Roll latest libvpx into Android.

Make the VP9 decoding 2X faster than the old one.
Checkout is from master branch(hash:242157c756314827ad9244952c7253e8900b9626).

Change-Id: Ibe67b3ee19f82b87df2416826b63a67f7f79b63a

18 files changed, 88 insertions, 260 deletions

diff --git a/libvpx/vp8/common/alloccommon.c b/libvpx/vp8/common/alloccommon.c
index 8af9e90..54afc13 100644
--- a/libvpx/vp8/common/alloccommon.c
+++ b/libvpx/vp8/common/alloccommon.c
@@ -173,7 +173,6 @@ void vp8_create_common(VP8_COMMON *oci)
     oci->use_bilinear_mc_filter = 0;
     oci->full_pixel = 0;
     oci->multi_token_partition = ONE_PARTITION;
-    oci->clr_type = REG_YUV;
     oci->clamp_type = RECON_CLAMP_REQUIRED;
 
     /* Initialize reference frame sign bias structure to defaults */
diff --git a/libvpx/vp8/common/onyxc_int.h b/libvpx/vp8/common/onyxc_int.h
index 276dd72..e9bb7af 100644
--- a/libvpx/vp8/common/onyxc_int.h
+++ b/libvpx/vp8/common/onyxc_int.h
@@ -72,7 +72,6 @@ typedef struct VP8Common
     int horiz_scale;
     int vert_scale;
 
-    YUV_TYPE clr_type;
     CLAMP_TYPE  clamp_type;
 
     YV12_BUFFER_CONFIG *frame_to_show;
@@ -115,9 +114,6 @@ typedef struct VP8Common
     int uvdc_delta_q;
     int uvac_delta_q;
 
-    unsigned int frames_since_golden;
-    unsigned int frames_till_alt_ref_frame;
-
     /* We allocate a MODE_INFO struct for each macroblock, together with
        an extra row on top and column on the left to simplify prediction. */
 
@@ -157,7 +153,6 @@ typedef struct VP8Common
 
     unsigned int current_video_frame;
 
-    int near_boffset[3];
     int version;
 
     TOKEN_PARTITION multi_token_partition;
@@ -165,8 +160,10 @@ typedef struct VP8Common
 #ifdef PACKET_TESTING
     VP8_HEADER oh;
 #endif
+#if CONFIG_POSTPROC_VISUALIZER
     double bitrate;
     double framerate;
+#endif
 
 #if CONFIG_MULTITHREAD
     int processor_core_count;
diff --git a/libvpx/vp8/common/postproc.c b/libvpx/vp8/common/postproc.c
index 0266f4c..dd998f1 100644
--- a/libvpx/vp8/common/postproc.c
+++ b/libvpx/vp8/common/postproc.c
@@ -923,7 +923,7 @@ int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t
     if (flags & VP8D_DEBUG_TXT_RATE_INFO)
     {
         char message[512];
-        sprintf(message, "Bitrate: %10.2f frame_rate: %10.2f ", oci->bitrate, oci->framerate);
+        sprintf(message, "Bitrate: %10.2f framerate: %10.2f ", oci->bitrate, oci->framerate);
         vp8_blit_text(message, oci->post_proc_buffer.y_buffer, oci->post_proc_buffer.y_stride);
     }
 
diff --git a/libvpx/vp8/common/vp8_asm_com_offsets.c b/libvpx/vp8/common/vp8_asm_com_offsets.c
deleted file mode 100644
index 7bab90f..0000000
--- a/libvpx/vp8/common/vp8_asm_com_offsets.c
+++ /dev/null
@@ -1,52 +0,0 @@
-/*
- *  Copyright (c) 2011 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/vpx_codec.h"
-#include "vpx_ports/asm_offsets.h"
-#include "vp8/common/blockd.h"
-
-#if CONFIG_POSTPROC
-#include "postproc.h"
-#endif /* CONFIG_POSTPROC */
-
-BEGIN
-
-#if CONFIG_POSTPROC
-/* mfqe.c / filter_by_weight */
-DEFINE(MFQE_PRECISION_VAL,                      MFQE_PRECISION);
-#endif /* CONFIG_POSTPROC */
-
-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_MEDIA
-/* switch case in vp8_intra4x4_predict_armv6 is based on these enumerated values */
-ct_assert(B_DC_PRED, B_DC_PRED == 0);
-ct_assert(B_TM_PRED, B_TM_PRED == 1);
-ct_assert(B_VE_PRED, B_VE_PRED == 2);
-ct_assert(B_HE_PRED, B_HE_PRED == 3);
-ct_assert(B_LD_PRED, B_LD_PRED == 4);
-ct_assert(B_RD_PRED, B_RD_PRED == 5);
-ct_assert(B_VR_PRED, B_VR_PRED == 6);
-ct_assert(B_VL_PRED, B_VL_PRED == 7);
-ct_assert(B_HD_PRED, B_HD_PRED == 8);
-ct_assert(B_HU_PRED, B_HU_PRED == 9);
-#endif
-
-#if HAVE_SSE2
-#if CONFIG_POSTPROC
-/* vp8_filter_by_weight16x16 and 8x8 */
-ct_assert(MFQE_PRECISION_VAL, MFQE_PRECISION == 4)
-#endif /* CONFIG_POSTPROC */
-#endif /* HAVE_SSE2 */
diff --git a/libvpx/vp8/decoder/dboolhuff.c b/libvpx/vp8/decoder/dboolhuff.c
index 546fb2d..0007d7a 100644
--- a/libvpx/vp8/decoder/dboolhuff.c
+++ b/libvpx/vp8/decoder/dboolhuff.c
@@ -47,8 +47,8 @@ void vp8dx_bool_decoder_fill(BOOL_DECODER *br)
     unsigned char decrypted[sizeof(VP8_BD_VALUE) + 1];
 
