Release 1.7.0 Mandarin Duck

Change-Id: I186440f3643a85694f45400393efb661f6d012fc
vpx_codec_enc_init_multi: fix segfault w/vp9
2018-01-24 14:25:44 -08:00 · 2018-01-23 18:14:23 -08:00 · 2018-01-23 18:14:14 -08:00 · 2018-01-17 20:17:33 +00:00 · 2018-01-17 11:28:24 -08:00 · 2018-01-16 18:27:42 -08:00
522 changed files with 86132 additions and 55585 deletions
--- a/.clang-format
+++ b/.clang-format
@@ -1,7 +1,7 @@
 ---
 Language:        Cpp
 # BasedOnStyle:  Google
-# Generated with clang-format 3.8.1
+# Generated with clang-format 4.0.1
 AccessModifierOffset: -1
 AlignAfterOpenBracket: Align
 AlignConsecutiveAssignments: false
@@ -37,6 +37,8 @@ BreakBeforeBinaryOperators: None
 BreakBeforeBraces: Attach
 BreakBeforeTernaryOperators: true
 BreakConstructorInitializersBeforeComma: false
+BreakAfterJavaFieldAnnotations: false
+BreakStringLiterals: true
 ColumnLimit:     80
 CommentPragmas:  '^ IWYU pragma:'
 ConstructorInitializerAllOnOneLineOrOnePerLine: false
@@ -54,9 +56,12 @@ IncludeCategories:
    Priority:        2
  - Regex:           '.*'
    Priority:        3
+IncludeIsMainRegex: '([-_](test|unittest))?$'
 IndentCaseLabels: true
 IndentWidth:     2
 IndentWrappedFunctionNames: false
+JavaScriptQuotes: Leave
+JavaScriptWrapImports: true
 KeepEmptyLinesAtTheStartOfBlocks: false
 MacroBlockBegin: ''
 MacroBlockEnd:   ''
@@ -75,6 +80,7 @@ PointerAlignment: Right
 ReflowComments:  true
 SortIncludes:    false
 SpaceAfterCStyleCast: false
+SpaceAfterTemplateKeyword: true
 SpaceBeforeAssignmentOperators: true
 SpaceBeforeParens: ControlStatements
 SpaceInEmptyParentheses: false
--- a/.mailmap
+++ b/.mailmap
@@ -3,6 +3,7 @@ Aℓex Converse <aconverse@google.com>
 Aℓex Converse <aconverse@google.com> <alex.converse@gmail.com>
 Alexis Ballier <aballier@gentoo.org> <alexis.ballier@gmail.com>
 Alpha Lam <hclam@google.com> <hclam@chromium.org>
+Chris Cunningham <chcunningham@chromium.org>
 Daniele Castagna <dcastagna@chromium.org> <dcastagna@google.com>
 Deb Mukherjee <debargha@google.com>
 Erik Niemeyer <erik.a.niemeyer@intel.com> <erik.a.niemeyer@gmail.com>
@@ -21,17 +22,21 @@ Marco Paniconi <marpan@google.com>
 Marco Paniconi <marpan@google.com> <marpan@chromium.org>
 Pascal Massimino <pascal.massimino@gmail.com>
 Paul Wilkins <paulwilkins@google.com>
+Peter Boström <pbos@chromium.org> <pbos@google.com>
 Peter de Rivaz <peter.derivaz@gmail.com>
 Peter de Rivaz <peter.derivaz@gmail.com> <peter.derivaz@argondesign.com>
 Ralph Giles <giles@xiph.org> <giles@entropywave.com>
 Ralph Giles <giles@xiph.org> <giles@mozilla.com>
 Ronald S. Bultje <rsbultje@gmail.com> <rbultje@google.com>
 Sami Pietilä <samipietila@google.com>
+Shiyou Yin <yinshiyou-hf@loongson.cn>
 Tamar Levy <tamar.levy@intel.com>
 Tamar Levy <tamar.levy@intel.com> <levytamar82@gmail.com>
 Tero Rintaluoma <teror@google.com> <tero.rintaluoma@on2.com>
 Timothy B. Terriberry <tterribe@xiph.org> <tterriberry@mozilla.com>
 Tom Finegan <tomfinegan@google.com>
 Tom Finegan <tomfinegan@google.com> <tomfinegan@chromium.org>
+Urvang Joshi <urvang@google.com> <urvang@chromium.org>
+Yaowu Xu <yaowu@google.com> <adam@xuyaowu.com>
 Yaowu Xu <yaowu@google.com> <yaowu@xuyaowu.com>
 Yaowu Xu <yaowu@google.com> <Yaowu Xu>
--- a/17
+++ b/17
@@ -3,13 +3,13 @@

 Aaron Watry <awatry@gmail.com>
 Abo Talib Mahfoodh <ab.mahfoodh@gmail.com>
-Adam Xu <adam@xuyaowu.com>
 Adrian Grange <agrange@google.com>
 Aℓex Converse <aconverse@google.com>
 Ahmad Sharif <asharif@google.com>
 Aleksey Vasenev <margtu-fivt@ya.ru>
 Alexander Potapenko <glider@google.com>
 Alexander Voronov <avoronov@graphics.cs.msu.ru>
+Alexandra Hájková <alexandra.khirnova@gmail.com>
 Alexis Ballier <aballier@gentoo.org>
 Alok Ahuja <waveletcoeff@gmail.com>
 Alpha Lam <hclam@google.com>
@@ -17,6 +17,7 @@ A.Mahfoodh <ab.mahfoodh@gmail.com>
 Ami Fischman <fischman@chromium.org>
 Andoni Morales Alastruey <ylatuya@gmail.com>
 Andres Mejia <mcitadel@gmail.com>
+Andrew Lewis <andrewlewis@google.com>
 Andrew Russell <anrussell@google.com>
 Angie Chiang <angiebird@google.com>
 Aron Rosenberg <arosenberg@logitech.com>
@@ -24,7 +25,9 @@ Attila Nagy <attilanagy@google.com>
 Brion Vibber <bvibber@wikimedia.org>
 changjun.yang <changjun.yang@intel.com>
 Charles 'Buck' Krasic <ckrasic@google.com>
+Cheng Chen <chengchen@google.com>
 chm <chm@rock-chips.com>
+Chris Cunningham <chcunningham@chromium.org>
 Christian Duvivier <cduvivier@google.com>
 Daniele Castagna <dcastagna@chromium.org>
 Daniel Kang <ddkang@google.com>
@@ -46,10 +49,12 @@ Geza Lore <gezalore@gmail.com>
 Ghislain MARY <ghislainmary2@gmail.com>
 Giuseppe Scrivano <gscrivano@gnu.org>
 Gordana Cmiljanovic <gordana.cmiljanovic@imgtec.com>
+Gregor Jasny <gjasny@gmail.com>
 Guillaume Martres <gmartres@google.com>
 Guillermo Ballester Valor <gbvalor@gmail.com>
 Hangyu Kuang <hkuang@google.com>
 Hanno Böck <hanno@hboeck.de>
+Han Shen <shenhan@google.com>
 Henrik Lundin <hlundin@google.com>
 Hui Su <huisu@google.com>
 Ivan Krasin <krasin@chromium.org>
@@ -83,6 +88,7 @@ Justin Clift <justin@salasaga.org>
 Justin Lebar <justin.lebar@gmail.com>
 Kaustubh Raste <kaustubh.raste@imgtec.com>
 KO Myung-Hun <komh@chollian.net>
+Kyle Siefring <kylesiefring@gmail.com>
 Lawrence Velázquez <larryv@macports.org>
 Linfeng Zhang <linfengz@google.com>
 Lou Quillio <louquillio@google.com>
@@ -101,6 +107,7 @@ Mikhal Shemer <mikhal@google.com>
 Min Chen <chenm003@gmail.com>
 Minghai Shang <minghai@google.com>
 Min Ye <yeemmi@google.com>
+Moriyoshi Koizumi <mozo@mozo.jp>
 Morton Jonuschat <yabawock@gmail.com>
 Nathan E. Egge <negge@mozilla.com>
 Nico Weber <thakis@chromium.org>
@@ -111,12 +118,15 @@ Paul Wilkins <paulwilkins@google.com>
 Pavol Rusnak <stick@gk2.sk>
 Paweł Hajdan <phajdan@google.com>
 Pengchong Jin <pengchong@google.com>
-Peter Boström <pbos@google.com>
+Peter Boström <pbos@chromium.org>
+Peter Collingbourne <pcc@chromium.org>
 Peter de Rivaz <peter.derivaz@gmail.com>
 Philip Jägenstedt <philipj@opera.com>
 Priit Laes <plaes@plaes.org>
 Rafael Ávila de Espíndola <rafael.espindola@gmail.com>
 Rafaël Carré <funman@videolan.org>
+Rafael de Lucena Valle <rafaeldelucena@gmail.com>
+Rahul Chaudhry <rahulchaudhry@google.com>
 Ralph Giles <giles@xiph.org>
 Ranjit Kumar Tulabandu <ranjit.tulabandu@ittiam.com>
 Rob Bradford <rob@linux.intel.com>
@@ -131,9 +141,11 @@ Sean McGovern <gseanmcg@gmail.com>
 Sergey Kolomenkin <kolomenkin@gmail.com>
 Sergey Ulanov <sergeyu@chromium.org>
 Shimon Doodkin <helpmepro1@gmail.com>
+Shiyou Yin <yinshiyou-hf@loongson.cn>
 Shunyao Li <shunyaoli@google.com>
 Stefan Holmer <holmer@google.com>
 Suman Sunkara <sunkaras@google.com>
+Sylvestre Ledru <sylvestre@mozilla.com>
 Taekhyun Kim <takim@nvidia.com>
 Takanori MATSUURA <t.matsuu@gmail.com>
 Tamar Levy <tamar.levy@intel.com>
@@ -146,6 +158,7 @@ Tom Finegan <tomfinegan@google.com>
 Tristan Matthews <le.businessman@gmail.com>
 Urvang Joshi <urvang@google.com>
 Vignesh Venkatasubramanian <vigneshv@google.com>
+Vlad Tsyrklevich <vtsyrklevich@chromium.org>
 Yaowu Xu <yaowu@google.com>
 Yi Luo <luoyi@google.com>
 Yongzhe Wang <yongzhe@google.com>
--- a/25
+++ b/25
@@ -1,3 +1,28 @@
+2017-01-04 v1.7.0 "Mandarin Duck"
+  This release focused on high bit depth performance (10/12 bit) and vp9
+  encoding improvements.
+
+  - Upgrading:
+    This release is ABI incompatible due to new vp9 encoder features.
+
+    Frame parallel decoding for vp9 has been removed.
+
+  - Enhancements:
+    vp9 encoding supports additional threads with --row-mt. This can be greater
+    than the number of tiles.
+
+    Two new vp9 encoder options have been added:
+      --corpus-complexity
+      --tune-content=film
+
+    Additional tooling for respecting the vp9 "level" profiles has been added.
+
+  - Bug fixes:
+    A variety of fuzzing issues.
+    vp8 threading fix for ARM.
+    Codec control VP9_SET_SKIP_LOOP_FILTER fixed.
+    Reject invalid multi resolution configurations.
+
 2017-01-09 v1.6.1 "Long Tailed Duck"
  This release improves upon the VP9 encoder and speeds up the encoding and
  decoding processes.
--- a/9
+++ b/9
@@ -1,4 +1,4 @@
-README - 9 January 2017
+README - 24 January 2018

 Welcome to the WebM VP8/VP9 Codec SDK!

@@ -58,10 +58,13 @@ COMPILING THE APPLICATIONS/LIBRARIES:
    armv7-win32-vs11
    armv7-win32-vs12
    armv7-win32-vs14
+    armv7-win32-vs15
    armv7s-darwin-gcc
    armv8-linux-gcc
    mips32-linux-gcc
    mips64-linux-gcc
+    ppc64-linux-gcc
+    ppc64le-linux-gcc
    sparc-solaris-gcc
    x86-android-gcc
    x86-darwin8-gcc
@@ -74,6 +77,7 @@ COMPILING THE APPLICATIONS/LIBRARIES:
    x86-darwin13-gcc
    x86-darwin14-gcc
    x86-darwin15-gcc
+    x86-darwin16-gcc
    x86-iphonesimulator-gcc
    x86-linux-gcc
    x86-linux-icc
@@ -84,6 +88,7 @@ COMPILING THE APPLICATIONS/LIBRARIES:
    x86-win32-vs11
    x86-win32-vs12
    x86-win32-vs14
+    x86-win32-vs15
    x86_64-android-gcc
    x86_64-darwin9-gcc
    x86_64-darwin10-gcc
@@ -92,6 +97,7 @@ COMPILING THE APPLICATIONS/LIBRARIES:
    x86_64-darwin13-gcc
    x86_64-darwin14-gcc
    x86_64-darwin15-gcc
+    x86_64-darwin16-gcc
    x86_64-iphonesimulator-gcc
    x86_64-linux-gcc
    x86_64-linux-icc
@@ -101,6 +107,7 @@ COMPILING THE APPLICATIONS/LIBRARIES:
    x86_64-win64-vs11
    x86_64-win64-vs12
    x86_64-win64-vs14
+    x86_64-win64-vs15
    generic-gnu

  The generic-gnu target, in conjunction with the CROSS environment variable,
--- a/build/make/Makefile
+++ b/build/make/Makefile
@@ -124,6 +124,7 @@ ifeq ($(TOOLCHAIN), x86-os2-gcc)
 CFLAGS += -mstackrealign
 endif

+# x86[_64]
 $(BUILD_PFX)%_mmx.c.d: CFLAGS += -mmmx
 $(BUILD_PFX)%_mmx.c.o: CFLAGS += -mmmx
 $(BUILD_PFX)%_sse2.c.d: CFLAGS += -msse2
@@ -138,6 +139,12 @@ $(BUILD_PFX)%_avx.c.d: CFLAGS += -mavx
 $(BUILD_PFX)%_avx.c.o: CFLAGS += -mavx
 $(BUILD_PFX)%_avx2.c.d: CFLAGS += -mavx2
 $(BUILD_PFX)%_avx2.c.o: CFLAGS += -mavx2
+$(BUILD_PFX)%_avx512.c.d: CFLAGS += -mavx512f -mavx512cd -mavx512bw -mavx512dq -mavx512vl
+$(BUILD_PFX)%_avx512.c.o: CFLAGS += -mavx512f -mavx512cd -mavx512bw -mavx512dq -mavx512vl
+
+# POWER
+$(BUILD_PFX)%_vsx.c.d: CFLAGS += -maltivec -mvsx
+$(BUILD_PFX)%_vsx.c.o: CFLAGS += -maltivec -mvsx

 $(BUILD_PFX)%.c.d: %.c
 	$(if $(quiet),@echo "    [DEP] $@")
--- a/build/make/configure.sh
+++ b/build/make/configure.sh
@@ -403,6 +403,23 @@ check_gcc_machine_option() {
  fi
 }

+# tests for -m$2, -m$3, -m$4... toggling the feature given in $1.
+check_gcc_machine_options() {
+  feature="$1"
+  shift
+  flags="-m$1"
+  shift
+  for opt in $*; do
+    flags="$flags -m$opt"
+  done
+
+  if enabled gcc && ! disabled "$feature" && ! check_cflags $flags; then
+    RTCD_OPTIONS="${RTCD_OPTIONS}--disable-$feature "
+  else
+    soft_enable "$feature"
+  fi
+}
+
 write_common_config_banner() {
  print_webm_license config.mk "##" ""
  echo '# This file automatically generated by configure. Do not edit!' >> config.mk
@@ -674,7 +691,6 @@ check_xcode_minimum_version() {
 process_common_toolchain() {
  if [ -z "$toolchain" ]; then
    gcctarget="${CHOST:-$(gcc -dumpmachine 2> /dev/null)}"
-
    # detect tgt_isa
    case "$gcctarget" in
      aarch64*)
@@ -697,6 +713,18 @@ process_common_toolchain() {
      *sparc*)
        tgt_isa=sparc
        ;;
+      power*64*-*)
+        tgt_isa=ppc64
+        ;;
+      power*)
+        tgt_isa=ppc
+        ;;
+      *mips64el*)
+        tgt_isa=mips64
+        ;;
+      *mips32el*)
+        tgt_isa=mips32
+        ;;
    esac

    # detect tgt_os
@@ -725,9 +753,16 @@ process_common_toolchain() {
        tgt_isa=x86_64
        tgt_os=darwin15
        ;;
+      *darwin16*)
+        tgt_isa=x86_64
+        tgt_os=darwin16
+        ;;
      x86_64*mingw32*)
        tgt_os=win64
        ;;
+      x86_64*cygwin*)
+        tgt_os=win64
+        ;;
      *mingw32*|*cygwin*)
        [ -z "$tgt_isa" ] && tgt_isa=x86
        tgt_os=win32
@@ -775,6 +810,9 @@ process_common_toolchain() {
    mips*)
      enable_feature mips
      ;;
+    ppc*)
+      enable_feature ppc
+      ;;
  esac

  # PIC is probably what we want when building shared libs
@@ -843,6 +881,10 @@ process_common_toolchain() {
      add_cflags  "-mmacosx-version-min=10.11"
      add_ldflags "-mmacosx-version-min=10.11"
      ;;
+    *-darwin16-*)
+      add_cflags  "-mmacosx-version-min=10.12"
+      add_ldflags "-mmacosx-version-min=10.12"
+      ;;
    *-iphonesimulator-*)
      add_cflags  "-miphoneos-version-min=${IOS_VERSION_MIN}"
      add_ldflags "-miphoneos-version-min=${IOS_VERSION_MIN}"
@@ -1144,10 +1186,20 @@ EOF
        fi
      fi

+      if enabled mmi; then
+        tgt_isa=loongson3a
+        check_add_ldflags -march=loongson3a
+      fi
+
      check_add_cflags -march=${tgt_isa}
      check_add_asflags -march=${tgt_isa}
      check_add_asflags -KPIC
      ;;
+    ppc*)
+      link_with_cc=gcc
+      setup_gnu_toolchain
+      check_gcc_machine_option "vsx"
+      ;;
    x86*)
      case  ${tgt_os} in
        win*)
@@ -1202,6 +1254,13 @@ EOF
          AS=msvs
          msvs_arch_dir=x86-msvs
          vc_version=${tgt_cc##vs}
+          case $vc_version in
+            7|8|9|10|11|12|13|14)
+              echo "${tgt_cc} does not support avx512, disabling....."
+              RTCD_OPTIONS="${RTCD_OPTIONS}--disable-avx512 "
+              soft_disable avx512
+              ;;
+          esac
          case $vc_version in
            7|8|9|10)
              echo "${tgt_cc} does not support avx/avx2, disabling....."
@@ -1246,9 +1305,18 @@ EOF
        elif disabled $ext; then
          disable_exts="yes"
        else
-          # use the shortened version for the flag: sse4_1 -> sse4
-          check_gcc_machine_option ${ext%_*} $ext
+          if [ "$ext" = "avx512" ]; then
+            check_gcc_machine_options $ext avx512f avx512cd avx512bw avx512dq avx512vl
+          else
+            # use the shortened version for the flag: sse4_1 -> sse4
+            check_gcc_machine_option ${ext%_*} $ext
+          fi
        fi
+
+        # https://bugs.chromium.org/p/webm/issues/detail?id=1464
+        # The assembly optimizations for vpx_sub_pixel_variance do not link with
+        # gcc 6.
+        enabled sse2 && soft_enable pic
      done

      if enabled external_build; then
@@ -1273,7 +1341,6 @@ EOF
        esac
        log_echo "  using $AS"
      fi
-      [ "${AS##*/}" = nasm ] && add_asflags -Ox
      AS_SFX=.asm
      case  ${tgt_os} in
        win32)
@@ -1282,7 +1349,7 @@ EOF
          EXE_SFX=.exe
          ;;
        win64)
-          add_asflags -f x64
+          add_asflags -f win64
          enabled debug && add_asflags -g cv8
          EXE_SFX=.exe
          ;;
@@ -1416,6 +1483,10 @@ EOF
          echo "msa optimizations are available only for little endian platforms"
          disable_feature msa
        fi
+        if enabled mmi; then
+          echo "mmi optimizations are available only for little endian platforms"
+          disable_feature mmi
+        fi
      fi
      ;;
  esac
--- a/build/make/gen_msvs_sln.sh
+++ b/build/make/gen_msvs_sln.sh
@@ -25,7 +25,7 @@ files.
 Options:
    --help                      Print this message
    --out=outfile               Redirect output to a file
-    --ver=version               Version (7,8,9,10,11,12,14) of visual studio to generate for
+    --ver=version               Version (7,8,9,10,11,12,14,15) of visual studio to generate for
    --target=isa-os-cc          Target specifier
 EOF
    exit 1
@@ -215,7 +215,7 @@ for opt in "$@"; do
    ;;
    --ver=*) vs_ver="$optval"
             case $optval in
-             10|11|12|14)
+             10|11|12|14|15)
             ;;
             *) die Unrecognized Visual Studio Version in $opt
             ;;
@@ -240,9 +240,12 @@ case "${vs_ver:-10}" in
    12) sln_vers="12.00"
       sln_vers_str="Visual Studio 2013"
    ;;
-    14) sln_vers="14.00"
+    14) sln_vers="12.00"
       sln_vers_str="Visual Studio 2015"
    ;;
+    15) sln_vers="12.00"
+       sln_vers_str="Visual Studio 2017"
+    ;;
 esac
 sfx=vcxproj

--- a/build/make/gen_msvs_vcxproj.sh
+++ b/build/make/gen_msvs_vcxproj.sh
@@ -34,7 +34,7 @@ Options:
    --name=project_name         Name of the project (required)
    --proj-guid=GUID            GUID to use for the project
    --module-def=filename       File containing export definitions (for DLLs)
-    --ver=version               Version (10,11,12,14) of visual studio to generate for
+    --ver=version               Version (10,11,12,14,15) of visual studio to generate for
    --src-path-bare=dir         Path to root of source tree
    -Ipath/to/include           Additional include directories
    -DFLAG[=value]              Preprocessor macros to define
@@ -168,7 +168,7 @@ for opt in "$@"; do
        --ver=*)
            vs_ver="$optval"
            case "$optval" in
-                10|11|12|14)
+                10|11|12|14|15)
                ;;
                *) die Unrecognized Visual Studio Version in $opt
                ;;
@@ -218,7 +218,7 @@ guid=${guid:-`generate_uuid`}
 asm_use_custom_step=false
 uses_asm=${uses_asm:-false}
 case "${vs_ver:-11}" in
-    10|11|12|14)
+    10|11|12|14|15)
       asm_use_custom_step=$uses_asm
    ;;
 esac
@@ -347,6 +347,9 @@ generate_vcxproj() {
            if [ "$vs_ver" = "14" ]; then
                tag_content PlatformToolset v140
            fi
+            if [ "$vs_ver" = "15" ]; then
+                tag_content PlatformToolset v141
+            fi
            tag_content CharacterSet Unicode
            if [ "$config" = "Release" ]; then
                tag_content WholeProgramOptimization true
--- a/build/make/iosbuild.sh
+++ b/build/make/iosbuild.sh
@@ -35,8 +35,8 @@ ARM_TARGETS="arm64-darwin-gcc
             armv7s-darwin-gcc"
 SIM_TARGETS="x86-iphonesimulator-gcc
             x86_64-iphonesimulator-gcc"
-OSX_TARGETS="x86-darwin15-gcc
-             x86_64-darwin15-gcc"
+OSX_TARGETS="x86-darwin16-gcc
+             x86_64-darwin16-gcc"
 TARGETS="${ARM_TARGETS} ${SIM_TARGETS}"

 # Configures for the target specified by $1, and invokes make with the dist
@@ -271,7 +271,7 @@ cat << EOF
    --help: Display this message and exit.
    --enable-shared: Build a dynamic framework for use on iOS 8 or later.
    --extra-configure-args <args>: Extra args to pass when configuring libvpx.
-    --macosx: Uses darwin15 targets instead of iphonesimulator targets for x86
+    --macosx: Uses darwin16 targets instead of iphonesimulator targets for x86
              and x86_64. Allows linking to framework when builds target MacOSX
              instead of iOS.
    --preserve-build-output: Do not delete the build directory.
--- a/build/make/rtcd.pl
+++ b/build/make/rtcd.pl
@@ -1,4 +1,13 @@
 #!/usr/bin/env perl
+##
+##  Copyright (c) 2017 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.
+##

 no strict 'refs';
 use warnings;
@@ -200,6 +209,7 @@ sub filter {
 sub common_top() {
  my $include_guard = uc($opts{sym})."_H_";
  print <<EOF;
+// This file is generated. Do not edit.
 #ifndef ${include_guard}
 #define ${include_guard}

@@ -335,6 +345,36 @@ EOF
  common_bottom;
 }

+sub ppc() {
+  determine_indirection("c", @ALL_ARCHS);
+
+  # Assign the helper variable for each enabled extension
+  foreach my $opt (@ALL_ARCHS) {
+    my $opt_uc = uc $opt;
+    eval "\$have_${opt}=\"flags & HAS_${opt_uc}\"";
+  }
+
+  common_top;
+  print <<EOF;
+#include "vpx_config.h"
+
+#ifdef RTCD_C
+#include "vpx_ports/ppc.h"
+static void setup_rtcd_internal(void)
+{
+    int flags = ppc_simd_caps();
+    (void)flags;
+EOF
+
+  set_function_pointers("c", @ALL_ARCHS);
+
+  print <<EOF;
+}
+#endif
+EOF
+  common_bottom;
+}
+
 sub unoptimized() {
  determine_indirection "c";
  common_top;
@@ -361,10 +401,10 @@ EOF

 &require("c");
 if ($opts{arch} eq 'x86') {
-  @ALL_ARCHS = filter(qw/mmx sse sse2 sse3 ssse3 sse4_1 avx avx2/);
+  @ALL_ARCHS = filter(qw/mmx sse sse2 sse3 ssse3 sse4_1 avx avx2 avx512/);
  x86;
 } elsif ($opts{arch} eq 'x86_64') {
-  @ALL_ARCHS = filter(qw/mmx sse sse2 sse3 ssse3 sse4_1 avx avx2/);
+  @ALL_ARCHS = filter(qw/mmx sse sse2 sse3 ssse3 sse4_1 avx avx2 avx512/);
  @REQUIRES = filter(keys %required ? keys %required : qw/mmx sse sse2/);
  &require(@REQUIRES);
  x86;
@@ -381,6 +421,10 @@ if ($opts{arch} eq 'x86') {
      @ALL_ARCHS = filter("$opts{arch}", qw/msa/);
      last;
    }
+    if (/HAVE_MMI=yes/) {
+      @ALL_ARCHS = filter("$opts{arch}", qw/mmi/);
+      last;
+    }
  }
  close CONFIG_FILE;
  mips;
@@ -390,6 +434,9 @@ if ($opts{arch} eq 'x86') {
 } elsif ($opts{arch} eq 'armv8' || $opts{arch} eq 'arm64' ) {
  @ALL_ARCHS = filter(qw/neon/);
  arm;
+} elsif ($opts{arch} =~ /^ppc/ ) {
+  @ALL_ARCHS = filter(qw/vsx/);
+  ppc;
 } else {
  unoptimized;
 }
--- a/build/make/version.sh
+++ b/build/make/version.sh
@@ -60,6 +60,7 @@ if [ ${bare} ]; then
    echo "${changelog_version}${git_version_id}" > $$.tmp
 else
    cat<<EOF>$$.tmp
+// This file is generated. Do not edit.
 #define VERSION_MAJOR  $major_version
 #define VERSION_MINOR  $minor_version
 #define VERSION_PATCH  $patch_version
--- a/37
+++ b/37
@@ -109,10 +109,13 @@ all_platforms="${all_platforms} armv7-none-rvct"     #neon Cortex-A8
 all_platforms="${all_platforms} armv7-win32-vs11"
 all_platforms="${all_platforms} armv7-win32-vs12"
 all_platforms="${all_platforms} armv7-win32-vs14"
+all_platforms="${all_platforms} armv7-win32-vs15"
 all_platforms="${all_platforms} armv7s-darwin-gcc"
 all_platforms="${all_platforms} armv8-linux-gcc"
 all_platforms="${all_platforms} mips32-linux-gcc"
 all_platforms="${all_platforms} mips64-linux-gcc"
+all_platforms="${all_platforms} ppc64-linux-gcc"
+all_platforms="${all_platforms} ppc64le-linux-gcc"
 all_platforms="${all_platforms} sparc-solaris-gcc"
 all_platforms="${all_platforms} x86-android-gcc"
 all_platforms="${all_platforms} x86-darwin8-gcc"
@@ -125,6 +128,7 @@ all_platforms="${all_platforms} x86-darwin12-gcc"
 all_platforms="${all_platforms} x86-darwin13-gcc"
 all_platforms="${all_platforms} x86-darwin14-gcc"
 all_platforms="${all_platforms} x86-darwin15-gcc"
+all_platforms="${all_platforms} x86-darwin16-gcc"
 all_platforms="${all_platforms} x86-iphonesimulator-gcc"
 all_platforms="${all_platforms} x86-linux-gcc"
 all_platforms="${all_platforms} x86-linux-icc"
@@ -135,6 +139,7 @@ all_platforms="${all_platforms} x86-win32-vs10"
 all_platforms="${all_platforms} x86-win32-vs11"
 all_platforms="${all_platforms} x86-win32-vs12"
 all_platforms="${all_platforms} x86-win32-vs14"
+all_platforms="${all_platforms} x86-win32-vs15"
 all_platforms="${all_platforms} x86_64-android-gcc"
 all_platforms="${all_platforms} x86_64-darwin9-gcc"
 all_platforms="${all_platforms} x86_64-darwin10-gcc"
@@ -143,6 +148,7 @@ all_platforms="${all_platforms} x86_64-darwin12-gcc"
 all_platforms="${all_platforms} x86_64-darwin13-gcc"
 all_platforms="${all_platforms} x86_64-darwin14-gcc"
 all_platforms="${all_platforms} x86_64-darwin15-gcc"
+all_platforms="${all_platforms} x86_64-darwin16-gcc"
 all_platforms="${all_platforms} x86_64-iphonesimulator-gcc"
 all_platforms="${all_platforms} x86_64-linux-gcc"
 all_platforms="${all_platforms} x86_64-linux-icc"
@@ -152,6 +158,7 @@ all_platforms="${all_platforms} x86_64-win64-vs10"
 all_platforms="${all_platforms} x86_64-win64-vs11"
 all_platforms="${all_platforms} x86_64-win64-vs12"
 all_platforms="${all_platforms} x86_64-win64-vs14"
+all_platforms="${all_platforms} x86_64-win64-vs15"
 all_platforms="${all_platforms} generic-gnu"

 # all_targets is a list of all targets that can be configured
@@ -163,11 +170,14 @@ for t in ${all_targets}; do
    [ -f "${source_path}/${t}.mk" ] && enable_feature ${t}
 done

+if ! diff --version >/dev/null; then
+  die "diff missing: Try installing diffutils via your package manager."
+fi
+
 if ! perl --version >/dev/null; then
    die "Perl is required to build"
 fi

-
 if [ "`cd \"${source_path}\" && pwd`" != "`pwd`" ]; then
  # test to see if source_path already configured
  if [ -f "${source_path}/vpx_config.h" ]; then
@@ -223,6 +233,7 @@ ARCH_LIST="
    mips
    x86
    x86_64
+    ppc
 "
 ARCH_EXT_LIST_X86="
    mmx
@@ -233,7 +244,13 @@ ARCH_EXT_LIST_X86="
    sse4_1
    avx
    avx2
+    avx512
 "
+
+ARCH_EXT_LIST_LOONGSON="
+    mmi
+"
+
 ARCH_EXT_LIST="
    neon
    neon_asm
@@ -244,6 +261,10 @@ ARCH_EXT_LIST="
    mips64

    ${ARCH_EXT_LIST_X86}
+
+    vsx
+
+    ${ARCH_EXT_LIST_LOONGSON}
 "
 HAVE_LIST="
    ${ARCH_EXT_LIST}
@@ -255,7 +276,6 @@ EXPERIMENT_LIST="
    spatial_svc
    fp_mb_stats
    emulate_hardware
-    misc_fixes
 "
 CONFIG_LIST="
    dependency_tracking
@@ -310,6 +330,7 @@ CONFIG_LIST="
    better_hw_compatibility
    experimental
    size_limit
+    always_adjust_bpm
    ${EXPERIMENT_LIST}
 "
 CMDLINE_SELECT="
@@ -369,6 +390,7 @@ CMDLINE_SELECT="
    better_hw_compatibility
    vp9_highbitdepth
    experimental
+    always_adjust_bpm
 "

 process_cmdline() {
@@ -570,6 +592,7 @@ process_toolchain() {
        check_add_cflags -Wdeclaration-after-statement
        check_add_cflags -Wdisabled-optimization
        check_add_cflags -Wfloat-conversion
+        check_add_cflags -Wparentheses-equality
        check_add_cflags -Wpointer-arith
        check_add_cflags -Wtype-limits
        check_add_cflags -Wcast-qual
@@ -588,11 +611,9 @@ process_toolchain() {
        if enabled mips || [ -z "${INLINE}" ]; then
          enabled extra_warnings || check_add_cflags -Wno-unused-function
        fi
-        if ! enabled vp9_highbitdepth; then
-          # Avoid this warning for third_party C++ sources. Some reorganization
-          # would be needed to apply this only to test/*.cc.
-          check_cflags -Wshorten-64-to-32 && add_cflags_only -Wshorten-64-to-32
-        fi
+        # Avoid this warning for third_party C++ sources. Some reorganization
+        # would be needed to apply this only to test/*.cc.
+        check_cflags -Wshorten-64-to-32 && add_cflags_only -Wshorten-64-to-32
    fi

    if enabled icc; then
@@ -644,7 +665,7 @@ process_toolchain() {
             gen_vcproj_cmd=${source_path}/build/make/gen_msvs_vcxproj.sh
             enabled werror && gen_vcproj_cmd="${gen_vcproj_cmd} --enable-werror"
             all_targets="${all_targets} solution"
-             INLINE="__forceinline"
+             INLINE="__inline"
        ;;
    esac

--- a/examples/vp8_multi_resolution_encoder.c
+++ b/examples/vp8_multi_resolution_encoder.c
@@ -151,7 +151,7 @@ static void write_ivf_frame_header(FILE *outfile,
  if (pkt->kind != VPX_CODEC_CX_FRAME_PKT) return;

  pts = pkt->data.frame.pts;
-  mem_put_le32(header, pkt->data.frame.sz);
+  mem_put_le32(header, (int)pkt->data.frame.sz);
  mem_put_le32(header + 4, pts & 0xFFFFFFFF);
  mem_put_le32(header + 8, pts >> 32);

@@ -190,7 +190,7 @@ static void set_temporal_layer_pattern(int num_temporal_layers,
      cfg->ts_layer_id[0] = 0;
      cfg->ts_layer_id[1] = 1;
      // Use 60/40 bit allocation as example.
-      cfg->ts_target_bitrate[0] = 0.6f * bitrate;
+      cfg->ts_target_bitrate[0] = (int)(0.6f * bitrate);
      cfg->ts_target_bitrate[1] = bitrate;

      /* 0=L, 1=GF */
@@ -241,8 +241,8 @@ static void set_temporal_layer_pattern(int num_temporal_layers,
      cfg->ts_layer_id[2] = 1;
      cfg->ts_layer_id[3] = 2;
      // Use 45/20/35 bit allocation as example.
-      cfg->ts_target_bitrate[0] = 0.45f * bitrate;
-      cfg->ts_target_bitrate[1] = 0.65f * bitrate;
+      cfg->ts_target_bitrate[0] = (int)(0.45f * bitrate);
+      cfg->ts_target_bitrate[1] = (int)(0.65f * bitrate);
      cfg->ts_target_bitrate[2] = bitrate;

      /* 0=L, 1=GF, 2=ARF */
@@ -294,8 +294,8 @@ int main(int argc, char **argv) {
  vpx_codec_err_t res[NUM_ENCODERS];

  int i;
-  long width;
-  long height;
+  int width;
+  int height;
  int length_frame;
  int frame_avail;
  int got_data;
@@ -347,9 +347,9 @@ int main(int argc, char **argv) {

  printf("Using %s\n", vpx_codec_iface_name(interface));

-  width = strtol(argv[1], NULL, 0);
-  height = strtol(argv[2], NULL, 0);
-  framerate = strtol(argv[3], NULL, 0);
+  width = (int)strtol(argv[1], NULL, 0);
+  height = (int)strtol(argv[2], NULL, 0);
+  framerate = (int)strtol(argv[3], NULL, 0);

  if (width < 16 || width % 2 || height < 16 || height % 2)
    die("Invalid resolution: %ldx%ld", width, height);
@@ -371,12 +371,13 @@ int main(int argc, char **argv) {

  // Bitrates per spatial layer: overwrite default rates above.
  for (i = 0; i < NUM_ENCODERS; i++) {
-    target_bitrate[i] = strtol(argv[NUM_ENCODERS + 5 + i], NULL, 0);
+    target_bitrate[i] = (int)strtol(argv[NUM_ENCODERS + 5 + i], NULL, 0);
  }

  // Temporal layers per spatial layers: overwrite default settings above.
  for (i = 0; i < NUM_ENCODERS; i++) {
-    num_temporal_layers[i] = strtol(argv[2 * NUM_ENCODERS + 5 + i], NULL, 0);
+    num_temporal_layers[i] =
+        (int)strtol(argv[2 * NUM_ENCODERS + 5 + i], NULL, 0);
    if (num_temporal_layers[i] < 1 || num_temporal_layers[i] > 3)
      die("Invalid temporal layers: %d, Must be 1, 2, or 3. \n",
          num_temporal_layers);
@@ -391,9 +392,9 @@ int main(int argc, char **argv) {
    downsampled_input[i] = fopen(filename, "wb");
  }

-  key_frame_insert = strtol(argv[3 * NUM_ENCODERS + 5], NULL, 0);
+  key_frame_insert = (int)strtol(argv[3 * NUM_ENCODERS + 5], NULL, 0);

-  show_psnr = strtol(argv[3 * NUM_ENCODERS + 6], NULL, 0);
+  show_psnr = (int)strtol(argv[3 * NUM_ENCODERS + 6], NULL, 0);

  /* Populate default encoder configuration */
  for (i = 0; i < NUM_ENCODERS; i++) {
@@ -469,7 +470,7 @@ int main(int argc, char **argv) {
    if (!vpx_img_alloc(&raw[i], VPX_IMG_FMT_I420, cfg[i].g_w, cfg[i].g_h, 32))
      die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h);

-  if (raw[0].stride[VPX_PLANE_Y] == raw[0].d_w)
+  if (raw[0].stride[VPX_PLANE_Y] == (int)raw[0].d_w)
    read_frame_p = read_frame;
  else
    read_frame_p = read_frame_by_row;
@@ -558,7 +559,8 @@ int main(int argc, char **argv) {
        /* Write out down-sampled input. */
        length_frame = cfg[i].g_w * cfg[i].g_h * 3 / 2;
        if (fwrite(raw[i].planes[0], 1, length_frame,
-                   downsampled_input[NUM_ENCODERS - i - 1]) != length_frame) {
+                   downsampled_input[NUM_ENCODERS - i - 1]) !=
+            (unsigned int)length_frame) {
          return EXIT_FAILURE;
        }
      }
@@ -619,10 +621,6 @@ int main(int argc, char **argv) {
            break;
          default: break;
        }
-        printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT &&
-                       (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)
-                   ? "K"
-                   : "");
        fflush(stdout);
      }
    }
@@ -663,7 +661,6 @@ int main(int argc, char **argv) {
      write_ivf_file_header(outfile[i], &cfg[i], frame_cnt - 1);
    fclose(outfile[i]);
  }
-  printf("\n");

  return EXIT_SUCCESS;
 }
--- a/examples/vp9_spatial_svc_encoder.c
+++ b/examples/vp9_spatial_svc_encoder.c
@@ -168,7 +168,7 @@ void usage_exit(void) {
 static void parse_command_line(int argc, const char **argv_,
                               AppInput *app_input, SvcContext *svc_ctx,
                               vpx_codec_enc_cfg_t *enc_cfg) {
-  struct arg arg = { 0 };
+  struct arg arg;
  char **argv = NULL;
  char **argi = NULL;
  char **argj = NULL;
@@ -509,7 +509,7 @@ static void printout_rate_control_summary(struct RateControlStats *rc,
 }

 vpx_codec_err_t parse_superframe_index(const uint8_t *data, size_t data_sz,
-                                       uint32_t sizes[8], int *count) {
+                                       uint64_t sizes[8], int *count) {
  // A chunk ending with a byte matching 0xc0 is an invalid chunk unless
  // it is a super frame index. If the last byte of real video compression
  // data is 0xc0 the encoder must add a 0 byte. If we have the marker but
@@ -606,9 +606,9 @@ void set_frame_flags_bypass_mode(int sl, int tl, int num_spatial_layers,
 }

 int main(int argc, const char **argv) {
-  AppInput app_input = { 0 };
+  AppInput app_input;
  VpxVideoWriter *writer = NULL;
-  VpxVideoInfo info = { 0 };
+  VpxVideoInfo info;
  vpx_codec_ctx_t codec;
  vpx_codec_enc_cfg_t enc_cfg;
  SvcContext svc_ctx;
@@ -640,8 +640,9 @@ int main(int argc, const char **argv) {

