoder/temporal_filter.c
+++ b/vp8/encoder/temporal_filter.c
@@ -24,7 +24,6 @@
 #include "segmentation.h"
 #include "vp8/common/g_common.h"
 #include "vpx_scale/yv12extend.h"
-#include "vp8/common/postproc.h"
 #include "vpx_mem/vpx_mem.h"
 #include "vp8/common/swapyv12buffer.h"
 #include "vp8/common/threading.h"
diff --git a/vp8/vp8_common.mk b/vp8/vp8_common.mk
index 91e1e6cf8..1a883b8ab 100644
--- a/vp8/vp8_common.mk
+++ b/vp8/vp8_common.mk
@@ -15,6 +15,7 @@ VP8_COMMON_SRCS-yes += common/ppflags.h
 VP8_COMMON_SRCS-yes += common/onyx.h
 VP8_COMMON_SRCS-yes += common/onyxd.h
 VP8_COMMON_SRCS-yes += common/alloccommon.c
+VP8_COMMON_SRCS-yes += common/asm_com_offsets.c
 VP8_COMMON_SRCS-yes += common/blockd.c
 VP8_COMMON_SRCS-yes += common/coefupdateprobs.h
 VP8_COMMON_SRCS-yes += common/debugmodes.c
@@ -101,19 +102,21 @@ VP8_COMMON_SRCS-$(HAVE_SSE2) += common/x86/postproc_sse2.asm
 endif
 
 # common (c)
-VP8_COMMON_SRCS-$(ARCH_ARM)  += common/asm_com_offsets.c
 ifeq ($(CONFIG_CSM),yes)
 VP8_COMMON_SRCS-yes += common/maskingmv.c
 VP8_COMMON_SRCS-$(HAVE_SSE3) += common/x86/mask_sse3.asm
 endif
 
 VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/arm_systemdependent.c
-
 VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/bilinearfilter_arm.c
 VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/bilinearfilter_arm.h
 VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/filter_arm.c
+VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/idct_arm.h
 VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/loopfilter_arm.c
+VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/loopfilter_arm.h
+VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/recon_arm.h
 VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/reconintra_arm.c
+VP8_COMMON_SRCS-$(ARCH_ARM)  += common/arm/subpixel_arm.h
 
 # common (armv6)
 VP8_COMMON_SRCS-$(HAVE_ARMV6)  += common/arm/armv6/bilinearfilter_v6$(ASM)
diff --git a/vp8/vp8_cx_iface.c b/vp8/vp8_cx_iface.c
index 199678817..9f703a3c4 100644
--- a/vp8/vp8_cx_iface.c
+++ b/vp8/vp8_cx_iface.c
@@ -40,6 +40,7 @@ struct vp8_extracfg
     unsigned int                experimental;
     vp8e_tuning                 tuning;
     unsigned int                cq_level;         /* constrained quality level */
+    unsigned int                rc_max_intra_bitrate_pct;
 
 };
 
@@ -73,6 +74,7 @@ static const struct extraconfig_map extracfg_map[] =
             0,                          /* experimental mode */
             0,                          /* tuning*/
             10,                         /* cq_level */
+            0,                          /* rc_max_intra_bitrate_pct */
         }
     }
 };
@@ -308,7 +310,7 @@ static vpx_codec_err_t set_vp8e_config(VP8_CONFIG *oxcf,
     }
 
     oxcf->target_bandwidth       = cfg.rc_target_bitrate;
-    oxcf->rc_max_intra_bitrate_pct = cfg.rc_max_intra_bitrate_pct;
+    oxcf->rc_max_intra_bitrate_pct = vp8_cfg.rc_max_intra_bitrate_pct;
 
     oxcf->best_allowed_q          = cfg.rc_min_quantizer;
     oxcf->worst_allowed_q         = cfg.rc_max_quantizer;
@@ -465,6 +467,7 @@ static vpx_codec_err_t set_param(vpx_codec_alg_priv_t *ctx,
         MAP(VP8E_SET_ARNR_TYPE     ,        xcfg.arnr_type);
         MAP(VP8E_SET_TUNING,                xcfg.tuning);
         MAP(VP8E_SET_CQ_LEVEL,              xcfg.cq_level);
+        MAP(VP8E_SET_MAX_INTRA_BITRATE_PCT, xcfg.rc_max_intra_bitrate_pct);
 