     if (br->decrypt_cb) {
-        int n = bytes_left > sizeof(decrypted) ? sizeof(decrypted) : bytes_left;
-        br->decrypt_cb(br->decrypt_state, bufptr, decrypted, n);
+        size_t n = bytes_left > sizeof(decrypted) ? sizeof(decrypted) : bytes_left;
+        br->decrypt_cb(br->decrypt_state, bufptr, decrypted, (int)n);
         bufptr = decrypted;
     }
 
diff --git a/libvpx/vp8/decoder/decodframe.c b/libvpx/vp8/decoder/decodframe.c
index 44c35ef..51eeb02 100644
--- a/libvpx/vp8/decoder/decodframe.c
+++ b/libvpx/vp8/decoder/decodframe.c
@@ -1095,7 +1095,7 @@ int vp8_decode_frame(VP8D_COMP *pbi)
         vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
                            "Failed to allocate bool decoder 0");
     if (pc->frame_type == KEY_FRAME) {
-        pc->clr_type    = (YUV_TYPE)vp8_read_bit(bc);
+        (void)vp8_read_bit(bc);  // colorspace
         pc->clamp_type  = (CLAMP_TYPE)vp8_read_bit(bc);
     }
 
diff --git a/libvpx/vp8/decoder/onyxd_if.c b/libvpx/vp8/decoder/onyxd_if.c
index 2db3096..2d9e343 100644
--- a/libvpx/vp8/decoder/onyxd_if.c
+++ b/libvpx/vp8/decoder/onyxd_if.c
@@ -430,7 +430,6 @@ int vp8dx_get_raw_frame(VP8D_COMP *pbi, YV12_BUFFER_CONFIG *sd, int64_t *time_st
     *time_stamp = pbi->last_time_stamp;
     *time_end_stamp = 0;
 
-    sd->clrtype = pbi->common.clr_type;
 #if CONFIG_POSTPROC
     ret = vp8_post_proc_frame(&pbi->common, sd, flags);
 #else
diff --git a/libvpx/vp8/decoder/vp8_asm_dec_offsets.c b/libvpx/vp8/decoder/vp8_asm_dec_offsets.c
deleted file mode 100644
index 842a0d5..0000000
--- a/libvpx/vp8/decoder/vp8_asm_dec_offsets.c
+++ /dev/null
@@ -1,26 +0,0 @@
-/*
- *  Copyright (c) 2011 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/asm_offsets.h"
-#include "onyxd_int.h"
-
-BEGIN
-
-DEFINE(bool_decoder_user_buffer_end,            offsetof(BOOL_DECODER, user_buffer_end));
-DEFINE(bool_decoder_user_buffer,                offsetof(BOOL_DECODER, user_buffer));
-DEFINE(bool_decoder_value,                      offsetof(BOOL_DECODER, value));
-DEFINE(bool_decoder_count,                      offsetof(BOOL_DECODER, count));
-DEFINE(bool_decoder_range,                      offsetof(BOOL_DECODER, range));
-
-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 */
diff --git a/libvpx/vp8/encoder/bitstream.c b/libvpx/vp8/encoder/bitstream.c
index 4707ae5..5f0c1f7 100644
--- a/libvpx/vp8/encoder/bitstream.c
+++ b/libvpx/vp8/encoder/bitstream.c
@@ -1322,7 +1322,7 @@ void vp8_pack_bitstream(VP8_COMP *cpi, unsigned char *dest, unsigned char * dest
         vp8_start_encode(bc, cx_data, cx_data_end);
 
         /* signal clr type */
-        vp8_write_bit(bc, pc->clr_type);
+        vp8_write_bit(bc, 0);
         vp8_write_bit(bc, pc->clamp_type);
 
     }
diff --git a/libvpx/vp8/encoder/firstpass.c b/libvpx/vp8/encoder/firstpass.c
index 433726d..ded0c43 100644
--- a/libvpx/vp8/encoder/firstpass.c
+++ b/libvpx/vp8/encoder/firstpass.c
@@ -1325,7 +1325,7 @@ static int estimate_kf_group_q(VP8_COMP *cpi, double section_err, int section_ta
     return Q;
 }
 
-extern void vp8_new_frame_rate(VP8_COMP *cpi, double framerate);
+extern void vp8_new_framerate(VP8_COMP *cpi, double framerate);
 
 void vp8_init_second_pass(VP8_COMP *cpi)
 {
@@ -1349,9 +1349,9 @@ void vp8_init_second_pass(VP8_COMP *cpi)
      * sum duration is not. Its calculated based on the actual durations of
      * all frames from the first pass.
      */
-    vp8_new_frame_rate(cpi, 10000000.0 * cpi->twopass.total_stats.count / cpi->twopass.total_stats.duration);
+    vp8_new_framerate(cpi, 10000000.0 * cpi->twopass.total_stats.count / cpi->twopass.total_stats.duration);
 
-    cpi->output_frame_rate = cpi->frame_rate;
+    cpi->output_framerate = cpi->framerate;
     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);
 
@@ -2398,7 +2398,7 @@ static void assign_std_frame_bits(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame)
     target_frame_size += cpi->min_frame_bandwidth;
 
     /* Every other frame gets a few extra bits */
-    if ( (cpi->common.frames_since_golden & 0x01) &&
+    if ( (cpi->frames_since_golden & 0x01) &&
          (cpi->frames_till_gf_update_due > 0) )
     {
         target_frame_size += cpi->twopass.alt_extra_bits;
@@ -2529,7 +2529,7 @@ void vp8_second_pass(VP8_COMP *cpi)
 
     /* Set nominal per second bandwidth for this frame */
     cpi->target_bandwidth = (int)
-    (cpi->per_frame_bandwidth * cpi->output_frame_rate);
+    (cpi->per_frame_bandwidth * cpi->output_framerate);
     if (cpi->target_bandwidth < 0)
         cpi->target_bandwidth = 0;
 