 // Allocate image buffer
 #if CONFIG_VP9_HIGHBITDEPTH
-  if (!vpx_img_alloc(&raw, enc_cfg.g_input_bit_depth == 8 ? VPX_IMG_FMT_I420
-                                                          : VPX_IMG_FMT_I42016,
+  if (!vpx_img_alloc(&raw,
+                     enc_cfg.g_input_bit_depth == 8 ? VPX_IMG_FMT_I420
+                                                    : VPX_IMG_FMT_I42016,
                     enc_cfg.g_w, enc_cfg.g_h, 32)) {
    die("Failed to allocate image %dx%d\n", enc_cfg.g_w, enc_cfg.g_h);
  }
@@ -697,12 +698,18 @@ int main(int argc, const char **argv) {

  if (svc_ctx.speed != -1)
    vpx_codec_control(&codec, VP8E_SET_CPUUSED, svc_ctx.speed);
-  if (svc_ctx.threads)
+  if (svc_ctx.threads) {
    vpx_codec_control(&codec, VP9E_SET_TILE_COLUMNS, (svc_ctx.threads >> 1));
+    if (svc_ctx.threads > 1)
+      vpx_codec_control(&codec, VP9E_SET_ROW_MT, 1);
+    else
+      vpx_codec_control(&codec, VP9E_SET_ROW_MT, 0);
+  }
  if (svc_ctx.speed >= 5 && svc_ctx.aqmode == 1)
    vpx_codec_control(&codec, VP9E_SET_AQ_MODE, 3);
  if (svc_ctx.speed >= 5)
    vpx_codec_control(&codec, VP8E_SET_STATIC_THRESHOLD, 1);
+  vpx_codec_control(&codec, VP8E_SET_MAX_INTRA_BITRATE_PCT, 900);

  // Encode frames
  while (!end_of_stream) {
@@ -763,7 +770,7 @@ int main(int argc, const char **argv) {
          SvcInternal_t *const si = (SvcInternal_t *)svc_ctx.internal;
          if (cx_pkt->data.frame.sz > 0) {
 #if OUTPUT_RC_STATS
-            uint32_t sizes[8];
+            uint64_t sizes[8];
            int count = 0;
 #endif
            vpx_video_writer_write_frame(writer, cx_pkt->data.frame.buf,
@@ -775,6 +782,8 @@ int main(int argc, const char **argv) {
              vpx_codec_control(&codec, VP9E_GET_SVC_LAYER_ID, &layer_id);
              parse_superframe_index(cx_pkt->data.frame.buf,
                                     cx_pkt->data.frame.sz, sizes, &count);
+              if (enc_cfg.ss_number_layers == 1)
+                sizes[0] = cx_pkt->data.frame.sz;
              // Note computing input_layer_frames here won't account for frame
              // drops in rate control stats.
              // TODO(marpan): Fix this for non-bypass mode so we can get stats
--- a/examples/vpx_temporal_svc_encoder.c
+++ b/examples/vpx_temporal_svc_encoder.c
@@ -26,17 +26,27 @@
 #include "../tools_common.h"
 #include "../video_writer.h"

+#define VP8_ROI_MAP 0
+
 static const char *exec_name;

 void usage_exit(void) { exit(EXIT_FAILURE); }

-// Denoiser states, for temporal denoising.
-enum denoiserState {
-  kDenoiserOff,
-  kDenoiserOnYOnly,
-  kDenoiserOnYUV,
-  kDenoiserOnYUVAggressive,
-  kDenoiserOnAdaptive
+// Denoiser states for vp8, for temporal denoising.
+enum denoiserStateVp8 {
+  kVp8DenoiserOff,
+  kVp8DenoiserOnYOnly,
+  kVp8DenoiserOnYUV,
+  kVp8DenoiserOnYUVAggressive,
+  kVp8DenoiserOnAdaptive
+};
+
+// Denoiser states for vp9, for temporal denoising.
+enum denoiserStateVp9 {
+  kVp9DenoiserOff,
+  kVp9DenoiserOnYOnly,
+  // For SVC: denoise the top two spatial layers.
+  kVp9DenoiserOnYTwoSpatialLayers
 };

 static int mode_to_num_layers[13] = { 1, 2, 2, 3, 3, 3, 3, 5, 2, 3, 3, 3, 3 };
@@ -154,6 +164,53 @@ static void printout_rate_control_summary(struct RateControlMetrics *rc,
    die("Error: Number of input frames not equal to output! \n");
 }

+#if VP8_ROI_MAP
+static void vp8_set_roi_map(vpx_codec_enc_cfg_t *cfg, vpx_roi_map_t *roi) {
+  unsigned int i, j;
+  memset(roi, 0, sizeof(*roi));
+
+  // ROI is based on the segments (4 for vp8, 8 for vp9), smallest unit for
+  // segment is 16x16 for vp8, 8x8 for vp9.
+  roi->rows = (cfg->g_h + 15) / 16;
+  roi->cols = (cfg->g_w + 15) / 16;
+
+  // Applies delta QP on the segment blocks, varies from -63 to 63.
+  // Setting to negative means lower QP (better quality).
+  // Below we set delta_q to the extreme (-63) to show strong effect.
+  roi->delta_q[0] = 0;
+  roi->delta_q[1] = -63;
+  roi->delta_q[2] = 0;
+  roi->delta_q[3] = 0;
+
+  // Applies delta loopfilter strength on the segment blocks, varies from -63 to
+  // 63. Setting to positive means stronger loopfilter.
+  roi->delta_lf[0] = 0;
+  roi->delta_lf[1] = 0;
+  roi->delta_lf[2] = 0;
+  roi->delta_lf[3] = 0;
+
+  // Applies skip encoding threshold on the segment blocks, varies from 0 to
+  // UINT_MAX. Larger value means more skipping of encoding is possible.
+  // This skip threshold only applies on delta frames.
+  roi->static_threshold[0] = 0;
+  roi->static_threshold[1] = 0;
+  roi->static_threshold[2] = 0;
+  roi->static_threshold[3] = 0;
+
+  // Use 2 states: 1 is center square, 0 is the rest.
+  roi->roi_map =
+      (uint8_t *)calloc(roi->rows * roi->cols, sizeof(*roi->roi_map));
+  for (i = 0; i < roi->rows; ++i) {
+    for (j = 0; j < roi->cols; ++j) {
+      if (i > (roi->rows >> 2) && i < ((roi->rows * 3) >> 2) &&
+          j > (roi->cols >> 2) && j < ((roi->cols * 3) >> 2)) {
+        roi->roi_map[i * roi->cols + j] = 1;
+      }
+    }
+  }
+}
+#endif
+
 // Temporal scaling parameters:
 // NOTE: The 3 prediction frames cannot be used interchangeably due to
 // differences in the way they are handled throughout the code. The
@@ -495,6 +552,7 @@ int main(int argc, char **argv) {
  vpx_codec_err_t res;
  unsigned int width;
  unsigned int height;
+  uint32_t error_resilient = 0;
  int speed;
  int frame_avail;
  int got_data;
@@ -505,16 +563,15 @@ int main(int argc, char **argv) {
  int layering_mode = 0;
  int layer_flags[VPX_TS_MAX_PERIODICITY] = { 0 };
  int flag_periodicity = 1;
-#if VPX_ENCODER_ABI_VERSION > (4 + VPX_CODEC_ABI_VERSION)
-  vpx_svc_layer_id_t layer_id = { 0, 0 };
-#else
-  vpx_svc_layer_id_t layer_id = { 0 };
+#if VP8_ROI_MAP
+  vpx_roi_map_t roi;
 #endif
+  vpx_svc_layer_id_t layer_id = { 0, 0 };
  const VpxInterface *encoder = NULL;
  FILE *infile = NULL;
  struct RateControlMetrics rc;
  int64_t cx_time = 0;
-  const int min_args_base = 12;
+  const int min_args_base = 13;
 #if CONFIG_VP9_HIGHBITDEPTH
  vpx_bit_depth_t bit_depth = VPX_BITS_8;
  int input_bit_depth = 8;
@@ -531,12 +588,14 @@ int main(int argc, char **argv) {
  if (argc < min_args) {
 #if CONFIG_VP9_HIGHBITDEPTH
    die("Usage: %s <infile> <outfile> <codec_type(vp8/vp9)> <width> <height> "
-        "<rate_num> <rate_den> <speed> <frame_drop_threshold> <threads> <mode> "
+        "<rate_num> <rate_den> <speed> <frame_drop_threshold> "
+        "<error_resilient> <threads> <mode> "
        "<Rate_0> ... <Rate_nlayers-1> <bit-depth> \n",
        argv[0]);
 #else
    die("Usage: %s <infile> <outfile> <codec_type(vp8/vp9)> <width> <height> "
-        "<rate_num> <rate_den> <speed> <frame_drop_threshold> <threads> <mode> "
+        "<rate_num> <rate_den> <speed> <frame_drop_threshold> "
+        "<error_resilient> <threads> <mode> "
        "<Rate_0> ... <Rate_nlayers-1> \n",
        argv[0]);
 #endif  // CONFIG_VP9_HIGHBITDEPTH
@@ -553,9 +612,9 @@ int main(int argc, char **argv) {
    die("Invalid resolution: %d x %d", width, height);
  }

-  layering_mode = (int)strtol(argv[11], NULL, 0);
+  layering_mode = (int)strtol(argv[12], NULL, 0);
  if (layering_mode < 0 || layering_mode > 13) {
-    die("Invalid layering mode (0..12) %s", argv[11]);
+    die("Invalid layering mode (0..12) %s", argv[12]);
  }

  if (argc != min_args + mode_to_num_layers[layering_mode]) {
@@ -619,11 +678,11 @@ int main(int argc, char **argv) {

  for (i = min_args_base;
       (int)i < min_args_base + mode_to_num_layers[layering_mode]; ++i) {
-    rc.layer_target_bitrate[i - 12] = (int)strtol(argv[i], NULL, 0);
+    rc.layer_target_bitrate[i - 13] = (int)strtol(argv[i], NULL, 0);
    if (strncmp(encoder->name, "vp8", 3) == 0)
-      cfg.ts_target_bitrate[i - 12] = rc.layer_target_bitrate[i - 12];
+      cfg.ts_target_bitrate[i - 13] = rc.layer_target_bitrate[i - 13];
    else if (strncmp(encoder->name, "vp9", 3) == 0)
-      cfg.layer_target_bitrate[i - 12] = rc.layer_target_bitrate[i - 12];
+      cfg.layer_target_bitrate[i - 13] = rc.layer_target_bitrate[i - 13];
  }

  // Real time parameters.
@@ -634,7 +693,7 @@ int main(int argc, char **argv) {
  if (strncmp(encoder->name, "vp9", 3) == 0) cfg.rc_max_quantizer = 52;
  cfg.rc_undershoot_pct = 50;
  cfg.rc_overshoot_pct = 50;
-  cfg.rc_buf_initial_sz = 500;
+  cfg.rc_buf_initial_sz = 600;
  cfg.rc_buf_optimal_sz = 600;
  cfg.rc_buf_sz = 1000;

@@ -642,10 +701,14 @@ int main(int argc, char **argv) {
  cfg.rc_resize_allowed = 0;

  // Use 1 thread as default.
-  cfg.g_threads = (unsigned int)strtoul(argv[10], NULL, 0);
+  cfg.g_threads = (unsigned int)strtoul(argv[11], NULL, 0);

+  error_resilient = (uint32_t)strtoul(argv[10], NULL, 0);
+  if (error_resilient != 0 && error_resilient != 1) {
+    die("Invalid value for error resilient (0, 1): %d.", error_resilient);
+  }
  // Enable error resilient mode.
-  cfg.g_error_resilient = 1;
+  cfg.g_error_resilient = error_resilient;
  cfg.g_lag_in_frames = 0;
  cfg.kf_mode = VPX_KF_AUTO;

@@ -700,9 +763,15 @@ int main(int argc, char **argv) {

  if (strncmp(encoder->name, "vp8", 3) == 0) {
    vpx_codec_control(&codec, VP8E_SET_CPUUSED, -speed);
-    vpx_codec_control(&codec, VP8E_SET_NOISE_SENSITIVITY, kDenoiserOff);
+    vpx_codec_control(&codec, VP8E_SET_NOISE_SENSITIVITY, kVp8DenoiserOff);
    vpx_codec_control(&codec, VP8E_SET_STATIC_THRESHOLD, 1);
    vpx_codec_control(&codec, VP8E_SET_GF_CBR_BOOST_PCT, 0);
+#if VP8_ROI_MAP
+    vp8_set_roi_map(&cfg, &roi);
+    if (vpx_codec_control(&codec, VP8E_SET_ROI_MAP, &roi))
+      die_codec(&codec, "Failed to set ROI map");
+#endif
+
  } else if (strncmp(encoder->name, "vp9", 3) == 0) {
    vpx_svc_extra_cfg_t svc_params;
    memset(&svc_params, 0, sizeof(svc_params));
@@ -711,10 +780,16 @@ int main(int argc, char **argv) {
    vpx_codec_control(&codec, VP9E_SET_GF_CBR_BOOST_PCT, 0);
    vpx_codec_control(&codec, VP9E_SET_FRAME_PARALLEL_DECODING, 0);
    vpx_codec_control(&codec, VP9E_SET_FRAME_PERIODIC_BOOST, 0);
-    vpx_codec_control(&codec, VP9E_SET_NOISE_SENSITIVITY, kDenoiserOff);
+    vpx_codec_control(&codec, VP9E_SET_NOISE_SENSITIVITY, kVp9DenoiserOff);
    vpx_codec_control(&codec, VP8E_SET_STATIC_THRESHOLD, 1);
    vpx_codec_control(&codec, VP9E_SET_TUNE_CONTENT, 0);
    vpx_codec_control(&codec, VP9E_SET_TILE_COLUMNS, (cfg.g_threads >> 1));
+    // TODO(marpan/jianj): There is an issue with row-mt for low resolutons at
+    // high speed settings, disable its use for those cases for now.
+    if (cfg.g_threads > 1 && ((cfg.g_w > 320 && cfg.g_h > 240) || speed < 7))
+      vpx_codec_control(&codec, VP9E_SET_ROW_MT, 1);
+    else
+      vpx_codec_control(&codec, VP9E_SET_ROW_MT, 0);
    if (vpx_codec_control(&codec, VP9E_SET_SVC, layering_mode > 0 ? 1 : 0))
      die_codec(&codec, "Failed to set SVC");
    for (i = 0; i < cfg.ts_number_layers; ++i) {
@@ -733,7 +808,7 @@ int main(int argc, char **argv) {
  // For generating smaller key frames, use a smaller max_intra_size_pct
  // value, like 100 or 200.
  {
-    const int max_intra_size_pct = 900;
+    const int max_intra_size_pct = 1000;
    vpx_codec_control(&codec, VP8E_SET_MAX_INTRA_BITRATE_PCT,
                      max_intra_size_pct);
  }
@@ -743,10 +818,8 @@ int main(int argc, char **argv) {
    struct vpx_usec_timer timer;
    vpx_codec_iter_t iter = NULL;
    const vpx_codec_cx_pkt_t *pkt;
-#if VPX_ENCODER_ABI_VERSION > (4 + VPX_CODEC_ABI_VERSION)
    // Update the temporal layer_id. No spatial layers in this test.
    layer_id.spatial_layer_id = 0;
-#endif
    layer_id.temporal_layer_id =
        cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity];
    if (strncmp(encoder->name, "vp9", 3) == 0) {
--- a/libs.mk
+++ b/libs.mk
@@ -149,6 +149,7 @@ CODEC_SRCS-yes += $(BUILD_PFX)vpx_config.c
 INSTALL-SRCS-no += $(BUILD_PFX)vpx_config.c
 ifeq ($(ARCH_X86)$(ARCH_X86_64),yes)
 INSTALL-SRCS-$(CONFIG_CODEC_SRCS) += third_party/x86inc/x86inc.asm
+INSTALL-SRCS-$(CONFIG_CODEC_SRCS) += vpx_dsp/x86/bitdepth_conversion_sse2.asm
 endif
 CODEC_EXPORTS-yes += vpx/exports_com
 CODEC_EXPORTS-$(CONFIG_ENCODERS) += vpx/exports_enc
@@ -187,6 +188,13 @@ libvpx_srcs.txt:
 	@echo $(CODEC_SRCS) | xargs -n1 echo | LC_ALL=C sort -u > $@
 CLEAN-OBJS += libvpx_srcs.txt

+# Assembly files that are included, but don't define symbols themselves.
+# Filtered out to avoid Windows build warnings.
+ASM_INCLUDES := \
+    third_party/x86inc/x86inc.asm \
+    vpx_config.asm \
+    vpx_ports/x86_abi_support.asm \
+    vpx_dsp/x86/bitdepth_conversion_sse2.asm \

 ifeq ($(CONFIG_EXTERNAL_BUILD),yes)
 ifeq ($(CONFIG_MSVS),yes)
@@ -198,13 +206,6 @@ vpx.def: $(call enabled,CODEC_EXPORTS)
            --out=$@ $^
 CLEAN-OBJS += vpx.def

-# Assembly files that are included, but don't define symbols themselves.
-# Filtered out to avoid Visual Studio build warnings.
-ASM_INCLUDES := \
-    third_party/x86inc/x86inc.asm \
-    vpx_config.asm \
-    vpx_ports/x86_abi_support.asm \
-
 vpx.$(VCPROJ_SFX): $(CODEC_SRCS) vpx.def
 	@echo "    [CREATE] $@"
 	$(qexec)$(GEN_VCPROJ) \
@@ -227,13 +228,13 @@ vpx.$(VCPROJ_SFX): $(RTCD)

 endif
 else
-LIBVPX_OBJS=$(call objs,$(CODEC_SRCS))
+LIBVPX_OBJS=$(call objs, $(filter-out $(ASM_INCLUDES), $(CODEC_SRCS)))
 OBJS-yes += $(LIBVPX_OBJS)
 LIBS-$(if yes,$(CONFIG_STATIC)) += $(BUILD_PFX)libvpx.a $(BUILD_PFX)libvpx_g.a
 $(BUILD_PFX)libvpx_g.a: $(LIBVPX_OBJS)

-SO_VERSION_MAJOR := 4
-SO_VERSION_MINOR := 1
+SO_VERSION_MAJOR := 5
+SO_VERSION_MINOR := 0
 SO_VERSION_PATCH := 0
 ifeq ($(filter darwin%,$(TGT_OS)),$(TGT_OS))
 LIBVPX_SO               := libvpx.$(SO_VERSION_MAJOR).dylib
@@ -404,8 +405,16 @@ CLEAN-OBJS += libvpx_test_srcs.txt

 $(LIBVPX_TEST_DATA): $(SRC_PATH_BARE)/test/test-data.sha1
 	@echo "    [DOWNLOAD] $@"
-	$(qexec)trap 'rm -f $@' INT TERM &&\
-            curl --retry 1 -L -o $@ $(call libvpx_test_data_url,$(@F))
+	# Attempt to download the file using curl, retrying once if it fails for a
+	# partial file (18).
+	$(qexec)( \
+	  trap 'rm -f $@' INT TERM; \
+	  curl="curl --retry 1 -L -o $@ $(call libvpx_test_data_url,$(@F))"; \
+	  $$curl; \
+	  case "$$?" in \
+	    18) $$curl -C -;; \
+	  esac \
+	)

 testdata:: $(LIBVPX_TEST_DATA)
 	$(qexec)[ -x "$$(which sha1sum)" ] && sha1sum=sha1sum;\
--- a/rate_hist.c
+++ b/rate_hist.c
@@ -37,7 +37,13 @@ struct rate_hist {
 struct rate_hist *init_rate_histogram(const vpx_codec_enc_cfg_t *cfg,
                                      const vpx_rational_t *fps) {
  int i;
-  struct rate_hist *hist = malloc(sizeof(*hist));
+  struct rate_hist *hist = calloc(1, sizeof(*hist));
+
+  if (hist == NULL || cfg == NULL || fps == NULL || fps->num == 0 ||
+      fps->den == 0) {
+    destroy_rate_histogram(hist);
+    return NULL;
+  }

  // Determine the number of samples in the buffer. Use the file's framerate
  // to determine the number of frames in rc_buf_sz milliseconds, with an
@@ -80,7 +86,11 @@ void update_rate_histogram(struct rate_hist *hist,
                      (uint64_t)cfg->g_timebase.num /
                      (uint64_t)cfg->g_timebase.den;

-  int idx = hist->frames++ % hist->samples;
+  int idx;
+
+  if (hist == NULL || cfg == NULL || pkt == NULL) return;
+
+  idx = hist->frames++ % hist->samples;
  hist->pts[idx] = now;
  hist->sz[idx] = (int)pkt->data.frame.sz;

@@ -116,9 +126,14 @@ void update_rate_histogram(struct rate_hist *hist,
 static int merge_hist_buckets(struct hist_bucket *bucket, int max_buckets,
                              int *num_buckets) {
  int small_bucket = 0, merge_bucket = INT_MAX, big_bucket = 0;
-  int buckets = *num_buckets;
+  int buckets;
  int i;

+  assert(bucket != NULL);
+  assert(num_buckets != NULL);
+
+  buckets = *num_buckets;
+
  /* Find the extrema for this list of buckets */
  big_bucket = small_bucket = 0;
  for (i = 0; i < buckets; i++) {
@@ -181,6 +196,8 @@ static void show_histogram(const struct hist_bucket *bucket, int buckets,
  const char *pat1, *pat2;
  int i;

+  assert(bucket != NULL);
+
  switch ((int)(log(bucket[buckets - 1].high) / log(10)) + 1) {
    case 1:
    case 2:
@@ -259,6 +276,8 @@ void show_rate_histogram(struct rate_hist *hist, const vpx_codec_enc_cfg_t *cfg,
  int i, scale;
  int buckets = 0;

+  if (hist == NULL || cfg == NULL) return;
+
  for (i = 0; i < RATE_BINS; i++) {
    if (hist->bucket[i].low == INT_MAX) continue;
    hist->bucket[buckets++] = hist->bucket[i];
--- a/test/acm_random.h
+++ b/test/acm_random.h
@@ -11,6 +11,10 @@
 #ifndef TEST_ACM_RANDOM_H_
 #define TEST_ACM_RANDOM_H_

+#include <assert.h>
+
+#include <limits>
+
 #include "third_party/googletest/src/include/gtest/gtest.h"

 #include "vpx/vpx_integer.h"
@@ -50,6 +54,13 @@ class ACMRandom {
    return r < 128 ? r << 4 : r >> 4;
  }

+  uint32_t RandRange(const uint32_t range) {
+    // testing::internal::Random::Generate provides values in the range
+    // testing::internal::Random::kMaxRange.
+    assert(range <= testing::internal::Random::kMaxRange);
+    return random_.Generate(range);
+  }
+
  int PseudoUniform(int range) { return random_.Generate(range); }

  int operator()(int n) { return PseudoUniform(n); }
--- a/test/android/Android.mk
+++ b/test/android/Android.mk
@@ -32,6 +32,7 @@ LOCAL_CPP_EXTENSION := .cc
 LOCAL_MODULE := gtest
 LOCAL_C_INCLUDES := $(LOCAL_PATH)/third_party/googletest/src/
 LOCAL_C_INCLUDES += $(LOCAL_PATH)/third_party/googletest/src/include/
+LOCAL_EXPORT_C_INCLUDES := $(LOCAL_PATH)/third_party/googletest/src/include/
 LOCAL_SRC_FILES := ./third_party/googletest/src/src/gtest-all.cc
 include $(BUILD_STATIC_LIBRARY)

--- a/test/avg_test.cc
+++ b/test/avg_test.cc
@@ -14,6 +14,7 @@

 #include "third_party/googletest/src/include/gtest/gtest.h"

+#include "./vp9_rtcd.h"
 #include "./vpx_config.h"
 #include "./vpx_dsp_rtcd.h"

@@ -22,6 +23,7 @@
 #include "test/register_state_check.h"
 #include "test/util.h"
 #include "vpx_mem/vpx_mem.h"
+#include "vpx_ports/vpx_timer.h"

 using libvpx_test::ACMRandom;

@@ -186,7 +188,7 @@ class IntProColTest : public AverageTestBase,
  int16_t sum_c_;
 };

-typedef int (*SatdFunc)(const int16_t *coeffs, int length);
+typedef int (*SatdFunc)(const tran_low_t *coeffs, int length);
 typedef std::tr1::tuple<int, SatdFunc> SatdTestParam;

 class SatdTest : public ::testing::Test,
@@ -196,7 +198,7 @@ class SatdTest : public ::testing::Test,
    satd_size_ = GET_PARAM(0);
    satd_func_ = GET_PARAM(1);
    rnd_.Reset(ACMRandom::DeterministicSeed());
-    src_ = reinterpret_cast<int16_t *>(
+    src_ = reinterpret_cast<tran_low_t *>(
        vpx_memalign(16, sizeof(*src_) * satd_size_));
    ASSERT_TRUE(src_ != NULL);
  }
@@ -206,12 +208,15 @@ class SatdTest : public ::testing::Test,
    vpx_free(src_);
  }

-  void FillConstant(const int16_t val) {
+  void FillConstant(const tran_low_t val) {
    for (int i = 0; i < satd_size_; ++i) src_[i] = val;
  }

  void FillRandom() {
-    for (int i = 0; i < satd_size_; ++i) src_[i] = rnd_.Rand16();
+    for (int i = 0; i < satd_size_; ++i) {
+      const int16_t tmp = rnd_.Rand16();
+      src_[i] = (tran_low_t)tmp;
+    }
  }

  void Check(const int expected) {
@@ -223,11 +228,66 @@ class SatdTest : public ::testing::Test,
  int satd_size_;

 private:
-  int16_t *src_;
+  tran_low_t *src_;
  SatdFunc satd_func_;
  ACMRandom rnd_;
 };

+typedef int64_t (*BlockErrorFunc)(const tran_low_t *coeff,
+                                  const tran_low_t *dqcoeff, int block_size);
+typedef std::tr1::tuple<int, BlockErrorFunc> BlockErrorTestFPParam;
+
+class BlockErrorTestFP
+    : public ::testing::Test,
+      public ::testing::WithParamInterface<BlockErrorTestFPParam> {
+ protected:
+  virtual void SetUp() {
+    txfm_size_ = GET_PARAM(0);
+    block_error_func_ = GET_PARAM(1);
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+    coeff_ = reinterpret_cast<tran_low_t *>(
+        vpx_memalign(16, sizeof(*coeff_) * txfm_size_));
+    dqcoeff_ = reinterpret_cast<tran_low_t *>(
+        vpx_memalign(16, sizeof(*dqcoeff_) * txfm_size_));
+    ASSERT_TRUE(coeff_ != NULL);
+    ASSERT_TRUE(dqcoeff_ != NULL);
+  }
+
+  virtual void TearDown() {
+    libvpx_test::ClearSystemState();
+    vpx_free(coeff_);
+    vpx_free(dqcoeff_);
+  }
+
+  void FillConstant(const tran_low_t coeff_val, const tran_low_t dqcoeff_val) {
+    for (int i = 0; i < txfm_size_; ++i) coeff_[i] = coeff_val;
+    for (int i = 0; i < txfm_size_; ++i) dqcoeff_[i] = dqcoeff_val;
+  }
+
+  void FillRandom() {
+    // Just two fixed seeds
+    rnd_.Reset(0xb0b9);
+    for (int i = 0; i < txfm_size_; ++i) coeff_[i] = rnd_.Rand16() >> 1;
+    rnd_.Reset(0xb0c8);
+    for (int i = 0; i < txfm_size_; ++i) dqcoeff_[i] = rnd_.Rand16() >> 1;
+  }
+
+  void Check(const int64_t expected) {
+    int64_t total;
+    ASM_REGISTER_STATE_CHECK(
+        total = block_error_func_(coeff_, dqcoeff_, txfm_size_));
+    EXPECT_EQ(expected, total);
+  }
+
+  int txfm_size_;
+
+ private:
+  tran_low_t *coeff_;
+  tran_low_t *dqcoeff_;
+  BlockErrorFunc block_error_func_;
+  ACMRandom rnd_;
+};
+
 uint8_t *AverageTestBase::source_data_ = NULL;

 TEST_P(AverageTest, MinValue) {
@@ -308,6 +368,66 @@ TEST_P(SatdTest, Random) {
  Check(expected);
 }

+TEST_P(SatdTest, DISABLED_Speed) {
+  const int kCountSpeedTestBlock = 20000;
+  vpx_usec_timer timer;
+  DECLARE_ALIGNED(16, tran_low_t, coeff[1024]);
+  const int blocksize = GET_PARAM(0);
+
+  vpx_usec_timer_start(&timer);
+  for (int i = 0; i < kCountSpeedTestBlock; ++i) {
+    GET_PARAM(1)(coeff, blocksize);
+  }
+  vpx_usec_timer_mark(&timer);
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("blocksize: %4d time: %4d us\n", blocksize, elapsed_time);
+}
+
+TEST_P(BlockErrorTestFP, MinValue) {
+  const int64_t kMin = -32640;
+  const int64_t expected = kMin * kMin * txfm_size_;
+  FillConstant(kMin, 0);
+  Check(expected);
+}
+
+TEST_P(BlockErrorTestFP, MaxValue) {
+  const int64_t kMax = 32640;
+  const int64_t expected = kMax * kMax * txfm_size_;
+  FillConstant(kMax, 0);
+  Check(expected);
+}
+
+TEST_P(BlockErrorTestFP, Random) {
+  int64_t expected;
+  switch (txfm_size_) {
+    case 16: expected = 2051681432; break;
+    case 64: expected = 11075114379; break;
+    case 256: expected = 44386271116; break;
+    case 1024: expected = 184774996089; break;
+    default:
+      FAIL() << "Invalid satd size (" << txfm_size_
+             << ") valid: 16/64/256/1024";
+  }
+  FillRandom();
+  Check(expected);
+}
+
+TEST_P(BlockErrorTestFP, DISABLED_Speed) {
+  const int kCountSpeedTestBlock = 20000;
+  vpx_usec_timer timer;
+  DECLARE_ALIGNED(16, tran_low_t, coeff[1024]);
+  DECLARE_ALIGNED(16, tran_low_t, dqcoeff[1024]);
+  const int blocksize = GET_PARAM(0);
+
+  vpx_usec_timer_start(&timer);
+  for (int i = 0; i < kCountSpeedTestBlock; ++i) {
+    GET_PARAM(1)(coeff, dqcoeff, blocksize);
+  }
+  vpx_usec_timer_mark(&timer);
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("blocksize: %4d time: %4d us\n", blocksize, elapsed_time);
+}
+
 using std::tr1::make_tuple;

 INSTANTIATE_TEST_CASE_P(
@@ -321,6 +441,13 @@ INSTANTIATE_TEST_CASE_P(C, SatdTest,
                                          make_tuple(256, &vpx_satd_c),
                                          make_tuple(1024, &vpx_satd_c)));

+INSTANTIATE_TEST_CASE_P(
+    C, BlockErrorTestFP,
+    ::testing::Values(make_tuple(16, &vp9_block_error_fp_c),
+                      make_tuple(64, &vp9_block_error_fp_c),
+                      make_tuple(256, &vp9_block_error_fp_c),
+                      make_tuple(1024, &vp9_block_error_fp_c)));
+
 #if HAVE_SSE2
 INSTANTIATE_TEST_CASE_P(
    SSE2, AverageTest,
@@ -350,6 +477,28 @@ INSTANTIATE_TEST_CASE_P(SSE2, SatdTest,
                                          make_tuple(64, &vpx_satd_sse2),
                                          make_tuple(256, &vpx_satd_sse2),
                                          make_tuple(1024, &vpx_satd_sse2)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSE2, BlockErrorTestFP,
+    ::testing::Values(make_tuple(16, &vp9_block_error_fp_sse2),
+                      make_tuple(64, &vp9_block_error_fp_sse2),
+                      make_tuple(256, &vp9_block_error_fp_sse2),
+                      make_tuple(1024, &vp9_block_error_fp_sse2)));
+#endif  // HAVE_SSE2
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, SatdTest,
+                        ::testing::Values(make_tuple(16, &vpx_satd_avx2),
+                                          make_tuple(64, &vpx_satd_avx2),
+                                          make_tuple(256, &vpx_satd_avx2),
+                                          make_tuple(1024, &vpx_satd_avx2)));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2, BlockErrorTestFP,
+    ::testing::Values(make_tuple(16, &vp9_block_error_fp_avx2),
+                      make_tuple(64, &vp9_block_error_fp_avx2),
+                      make_tuple(256, &vp9_block_error_fp_avx2),
+                      make_tuple(1024, &vp9_block_error_fp_avx2)));
 #endif

 #if HAVE_NEON
@@ -381,7 +530,18 @@ INSTANTIATE_TEST_CASE_P(NEON, SatdTest,
                                          make_tuple(64, &vpx_satd_neon),
                                          make_tuple(256, &vpx_satd_neon),
                                          make_tuple(1024, &vpx_satd_neon)));
-#endif
+
+// TODO(jianj): Remove the highbitdepth flag once the SIMD functions are
+// in place.
+#if !CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    NEON, BlockErrorTestFP,
+    ::testing::Values(make_tuple(16, &vp9_block_error_fp_neon),
+                      make_tuple(64, &vp9_block_error_fp_neon),
+                      make_tuple(256, &vp9_block_error_fp_neon),
+                      make_tuple(1024, &vp9_block_error_fp_neon)));
+#endif  // !CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_NEON

 #if HAVE_MSA
 INSTANTIATE_TEST_CASE_P(
@@ -392,6 +552,30 @@ INSTANTIATE_TEST_CASE_P(
                      make_tuple(16, 16, 0, 4, &vpx_avg_4x4_msa),
                      make_tuple(16, 16, 5, 4, &vpx_avg_4x4_msa),
                      make_tuple(32, 32, 15, 4, &vpx_avg_4x4_msa)));
-#endif
+
+INSTANTIATE_TEST_CASE_P(
+    MSA, IntProRowTest,
+    ::testing::Values(make_tuple(16, &vpx_int_pro_row_msa, &vpx_int_pro_row_c),
+                      make_tuple(32, &vpx_int_pro_row_msa, &vpx_int_pro_row_c),
+                      make_tuple(64, &vpx_int_pro_row_msa,
+                                 &vpx_int_pro_row_c)));
+
+INSTANTIATE_TEST_CASE_P(
+    MSA, IntProColTest,
+    ::testing::Values(make_tuple(16, &vpx_int_pro_col_msa, &vpx_int_pro_col_c),
+                      make_tuple(32, &vpx_int_pro_col_msa, &vpx_int_pro_col_c),
+                      make_tuple(64, &vpx_int_pro_col_msa,
+                                 &vpx_int_pro_col_c)));
+
+// TODO(jingning): Remove the highbitdepth flag once the SIMD functions are
+// in place.
+#if !CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(MSA, SatdTest,
+                        ::testing::Values(make_tuple(16, &vpx_satd_msa),
+                                          make_tuple(64, &vpx_satd_msa),
+                                          make_tuple(256, &vpx_satd_msa),
+                                          make_tuple(1024, &vpx_satd_msa)));
+#endif  // !CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_MSA