     }
 
@@ -745,7 +748,7 @@ static vpx_codec_err_t vp8e_encode(vpx_codec_alg_priv_t  *ctx,
     {
         unsigned int lib_flags;
         YV12_BUFFER_CONFIG sd;
-        INT64 dst_time_stamp, dst_end_time_stamp;
+        int64_t dst_time_stamp, dst_end_time_stamp;
         unsigned long size, cx_data_sz;
         unsigned char *cx_data;
 
@@ -1107,6 +1110,7 @@ static vpx_codec_ctrl_fn_map_t vp8e_ctf_maps[] =
     {VP8E_SET_ARNR_TYPE     ,           set_param},
     {VP8E_SET_TUNING,                   set_param},
     {VP8E_SET_CQ_LEVEL,                 set_param},
+    {VP8E_SET_MAX_INTRA_BITRATE_PCT,    set_param},
     { -1, NULL},
 };
 
@@ -1139,7 +1143,6 @@ static vpx_codec_enc_cfg_map_t vp8e_usage_cfg_map[] =
         {0},                /* rc_twopass_stats_in */
 #endif
         256,                /* rc_target_bandwidth */
-        0,                  /* rc_max_intra_bitrate_pct */
         4,                  /* rc_min_quantizer */
         63,                 /* rc_max_quantizer */
         100,                /* rc_undershoot_pct */
diff --git a/vp8/vp8_dx_iface.c b/vp8/vp8_dx_iface.c
index 13a072bff..ad8cd5e95 100644
--- a/vp8/vp8_dx_iface.c
+++ b/vp8/vp8_dx_iface.c
@@ -427,7 +427,7 @@ static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t  *ctx,
     if (!res && ctx->pbi)
     {
         YV12_BUFFER_CONFIG sd;
-        INT64 time_stamp = 0, time_end_stamp = 0;
+        int64_t time_stamp = 0, time_end_stamp = 0;
         vp8_ppflags_t flags = {0};
 
         if (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC)
diff --git a/vp8/vp8cx_arm.mk b/vp8/vp8cx_arm.mk
index 599bf4753..e8dbd5d7e 100644
--- a/vp8/vp8cx_arm.mk
+++ b/vp8/vp8cx_arm.mk
@@ -15,9 +15,12 @@
 # encoder
 VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/arm_csystemdependent.c
 
-VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/quantize_arm.c
-VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/picklpf_arm.c
 VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/dct_arm.c
+VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/dct_arm.h
+VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/encodemb_arm.h
+VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/picklpf_arm.c
+VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/quantize_arm.c
+VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/quantize_arm.h
 VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/variance_arm.c
 VP8_CX_SRCS-$(ARCH_ARM)  += encoder/arm/variance_arm.h
 
diff --git a/vp8/vp8dx.mk b/vp8/vp8dx.mk
index 85d6f513e..d88b595fb 100644
--- a/vp8/vp8dx.mk
+++ b/vp8/vp8dx.mk
@@ -48,6 +48,7 @@ VP8_DX_SRCS-yes += vp8_dx_iface.c
 #INCLUDES += common
 #INCLUDES += decoder
 
+VP8_DX_SRCS-yes += decoder/asm_dec_offsets.c
 VP8_DX_SRCS-yes += decoder/dboolhuff.c
 VP8_DX_SRCS-yes += decoder/decodemv.c
 VP8_DX_SRCS-yes += decoder/decodframe.c
diff --git a/vp8/vp8dx_arm.mk b/vp8/vp8dx_arm.mk
index 583562b97..6bde42f4c 100644
--- a/vp8/vp8dx_arm.mk
+++ b/vp8/vp8dx_arm.mk
@@ -12,9 +12,8 @@
 #VP8_DX_SRCS list is modified according to different platforms.
 
 VP8_DX_SRCS-$(ARCH_ARM)  += decoder/arm/arm_dsystemdependent.c
-VP8_CX_SRCS-$(ARCH_ARM)  += decoder/asm_dec_offsets.c
-
 VP8_DX_SRCS-$(ARCH_ARM)  += decoder/arm/dequantize_arm.c
+VP8_DX_SRCS-$(ARCH_ARM)  += decoder/arm/dequantize_arm.h
 
 #File list for armv6
 VP8_DX_SRCS-$(HAVE_ARMV6)  += decoder/arm/armv6/dequant_dc_idct_v6$(ASM)