@@ -3185,7 +3185,7 @@ static void find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame)
 
         /* Convert to a per second bitrate */
         cpi->target_bandwidth = (int)(cpi->twopass.kf_bits *
-                                      cpi->output_frame_rate);
+                                      cpi->output_framerate);
     }
 
     /* Note the total error score of the kf group minus the key frame itself */
@@ -3224,7 +3224,7 @@ static void find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame)
         cpi->common.vert_scale = NORMAL;
 
         /* Calculate Average bits per frame. */
-        av_bits_per_frame = cpi->oxcf.target_bandwidth / DOUBLE_DIVIDE_CHECK((double)cpi->frame_rate);
+        av_bits_per_frame = cpi->oxcf.target_bandwidth / DOUBLE_DIVIDE_CHECK((double)cpi->framerate);
 
         /* CBR... Use the clip average as the target for deciding resample */
         if (cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER)
@@ -3299,7 +3299,7 @@ static void find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame)
         }
         else
         {
-            int64_t clip_bits = (int64_t)(cpi->twopass.total_stats.count * cpi->oxcf.target_bandwidth / DOUBLE_DIVIDE_CHECK((double)cpi->frame_rate));
+            int64_t clip_bits = (int64_t)(cpi->twopass.total_stats.count * cpi->oxcf.target_bandwidth / DOUBLE_DIVIDE_CHECK((double)cpi->framerate));
             int64_t over_spend = cpi->oxcf.starting_buffer_level - cpi->buffer_level;
 
             /* If triggered last time the threshold for triggering again is
diff --git a/libvpx/vp8/encoder/onyx_if.c b/libvpx/vp8/encoder/onyx_if.c
index 73f6583..7c07975 100644
--- a/libvpx/vp8/encoder/onyx_if.c
+++ b/libvpx/vp8/encoder/onyx_if.c
@@ -301,11 +301,11 @@ static int rescale(int val, int num, int denom)
 static void init_temporal_layer_context(VP8_COMP *cpi,
                                         VP8_CONFIG *oxcf,
                                         const int layer,
-                                        double prev_layer_frame_rate)
+                                        double prev_layer_framerate)
 {
     LAYER_CONTEXT *lc = &cpi->layer_context[layer];
 
-    lc->frame_rate = cpi->output_frame_rate / cpi->oxcf.rate_decimator[layer];
+    lc->framerate = cpi->output_framerate / cpi->oxcf.rate_decimator[layer];
     lc->target_bandwidth = cpi->oxcf.target_bitrate[layer] * 1000;
 
     lc->starting_buffer_level_in_ms = oxcf->starting_buffer_level;
@@ -335,7 +335,7 @@ static void init_temporal_layer_context(VP8_COMP *cpi,
       lc->avg_frame_size_for_layer =
           (int)((cpi->oxcf.target_bitrate[layer] -
                 cpi->oxcf.target_bitrate[layer-1]) * 1000 /
-                (lc->frame_rate - prev_layer_frame_rate));
+                (lc->framerate - prev_layer_framerate));
 
      lc->active_worst_quality         = cpi->oxcf.worst_allowed_q;
      lc->active_best_quality          = cpi->oxcf.best_allowed_q;
@@ -363,7 +363,7 @@ static void reset_temporal_layer_change(VP8_COMP *cpi,
                                         const int prev_num_layers)
 {
     int i;
-    double prev_layer_frame_rate = 0;
+    double prev_layer_framerate = 0;
     const int curr_num_layers = cpi->oxcf.number_of_layers;
     // If the previous state was 1 layer, get current layer context from cpi.
     // We need this to set the layer context for the new layers below.
@@ -377,7 +377,7 @@ static void reset_temporal_layer_change(VP8_COMP *cpi,
         LAYER_CONTEXT *lc = &cpi->layer_context[i];
         if (i >= prev_num_layers)
         {
-           init_temporal_layer_context(cpi, oxcf, i, prev_layer_frame_rate);
+           init_temporal_layer_context(cpi, oxcf, i, prev_layer_framerate);
         }
         // The initial buffer levels are set based on their starting levels.
         // We could set the buffer levels based on the previous state (normalized
@@ -403,8 +403,8 @@ static void reset_temporal_layer_change(VP8_COMP *cpi,
             lc->bits_off_target = lc->buffer_level;
             restore_layer_context(cpi, 0);
         }
-        prev_layer_frame_rate =  cpi->output_frame_rate /
-                                 cpi->oxcf.rate_decimator[i];
+        prev_layer_framerate = cpi->output_framerate /
+                               cpi->oxcf.rate_decimator[i];
     }
 }
 
@@ -1282,21 +1282,21 @@ int vp8_reverse_trans(int x)
 
     return 63;
 }
-void vp8_new_frame_rate(VP8_COMP *cpi, double framerate)
+void vp8_new_framerate(VP8_COMP *cpi, double framerate)
 {
     if(framerate < .1)
         framerate = 30;
 
-    cpi->frame_rate             = framerate;
-    cpi->output_frame_rate      = framerate;
+    cpi->framerate              = framerate;
+    cpi->output_framerate       = framerate;
     cpi->per_frame_bandwidth    = (int)(cpi->oxcf.target_bandwidth /
-                                  cpi->output_frame_rate);
+                                  cpi->output_framerate);
     cpi->av_per_frame_bandwidth = cpi->per_frame_bandwidth;
     cpi->min_frame_bandwidth    = (int)(cpi->av_per_frame_bandwidth *
                                   cpi->oxcf.two_pass_vbrmin_section / 100);
 