 }  // namespace
--- a/test/buffer.h
+++ b/test/buffer.h
@@ -0,0 +1,382 @@
+/*
+ *  Copyright (c) 2016 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef TEST_BUFFER_H_
+#define TEST_BUFFER_H_
+
+#include <stdio.h>
+
+#include <limits>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "test/acm_random.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_mem/vpx_mem.h"
+
+namespace libvpx_test {
+
+template <typename T>
+class Buffer {
+ public:
+  Buffer(int width, int height, int top_padding, int left_padding,
+         int right_padding, int bottom_padding)
+      : width_(width), height_(height), top_padding_(top_padding),
+        left_padding_(left_padding), right_padding_(right_padding),
+        bottom_padding_(bottom_padding), alignment_(0), padding_value_(0),
+        stride_(0), raw_size_(0), num_elements_(0), raw_buffer_(NULL) {}
+
+  Buffer(int width, int height, int top_padding, int left_padding,
+         int right_padding, int bottom_padding, unsigned int alignment)
+      : width_(width), height_(height), top_padding_(top_padding),
+        left_padding_(left_padding), right_padding_(right_padding),
+        bottom_padding_(bottom_padding), alignment_(alignment),
+        padding_value_(0), stride_(0), raw_size_(0), num_elements_(0),
+        raw_buffer_(NULL) {}
+
+  Buffer(int width, int height, int padding)
+      : width_(width), height_(height), top_padding_(padding),
+        left_padding_(padding), right_padding_(padding),
+        bottom_padding_(padding), alignment_(0), padding_value_(0), stride_(0),
+        raw_size_(0), num_elements_(0), raw_buffer_(NULL) {}
+
+  Buffer(int width, int height, int padding, unsigned int alignment)
+      : width_(width), height_(height), top_padding_(padding),
+        left_padding_(padding), right_padding_(padding),
+        bottom_padding_(padding), alignment_(alignment), padding_value_(0),
+        stride_(0), raw_size_(0), num_elements_(0), raw_buffer_(NULL) {}
+
+  ~Buffer() {
+    if (alignment_) {
+      vpx_free(raw_buffer_);
+    } else {
+      delete[] raw_buffer_;
+    }
+  }
+
+  T *TopLeftPixel() const;
+
+  int stride() const { return stride_; }
+
+  // Set the buffer (excluding padding) to 'value'.
+  void Set(const T value);
+
+  // Set the buffer (excluding padding) to the output of ACMRandom function
+  // 'rand_func'.
+  void Set(ACMRandom *rand_class, T (ACMRandom::*rand_func)());
+
+  // Set the buffer (excluding padding) to the output of ACMRandom function
+  // 'RandRange' with range 'low' to 'high' which typically must be within
+  // testing::internal::Random::kMaxRange (1u << 31). However, because we want
+  // to allow negative low (and high) values, it is restricted to INT32_MAX
+  // here.
+  void Set(ACMRandom *rand_class, const T low, const T high);
+
+  // Copy the contents of Buffer 'a' (excluding padding).
+  void CopyFrom(const Buffer<T> &a);
+
+  void DumpBuffer() const;
+
+  // Highlight the differences between two buffers if they are the same size.
+  void PrintDifference(const Buffer<T> &a) const;
+
+  bool HasPadding() const;
+
+  // Sets all the values in the buffer to 'padding_value'.
+  void SetPadding(const T padding_value);
+
+  // Checks if all the values (excluding padding) are equal to 'value' if the
+  // Buffers are the same size.
+  bool CheckValues(const T value) const;
+
+  // Check that padding matches the expected value or there is no padding.
+  bool CheckPadding() const;
+
+  // Compare the non-padding portion of two buffers if they are the same size.
+  bool CheckValues(const Buffer<T> &a) const;
+
+  bool Init() {
+    if (raw_buffer_ != NULL) return false;
+    EXPECT_GT(width_, 0);
+    EXPECT_GT(height_, 0);
+    EXPECT_GE(top_padding_, 0);
+    EXPECT_GE(left_padding_, 0);
+    EXPECT_GE(right_padding_, 0);
+    EXPECT_GE(bottom_padding_, 0);
+    stride_ = left_padding_ + width_ + right_padding_;
+    num_elements_ = stride_ * (top_padding_ + height_ + bottom_padding_);
+    raw_size_ = num_elements_ * sizeof(T);
+    if (alignment_) {
+      EXPECT_GE(alignment_, sizeof(T));
+      // Ensure alignment of the first value will be preserved.
+      EXPECT_EQ((left_padding_ * sizeof(T)) % alignment_, 0u);
+      // Ensure alignment of the subsequent rows will be preserved when there is
+      // a stride.
+      if (stride_ != width_) {
+        EXPECT_EQ((stride_ * sizeof(T)) % alignment_, 0u);
+      }
+      raw_buffer_ = reinterpret_cast<T *>(vpx_memalign(alignment_, raw_size_));
+    } else {
+      raw_buffer_ = new (std::nothrow) T[num_elements_];
+    }
+    EXPECT_TRUE(raw_buffer_ != NULL);
+    SetPadding(std::numeric_limits<T>::max());
+    return !::testing::Test::HasFailure();
+  }
+
+ private:
+  bool BufferSizesMatch(const Buffer<T> &a) const;
+
+  const int width_;
+  const int height_;
+  const int top_padding_;
+  const int left_padding_;
+  const int right_padding_;
+  const int bottom_padding_;
+  const unsigned int alignment_;
+  T padding_value_;
+  int stride_;
+  int raw_size_;
+  int num_elements_;
+  T *raw_buffer_;
+};
+
+template <typename T>
+T *Buffer<T>::TopLeftPixel() const {
+  if (!raw_buffer_) return NULL;
+  return raw_buffer_ + (top_padding_ * stride_) + left_padding_;
+}
+
+template <typename T>
+void Buffer<T>::Set(const T value) {
+  if (!raw_buffer_) return;
+  T *src = TopLeftPixel();
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < width_; ++width) {
+      src[width] = value;
+    }
+    src += stride_;
+  }
+}
+
+template <typename T>
+void Buffer<T>::Set(ACMRandom *rand_class, T (ACMRandom::*rand_func)()) {
+  if (!raw_buffer_) return;
+  T *src = TopLeftPixel();
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < width_; ++width) {
+      src[width] = (*rand_class.*rand_func)();
+    }
+    src += stride_;
+  }
+}
+
+template <typename T>
+void Buffer<T>::Set(ACMRandom *rand_class, const T low, const T high) {
+  if (!raw_buffer_) return;
+
+  EXPECT_LE(low, high);
+  EXPECT_LE(static_cast<int64_t>(high) - low,
+            std::numeric_limits<int32_t>::max());
+
+  T *src = TopLeftPixel();
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < width_; ++width) {
+      // 'low' will be promoted to unsigned given the return type of RandRange.
+      // Store the value as an int to avoid unsigned overflow warnings when
+      // 'low' is negative.
+      const int32_t value =
+          static_cast<int32_t>((*rand_class).RandRange(high - low));
+      src[width] = static_cast<T>(value + low);
+    }
+    src += stride_;
+  }
+}
+
+template <typename T>
+void Buffer<T>::CopyFrom(const Buffer<T> &a) {
+  if (!raw_buffer_) return;
+  if (!BufferSizesMatch(a)) return;
+
+  T *a_src = a.TopLeftPixel();
+  T *b_src = this->TopLeftPixel();
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < width_; ++width) {
+      b_src[width] = a_src[width];
+    }
+    a_src += a.stride();
+    b_src += this->stride();
+  }
+}
+
+template <typename T>
+void Buffer<T>::DumpBuffer() const {
+  if (!raw_buffer_) return;
+  for (int height = 0; height < height_ + top_padding_ + bottom_padding_;
+       ++height) {
+    for (int width = 0; width < stride_; ++width) {
+      printf("%4d", raw_buffer_[height + width * stride_]);
+    }
+    printf("\n");
+  }
+}
+
+template <typename T>
+bool Buffer<T>::HasPadding() const {
+  if (!raw_buffer_) return false;
+  return top_padding_ || left_padding_ || right_padding_ || bottom_padding_;
+}
+
+template <typename T>
+void Buffer<T>::PrintDifference(const Buffer<T> &a) const {
+  if (!raw_buffer_) return;
+  if (!BufferSizesMatch(a)) return;
+
+  T *a_src = a.TopLeftPixel();
+  T *b_src = TopLeftPixel();
+
+  printf("This buffer:\n");
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < width_; ++width) {
+      if (a_src[width] != b_src[width]) {
+        printf("*%3d", b_src[width]);
+      } else {
+        printf("%4d", b_src[width]);
+      }
+    }
+    printf("\n");
+    a_src += a.stride();
+    b_src += this->stride();
+  }
+
+  a_src = a.TopLeftPixel();
+  b_src = TopLeftPixel();
+
+  printf("Reference buffer:\n");
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < width_; ++width) {
+      if (a_src[width] != b_src[width]) {
+        printf("*%3d", a_src[width]);
+      } else {
+        printf("%4d", a_src[width]);
+      }
+    }
+    printf("\n");
+    a_src += a.stride();
+    b_src += this->stride();
+  }
+}
+
+template <typename T>
+void Buffer<T>::SetPadding(const T padding_value) {
+  if (!raw_buffer_) return;
+  padding_value_ = padding_value;
+
+  T *src = raw_buffer_;
+  for (int i = 0; i < num_elements_; ++i) {
+    src[i] = padding_value;
+  }
+}
+
+template <typename T>
+bool Buffer<T>::CheckValues(const T value) const {
+  if (!raw_buffer_) return false;
+  T *src = TopLeftPixel();
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < width_; ++width) {
+      if (value != src[width]) {
+        return false;
+      }
+    }
+    src += stride_;
+  }
+  return true;
+}
+
+template <typename T>
+bool Buffer<T>::CheckPadding() const {
+  if (!raw_buffer_) return false;
+  if (!HasPadding()) return true;
+
+  // Top padding.
+  T const *top = raw_buffer_;
+  for (int i = 0; i < stride_ * top_padding_; ++i) {
+    if (padding_value_ != top[i]) {
+      return false;
+    }
+  }
+
+  // Left padding.
+  T const *left = TopLeftPixel() - left_padding_;
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < left_padding_; ++width) {
+      if (padding_value_ != left[width]) {
+        return false;
+      }
+    }
+    left += stride_;
+  }
+
+  // Right padding.
+  T const *right = TopLeftPixel() + width_;
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < right_padding_; ++width) {
+      if (padding_value_ != right[width]) {
+        return false;
+      }
+    }
+    right += stride_;
+  }
+
+  // Bottom padding
+  T const *bottom = raw_buffer_ + (top_padding_ + height_) * stride_;
+  for (int i = 0; i < stride_ * bottom_padding_; ++i) {
+    if (padding_value_ != bottom[i]) {
+      return false;
+    }
+  }
+
+  return true;
+}
+
+template <typename T>
+bool Buffer<T>::CheckValues(const Buffer<T> &a) const {
+  if (!raw_buffer_) return false;
+  if (!BufferSizesMatch(a)) return false;
+
+  T *a_src = a.TopLeftPixel();
+  T *b_src = this->TopLeftPixel();
+  for (int height = 0; height < height_; ++height) {
+    for (int width = 0; width < width_; ++width) {
+      if (a_src[width] != b_src[width]) {
+        return false;
+      }
+    }
+    a_src += a.stride();
+    b_src += this->stride();
+  }
+  return true;
+}
+
+template <typename T>
+bool Buffer<T>::BufferSizesMatch(const Buffer<T> &a) const {
+  if (!raw_buffer_) return false;
+  if (a.width_ != this->width_ || a.height_ != this->height_) {
+    printf(
+        "Reference buffer of size %dx%d does not match this buffer which is "
+        "size %dx%d\n",
+        a.width_, a.height_, this->width_, this->height_);
+    return false;
+  }
+
+  return true;
+}
+}  // namespace libvpx_test
+#endif  // TEST_BUFFER_H_
--- a/test/byte_alignment_test.cc
+++ b/test/byte_alignment_test.cc
@@ -128,8 +128,8 @@ class ByteAlignmentTest
  // TODO(fgalligan): Move the MD5 testing code into another class.
  void OpenMd5File(const std::string &md5_file_name_) {
    md5_file_ = libvpx_test::OpenTestDataFile(md5_file_name_);
-    ASSERT_TRUE(md5_file_ != NULL) << "MD5 file open failed. Filename: "
-                                   << md5_file_name_;
+    ASSERT_TRUE(md5_file_ != NULL)
+        << "MD5 file open failed. Filename: " << md5_file_name_;
  }

  void CheckMd5(const vpx_image_t &img) {
--- a/test/comp_avg_pred_test.cc
+++ b/test/comp_avg_pred_test.cc
@@ -0,0 +1,182 @@
+/*
+ *  Copyright (c) 2017 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 "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vpx_dsp_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/buffer.h"
+#include "test/register_state_check.h"
+#include "vpx_ports/vpx_timer.h"
+
+namespace {
+
+using ::libvpx_test::ACMRandom;
+using ::libvpx_test::Buffer;
+
+typedef void (*AvgPredFunc)(uint8_t *a, const uint8_t *b, int w, int h,
+                            const uint8_t *c, int c_stride);
+
+uint8_t avg_with_rounding(uint8_t a, uint8_t b) { return (a + b + 1) >> 1; }
+
+void reference_pred(const Buffer<uint8_t> &pred, const Buffer<uint8_t> &ref,
+                    int width, int height, Buffer<uint8_t> *avg) {
+  for (int y = 0; y < height; ++y) {
+    for (int x = 0; x < width; ++x) {
+      avg->TopLeftPixel()[y * avg->stride() + x] =
+          avg_with_rounding(pred.TopLeftPixel()[y * pred.stride() + x],
+                            ref.TopLeftPixel()[y * ref.stride() + x]);
+    }
+  }
+}
+
+class AvgPredTest : public ::testing::TestWithParam<AvgPredFunc> {
+ public:
+  virtual void SetUp() {
+    avg_pred_func_ = GetParam();
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+  }
+
+ protected:
+  AvgPredFunc avg_pred_func_;
+  ACMRandom rnd_;
+};
+
+TEST_P(AvgPredTest, SizeCombinations) {
+  // This is called as part of the sub pixel variance. As such it must be one of
+  // the variance block sizes.
+
+  for (int width_pow = 2; width_pow <= 6; ++width_pow) {
+    for (int height_pow = width_pow - 1; height_pow <= width_pow + 1;
+         ++height_pow) {
+      // Don't test 4x2 or 64x128
+      if (height_pow == 1 || height_pow == 7) continue;
+
+      // The sse2 special-cases when ref width == stride, so make sure to test
+      // it.
+      for (int ref_padding = 0; ref_padding < 2; ref_padding++) {
+        const int width = 1 << width_pow;
+        const int height = 1 << height_pow;
+        // Only the reference buffer may have a stride not equal to width.
+        Buffer<uint8_t> ref =
+            Buffer<uint8_t>(width, height, ref_padding ? 8 : 0);
+        ASSERT_TRUE(ref.Init());
+        Buffer<uint8_t> pred = Buffer<uint8_t>(width, height, 0, 16);
+        ASSERT_TRUE(pred.Init());
+        Buffer<uint8_t> avg_ref = Buffer<uint8_t>(width, height, 0, 16);
+        ASSERT_TRUE(avg_ref.Init());
+        Buffer<uint8_t> avg_chk = Buffer<uint8_t>(width, height, 0, 16);
+        ASSERT_TRUE(avg_chk.Init());
+
+        ref.Set(&rnd_, &ACMRandom::Rand8);
+        pred.Set(&rnd_, &ACMRandom::Rand8);
+
+        reference_pred(pred, ref, width, height, &avg_ref);
+        ASM_REGISTER_STATE_CHECK(
+            avg_pred_func_(avg_chk.TopLeftPixel(), pred.TopLeftPixel(), width,
+                           height, ref.TopLeftPixel(), ref.stride()));
+
+        EXPECT_TRUE(avg_chk.CheckValues(avg_ref));
+        if (HasFailure()) {
+          printf("Width: %d Height: %d\n", width, height);
+          avg_chk.PrintDifference(avg_ref);
+          return;
+        }
+      }
+    }
+  }
+}
+
+TEST_P(AvgPredTest, CompareReferenceRandom) {
+  const int width = 64;
+  const int height = 32;
+  Buffer<uint8_t> ref = Buffer<uint8_t>(width, height, 8);
+  ASSERT_TRUE(ref.Init());
+  Buffer<uint8_t> pred = Buffer<uint8_t>(width, height, 0, 16);
+  ASSERT_TRUE(pred.Init());
+  Buffer<uint8_t> avg_ref = Buffer<uint8_t>(width, height, 0, 16);
+  ASSERT_TRUE(avg_ref.Init());
+  Buffer<uint8_t> avg_chk = Buffer<uint8_t>(width, height, 0, 16);
+  ASSERT_TRUE(avg_chk.Init());
+
+  for (int i = 0; i < 500; ++i) {
+    ref.Set(&rnd_, &ACMRandom::Rand8);
+    pred.Set(&rnd_, &ACMRandom::Rand8);
+
+    reference_pred(pred, ref, width, height, &avg_ref);
+    ASM_REGISTER_STATE_CHECK(avg_pred_func_(avg_chk.TopLeftPixel(),
+                                            pred.TopLeftPixel(), width, height,
+                                            ref.TopLeftPixel(), ref.stride()));
+    EXPECT_TRUE(avg_chk.CheckValues(avg_ref));
+    if (HasFailure()) {
+      printf("Width: %d Height: %d\n", width, height);
+      avg_chk.PrintDifference(avg_ref);
+      return;
+    }
+  }
+}
+
+TEST_P(AvgPredTest, DISABLED_Speed) {
+  for (int width_pow = 2; width_pow <= 6; ++width_pow) {
+    for (int height_pow = width_pow - 1; height_pow <= width_pow + 1;
+         ++height_pow) {
+      // Don't test 4x2 or 64x128
+      if (height_pow == 1 || height_pow == 7) continue;
+
+      for (int ref_padding = 0; ref_padding < 2; ref_padding++) {
+        const int width = 1 << width_pow;
+        const int height = 1 << height_pow;
+        Buffer<uint8_t> ref =
+            Buffer<uint8_t>(width, height, ref_padding ? 8 : 0);
+        ASSERT_TRUE(ref.Init());
+        Buffer<uint8_t> pred = Buffer<uint8_t>(width, height, 0, 16);
+        ASSERT_TRUE(pred.Init());
+        Buffer<uint8_t> avg = Buffer<uint8_t>(width, height, 0, 16);
+        ASSERT_TRUE(avg.Init());
+
+        ref.Set(&rnd_, &ACMRandom::Rand8);
+        pred.Set(&rnd_, &ACMRandom::Rand8);
+
+        vpx_usec_timer timer;
+        vpx_usec_timer_start(&timer);
+        for (int i = 0; i < 10000000 / (width * height); ++i) {
+          avg_pred_func_(avg.TopLeftPixel(), pred.TopLeftPixel(), width, height,
+                         ref.TopLeftPixel(), ref.stride());
+        }
+        vpx_usec_timer_mark(&timer);
+
+        const int elapsed_time =
+            static_cast<int>(vpx_usec_timer_elapsed(&timer));
+        printf("Average Test (ref_padding: %d) %dx%d time: %5d us\n",
+               ref_padding, width, height, elapsed_time);
+      }
+    }
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(C, AvgPredTest,
+                        ::testing::Values(&vpx_comp_avg_pred_c));
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, AvgPredTest,
+                        ::testing::Values(&vpx_comp_avg_pred_sse2));
+#endif  // HAVE_SSE2
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, AvgPredTest,
+                        ::testing::Values(&vpx_comp_avg_pred_neon));
+#endif  // HAVE_NEON
+
+#if HAVE_VSX
+INSTANTIATE_TEST_CASE_P(VSX, AvgPredTest,
+                        ::testing::Values(&vpx_comp_avg_pred_vsx));
+#endif  // HAVE_VSX
+}  // namespace
--- a/test/convolve_test.cc
+++ b/test/convolve_test.cc
@@ -25,6 +25,7 @@
 #include "vpx_dsp/vpx_filter.h"
 #include "vpx_mem/vpx_mem.h"
 #include "vpx_ports/mem.h"
+#include "vpx_ports/vpx_timer.h"

 namespace {

@@ -32,9 +33,9 @@ static const unsigned int kMaxDimension = 64;

 typedef void (*ConvolveFunc)(const uint8_t *src, ptrdiff_t src_stride,
                             uint8_t *dst, ptrdiff_t dst_stride,
-                             const int16_t *filter_x, int filter_x_stride,
-                             const int16_t *filter_y, int filter_y_stride,
-                             int w, int h);
+                             const InterpKernel *filter, int x0_q4,
+                             int x_step_q4, int y0_q4, int y_step_q4, int w,
+                             int h);

 typedef void (*WrapperFilterBlock2d8Func)(
    const uint8_t *src_ptr, const unsigned int src_stride,
@@ -300,9 +301,9 @@ void wrapper_filter_average_block2d_8_c(
    filter_average_block2d_8_c(src_ptr, src_stride, hfilter, vfilter, dst_ptr,
                               dst_stride, output_width, output_height);
  } else {
-    highbd_filter_average_block2d_8_c(CONVERT_TO_SHORTPTR(src_ptr), src_stride,
+    highbd_filter_average_block2d_8_c(CAST_TO_SHORTPTR(src_ptr), src_stride,
                                      hfilter, vfilter,
-                                      CONVERT_TO_SHORTPTR(dst_ptr), dst_stride,
+                                      CAST_TO_SHORTPTR(dst_ptr), dst_stride,
                                      output_width, output_height, use_highbd);
  }
 #else
@@ -323,8 +324,8 @@ void wrapper_filter_block2d_8_c(const uint8_t *src_ptr,
    filter_block2d_8_c(src_ptr, src_stride, hfilter, vfilter, dst_ptr,
                       dst_stride, output_width, output_height);
  } else {
-    highbd_filter_block2d_8_c(CONVERT_TO_SHORTPTR(src_ptr), src_stride, hfilter,
-                              vfilter, CONVERT_TO_SHORTPTR(dst_ptr), dst_stride,
+    highbd_filter_block2d_8_c(CAST_TO_SHORTPTR(src_ptr), src_stride, hfilter,
+                              vfilter, CAST_TO_SHORTPTR(dst_ptr), dst_stride,
                              output_width, output_height, use_highbd);
  }
 #else
@@ -459,7 +460,7 @@ class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
    if (UUT_->use_highbd_ == 0) {
      return input_ + offset;
    } else {
-      return CONVERT_TO_BYTEPTR(input16_) + offset;
+      return CAST_TO_BYTEPTR(input16_ + offset);
    }
 #else
    return input_ + offset;
@@ -472,7 +473,7 @@ class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
    if (UUT_->use_highbd_ == 0) {
      return output_ + offset;
    } else {
-      return CONVERT_TO_BYTEPTR(output16_) + offset;
+      return CAST_TO_BYTEPTR(output16_ + offset);
    }
 #else
    return output_ + offset;
@@ -485,7 +486,7 @@ class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
    if (UUT_->use_highbd_ == 0) {
      return output_ref_ + offset;
    } else {
-      return CONVERT_TO_BYTEPTR(output16_ref_) + offset;
+      return CAST_TO_BYTEPTR(output16_ref_ + offset);
    }
 #else
    return output_ref_ + offset;
@@ -497,7 +498,7 @@ class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
    if (UUT_->use_highbd_ == 0) {
      return list[index];
    } else {
-      return CONVERT_TO_SHORTPTR(list)[index];
+      return CAST_TO_SHORTPTR(list)[index];
    }
 #else
    return list[index];
@@ -509,7 +510,7 @@ class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
    if (UUT_->use_highbd_ == 0) {
      list[index] = (uint8_t)val;
    } else {
-      CONVERT_TO_SHORTPTR(list)[index] = val;
+      CAST_TO_SHORTPTR(list)[index] = val;
    }
 #else
    list[index] = (uint8_t)val;
@@ -539,12 +540,167 @@ uint16_t *ConvolveTest::output16_ref_ = NULL;

 TEST_P(ConvolveTest, GuardBlocks) { CheckGuardBlocks(); }

+TEST_P(ConvolveTest, DISABLED_Copy_Speed) {
+  const uint8_t *const in = input();
+  uint8_t *const out = output();
+  const int kNumTests = 5000000;
+  const int width = Width();
+  const int height = Height();
+  vpx_usec_timer timer;
+
+  vpx_usec_timer_start(&timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    UUT_->copy_[0](in, kInputStride, out, kOutputStride, NULL, 0, 0, 0, 0,
+                   width, height);
+  }
+  vpx_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("convolve_copy_%dx%d_%d: %d us\n", width, height,
+         UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
+}
+
+TEST_P(ConvolveTest, DISABLED_Avg_Speed) {
+  const uint8_t *const in = input();
+  uint8_t *const out = output();
+  const int kNumTests = 5000000;
+  const int width = Width();
+  const int height = Height();
+  vpx_usec_timer timer;
+
+  vpx_usec_timer_start(&timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    UUT_->copy_[1](in, kInputStride, out, kOutputStride, NULL, 0, 0, 0, 0,
+                   width, height);
+  }
+  vpx_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("convolve_avg_%dx%d_%d: %d us\n", width, height,
+         UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
+}
+
+TEST_P(ConvolveTest, DISABLED_Scale_Speed) {
+  const uint8_t *const in = input();
+  uint8_t *const out = output();
+  const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP];
+  const int kNumTests = 5000000;
+  const int width = Width();
+  const int height = Height();
+  vpx_usec_timer timer;
+
+  SetConstantInput(127);
+
+  vpx_usec_timer_start(&timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    UUT_->shv8_[0](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
+                   width, height);
+  }
+  vpx_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("convolve_scale_%dx%d_%d: %d us\n", width, height,
+         UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
+}
+
+TEST_P(ConvolveTest, DISABLED_8Tap_Speed) {
+  const uint8_t *const in = input();
+  uint8_t *const out = output();
+  const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP_SHARP];
+  const int kNumTests = 5000000;
+  const int width = Width();
+  const int height = Height();
+  vpx_usec_timer timer;
+
+  SetConstantInput(127);
+
+  vpx_usec_timer_start(&timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    UUT_->hv8_[0](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
+                  width, height);
+  }
+  vpx_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("convolve8_%dx%d_%d: %d us\n", width, height,
+         UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
+}
+
+TEST_P(ConvolveTest, DISABLED_8Tap_Horiz_Speed) {
+  const uint8_t *const in = input();
+  uint8_t *const out = output();
+  const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP_SHARP];
+  const int kNumTests = 5000000;
+  const int width = Width();
+  const int height = Height();
+  vpx_usec_timer timer;
+
+  SetConstantInput(127);
+
+  vpx_usec_timer_start(&timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    UUT_->h8_[0](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
+                 width, height);
+  }
+  vpx_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("convolve8_horiz_%dx%d_%d: %d us\n", width, height,
+         UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
+}
+
+TEST_P(ConvolveTest, DISABLED_8Tap_Vert_Speed) {
+  const uint8_t *const in = input();
+  uint8_t *const out = output();
+  const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP_SHARP];
+  const int kNumTests = 5000000;
+  const int width = Width();
+  const int height = Height();
+  vpx_usec_timer timer;
+
+  SetConstantInput(127);
+
+  vpx_usec_timer_start(&timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    UUT_->v8_[0](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
+                 width, height);
+  }
+  vpx_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("convolve8_vert_%dx%d_%d: %d us\n", width, height,
+         UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
+}
+
+TEST_P(ConvolveTest, DISABLED_8Tap_Avg_Speed) {
+  const uint8_t *const in = input();
+  uint8_t *const out = output();
+  const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP_SHARP];
+  const int kNumTests = 5000000;
+  const int width = Width();
+  const int height = Height();
+  vpx_usec_timer timer;
+
+  SetConstantInput(127);
+
+  vpx_usec_timer_start(&timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    UUT_->hv8_[1](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
+                  width, height);
+  }
+  vpx_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("convolve8_avg_%dx%d_%d: %d us\n", width, height,
+         UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
+}
+
 TEST_P(ConvolveTest, Copy) {
  uint8_t *const in = input();
  uint8_t *const out = output();

  ASM_REGISTER_STATE_CHECK(UUT_->copy_[0](in, kInputStride, out, kOutputStride,
-                                          NULL, 0, NULL, 0, Width(), Height()));
+                                          NULL, 0, 0, 0, 0, Width(), Height()));

  CheckGuardBlocks();

@@ -563,7 +719,7 @@ TEST_P(ConvolveTest, Avg) {
  CopyOutputToRef();

  ASM_REGISTER_STATE_CHECK(UUT_->copy_[1](in, kInputStride, out, kOutputStride,
-                                          NULL, 0, NULL, 0, Width(), Height()));
+                                          NULL, 0, 0, 0, 0, Width(), Height()));

  CheckGuardBlocks();

@@ -580,12 +736,10 @@ TEST_P(ConvolveTest, Avg) {
 TEST_P(ConvolveTest, CopyHoriz) {
  uint8_t *const in = input();
  uint8_t *const out = output();
-  DECLARE_ALIGNED(256, const int16_t,
-                  filter8[8]) = { 0, 0, 0, 128, 0, 0, 0, 0 };

  ASM_REGISTER_STATE_CHECK(UUT_->sh8_[0](in, kInputStride, out, kOutputStride,
-                                         filter8, 16, filter8, 16, Width(),
-                                         Height()));
+                                         vp9_filter_kernels[0], 0, 16, 0, 16,
+                                         Width(), Height()));

  CheckGuardBlocks();

@@ -600,12 +754,10 @@ TEST_P(ConvolveTest, CopyHoriz) {
 TEST_P(ConvolveTest, CopyVert) {
  uint8_t *const in = input();
  uint8_t *const out = output();
-  DECLARE_ALIGNED(256, const int16_t,
-                  filter8[8]) = { 0, 0, 0, 128, 0, 0, 0, 0 };

  ASM_REGISTER_STATE_CHECK(UUT_->sv8_[0](in, kInputStride, out, kOutputStride,
-                                         filter8, 16, filter8, 16, Width(),
-                                         Height()));
+                                         vp9_filter_kernels[0], 0, 16, 0, 16,
+                                         Width(), Height()));

  CheckGuardBlocks();

@@ -620,12 +772,10 @@ TEST_P(ConvolveTest, CopyVert) {
 TEST_P(ConvolveTest, Copy2D) {
  uint8_t *const in = input();
  uint8_t *const out = output();
-  DECLARE_ALIGNED(256, const int16_t,
-                  filter8[8]) = { 0, 0, 0, 128, 0, 0, 0, 0 };

  ASM_REGISTER_STATE_CHECK(UUT_->shv8_[0](in, kInputStride, out, kOutputStride,
-                                          filter8, 16, filter8, 16, Width(),
-                                          Height()));
+                                          vp9_filter_kernels[0], 0, 16, 0, 16,
+                                          Width(), Height()));

  CheckGuardBlocks();

@@ -661,7 +811,6 @@ TEST(ConvolveTest, FiltersWontSaturateWhenAddedPairwise) {
  }
 }

-const int16_t kInvalidFilter[8] = { 0 };
 const WrapperFilterBlock2d8Func wrapper_filter_block2d_8[2] = {
  wrapper_filter_block2d_8_c, wrapper_filter_average_block2d_8_c
 };
@@ -677,7 +826,7 @@ TEST_P(ConvolveTest, MatchesReferenceSubpixelFilter) {
    if (UUT_->use_highbd_ == 0) {
      ref = ref8;
    } else {
-      ref = CONVERT_TO_BYTEPTR(ref16);
+      ref = CAST_TO_BYTEPTR(ref16);
    }
 #else
    uint8_t ref[kOutputStride * kMaxDimension];
@@ -714,21 +863,21 @@ TEST_P(ConvolveTest, MatchesReferenceSubpixelFilter) {
                                      Width(), Height(), UUT_->use_highbd_);

          if (filter_x && filter_y)
-            ASM_REGISTER_STATE_CHECK(UUT_->hv8_[i](
-                in, kInputStride, out, kOutputStride, filters[filter_x], 16,
-                filters[filter_y], 16, Width(), Height()));
+            ASM_REGISTER_STATE_CHECK(
+                UUT_->hv8_[i](in, kInputStride, out, kOutputStride, filters,
+                              filter_x, 16, filter_y, 16, Width(), Height()));
          else if (filter_y)
-            ASM_REGISTER_STATE_CHECK(UUT_->v8_[i](
-                in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
-                filters[filter_y], 16, Width(), Height()));
+            ASM_REGISTER_STATE_CHECK(
+                UUT_->v8_[i](in, kInputStride, out, kOutputStride, filters, 0,
+                             16, filter_y, 16, Width(), Height()));
          else if (filter_x)
-            ASM_REGISTER_STATE_CHECK(UUT_->h8_[i](
-                in, kInputStride, out, kOutputStride, filters[filter_x], 16,
-                kInvalidFilter, 16, Width(), Height()));
+            ASM_REGISTER_STATE_CHECK(
+                UUT_->h8_[i](in, kInputStride, out, kOutputStride, filters,
+                             filter_x, 16, 0, 16, Width(), Height()));
          else
-            ASM_REGISTER_STATE_CHECK(UUT_->copy_[i](
-                in, kInputStride, out, kOutputStride, kInvalidFilter, 0,
-                kInvalidFilter, 0, Width(), Height()));
+            ASM_REGISTER_STATE_CHECK(UUT_->copy_[i](in, kInputStride, out,
+                                                    kOutputStride, NULL, 0, 0,
+                                                    0, 0, Width(), Height()));

          CheckGuardBlocks();

@@ -756,7 +905,7 @@ TEST_P(ConvolveTest, FilterExtremes) {
  if (UUT_->use_highbd_ == 0) {
    ref = ref8;
  } else {
-    ref = CONVERT_TO_BYTEPTR(ref16);
+    ref = CAST_TO_BYTEPTR(ref16);
  }
 #else
  uint8_t ref[kOutputStride * kMaxDimension];
@@ -812,21 +961,21 @@ TEST_P(ConvolveTest, FilterExtremes) {
                                       filters[filter_y], ref, kOutputStride,
                                       Width(), Height(), UUT_->use_highbd_);
            if (filter_x && filter_y)
-              ASM_REGISTER_STATE_CHECK(UUT_->hv8_[0](
-                  in, kInputStride, out, kOutputStride, filters[filter_x], 16,
-                  filters[filter_y], 16, Width(), Height()));
+              ASM_REGISTER_STATE_CHECK(
+                  UUT_->hv8_[0](in, kInputStride, out, kOutputStride, filters,
+                                filter_x, 16, filter_y, 16, Width(), Height()));
            else if (filter_y)
-              ASM_REGISTER_STATE_CHECK(UUT_->v8_[0](
-                  in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
-                  filters[filter_y], 16, Width(), Height()));
+              ASM_REGISTER_STATE_CHECK(
+                  UUT_->v8_[0](in, kInputStride, out, kOutputStride, filters, 0,
+                               16, filter_y, 16, Width(), Height()));
            else if (filter_x)
-              ASM_REGISTER_STATE_CHECK(UUT_->h8_[0](
-                  in, kInputStride, out, kOutputStride, filters[filter_x], 16,
-                  kInvalidFilter, 16, Width(), Height()));
+              ASM_REGISTER_STATE_CHECK(
+                  UUT_->h8_[0](in, kInputStride, out, kOutputStride, filters,
+                               filter_x, 16, 0, 16, Width(), Height()));
            else
-              ASM_REGISTER_STATE_CHECK(UUT_->copy_[0](
-                  in, kInputStride, out, kOutputStride, kInvalidFilter, 0,
-                  kInvalidFilter, 0, Width(), Height()));
+              ASM_REGISTER_STATE_CHECK(UUT_->copy_[0](in, kInputStride, out,
+                                                      kOutputStride, NULL, 0, 0,
+                                                      0, 0, Width(), Height()));

            for (int y = 0; y < Height(); ++y) {
              for (int x = 0; x < Width(); ++x)
@@ -845,33 +994,51 @@ TEST_P(ConvolveTest, FilterExtremes) {

 /* This test exercises that enough rows and columns are filtered with every
   possible initial fractional positions and scaling steps. */
+#if !CONFIG_VP9_HIGHBITDEPTH
+static const ConvolveFunc scaled_2d_c_funcs[2] = { vpx_scaled_2d_c,
+                                                   vpx_scaled_avg_2d_c };
+
 TEST_P(ConvolveTest, CheckScalingFiltering) {
  uint8_t *const in = input();
  uint8_t *const out = output();
-  const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP];
+  uint8_t ref[kOutputStride * kMaxDimension];

-  SetConstantInput(127);
+  ::libvpx_test::ACMRandom prng;
+  for (int y = 0; y < Height(); ++y) {
+    for (int x = 0; x < Width(); ++x) {
+      const uint16_t r = prng.Rand8Extremes();
+      assign_val(in, y * kInputStride + x, r);
+    }
+  }

-  for (int frac = 0; frac < 16; ++frac) {
-    for (int step = 1; step <= 32; ++step) {
-      /* Test the horizontal and vertical filters in combination. */
-      ASM_REGISTER_STATE_CHECK(
-          UUT_->shv8_[0](in, kInputStride, out, kOutputStride, eighttap[frac],
-                         step, eighttap[frac], step, Width(), Height()));
+  for (int i = 0; i < 2; ++i) {
+    for (INTERP_FILTER filter_type = 0; filter_type < 4; ++filter_type) {
+      const InterpKernel *const eighttap = vp9_filter_kernels[filter_type];
+      for (int frac = 0; frac < 16; ++frac) {
+        for (int step = 1; step <= 32; ++step) {
+          /* Test the horizontal and vertical filters in combination. */
+          scaled_2d_c_funcs[i](in, kInputStride, ref, kOutputStride, eighttap,
+                               frac, step, frac, step, Width(), Height());
+          ASM_REGISTER_STATE_CHECK(
+              UUT_->shv8_[i](in, kInputStride, out, kOutputStride, eighttap,
+                             frac, step, frac, step, Width(), Height()));

-      CheckGuardBlocks();
+          CheckGuardBlocks();

-      for (int y = 0; y < Height(); ++y) {
-        for (int x = 0; x < Width(); ++x) {
-          ASSERT_EQ(lookup(in, y * kInputStride + x),
-                    lookup(out, y * kOutputStride + x))
-              << "x == " << x << ", y == " << y << ", frac == " << frac
-              << ", step == " << step;
+          for (int y = 0; y < Height(); ++y) {
+            for (int x = 0; x < Width(); ++x) {
+              ASSERT_EQ(lookup(ref, y * kOutputStride + x),
+                        lookup(out, y * kOutputStride + x))
+                  << "x == " << x << ", y == " << y << ", frac == " << frac
+                  << ", step == " << step;
+            }
+          }
        }
      }
    }
  }
 }
+#endif

 using std::tr1::make_tuple;

@@ -879,10 +1046,11 @@ using std::tr1::make_tuple;
 #define WRAP(func, bd)                                                       \
  void wrap_##func##_##bd(                                                   \
      const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,                \
-      ptrdiff_t dst_stride, const int16_t *filter_x, int filter_x_stride,    \
-      const int16_t *filter_y, int filter_y_stride, int w, int h) {          \
-    vpx_highbd_##func(src, src_stride, dst, dst_stride, filter_x,            \
-                      filter_x_stride, filter_y, filter_y_stride, w, h, bd); \
+      ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4,           \
+      int x_step_q4, int y0_q4, int y_step_q4, int w, int h) {               \
+    vpx_highbd_##func(reinterpret_cast<const uint16_t *>(src), src_stride,   \
+                      reinterpret_cast<uint16_t *>(dst), dst_stride, filter, \
+                      x0_q4, x_step_q4, y0_q4, y_step_q4, w, h, bd);         \
  }

 #if HAVE_SSE2 && ARCH_X86_64
@@ -912,6 +1080,35 @@ WRAP(convolve8_sse2, 12)
 WRAP(convolve8_avg_sse2, 12)
 #endif  // HAVE_SSE2 && ARCH_X86_64

+#if HAVE_AVX2
+WRAP(convolve_copy_avx2, 8)
+WRAP(convolve_avg_avx2, 8)
+WRAP(convolve8_horiz_avx2, 8)
+WRAP(convolve8_avg_horiz_avx2, 8)
+WRAP(convolve8_vert_avx2, 8)
+WRAP(convolve8_avg_vert_avx2, 8)
+WRAP(convolve8_avx2, 8)
+WRAP(convolve8_avg_avx2, 8)
+
+WRAP(convolve_copy_avx2, 10)
+WRAP(convolve_avg_avx2, 10)
+WRAP(convolve8_avx2, 10)
+WRAP(convolve8_horiz_avx2, 10)
+WRAP(convolve8_vert_avx2, 10)
+WRAP(convolve8_avg_avx2, 10)
+WRAP(convolve8_avg_horiz_avx2, 10)
+WRAP(convolve8_avg_vert_avx2, 10)
+
+WRAP(convolve_copy_avx2, 12)
+WRAP(convolve_avg_avx2, 12)
+WRAP(convolve8_avx2, 12)
+WRAP(convolve8_horiz_avx2, 12)
+WRAP(convolve8_vert_avx2, 12)
+WRAP(convolve8_avg_avx2, 12)
+WRAP(convolve8_avg_horiz_avx2, 12)
+WRAP(convolve8_avg_vert_avx2, 12)
+#endif  // HAVE_AVX2
+
 #if HAVE_NEON
 WRAP(convolve_copy_neon, 8)
 WRAP(convolve_avg_neon, 8)
@@ -1057,18 +1254,48 @@ INSTANTIATE_TEST_CASE_P(SSSE3, ConvolveTest,
                        ::testing::ValuesIn(kArrayConvolve8_ssse3));
 #endif

-#if HAVE_AVX2 && HAVE_SSSE3
+#if HAVE_AVX2
+#if CONFIG_VP9_HIGHBITDEPTH
+const ConvolveFunctions convolve8_avx2(
+    wrap_convolve_copy_avx2_8, wrap_convolve_avg_avx2_8,
+    wrap_convolve8_horiz_avx2_8, wrap_convolve8_avg_horiz_avx2_8,
+    wrap_convolve8_vert_avx2_8, wrap_convolve8_avg_vert_avx2_8,
+    wrap_convolve8_avx2_8, wrap_convolve8_avg_avx2_8, wrap_convolve8_horiz_c_8,
+    wrap_convolve8_avg_horiz_c_8, wrap_convolve8_vert_c_8,
+    wrap_convolve8_avg_vert_c_8, wrap_convolve8_c_8, wrap_convolve8_avg_c_8, 8);
+const ConvolveFunctions convolve10_avx2(
+    wrap_convolve_copy_avx2_10, wrap_convolve_avg_avx2_10,
+    wrap_convolve8_horiz_avx2_10, wrap_convolve8_avg_horiz_avx2_10,
+    wrap_convolve8_vert_avx2_10, wrap_convolve8_avg_vert_avx2_10,
+    wrap_convolve8_avx2_10, wrap_convolve8_avg_avx2_10,
+    wrap_convolve8_horiz_c_10, wrap_convolve8_avg_horiz_c_10,
+    wrap_convolve8_vert_c_10, wrap_convolve8_avg_vert_c_10, wrap_convolve8_c_10,
+    wrap_convolve8_avg_c_10, 10);
+const ConvolveFunctions convolve12_avx2(
+    wrap_convolve_copy_avx2_12, wrap_convolve_avg_avx2_12,
+    wrap_convolve8_horiz_avx2_12, wrap_convolve8_avg_horiz_avx2_12,
+    wrap_convolve8_vert_avx2_12, wrap_convolve8_avg_vert_avx2_12,
+    wrap_convolve8_avx2_12, wrap_convolve8_avg_avx2_12,
+    wrap_convolve8_horiz_c_12, wrap_convolve8_avg_horiz_c_12,
+    wrap_convolve8_vert_c_12, wrap_convolve8_avg_vert_c_12, wrap_convolve8_c_12,
+    wrap_convolve8_avg_c_12, 12);
+const ConvolveParam kArrayConvolve8_avx2[] = { ALL_SIZES(convolve8_avx2),
+                                               ALL_SIZES(convolve10_avx2),
+                                               ALL_SIZES(convolve12_avx2) };
+INSTANTIATE_TEST_CASE_P(AVX2, ConvolveTest,
+                        ::testing::ValuesIn(kArrayConvolve8_avx2));
+#else   // !CONFIG_VP9_HIGHBITDEPTH
 const ConvolveFunctions convolve8_avx2(
    vpx_convolve_copy_c, vpx_convolve_avg_c, vpx_convolve8_horiz_avx2,
-    vpx_convolve8_avg_horiz_ssse3, vpx_convolve8_vert_avx2,
-    vpx_convolve8_avg_vert_ssse3, vpx_convolve8_avx2, vpx_convolve8_avg_ssse3,
+    vpx_convolve8_avg_horiz_avx2, vpx_convolve8_vert_avx2,
+    vpx_convolve8_avg_vert_avx2, vpx_convolve8_avx2, vpx_convolve8_avg_avx2,
    vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
    vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
-
 const ConvolveParam kArrayConvolve8_avx2[] = { ALL_SIZES(convolve8_avx2) };
 INSTANTIATE_TEST_CASE_P(AVX2, ConvolveTest,
                        ::testing::ValuesIn(kArrayConvolve8_avx2));
-#endif  // HAVE_AVX2 && HAVE_SSSE3
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_AVX2

 #if HAVE_NEON
 #if CONFIG_VP9_HIGHBITDEPTH
@@ -1105,7 +1332,7 @@ const ConvolveFunctions convolve8_neon(
    vpx_convolve8_avg_horiz_neon, vpx_convolve8_vert_neon,
    vpx_convolve8_avg_vert_neon, vpx_convolve8_neon, vpx_convolve8_avg_neon,
    vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
-    vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
+    vpx_scaled_avg_vert_c, vpx_scaled_2d_neon, vpx_scaled_avg_2d_c, 0);

 const ConvolveParam kArrayConvolve_neon[] = { ALL_SIZES(convolve8_neon) };
 #endif  // CONFIG_VP9_HIGHBITDEPTH
@@ -1132,10 +1359,22 @@ const ConvolveFunctions convolve8_msa(
    vpx_convolve8_avg_horiz_msa, vpx_convolve8_vert_msa,
    vpx_convolve8_avg_vert_msa, vpx_convolve8_msa, vpx_convolve8_avg_msa,
    vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
-    vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
+    vpx_scaled_avg_vert_c, vpx_scaled_2d_msa, vpx_scaled_avg_2d_c, 0);

 const ConvolveParam kArrayConvolve8_msa[] = { ALL_SIZES(convolve8_msa) };
 INSTANTIATE_TEST_CASE_P(MSA, ConvolveTest,
                        ::testing::ValuesIn(kArrayConvolve8_msa));
 #endif  // HAVE_MSA
+
+#if HAVE_VSX
+const ConvolveFunctions convolve8_vsx(
+    vpx_convolve_copy_vsx, vpx_convolve_avg_vsx, vpx_convolve8_horiz_vsx,
+    vpx_convolve8_avg_horiz_vsx, vpx_convolve8_vert_vsx,
+    vpx_convolve8_avg_vert_vsx, vpx_convolve8_vsx, vpx_convolve8_avg_vsx,
+    vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
+    vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
+const ConvolveParam kArrayConvolve_vsx[] = { ALL_SIZES(convolve8_vsx) };
+INSTANTIATE_TEST_CASE_P(VSX, ConvolveTest,
+                        ::testing::ValuesIn(kArrayConvolve_vsx));
+#endif  // HAVE_VSX
 }  // namespace
--- a/test/datarate_test.cc
+++ b/test/datarate_test.cc
@@ -44,6 +44,7 @@ class DatarateTestLarge
    denoiser_offon_test_ = 0;
    denoiser_offon_period_ = -1;
    gf_boost_ = 0;
+    use_roi_ = 0;
  }

  virtual void PreEncodeFrameHook(::libvpx_test::VideoSource *video,
@@ -54,6 +55,12 @@ class DatarateTestLarge
      encoder->Control(VP8E_SET_GF_CBR_BOOST_PCT, gf_boost_);
    }

+#if CONFIG_VP8_ENCODER
+    if (use_roi_ == 1) {
+      encoder->Control(VP8E_SET_ROI_MAP, &roi_);
+    }
+#endif
+
    if (denoiser_offon_test_) {
      ASSERT_GT(denoiser_offon_period_, 0)
          << "denoiser_offon_period_ is not positive.";
@@ -91,8 +98,8 @@ class DatarateTestLarge
    const bool key_frame =
        (pkt->data.frame.flags & VPX_FRAME_IS_KEY) ? true : false;
    if (!key_frame) {
-      ASSERT_GE(bits_in_buffer_model_, 0) << "Buffer Underrun at frame "
-                                          << pkt->data.frame.pts;
+      ASSERT_GE(bits_in_buffer_model_, 0)
+          << "Buffer Underrun at frame " << pkt->data.frame.pts;
    }

    const int64_t frame_size_in_bits = pkt->data.frame.sz * 8;
@@ -145,6 +152,8 @@ class DatarateTestLarge
  int denoiser_offon_period_;
  int set_cpu_used_;
  int gf_boost_;
+  int use_roi_;
+  vpx_roi_map_t roi_;
 };

 #if CONFIG_TEMPORAL_DENOISING
@@ -258,14 +267,6 @@ TEST_P(DatarateTestLarge, ChangingDropFrameThresh) {
  }
 }

-// Disabled for tsan, see:
-// https://bugs.chromium.org/p/webm/issues/detail?id=1049
-#if defined(__has_feature)
-#if __has_feature(thread_sanitizer)
-#define BUILDING_WITH_TSAN
-#endif
-#endif
-#ifndef BUILDING_WITH_TSAN
 TEST_P(DatarateTestLarge, DropFramesMultiThreads) {
  denoiser_on_ = 0;
  cfg_.rc_buf_initial_sz = 500;
@@ -285,7 +286,6 @@ TEST_P(DatarateTestLarge, DropFramesMultiThreads) {
  ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.4)
      << " The datarate for the file missed the target!";
 }
-#endif  // !BUILDING_WITH_TSAN

 class DatarateTestRealTime : public DatarateTestLarge {
 public:
@@ -402,10 +402,6 @@ TEST_P(DatarateTestRealTime, ChangingDropFrameThresh) {
  }
 }

-// Disabled for tsan, see:
-// https://bugs.chromium.org/p/webm/issues/detail?id=1049
-
-#ifndef BUILDING_WITH_TSAN
 TEST_P(DatarateTestRealTime, DropFramesMultiThreads) {
  denoiser_on_ = 0;
  cfg_.rc_buf_initial_sz = 500;
@@ -426,7 +422,67 @@ TEST_P(DatarateTestRealTime, DropFramesMultiThreads) {
  ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.4)
      << " The datarate for the file missed the target!";
 }
-#endif
+
+TEST_P(DatarateTestRealTime, RegionOfInterest) {
+  denoiser_on_ = 0;
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_dropframe_thresh = 0;
+  cfg_.rc_max_quantizer = 56;
+  cfg_.rc_end_usage = VPX_CBR;
+  // Encode using multiple threads.
+  cfg_.g_threads = 2;
+
+  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 300);
+  cfg_.rc_target_bitrate = 450;
+  cfg_.g_w = 352;
+  cfg_.g_h = 288;
+
+  ResetModel();
+
+  // Set ROI parameters
+  use_roi_ = 1;
+  memset(&roi_, 0, sizeof(roi_));
+
+  roi_.rows = (cfg_.g_h + 15) / 16;
+  roi_.cols = (cfg_.g_w + 15) / 16;
+
+  roi_.delta_q[0] = 0;
+  roi_.delta_q[1] = -20;
+  roi_.delta_q[2] = 0;
+  roi_.delta_q[3] = 0;
+
+  roi_.delta_lf[0] = 0;
+  roi_.delta_lf[1] = -20;
+  roi_.delta_lf[2] = 0;
+  roi_.delta_lf[3] = 0;
+
+  roi_.static_threshold[0] = 0;
+  roi_.static_threshold[1] = 1000;
+  roi_.static_threshold[2] = 0;
+  roi_.static_threshold[3] = 0;
+
+  // Use 2 states: 1 is center square, 0 is the rest.
+  roi_.roi_map =
+      (uint8_t *)calloc(roi_.rows * roi_.cols, sizeof(*roi_.roi_map));
+  for (unsigned int i = 0; i < roi_.rows; ++i) {
+    for (unsigned int j = 0; j < roi_.cols; ++j) {
+      if (i > (roi_.rows >> 2) && i < ((roi_.rows * 3) >> 2) &&
+          j > (roi_.cols >> 2) && j < ((roi_.cols * 3) >> 2)) {
+        roi_.roi_map[i * roi_.cols + j] = 1;
+      }
+    }
+  }
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  ASSERT_GE(cfg_.rc_target_bitrate, effective_datarate_ * 0.95)
+      << " The datarate for the file exceeds the target!";
+
+  ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.4)
+      << " The datarate for the file missed the target!";
+
+  free(roi_.roi_map);
+}

 TEST_P(DatarateTestRealTime, GFBoost) {
  denoiser_on_ = 0;
@@ -482,6 +538,7 @@ class DatarateTestVP9Large
    }
    denoiser_offon_test_ = 0;
    denoiser_offon_period_ = -1;
+    frame_parallel_decoding_mode_ = 1;
  }

  //
@@ -496,8 +553,8 @@ class DatarateTestVP9Large
  //      2           6
  // 0          4            ....
  // LAST is always update on base/layer 0, GOLDEN is updated on layer 1.
-  // For this 3 layer example, the 2nd enhancement layer (layer 2) does not
-  // update any reference frames.
+  // For this 3 layer example, the 2nd enhancement layer (layer 2) updates
+  // the altref frame.
  int SetFrameFlags(int frame_num, int num_temp_layers) {
    int frame_flags = 0;
    if (num_temp_layers == 2) {
@@ -519,9 +576,8 @@ class DatarateTestVP9Large
        // Layer 1: predict from L, G, ARF; update G.
        frame_flags = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
      } else if ((frame_num - 1) % 2 == 0) {
-        // Layer 2: predict from L, G, ARF; update none.
-        frame_flags =
-            VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
+        // Layer 2: predict from L, G, ARF; update ARF.
+        frame_flags = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_LAST;
      }
    }
    return frame_flags;
@@ -561,6 +617,9 @@ class DatarateTestVP9Large
    }

    encoder->Control(VP9E_SET_NOISE_SENSITIVITY, denoiser_on_);
+    encoder->Control(VP9E_SET_TILE_COLUMNS, (cfg_.g_threads >> 1));
+    encoder->Control(VP9E_SET_FRAME_PARALLEL_DECODING,
+                     frame_parallel_decoding_mode_);

    if (cfg_.ts_number_layers > 1) {
      if (video->frame() == 0) {
@@ -599,8 +658,8 @@ class DatarateTestVP9Large
        duration * timebase_ * cfg_.rc_target_bitrate * 1000);

    // Buffer should not go negative.
-    ASSERT_GE(bits_in_buffer_model_, 0) << "Buffer Underrun at frame "
-                                        << pkt->data.frame.pts;
+    ASSERT_GE(bits_in_buffer_model_, 0)
+        << "Buffer Underrun at frame " << pkt->data.frame.pts;

    const size_t frame_size_in_bits = pkt->data.frame.sz * 8;

@@ -641,6 +700,7 @@ class DatarateTestVP9Large
  int denoiser_on_;
  int denoiser_offon_test_;
  int denoiser_offon_period_;
+  int frame_parallel_decoding_mode_;
 };

 // Check basic rate targeting for VBR mode with 0 lag.
@@ -659,7 +719,7 @@ TEST_P(DatarateTestVP9Large, BasicRateTargetingVBRLagZero) {
    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
    ASSERT_GE(effective_datarate_[0], cfg_.rc_target_bitrate * 0.75)
        << " The datarate for the file is lower than target by too much!";
-    ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.25)
+    ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.30)
        << " The datarate for the file is greater than target by too much!";
  }
 }
@@ -686,7 +746,37 @@ TEST_P(DatarateTestVP9Large, BasicRateTargetingVBRLagNonZero) {
    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
    ASSERT_GE(effective_datarate_[0], cfg_.rc_target_bitrate * 0.75)
        << " The datarate for the file is lower than target by too much!";
-    ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.25)
+    ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.30)
+        << " The datarate for the file is greater than target by too much!";
+  }
+}
+
+// Check basic rate targeting for VBR mode with non-zero lag, with
+// frame_parallel_decoding_mode off. This enables the adapt_coeff/mode/mv probs
+// since error_resilience is off.
+TEST_P(DatarateTestVP9Large, BasicRateTargetingVBRLagNonZeroFrameParDecOff) {
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 63;
+  cfg_.g_error_resilient = 0;
+  cfg_.rc_end_usage = VPX_VBR;
+  // For non-zero lag, rate control will work (be within bounds) for
+  // real-time mode.
+  if (deadline_ == VPX_DL_REALTIME) {
+    cfg_.g_lag_in_frames = 15;
+  } else {
+    cfg_.g_lag_in_frames = 0;
+  }
+
+  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 300);
+  for (int i = 400; i <= 800; i += 400) {
+    cfg_.rc_target_bitrate = i;
+    ResetModel();
+    frame_parallel_decoding_mode_ = 0;
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    ASSERT_GE(effective_datarate_[0], cfg_.rc_target_bitrate * 0.75)
+        << " The datarate for the file is lower than target by too much!";
+    ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.30)
        << " The datarate for the file is greater than target by too much!";
  }
 }
@@ -715,6 +805,33 @@ TEST_P(DatarateTestVP9Large, BasicRateTargeting) {
  }
 }

+// Check basic rate targeting for CBR mode, with frame_parallel_decoding_mode
+// off( and error_resilience off).
+TEST_P(DatarateTestVP9Large, BasicRateTargetingFrameParDecOff) {
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_buf_optimal_sz = 500;
+  cfg_.rc_buf_sz = 1000;
+  cfg_.rc_dropframe_thresh = 1;
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 63;
+  cfg_.rc_end_usage = VPX_CBR;
+  cfg_.g_lag_in_frames = 0;
+  cfg_.g_error_resilient = 0;
+
+  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 140);
+  for (int i = 150; i < 800; i += 200) {
+    cfg_.rc_target_bitrate = i;
+    ResetModel();
+    frame_parallel_decoding_mode_ = 0;
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    ASSERT_GE(effective_datarate_[0], cfg_.rc_target_bitrate * 0.85)
+        << " The datarate for the file is lower than target by too much!";
+    ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.15)
+        << " The datarate for the file is greater than target by too much!";
+  }
+}
+
 // Check basic rate targeting for CBR mode, with 2 threads and dropped frames.
 TEST_P(DatarateTestVP9Large, BasicRateTargetingDropFramesMultiThreads) {
  cfg_.rc_buf_initial_sz = 500;
@@ -759,7 +876,7 @@ TEST_P(DatarateTestVP9Large, BasicRateTargeting444) {
    ResetModel();
    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
    ASSERT_GE(static_cast<double>(cfg_.rc_target_bitrate),
-              effective_datarate_[0] * 0.85)
+              effective_datarate_[0] * 0.80)
        << " The datarate for the file exceeds the target by too much!";
    ASSERT_LE(static_cast<double>(cfg_.rc_target_bitrate),
              effective_datarate_[0] * 1.15)
@@ -792,26 +909,29 @@ TEST_P(DatarateTestVP9Large, ChangingDropFrameThresh) {
                                       30, 1, 0, 140);

  const int kDropFrameThreshTestStep = 30;
-  vpx_codec_pts_t last_drop = 140;
-  int last_num_drops = 0;
-  for (int i = 10; i < 100; i += kDropFrameThreshTestStep) {
-    cfg_.rc_dropframe_thresh = i;
-    ResetModel();
-    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-    ASSERT_GE(effective_datarate_[0], cfg_.rc_target_bitrate * 0.85)
-        << " The datarate for the file is lower than target by too much!";
-    ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.15)
-        << " The datarate for the file is greater than target by too much!";
-    ASSERT_LE(first_drop_, last_drop)
-        << " The first dropped frame for drop_thresh " << i
-        << " > first dropped frame for drop_thresh "
-        << i - kDropFrameThreshTestStep;
-    ASSERT_GE(num_drops_, last_num_drops * 0.85)
-        << " The number of dropped frames for drop_thresh " << i
-        << " < number of dropped frames for drop_thresh "
-        << i - kDropFrameThreshTestStep;
-    last_drop = first_drop_;
-    last_num_drops = num_drops_;
+  for (int j = 50; j <= 150; j += 100) {
+    cfg_.rc_target_bitrate = j;
+    vpx_codec_pts_t last_drop = 140;
+    int last_num_drops = 0;
+    for (int i = 10; i < 100; i += kDropFrameThreshTestStep) {
+      cfg_.rc_dropframe_thresh = i;
+      ResetModel();
+      ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+      ASSERT_GE(effective_datarate_[0], cfg_.rc_target_bitrate * 0.85)
+          << " The datarate for the file is lower than target by too much!";
+      ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.25)
+          << " The datarate for the file is greater than target by too much!";
+      ASSERT_LE(first_drop_, last_drop)
+          << " The first dropped frame for drop_thresh " << i
+          << " > first dropped frame for drop_thresh "
+          << i - kDropFrameThreshTestStep;
+      ASSERT_GE(num_drops_, last_num_drops * 0.85)
+          << " The number of dropped frames for drop_thresh " << i
+          << " < number of dropped frames for drop_thresh "
+          << i - kDropFrameThreshTestStep;
+      last_drop = first_drop_;
+      last_num_drops = num_drops_;
+    }
  }
 }

@@ -988,7 +1108,7 @@ TEST_P(DatarateTestVP9LargeDenoiser, LowNoise) {
 }

 // Check basic datarate targeting, for a single bitrate, when denoiser is on,
-// for clip with high noise level.
+// for clip with high noise level. Use 2 threads.
 TEST_P(DatarateTestVP9LargeDenoiser, HighNoise) {
  cfg_.rc_buf_initial_sz = 500;
  cfg_.rc_buf_optimal_sz = 500;
@@ -998,9 +1118,39 @@ TEST_P(DatarateTestVP9LargeDenoiser, HighNoise) {
  cfg_.rc_max_quantizer = 56;
  cfg_.rc_end_usage = VPX_CBR;
  cfg_.g_lag_in_frames = 0;
+  cfg_.g_threads = 2;

  ::libvpx_test::Y4mVideoSource video("noisy_clip_640_360.y4m", 0, 200);

+  // For the temporal denoiser (#if CONFIG_VP9_TEMPORAL_DENOISING),
+  // there is only one denoiser mode: kDenoiserOnYOnly(which is 1),
+  // but may add more modes in the future.
+  cfg_.rc_target_bitrate = 1000;
+  ResetModel();
+  // Turn on the denoiser.
+  denoiser_on_ = 1;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  ASSERT_GE(effective_datarate_[0], cfg_.rc_target_bitrate * 0.85)
+      << " The datarate for the file is lower than target by too much!";
+  ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.15)
+      << " The datarate for the file is greater than target by too much!";
+}
+
+// Check basic datarate targeting, for a single bitrate, when denoiser is on,
+// for 1280x720 clip with 4 threads.
+TEST_P(DatarateTestVP9LargeDenoiser, 4threads) {
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_buf_optimal_sz = 500;
+  cfg_.rc_buf_sz = 1000;
+  cfg_.rc_dropframe_thresh = 1;
+  cfg_.rc_min_quantizer = 2;
+  cfg_.rc_max_quantizer = 56;
+  cfg_.rc_end_usage = VPX_CBR;
+  cfg_.g_lag_in_frames = 0;
+  cfg_.g_threads = 4;
+
+  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300);
+
  // For the temporal denoiser (#if CONFIG_VP9_TEMPORAL_DENOISING),
  // there is only one denoiser mode: denoiserYonly(which is 1),
  // but may add more modes in the future.
@@ -1011,7 +1161,7 @@ TEST_P(DatarateTestVP9LargeDenoiser, HighNoise) {
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
  ASSERT_GE(effective_datarate_[0], cfg_.rc_target_bitrate * 0.85)
      << " The datarate for the file is lower than target by too much!";
-  ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.15)
+  ASSERT_LE(effective_datarate_[0], cfg_.rc_target_bitrate * 1.29)
      << " The datarate for the file is greater than target by too much!";
 }

@@ -1066,13 +1216,17 @@ class DatarateOnePassCbrSvc
  }
  virtual void ResetModel() {
    last_pts_ = 0;
-    bits_in_buffer_model_ = cfg_.rc_target_bitrate * cfg_.rc_buf_initial_sz;
-    frame_number_ = 0;
-    first_drop_ = 0;
-    bits_total_ = 0;
    duration_ = 0.0;
    mismatch_psnr_ = 0.0;
    mismatch_nframes_ = 0;
+    denoiser_on_ = 0;
+    tune_content_ = 0;
+    base_speed_setting_ = 5;
+    spatial_layer_id_ = 0;
+    temporal_layer_id_ = 0;
+    memset(bits_in_buffer_model_, 0, sizeof(bits_in_buffer_model_));
+    memset(bits_total_, 0, sizeof(bits_total_));
+    memset(layer_target_avg_bandwidth_, 0, sizeof(layer_target_avg_bandwidth_));
  }
  virtual void BeginPassHook(unsigned int /*pass*/) {}
  virtual void PreEncodeFrameHook(::libvpx_test::VideoSource *video,
@@ -1083,48 +1237,114 @@ class DatarateOnePassCbrSvc
        svc_params_.max_quantizers[i] = 63;
        svc_params_.min_quantizers[i] = 0;
      }
-      svc_params_.speed_per_layer[0] = 5;
+      svc_params_.speed_per_layer[0] = base_speed_setting_;
      for (i = 1; i < VPX_SS_MAX_LAYERS; ++i) {
        svc_params_.speed_per_layer[i] = speed_setting_;
      }
+
+      encoder->Control(VP9E_SET_NOISE_SENSITIVITY, denoiser_on_);
      encoder->Control(VP9E_SET_SVC, 1);
      encoder->Control(VP9E_SET_SVC_PARAMETERS, &svc_params_);
      encoder->Control(VP8E_SET_CPUUSED, speed_setting_);
      encoder->Control(VP9E_SET_TILE_COLUMNS, 0);
      encoder->Control(VP8E_SET_MAX_INTRA_BITRATE_PCT, 300);
      encoder->Control(VP9E_SET_TILE_COLUMNS, (cfg_.g_threads >> 1));
+      encoder->Control(VP9E_SET_ROW_MT, 1);
      encoder->Control(VP8E_SET_STATIC_THRESHOLD, 1);
+      encoder->Control(VP9E_SET_TUNE_CONTENT, tune_content_);
    }
    const vpx_rational_t tb = video->timebase();
    timebase_ = static_cast<double>(tb.num) / tb.den;
    duration_ = 0;
  }
+
+  virtual void PostEncodeFrameHook(::libvpx_test::Encoder *encoder) {
+    vpx_svc_layer_id_t layer_id;
+    encoder->Control(VP9E_GET_SVC_LAYER_ID, &layer_id);
+    spatial_layer_id_ = layer_id.spatial_layer_id;
+    temporal_layer_id_ = layer_id.temporal_layer_id;
+    // Update buffer with per-layer target frame bandwidth, this is done
+    // for every frame passed to the encoder (encoded or dropped).
+    // For temporal layers, update the cumulative buffer level.
+    for (int sl = 0; sl < number_spatial_layers_; ++sl) {
+      for (int tl = temporal_layer_id_; tl < number_temporal_layers_; ++tl) {
+        const int layer = sl * number_temporal_layers_ + tl;
+        bits_in_buffer_model_[layer] +=
+            static_cast<int64_t>(layer_target_avg_bandwidth_[layer]);
+      }
+    }
+  }
+
+  vpx_codec_err_t parse_superframe_index(const uint8_t *data, size_t data_sz,
+                                         uint32_t sizes[8], int *count) {
+    uint8_t marker;
+    marker = *(data + data_sz - 1);
+    *count = 0;
+    if ((marker & 0xe0) == 0xc0) {
+      const uint32_t frames = (marker & 0x7) + 1;
+      const uint32_t mag = ((marker >> 3) & 0x3) + 1;
+      const size_t index_sz = 2 + mag * frames;
+      // This chunk is marked as having a superframe index but doesn't have
+      // enough data for it, thus it's an invalid superframe index.
+      if (data_sz < index_sz) return VPX_CODEC_CORRUPT_FRAME;
+      {
+        const uint8_t marker2 = *(data + data_sz - index_sz);
+        // This chunk is marked as having a superframe index but doesn't have
+        // the matching marker byte at the front of the index therefore it's an
+        // invalid chunk.
+        if (marker != marker2) return VPX_CODEC_CORRUPT_FRAME;
+      }
+      {
+        uint32_t i, j;
+        const uint8_t *x = &data[data_sz - index_sz + 1];
+        for (i = 0; i < frames; ++i) {
+          uint32_t this_sz = 0;
+
+          for (j = 0; j < mag; ++j) this_sz |= (*x++) << (j * 8);
+          sizes[i] = this_sz;
+        }
+        *count = frames;
+      }
+    }
+    return VPX_CODEC_OK;
+  }
+
  virtual void FramePktHook(const vpx_codec_cx_pkt_t *pkt) {
-    vpx_codec_pts_t duration = pkt->data.frame.pts - last_pts_;
-    if (last_pts_ == 0) duration = 1;
-    bits_in_buffer_model_ += static_cast<int64_t>(
-        duration * timebase_ * cfg_.rc_target_bitrate * 1000);
+    uint32_t sizes[8] = { 0 };
+    int count = 0;
+    last_pts_ = pkt->data.frame.pts;
    const bool key_frame =
        (pkt->data.frame.flags & VPX_FRAME_IS_KEY) ? true : false;
-    if (!key_frame) {
-      // TODO(marpan): This check currently fails for some of the SVC tests,
-      // re-enable when issue (webm:1350) is resolved.
-      //  ASSERT_GE(bits_in_buffer_model_, 0) << "Buffer Underrun at frame "
-      //                                      << pkt->data.frame.pts;
+    parse_superframe_index(static_cast<const uint8_t *>(pkt->data.frame.buf),
+                           pkt->data.frame.sz, sizes, &count);
+    ASSERT_EQ(count, number_spatial_layers_);
+    for (int sl = 0; sl < number_spatial_layers_; ++sl) {
+      sizes[sl] = sizes[sl] << 3;
+      // Update the total encoded bits per layer.
+      // For temporal layers, update the cumulative encoded bits per layer.
+      for (int tl = temporal_layer_id_; tl < number_temporal_layers_; ++tl) {
+        const int layer = sl * number_temporal_layers_ + tl;
+        bits_total_[layer] += static_cast<int64_t>(sizes[sl]);
+        // Update the per-layer buffer level with the encoded frame size.
+        bits_in_buffer_model_[layer] -= static_cast<int64_t>(sizes[sl]);
+        // There should be no buffer underrun, except on the base
+        // temporal layer, since there may be key frames there.
+        if (!key_frame && tl > 0) {
+          ASSERT_GE(bits_in_buffer_model_[layer], 0)
+              << "Buffer Underrun at frame " << pkt->data.frame.pts;
+        }
+      }
    }
-    const size_t frame_size_in_bits = pkt->data.frame.sz * 8;
-    bits_in_buffer_model_ -= static_cast<int64_t>(frame_size_in_bits);
-    bits_total_ += frame_size_in_bits;
-    if (!first_drop_ && duration > 1) first_drop_ = last_pts_ + 1;
-    last_pts_ = pkt->data.frame.pts;
-    bits_in_last_frame_ = frame_size_in_bits;
-    ++frame_number_;
  }
+
  virtual void EndPassHook(void) {
-    if (bits_total_) {
-      const double file_size_in_kb = bits_total_ / 1000.;  // bits per kilobit
-      duration_ = (last_pts_ + 1) * timebase_;
-      file_datarate_ = file_size_in_kb / duration_;
+    for (int sl = 0; sl < number_spatial_layers_; ++sl) {
+      for (int tl = 0; tl < number_temporal_layers_; ++tl) {
+        const int layer = sl * number_temporal_layers_ + tl;
+        const double file_size_in_kb = bits_total_[layer] / 1000.;
+        duration_ = (last_pts_ + 1) * timebase_;
+        file_datarate_[layer] = file_size_in_kb / duration_;
+      }
    }
  }

@@ -1137,26 +1357,35 @@ class DatarateOnePassCbrSvc
  unsigned int GetMismatchFrames() { return mismatch_nframes_; }

  vpx_codec_pts_t last_pts_;
-  int64_t bits_in_buffer_model_;
+  int64_t bits_in_buffer_model_[VPX_MAX_LAYERS];
  double timebase_;
-  int frame_number_;
-  vpx_codec_pts_t first_drop_;
-  int64_t bits_total_;
+  int64_t bits_total_[VPX_MAX_LAYERS];
  double duration_;
-  double file_datarate_;
+  double file_datarate_[VPX_MAX_LAYERS];
  size_t bits_in_last_frame_;
  vpx_svc_extra_cfg_t svc_params_;
  int speed_setting_;
  double mismatch_psnr_;
  int mismatch_nframes_;
+  int denoiser_on_;
+  int tune_content_;
+  int base_speed_setting_;
+  int spatial_layer_id_;
+  int temporal_layer_id_;
+  int number_spatial_layers_;
+  int number_temporal_layers_;
+  int layer_target_avg_bandwidth_[VPX_MAX_LAYERS];
 };
 static void assign_layer_bitrates(vpx_codec_enc_cfg_t *const enc_cfg,
                                  const vpx_svc_extra_cfg_t *svc_params,
                                  int spatial_layers, int temporal_layers,
-                                  int temporal_layering_mode) {
+                                  int temporal_layering_mode,
+                                  int *layer_target_avg_bandwidth,
+                                  int64_t *bits_in_buffer_model) {
  int sl, spatial_layer_target;
  float total = 0;
  float alloc_ratio[VPX_MAX_LAYERS] = { 0 };
+  float framerate = 30.0;
  for (sl = 0; sl < spatial_layers; ++sl) {
    if (svc_params->scaling_factor_den[sl] > 0) {
      alloc_ratio[sl] = (float)(svc_params->scaling_factor_num[sl] * 1.0 /
@@ -1176,13 +1405,85 @@ static void assign_layer_bitrates(vpx_codec_enc_cfg_t *const enc_cfg,
    } else if (temporal_layering_mode == 2) {
      enc_cfg->layer_target_bitrate[index] = spatial_layer_target * 2 / 3;
      enc_cfg->layer_target_bitrate[index + 1] = spatial_layer_target;
+    } else if (temporal_layering_mode <= 1) {
+      enc_cfg->layer_target_bitrate[index] = spatial_layer_target;
+    }
+  }
+  for (sl = 0; sl < spatial_layers; ++sl) {
+    for (int tl = 0; tl < temporal_layers; ++tl) {
+      const int layer = sl * temporal_layers + tl;
+      float layer_framerate = framerate;
+      if (temporal_layers == 2 && tl == 0) layer_framerate = framerate / 2;
+      if (temporal_layers == 3 && tl == 0) layer_framerate = framerate / 4;
+      if (temporal_layers == 3 && tl == 1) layer_framerate = framerate / 2;
+      layer_target_avg_bandwidth[layer] = static_cast<int>(
+          enc_cfg->layer_target_bitrate[layer] * 1000.0 / layer_framerate);
+      bits_in_buffer_model[layer] =
+          enc_cfg->layer_target_bitrate[layer] * enc_cfg->rc_buf_initial_sz;
    }
  }
 }

+static void CheckLayerRateTargeting(vpx_codec_enc_cfg_t *const cfg,
+                                    int number_spatial_layers,
+                                    int number_temporal_layers,
+                                    double *file_datarate,
+                                    double thresh_overshoot,
+                                    double thresh_undershoot) {
+  for (int sl = 0; sl < number_spatial_layers; ++sl)
+    for (int tl = 0; tl < number_temporal_layers; ++tl) {
+      const int layer = sl * number_temporal_layers + tl;
+      ASSERT_GE(cfg->layer_target_bitrate[layer],
+                file_datarate[layer] * thresh_overshoot)
+          << " The datarate for the file exceeds the target by too much!";
+      ASSERT_LE(cfg->layer_target_bitrate[layer],
+                file_datarate[layer] * thresh_undershoot)
+          << " The datarate for the file is lower than the target by too much!";
+    }
+}
+
+// Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and 1
+// temporal layer, with screen content mode on and same speed setting for all
+// layers.
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL1TLScreenContent1) {
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_buf_optimal_sz = 500;
+  cfg_.rc_buf_sz = 1000;
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 63;
+  cfg_.rc_end_usage = VPX_CBR;
+  cfg_.g_lag_in_frames = 0;
+  cfg_.ss_number_layers = 2;
+  cfg_.ts_number_layers = 1;
+  cfg_.ts_rate_decimator[0] = 1;
+  cfg_.g_error_resilient = 1;
+  cfg_.g_threads = 1;
+  cfg_.temporal_layering_mode = 0;
+  svc_params_.scaling_factor_num[0] = 144;
+  svc_params_.scaling_factor_den[0] = 288;
+  svc_params_.scaling_factor_num[1] = 288;
+  svc_params_.scaling_factor_den[1] = 288;
+  cfg_.rc_dropframe_thresh = 10;
+  cfg_.kf_max_dist = 9999;
+  number_spatial_layers_ = cfg_.ss_number_layers;
+  number_temporal_layers_ = cfg_.ts_number_layers;
+  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 60);
+  cfg_.rc_target_bitrate = 500;
+  ResetModel();
+  tune_content_ = 1;
+  base_speed_setting_ = speed_setting_;
+  assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
+                        cfg_.ts_number_layers, cfg_.temporal_layering_mode,
+                        layer_target_avg_bandwidth_, bits_in_buffer_model_);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  CheckLayerRateTargeting(&cfg_, number_spatial_layers_,
+                          number_temporal_layers_, file_datarate_, 0.78, 1.15);
+  EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+}
+
 // Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and
 // 3 temporal layers. Run CIF clip with 1 thread.
-TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers) {
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL3TL) {
  cfg_.rc_buf_initial_sz = 500;
  cfg_.rc_buf_optimal_sz = 500;
  cfg_.rc_buf_sz = 1000;
@@ -1202,29 +1503,93 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers) {
  svc_params_.scaling_factor_den[0] = 288;
  svc_params_.scaling_factor_num[1] = 288;
  svc_params_.scaling_factor_den[1] = 288;
-  cfg_.rc_dropframe_thresh = 10;
+  cfg_.rc_dropframe_thresh = 0;
  cfg_.kf_max_dist = 9999;
-  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
-                                       30, 1, 0, 200);
+  number_spatial_layers_ = cfg_.ss_number_layers;
+  number_temporal_layers_ = cfg_.ts_number_layers;
+  ::libvpx_test::I420VideoSource video("niklas_640_480_30.yuv", 640, 480, 30, 1,
+                                       0, 400);
  // TODO(marpan): Check that effective_datarate for each layer hits the
  // layer target_bitrate.
  for (int i = 200; i <= 800; i += 200) {
    cfg_.rc_target_bitrate = i;
    ResetModel();
    assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
-                          cfg_.ts_number_layers, cfg_.temporal_layering_mode);
+                          cfg_.ts_number_layers, cfg_.temporal_layering_mode,
+                          layer_target_avg_bandwidth_, bits_in_buffer_model_);
    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-    ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85)
-        << " The datarate for the file exceeds the target by too much!";
-    ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.15)
-        << " The datarate for the file is lower than the target by too much!";
-    EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+    CheckLayerRateTargeting(&cfg_, number_spatial_layers_,
+                            number_temporal_layers_, file_datarate_, 0.78,
+                            1.15);
+#if CONFIG_VP9_DECODER
+    // Number of temporal layers > 1, so half of the frames in this SVC pattern
+    // will be non-reference frame and hence encoder will avoid loopfilter.
+    // Since frame dropper is off, we can expect 200 (half of the sequence)
+    // mismatched frames.
+    EXPECT_EQ(static_cast<unsigned int>(200), GetMismatchFrames());
+#endif
+  }
+}
+
+// Check basic rate targeting for 1 pass CBR SVC with denoising.
+// 2 spatial layers and 3 temporal layer. Run HD clip with 2 threads.
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL3TLDenoiserOn) {
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_buf_optimal_sz = 500;
+  cfg_.rc_buf_sz = 1000;
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 63;
+  cfg_.rc_end_usage = VPX_CBR;
+  cfg_.g_lag_in_frames = 0;
+  cfg_.ss_number_layers = 2;
+  cfg_.ts_number_layers = 3;
+  cfg_.ts_rate_decimator[0] = 4;
+  cfg_.ts_rate_decimator[1] = 2;
+  cfg_.ts_rate_decimator[2] = 1;
+  cfg_.g_error_resilient = 1;
+  cfg_.g_threads = 2;
+  cfg_.temporal_layering_mode = 3;
+  svc_params_.scaling_factor_num[0] = 144;
+  svc_params_.scaling_factor_den[0] = 288;
+  svc_params_.scaling_factor_num[1] = 288;
+  svc_params_.scaling_factor_den[1] = 288;
+  cfg_.rc_dropframe_thresh = 0;
+  cfg_.kf_max_dist = 9999;
+  number_spatial_layers_ = cfg_.ss_number_layers;
+  number_temporal_layers_ = cfg_.ts_number_layers;
+  ::libvpx_test::I420VideoSource video("niklas_640_480_30.yuv", 640, 480, 30, 1,
+                                       0, 400);
+  // TODO(marpan): Check that effective_datarate for each layer hits the
+  // layer target_bitrate.
+  // For SVC, noise_sen = 1 means denoising only the top spatial layer
+  // noise_sen = 2 means denoising the two top spatial layers.
+  for (int noise_sen = 1; noise_sen <= 2; noise_sen++) {
+    for (int i = 600; i <= 1000; i += 200) {
+      cfg_.rc_target_bitrate = i;
+      ResetModel();
+      denoiser_on_ = noise_sen;
+      assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
+                            cfg_.ts_number_layers, cfg_.temporal_layering_mode,
+                            layer_target_avg_bandwidth_, bits_in_buffer_model_);
+      ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+      CheckLayerRateTargeting(&cfg_, number_spatial_layers_,
+                              number_temporal_layers_, file_datarate_, 0.78,
+                              1.15);
+#if CONFIG_VP9_DECODER
+      // Number of temporal layers > 1, so half of the frames in this SVC
+      // pattern
+      // will be non-reference frame and hence encoder will avoid loopfilter.
+      // Since frame dropper is off, we can expect 200 (half of the sequence)
+      // mismatched frames.
+      EXPECT_EQ(static_cast<unsigned int>(200), GetMismatchFrames());
+#endif
+    }
  }
 }

 // Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and 3
 // temporal layers. Run CIF clip with 1 thread, and few short key frame periods.
-TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayersSmallKf) {
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL3TLSmallKf) {
  cfg_.rc_buf_initial_sz = 500;
  cfg_.rc_buf_optimal_sz = 500;
  cfg_.rc_buf_sz = 1000;
@@ -1245,28 +1610,29 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayersSmallKf) {
  svc_params_.scaling_factor_num[1] = 288;
  svc_params_.scaling_factor_den[1] = 288;
  cfg_.rc_dropframe_thresh = 10;
-  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
-                                       30, 1, 0, 200);
  cfg_.rc_target_bitrate = 400;
+  number_spatial_layers_ = cfg_.ss_number_layers;
+  number_temporal_layers_ = cfg_.ts_number_layers;
+  ::libvpx_test::I420VideoSource video("niklas_640_480_30.yuv", 640, 480, 30, 1,
+                                       0, 400);
  // For this 3 temporal layer case, pattern repeats every 4 frames, so choose
  // 4 key neighboring key frame periods (so key frame will land on 0-2-1-2).
  for (int j = 64; j <= 67; j++) {
    cfg_.kf_max_dist = j;
    ResetModel();
    assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
-                          cfg_.ts_number_layers, cfg_.temporal_layering_mode);
+                          cfg_.ts_number_layers, cfg_.temporal_layering_mode,
+                          layer_target_avg_bandwidth_, bits_in_buffer_model_);
    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-    ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85)
-        << " The datarate for the file exceeds the target by too much!";
-    ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.15)
-        << " The datarate for the file is lower than the target by too much!";
-    EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+    CheckLayerRateTargeting(&cfg_, number_spatial_layers_,
+                            number_temporal_layers_, file_datarate_, 0.78,
+                            1.15);
  }
 }

 // Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and
 // 3 temporal layers. Run HD clip with 4 threads.
-TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers4threads) {
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL3TL4Threads) {
  cfg_.rc_buf_initial_sz = 500;
  cfg_.rc_buf_optimal_sz = 500;
  cfg_.rc_buf_sz = 1000;
@@ -1286,24 +1652,31 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers4threads) {
  svc_params_.scaling_factor_den[0] = 288;
  svc_params_.scaling_factor_num[1] = 288;
  svc_params_.scaling_factor_den[1] = 288;
-  cfg_.rc_dropframe_thresh = 10;
+  cfg_.rc_dropframe_thresh = 0;
  cfg_.kf_max_dist = 9999;
-  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300);
+  number_spatial_layers_ = cfg_.ss_number_layers;
+  number_temporal_layers_ = cfg_.ts_number_layers;
+  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 60);
  cfg_.rc_target_bitrate = 800;
  ResetModel();
  assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
-                        cfg_.ts_number_layers, cfg_.temporal_layering_mode);
+                        cfg_.ts_number_layers, cfg_.temporal_layering_mode,
+                        layer_target_avg_bandwidth_, bits_in_buffer_model_);
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85)
-      << " The datarate for the file exceeds the target by too much!";
-  ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.15)
-      << " The datarate for the file is lower than the target by too much!";
-  EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+  CheckLayerRateTargeting(&cfg_, number_spatial_layers_,
+                          number_temporal_layers_, file_datarate_, 0.78, 1.15);
+#if CONFIG_VP9_DECODER
+  // Number of temporal layers > 1, so half of the frames in this SVC pattern
+  // will be non-reference frame and hence encoder will avoid loopfilter.
+  // Since frame dropper is off, we can expect 30 (half of the sequence)
+  // mismatched frames.
+  EXPECT_EQ(static_cast<unsigned int>(30), GetMismatchFrames());
+#endif
 }

 // Check basic rate targeting for 1 pass CBR SVC: 3 spatial layers and
 // 3 temporal layers. Run CIF clip with 1 thread.
-TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers) {
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SL3TL) {
  cfg_.rc_buf_initial_sz = 500;
  cfg_.rc_buf_optimal_sz = 500;
  cfg_.rc_buf_sz = 1000;
@@ -1325,24 +1698,32 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers) {
  svc_params_.scaling_factor_den[1] = 288;
  svc_params_.scaling_factor_num[2] = 288;
  svc_params_.scaling_factor_den[2] = 288;
-  cfg_.rc_dropframe_thresh = 10;
+  cfg_.rc_dropframe_thresh = 0;
  cfg_.kf_max_dist = 9999;
-  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300);
+  number_spatial_layers_ = cfg_.ss_number_layers;
+  number_temporal_layers_ = cfg_.ts_number_layers;
+  ::libvpx_test::I420VideoSource video("niklas_640_480_30.yuv", 640, 480, 30, 1,
+                                       0, 400);
  cfg_.rc_target_bitrate = 800;
  ResetModel();
  assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
-                        cfg_.ts_number_layers, cfg_.temporal_layering_mode);
+                        cfg_.ts_number_layers, cfg_.temporal_layering_mode,
+                        layer_target_avg_bandwidth_, bits_in_buffer_model_);
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85)
-      << " The datarate for the file exceeds the target by too much!";
-  ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.22)
-      << " The datarate for the file is lower than the target by too much!";
-  EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+  CheckLayerRateTargeting(&cfg_, number_spatial_layers_,
+                          number_temporal_layers_, file_datarate_, 0.78, 1.15);
+#if CONFIG_VP9_DECODER
+  // Number of temporal layers > 1, so half of the frames in this SVC pattern
+  // will be non-reference frame and hence encoder will avoid loopfilter.
+  // Since frame dropper is off, we can expect 200 (half of the sequence)
+  // mismatched frames.
+  EXPECT_EQ(static_cast<unsigned int>(200), GetMismatchFrames());
+#endif
 }