     /* Set Maximum gf/arf interval */
-    cpi->max_gf_interval = ((int)(cpi->output_frame_rate / 2.0) + 2);
+    cpi->max_gf_interval = ((int)(cpi->output_framerate / 2.0) + 2);
 
     if(cpi->max_gf_interval < 12)
         cpi->max_gf_interval = 12;
@@ -1337,13 +1337,13 @@ static void init_config(VP8_COMP *cpi, VP8_CONFIG *oxcf)
      * seems like a reasonable framerate, then use that as a guess, otherwise
      * use 30.
      */
-    cpi->frame_rate = (double)(oxcf->timebase.den) /
-                      (double)(oxcf->timebase.num);
+    cpi->framerate = (double)(oxcf->timebase.den) /
+                     (double)(oxcf->timebase.num);
 
-    if (cpi->frame_rate > 180)
-        cpi->frame_rate = 30;
+    if (cpi->framerate > 180)
+        cpi->framerate = 30;
 
-    cpi->ref_frame_rate = cpi->frame_rate;
+    cpi->ref_framerate = cpi->framerate;
 
     /* change includes all joint functionality */
     vp8_change_config(cpi, oxcf);
@@ -1369,13 +1369,13 @@ static void init_config(VP8_COMP *cpi, VP8_CONFIG *oxcf)
     if (cpi->oxcf.number_of_layers > 1)
     {
         unsigned int i;
-        double prev_layer_frame_rate=0;
+        double prev_layer_framerate=0;
 
         for (i=0; i<cpi->oxcf.number_of_layers; i++)
         {
-            init_temporal_layer_context(cpi, oxcf, i, prev_layer_frame_rate);
-            prev_layer_frame_rate = cpi->output_frame_rate /
-                                    cpi->oxcf.rate_decimator[i];
+            init_temporal_layer_context(cpi, oxcf, i, prev_layer_framerate);
+            prev_layer_framerate = cpi->output_framerate /
+                                   cpi->oxcf.rate_decimator[i];
         }
     }
 
@@ -1399,14 +1399,14 @@ static void update_layer_contexts (VP8_COMP *cpi)
     if (oxcf->number_of_layers > 1)
     {
         unsigned int i;
-        double prev_layer_frame_rate=0;
+        double prev_layer_framerate=0;
 
         for (i=0; i<oxcf->number_of_layers; i++)
         {
             LAYER_CONTEXT *lc = &cpi->layer_context[i];
 
-            lc->frame_rate =
-                cpi->ref_frame_rate / oxcf->rate_decimator[i];
+            lc->framerate =
+                cpi->ref_framerate / oxcf->rate_decimator[i];
             lc->target_bandwidth = oxcf->target_bitrate[i] * 1000;
 
             lc->starting_buffer_level = rescale(
@@ -1432,9 +1432,9 @@ static void update_layer_contexts (VP8_COMP *cpi)
                 lc->avg_frame_size_for_layer =
                    (int)((oxcf->target_bitrate[i] -
                           oxcf->target_bitrate[i-1]) * 1000 /
-                          (lc->frame_rate - prev_layer_frame_rate));
+                          (lc->framerate - prev_layer_framerate));
 
-            prev_layer_frame_rate = lc->frame_rate;
+            prev_layer_framerate = lc->framerate;
         }
     }
 }
@@ -1625,7 +1625,7 @@ void vp8_change_config(VP8_COMP *cpi, VP8_CONFIG *oxcf)
                     cpi->oxcf.target_bandwidth, 1000);
 
     /* Set up frame rate and related parameters rate control values. */
-    vp8_new_frame_rate(cpi, cpi->frame_rate);
+    vp8_new_framerate(cpi, cpi->framerate);
 
     /* Set absolute upper and lower quality limits */
     cpi->worst_quality               = cpi->oxcf.worst_allowed_q;
@@ -1945,7 +1945,7 @@ struct VP8_COMP* vp8_create_compressor(VP8_CONFIG *oxcf)
 
     for (i = 0; i < KEY_FRAME_CONTEXT; i++)
     {
-        cpi->prior_key_frame_distance[i] = (int)cpi->output_frame_rate;
+        cpi->prior_key_frame_distance[i] = (int)cpi->output_framerate;
     }
 
 #ifdef OUTPUT_YUV_SRC
@@ -2273,7 +2273,7 @@ void vp8_remove_compressor(VP8_COMP **ptr)
         {
             extern int count_mb_seg[4];
             FILE *f = fopen("modes.stt", "a");
-            double dr = (double)cpi->frame_rate * (double)bytes * (double)8 / (double)count / (double)1000 ;
+            double dr = (double)cpi->framerate * (double)bytes * (double)8 / (double)count / (double)1000 ;
             fprintf(f, "intra_mode in Intra Frames:\n");
             fprintf(f, "Y: %8d, %8d, %8d, %8d, %8d\n", y_modes[0], y_modes[1], y_modes[2], y_modes[3], y_modes[4]);
             fprintf(f, "UV:%8d, %8d, %8d, %8d\n", uv_modes[0], uv_modes[1], uv_modes[2], uv_modes[3]);
@@ -2750,7 +2750,7 @@ static void update_alt_ref_frame_stats(VP8_COMP *cpi)
     cpi->gf_active_count = cm->mb_rows * cm->mb_cols;
 
     /* this frame refreshes means next frames don't unless specified by user */
-    cpi->common.frames_since_golden = 0;
+    cpi->frames_since_golden = 0;
 
     /* Clear the alternate reference update pending flag. */
     cpi->source_alt_ref_pending = 0;
@@ -2802,7 +2802,7 @@ static void update_golden_frame_stats(VP8_COMP *cpi)
          * user
          */
         cm->refresh_golden_frame = 0;
-        cpi->common.frames_since_golden = 0;
+        cpi->frames_since_golden = 0;
 
         cpi->recent_ref_frame_usage[INTRA_FRAME] = 1;
         cpi->recent_ref_frame_usage[LAST_FRAME] = 1;
@@ -2834,12 +2834,12 @@ static void update_golden_frame_stats(VP8_COMP *cpi)
         if (cpi->frames_till_gf_update_due > 0)
             cpi->frames_till_gf_update_due--;
 