 // Check basic rate targeting for 1 pass CBR SVC: 3 spatial layers and 3
 // temporal layers. Run CIF clip with 1 thread, and few short key frame periods.
-TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayersSmallKf) {
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SL3TLSmallKf) {
  cfg_.rc_buf_initial_sz = 500;
  cfg_.rc_buf_optimal_sz = 500;
  cfg_.rc_buf_sz = 1000;
@@ -1365,27 +1746,29 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayersSmallKf) {
  svc_params_.scaling_factor_num[2] = 288;
  svc_params_.scaling_factor_den[2] = 288;
  cfg_.rc_dropframe_thresh = 10;
-  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300);
  cfg_.rc_target_bitrate = 800;
+  number_spatial_layers_ = cfg_.ss_number_layers;
+  number_temporal_layers_ = cfg_.ts_number_layers;
+  ::libvpx_test::I420VideoSource video("niklas_640_480_30.yuv", 640, 480, 30, 1,
+                                       0, 400);
  // For this 3 temporal layer case, pattern repeats every 4 frames, so choose
  // 4 key neighboring key frame periods (so key frame will land on 0-2-1-2).
  for (int j = 32; j <= 35; j++) {
    cfg_.kf_max_dist = j;
    ResetModel();
    assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
-                          cfg_.ts_number_layers, cfg_.temporal_layering_mode);
+                          cfg_.ts_number_layers, cfg_.temporal_layering_mode,
+                          layer_target_avg_bandwidth_, bits_in_buffer_model_);
    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-    ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85)
-        << " The datarate for the file exceeds the target by too much!";
-    ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.30)
-        << " The datarate for the file is lower than the target by too much!";
-    EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+    CheckLayerRateTargeting(&cfg_, number_spatial_layers_,
+                            number_temporal_layers_, file_datarate_, 0.78,
+                            1.15);
  }
 }

 // Check basic rate targeting for 1 pass CBR SVC: 3 spatial layers and
 // 3 temporal layers. Run HD clip with 4 threads.
-TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers4threads) {
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SL3TL4threads) {
  cfg_.rc_buf_initial_sz = 500;
  cfg_.rc_buf_optimal_sz = 500;
  cfg_.rc_buf_sz = 1000;
@@ -1407,24 +1790,31 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SpatialLayers4threads) {
  svc_params_.scaling_factor_den[1] = 288;
  svc_params_.scaling_factor_num[2] = 288;
  svc_params_.scaling_factor_den[2] = 288;
-  cfg_.rc_dropframe_thresh = 10;
+  cfg_.rc_dropframe_thresh = 0;
  cfg_.kf_max_dist = 9999;
-  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300);
+  number_spatial_layers_ = cfg_.ss_number_layers;
+  number_temporal_layers_ = cfg_.ts_number_layers;
+  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 60);
  cfg_.rc_target_bitrate = 800;
  ResetModel();
  assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
-                        cfg_.ts_number_layers, cfg_.temporal_layering_mode);
+                        cfg_.ts_number_layers, cfg_.temporal_layering_mode,
+                        layer_target_avg_bandwidth_, bits_in_buffer_model_);
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  ASSERT_GE(cfg_.rc_target_bitrate, file_datarate_ * 0.85)
-      << " The datarate for the file exceeds the target by too much!";
-  ASSERT_LE(cfg_.rc_target_bitrate, file_datarate_ * 1.22)
-      << " The datarate for the file is lower than the target by too much!";
-  EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+  CheckLayerRateTargeting(&cfg_, number_spatial_layers_,
+                          number_temporal_layers_, file_datarate_, 0.78, 1.15);
+#if CONFIG_VP9_DECODER
+  // Number of temporal layers > 1, so half of the frames in this SVC pattern
+  // will be non-reference frame and hence encoder will avoid loopfilter.
+  // Since frame dropper is off, we can expect 30 (half of the sequence)
+  // mismatched frames.
+  EXPECT_EQ(static_cast<unsigned int>(30), GetMismatchFrames());
+#endif
 }

 // Run SVC encoder for 1 temporal layer, 2 spatial layers, with spatial
 // downscale 5x5.
-TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers5x5MultipleRuns) {
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SL1TL5x5MultipleRuns) {
  cfg_.rc_buf_initial_sz = 500;
  cfg_.rc_buf_optimal_sz = 500;
  cfg_.rc_buf_sz = 1000;
@@ -1442,7 +1832,7 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers5x5MultipleRuns) {
  svc_params_.scaling_factor_den[0] = 1280;
  svc_params_.scaling_factor_num[1] = 1280;
  svc_params_.scaling_factor_den[1] = 1280;
-  cfg_.rc_dropframe_thresh = 0;
+  cfg_.rc_dropframe_thresh = 10;
  cfg_.kf_max_dist = 999999;
  cfg_.kf_min_dist = 0;
  cfg_.ss_target_bitrate[0] = 300;
@@ -1450,9 +1840,19 @@ TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc2SpatialLayers5x5MultipleRuns) {
  cfg_.layer_target_bitrate[0] = 300;
  cfg_.layer_target_bitrate[1] = 1400;
  cfg_.rc_target_bitrate = 1700;
-  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 300);
+  number_spatial_layers_ = cfg_.ss_number_layers;
+  number_temporal_layers_ = cfg_.ts_number_layers;
  ResetModel();
+  layer_target_avg_bandwidth_[0] = cfg_.layer_target_bitrate[0] * 1000 / 30;
+  bits_in_buffer_model_[0] =
+      cfg_.layer_target_bitrate[0] * cfg_.rc_buf_initial_sz;
+  layer_target_avg_bandwidth_[1] = cfg_.layer_target_bitrate[1] * 1000 / 30;
+  bits_in_buffer_model_[1] =
+      cfg_.layer_target_bitrate[1] * cfg_.rc_buf_initial_sz;
+  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 60);
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  CheckLayerRateTargeting(&cfg_, number_spatial_layers_,
+                          number_temporal_layers_, file_datarate_, 0.78, 1.15);
  EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
 }

--- a/test/dct16x16_test.cc
+++ b/test/dct16x16_test.cc
@@ -255,11 +255,11 @@ void iht16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,

 #if CONFIG_VP9_HIGHBITDEPTH
 void idct16x16_10(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct16x16_256_add_c(in, out, stride, 10);
+  vpx_highbd_idct16x16_256_add_c(in, CAST_TO_SHORTPTR(out), stride, 10);
 }

 void idct16x16_12(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct16x16_256_add_c(in, out, stride, 12);
+  vpx_highbd_idct16x16_256_add_c(in, CAST_TO_SHORTPTR(out), stride, 12);
 }

 void idct16x16_10_ref(const tran_low_t *in, uint8_t *out, int stride,
@@ -273,36 +273,36 @@ void idct16x16_12_ref(const tran_low_t *in, uint8_t *out, int stride,
 }

 void iht16x16_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
-  vp9_highbd_iht16x16_256_add_c(in, out, stride, tx_type, 10);
+  vp9_highbd_iht16x16_256_add_c(in, CAST_TO_SHORTPTR(out), stride, tx_type, 10);
 }

 void iht16x16_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
-  vp9_highbd_iht16x16_256_add_c(in, out, stride, tx_type, 12);
+  vp9_highbd_iht16x16_256_add_c(in, CAST_TO_SHORTPTR(out), stride, tx_type, 12);
 }

 #if HAVE_SSE2
 void idct16x16_10_add_10_c(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct16x16_10_add_c(in, out, stride, 10);
+  vpx_highbd_idct16x16_10_add_c(in, CAST_TO_SHORTPTR(out), stride, 10);
 }

 void idct16x16_10_add_12_c(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct16x16_10_add_c(in, out, stride, 12);
+  vpx_highbd_idct16x16_10_add_c(in, CAST_TO_SHORTPTR(out), stride, 12);
 }

 void idct16x16_256_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct16x16_256_add_sse2(in, out, stride, 10);
+  vpx_highbd_idct16x16_256_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 10);
 }

 void idct16x16_256_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct16x16_256_add_sse2(in, out, stride, 12);
+  vpx_highbd_idct16x16_256_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 12);
 }

 void idct16x16_10_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct16x16_10_add_sse2(in, out, stride, 10);
+  vpx_highbd_idct16x16_10_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 10);
 }

 void idct16x16_10_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct16x16_10_add_sse2(in, out, stride, 12);
+  vpx_highbd_idct16x16_10_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 12);
 }
 #endif  // HAVE_SSE2
 #endif  // CONFIG_VP9_HIGHBITDEPTH
@@ -353,7 +353,7 @@ class Trans16x16TestBase {
 #if CONFIG_VP9_HIGHBITDEPTH
      } else {
        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
+            RunInvTxfm(test_temp_block, CAST_TO_BYTEPTR(dst16), pitch_));
 #endif
      }

@@ -475,10 +475,10 @@ class Trans16x16TestBase {
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
 #if CONFIG_VP9_HIGHBITDEPTH
      } else {
-        inv_txfm_ref(output_ref_block, CONVERT_TO_BYTEPTR(ref16), pitch_,
+        inv_txfm_ref(output_ref_block, CAST_TO_BYTEPTR(ref16), pitch_,
                     tx_type_);
        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(output_ref_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
+            RunInvTxfm(output_ref_block, CAST_TO_BYTEPTR(dst16), pitch_));
 #endif
      }
      if (bit_depth_ == VPX_BITS_8) {
@@ -530,8 +530,7 @@ class Trans16x16TestBase {
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
 #if CONFIG_VP9_HIGHBITDEPTH
      } else {
-        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16), 16));
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CAST_TO_BYTEPTR(dst16), 16));
 #endif  // CONFIG_VP9_HIGHBITDEPTH
      }

@@ -543,8 +542,8 @@ class Trans16x16TestBase {
        const uint32_t diff = dst[j] - src[j];
 #endif  // CONFIG_VP9_HIGHBITDEPTH
        const uint32_t error = diff * diff;
-        EXPECT_GE(1u, error) << "Error: 16x16 IDCT has error " << error
-                             << " at index " << j;
+        EXPECT_GE(1u, error)
+            << "Error: 16x16 IDCT has error " << error << " at index " << j;
      }
    }
  }
@@ -585,9 +584,9 @@ class Trans16x16TestBase {
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
      } else {
 #if CONFIG_VP9_HIGHBITDEPTH
-        ref_txfm(coeff, CONVERT_TO_BYTEPTR(ref16), pitch_);
+        ref_txfm(coeff, CAST_TO_BYTEPTR(ref16), pitch_);
        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16), pitch_));
+            RunInvTxfm(coeff, CAST_TO_BYTEPTR(dst16), pitch_));
 #endif  // CONFIG_VP9_HIGHBITDEPTH
      }

@@ -745,66 +744,6 @@ TEST_P(InvTrans16x16DCT, CompareReference) {
  CompareInvReference(ref_txfm_, thresh_);
 }

-class PartialTrans16x16Test : public ::testing::TestWithParam<
-                                  std::tr1::tuple<FdctFunc, vpx_bit_depth_t> > {
- public:
-  virtual ~PartialTrans16x16Test() {}
-  virtual void SetUp() {
-    fwd_txfm_ = GET_PARAM(0);
-    bit_depth_ = GET_PARAM(1);
-  }
-
-  virtual void TearDown() { libvpx_test::ClearSystemState(); }
-
- protected:
-  vpx_bit_depth_t bit_depth_;
-  FdctFunc fwd_txfm_;
-};
-
-TEST_P(PartialTrans16x16Test, Extremes) {
-#if CONFIG_VP9_HIGHBITDEPTH
-  const int16_t maxval =
-      static_cast<int16_t>(clip_pixel_highbd(1 << 30, bit_depth_));
-#else
-  const int16_t maxval = 255;
-#endif
-  const int minval = -maxval;
-  DECLARE_ALIGNED(16, int16_t, input[kNumCoeffs]);
-  DECLARE_ALIGNED(16, tran_low_t, output[kNumCoeffs]);
-
-  for (int i = 0; i < kNumCoeffs; ++i) input[i] = maxval;
-  output[0] = 0;
-  ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 16));
-  EXPECT_EQ((maxval * kNumCoeffs) >> 1, output[0]);
-
-  for (int i = 0; i < kNumCoeffs; ++i) input[i] = minval;
-  output[0] = 0;
-  ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 16));
-  EXPECT_EQ((minval * kNumCoeffs) >> 1, output[0]);
-}
-
-TEST_P(PartialTrans16x16Test, Random) {
-#if CONFIG_VP9_HIGHBITDEPTH
-  const int16_t maxval =
-      static_cast<int16_t>(clip_pixel_highbd(1 << 30, bit_depth_));
-#else
-  const int16_t maxval = 255;
-#endif
-  DECLARE_ALIGNED(16, int16_t, input[kNumCoeffs]);
-  DECLARE_ALIGNED(16, tran_low_t, output[kNumCoeffs]);
-  ACMRandom rnd(ACMRandom::DeterministicSeed());
-
-  int sum = 0;
-  for (int i = 0; i < kNumCoeffs; ++i) {
-    const int val = (i & 1) ? -rnd(maxval + 1) : rnd(maxval + 1);
-    input[i] = val;
-    sum += val;
-  }
-  output[0] = 0;
-  ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 16));
-  EXPECT_EQ(sum >> 1, output[0]);
-}
-
 using std::tr1::make_tuple;

 #if CONFIG_VP9_HIGHBITDEPTH
@@ -837,11 +776,6 @@ INSTANTIATE_TEST_CASE_P(
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(
-    C, PartialTrans16x16Test,
-    ::testing::Values(make_tuple(&vpx_highbd_fdct16x16_1_c, VPX_BITS_8),
-                      make_tuple(&vpx_highbd_fdct16x16_1_c, VPX_BITS_10),
-                      make_tuple(&vpx_highbd_fdct16x16_1_c, VPX_BITS_12)));
 #else
 INSTANTIATE_TEST_CASE_P(
    C, Trans16x16HT,
@@ -850,17 +784,14 @@ INSTANTIATE_TEST_CASE_P(
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(C, PartialTrans16x16Test,
-                        ::testing::Values(make_tuple(&vpx_fdct16x16_1_c,
-                                                     VPX_BITS_8)));
 #endif  // CONFIG_VP9_HIGHBITDEPTH

-#if HAVE_NEON && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+#if HAVE_NEON && !CONFIG_EMULATE_HARDWARE
 INSTANTIATE_TEST_CASE_P(
    NEON, Trans16x16DCT,
-    ::testing::Values(make_tuple(&vpx_fdct16x16_c, &vpx_idct16x16_256_add_neon,
-                                 0, VPX_BITS_8)));
-#endif
+    ::testing::Values(make_tuple(&vpx_fdct16x16_neon,
+                                 &vpx_idct16x16_256_add_neon, 0, VPX_BITS_8)));
+#endif  // HAVE_NEON && !CONFIG_EMULATE_HARDWARE

 #if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 INSTANTIATE_TEST_CASE_P(
@@ -877,9 +808,6 @@ INSTANTIATE_TEST_CASE_P(
                                 2, VPX_BITS_8),
                      make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2,
                                 3, VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(SSE2, PartialTrans16x16Test,
-                        ::testing::Values(make_tuple(&vpx_fdct16x16_1_sse2,
-                                                     VPX_BITS_8)));
 #endif  // HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE

 #if HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
@@ -914,9 +842,6 @@ INSTANTIATE_TEST_CASE_P(
                                 &idct16x16_10_add_12_sse2, 3167, VPX_BITS_12),
                      make_tuple(&idct16x16_12, &idct16x16_256_add_12_sse2,
                                 3167, VPX_BITS_12)));
-INSTANTIATE_TEST_CASE_P(SSE2, PartialTrans16x16Test,
-                        ::testing::Values(make_tuple(&vpx_fdct16x16_1_sse2,
-                                                     VPX_BITS_8)));
 #endif  // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE

 #if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
@@ -932,8 +857,12 @@ INSTANTIATE_TEST_CASE_P(
        make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 2, VPX_BITS_8),
        make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 3,
                   VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(MSA, PartialTrans16x16Test,
-                        ::testing::Values(make_tuple(&vpx_fdct16x16_1_msa,
-                                                     VPX_BITS_8)));
 #endif  // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+
+#if HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+INSTANTIATE_TEST_CASE_P(VSX, Trans16x16DCT,
+                        ::testing::Values(make_tuple(&vpx_fdct16x16_c,
+                                                     &vpx_idct16x16_256_add_vsx,
+                                                     0, VPX_BITS_8)));
+#endif  // HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 }  // namespace
--- a/test/dct32x32_test.cc
+++ b/test/dct32x32_test.cc
@@ -71,11 +71,11 @@ typedef std::tr1::tuple<FwdTxfmFunc, InvTxfmFunc, int, vpx_bit_depth_t>

 #if CONFIG_VP9_HIGHBITDEPTH
 void idct32x32_10(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct32x32_1024_add_c(in, out, stride, 10);
+  vpx_highbd_idct32x32_1024_add_c(in, CAST_TO_SHORTPTR(out), stride, 10);
 }

 void idct32x32_12(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct32x32_1024_add_c(in, out, stride, 12);
+  vpx_highbd_idct32x32_1024_add_c(in, CAST_TO_SHORTPTR(out), stride, 12);
 }
 #endif  // CONFIG_VP9_HIGHBITDEPTH

@@ -137,7 +137,7 @@ TEST_P(Trans32x32Test, AccuracyCheck) {
 #if CONFIG_VP9_HIGHBITDEPTH
    } else {
      ASM_REGISTER_STATE_CHECK(
-          inv_txfm_(test_temp_block, CONVERT_TO_BYTEPTR(dst16), 32));
+          inv_txfm_(test_temp_block, CAST_TO_BYTEPTR(dst16), 32));
 #endif
    }

@@ -275,7 +275,7 @@ TEST_P(Trans32x32Test, InverseAccuracy) {
      ASM_REGISTER_STATE_CHECK(inv_txfm_(coeff, dst, 32));
 #if CONFIG_VP9_HIGHBITDEPTH
    } else {
-      ASM_REGISTER_STATE_CHECK(inv_txfm_(coeff, CONVERT_TO_BYTEPTR(dst16), 32));
+      ASM_REGISTER_STATE_CHECK(inv_txfm_(coeff, CAST_TO_BYTEPTR(dst16), 32));
 #endif
    }
    for (int j = 0; j < kNumCoeffs; ++j) {
@@ -292,67 +292,6 @@ TEST_P(Trans32x32Test, InverseAccuracy) {
  }
 }

-class PartialTrans32x32Test
-    : public ::testing::TestWithParam<
-          std::tr1::tuple<FwdTxfmFunc, vpx_bit_depth_t> > {
- public:
-  virtual ~PartialTrans32x32Test() {}
-  virtual void SetUp() {
-    fwd_txfm_ = GET_PARAM(0);
-    bit_depth_ = GET_PARAM(1);
-  }
-
-  virtual void TearDown() { libvpx_test::ClearSystemState(); }
-
- protected:
-  vpx_bit_depth_t bit_depth_;
-  FwdTxfmFunc fwd_txfm_;
-};
-
-TEST_P(PartialTrans32x32Test, Extremes) {
-#if CONFIG_VP9_HIGHBITDEPTH
-  const int16_t maxval =
-      static_cast<int16_t>(clip_pixel_highbd(1 << 30, bit_depth_));
-#else
-  const int16_t maxval = 255;
-#endif
-  const int minval = -maxval;
-  DECLARE_ALIGNED(16, int16_t, input[kNumCoeffs]);
-  DECLARE_ALIGNED(16, tran_low_t, output[kNumCoeffs]);
-
-  for (int i = 0; i < kNumCoeffs; ++i) input[i] = maxval;
-  output[0] = 0;
-  ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 32));
-  EXPECT_EQ((maxval * kNumCoeffs) >> 3, output[0]);
-
-  for (int i = 0; i < kNumCoeffs; ++i) input[i] = minval;
-  output[0] = 0;
-  ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 32));
-  EXPECT_EQ((minval * kNumCoeffs) >> 3, output[0]);
-}
-
-TEST_P(PartialTrans32x32Test, Random) {
-#if CONFIG_VP9_HIGHBITDEPTH
-  const int16_t maxval =
-      static_cast<int16_t>(clip_pixel_highbd(1 << 30, bit_depth_));
-#else
-  const int16_t maxval = 255;
-#endif
-  DECLARE_ALIGNED(16, int16_t, input[kNumCoeffs]);
-  DECLARE_ALIGNED(16, tran_low_t, output[kNumCoeffs]);
-  ACMRandom rnd(ACMRandom::DeterministicSeed());
-
-  int sum = 0;
-  for (int i = 0; i < kNumCoeffs; ++i) {
-    const int val = (i & 1) ? -rnd(maxval + 1) : rnd(maxval + 1);
-    input[i] = val;
-    sum += val;
-  }
-  output[0] = 0;
-  ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 32));
-  EXPECT_EQ(sum >> 3, output[0]);
-}
-
 using std::tr1::make_tuple;

 #if CONFIG_VP9_HIGHBITDEPTH
@@ -366,11 +305,6 @@ INSTANTIATE_TEST_CASE_P(
        make_tuple(&vpx_fdct32x32_c, &vpx_idct32x32_1024_add_c, 0, VPX_BITS_8),
        make_tuple(&vpx_fdct32x32_rd_c, &vpx_idct32x32_1024_add_c, 1,
                   VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(
-    C, PartialTrans32x32Test,
-    ::testing::Values(make_tuple(&vpx_highbd_fdct32x32_1_c, VPX_BITS_8),
-                      make_tuple(&vpx_highbd_fdct32x32_1_c, VPX_BITS_10),
-                      make_tuple(&vpx_highbd_fdct32x32_1_c, VPX_BITS_12)));
 #else
 INSTANTIATE_TEST_CASE_P(
    C, Trans32x32Test,
@@ -378,19 +312,16 @@ INSTANTIATE_TEST_CASE_P(
                                 VPX_BITS_8),
                      make_tuple(&vpx_fdct32x32_rd_c, &vpx_idct32x32_1024_add_c,
                                 1, VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(C, PartialTrans32x32Test,
-                        ::testing::Values(make_tuple(&vpx_fdct32x32_1_c,
-                                                     VPX_BITS_8)));
 #endif  // CONFIG_VP9_HIGHBITDEPTH

-#if HAVE_NEON && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+#if HAVE_NEON && !CONFIG_EMULATE_HARDWARE
 INSTANTIATE_TEST_CASE_P(
    NEON, Trans32x32Test,
-    ::testing::Values(make_tuple(&vpx_fdct32x32_c, &vpx_idct32x32_1024_add_neon,
-                                 0, VPX_BITS_8),
-                      make_tuple(&vpx_fdct32x32_rd_c,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_neon,
+                                 &vpx_idct32x32_1024_add_neon, 0, VPX_BITS_8),
+                      make_tuple(&vpx_fdct32x32_rd_neon,
                                 &vpx_idct32x32_1024_add_neon, 1, VPX_BITS_8)));
-#endif  // HAVE_NEON && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+#endif  // HAVE_NEON && !CONFIG_EMULATE_HARDWARE

 #if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 INSTANTIATE_TEST_CASE_P(
@@ -399,9 +330,6 @@ INSTANTIATE_TEST_CASE_P(
                                 &vpx_idct32x32_1024_add_sse2, 0, VPX_BITS_8),
                      make_tuple(&vpx_fdct32x32_rd_sse2,
                                 &vpx_idct32x32_1024_add_sse2, 1, VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(SSE2, PartialTrans32x32Test,
-                        ::testing::Values(make_tuple(&vpx_fdct32x32_1_sse2,
-                                                     VPX_BITS_8)));
 #endif  // HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE

 #if HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
@@ -418,9 +346,6 @@ INSTANTIATE_TEST_CASE_P(
                   VPX_BITS_8),
        make_tuple(&vpx_fdct32x32_rd_sse2, &vpx_idct32x32_1024_add_c, 1,
                   VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(SSE2, PartialTrans32x32Test,
-                        ::testing::Values(make_tuple(&vpx_fdct32x32_1_sse2,
-                                                     VPX_BITS_8)));
 #endif  // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE

 #if HAVE_AVX2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
@@ -439,8 +364,14 @@ INSTANTIATE_TEST_CASE_P(
                                 &vpx_idct32x32_1024_add_msa, 0, VPX_BITS_8),
                      make_tuple(&vpx_fdct32x32_rd_msa,
                                 &vpx_idct32x32_1024_add_msa, 1, VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(MSA, PartialTrans32x32Test,
-                        ::testing::Values(make_tuple(&vpx_fdct32x32_1_msa,
-                                                     VPX_BITS_8)));
 #endif  // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+
+#if HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+INSTANTIATE_TEST_CASE_P(
+    VSX, Trans32x32Test,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_c, &vpx_idct32x32_1024_add_vsx,
+                                 0, VPX_BITS_8),
+                      make_tuple(&vpx_fdct32x32_rd_c,
+                                 &vpx_idct32x32_1024_add_vsx, 1, VPX_BITS_8)));
+#endif  // HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 }  // namespace
--- a/test/dct_partial_test.cc
+++ b/test/dct_partial_test.cc
@@ -0,0 +1,169 @@
+/*
+ *  Copyright (c) 2017 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 <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <limits>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vpx_dsp_rtcd.h"
+#include "test/acm_random.h"
+#include "test/buffer.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "vpx/vpx_codec.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_dsp/vpx_dsp_common.h"
+
+using libvpx_test::ACMRandom;
+using libvpx_test::Buffer;
+using std::tr1::tuple;
+using std::tr1::make_tuple;
+
+namespace {
+typedef void (*PartialFdctFunc)(const int16_t *in, tran_low_t *out, int stride);
+
+typedef tuple<PartialFdctFunc, int /* size */, vpx_bit_depth_t>
+    PartialFdctParam;
+
+tran_low_t partial_fdct_ref(const Buffer<int16_t> &in, int size) {
+  int64_t sum = 0;
+  for (int y = 0; y < size; ++y) {
+    for (int x = 0; x < size; ++x) {
+      sum += in.TopLeftPixel()[y * in.stride() + x];
+    }
+  }
+
+  switch (size) {
+    case 4: sum *= 2; break;
+    case 8: /*sum = sum;*/ break;
+    case 16: sum >>= 1; break;
+    case 32: sum >>= 3; break;
+  }
+
+  return static_cast<tran_low_t>(sum);
+}
+
+class PartialFdctTest : public ::testing::TestWithParam<PartialFdctParam> {
+ public:
+  PartialFdctTest() {
+    fwd_txfm_ = GET_PARAM(0);
+    size_ = GET_PARAM(1);
+    bit_depth_ = GET_PARAM(2);
+  }
+
+  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  void RunTest() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int16_t maxvalue =
+        clip_pixel_highbd(std::numeric_limits<int16_t>::max(), bit_depth_);
+    const int16_t minvalue = -maxvalue;
+    Buffer<int16_t> input_block =
+        Buffer<int16_t>(size_, size_, 8, size_ == 4 ? 0 : 16);
+    ASSERT_TRUE(input_block.Init());
+    Buffer<tran_low_t> output_block = Buffer<tran_low_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(output_block.Init());
+
+    for (int i = 0; i < 100; ++i) {
+      if (i == 0) {
+        input_block.Set(maxvalue);
+      } else if (i == 1) {
+        input_block.Set(minvalue);
+      } else {
+        input_block.Set(&rnd, minvalue, maxvalue);
+      }
+
+      ASM_REGISTER_STATE_CHECK(fwd_txfm_(input_block.TopLeftPixel(),
+                                         output_block.TopLeftPixel(),
+                                         input_block.stride()));
+
+      EXPECT_EQ(partial_fdct_ref(input_block, size_),
+                output_block.TopLeftPixel()[0]);
+    }
+  }
+
+  PartialFdctFunc fwd_txfm_;
+  vpx_bit_depth_t bit_depth_;
+  int size_;
+};
+
+TEST_P(PartialFdctTest, PartialFdctTest) { RunTest(); }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    C, PartialFdctTest,
+    ::testing::Values(make_tuple(&vpx_highbd_fdct32x32_1_c, 32, VPX_BITS_12),
+                      make_tuple(&vpx_highbd_fdct32x32_1_c, 32, VPX_BITS_10),
+                      make_tuple(&vpx_fdct32x32_1_c, 32, VPX_BITS_8),
+                      make_tuple(&vpx_highbd_fdct16x16_1_c, 16, VPX_BITS_12),
+                      make_tuple(&vpx_highbd_fdct16x16_1_c, 16, VPX_BITS_10),
+                      make_tuple(&vpx_fdct16x16_1_c, 16, VPX_BITS_8),
+                      make_tuple(&vpx_highbd_fdct8x8_1_c, 8, VPX_BITS_12),
+                      make_tuple(&vpx_highbd_fdct8x8_1_c, 8, VPX_BITS_10),
+                      make_tuple(&vpx_fdct8x8_1_c, 8, VPX_BITS_8),
+                      make_tuple(&vpx_fdct4x4_1_c, 4, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, PartialFdctTest,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_1_c, 32, VPX_BITS_8),
+                      make_tuple(&vpx_fdct16x16_1_c, 16, VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_1_c, 8, VPX_BITS_8),
+                      make_tuple(&vpx_fdct4x4_1_c, 4, VPX_BITS_8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, PartialFdctTest,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_1_sse2, 32, VPX_BITS_8),
+                      make_tuple(&vpx_fdct16x16_1_sse2, 16, VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_1_sse2, 8, VPX_BITS_8),
+                      make_tuple(&vpx_fdct4x4_1_sse2, 4, VPX_BITS_8)));
+#endif  // HAVE_SSE2
+
+#if HAVE_NEON
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    NEON, PartialFdctTest,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_1_neon, 32, VPX_BITS_8),
+                      make_tuple(&vpx_fdct16x16_1_neon, 16, VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_1_neon, 8, VPX_BITS_12),
+                      make_tuple(&vpx_fdct8x8_1_neon, 8, VPX_BITS_10),
+                      make_tuple(&vpx_fdct8x8_1_neon, 8, VPX_BITS_8),
+                      make_tuple(&vpx_fdct4x4_1_neon, 4, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    NEON, PartialFdctTest,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_1_neon, 32, VPX_BITS_8),
+                      make_tuple(&vpx_fdct16x16_1_neon, 16, VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_1_neon, 8, VPX_BITS_8),
+                      make_tuple(&vpx_fdct4x4_1_neon, 4, VPX_BITS_8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_NEON
+
+#if HAVE_MSA
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(MSA, PartialFdctTest,
+                        ::testing::Values(make_tuple(&vpx_fdct8x8_1_msa, 8,
+                                                     VPX_BITS_8)));
+#else   // !CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    MSA, PartialFdctTest,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_1_msa, 32, VPX_BITS_8),
+                      make_tuple(&vpx_fdct16x16_1_msa, 16, VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_1_msa, 8, VPX_BITS_8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_MSA
+}  // namespace
--- a/test/dct_test.cc
+++ b/test/dct_test.cc
@@ -0,0 +1,737 @@
+/*
+ *  Copyright (c) 2017 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 <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
+#include "test/acm_random.h"
+#include "test/buffer.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "vp9/common/vp9_entropy.h"
+#include "vpx/vpx_codec.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+
+using libvpx_test::ACMRandom;
+using libvpx_test::Buffer;
+using std::tr1::tuple;
+using std::tr1::make_tuple;
+
+namespace {
+typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
+typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
+typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
+                        int tx_type);
+typedef void (*FhtFuncRef)(const Buffer<int16_t> &in, Buffer<tran_low_t> *out,
+                           int size, int tx_type);
+typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
+                        int tx_type);
+
+/* forward transform, inverse transform, size, transform type, bit depth */
+typedef tuple<FdctFunc, IdctFunc, int, int, vpx_bit_depth_t> DctParam;
+typedef tuple<FhtFunc, IhtFunc, int, int, vpx_bit_depth_t> HtParam;
+
+void fdct_ref(const Buffer<int16_t> &in, Buffer<tran_low_t> *out, int size,
+              int /*tx_type*/) {
+  const int16_t *i = in.TopLeftPixel();
+  const int i_stride = in.stride();
+  tran_low_t *o = out->TopLeftPixel();
+  if (size == 4) {
+    vpx_fdct4x4_c(i, o, i_stride);
+  } else if (size == 8) {
+    vpx_fdct8x8_c(i, o, i_stride);
+  } else if (size == 16) {
+    vpx_fdct16x16_c(i, o, i_stride);
+  } else if (size == 32) {
+    vpx_fdct32x32_c(i, o, i_stride);
+  }
+}
+
+void fht_ref(const Buffer<int16_t> &in, Buffer<tran_low_t> *out, int size,
+             int tx_type) {
+  const int16_t *i = in.TopLeftPixel();
+  const int i_stride = in.stride();
+  tran_low_t *o = out->TopLeftPixel();
+  if (size == 4) {
+    vp9_fht4x4_c(i, o, i_stride, tx_type);
+  } else if (size == 8) {
+    vp9_fht8x8_c(i, o, i_stride, tx_type);
+  } else if (size == 16) {
+    vp9_fht16x16_c(i, o, i_stride, tx_type);
+  }
+}
+
+void fwht_ref(const Buffer<int16_t> &in, Buffer<tran_low_t> *out, int size,
+              int /*tx_type*/) {
+  ASSERT_EQ(size, 4);
+  vp9_fwht4x4_c(in.TopLeftPixel(), out->TopLeftPixel(), in.stride());
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+#define idctNxN(n, coeffs, bitdepth)                                       \
+  void idct##n##x##n##_##bitdepth(const tran_low_t *in, uint8_t *out,      \
+                                  int stride) {                            \
+    vpx_highbd_idct##n##x##n##_##coeffs##_add_c(in, CAST_TO_SHORTPTR(out), \
+                                                stride, bitdepth);         \
+  }
+
+idctNxN(4, 16, 10);
+idctNxN(4, 16, 12);
+idctNxN(8, 64, 10);
+idctNxN(8, 64, 12);
+idctNxN(16, 256, 10);
+idctNxN(16, 256, 12);
+idctNxN(32, 1024, 10);
+idctNxN(32, 1024, 12);
+
+#define ihtNxN(n, coeffs, bitdepth)                                        \
+  void iht##n##x##n##_##bitdepth(const tran_low_t *in, uint8_t *out,       \
+                                 int stride, int tx_type) {                \
+    vp9_highbd_iht##n##x##n##_##coeffs##_add_c(in, CAST_TO_SHORTPTR(out),  \
+                                               stride, tx_type, bitdepth); \
+  }
+
+ihtNxN(4, 16, 10);
+ihtNxN(4, 16, 12);
+ihtNxN(8, 64, 10);
+ihtNxN(8, 64, 12);
+ihtNxN(16, 256, 10);
+// ihtNxN(16, 256, 12);
+
+void iwht4x4_10(const tran_low_t *in, uint8_t *out, int stride) {
+  vpx_highbd_iwht4x4_16_add_c(in, CAST_TO_SHORTPTR(out), stride, 10);
+}
+
+void iwht4x4_12(const tran_low_t *in, uint8_t *out, int stride) {
+  vpx_highbd_iwht4x4_16_add_c(in, CAST_TO_SHORTPTR(out), stride, 12);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+class TransTestBase {
+ public:
+  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+
+ protected:
+  virtual void RunFwdTxfm(const Buffer<int16_t> &in,
+                          Buffer<tran_low_t> *out) = 0;
+
+  virtual void RunInvTxfm(const Buffer<tran_low_t> &in, uint8_t *out) = 0;
+
+  void RunAccuracyCheck(int limit) {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    Buffer<int16_t> test_input_block =
+        Buffer<int16_t>(size_, size_, 8, size_ == 4 ? 0 : 16);
+    ASSERT_TRUE(test_input_block.Init());
+    Buffer<tran_low_t> test_temp_block =
+        Buffer<tran_low_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(test_temp_block.Init());
+    Buffer<uint8_t> dst = Buffer<uint8_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(dst.Init());
+    Buffer<uint8_t> src = Buffer<uint8_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(src.Init());
+#if CONFIG_VP9_HIGHBITDEPTH
+    Buffer<uint16_t> dst16 = Buffer<uint16_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(dst16.Init());
+    Buffer<uint16_t> src16 = Buffer<uint16_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(src16.Init());
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+    uint32_t max_error = 0;
+    int64_t total_error = 0;
+    const int count_test_block = 10000;
+    for (int i = 0; i < count_test_block; ++i) {
+      if (bit_depth_ == 8) {
+        src.Set(&rnd, &ACMRandom::Rand8);
+        dst.Set(&rnd, &ACMRandom::Rand8);
+        // Initialize a test block with input range [-255, 255].
+        for (int h = 0; h < size_; ++h) {
+          for (int w = 0; w < size_; ++w) {
+            test_input_block.TopLeftPixel()[h * test_input_block.stride() + w] =
+                src.TopLeftPixel()[h * src.stride() + w] -
+                dst.TopLeftPixel()[h * dst.stride() + w];
+          }
+        }
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        src16.Set(&rnd, 0, max_pixel_value_);
+        dst16.Set(&rnd, 0, max_pixel_value_);
+        for (int h = 0; h < size_; ++h) {
+          for (int w = 0; w < size_; ++w) {
+            test_input_block.TopLeftPixel()[h * test_input_block.stride() + w] =
+                src16.TopLeftPixel()[h * src16.stride() + w] -
+                dst16.TopLeftPixel()[h * dst16.stride() + w];
+          }
+        }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+      }
+
+      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block, &test_temp_block));
+      if (bit_depth_ == VPX_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(test_temp_block, dst.TopLeftPixel()));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(test_temp_block, CAST_TO_BYTEPTR(dst16.TopLeftPixel())));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+      }
+
+      for (int h = 0; h < size_; ++h) {
+        for (int w = 0; w < size_; ++w) {
+          int diff;
+#if CONFIG_VP9_HIGHBITDEPTH
+          if (bit_depth_ != 8) {
+            diff = dst16.TopLeftPixel()[h * dst16.stride() + w] -
+                   src16.TopLeftPixel()[h * src16.stride() + w];
+          } else {
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+            diff = dst.TopLeftPixel()[h * dst.stride() + w] -
+                   src.TopLeftPixel()[h * src.stride() + w];
+#if CONFIG_VP9_HIGHBITDEPTH
+          }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+          const uint32_t error = diff * diff;
+          if (max_error < error) max_error = error;
+          total_error += error;
+        }
+      }
+    }
+
+    EXPECT_GE(static_cast<uint32_t>(limit), max_error)
+        << "Error: 4x4 FHT/IHT has an individual round trip error > " << limit;
+
+    EXPECT_GE(count_test_block * limit, total_error)
+        << "Error: 4x4 FHT/IHT has average round trip error > " << limit
+        << " per block";
+  }
+
+  void RunCoeffCheck() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int count_test_block = 5000;
+    Buffer<int16_t> input_block =
+        Buffer<int16_t>(size_, size_, 8, size_ == 4 ? 0 : 16);
+    ASSERT_TRUE(input_block.Init());
+    Buffer<tran_low_t> output_ref_block = Buffer<tran_low_t>(size_, size_, 0);
+    ASSERT_TRUE(output_ref_block.Init());
+    Buffer<tran_low_t> output_block = Buffer<tran_low_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(output_block.Init());
+
+    for (int i = 0; i < count_test_block; ++i) {
+      // Initialize a test block with input range [-max_pixel_value_,
+      // max_pixel_value_].
+      input_block.Set(&rnd, -max_pixel_value_, max_pixel_value_);
+
+      fwd_txfm_ref(input_block, &output_ref_block, size_, tx_type_);
+      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, &output_block));
+
+      // The minimum quant value is 4.
+      EXPECT_TRUE(output_block.CheckValues(output_ref_block));
+      if (::testing::Test::HasFailure()) {
+        printf("Size: %d Transform type: %d\n", size_, tx_type_);
+        output_block.PrintDifference(output_ref_block);
+        return;
+      }
+    }
+  }
+
+  void RunMemCheck() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int count_test_block = 5000;
+    Buffer<int16_t> input_extreme_block =
+        Buffer<int16_t>(size_, size_, 8, size_ == 4 ? 0 : 16);
+    ASSERT_TRUE(input_extreme_block.Init());
+    Buffer<tran_low_t> output_ref_block = Buffer<tran_low_t>(size_, size_, 0);
+    ASSERT_TRUE(output_ref_block.Init());
+    Buffer<tran_low_t> output_block = Buffer<tran_low_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(output_block.Init());
+
+    for (int i = 0; i < count_test_block; ++i) {
+      // Initialize a test block with -max_pixel_value_ or max_pixel_value_.
+      if (i == 0) {
+        input_extreme_block.Set(max_pixel_value_);
+      } else if (i == 1) {
+        input_extreme_block.Set(-max_pixel_value_);
+      } else {
+        for (int h = 0; h < size_; ++h) {
+          for (int w = 0; w < size_; ++w) {
+            input_extreme_block
+                .TopLeftPixel()[h * input_extreme_block.stride() + w] =
+                rnd.Rand8() % 2 ? max_pixel_value_ : -max_pixel_value_;
+          }
+        }
+      }
+
+      fwd_txfm_ref(input_extreme_block, &output_ref_block, size_, tx_type_);
+      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block, &output_block));
+
+      // The minimum quant value is 4.
+      EXPECT_TRUE(output_block.CheckValues(output_ref_block));
+      for (int h = 0; h < size_; ++h) {
+        for (int w = 0; w < size_; ++w) {
+          EXPECT_GE(
+              4 * DCT_MAX_VALUE << (bit_depth_ - 8),
+              abs(output_block.TopLeftPixel()[h * output_block.stride() + w]))
+              << "Error: 4x4 FDCT has coefficient larger than "
+                 "4*DCT_MAX_VALUE"
+              << " at " << w << "," << h;
+          if (::testing::Test::HasFailure()) {
+            printf("Size: %d Transform type: %d\n", size_, tx_type_);
+            output_block.DumpBuffer();
+            return;
+          }
+        }
+      }
+    }
+  }
+
+  void RunInvAccuracyCheck(int limit) {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int count_test_block = 1000;
+    Buffer<int16_t> in = Buffer<int16_t>(size_, size_, 4);
+    ASSERT_TRUE(in.Init());
+    Buffer<tran_low_t> coeff = Buffer<tran_low_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(coeff.Init());
+    Buffer<uint8_t> dst = Buffer<uint8_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(dst.Init());
+    Buffer<uint8_t> src = Buffer<uint8_t>(size_, size_, 0);
+    ASSERT_TRUE(src.Init());
+    Buffer<uint16_t> dst16 = Buffer<uint16_t>(size_, size_, 0, 16);
+    ASSERT_TRUE(dst16.Init());
+    Buffer<uint16_t> src16 = Buffer<uint16_t>(size_, size_, 0);
+    ASSERT_TRUE(src16.Init());
+
+    for (int i = 0; i < count_test_block; ++i) {
+      // Initialize a test block with input range [-max_pixel_value_,
+      // max_pixel_value_].
+      if (bit_depth_ == VPX_BITS_8) {
+        src.Set(&rnd, &ACMRandom::Rand8);
+        dst.Set(&rnd, &ACMRandom::Rand8);
+        for (int h = 0; h < size_; ++h) {
+          for (int w = 0; w < size_; ++w) {
+            in.TopLeftPixel()[h * in.stride() + w] =
+                src.TopLeftPixel()[h * src.stride() + w] -
+                dst.TopLeftPixel()[h * dst.stride() + w];
+          }
+        }
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        src16.Set(&rnd, 0, max_pixel_value_);
+        dst16.Set(&rnd, 0, max_pixel_value_);
+        for (int h = 0; h < size_; ++h) {
+          for (int w = 0; w < size_; ++w) {
+            in.TopLeftPixel()[h * in.stride() + w] =
+                src16.TopLeftPixel()[h * src16.stride() + w] -
+                dst16.TopLeftPixel()[h * dst16.stride() + w];
+          }
+        }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+      }
+
+      fwd_txfm_ref(in, &coeff, size_, tx_type_);
+
+      if (bit_depth_ == VPX_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst.TopLeftPixel()));
+#if CONFIG_VP9_HIGHBITDEPTH
+      } else {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(coeff, CAST_TO_BYTEPTR(dst16.TopLeftPixel())));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+      }
+
+      for (int h = 0; h < size_; ++h) {
+        for (int w = 0; w < size_; ++w) {
+          int diff;
+#if CONFIG_VP9_HIGHBITDEPTH
+          if (bit_depth_ != 8) {
+            diff = dst16.TopLeftPixel()[h * dst16.stride() + w] -
+                   src16.TopLeftPixel()[h * src16.stride() + w];
+          } else {
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+            diff = dst.TopLeftPixel()[h * dst.stride() + w] -
+                   src.TopLeftPixel()[h * src.stride() + w];
+#if CONFIG_VP9_HIGHBITDEPTH
+          }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+          const uint32_t error = diff * diff;
+          EXPECT_GE(static_cast<uint32_t>(limit), error)
+              << "Error: " << size_ << "x" << size_ << " IDCT has error "
+              << error << " at " << w << "," << h;
+        }
+      }
+    }
+  }
+
+  FhtFuncRef fwd_txfm_ref;
+  vpx_bit_depth_t bit_depth_;
+  int tx_type_;
+  int max_pixel_value_;
+  int size_;
+};
+
+class TransDCT : public TransTestBase,
+                 public ::testing::TestWithParam<DctParam> {
+ public:
+  TransDCT() {
+    fwd_txfm_ref = fdct_ref;
+    fwd_txfm_ = GET_PARAM(0);
+    inv_txfm_ = GET_PARAM(1);
+    size_ = GET_PARAM(2);
+    tx_type_ = GET_PARAM(3);
+    bit_depth_ = GET_PARAM(4);
+    max_pixel_value_ = (1 << bit_depth_) - 1;
+  }
+
+ protected:
+  void RunFwdTxfm(const Buffer<int16_t> &in, Buffer<tran_low_t> *out) {
+    fwd_txfm_(in.TopLeftPixel(), out->TopLeftPixel(), in.stride());
+  }
+
+  void RunInvTxfm(const Buffer<tran_low_t> &in, uint8_t *out) {
+    inv_txfm_(in.TopLeftPixel(), out, in.stride());
+  }
+
+  FdctFunc fwd_txfm_;
+  IdctFunc inv_txfm_;
+};
+
+TEST_P(TransDCT, AccuracyCheck) { RunAccuracyCheck(1); }
+
+TEST_P(TransDCT, CoeffCheck) { RunCoeffCheck(); }
+
+TEST_P(TransDCT, MemCheck) { RunMemCheck(); }
+
+TEST_P(TransDCT, InvAccuracyCheck) { RunInvAccuracyCheck(1); }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    C, TransDCT,
+    ::testing::Values(
+        make_tuple(&vpx_highbd_fdct32x32_c, &idct32x32_10, 32, 0, VPX_BITS_10),
+        make_tuple(&vpx_highbd_fdct32x32_c, &idct32x32_12, 32, 0, VPX_BITS_10),
+        make_tuple(&vpx_fdct32x32_c, &vpx_idct32x32_1024_add_c, 32, 0,
+                   VPX_BITS_8),
+        make_tuple(&vpx_highbd_fdct16x16_c, &idct16x16_10, 16, 0, VPX_BITS_10),
+        make_tuple(&vpx_highbd_fdct16x16_c, &idct16x16_12, 16, 0, VPX_BITS_10),
+        make_tuple(&vpx_fdct16x16_c, &vpx_idct16x16_256_add_c, 16, 0,
+                   VPX_BITS_8),
+        make_tuple(&vpx_highbd_fdct8x8_c, &idct8x8_10, 8, 0, VPX_BITS_10),
+        make_tuple(&vpx_highbd_fdct8x8_c, &idct8x8_12, 8, 0, VPX_BITS_10),
+        make_tuple(&vpx_fdct8x8_c, &vpx_idct8x8_64_add_c, 8, 0, VPX_BITS_8),
+        make_tuple(&vpx_highbd_fdct4x4_c, &idct4x4_10, 4, 0, VPX_BITS_10),
+        make_tuple(&vpx_highbd_fdct4x4_c, &idct4x4_12, 4, 0, VPX_BITS_12),
+        make_tuple(&vpx_fdct4x4_c, &vpx_idct4x4_16_add_c, 4, 0, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, TransDCT,
+    ::testing::Values(
+        make_tuple(&vpx_fdct32x32_c, &vpx_idct32x32_1024_add_c, 32, 0,
+                   VPX_BITS_8),
+        make_tuple(&vpx_fdct16x16_c, &vpx_idct16x16_256_add_c, 16, 0,
+                   VPX_BITS_8),
+        make_tuple(&vpx_fdct8x8_c, &vpx_idct8x8_64_add_c, 8, 0, VPX_BITS_8),
+        make_tuple(&vpx_fdct4x4_c, &vpx_idct4x4_16_add_c, 4, 0, VPX_BITS_8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if HAVE_SSE2
+#if !CONFIG_EMULATE_HARDWARE
+#if CONFIG_VP9_HIGHBITDEPTH
+/* TODO:(johannkoenig) Determine why these fail AccuracyCheck
+   make_tuple(&vpx_highbd_fdct32x32_sse2, &idct32x32_12, 32, 0, VPX_BITS_12),
+   make_tuple(&vpx_highbd_fdct16x16_sse2, &idct16x16_12, 16, 0, VPX_BITS_12),
+*/
+INSTANTIATE_TEST_CASE_P(
+    SSE2, TransDCT,
+    ::testing::Values(
+        make_tuple(&vpx_highbd_fdct32x32_sse2, &idct32x32_10, 32, 0,
+                   VPX_BITS_10),
+        make_tuple(&vpx_fdct32x32_sse2, &vpx_idct32x32_1024_add_sse2, 32, 0,
+                   VPX_BITS_8),
+        make_tuple(&vpx_highbd_fdct16x16_sse2, &idct16x16_10, 16, 0,
+                   VPX_BITS_10),
+        make_tuple(&vpx_fdct16x16_sse2, &vpx_idct16x16_256_add_sse2, 16, 0,
+                   VPX_BITS_8),
+        make_tuple(&vpx_highbd_fdct8x8_sse2, &idct8x8_10, 8, 0, VPX_BITS_10),
+        make_tuple(&vpx_highbd_fdct8x8_sse2, &idct8x8_12, 8, 0, VPX_BITS_12),
+        make_tuple(&vpx_fdct8x8_sse2, &vpx_idct8x8_64_add_sse2, 8, 0,
+                   VPX_BITS_8),
+        make_tuple(&vpx_highbd_fdct4x4_sse2, &idct4x4_10, 4, 0, VPX_BITS_10),
+        make_tuple(&vpx_highbd_fdct4x4_sse2, &idct4x4_12, 4, 0, VPX_BITS_12),
+        make_tuple(&vpx_fdct4x4_sse2, &vpx_idct4x4_16_add_sse2, 4, 0,
+                   VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    SSE2, TransDCT,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_sse2,
+                                 &vpx_idct32x32_1024_add_sse2, 32, 0,
+                                 VPX_BITS_8),
+                      make_tuple(&vpx_fdct16x16_sse2,
+                                 &vpx_idct16x16_256_add_sse2, 16, 0,
+                                 VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_sse2, &vpx_idct8x8_64_add_sse2, 8,
+                                 0, VPX_BITS_8),
+                      make_tuple(&vpx_fdct4x4_sse2, &vpx_idct4x4_16_add_sse2, 4,
+                                 0, VPX_BITS_8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // !CONFIG_EMULATE_HARDWARE
+#endif  // HAVE_SSE2
+
+#if !CONFIG_VP9_HIGHBITDEPTH
+#if HAVE_SSSE3 && !CONFIG_EMULATE_HARDWARE
+#if !ARCH_X86_64
+// TODO(johannkoenig): high bit depth fdct8x8.
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, TransDCT,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_c, &vpx_idct32x32_1024_add_sse2,
+                                 32, 0, VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_c, &vpx_idct8x8_64_add_sse2, 8, 0,
+                                 VPX_BITS_8)));
+#else
+// vpx_fdct8x8_ssse3 is only available in 64 bit builds.
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, TransDCT,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_c, &vpx_idct32x32_1024_add_sse2,
+                                 32, 0, VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_ssse3, &vpx_idct8x8_64_add_sse2,
+                                 8, 0, VPX_BITS_8)));
+#endif  // !ARCH_X86_64
+#endif  // HAVE_SSSE3 && !CONFIG_EMULATE_HARDWARE
+#endif  // !CONFIG_VP9_HIGHBITDEPTH
+
+#if !CONFIG_VP9_HIGHBITDEPTH && HAVE_AVX2 && !CONFIG_EMULATE_HARDWARE
+// TODO(johannkoenig): high bit depth fdct32x32.
+INSTANTIATE_TEST_CASE_P(
+    AVX2, TransDCT, ::testing::Values(make_tuple(&vpx_fdct32x32_avx2,
+                                                 &vpx_idct32x32_1024_add_sse2,
+                                                 32, 0, VPX_BITS_8)));
+
+#endif  // !CONFIG_VP9_HIGHBITDEPTH && HAVE_AVX2 && !CONFIG_EMULATE_HARDWARE
+
+#if HAVE_NEON
+#if !CONFIG_EMULATE_HARDWARE
+INSTANTIATE_TEST_CASE_P(
+    NEON, TransDCT,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_neon,
+                                 &vpx_idct32x32_1024_add_neon, 32, 0,
+                                 VPX_BITS_8),
+                      make_tuple(&vpx_fdct16x16_neon,
+                                 &vpx_idct16x16_256_add_neon, 16, 0,
+                                 VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_neon, &vpx_idct8x8_64_add_neon, 8,
+                                 0, VPX_BITS_8),
+                      make_tuple(&vpx_fdct4x4_neon, &vpx_idct4x4_16_add_neon, 4,
+                                 0, VPX_BITS_8)));
+#endif  // !CONFIG_EMULATE_HARDWARE
+#endif  // HAVE_NEON
+
+#if HAVE_MSA
+#if !CONFIG_VP9_HIGHBITDEPTH
+#if !CONFIG_EMULATE_HARDWARE
+INSTANTIATE_TEST_CASE_P(
+    MSA, TransDCT,
+    ::testing::Values(
+        make_tuple(&vpx_fdct32x32_msa, &vpx_idct32x32_1024_add_msa, 32, 0,
+                   VPX_BITS_8),
+        make_tuple(&vpx_fdct16x16_msa, &vpx_idct16x16_256_add_msa, 16, 0,
+                   VPX_BITS_8),
+        make_tuple(&vpx_fdct8x8_msa, &vpx_idct8x8_64_add_msa, 8, 0, VPX_BITS_8),
+        make_tuple(&vpx_fdct4x4_msa, &vpx_idct4x4_16_add_msa, 4, 0,
+                   VPX_BITS_8)));
+#endif  // !CONFIG_EMULATE_HARDWARE
+#endif  // !CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_MSA
+
+#if HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+INSTANTIATE_TEST_CASE_P(VSX, TransDCT,
+                        ::testing::Values(make_tuple(&vpx_fdct4x4_c,
+                                                     &vpx_idct4x4_16_add_vsx, 4,
+                                                     0, VPX_BITS_8)));
+#endif  // HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+
+class TransHT : public TransTestBase, public ::testing::TestWithParam<HtParam> {
+ public:
+  TransHT() {
+    fwd_txfm_ref = fht_ref;
+    fwd_txfm_ = GET_PARAM(0);
+    inv_txfm_ = GET_PARAM(1);
+    size_ = GET_PARAM(2);
+    tx_type_ = GET_PARAM(3);
+    bit_depth_ = GET_PARAM(4);
+    max_pixel_value_ = (1 << bit_depth_) - 1;
+  }
+
+ protected:
+  void RunFwdTxfm(const Buffer<int16_t> &in, Buffer<tran_low_t> *out) {
+    fwd_txfm_(in.TopLeftPixel(), out->TopLeftPixel(), in.stride(), tx_type_);
+  }
+
+  void RunInvTxfm(const Buffer<tran_low_t> &in, uint8_t *out) {
+    inv_txfm_(in.TopLeftPixel(), out, in.stride(), tx_type_);
+  }
+
+  FhtFunc fwd_txfm_;
+  IhtFunc inv_txfm_;
+};
+
+TEST_P(TransHT, AccuracyCheck) { RunAccuracyCheck(1); }
+
+TEST_P(TransHT, CoeffCheck) { RunCoeffCheck(); }
+
+TEST_P(TransHT, MemCheck) { RunMemCheck(); }
+
+TEST_P(TransHT, InvAccuracyCheck) { RunInvAccuracyCheck(1); }
+
+/* TODO:(johannkoenig) Determine why these fail AccuracyCheck
+   make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 16, 0, VPX_BITS_12),
+   make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 16, 1, VPX_BITS_12),
+   make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 16, 2, VPX_BITS_12),
+   make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 16, 3, VPX_BITS_12),
+  */
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    C, TransHT,
+    ::testing::Values(
+        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 16, 0, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 16, 1, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 16, 2, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 16, 3, VPX_BITS_10),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 16, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 16, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 16, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 16, 3, VPX_BITS_8),
+        make_tuple(&vp9_highbd_fht8x8_c, &iht8x8_10, 8, 0, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht8x8_c, &iht8x8_10, 8, 1, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht8x8_c, &iht8x8_10, 8, 2, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht8x8_c, &iht8x8_10, 8, 3, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht8x8_c, &iht8x8_12, 8, 0, VPX_BITS_12),
+        make_tuple(&vp9_highbd_fht8x8_c, &iht8x8_12, 8, 1, VPX_BITS_12),
+        make_tuple(&vp9_highbd_fht8x8_c, &iht8x8_12, 8, 2, VPX_BITS_12),
+        make_tuple(&vp9_highbd_fht8x8_c, &iht8x8_12, 8, 3, VPX_BITS_12),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 8, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 8, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 8, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 8, 3, VPX_BITS_8),
+        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_10, 4, 0, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_10, 4, 1, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_10, 4, 2, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_10, 4, 3, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_12, 4, 0, VPX_BITS_12),
+        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_12, 4, 1, VPX_BITS_12),
+        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_12, 4, 2, VPX_BITS_12),
+        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_12, 4, 3, VPX_BITS_12),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 4, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 4, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 4, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 4, 3, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    C, TransHT,
+    ::testing::Values(
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 16, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 16, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 16, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 16, 3, VPX_BITS_8),
+
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 8, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 8, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 8, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_c, 8, 3, VPX_BITS_8),
+
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 4, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 4, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 4, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 4, 3, VPX_BITS_8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, TransHT,
+    ::testing::Values(
+        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 16, 0,
+                   VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 16, 1,
+                   VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 16, 2,
+                   VPX_BITS_8),
+        make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 16, 3,
+                   VPX_BITS_8),
+
+        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 8, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 8, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 8, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht8x8_sse2, &vp9_iht8x8_64_add_sse2, 8, 3, VPX_BITS_8),
+
+        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 4, 0, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 4, 1, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 4, 2, VPX_BITS_8),
+        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 4, 3,
+                   VPX_BITS_8)));
+#endif  // HAVE_SSE2
+
+class TransWHT : public TransTestBase,
+                 public ::testing::TestWithParam<DctParam> {
+ public:
+  TransWHT() {
+    fwd_txfm_ref = fwht_ref;
+    fwd_txfm_ = GET_PARAM(0);
+    inv_txfm_ = GET_PARAM(1);
+    size_ = GET_PARAM(2);
+    tx_type_ = GET_PARAM(3);
+    bit_depth_ = GET_PARAM(4);
+    max_pixel_value_ = (1 << bit_depth_) - 1;
+  }
+
+ protected:
+  void RunFwdTxfm(const Buffer<int16_t> &in, Buffer<tran_low_t> *out) {
+    fwd_txfm_(in.TopLeftPixel(), out->TopLeftPixel(), in.stride());
+  }
+
+  void RunInvTxfm(const Buffer<tran_low_t> &in, uint8_t *out) {
+    inv_txfm_(in.TopLeftPixel(), out, in.stride());
+  }
+
+  FdctFunc fwd_txfm_;
+  IdctFunc inv_txfm_;
+};
+
+TEST_P(TransWHT, AccuracyCheck) { RunAccuracyCheck(0); }
+
+TEST_P(TransWHT, CoeffCheck) { RunCoeffCheck(); }
+
+TEST_P(TransWHT, MemCheck) { RunMemCheck(); }
+
+TEST_P(TransWHT, InvAccuracyCheck) { RunInvAccuracyCheck(0); }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    C, TransWHT,
+    ::testing::Values(
+        make_tuple(&vp9_highbd_fwht4x4_c, &iwht4x4_10, 4, 0, VPX_BITS_10),
+        make_tuple(&vp9_highbd_fwht4x4_c, &iwht4x4_12, 4, 0, VPX_BITS_12),
+        make_tuple(&vp9_fwht4x4_c, &vpx_iwht4x4_16_add_c, 4, 0, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(C, TransWHT,
+                        ::testing::Values(make_tuple(&vp9_fwht4x4_c,
+                                                     &vpx_iwht4x4_16_add_c, 4,
+                                                     0, VPX_BITS_8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, TransWHT,
+                        ::testing::Values(make_tuple(&vp9_fwht4x4_sse2,
+                                                     &vpx_iwht4x4_16_add_sse2,
+                                                     4, 0, VPX_BITS_8)));
+#endif  // HAVE_SSE2
+}  // namespace
--- a/test/decode_api_test.cc
+++ b/test/decode_api_test.cc
@@ -172,4 +172,21 @@ TEST(DecodeAPI, Vp9PeekSI) {
 }
 #endif  // CONFIG_VP9_DECODER

+TEST(DecodeAPI, HighBitDepthCapability) {
+// VP8 should not claim VP9 HBD as a capability.
+#if CONFIG_VP8_DECODER
+  const vpx_codec_caps_t vp8_caps = vpx_codec_get_caps(&vpx_codec_vp8_dx_algo);
+  EXPECT_EQ(vp8_caps & VPX_CODEC_CAP_HIGHBITDEPTH, 0);
+#endif
+
+#if CONFIG_VP9_DECODER
+  const vpx_codec_caps_t vp9_caps = vpx_codec_get_caps(&vpx_codec_vp9_dx_algo);
+#if CONFIG_VP9_HIGHBITDEPTH
+  EXPECT_EQ(vp9_caps & VPX_CODEC_CAP_HIGHBITDEPTH, VPX_CODEC_CAP_HIGHBITDEPTH);
+#else
+  EXPECT_EQ(vp9_caps & VPX_CODEC_CAP_HIGHBITDEPTH, 0);
+#endif
+#endif
+}
+
 }  // namespace
--- a/test/decode_test_driver.cc
+++ b/test/decode_test_driver.cc
@@ -53,13 +53,13 @@ void DecoderTest::HandlePeekResult(Decoder *const decoder,
     * pass it is not a keyframe, so we only expect VPX_CODEC_OK on the first
     * frame, which must be a keyframe. */
    if (video->frame_number() == 0)
-      ASSERT_EQ(VPX_CODEC_OK, res_peek) << "Peek return failed: "
-                                        << vpx_codec_err_to_string(res_peek);
+      ASSERT_EQ(VPX_CODEC_OK, res_peek)
+          << "Peek return failed: " << vpx_codec_err_to_string(res_peek);
  } else {
    /* The Vp9 implementation of PeekStream returns an error only if the
     * data passed to it isn't a valid Vp9 chunk. */
-    ASSERT_EQ(VPX_CODEC_OK, res_peek) << "Peek return failed: "
-                                      << vpx_codec_err_to_string(res_peek);
+    ASSERT_EQ(VPX_CODEC_OK, res_peek)
+        << "Peek return failed: " << vpx_codec_err_to_string(res_peek);
  }
 }

--- a/test/encode_api_test.cc
+++ b/test/encode_api_test.cc
@@ -62,4 +62,134 @@ TEST(EncodeAPI, InvalidParams) {
  }
 }

+TEST(EncodeAPI, HighBitDepthCapability) {
+// VP8 should not claim VP9 HBD as a capability.
+#if CONFIG_VP8_ENCODER
+  const vpx_codec_caps_t vp8_caps = vpx_codec_get_caps(&vpx_codec_vp8_cx_algo);
+  EXPECT_EQ(vp8_caps & VPX_CODEC_CAP_HIGHBITDEPTH, 0);
+#endif
+
+#if CONFIG_VP9_ENCODER
+  const vpx_codec_caps_t vp9_caps = vpx_codec_get_caps(&vpx_codec_vp9_cx_algo);
+#if CONFIG_VP9_HIGHBITDEPTH
+  EXPECT_EQ(vp9_caps & VPX_CODEC_CAP_HIGHBITDEPTH, VPX_CODEC_CAP_HIGHBITDEPTH);
+#else
+  EXPECT_EQ(vp9_caps & VPX_CODEC_CAP_HIGHBITDEPTH, 0);
+#endif
+#endif
+}
+
+#if CONFIG_VP8_ENCODER
+TEST(EncodeAPI, ImageSizeSetting) {
+  const int width = 711;
+  const int height = 360;
+  const int bps = 12;
+  vpx_image_t img;
+  vpx_codec_ctx_t enc;
+  vpx_codec_enc_cfg_t cfg;
+  uint8_t *img_buf = reinterpret_cast<uint8_t *>(
+      calloc(width * height * bps / 8, sizeof(*img_buf)));
+  vpx_codec_enc_config_default(vpx_codec_vp8_cx(), &cfg, 0);
+
+  cfg.g_w = width;
+  cfg.g_h = height;
+
+  vpx_img_wrap(&img, VPX_IMG_FMT_I420, width, height, 1, img_buf);
+
+  vpx_codec_enc_init(&enc, vpx_codec_vp8_cx(), &cfg, 0);
+
+  EXPECT_EQ(VPX_CODEC_OK, vpx_codec_encode(&enc, &img, 0, 1, 0, 0));
+
+  free(img_buf);
+
+  vpx_codec_destroy(&enc);
+}
+#endif
+
+// Set up 2 spatial streams with 2 temporal layers per stream, and generate
+// invalid configuration by setting the temporal layer rate allocation
+// (ts_target_bitrate[]) to 0 for both layers. This should fail independent of
+// CONFIG_MULTI_RES_ENCODING.
+TEST(EncodeAPI, MultiResEncode) {
+  static const vpx_codec_iface_t *kCodecs[] = {
+#if CONFIG_VP8_ENCODER
+    &vpx_codec_vp8_cx_algo,
+#endif
+#if CONFIG_VP9_ENCODER
+    &vpx_codec_vp9_cx_algo,
+#endif
+  };
+  const int width = 1280;
+  const int height = 720;
+  const int width_down = width / 2;
+  const int height_down = height / 2;
+  const int target_bitrate = 1000;
+  const int framerate = 30;
+
+  for (int c = 0; c < NELEMENTS(kCodecs); ++c) {
+    const vpx_codec_iface_t *const iface = kCodecs[c];
+    vpx_codec_ctx_t enc[2];
+    vpx_codec_enc_cfg_t cfg[2];
+    vpx_rational_t dsf[2] = { { 2, 1 }, { 2, 1 } };
+
+    memset(enc, 0, sizeof(enc));
+
+    for (int i = 0; i < 2; i++) {
+      vpx_codec_enc_config_default(iface, &cfg[i], 0);
+    }
+
+    /* Highest-resolution encoder settings */
+    cfg[0].g_w = width;
+    cfg[0].g_h = height;
+    cfg[0].rc_dropframe_thresh = 0;
+    cfg[0].rc_end_usage = VPX_CBR;
+    cfg[0].rc_resize_allowed = 0;
+    cfg[0].rc_min_quantizer = 2;
+    cfg[0].rc_max_quantizer = 56;
+    cfg[0].rc_undershoot_pct = 100;
+    cfg[0].rc_overshoot_pct = 15;
+    cfg[0].rc_buf_initial_sz = 500;
+    cfg[0].rc_buf_optimal_sz = 600;
+    cfg[0].rc_buf_sz = 1000;
+    cfg[0].g_error_resilient = 1; /* Enable error resilient mode */
+    cfg[0].g_lag_in_frames = 0;
+
+    cfg[0].kf_mode = VPX_KF_AUTO;
+    cfg[0].kf_min_dist = 3000;
+    cfg[0].kf_max_dist = 3000;
+
+    cfg[0].rc_target_bitrate = target_bitrate; /* Set target bitrate */
+    cfg[0].g_timebase.num = 1;                 /* Set fps */
+    cfg[0].g_timebase.den = framerate;
+
+    memcpy(&cfg[1], &cfg[0], sizeof(cfg[0]));
+    cfg[1].rc_target_bitrate = 500;
+    cfg[1].g_w = width_down;
+    cfg[1].g_h = height_down;
+
+    for (int i = 0; i < 2; i++) {
+      cfg[i].ts_number_layers = 2;
+      cfg[i].ts_periodicity = 2;
+      cfg[i].ts_rate_decimator[0] = 2;
+      cfg[i].ts_rate_decimator[1] = 1;
+      cfg[i].ts_layer_id[0] = 0;
+      cfg[i].ts_layer_id[1] = 1;
+      // Invalid parameters.
+      cfg[i].ts_target_bitrate[0] = 0;
+      cfg[i].ts_target_bitrate[1] = 0;
+    }
+
+    // VP9 should report incapable, VP8 invalid for all configurations.
+    const char kVP9Name[] = "WebM Project VP9";
+    const bool is_vp9 = strncmp(kVP9Name, vpx_codec_iface_name(iface),
+                                sizeof(kVP9Name) - 1) == 0;
+    EXPECT_EQ(is_vp9 ? VPX_CODEC_INCAPABLE : VPX_CODEC_INVALID_PARAM,
+              vpx_codec_enc_init_multi(&enc[0], iface, &cfg[0], 2, 0, &dsf[0]));
+
+    for (int i = 0; i < 2; i++) {
+      vpx_codec_destroy(&enc[i]);
+    }
+  }
+}
+
 }  // namespace
--- a/test/encode_test_driver.cc
+++ b/test/encode_test_driver.cc
@@ -201,6 +201,8 @@ void EncoderTest::RunLoop(VideoSource *video) {
      PreEncodeFrameHook(video, encoder.get());
      encoder->EncodeFrame(video, frame_flags_);

+      PostEncodeFrameHook(encoder.get());
+
      CxDataIterator iter = encoder->GetCxData();

      bool has_cxdata = false;
@@ -226,6 +228,8 @@ void EncoderTest::RunLoop(VideoSource *video) {

          case VPX_CODEC_PSNR_PKT: PSNRPktHook(pkt); break;

+          case VPX_CODEC_STATS_PKT: StatsPktHook(pkt); break;
+
          default: break;
        }
      }
--- a/test/encode_test_driver.h
+++ b/test/encode_test_driver.h
@@ -139,6 +139,13 @@ class Encoder {
  }
 #endif

+#if CONFIG_VP8_ENCODER
+  void Control(int ctrl_id, vpx_roi_map_t *arg) {
+    const vpx_codec_err_t res = vpx_codec_control_(&encoder_, ctrl_id, arg);
+    ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
+  }
+#endif
+
  void Config(const vpx_codec_enc_cfg_t *cfg) {
    const vpx_codec_err_t res = vpx_codec_enc_config_set(&encoder_, cfg);
    ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
@@ -212,12 +219,17 @@ class EncoderTest {
  virtual void PreEncodeFrameHook(VideoSource * /*video*/,
                                  Encoder * /*encoder*/) {}

+  virtual void PostEncodeFrameHook(Encoder * /*encoder*/) {}
+
  // Hook to be called on every compressed data packet.
  virtual void FramePktHook(const vpx_codec_cx_pkt_t * /*pkt*/) {}

  // Hook to be called on every PSNR packet.
  virtual void PSNRPktHook(const vpx_codec_cx_pkt_t * /*pkt*/) {}

+  // Hook to be called on every first pass stats packet.
+  virtual void StatsPktHook(const vpx_codec_cx_pkt_t * /*pkt*/) {}
+
  // Hook to determine whether the encode loop should continue.
  virtual bool Continue() const {
    return !(::testing::Test::HasFatalFailure() || abort_);
--- a/test/external_frame_buffer_test.cc
+++ b/test/external_frame_buffer_test.cc
@@ -34,7 +34,8 @@ struct ExternalFrameBuffer {
 // Class to manipulate a list of external frame buffers.
 class ExternalFrameBufferList {
 public:
-  ExternalFrameBufferList() : num_buffers_(0), ext_fb_list_(NULL) {}
+  ExternalFrameBufferList()
+      : num_buffers_(0), num_used_buffers_(0), ext_fb_list_(NULL) {}

  virtual ~ExternalFrameBufferList() {
    for (int i = 0; i < num_buffers_; ++i) {
@@ -71,6 +72,8 @@ class ExternalFrameBufferList {
    }

    SetFrameBuffer(idx, fb);
+
+    num_used_buffers_++;
    return 0;
  }

@@ -106,6 +109,7 @@ class ExternalFrameBufferList {
    }
    EXPECT_EQ(1, ext_fb->in_use);
    ext_fb->in_use = 0;
+    num_used_buffers_--;
    return 0;
  }

@@ -121,6 +125,8 @@ class ExternalFrameBufferList {
    }
  }

+  int num_used_buffers() const { return num_used_buffers_; }
+
 private:
  // Returns the index of the first free frame buffer. Returns |num_buffers_|
  // if there are no free frame buffers.
@@ -145,6 +151,7 @@ class ExternalFrameBufferList {
  }

  int num_buffers_;
+  int num_used_buffers_;
  ExternalFrameBuffer *ext_fb_list_;
 };

@@ -220,8 +227,8 @@ class ExternalFrameBufferMD5Test

  void OpenMD5File(const std::string &md5_file_name_) {
    md5_file_ = libvpx_test::OpenTestDataFile(md5_file_name_);
-    ASSERT_TRUE(md5_file_ != NULL) << "Md5 file open failed. Filename: "
-                                   << md5_file_name_;
+    ASSERT_TRUE(md5_file_ != NULL)
+        << "Md5 file open failed. Filename: " << md5_file_name_;
  }

  virtual void DecompressedFrameHook(const vpx_image_t &img,
@@ -273,6 +280,7 @@ class ExternalFrameBufferMD5Test

 #if CONFIG_WEBM_IO
 const char kVP9TestFile[] = "vp90-2-02-size-lf-1920x1080.webm";
+const char kVP9NonRefTestFile[] = "vp90-2-22-svc_1280x720_1.webm";

 // Class for testing passing in external frame buffers to libvpx.
 class ExternalFrameBufferTest : public ::testing::Test {
@@ -292,7 +300,9 @@ class ExternalFrameBufferTest : public ::testing::Test {

  virtual void TearDown() {
    delete decoder_;
+    decoder_ = NULL;
    delete video_;
+    video_ = NULL;
  }

  // Passes the external frame buffer information to libvpx.
@@ -325,7 +335,7 @@ class ExternalFrameBufferTest : public ::testing::Test {
    return VPX_CODEC_OK;
  }

- private:
+ protected:
  void CheckDecodedFrames() {
    libvpx_test::DxDataIterator dec_iter = decoder_->GetDxData();
    const vpx_image_t *img = NULL;
@@ -341,6 +351,25 @@ class ExternalFrameBufferTest : public ::testing::Test {
  int num_buffers_;
  ExternalFrameBufferList fb_list_;
 };
+
+class ExternalFrameBufferNonRefTest : public ExternalFrameBufferTest {
+ protected:
+  virtual void SetUp() {
+    video_ = new libvpx_test::WebMVideoSource(kVP9NonRefTestFile);
+    ASSERT_TRUE(video_ != NULL);
+    video_->Init();
+    video_->Begin();
+
+    vpx_codec_dec_cfg_t cfg = vpx_codec_dec_cfg_t();
+    decoder_ = new libvpx_test::VP9Decoder(cfg, 0);
+    ASSERT_TRUE(decoder_ != NULL);
+  }
+
+  virtual void CheckFrameBufferRelease() {
+    TearDown();
+    ASSERT_EQ(0, fb_list_.num_used_buffers());
+  }
+};
 #endif  // CONFIG_WEBM_IO

 // This test runs through the set of test vectors, and decodes them.
@@ -419,6 +448,8 @@ TEST_F(ExternalFrameBufferTest, NotEnoughBuffers) {
            SetFrameBufferFunctions(num_buffers, get_vp9_frame_buffer,
                                    release_vp9_frame_buffer));
  ASSERT_EQ(VPX_CODEC_OK, DecodeOneFrame());
+  // Only run this on long clips. Decoding a very short clip will return
+  // VPX_CODEC_OK even with only 2 buffers.
  ASSERT_EQ(VPX_CODEC_MEM_ERROR, DecodeRemainingFrames());
 }

@@ -467,6 +498,15 @@ TEST_F(ExternalFrameBufferTest, SetAfterDecode) {
            SetFrameBufferFunctions(num_buffers, get_vp9_frame_buffer,
                                    release_vp9_frame_buffer));
 }
+
+TEST_F(ExternalFrameBufferNonRefTest, ReleaseNonRefFrameBuffer) {
+  const int num_buffers = VP9_MAXIMUM_REF_BUFFERS + VPX_MAXIMUM_WORK_BUFFERS;
+  ASSERT_EQ(VPX_CODEC_OK,
+            SetFrameBufferFunctions(num_buffers, get_vp9_frame_buffer,
+                                    release_vp9_frame_buffer));
+  ASSERT_EQ(VPX_CODEC_OK, DecodeRemainingFrames());
+  CheckFrameBufferRelease();
+}
 #endif  // CONFIG_WEBM_IO