-        if (cpi->common.frames_till_alt_ref_frame)
-            cpi->common.frames_till_alt_ref_frame --;
+        if (cpi->frames_till_alt_ref_frame)
+            cpi->frames_till_alt_ref_frame --;
 
-        cpi->common.frames_since_golden ++;
+        cpi->frames_since_golden ++;
 
-        if (cpi->common.frames_since_golden > 1)
+        if (cpi->frames_since_golden > 1)
         {
             cpi->recent_ref_frame_usage[INTRA_FRAME] +=
                 cpi->mb.count_mb_ref_frame_usage[INTRA_FRAME];
@@ -2890,11 +2890,11 @@ static void update_rd_ref_frame_probs(VP8_COMP *cpi)
             cpi->prob_last_coded = 200;
             cpi->prob_gf_coded = 1;
         }
-        else if (cpi->common.frames_since_golden == 0)
+        else if (cpi->frames_since_golden == 0)
         {
             cpi->prob_last_coded = 214;
         }
-        else if (cpi->common.frames_since_golden == 1)
+        else if (cpi->frames_since_golden == 1)
         {
             cpi->prob_last_coded = 192;
             cpi->prob_gf_coded = 220;
@@ -3368,12 +3368,12 @@ static void encode_frame_to_data_rate
             cpi->per_frame_bandwidth = cpi->twopass.gf_bits;
             /* per second target bitrate */
             cpi->target_bandwidth = (int)(cpi->twopass.gf_bits *
-                                          cpi->output_frame_rate);
+                                          cpi->output_framerate);
         }
     }
     else
 #endif
-        cpi->per_frame_bandwidth  = (int)(cpi->target_bandwidth / cpi->output_frame_rate);
+        cpi->per_frame_bandwidth  = (int)(cpi->target_bandwidth / cpi->output_framerate);
 
     /* Default turn off buffer to buffer copying */
     cm->copy_buffer_to_gf = 0;
@@ -4557,7 +4557,7 @@ static void encode_frame_to_data_rate
         {
             LAYER_CONTEXT *lc = &cpi->layer_context[i];
             int bits_off_for_this_layer =
-               (int)(lc->target_bandwidth / lc->frame_rate -
+               (int)(lc->target_bandwidth / lc->framerate -
                      cpi->projected_frame_size);
 
             lc->bits_off_target += bits_off_for_this_layer;
@@ -4805,7 +4805,7 @@ 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 += (int64_t)(two_pass_min_rate / cpi->frame_rate);
+        cpi->twopass.bits_left += (int64_t)(two_pass_min_rate / cpi->framerate);
     }
 }
 #endif
@@ -4821,8 +4821,10 @@ int vp8_receive_raw_frame(VP8_COMP *cpi, unsigned int frame_flags, YV12_BUFFER_C
 {
 #if HAVE_NEON
     int64_t store_reg[8];
-#endif
+#if CONFIG_RUNTIME_CPU_DETECT
     VP8_COMMON            *cm = &cpi->common;
+#endif
+#endif
     struct vpx_usec_timer  timer;
     int                    res = 0;
 
@@ -4848,7 +4850,6 @@ int vp8_receive_raw_frame(VP8_COMP *cpi, unsigned int frame_flags, YV12_BUFFER_C
     if(vp8_lookahead_push(cpi->lookahead, sd, time_stamp, end_time,
                           frame_flags, cpi->active_map_enabled ? cpi->active_map : NULL))
         res = -1;
-    cm->clr_type = sd->clrtype;
     vpx_usec_timer_mark(&timer);
     cpi->time_receive_data += vpx_usec_timer_elapsed(&timer);
 
@@ -4933,7 +4934,7 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l
                                               cpi->frames_till_gf_update_due);
                 force_src_buffer = &cpi->alt_ref_buffer;
             }
-            cm->frames_till_alt_ref_frame = cpi->frames_till_gf_update_due;
+            cpi->frames_till_alt_ref_frame = cpi->frames_till_gf_update_due;
             cm->refresh_alt_ref_frame = 1;
             cm->refresh_golden_frame = 0;
             cm->refresh_last_frame = 0;
@@ -5038,7 +5039,7 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l
         if (this_duration)
         {
             if (step)
-                cpi->ref_frame_rate = 10000000.0 / this_duration;
+                cpi->ref_framerate = 10000000.0 / this_duration;
             else
             {
                 double avg_duration, interval;
@@ -5052,11 +5053,11 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l
                 if(interval > 10000000.0)
                     interval = 10000000;
 
-                avg_duration = 10000000.0 / cpi->ref_frame_rate;
+                avg_duration = 10000000.0 / cpi->ref_framerate;
                 avg_duration *= (interval - avg_duration + this_duration);
                 avg_duration /= interval;
 
-                cpi->ref_frame_rate = 10000000.0 / avg_duration;
+                cpi->ref_framerate = 10000000.0 / avg_duration;
             }
 
             if (cpi->oxcf.number_of_layers > 1)
@@ -5067,12 +5068,12 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l
                 for (i=0; i<cpi->oxcf.number_of_layers; i++)
                 {
                     LAYER_CONTEXT *lc = &cpi->layer_context[i];
-                    lc->frame_rate = cpi->ref_frame_rate /
-                                  cpi->oxcf.rate_decimator[i];
+                    lc->framerate = cpi->ref_framerate /
+                                    cpi->oxcf.rate_decimator[i];
                 }
             }
             else
-                vp8_new_frame_rate(cpi, cpi->ref_frame_rate);
+                vp8_new_framerate(cpi, cpi->ref_framerate);
         }
 
         cpi->last_time_stamp_seen = cpi->source->ts_start;
@@ -5089,7 +5090,7 @@ int vp8_get_compressed_data(VP8_COMP *cpi, unsigned int *frame_flags, unsigned l
         layer = cpi->oxcf.layer_id[
                 cpi->temporal_pattern_counter % cpi->oxcf.periodicity];
         restore_layer_context (cpi, layer);
-        vp8_new_frame_rate (cpi, cpi->layer_context[layer].frame_rate);
+        vp8_new_framerate(cpi, cpi->layer_context[layer].framerate);
     }
 
     if (cpi->compressor_speed == 2)
diff --git a/libvpx/vp8/encoder/onyx_int.h b/libvpx/vp8/encoder/onyx_int.h
index 5120fcc..3ab0fe8 100644
--- a/libvpx/vp8/encoder/onyx_int.h
+++ b/libvpx/vp8/encoder/onyx_int.h
@@ -232,7 +232,7 @@ enum
 typedef struct
 {
     /* Layer configuration */
-    double frame_rate;
+    double framerate;
     int target_bandwidth;
 
     /* Layer specific coding parameters */
@@ -320,6 +320,7 @@ typedef struct VP8_COMP
     YV12_BUFFER_CONFIG scaled_source;
     YV12_BUFFER_CONFIG *last_frame_unscaled_source;
 
+    unsigned int frames_till_alt_ref_frame;
     /* frame in src_buffers has been identified to be encoded as an alt ref */
     int source_alt_ref_pending;
     /* an alt ref frame has been encoded and is usable */
@@ -369,6 +370,7 @@ typedef struct VP8_COMP
     double key_frame_rate_correction_factor;
     double gf_rate_correction_factor;
 
+    unsigned int frames_since_golden;
     /* Count down till next GF */
     int frames_till_gf_update_due;
 
@@ -401,7 +403,7 @@ typedef struct VP8_COMP
     /* Minimum allocation that should be used for any frame */
     int min_frame_bandwidth;
     int inter_frame_target;
-    double output_frame_rate;
+    double output_framerate;
     int64_t last_time_stamp_seen;
     int64_t last_end_time_stamp_seen;
     int64_t first_time_stamp_ever;
@@ -415,8 +417,8 @@ typedef struct VP8_COMP
 
     int buffered_mode;
 
-    double frame_rate;
-    double ref_frame_rate;
+    double framerate;
+    double ref_framerate;
     int64_t buffer_level;
     int64_t bits_off_target;
 
diff --git a/libvpx/vp8/encoder/ratectrl.c b/libvpx/vp8/encoder/ratectrl.c
index 8e3c01d..1e8259c 100644
--- a/libvpx/vp8/encoder/ratectrl.c
+++ b/libvpx/vp8/encoder/ratectrl.c
@@ -234,7 +234,7 @@ void vp8_save_coding_context(VP8_COMP *cpi)
     cc->frames_since_key          = cpi->frames_since_key;
     cc->filter_level             = cpi->common.filter_level;
     cc->frames_till_gf_update_due   = cpi->frames_till_gf_update_due;
-    cc->frames_since_golden       = cpi->common.frames_since_golden;
+    cc->frames_since_golden       = cpi->frames_since_golden;
 
     vp8_copy(cc->mvc,      cpi->common.fc.mvc);
     vp8_copy(cc->mvcosts,  cpi->rd_costs.mvcosts);
@@ -271,7 +271,7 @@ void vp8_restore_coding_context(VP8_COMP *cpi)
     cpi->frames_since_key         =   cc->frames_since_key;
     cpi->common.filter_level     =   cc->filter_level;
     cpi->frames_till_gf_update_due  =   cc->frames_till_gf_update_due;
-    cpi->common.frames_since_golden       =   cc->frames_since_golden;
+    cpi->frames_since_golden       =   cc->frames_since_golden;
 
     vp8_copy(cpi->common.fc.mvc, cc->mvc);
 
@@ -388,7 +388,7 @@ static void calc_iframe_target_size(VP8_COMP *cpi)
         int initial_boost = 32; /* |3.0 * per_frame_bandwidth| */
         /* Boost depends somewhat on frame rate: only used for 1 layer case. */
         if (cpi->oxcf.number_of_layers == 1) {
-          kf_boost = MAX(initial_boost, (int)(2 * cpi->output_frame_rate - 16));
+          kf_boost = MAX(initial_boost, (int)(2 * cpi->output_framerate - 16));
         }
         else {
           /* Initial factor: set target size to: |3.0 * per_frame_bandwidth|. */
@@ -399,9 +399,9 @@ static void calc_iframe_target_size(VP8_COMP *cpi)
         kf_boost = kf_boost * kf_boost_qadjustment[Q] / 100;
 
         /* frame separation adjustment ( down) */
-        if (cpi->frames_since_key  < cpi->output_frame_rate / 2)
+        if (cpi->frames_since_key  < cpi->output_framerate / 2)
             kf_boost = (int)(kf_boost
-                       * cpi->frames_since_key / (cpi->output_frame_rate / 2));
+                       * cpi->frames_since_key / (cpi->output_framerate / 2));
 
         /* Minimal target size is |2* per_frame_bandwidth|. */
         if (kf_boost < 16)
@@ -715,7 +715,7 @@ static void calc_pframe_target_size(VP8_COMP *cpi)
                 if (Adjustment > (cpi->this_frame_target - min_frame_target))
                     Adjustment = (cpi->this_frame_target - min_frame_target);
 