 VP9_INSTANTIATE_TEST_CASE(
--- a/test/fdct4x4_test.cc
+++ b/test/fdct4x4_test.cc
@@ -1,511 +0,0 @@
-/*
- *  Copyright (c) 2012 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 <math.h>
-#include <stdlib.h>
-#include <string.h>
-
-#include "third_party/googletest/src/include/gtest/gtest.h"
-
-#include "./vp9_rtcd.h"
-#include "./vpx_dsp_rtcd.h"
-#include "test/acm_random.h"
-#include "test/clear_system_state.h"
-#include "test/register_state_check.h"
-#include "test/util.h"
-#include "vp9/common/vp9_entropy.h"
-#include "vpx/vpx_codec.h"
-#include "vpx/vpx_integer.h"
-#include "vpx_ports/mem.h"
-
-using libvpx_test::ACMRandom;
-
-namespace {
-const int kNumCoeffs = 16;
-typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
-typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
-typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
-                        int tx_type);
-typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
-                        int tx_type);
-
-typedef std::tr1::tuple<FdctFunc, IdctFunc, int, vpx_bit_depth_t> Dct4x4Param;
-typedef std::tr1::tuple<FhtFunc, IhtFunc, int, vpx_bit_depth_t> Ht4x4Param;
-
-void fdct4x4_ref(const int16_t *in, tran_low_t *out, int stride,
-                 int /*tx_type*/) {
-  vpx_fdct4x4_c(in, out, stride);
-}
-
-void fht4x4_ref(const int16_t *in, tran_low_t *out, int stride, int tx_type) {
-  vp9_fht4x4_c(in, out, stride, tx_type);
-}
-
-void fwht4x4_ref(const int16_t *in, tran_low_t *out, int stride,
-                 int /*tx_type*/) {
-  vp9_fwht4x4_c(in, out, stride);
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-void idct4x4_10(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct4x4_16_add_c(in, out, stride, 10);
-}
-
-void idct4x4_12(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct4x4_16_add_c(in, out, stride, 12);
-}
-
-void iht4x4_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
-  vp9_highbd_iht4x4_16_add_c(in, out, stride, tx_type, 10);
-}
-
-void iht4x4_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
-  vp9_highbd_iht4x4_16_add_c(in, out, stride, tx_type, 12);
-}
-
-void iwht4x4_10(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_iwht4x4_16_add_c(in, out, stride, 10);
-}
-
-void iwht4x4_12(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_iwht4x4_16_add_c(in, out, stride, 12);
-}
-
-#if HAVE_SSE2
-void idct4x4_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct4x4_16_add_sse2(in, out, stride, 10);
-}
-
-void idct4x4_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct4x4_16_add_sse2(in, out, stride, 12);
-}
-#endif  // HAVE_SSE2
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-class Trans4x4TestBase {
- public:
-  virtual ~Trans4x4TestBase() {}
-
- protected:
-  virtual void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) = 0;
-
-  virtual void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) = 0;
-
-  void RunAccuracyCheck(int limit) {
-    ACMRandom rnd(ACMRandom::DeterministicSeed());
-    uint32_t max_error = 0;
-    int64_t total_error = 0;
-    const int count_test_block = 10000;
-    for (int i = 0; i < count_test_block; ++i) {
-      DECLARE_ALIGNED(16, int16_t, test_input_block[kNumCoeffs]);
-      DECLARE_ALIGNED(16, tran_low_t, test_temp_block[kNumCoeffs]);
-      DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
-      DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
-#if CONFIG_VP9_HIGHBITDEPTH
-      DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
-      DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
-#endif
-
-      // Initialize a test block with input range [-255, 255].
-      for (int j = 0; j < kNumCoeffs; ++j) {
-        if (bit_depth_ == VPX_BITS_8) {
-          src[j] = rnd.Rand8();
-          dst[j] = rnd.Rand8();
-          test_input_block[j] = src[j] - dst[j];
-#if CONFIG_VP9_HIGHBITDEPTH
-        } else {
-          src16[j] = rnd.Rand16() & mask_;
-          dst16[j] = rnd.Rand16() & mask_;
-          test_input_block[j] = src16[j] - dst16[j];
-#endif
-        }
-      }
-
-      ASM_REGISTER_STATE_CHECK(
-          RunFwdTxfm(test_input_block, test_temp_block, pitch_));
-      if (bit_depth_ == VPX_BITS_8) {
-        ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
-#if CONFIG_VP9_HIGHBITDEPTH
-      } else {
-        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
-#endif
-      }
-
-      for (int j = 0; j < kNumCoeffs; ++j) {
-#if CONFIG_VP9_HIGHBITDEPTH
-        const int diff =
-            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
-#else
-        ASSERT_EQ(VPX_BITS_8, bit_depth_);
-        const int diff = dst[j] - src[j];
-#endif
-        const uint32_t error = diff * diff;
-        if (max_error < error) max_error = error;
-        total_error += error;
-      }
-    }
-
-    EXPECT_GE(static_cast<uint32_t>(limit), max_error)
-        << "Error: 4x4 FHT/IHT has an individual round trip error > " << limit;
-
-    EXPECT_GE(count_test_block * limit, total_error)
-        << "Error: 4x4 FHT/IHT has average round trip error > " << limit
-        << " per block";
-  }
-
-  void RunCoeffCheck() {
-    ACMRandom rnd(ACMRandom::DeterministicSeed());
-    const int count_test_block = 5000;
-    DECLARE_ALIGNED(16, int16_t, input_block[kNumCoeffs]);
-    DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
-    DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
-
-    for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-mask_, mask_].
-      for (int j = 0; j < kNumCoeffs; ++j) {
-        input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
-      }
-
-      fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
-      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
-
-      // The minimum quant value is 4.
-      for (int j = 0; j < kNumCoeffs; ++j)
-        EXPECT_EQ(output_block[j], output_ref_block[j]);
-    }
-  }
-
-  void RunMemCheck() {
-    ACMRandom rnd(ACMRandom::DeterministicSeed());
-    const int count_test_block = 5000;
-    DECLARE_ALIGNED(16, int16_t, input_extreme_block[kNumCoeffs]);
-    DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
-    DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
-
-    for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-mask_, mask_].
-      for (int j = 0; j < kNumCoeffs; ++j) {
-        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
-      }
-      if (i == 0) {
-        for (int j = 0; j < kNumCoeffs; ++j) input_extreme_block[j] = mask_;
-      } else if (i == 1) {
-        for (int j = 0; j < kNumCoeffs; ++j) input_extreme_block[j] = -mask_;
-      }
-
-      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
-      ASM_REGISTER_STATE_CHECK(
-          RunFwdTxfm(input_extreme_block, output_block, pitch_));
-
-      // The minimum quant value is 4.
-      for (int j = 0; j < kNumCoeffs; ++j) {
-        EXPECT_EQ(output_block[j], output_ref_block[j]);
-        EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
-            << "Error: 4x4 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
-      }
-    }
-  }
-
-  void RunInvAccuracyCheck(int limit) {
-    ACMRandom rnd(ACMRandom::DeterministicSeed());
-    const int count_test_block = 1000;
-    DECLARE_ALIGNED(16, int16_t, in[kNumCoeffs]);
-    DECLARE_ALIGNED(16, tran_low_t, coeff[kNumCoeffs]);
-    DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
-    DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
-#if CONFIG_VP9_HIGHBITDEPTH
-    DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
-    DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
-#endif
-
-    for (int i = 0; i < count_test_block; ++i) {
-      // Initialize a test block with input range [-mask_, mask_].
-      for (int j = 0; j < kNumCoeffs; ++j) {
-        if (bit_depth_ == VPX_BITS_8) {
-          src[j] = rnd.Rand8();
-          dst[j] = rnd.Rand8();
-          in[j] = src[j] - dst[j];
-#if CONFIG_VP9_HIGHBITDEPTH
-        } else {
-          src16[j] = rnd.Rand16() & mask_;
-          dst16[j] = rnd.Rand16() & mask_;
-          in[j] = src16[j] - dst16[j];
-#endif
-        }
-      }
-
-      fwd_txfm_ref(in, coeff, pitch_, tx_type_);
-
-      if (bit_depth_ == VPX_BITS_8) {
-        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
-#if CONFIG_VP9_HIGHBITDEPTH
-      } else {
-        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16), pitch_));
-#endif
-      }
-
-      for (int j = 0; j < kNumCoeffs; ++j) {
-#if CONFIG_VP9_HIGHBITDEPTH
-        const int diff =
-            bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
-#else
-        const int diff = dst[j] - src[j];
-#endif
-        const uint32_t error = diff * diff;
-        EXPECT_GE(static_cast<uint32_t>(limit), error)
-            << "Error: 4x4 IDCT has error " << error << " at index " << j;
-      }
-    }
-  }
-
-  int pitch_;
-  int tx_type_;
-  FhtFunc fwd_txfm_ref;
-  vpx_bit_depth_t bit_depth_;
-  int mask_;
-};
-
-class Trans4x4DCT : public Trans4x4TestBase,
-                    public ::testing::TestWithParam<Dct4x4Param> {
- public:
-  virtual ~Trans4x4DCT() {}
-
-  virtual void SetUp() {
-    fwd_txfm_ = GET_PARAM(0);
-    inv_txfm_ = GET_PARAM(1);
-    tx_type_ = GET_PARAM(2);
-    pitch_ = 4;
-    fwd_txfm_ref = fdct4x4_ref;
-    bit_depth_ = GET_PARAM(3);
-    mask_ = (1 << bit_depth_) - 1;
-  }
-  virtual void TearDown() { libvpx_test::ClearSystemState(); }
-
- protected:
-  void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) {
-    fwd_txfm_(in, out, stride);
-  }
-  void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) {
-    inv_txfm_(out, dst, stride);
-  }
-
-  FdctFunc fwd_txfm_;
-  IdctFunc inv_txfm_;
-};
-
-TEST_P(Trans4x4DCT, AccuracyCheck) { RunAccuracyCheck(1); }
-
-TEST_P(Trans4x4DCT, CoeffCheck) { RunCoeffCheck(); }
-
-TEST_P(Trans4x4DCT, MemCheck) { RunMemCheck(); }
-
-TEST_P(Trans4x4DCT, InvAccuracyCheck) { RunInvAccuracyCheck(1); }
-
-class Trans4x4HT : public Trans4x4TestBase,
-                   public ::testing::TestWithParam<Ht4x4Param> {
- public:
-  virtual ~Trans4x4HT() {}
-
-  virtual void SetUp() {
-    fwd_txfm_ = GET_PARAM(0);
-    inv_txfm_ = GET_PARAM(1);
-    tx_type_ = GET_PARAM(2);
-    pitch_ = 4;
-    fwd_txfm_ref = fht4x4_ref;
-    bit_depth_ = GET_PARAM(3);
-    mask_ = (1 << bit_depth_) - 1;
-  }
-  virtual void TearDown() { libvpx_test::ClearSystemState(); }
-
- protected:
-  void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) {
-    fwd_txfm_(in, out, stride, tx_type_);
-  }
-
-  void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) {
-    inv_txfm_(out, dst, stride, tx_type_);
-  }
-
-  FhtFunc fwd_txfm_;
-  IhtFunc inv_txfm_;
-};
-
-TEST_P(Trans4x4HT, AccuracyCheck) { RunAccuracyCheck(1); }
-
-TEST_P(Trans4x4HT, CoeffCheck) { RunCoeffCheck(); }
-
-TEST_P(Trans4x4HT, MemCheck) { RunMemCheck(); }
-
-TEST_P(Trans4x4HT, InvAccuracyCheck) { RunInvAccuracyCheck(1); }
-
-class Trans4x4WHT : public Trans4x4TestBase,
-                    public ::testing::TestWithParam<Dct4x4Param> {
- public:
-  virtual ~Trans4x4WHT() {}
-
-  virtual void SetUp() {
-    fwd_txfm_ = GET_PARAM(0);
-    inv_txfm_ = GET_PARAM(1);
-    tx_type_ = GET_PARAM(2);
-    pitch_ = 4;
-    fwd_txfm_ref = fwht4x4_ref;
-    bit_depth_ = GET_PARAM(3);
-    mask_ = (1 << bit_depth_) - 1;
-  }
-  virtual void TearDown() { libvpx_test::ClearSystemState(); }
-
- protected:
-  void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) {
-    fwd_txfm_(in, out, stride);
-  }
-  void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) {
-    inv_txfm_(out, dst, stride);
-  }
-
-  FdctFunc fwd_txfm_;
-  IdctFunc inv_txfm_;
-};
-
-TEST_P(Trans4x4WHT, AccuracyCheck) { RunAccuracyCheck(0); }
-
-TEST_P(Trans4x4WHT, CoeffCheck) { RunCoeffCheck(); }
-
-TEST_P(Trans4x4WHT, MemCheck) { RunMemCheck(); }
-
-TEST_P(Trans4x4WHT, InvAccuracyCheck) { RunInvAccuracyCheck(0); }
-using std::tr1::make_tuple;
-
-#if CONFIG_VP9_HIGHBITDEPTH
-INSTANTIATE_TEST_CASE_P(
-    C, Trans4x4DCT,
-    ::testing::Values(
-        make_tuple(&vpx_highbd_fdct4x4_c, &idct4x4_10, 0, VPX_BITS_10),
-        make_tuple(&vpx_highbd_fdct4x4_c, &idct4x4_12, 0, VPX_BITS_12),
-        make_tuple(&vpx_fdct4x4_c, &vpx_idct4x4_16_add_c, 0, VPX_BITS_8)));
-#else
-INSTANTIATE_TEST_CASE_P(C, Trans4x4DCT,
-                        ::testing::Values(make_tuple(&vpx_fdct4x4_c,
-                                                     &vpx_idct4x4_16_add_c, 0,
-                                                     VPX_BITS_8)));
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-#if CONFIG_VP9_HIGHBITDEPTH
-INSTANTIATE_TEST_CASE_P(
-    C, Trans4x4HT,
-    ::testing::Values(
-        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_10, 0, VPX_BITS_10),
-        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_10, 1, VPX_BITS_10),
-        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_10, 2, VPX_BITS_10),
-        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_10, 3, VPX_BITS_10),
-        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_12, 0, VPX_BITS_12),
-        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_12, 1, VPX_BITS_12),
-        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_12, 2, VPX_BITS_12),
-        make_tuple(&vp9_highbd_fht4x4_c, &iht4x4_12, 3, VPX_BITS_12),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 0, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 1, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 2, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 3, VPX_BITS_8)));
-#else
-INSTANTIATE_TEST_CASE_P(
-    C, Trans4x4HT,
-    ::testing::Values(
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 0, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 1, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 2, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_c, 3, VPX_BITS_8)));
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-#if CONFIG_VP9_HIGHBITDEPTH
-INSTANTIATE_TEST_CASE_P(
-    C, Trans4x4WHT,
-    ::testing::Values(
-        make_tuple(&vp9_highbd_fwht4x4_c, &iwht4x4_10, 0, VPX_BITS_10),
-        make_tuple(&vp9_highbd_fwht4x4_c, &iwht4x4_12, 0, VPX_BITS_12),
-        make_tuple(&vp9_fwht4x4_c, &vpx_iwht4x4_16_add_c, 0, VPX_BITS_8)));
-#else
-INSTANTIATE_TEST_CASE_P(C, Trans4x4WHT,
-                        ::testing::Values(make_tuple(&vp9_fwht4x4_c,
-                                                     &vpx_iwht4x4_16_add_c, 0,
-                                                     VPX_BITS_8)));
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
-#if HAVE_NEON && !CONFIG_EMULATE_HARDWARE
-INSTANTIATE_TEST_CASE_P(NEON, Trans4x4DCT,
-                        ::testing::Values(make_tuple(&vpx_fdct4x4_c,
-                                                     &vpx_idct4x4_16_add_neon,
-                                                     0, VPX_BITS_8)));
-#if !CONFIG_VP9_HIGHBITDEPTH
-INSTANTIATE_TEST_CASE_P(
-    NEON, Trans4x4HT,
-    ::testing::Values(
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 0, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 1, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 2, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_c, &vp9_iht4x4_16_add_neon, 3, VPX_BITS_8)));
-#endif  // !CONFIG_VP9_HIGHBITDEPTH
-#endif  // HAVE_NEON && !CONFIG_EMULATE_HARDWARE
-
-#if HAVE_SSE2 && !CONFIG_EMULATE_HARDWARE
-INSTANTIATE_TEST_CASE_P(
-    SSE2, Trans4x4WHT,
-    ::testing::Values(
-        make_tuple(&vp9_fwht4x4_sse2, &vpx_iwht4x4_16_add_c, 0, VPX_BITS_8),
-        make_tuple(&vp9_fwht4x4_c, &vpx_iwht4x4_16_add_sse2, 0, VPX_BITS_8)));
-#endif
-
-#if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
-INSTANTIATE_TEST_CASE_P(SSE2, Trans4x4DCT,
-                        ::testing::Values(make_tuple(&vpx_fdct4x4_sse2,
-                                                     &vpx_idct4x4_16_add_sse2,
-                                                     0, VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(
-    SSE2, Trans4x4HT,
-    ::testing::Values(
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 0, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 1, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 2, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_sse2, 3, VPX_BITS_8)));
-#endif  // HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
-
-#if HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
-INSTANTIATE_TEST_CASE_P(
-    SSE2, Trans4x4DCT,
-    ::testing::Values(
-        make_tuple(&vpx_highbd_fdct4x4_c, &idct4x4_10_sse2, 0, VPX_BITS_10),
-        make_tuple(&vpx_highbd_fdct4x4_sse2, &idct4x4_10_sse2, 0, VPX_BITS_10),
-        make_tuple(&vpx_highbd_fdct4x4_c, &idct4x4_12_sse2, 0, VPX_BITS_12),
-        make_tuple(&vpx_highbd_fdct4x4_sse2, &idct4x4_12_sse2, 0, VPX_BITS_12),
-        make_tuple(&vpx_fdct4x4_sse2, &vpx_idct4x4_16_add_c, 0, VPX_BITS_8)));
-
-INSTANTIATE_TEST_CASE_P(
-    SSE2, Trans4x4HT,
-    ::testing::Values(
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_c, 0, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_c, 1, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_c, 2, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_c, 3, VPX_BITS_8)));
-#endif  // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
-
-#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
-INSTANTIATE_TEST_CASE_P(MSA, Trans4x4DCT,
-                        ::testing::Values(make_tuple(&vpx_fdct4x4_msa,
-                                                     &vpx_idct4x4_16_add_msa, 0,
-                                                     VPX_BITS_8)));
-INSTANTIATE_TEST_CASE_P(
-    MSA, Trans4x4HT,
-    ::testing::Values(
-        make_tuple(&vp9_fht4x4_msa, &vp9_iht4x4_16_add_msa, 0, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_msa, &vp9_iht4x4_16_add_msa, 1, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_msa, &vp9_iht4x4_16_add_msa, 2, VPX_BITS_8),
-        make_tuple(&vp9_fht4x4_msa, &vp9_iht4x4_16_add_msa, 3, VPX_BITS_8)));
-#endif  // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
-}  // namespace
--- a/test/fdct8x8_test.cc
+++ b/test/fdct8x8_test.cc
@@ -88,45 +88,45 @@ void fht8x8_ref(const int16_t *in, tran_low_t *out, int stride, int tx_type) {

 #if CONFIG_VP9_HIGHBITDEPTH
 void idct8x8_10(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct8x8_64_add_c(in, out, stride, 10);
+  vpx_highbd_idct8x8_64_add_c(in, CAST_TO_SHORTPTR(out), stride, 10);
 }

 void idct8x8_12(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct8x8_64_add_c(in, out, stride, 12);
+  vpx_highbd_idct8x8_64_add_c(in, CAST_TO_SHORTPTR(out), stride, 12);
 }

 void iht8x8_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
-  vp9_highbd_iht8x8_64_add_c(in, out, stride, tx_type, 10);
+  vp9_highbd_iht8x8_64_add_c(in, CAST_TO_SHORTPTR(out), stride, tx_type, 10);
 }

 void iht8x8_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
-  vp9_highbd_iht8x8_64_add_c(in, out, stride, tx_type, 12);
+  vp9_highbd_iht8x8_64_add_c(in, CAST_TO_SHORTPTR(out), stride, tx_type, 12);
 }

 #if HAVE_SSE2

 void idct8x8_12_add_10_c(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct8x8_12_add_c(in, out, stride, 10);
+  vpx_highbd_idct8x8_12_add_c(in, CAST_TO_SHORTPTR(out), stride, 10);
 }

 void idct8x8_12_add_12_c(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct8x8_12_add_c(in, out, stride, 12);
+  vpx_highbd_idct8x8_12_add_c(in, CAST_TO_SHORTPTR(out), stride, 12);
 }

 void idct8x8_12_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct8x8_12_add_sse2(in, out, stride, 10);
+  vpx_highbd_idct8x8_12_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 10);
 }

 void idct8x8_12_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct8x8_12_add_sse2(in, out, stride, 12);
+  vpx_highbd_idct8x8_12_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 12);
 }

 void idct8x8_64_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct8x8_64_add_sse2(in, out, stride, 10);
+  vpx_highbd_idct8x8_64_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 10);
 }

 void idct8x8_64_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
-  vpx_highbd_idct8x8_64_add_sse2(in, out, stride, 12);
+  vpx_highbd_idct8x8_64_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 12);
 }
 #endif  // HAVE_SSE2
 #endif  // CONFIG_VP9_HIGHBITDEPTH
@@ -257,7 +257,7 @@ class FwdTrans8x8TestBase {
 #if CONFIG_VP9_HIGHBITDEPTH
      } else {
        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
+            RunInvTxfm(test_temp_block, CAST_TO_BYTEPTR(dst16), pitch_));
 #endif
      }

@@ -340,7 +340,7 @@ class FwdTrans8x8TestBase {
 #if CONFIG_VP9_HIGHBITDEPTH
      } else {
        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
+            RunInvTxfm(test_temp_block, CAST_TO_BYTEPTR(dst16), pitch_));
 #endif
      }

@@ -413,7 +413,7 @@ class FwdTrans8x8TestBase {
 #if CONFIG_VP9_HIGHBITDEPTH
      } else {
        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16), pitch_));
+            RunInvTxfm(coeff, CAST_TO_BYTEPTR(dst16), pitch_));
 #endif
      }

@@ -497,9 +497,9 @@ class FwdTrans8x8TestBase {
        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
 #if CONFIG_VP9_HIGHBITDEPTH
      } else {
-        ref_txfm(coeff, CONVERT_TO_BYTEPTR(ref16), pitch_);
+        ref_txfm(coeff, CAST_TO_BYTEPTR(ref16), pitch_);
        ASM_REGISTER_STATE_CHECK(
-            RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16), pitch_));
+            RunInvTxfm(coeff, CAST_TO_BYTEPTR(dst16), pitch_));
 #endif
      }

@@ -511,8 +511,8 @@ class FwdTrans8x8TestBase {
        const int diff = dst[j] - ref[j];
 #endif
        const uint32_t error = diff * diff;
-        EXPECT_EQ(0u, error) << "Error: 8x8 IDCT has error " << error
-                             << " at index " << j;
+        EXPECT_EQ(0u, error)
+            << "Error: 8x8 IDCT has error " << error << " at index " << j;
      }
    }
  }
@@ -671,16 +671,11 @@ INSTANTIATE_TEST_CASE_P(
 #endif  // CONFIG_VP9_HIGHBITDEPTH

 #if HAVE_NEON && !CONFIG_EMULATE_HARDWARE
-#if CONFIG_VP9_HIGHBITDEPTH
-INSTANTIATE_TEST_CASE_P(NEON, FwdTrans8x8DCT,
-                        ::testing::Values(make_tuple(&vpx_fdct8x8_c,
-                                                     &vpx_idct8x8_64_add_neon,
-                                                     0, VPX_BITS_8)));
-#else   // !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(NEON, FwdTrans8x8DCT,
                        ::testing::Values(make_tuple(&vpx_fdct8x8_neon,
                                                     &vpx_idct8x8_64_add_neon,
                                                     0, VPX_BITS_8)));
+#if !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(
    NEON, FwdTrans8x8HT,
    ::testing::Values(
@@ -688,8 +683,8 @@ INSTANTIATE_TEST_CASE_P(
        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 1, VPX_BITS_8),
        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 2, VPX_BITS_8),
        make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_neon, 3, VPX_BITS_8)));
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-#endif  // HAVE_NEON && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+#endif  // !CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_NEON && !CONFIG_EMULATE_HARDWARE

 #if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 INSTANTIATE_TEST_CASE_P(SSE2, FwdTrans8x8DCT,
@@ -744,7 +739,7 @@ INSTANTIATE_TEST_CASE_P(
    !CONFIG_EMULATE_HARDWARE
 INSTANTIATE_TEST_CASE_P(SSSE3, FwdTrans8x8DCT,
                        ::testing::Values(make_tuple(&vpx_fdct8x8_ssse3,
-                                                     &vpx_idct8x8_64_add_ssse3,
+                                                     &vpx_idct8x8_64_add_sse2,
                                                     0, VPX_BITS_8)));
 #endif

@@ -761,4 +756,11 @@ INSTANTIATE_TEST_CASE_P(
        make_tuple(&vp9_fht8x8_msa, &vp9_iht8x8_64_add_msa, 2, VPX_BITS_8),
        make_tuple(&vp9_fht8x8_msa, &vp9_iht8x8_64_add_msa, 3, VPX_BITS_8)));
 #endif  // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+
+#if HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
+INSTANTIATE_TEST_CASE_P(VSX, FwdTrans8x8DCT,
+                        ::testing::Values(make_tuple(&vpx_fdct8x8_c,
+                                                     &vpx_idct8x8_64_add_vsx, 0,
+                                                     VPX_BITS_8)));
+#endif  // HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 }  // namespace
--- a/test/hadamard_test.cc
+++ b/test/hadamard_test.cc
@@ -13,6 +13,7 @@
 #include "third_party/googletest/src/include/gtest/gtest.h"

 #include "./vpx_dsp_rtcd.h"
+#include "vpx_ports/vpx_timer.h"

 #include "test/acm_random.h"
 #include "test/register_state_check.h"
@@ -21,7 +22,8 @@ namespace {

 using ::libvpx_test::ACMRandom;

-typedef void (*HadamardFunc)(const int16_t *a, int a_stride, int16_t *b);
+typedef void (*HadamardFunc)(const int16_t *a, ptrdiff_t a_stride,
+                             tran_low_t *b);

 void hadamard_loop(const int16_t *a, int a_stride, int16_t *out) {
  int16_t b[8];
@@ -46,18 +48,16 @@ void hadamard_loop(const int16_t *a, int a_stride, int16_t *out) {
  out[5] = c[3] - c[7];
 }

-void reference_hadamard8x8(const int16_t *a, int a_stride, int16_t *b) {
+void reference_hadamard8x8(const int16_t *a, int a_stride, tran_low_t *b) {
  int16_t buf[64];
-  for (int i = 0; i < 8; ++i) {
-    hadamard_loop(a + i, a_stride, buf + i * 8);
-  }
+  int16_t buf2[64];
+  for (int i = 0; i < 8; ++i) hadamard_loop(a + i, a_stride, buf + i * 8);
+  for (int i = 0; i < 8; ++i) hadamard_loop(buf + i, 8, buf2 + i * 8);

-  for (int i = 0; i < 8; ++i) {
-    hadamard_loop(buf + i, 8, b + i * 8);
-  }
+  for (int i = 0; i < 64; ++i) b[i] = (tran_low_t)buf2[i];
 }

-void reference_hadamard16x16(const int16_t *a, int a_stride, int16_t *b) {
+void reference_hadamard16x16(const int16_t *a, int a_stride, tran_low_t *b) {
  /* The source is a 16x16 block. The destination is rearranged to 8x32.
   * Input is 9 bit. */
  reference_hadamard8x8(a + 0 + 0 * a_stride, a_stride, b + 0);
@@ -68,16 +68,16 @@ void reference_hadamard16x16(const int16_t *a, int a_stride, int16_t *b) {
  /* Overlay the 8x8 blocks and combine. */
  for (int i = 0; i < 64; ++i) {
    /* 8x8 steps the range up to 15 bits. */
-    const int16_t a0 = b[0];
-    const int16_t a1 = b[64];
-    const int16_t a2 = b[128];
-    const int16_t a3 = b[192];
+    const tran_low_t a0 = b[0];
+    const tran_low_t a1 = b[64];
+    const tran_low_t a2 = b[128];
+    const tran_low_t a3 = b[192];

    /* Prevent the result from escaping int16_t. */
-    const int16_t b0 = (a0 + a1) >> 1;
-    const int16_t b1 = (a0 - a1) >> 1;
-    const int16_t b2 = (a2 + a3) >> 1;
-    const int16_t b3 = (a2 - a3) >> 1;
+    const tran_low_t b0 = (a0 + a1) >> 1;
+    const tran_low_t b1 = (a0 - a1) >> 1;
+    const tran_low_t b2 = (a2 + a3) >> 1;
+    const tran_low_t b3 = (a2 - a3) >> 1;

    /* Store a 16 bit value. */
    b[0] = b0 + b2;
@@ -101,12 +101,35 @@ class HadamardTestBase : public ::testing::TestWithParam<HadamardFunc> {
  ACMRandom rnd_;
 };

+void HadamardSpeedTest(const char *name, HadamardFunc const func,
+                       const int16_t *input, int stride, tran_low_t *output,
+                       int times) {
+  int i;
+  vpx_usec_timer timer;
+
+  vpx_usec_timer_start(&timer);
+  for (i = 0; i < times; ++i) {
+    func(input, stride, output);
+  }
+  vpx_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+  printf("%s[%12d runs]: %d us\n", name, times, elapsed_time);
+}
+
 class Hadamard8x8Test : public HadamardTestBase {};

+void HadamardSpeedTest8x8(HadamardFunc const func, int times) {
+  DECLARE_ALIGNED(16, int16_t, input[64]);
+  DECLARE_ALIGNED(16, tran_low_t, output[64]);
+  memset(input, 1, sizeof(input));
+  HadamardSpeedTest("Hadamard8x8", func, input, 8, output, times);
+}
+
 TEST_P(Hadamard8x8Test, CompareReferenceRandom) {
  DECLARE_ALIGNED(16, int16_t, a[64]);
-  DECLARE_ALIGNED(16, int16_t, b[64]);
-  int16_t b_ref[64];
+  DECLARE_ALIGNED(16, tran_low_t, b[64]);
+  tran_low_t b_ref[64];
  for (int i = 0; i < 64; ++i) {
    a[i] = rnd_.Rand9Signed();
  }
@@ -124,8 +147,8 @@ TEST_P(Hadamard8x8Test, CompareReferenceRandom) {

 TEST_P(Hadamard8x8Test, VaryStride) {
  DECLARE_ALIGNED(16, int16_t, a[64 * 8]);
-  DECLARE_ALIGNED(16, int16_t, b[64]);
-  int16_t b_ref[64];
+  DECLARE_ALIGNED(16, tran_low_t, b[64]);
+  tran_low_t b_ref[64];
  for (int i = 0; i < 64 * 8; ++i) {
    a[i] = rnd_.Rand9Signed();
  }
@@ -144,6 +167,12 @@ TEST_P(Hadamard8x8Test, VaryStride) {
  }
 }

+TEST_P(Hadamard8x8Test, DISABLED_Speed) {
+  HadamardSpeedTest8x8(h_func_, 10);
+  HadamardSpeedTest8x8(h_func_, 10000);
+  HadamardSpeedTest8x8(h_func_, 10000000);
+}
+
 INSTANTIATE_TEST_CASE_P(C, Hadamard8x8Test,
                        ::testing::Values(&vpx_hadamard_8x8_c));

@@ -162,12 +191,33 @@ INSTANTIATE_TEST_CASE_P(NEON, Hadamard8x8Test,
                        ::testing::Values(&vpx_hadamard_8x8_neon));
 #endif  // HAVE_NEON

+// TODO(jingning): Remove highbitdepth flag when the SIMD functions are
+// in place and turn on the unit test.
+#if !CONFIG_VP9_HIGHBITDEPTH
+#if HAVE_MSA
+INSTANTIATE_TEST_CASE_P(MSA, Hadamard8x8Test,
+                        ::testing::Values(&vpx_hadamard_8x8_msa));
+#endif  // HAVE_MSA
+#endif  // !CONFIG_VP9_HIGHBITDEPTH
+
+#if HAVE_VSX
+INSTANTIATE_TEST_CASE_P(VSX, Hadamard8x8Test,
+                        ::testing::Values(&vpx_hadamard_8x8_vsx));
+#endif  // HAVE_VSX
+
 class Hadamard16x16Test : public HadamardTestBase {};

+void HadamardSpeedTest16x16(HadamardFunc const func, int times) {
+  DECLARE_ALIGNED(16, int16_t, input[256]);
+  DECLARE_ALIGNED(16, tran_low_t, output[256]);
+  memset(input, 1, sizeof(input));
+  HadamardSpeedTest("Hadamard16x16", func, input, 16, output, times);
+}
+
 TEST_P(Hadamard16x16Test, CompareReferenceRandom) {
  DECLARE_ALIGNED(16, int16_t, a[16 * 16]);
-  DECLARE_ALIGNED(16, int16_t, b[16 * 16]);
-  int16_t b_ref[16 * 16];
+  DECLARE_ALIGNED(16, tran_low_t, b[16 * 16]);
+  tran_low_t b_ref[16 * 16];
  for (int i = 0; i < 16 * 16; ++i) {
    a[i] = rnd_.Rand9Signed();
  }
@@ -185,8 +235,8 @@ TEST_P(Hadamard16x16Test, CompareReferenceRandom) {

 TEST_P(Hadamard16x16Test, VaryStride) {
  DECLARE_ALIGNED(16, int16_t, a[16 * 16 * 8]);
-  DECLARE_ALIGNED(16, int16_t, b[16 * 16]);
-  int16_t b_ref[16 * 16];
+  DECLARE_ALIGNED(16, tran_low_t, b[16 * 16]);
+  tran_low_t b_ref[16 * 16];
  for (int i = 0; i < 16 * 16 * 8; ++i) {
    a[i] = rnd_.Rand9Signed();
  }
@@ -205,6 +255,12 @@ TEST_P(Hadamard16x16Test, VaryStride) {
  }
 }

+TEST_P(Hadamard16x16Test, DISABLED_Speed) {
+  HadamardSpeedTest16x16(h_func_, 10);
+  HadamardSpeedTest16x16(h_func_, 10000);
+  HadamardSpeedTest16x16(h_func_, 10000000);
+}
+
 INSTANTIATE_TEST_CASE_P(C, Hadamard16x16Test,
                        ::testing::Values(&vpx_hadamard_16x16_c));

@@ -213,8 +269,25 @@ INSTANTIATE_TEST_CASE_P(SSE2, Hadamard16x16Test,
                        ::testing::Values(&vpx_hadamard_16x16_sse2));
 #endif  // HAVE_SSE2

+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, Hadamard16x16Test,
+                        ::testing::Values(&vpx_hadamard_16x16_avx2));
+#endif  // HAVE_AVX2
+
+#if HAVE_VSX
+INSTANTIATE_TEST_CASE_P(VSX, Hadamard16x16Test,
+                        ::testing::Values(&vpx_hadamard_16x16_vsx));
+#endif  // HAVE_VSX
+
 #if HAVE_NEON
 INSTANTIATE_TEST_CASE_P(NEON, Hadamard16x16Test,
                        ::testing::Values(&vpx_hadamard_16x16_neon));
 #endif  // HAVE_NEON
+
+#if !CONFIG_VP9_HIGHBITDEPTH
+#if HAVE_MSA
+INSTANTIATE_TEST_CASE_P(MSA, Hadamard16x16Test,
+                        ::testing::Values(&vpx_hadamard_16x16_msa));
+#endif  // HAVE_MSA
+#endif  // !CONFIG_VP9_HIGHBITDEPTH
 }  // namespace
--- a/test/idct_test.cc
+++ b/test/idct_test.cc
@@ -13,6 +13,7 @@

 #include "third_party/googletest/src/include/gtest/gtest.h"

+#include "test/buffer.h"
 #include "test/clear_system_state.h"
 #include "test/register_state_check.h"
 #include "vpx/vpx_integer.h"
@@ -21,110 +22,156 @@ typedef void (*IdctFunc)(int16_t *input, unsigned char *pred_ptr,
                         int pred_stride, unsigned char *dst_ptr,
                         int dst_stride);
 namespace {
+
+using libvpx_test::Buffer;
+
 class IDCTTest : public ::testing::TestWithParam<IdctFunc> {
 protected:
  virtual void SetUp() {
-    int i;
-
    UUT = GetParam();
-    memset(input, 0, sizeof(input));
-    /* Set up guard blocks */
-    for (i = 0; i < 256; i++) output[i] = ((i & 0xF) < 4 && (i < 64)) ? 0 : -1;
+
+    input = new Buffer<int16_t>(4, 4, 0);
+    ASSERT_TRUE(input != NULL);
+    ASSERT_TRUE(input->Init());
+    predict = new Buffer<uint8_t>(4, 4, 3);
+    ASSERT_TRUE(predict != NULL);
+    ASSERT_TRUE(predict->Init());
+    output = new Buffer<uint8_t>(4, 4, 3);
+    ASSERT_TRUE(output != NULL);
+    ASSERT_TRUE(output->Init());
  }

-  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+  virtual void TearDown() {
+    delete input;
+    delete predict;
+    delete output;
+    libvpx_test::ClearSystemState();
+  }

  IdctFunc UUT;
-  int16_t input[16];
-  unsigned char output[256];
-  unsigned char predict[256];
+  Buffer<int16_t> *input;
+  Buffer<uint8_t> *predict;
+  Buffer<uint8_t> *output;
 };

-TEST_P(IDCTTest, TestGuardBlocks) {
-  int i;
-
-  for (i = 0; i < 256; i++) {
-    if ((i & 0xF) < 4 && i < 64)
-      EXPECT_EQ(0, output[i]) << i;
-    else
-      EXPECT_EQ(255, output[i]);
-  }
-}
-
 TEST_P(IDCTTest, TestAllZeros) {
-  int i;
+  // When the input is '0' the output will be '0'.
+  input->Set(0);
+  predict->Set(0);
+  output->Set(0);

-  ASM_REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
+  ASM_REGISTER_STATE_CHECK(UUT(input->TopLeftPixel(), predict->TopLeftPixel(),
+                               predict->stride(), output->TopLeftPixel(),
+                               output->stride()));

-  for (i = 0; i < 256; i++) {
-    if ((i & 0xF) < 4 && i < 64)
-      EXPECT_EQ(0, output[i]) << "i==" << i;
-    else
-      EXPECT_EQ(255, output[i]) << "i==" << i;
-  }
+  ASSERT_TRUE(input->CheckValues(0));
+  ASSERT_TRUE(input->CheckPadding());
+  ASSERT_TRUE(output->CheckValues(0));
+  ASSERT_TRUE(output->CheckPadding());
 }

 TEST_P(IDCTTest, TestAllOnes) {
-  int i;
+  input->Set(0);
+  // When the first element is '4' it will fill the output buffer with '1'.
+  input->TopLeftPixel()[0] = 4;
+  predict->Set(0);
+  output->Set(0);

-  input[0] = 4;
-  ASM_REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
+  ASM_REGISTER_STATE_CHECK(UUT(input->TopLeftPixel(), predict->TopLeftPixel(),
+                               predict->stride(), output->TopLeftPixel(),
+                               output->stride()));

-  for (i = 0; i < 256; i++) {
-    if ((i & 0xF) < 4 && i < 64)
-      EXPECT_EQ(1, output[i]) << "i==" << i;
-    else
-      EXPECT_EQ(255, output[i]) << "i==" << i;
-  }
+  ASSERT_TRUE(output->CheckValues(1));
+  ASSERT_TRUE(output->CheckPadding());
 }

 TEST_P(IDCTTest, TestAddOne) {
-  int i;
+  // Set the transform output to '1' and make sure it gets added to the
+  // prediction buffer.
+  input->Set(0);
+  input->TopLeftPixel()[0] = 4;
+  output->Set(0);

-  for (i = 0; i < 256; i++) predict[i] = i;
-  input[0] = 4;
-  ASM_REGISTER_STATE_CHECK(UUT(input, predict, 16, output, 16));
-
-  for (i = 0; i < 256; i++) {
-    if ((i & 0xF) < 4 && i < 64)
-      EXPECT_EQ(i + 1, output[i]) << "i==" << i;
-    else
-      EXPECT_EQ(255, output[i]) << "i==" << i;
+  uint8_t *pred = predict->TopLeftPixel();
+  for (int y = 0; y < 4; ++y) {
+    for (int x = 0; x < 4; ++x) {
+      pred[y * predict->stride() + x] = y * 4 + x;
+    }
  }
+
+  ASM_REGISTER_STATE_CHECK(UUT(input->TopLeftPixel(), predict->TopLeftPixel(),
+                               predict->stride(), output->TopLeftPixel(),
+                               output->stride()));
+
+  uint8_t const *out = output->TopLeftPixel();
+  for (int y = 0; y < 4; ++y) {
+    for (int x = 0; x < 4; ++x) {
+      EXPECT_EQ(1 + y * 4 + x, out[y * output->stride() + x]);
+    }
+  }
+
+  if (HasFailure()) {
+    output->DumpBuffer();
+  }
+
+  ASSERT_TRUE(output->CheckPadding());
 }

 TEST_P(IDCTTest, TestWithData) {
-  int i;
+  // Test a single known input.
+  predict->Set(0);

-  for (i = 0; i < 16; i++) input[i] = i;
-
-  ASM_REGISTER_STATE_CHECK(UUT(input, output, 16, output, 16));
-
-  for (i = 0; i < 256; i++) {
-    if ((i & 0xF) > 3 || i > 63)
-      EXPECT_EQ(255, output[i]) << "i==" << i;
-    else if (i == 0)
-      EXPECT_EQ(11, output[i]) << "i==" << i;
-    else if (i == 34)
-      EXPECT_EQ(1, output[i]) << "i==" << i;
-    else if (i == 2 || i == 17 || i == 32)
-      EXPECT_EQ(3, output[i]) << "i==" << i;
-    else
-      EXPECT_EQ(0, output[i]) << "i==" << i;
+  int16_t *in = input->TopLeftPixel();
+  for (int y = 0; y < 4; ++y) {
+    for (int x = 0; x < 4; ++x) {
+      in[y * input->stride() + x] = y * 4 + x;
+    }
  }
+
+  ASM_REGISTER_STATE_CHECK(UUT(input->TopLeftPixel(), predict->TopLeftPixel(),
+                               predict->stride(), output->TopLeftPixel(),
+                               output->stride()));
+
+  uint8_t *out = output->TopLeftPixel();
+  for (int y = 0; y < 4; ++y) {
+    for (int x = 0; x < 4; ++x) {
+      switch (y * 4 + x) {
+        case 0: EXPECT_EQ(11, out[y * output->stride() + x]); break;
+        case 2:
+        case 5:
+        case 8: EXPECT_EQ(3, out[y * output->stride() + x]); break;
+        case 10: EXPECT_EQ(1, out[y * output->stride() + x]); break;
+        default: EXPECT_EQ(0, out[y * output->stride() + x]);
+      }
+    }
+  }
+
+  if (HasFailure()) {
+    output->DumpBuffer();
+  }
+
+  ASSERT_TRUE(output->CheckPadding());
 }

 INSTANTIATE_TEST_CASE_P(C, IDCTTest, ::testing::Values(vp8_short_idct4x4llm_c));
+
 #if HAVE_NEON
 INSTANTIATE_TEST_CASE_P(NEON, IDCTTest,
                        ::testing::Values(vp8_short_idct4x4llm_neon));
-#endif
+#endif  // HAVE_NEON
+
 #if HAVE_MMX
 INSTANTIATE_TEST_CASE_P(MMX, IDCTTest,
                        ::testing::Values(vp8_short_idct4x4llm_mmx));
-#endif
+#endif  // HAVE_MMX
+
 #if HAVE_MSA
 INSTANTIATE_TEST_CASE_P(MSA, IDCTTest,
                        ::testing::Values(vp8_short_idct4x4llm_msa));
-#endif
+#endif  // HAVE_MSA
+
+#if HAVE_MMI
+INSTANTIATE_TEST_CASE_P(MMI, IDCTTest,
+                        ::testing::Values(vp8_short_idct4x4llm_mmi));
+#endif  // HAVE_MMI
 }
--- a/test/invalid_file_test.cc
+++ b/test/invalid_file_test.cc
@@ -45,8 +45,8 @@ class InvalidFileTest : public ::libvpx_test::DecoderTest,

  void OpenResFile(const std::string &res_file_name_) {
    res_file_ = libvpx_test::OpenTestDataFile(res_file_name_);
-    ASSERT_TRUE(res_file_ != NULL) << "Result file open failed. Filename: "
-                                   << res_file_name_;
+    ASSERT_TRUE(res_file_ != NULL)
+        << "Result file open failed. Filename: " << res_file_name_;
  }

  virtual bool HandleDecodeResult(
@@ -120,11 +120,23 @@ class InvalidFileTest : public ::libvpx_test::DecoderTest,

 TEST_P(InvalidFileTest, ReturnCode) { RunTest(); }

+#if CONFIG_VP8_DECODER
+const DecodeParam kVP8InvalidFileTests[] = {
+  { 1, "invalid-bug-1443.ivf" },
+};
+
+VP8_INSTANTIATE_TEST_CASE(InvalidFileTest,
+                          ::testing::ValuesIn(kVP8InvalidFileTests));
+#endif  // CONFIG_VP8_DECODER
+
 #if CONFIG_VP9_DECODER
 const DecodeParam kVP9InvalidFileTests[] = {
  { 1, "invalid-vp90-02-v2.webm" },
 #if CONFIG_VP9_HIGHBITDEPTH
  { 1, "invalid-vp90-2-00-quantizer-00.webm.ivf.s5861_r01-05_b6-.v2.ivf" },
+  { 1,
+    "invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-."
+    "ivf" },
 #endif
  { 1, "invalid-vp90-03-v3.webm" },
  { 1, "invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-.ivf" },
@@ -164,12 +176,12 @@ class InvalidFileInvalidPeekTest : public InvalidFileTest {
 TEST_P(InvalidFileInvalidPeekTest, ReturnCode) { RunTest(); }

 #if CONFIG_VP8_DECODER
-const DecodeParam kVP8InvalidFileTests[] = {
+const DecodeParam kVP8InvalidPeekTests[] = {
  { 1, "invalid-vp80-00-comprehensive-018.ivf.2kf_0x6.ivf" },
 };