-                if (cpi->common.frames_since_golden == (cpi->current_gf_interval >> 1))
+                if (cpi->frames_since_golden == (cpi->current_gf_interval >> 1))
                     cpi->this_frame_target += ((cpi->current_gf_interval - 1) * Adjustment);
                 else
                     cpi->this_frame_target -= Adjustment;
@@ -1360,7 +1360,7 @@ static int estimate_keyframe_frequency(VP8_COMP *cpi)
          * whichever is smaller.
          */
         int key_freq = cpi->oxcf.key_freq>0 ? cpi->oxcf.key_freq : 1;
-        av_key_frame_frequency = 1 + (int)cpi->output_frame_rate * 2;
+        av_key_frame_frequency = 1 + (int)cpi->output_framerate * 2;
 
         if (cpi->oxcf.auto_key && av_key_frame_frequency > key_freq)
             av_key_frame_frequency = key_freq;
diff --git a/libvpx/vp8/encoder/rdopt.c b/libvpx/vp8/encoder/rdopt.c
index 8579614..521e84f 100644
--- a/libvpx/vp8/encoder/rdopt.c
+++ b/libvpx/vp8/encoder/rdopt.c
@@ -341,7 +341,7 @@ void vp8_initialize_rd_consts(VP8_COMP *cpi, MACROBLOCK *x, int Qvalue)
 
 void vp8_auto_select_speed(VP8_COMP *cpi)
 {
-    int milliseconds_for_compress = (int)(1000000 / cpi->frame_rate);
+    int milliseconds_for_compress = (int)(1000000 / cpi->framerate);
 
     milliseconds_for_compress = milliseconds_for_compress * (16 - cpi->oxcf.cpu_used) / 16;
 
diff --git a/libvpx/vp8/vp8_common.mk b/libvpx/vp8/vp8_common.mk
index cde2651..f98eb31 100644
--- a/libvpx/vp8/vp8_common.mk
+++ b/libvpx/vp8/vp8_common.mk
@@ -66,7 +66,6 @@ VP8_COMMON_SRCS-yes += common/setupintrarecon.c
 VP8_COMMON_SRCS-yes += common/swapyv12buffer.c
 VP8_COMMON_SRCS-yes += common/variance_c.c
 VP8_COMMON_SRCS-yes += common/variance.h
-VP8_COMMON_SRCS-yes += common/vp8_asm_com_offsets.c
 VP8_COMMON_SRCS-yes += common/vp8_entropymodedata.h
 
 
@@ -192,7 +191,4 @@ VP8_COMMON_SRCS-$(HAVE_NEON)  += common/arm/neon/vp8_subpixelvariance8x8_neon$(A
 VP8_COMMON_SRCS-$(HAVE_NEON)  += common/arm/neon/vp8_subpixelvariance16x16_neon$(ASM)
 VP8_COMMON_SRCS-$(HAVE_NEON)  += common/arm/neon/vp8_subpixelvariance16x16s_neon$(ASM)
 
-$(eval $(call asm_offsets_template,\
-         vp8_asm_com_offsets.asm, $(VP8_PREFIX)common/vp8_asm_com_offsets.c))
-
 $(eval $(call rtcd_h_template,vp8_rtcd,vp8/common/rtcd_defs.sh))
diff --git a/libvpx/vp8/vp8_cx_iface.c b/libvpx/vp8/vp8_cx_iface.c
index 4531d5a..9a7b9c5 100644
--- a/libvpx/vp8/vp8_cx_iface.c
+++ b/libvpx/vp8/vp8_cx_iface.c
@@ -695,7 +695,6 @@ static vpx_codec_err_t image2yuvconfig(const vpx_image_t   *img,
     yv12->uv_stride = img->stride[VPX_PLANE_U];
 
     yv12->border  = (img->stride[VPX_PLANE_Y] - img->w) / 2;
-    yv12->clrtype = (img->fmt == VPX_IMG_FMT_VPXI420 || img->fmt == VPX_IMG_FMT_VPXYV12);
     return res;
 }
 
@@ -1079,11 +1078,7 @@ static vpx_image_t *vp8e_get_preview(vpx_codec_alg_priv_t *ctx)
         ctx->preview_img.planes[VPX_PLANE_U] = sd.u_buffer;
         ctx->preview_img.planes[VPX_PLANE_V] = sd.v_buffer;
 
-        if (sd.clrtype == REG_YUV)
-            ctx->preview_img.fmt = VPX_IMG_FMT_I420;
-        else
-            ctx->preview_img.fmt = VPX_IMG_FMT_VPXI420;
-
+        ctx->preview_img.fmt = VPX_IMG_FMT_I420;
         ctx->preview_img.x_chroma_shift = 1;
         ctx->preview_img.y_chroma_shift = 1;
 
diff --git a/libvpx/vp8/vp8_dx_iface.c b/libvpx/vp8/vp8_dx_iface.c
index c826f69..871b8d3 100644
--- a/libvpx/vp8/vp8_dx_iface.c
+++ b/libvpx/vp8/vp8_dx_iface.c
@@ -41,15 +41,6 @@ typedef enum
 
 static unsigned long vp8_priv_sz(const vpx_codec_dec_cfg_t *si, vpx_codec_flags_t);
 
-typedef struct
-{
-    unsigned int   id;
-    unsigned long  sz;
-    unsigned int   align;
-    unsigned int   flags;
-    unsigned long(*calc_sz)(const vpx_codec_dec_cfg_t *, vpx_codec_flags_t);
-} mem_req_t;
-
 static const mem_req_t vp8_mem_req_segs[] =
 {
     {VP8_SEG_ALG_PRIV,    0, 8, VPX_CODEC_MEM_ZERO, vp8_priv_sz},
@@ -93,65 +84,6 @@ static unsigned long vp8_priv_sz(const vpx_codec_dec_cfg_t *si, vpx_codec_flags_
     return sizeof(vpx_codec_alg_priv_t);
 }
 
-
-static void vp8_mmap_dtor(vpx_codec_mmap_t *mmap)
-{
-    free(mmap->priv);
-}
-
-static vpx_codec_err_t vp8_mmap_alloc(vpx_codec_mmap_t *mmap)
-{
-    vpx_codec_err_t  res;
-    unsigned int   align;
-
-    align = mmap->align ? mmap->align - 1 : 0;
-
-    if (mmap->flags & VPX_CODEC_MEM_ZERO)
-        mmap->priv = calloc(1, mmap->sz + align);
-    else
-        mmap->priv = malloc(mmap->sz + align);
-
-    res = (mmap->priv) ? VPX_CODEC_OK : VPX_CODEC_MEM_ERROR;
-    mmap->base = (void *)((((uintptr_t)mmap->priv) + align) & ~(uintptr_t)align);
-    mmap->dtor = vp8_mmap_dtor;
-    return res;
-}
-
-static vpx_codec_err_t vp8_validate_mmaps(const vp8_stream_info_t *si,
-        const vpx_codec_mmap_t        *mmaps,
-        vpx_codec_flags_t              init_flags)
-{
-    int i;
-    vpx_codec_err_t res = VPX_CODEC_OK;
-
-    for (i = 0; i < NELEMENTS(vp8_mem_req_segs) - 1; i++)
-    {
-        /* Ensure the segment has been allocated */
-        if (!mmaps[i].base)
-        {
-            res = VPX_CODEC_MEM_ERROR;
-            break;
-        }
-
-        /* Verify variable size segment is big enough for the current si. */
-        if (vp8_mem_req_segs[i].calc_sz)
-        {
-            vpx_codec_dec_cfg_t cfg;
-
-            cfg.w = si->w;
-            cfg.h = si->h;
-
-            if (mmaps[i].sz < vp8_mem_req_segs[i].calc_sz(&cfg, init_flags))
-            {
-                res = VPX_CODEC_MEM_ERROR;
-                break;
-            }
-        }
-    }
-
-    return res;
-}
-
 static void vp8_init_ctx(vpx_codec_ctx_t *ctx, const vpx_codec_mmap_t *mmap)
 {
     int i;
@@ -178,16 +110,6 @@ static void vp8_init_ctx(vpx_codec_ctx_t *ctx, const vpx_codec_mmap_t *mmap)
     }
 }
 
-static void *mmap_lkup(vpx_codec_alg_priv_t *ctx, unsigned int id)
-{
-    int i;
-
-    for (i = 0; i < NELEMENTS(ctx->mmaps); i++)
-        if (ctx->mmaps[i].id == id)
-            return ctx->mmaps[i].base;
-
-    return NULL;
-}
 static void vp8_finalize_mmaps(vpx_codec_alg_priv_t *ctx)
 {
     /* nothing to clean up */
@@ -214,7 +136,7 @@ static vpx_codec_err_t vp8_init(vpx_codec_ctx_t *ctx,
         mmap.align = vp8_mem_req_segs[0].align;
         mmap.flags = vp8_mem_req_segs[0].flags;
 
-        res = vp8_mmap_alloc(&mmap);
+        res = vpx_mmap_alloc(&mmap);
         if (res != VPX_CODEC_OK) return res;
 
         vp8_init_ctx(ctx, &mmap);
@@ -366,8 +288,7 @@ static void yuvconfig2image(vpx_image_t               *img,
       * the Y, U, and V planes, nor other alignment adjustments that
       * might be representable by a YV12_BUFFER_CONFIG, so we just
       * initialize all the fields.*/
-    img->fmt = yv12->clrtype == REG_YUV ?
-        VPX_IMG_FMT_I420 : VPX_IMG_FMT_VPXI420;
+    img->fmt = VPX_IMG_FMT_I420;
     img->w = yv12->y_stride;
     img->h = (yv12->y_height + 2 * VP8BORDERINPIXELS + 15) & ~15;
     img->d_w = yv12->y_width;
@@ -488,7 +409,7 @@ static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t  *ctx,
                 ctx->mmaps[i].sz = vp8_mem_req_segs[i].calc_sz(&cfg,
                                    ctx->base.init_flags);
 
-            res = vp8_mmap_alloc(&ctx->mmaps[i]);
+            res = vpx_mmap_alloc(&ctx->mmaps[i]);
         }
 
         if (!res)
@@ -500,7 +421,9 @@ static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t  *ctx,
     /* Initialize the decoder instance on the first frame*/
     if (!res && !ctx->decoder_init)
     {
-        res = vp8_validate_mmaps(&ctx->si, ctx->mmaps, ctx->base.init_flags);
+        res = vpx_validate_mmaps(&ctx->si, ctx->mmaps,
+                                 vp8_mem_req_segs, NELEMENTS(vp8_mem_req_segs),
+                                 ctx->base.init_flags);
 
         if (!res)
         {
@@ -797,8 +720,6 @@ static vpx_codec_err_t image2yuvconfig(const vpx_image_t   *img,
     yv12->uv_stride = img->stride[VPX_PLANE_U];
 
     yv12->border  = (img->stride[VPX_PLANE_Y] - img->d_w) / 2;
-    yv12->clrtype = (img->fmt == VPX_IMG_FMT_VPXI420 || img->fmt == VPX_IMG_FMT_VPXYV12);
-
     return res;
 }
 
diff --git a/libvpx/vp8/vp8dx.mk b/libvpx/vp8/vp8dx.mk
index c26f42d..4a8f467 100644
--- a/libvpx/vp8/vp8dx.mk
+++ b/libvpx/vp8/vp8dx.mk
@@ -35,9 +35,5 @@ VP8_DX_SRCS-yes += decoder/onyxd_int.h
 VP8_DX_SRCS-yes += decoder/treereader.h
 VP8_DX_SRCS-yes += decoder/onyxd_if.c
 VP8_DX_SRCS-$(CONFIG_MULTITHREAD) += decoder/threading.c
-VP8_DX_SRCS-yes += decoder/vp8_asm_dec_offsets.c
 
 VP8_DX_SRCS-yes := $(filter-out $(VP8_DX_SRCS_REMOVE-yes),$(VP8_DX_SRCS-yes))
-
-$(eval $(call asm_offsets_template,\
-         vp8_asm_dec_offsets.asm, $(VP8_PREFIX)decoder/vp8_asm_dec_offsets.c))