 VP8_INSTANTIATE_TEST_CASE(InvalidFileInvalidPeekTest,
-                          ::testing::ValuesIn(kVP8InvalidFileTests));
+                          ::testing::ValuesIn(kVP8InvalidPeekTests));
 #endif  // CONFIG_VP8_DECODER

 #if CONFIG_VP9_DECODER
--- a/test/ivf_video_source.h
+++ b/test/ivf_video_source.h
@@ -47,8 +47,8 @@ class IVFVideoSource : public CompressedVideoSource {

  virtual void Begin() {
    input_file_ = OpenTestDataFile(file_name_);
-    ASSERT_TRUE(input_file_ != NULL) << "Input file open failed. Filename: "
-                                     << file_name_;
+    ASSERT_TRUE(input_file_ != NULL)
+        << "Input file open failed. Filename: " << file_name_;

    // Read file header
    uint8_t file_hdr[kIvfFileHdrSize];
--- a/test/keyframe_test.cc
+++ b/test/keyframe_test.cc
@@ -135,8 +135,8 @@ TEST_P(KeyframeTest, TestAutoKeyframe) {
  for (std::vector<vpx_codec_pts_t>::const_iterator iter = kf_pts_list_.begin();
       iter != kf_pts_list_.end(); ++iter) {
    if (deadline_ == VPX_DL_REALTIME && *iter > 0)
-      EXPECT_EQ(0, (*iter - 1) % 30) << "Unexpected keyframe at frame "
-                                     << *iter;
+      EXPECT_EQ(0, (*iter - 1) % 30)
+          << "Unexpected keyframe at frame " << *iter;
    else
      EXPECT_EQ(0, *iter % 30) << "Unexpected keyframe at frame " << *iter;
  }
--- a/test/level_test.cc
+++ b/test/level_test.cc
@@ -66,34 +66,34 @@ class LevelTest
  int level_;
 };

-TEST_P(LevelTest, TestTargetLevel11) {
+TEST_P(LevelTest, TestTargetLevel11Large) {
  ASSERT_NE(encoding_mode_, ::libvpx_test::kRealTime);
  ::libvpx_test::I420VideoSource video("hantro_odd.yuv", 208, 144, 30, 1, 0,
-                                       90);
+                                       60);
  target_level_ = 11;
  cfg_.rc_target_bitrate = 150;
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  ASSERT_EQ(target_level_, level_);
+  ASSERT_GE(target_level_, level_);
 }

-TEST_P(LevelTest, TestTargetLevel20) {
+TEST_P(LevelTest, TestTargetLevel20Large) {
  ASSERT_NE(encoding_mode_, ::libvpx_test::kRealTime);
  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
-                                       30, 1, 0, 90);
+                                       30, 1, 0, 60);
  target_level_ = 20;
  cfg_.rc_target_bitrate = 1200;
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  ASSERT_EQ(target_level_, level_);
+  ASSERT_GE(target_level_, level_);
 }

-TEST_P(LevelTest, TestTargetLevel31) {
+TEST_P(LevelTest, TestTargetLevel31Large) {
  ASSERT_NE(encoding_mode_, ::libvpx_test::kRealTime);
  ::libvpx_test::I420VideoSource video("niklas_1280_720_30.y4m", 1280, 720, 30,
                                       1, 0, 60);
  target_level_ = 31;
  cfg_.rc_target_bitrate = 8000;
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  ASSERT_EQ(target_level_, level_);
+  ASSERT_GE(target_level_, level_);
 }

 // Test for keeping level stats only
@@ -103,11 +103,11 @@ TEST_P(LevelTest, TestTargetLevel0) {
  target_level_ = 0;
  min_gf_internal_ = 4;
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  ASSERT_EQ(11, level_);
+  ASSERT_GE(11, level_);

  cfg_.rc_target_bitrate = 1600;
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  ASSERT_EQ(20, level_);
+  ASSERT_GE(20, level_);
 }

 // Test for level control being turned off
@@ -130,7 +130,7 @@ TEST_P(LevelTest, TestTargetLevelApi) {
    if (level == 10 || level == 11 || level == 20 || level == 21 ||
        level == 30 || level == 31 || level == 40 || level == 41 ||
        level == 50 || level == 51 || level == 52 || level == 60 ||
-        level == 61 || level == 62 || level == 0 || level == 255)
+        level == 61 || level == 62 || level == 0 || level == 1 || level == 255)
      EXPECT_EQ(VPX_CODEC_OK,
                vpx_codec_control(&enc, VP9E_SET_TARGET_LEVEL, level));
    else
--- a/test/lpf_test.cc
+++ b/test/lpf_test.cc
@@ -114,6 +114,18 @@ void InitInput(Pixel *s, Pixel *ref_s, ACMRandom *rnd, const uint8_t limit,
  }
 }

+uint8_t GetOuterThresh(ACMRandom *rnd) {
+  return static_cast<uint8_t>(rnd->RandRange(3 * MAX_LOOP_FILTER + 5));
+}
+
+uint8_t GetInnerThresh(ACMRandom *rnd) {
+  return static_cast<uint8_t>(rnd->RandRange(MAX_LOOP_FILTER + 1));
+}
+
+uint8_t GetHevThresh(ACMRandom *rnd) {
+  return static_cast<uint8_t>(rnd->RandRange(MAX_LOOP_FILTER + 1) >> 4);
+}
+
 class Loop8Test6Param : public ::testing::TestWithParam<loop8_param_t> {
 public:
  virtual ~Loop8Test6Param() {}
@@ -162,15 +174,15 @@ TEST_P(Loop8Test6Param, OperationCheck) {
  int first_failure = -1;
  for (int i = 0; i < count_test_block; ++i) {
    int err_count = 0;
-    uint8_t tmp = static_cast<uint8_t>(rnd(3 * MAX_LOOP_FILTER + 4));
+    uint8_t tmp = GetOuterThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    blimit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = static_cast<uint8_t>(rnd(MAX_LOOP_FILTER));
+    tmp = GetInnerThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    limit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                   tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = rnd.Rand8();
+    tmp = GetHevThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    thresh[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
@@ -221,15 +233,15 @@ TEST_P(Loop8Test6Param, ValueCheck) {

  for (int i = 0; i < count_test_block; ++i) {
    int err_count = 0;
-    uint8_t tmp = static_cast<uint8_t>(rnd(3 * MAX_LOOP_FILTER + 4));
+    uint8_t tmp = GetOuterThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    blimit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = static_cast<uint8_t>(rnd(MAX_LOOP_FILTER));
+    tmp = GetInnerThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    limit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                   tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = rnd.Rand8();
+    tmp = GetHevThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    thresh[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
@@ -271,27 +283,27 @@ TEST_P(Loop8Test9Param, OperationCheck) {
  int first_failure = -1;
  for (int i = 0; i < count_test_block; ++i) {
    int err_count = 0;
-    uint8_t tmp = static_cast<uint8_t>(rnd(3 * MAX_LOOP_FILTER + 4));
+    uint8_t tmp = GetOuterThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    blimit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = static_cast<uint8_t>(rnd(MAX_LOOP_FILTER));
+    tmp = GetInnerThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    limit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = rnd.Rand8();
+    tmp = GetHevThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    thresh0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = static_cast<uint8_t>(rnd(3 * MAX_LOOP_FILTER + 4));
+    tmp = GetOuterThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    blimit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = static_cast<uint8_t>(rnd(MAX_LOOP_FILTER));
+    tmp = GetInnerThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    limit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = rnd.Rand8();
+    tmp = GetHevThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    thresh1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
@@ -334,27 +346,27 @@ TEST_P(Loop8Test9Param, ValueCheck) {
  int first_failure = -1;
  for (int i = 0; i < count_test_block; ++i) {
    int err_count = 0;
-    uint8_t tmp = static_cast<uint8_t>(rnd(3 * MAX_LOOP_FILTER + 4));
+    uint8_t tmp = GetOuterThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    blimit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = static_cast<uint8_t>(rnd(MAX_LOOP_FILTER));
+    tmp = GetInnerThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    limit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = rnd.Rand8();
+    tmp = GetHevThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    thresh0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = static_cast<uint8_t>(rnd(3 * MAX_LOOP_FILTER + 4));
+    tmp = GetOuterThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    blimit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = static_cast<uint8_t>(rnd(MAX_LOOP_FILTER));
+    tmp = GetInnerThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    limit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
-    tmp = rnd.Rand8();
+    tmp = GetHevThresh(&rnd);
    DECLARE_ALIGNED(16, const uint8_t,
                    thresh1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,
                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };
--- a/test/minmax_test.cc
+++ b/test/minmax_test.cc
@@ -107,10 +107,10 @@ TEST_P(MinMaxTest, CompareReferenceAndVaryStride) {
      int min_ref, max_ref, min, max;
      reference_minmax(a, a_stride, b, b_stride, &min_ref, &max_ref);
      ASM_REGISTER_STATE_CHECK(mm_func_(a, a_stride, b, b_stride, &min, &max));
-      EXPECT_EQ(max_ref, max) << "when a_stride = " << a_stride
-                              << " and b_stride = " << b_stride;
-      EXPECT_EQ(min_ref, min) << "when a_stride = " << a_stride
-                              << " and b_stride = " << b_stride;
+      EXPECT_EQ(max_ref, max)
+          << "when a_stride = " << a_stride << " and b_stride = " << b_stride;
+      EXPECT_EQ(min_ref, min)
+          << "when a_stride = " << a_stride << " and b_stride = " << b_stride;
    }
  }
 }
@@ -127,4 +127,9 @@ INSTANTIATE_TEST_CASE_P(NEON, MinMaxTest,
                        ::testing::Values(&vpx_minmax_8x8_neon));
 #endif

+#if HAVE_MSA
+INSTANTIATE_TEST_CASE_P(MSA, MinMaxTest,
+                        ::testing::Values(&vpx_minmax_8x8_msa));
+#endif
+
 }  // namespace
--- a/test/partial_idct_test.cc
+++ b/test/partial_idct_test.cc
@@ -43,9 +43,11 @@ void wrapper(const tran_low_t *in, uint8_t *out, int stride, int bd) {
 }

 #if CONFIG_VP9_HIGHBITDEPTH
-template <InvTxfmWithBdFunc fn>
+typedef void (*InvTxfmHighbdFunc)(const tran_low_t *in, uint16_t *out,
+                                  int stride, int bd);
+template <InvTxfmHighbdFunc fn>
 void highbd_wrapper(const tran_low_t *in, uint8_t *out, int stride, int bd) {
-  fn(in, CONVERT_TO_BYTEPTR(out), stride, bd);
+  fn(in, CAST_TO_SHORTPTR(out), stride, bd);
 }
 #endif

@@ -55,41 +57,14 @@ typedef std::tr1::tuple<FwdTxfmFunc, InvTxfmWithBdFunc, InvTxfmWithBdFunc,
 const int kMaxNumCoeffs = 1024;
 const int kCountTestBlock = 1000;

-// https://bugs.chromium.org/p/webm/issues/detail?id=1332
-// The functions specified do not pass with INT16_MIN/MAX. They fail at the
-// value specified, but pass when 1 is added/subtracted.
-int16_t MaxSupportedCoeff(InvTxfmWithBdFunc a) {
-#if HAVE_SSSE3 && ARCH_X86_64 && !CONFIG_EMULATE_HARDWARE
-  if (a == &wrapper<vpx_idct8x8_64_add_ssse3> ||
-      a == &wrapper<vpx_idct8x8_12_add_ssse3>) {
-    return 23625 - 1;
-  }
-#else
-  (void)a;
-#endif
-  return std::numeric_limits<int16_t>::max();
-}
-
-int16_t MinSupportedCoeff(InvTxfmWithBdFunc a) {
-#if HAVE_SSSE3 && ARCH_X86_64 && !CONFIG_EMULATE_HARDWARE
-  if (a == &wrapper<vpx_idct8x8_64_add_ssse3> ||
-      a == &wrapper<vpx_idct8x8_12_add_ssse3>) {
-    return -23625 + 1;
-  }
-#else
-  (void)a;
-#endif
-  return std::numeric_limits<int16_t>::min();
-}
-
 class PartialIDctTest : public ::testing::TestWithParam<PartialInvTxfmParam> {
 public:
  virtual ~PartialIDctTest() {}
  virtual void SetUp() {
    rnd_.Reset(ACMRandom::DeterministicSeed());
-    ftxfm_ = GET_PARAM(0);
-    full_itxfm_ = GET_PARAM(1);
-    partial_itxfm_ = GET_PARAM(2);
+    fwd_txfm_ = GET_PARAM(0);
+    full_inv_txfm_ = GET_PARAM(1);
+    partial_inv_txfm_ = GET_PARAM(2);
    tx_size_ = GET_PARAM(3);
    last_nonzero_ = GET_PARAM(4);
    bit_depth_ = GET_PARAM(5);
@@ -153,12 +128,12 @@ class PartialIDctTest : public ::testing::TestWithParam<PartialInvTxfmParam> {
  }

  void InitInput() {
-    const int max_coeff = 32766 / 4;
-    int max_energy_leftover = max_coeff * max_coeff;
+    const int64_t max_coeff = (32766 << (bit_depth_ - 8)) / 4;
+    int64_t max_energy_leftover = max_coeff * max_coeff;
    for (int j = 0; j < last_nonzero_; ++j) {
-      int16_t coeff = static_cast<int16_t>(sqrt(1.0 * max_energy_leftover) *
-                                           (rnd_.Rand16() - 32768) / 65536);
-      max_energy_leftover -= coeff * coeff;
+      tran_low_t coeff = static_cast<tran_low_t>(
+          sqrt(1.0 * max_energy_leftover) * (rnd_.Rand16() - 32768) / 65536);
+      max_energy_leftover -= static_cast<int64_t>(coeff) * coeff;
      if (max_energy_leftover < 0) {
        max_energy_leftover = 0;
        coeff = 0;
@@ -167,6 +142,36 @@ class PartialIDctTest : public ::testing::TestWithParam<PartialInvTxfmParam> {
    }
  }

+  void PrintDiff() {
+    if (memcmp(output_block_ref_, output_block_,
+               pixel_size_ * output_block_size_)) {
+      uint16_t ref, opt;
+      for (int y = 0; y < size_; y++) {
+        for (int x = 0; x < size_; x++) {
+          if (pixel_size_ == 1) {
+            ref = output_block_ref_[y * stride_ + x];
+            opt = output_block_[y * stride_ + x];
+          } else {
+            ref = reinterpret_cast<uint16_t *>(
+                output_block_ref_)[y * stride_ + x];
+            opt = reinterpret_cast<uint16_t *>(output_block_)[y * stride_ + x];
+          }
+          if (ref != opt) {
+            printf("dest[%d][%d] diff:%6d (ref),%6d (opt)\n", y, x, ref, opt);
+          }
+        }
+      }
+
+      printf("\ninput_block_:\n");
+      for (int y = 0; y < size_; y++) {
+        for (int x = 0; x < size_; x++) {
+          printf("%6d,", input_block_[y * size_ + x]);
+        }
+        printf("\n");
+      }
+    }
+  }
+
 protected:
  int last_nonzero_;
  TX_SIZE tx_size_;
@@ -180,34 +185,43 @@ class PartialIDctTest : public ::testing::TestWithParam<PartialInvTxfmParam> {
  int output_block_size_;
  int bit_depth_;
  int mask_;
-  FwdTxfmFunc ftxfm_;
-  InvTxfmWithBdFunc full_itxfm_;
-  InvTxfmWithBdFunc partial_itxfm_;
+  FwdTxfmFunc fwd_txfm_;
+  InvTxfmWithBdFunc full_inv_txfm_;
+  InvTxfmWithBdFunc partial_inv_txfm_;
  ACMRandom rnd_;
 };

 TEST_P(PartialIDctTest, RunQuantCheck) {
+  const int count_test_block = (size_ != 4) ? kCountTestBlock : 65536;
  DECLARE_ALIGNED(16, int16_t, input_extreme_block[kMaxNumCoeffs]);
  DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kMaxNumCoeffs]);

  InitMem();
-  for (int i = 0; i < kCountTestBlock * kCountTestBlock; ++i) {
+
+  for (int i = 0; i < count_test_block; ++i) {
    // Initialize a test block with input range [-mask_, mask_].
-    if (i == 0) {
-      for (int k = 0; k < input_block_size_; ++k) {
-        input_extreme_block[k] = mask_;
-      }
-    } else if (i == 1) {
-      for (int k = 0; k < input_block_size_; ++k) {
-        input_extreme_block[k] = -mask_;
+    if (size_ != 4) {
+      if (i == 0) {
+        for (int k = 0; k < input_block_size_; ++k) {
+          input_extreme_block[k] = mask_;
+        }
+      } else if (i == 1) {
+        for (int k = 0; k < input_block_size_; ++k) {
+          input_extreme_block[k] = -mask_;
+        }
+      } else {
+        for (int k = 0; k < input_block_size_; ++k) {
+          input_extreme_block[k] = rnd_.Rand8() % 2 ? mask_ : -mask_;
+        }
      }
    } else {
+      // Try all possible combinations.
      for (int k = 0; k < input_block_size_; ++k) {
-        input_extreme_block[k] = rnd_.Rand8() % 2 ? mask_ : -mask_;
+        input_extreme_block[k] = (i & (1 << k)) ? mask_ : -mask_;
      }
    }

-    ftxfm_(input_extreme_block, output_ref_block, size_);
+    fwd_txfm_(input_extreme_block, output_ref_block, size_);

    // quantization with minimum allowed step sizes
    input_block_[0] = (output_ref_block[0] / 4) * 4;
@@ -217,9 +231,9 @@ TEST_P(PartialIDctTest, RunQuantCheck) {
    }

    ASM_REGISTER_STATE_CHECK(
-        full_itxfm_(input_block_, output_block_ref_, stride_, bit_depth_));
+        full_inv_txfm_(input_block_, output_block_ref_, stride_, bit_depth_));
    ASM_REGISTER_STATE_CHECK(
-        partial_itxfm_(input_block_, output_block_, stride_, bit_depth_));
+        partial_inv_txfm_(input_block_, output_block_, stride_, bit_depth_));
    ASSERT_EQ(0, memcmp(output_block_ref_, output_block_,
                        pixel_size_ * output_block_size_))
        << "Error: partial inverse transform produces different results";
@@ -232,9 +246,9 @@ TEST_P(PartialIDctTest, ResultsMatch) {
    InitInput();

    ASM_REGISTER_STATE_CHECK(
-        full_itxfm_(input_block_, output_block_ref_, stride_, bit_depth_));
+        full_inv_txfm_(input_block_, output_block_ref_, stride_, bit_depth_));
    ASM_REGISTER_STATE_CHECK(
-        partial_itxfm_(input_block_, output_block_, stride_, bit_depth_));
+        partial_inv_txfm_(input_block_, output_block_, stride_, bit_depth_));
    ASSERT_EQ(0, memcmp(output_block_ref_, output_block_,
                        pixel_size_ * output_block_size_))
        << "Error: partial inverse transform produces different results";
@@ -249,9 +263,9 @@ TEST_P(PartialIDctTest, AddOutputBlock) {
    }

    ASM_REGISTER_STATE_CHECK(
-        full_itxfm_(input_block_, output_block_ref_, stride_, bit_depth_));
+        full_inv_txfm_(input_block_, output_block_ref_, stride_, bit_depth_));
    ASM_REGISTER_STATE_CHECK(
-        partial_itxfm_(input_block_, output_block_, stride_, bit_depth_));
+        partial_inv_txfm_(input_block_, output_block_, stride_, bit_depth_));
    ASSERT_EQ(0, memcmp(output_block_ref_, output_block_,
                        pixel_size_ * output_block_size_))
        << "Error: Transform results are not correctly added to output.";
@@ -259,8 +273,8 @@ TEST_P(PartialIDctTest, AddOutputBlock) {
 }

 TEST_P(PartialIDctTest, SingleExtremeCoeff) {
-  const int16_t max_coeff = MaxSupportedCoeff(partial_itxfm_);
-  const int16_t min_coeff = MinSupportedCoeff(partial_itxfm_);
+  const int16_t max_coeff = std::numeric_limits<int16_t>::max();
+  const int16_t min_coeff = std::numeric_limits<int16_t>::min();
  for (int i = 0; i < last_nonzero_; ++i) {
    memset(input_block_, 0, sizeof(*input_block_) * input_block_size_);
    // Run once for min and once for max.
@@ -272,9 +286,9 @@ TEST_P(PartialIDctTest, SingleExtremeCoeff) {
      input_block_[vp9_default_scan_orders[tx_size_].scan[i]] = coeff;

      ASM_REGISTER_STATE_CHECK(
-          full_itxfm_(input_block_, output_block_ref_, stride_, bit_depth_));
+          full_inv_txfm_(input_block_, output_block_ref_, stride_, bit_depth_));
      ASM_REGISTER_STATE_CHECK(
-          partial_itxfm_(input_block_, output_block_, stride_, bit_depth_));
+          partial_inv_txfm_(input_block_, output_block_, stride_, bit_depth_));
      ASSERT_EQ(0, memcmp(output_block_ref_, output_block_,
                          pixel_size_ * output_block_size_))
          << "Error: Fails with single coeff of " << coeff << " at " << i
@@ -291,20 +305,20 @@ TEST_P(PartialIDctTest, DISABLED_Speed) {

  for (int i = 0; i < kCountSpeedTestBlock; ++i) {
    ASM_REGISTER_STATE_CHECK(
-        full_itxfm_(input_block_, output_block_ref_, stride_, bit_depth_));
+        full_inv_txfm_(input_block_, output_block_ref_, stride_, bit_depth_));
  }
  vpx_usec_timer timer;
  vpx_usec_timer_start(&timer);
  for (int i = 0; i < kCountSpeedTestBlock; ++i) {
-    partial_itxfm_(input_block_, output_block_, stride_, bit_depth_);
+    partial_inv_txfm_(input_block_, output_block_, stride_, bit_depth_);
  }
  libvpx_test::ClearSystemState();
  vpx_usec_timer_mark(&timer);
  const int elapsed_time =
      static_cast<int>(vpx_usec_timer_elapsed(&timer) / 1000);
-  printf("idct%dx%d_%d (bitdepth %d) time: %5d ms ", size_, size_,
-         last_nonzero_, bit_depth_, elapsed_time);
-
+  printf("idct%dx%d_%d (%s %d) time: %5d ms\n", size_, size_, last_nonzero_,
+         (pixel_size_ == 1) ? "bitdepth" : "high bitdepth", bit_depth_,
+         elapsed_time);
  ASSERT_EQ(0, memcmp(output_block_ref_, output_block_,
                      pixel_size_ * output_block_size_))
      << "Error: partial inverse transform produces different results";
@@ -323,6 +337,15 @@ const PartialInvTxfmParam c_partial_idct_tests[] = {
  make_tuple(
      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
      &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>, TX_32X32, 1024, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>, TX_32X32, 135, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>, TX_32X32, 135, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>, TX_32X32, 135, 12, 2),
  make_tuple(
      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
      &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>, TX_32X32, 34, 8, 2),
@@ -350,6 +373,15 @@ const PartialInvTxfmParam c_partial_idct_tests[] = {
  make_tuple(
      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
      &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>, TX_16X16, 256, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>, TX_16X16, 38, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>, TX_16X16, 38, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>, TX_16X16, 38, 12, 2),
  make_tuple(
      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
      &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>, TX_16X16, 10, 8, 2),
@@ -424,6 +456,8 @@ const PartialInvTxfmParam c_partial_idct_tests[] = {
             &wrapper<vpx_idct32x32_1_add_c>, TX_32X32, 1, 8, 1),
  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
             &wrapper<vpx_idct16x16_256_add_c>, TX_16X16, 256, 8, 1),
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
+             &wrapper<vpx_idct16x16_38_add_c>, TX_16X16, 38, 8, 1),
  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
             &wrapper<vpx_idct16x16_10_add_c>, TX_16X16, 10, 8, 1),
  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
@@ -443,9 +477,86 @@ const PartialInvTxfmParam c_partial_idct_tests[] = {
 INSTANTIATE_TEST_CASE_P(C, PartialIDctTest,
                        ::testing::ValuesIn(c_partial_idct_tests));

-#if HAVE_NEON && !CONFIG_EMULATE_HARDWARE
+#if !CONFIG_EMULATE_HARDWARE
+
+#if HAVE_NEON
 const PartialInvTxfmParam neon_partial_idct_tests[] = {
 #if CONFIG_VP9_HIGHBITDEPTH
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_neon>, TX_32X32,
+             1024, 8, 2),
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_neon>, TX_32X32,
+             1024, 10, 2),
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_neon>, TX_32X32,
+             1024, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_135_add_neon>, TX_32X32, 135, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_135_add_neon>, TX_32X32, 135, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_135_add_neon>, TX_32X32, 135, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_34_add_neon>, TX_32X32, 34, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_34_add_neon>, TX_32X32, 34, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_34_add_neon>, TX_32X32, 34, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_1_add_neon>, TX_32X32, 1, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_1_add_neon>, TX_32X32, 1, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_1_add_neon>, TX_32X32, 1, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_256_add_neon>, TX_16X16, 256, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_256_add_neon>, TX_16X16, 256, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_256_add_neon>, TX_16X16, 256, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_neon>, TX_16X16, 38, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_neon>, TX_16X16, 38, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_neon>, TX_16X16, 38, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_10_add_neon>, TX_16X16, 10, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_10_add_neon>, TX_16X16, 10, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_10_add_neon>, TX_16X16, 10, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_1_add_neon>, TX_16X16, 1, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_1_add_neon>, TX_16X16, 1, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_1_add_neon>, TX_16X16, 1, 12, 2),
  make_tuple(&vpx_highbd_fdct8x8_c,
             &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
             &highbd_wrapper<vpx_highbd_idct8x8_64_add_neon>, TX_8X8, 64, 8, 2),
@@ -488,46 +599,78 @@ const PartialInvTxfmParam neon_partial_idct_tests[] = {
 #endif  // CONFIG_VP9_HIGHBITDEPTH
  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
             &wrapper<vpx_idct32x32_1024_add_neon>, TX_32X32, 1024, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_135_add_c>,
             &wrapper<vpx_idct32x32_135_add_neon>, TX_32X32, 135, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_34_add_c>,
             &wrapper<vpx_idct32x32_34_add_neon>, TX_32X32, 34, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1_add_c>,
             &wrapper<vpx_idct32x32_1_add_neon>, TX_32X32, 1, 8, 1),
  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
             &wrapper<vpx_idct16x16_256_add_neon>, TX_16X16, 256, 8, 1),
-  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_38_add_c>,
+             &wrapper<vpx_idct16x16_38_add_neon>, TX_16X16, 38, 8, 1),
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_10_add_c>,
             &wrapper<vpx_idct16x16_10_add_neon>, TX_16X16, 10, 8, 1),
-  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_1_add_c>,
             &wrapper<vpx_idct16x16_1_add_neon>, TX_16X16, 1, 8, 1),
  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
             &wrapper<vpx_idct8x8_64_add_neon>, TX_8X8, 64, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
+  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_12_add_c>,
             &wrapper<vpx_idct8x8_12_add_neon>, TX_8X8, 12, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
+  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_1_add_c>,
             &wrapper<vpx_idct8x8_1_add_neon>, TX_8X8, 1, 8, 1),
  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_16_add_c>,
             &wrapper<vpx_idct4x4_16_add_neon>, TX_4X4, 16, 8, 1),
-  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_16_add_c>,
+  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_1_add_c>,
             &wrapper<vpx_idct4x4_1_add_neon>, TX_4X4, 1, 8, 1)
 };

 INSTANTIATE_TEST_CASE_P(NEON, PartialIDctTest,
                        ::testing::ValuesIn(neon_partial_idct_tests));
-#endif  // HAVE_NEON && !CONFIG_EMULATE_HARDWARE
+#endif  // HAVE_NEON

-#if HAVE_SSE2 && !CONFIG_EMULATE_HARDWARE
+#if HAVE_SSE2
 // 32x32_135_ is implemented using the 1024 version.
 const PartialInvTxfmParam sse2_partial_idct_tests[] = {
 #if CONFIG_VP9_HIGHBITDEPTH
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_sse2>, TX_32X32,
+             1024, 8, 2),
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_sse2>, TX_32X32,
+             1024, 10, 2),
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_sse2>, TX_32X32,
+             1024, 12, 2),
  make_tuple(
-      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_135_add_sse2>, TX_32X32, 135, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_135_add_sse2>, TX_32X32, 135, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_135_add_sse2>, TX_32X32, 135, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_34_add_sse2>, TX_32X32, 34, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_34_add_sse2>, TX_32X32, 34, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_34_add_sse2>, TX_32X32, 34, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1_add_c>,
      &highbd_wrapper<vpx_highbd_idct32x32_1_add_sse2>, TX_32X32, 1, 8, 2),
  make_tuple(
-      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1_add_c>,
      &highbd_wrapper<vpx_highbd_idct32x32_1_add_sse2>, TX_32X32, 1, 10, 2),
  make_tuple(
-      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_1_add_c>,
      &highbd_wrapper<vpx_highbd_idct32x32_1_add_sse2>, TX_32X32, 1, 12, 2),
  make_tuple(
      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
@@ -539,14 +682,32 @@ const PartialInvTxfmParam sse2_partial_idct_tests[] = {
      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
      &highbd_wrapper<vpx_highbd_idct16x16_256_add_sse2>, TX_16X16, 256, 12, 2),
  make_tuple(
-      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_sse2>, TX_16X16, 38, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_sse2>, TX_16X16, 38, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_sse2>, TX_16X16, 38, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
      &highbd_wrapper<vpx_highbd_idct16x16_10_add_sse2>, TX_16X16, 10, 8, 2),
  make_tuple(
-      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
      &highbd_wrapper<vpx_highbd_idct16x16_10_add_sse2>, TX_16X16, 10, 10, 2),
  make_tuple(
-      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
      &highbd_wrapper<vpx_highbd_idct16x16_10_add_sse2>, TX_16X16, 10, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_1_add_sse2>, TX_16X16, 1, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_1_add_sse2>, TX_16X16, 1, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_1_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_1_add_sse2>, TX_16X16, 1, 12, 2),
  make_tuple(&vpx_highbd_fdct8x8_c,
             &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
             &highbd_wrapper<vpx_highbd_idct8x8_64_add_sse2>, TX_8X8, 64, 8, 2),
@@ -557,14 +718,20 @@ const PartialInvTxfmParam sse2_partial_idct_tests[] = {
      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
      &highbd_wrapper<vpx_highbd_idct8x8_64_add_sse2>, TX_8X8, 64, 12, 2),
  make_tuple(&vpx_highbd_fdct8x8_c,
-             &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
+             &highbd_wrapper<vpx_highbd_idct8x8_12_add_c>,
             &highbd_wrapper<vpx_highbd_idct8x8_12_add_sse2>, TX_8X8, 12, 8, 2),
  make_tuple(
-      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
+      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_12_add_c>,
      &highbd_wrapper<vpx_highbd_idct8x8_12_add_sse2>, TX_8X8, 12, 10, 2),
  make_tuple(
-      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
+      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_12_add_c>,
      &highbd_wrapper<vpx_highbd_idct8x8_12_add_sse2>, TX_8X8, 12, 12, 2),
+  make_tuple(&vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct8x8_1_add_sse2>, TX_8X8, 1, 8, 2),
+  make_tuple(&vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct8x8_1_add_sse2>, TX_8X8, 1, 10, 2),
+  make_tuple(&vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct8x8_1_add_sse2>, TX_8X8, 1, 12, 2),
  make_tuple(&vpx_highbd_fdct4x4_c,
             &highbd_wrapper<vpx_highbd_idct4x4_16_add_c>,
             &highbd_wrapper<vpx_highbd_idct4x4_16_add_sse2>, TX_4X4, 16, 8, 2),
@@ -574,119 +741,219 @@ const PartialInvTxfmParam sse2_partial_idct_tests[] = {
  make_tuple(
      &vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_16_add_c>,
      &highbd_wrapper<vpx_highbd_idct4x4_16_add_sse2>, TX_4X4, 16, 12, 2),
+  make_tuple(&vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct4x4_1_add_sse2>, TX_4X4, 1, 8, 2),
+  make_tuple(&vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct4x4_1_add_sse2>, TX_4X4, 1, 10, 2),
+  make_tuple(&vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_1_add_c>,
+             &highbd_wrapper<vpx_highbd_idct4x4_1_add_sse2>, TX_4X4, 1, 12, 2),
 #endif  // CONFIG_VP9_HIGHBITDEPTH
  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
             &wrapper<vpx_idct32x32_1024_add_sse2>, TX_32X32, 1024, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
-             &wrapper<vpx_idct32x32_1024_add_sse2>, TX_32X32, 135, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_135_add_c>,
+             &wrapper<vpx_idct32x32_135_add_sse2>, TX_32X32, 135, 8, 1),
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_34_add_c>,
             &wrapper<vpx_idct32x32_34_add_sse2>, TX_32X32, 34, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1_add_c>,
             &wrapper<vpx_idct32x32_1_add_sse2>, TX_32X32, 1, 8, 1),
  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
             &wrapper<vpx_idct16x16_256_add_sse2>, TX_16X16, 256, 8, 1),
-  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_38_add_c>,
+             &wrapper<vpx_idct16x16_38_add_sse2>, TX_16X16, 38, 8, 1),
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_10_add_c>,
             &wrapper<vpx_idct16x16_10_add_sse2>, TX_16X16, 10, 8, 1),
-  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_1_add_c>,
             &wrapper<vpx_idct16x16_1_add_sse2>, TX_16X16, 1, 8, 1),
  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
             &wrapper<vpx_idct8x8_64_add_sse2>, TX_8X8, 64, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
+  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_12_add_c>,
             &wrapper<vpx_idct8x8_12_add_sse2>, TX_8X8, 12, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
+  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_1_add_c>,
             &wrapper<vpx_idct8x8_1_add_sse2>, TX_8X8, 1, 8, 1),
  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_16_add_c>,
             &wrapper<vpx_idct4x4_16_add_sse2>, TX_4X4, 16, 8, 1),
-  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_16_add_c>,
+  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_1_add_c>,
             &wrapper<vpx_idct4x4_1_add_sse2>, TX_4X4, 1, 8, 1)
 };

 INSTANTIATE_TEST_CASE_P(SSE2, PartialIDctTest,
                        ::testing::ValuesIn(sse2_partial_idct_tests));

-#endif  // HAVE_SSE2 && !CONFIG_EMULATE_HARDWARE
+#endif  // HAVE_SSE2

-#if HAVE_SSSE3 && ARCH_X86_64 && !CONFIG_EMULATE_HARDWARE
+#if HAVE_SSSE3
 const PartialInvTxfmParam ssse3_partial_idct_tests[] = {
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
-             &wrapper<vpx_idct32x32_1024_add_ssse3>, TX_32X32, 1024, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_135_add_c>,
             &wrapper<vpx_idct32x32_135_add_ssse3>, TX_32X32, 135, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_34_add_c>,
             &wrapper<vpx_idct32x32_34_add_ssse3>, TX_32X32, 34, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
-             &wrapper<vpx_idct8x8_64_add_ssse3>, TX_8X8, 64, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
+  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_12_add_c>,
             &wrapper<vpx_idct8x8_12_add_ssse3>, TX_8X8, 12, 8, 1)
 };

 INSTANTIATE_TEST_CASE_P(SSSE3, PartialIDctTest,
                        ::testing::ValuesIn(ssse3_partial_idct_tests));
-#endif  // HAVE_SSSE3 && ARCH_X86_64 && !CONFIG_EMULATE_HARDWARE
+#endif  // HAVE_SSSE3

-#if HAVE_DSPR2 && !CONFIG_EMULATE_HARDWARE && !CONFIG_VP9_HIGHBITDEPTH
+#if HAVE_SSE4_1 && CONFIG_VP9_HIGHBITDEPTH
+const PartialInvTxfmParam sse4_1_partial_idct_tests[] = {
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_sse4_1>, TX_32X32,
+             1024, 8, 2),
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_sse4_1>, TX_32X32,
+             1024, 10, 2),
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_1024_add_sse4_1>, TX_32X32,
+             1024, 12, 2),
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_135_add_sse4_1>, TX_32X32,
+             135, 8, 2),
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_135_add_sse4_1>, TX_32X32,
+             135, 10, 2),
+  make_tuple(&vpx_highbd_fdct32x32_c,
+             &highbd_wrapper<vpx_highbd_idct32x32_135_add_c>,
+             &highbd_wrapper<vpx_highbd_idct32x32_135_add_sse4_1>, TX_32X32,
+             135, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_34_add_sse4_1>, TX_32X32, 34, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_34_add_sse4_1>, TX_32X32, 34, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct32x32_c, &highbd_wrapper<vpx_highbd_idct32x32_34_add_c>,
+      &highbd_wrapper<vpx_highbd_idct32x32_34_add_sse4_1>, TX_32X32, 34, 12, 2),
+  make_tuple(&vpx_highbd_fdct16x16_c,
+             &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+             &highbd_wrapper<vpx_highbd_idct16x16_256_add_sse4_1>, TX_16X16,
+             256, 8, 2),
+  make_tuple(&vpx_highbd_fdct16x16_c,
+             &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+             &highbd_wrapper<vpx_highbd_idct16x16_256_add_sse4_1>, TX_16X16,
+             256, 10, 2),
+  make_tuple(&vpx_highbd_fdct16x16_c,
+             &highbd_wrapper<vpx_highbd_idct16x16_256_add_c>,
+             &highbd_wrapper<vpx_highbd_idct16x16_256_add_sse4_1>, TX_16X16,
+             256, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_sse4_1>, TX_16X16, 38, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_sse4_1>, TX_16X16, 38, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_38_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_38_add_sse4_1>, TX_16X16, 38, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_10_add_sse4_1>, TX_16X16, 10, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_10_add_sse4_1>, TX_16X16, 10, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct16x16_c, &highbd_wrapper<vpx_highbd_idct16x16_10_add_c>,
+      &highbd_wrapper<vpx_highbd_idct16x16_10_add_sse4_1>, TX_16X16, 10, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
+      &highbd_wrapper<vpx_highbd_idct8x8_64_add_sse4_1>, TX_8X8, 64, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
+      &highbd_wrapper<vpx_highbd_idct8x8_64_add_sse4_1>, TX_8X8, 64, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_64_add_c>,
+      &highbd_wrapper<vpx_highbd_idct8x8_64_add_sse4_1>, TX_8X8, 64, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_12_add_c>,
+      &highbd_wrapper<vpx_highbd_idct8x8_12_add_sse4_1>, TX_8X8, 12, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_12_add_c>,
+      &highbd_wrapper<vpx_highbd_idct8x8_12_add_sse4_1>, TX_8X8, 12, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct8x8_c, &highbd_wrapper<vpx_highbd_idct8x8_12_add_c>,
+      &highbd_wrapper<vpx_highbd_idct8x8_12_add_sse4_1>, TX_8X8, 12, 12, 2),
+  make_tuple(
+      &vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_16_add_c>,
+      &highbd_wrapper<vpx_highbd_idct4x4_16_add_sse4_1>, TX_4X4, 16, 8, 2),
+  make_tuple(
+      &vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_16_add_c>,
+      &highbd_wrapper<vpx_highbd_idct4x4_16_add_sse4_1>, TX_4X4, 16, 10, 2),
+  make_tuple(
+      &vpx_highbd_fdct4x4_c, &highbd_wrapper<vpx_highbd_idct4x4_16_add_c>,
+      &highbd_wrapper<vpx_highbd_idct4x4_16_add_sse4_1>, TX_4X4, 16, 12, 2)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, PartialIDctTest,
+                        ::testing::ValuesIn(sse4_1_partial_idct_tests));
+#endif  // HAVE_SSE4_1 && CONFIG_VP9_HIGHBITDEPTH
+
+#if HAVE_DSPR2 && !CONFIG_VP9_HIGHBITDEPTH
 const PartialInvTxfmParam dspr2_partial_idct_tests[] = {
  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
             &wrapper<vpx_idct32x32_1024_add_dspr2>, TX_32X32, 1024, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
-             &wrapper<vpx_idct32x32_1024_add_dspr2>, TX_32X32, 135, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_34_add_c>,
             &wrapper<vpx_idct32x32_34_add_dspr2>, TX_32X32, 34, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1_add_c>,
             &wrapper<vpx_idct32x32_1_add_dspr2>, TX_32X32, 1, 8, 1),
  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
             &wrapper<vpx_idct16x16_256_add_dspr2>, TX_16X16, 256, 8, 1),
-  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_10_add_c>,
             &wrapper<vpx_idct16x16_10_add_dspr2>, TX_16X16, 10, 8, 1),
-  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_1_add_c>,
             &wrapper<vpx_idct16x16_1_add_dspr2>, TX_16X16, 1, 8, 1),
  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
             &wrapper<vpx_idct8x8_64_add_dspr2>, TX_8X8, 64, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
+  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_12_add_c>,
             &wrapper<vpx_idct8x8_12_add_dspr2>, TX_8X8, 12, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
+  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_1_add_c>,
             &wrapper<vpx_idct8x8_1_add_dspr2>, TX_8X8, 1, 8, 1),
  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_16_add_c>,
             &wrapper<vpx_idct4x4_16_add_dspr2>, TX_4X4, 16, 8, 1),
-  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_16_add_c>,
+  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_1_add_c>,
             &wrapper<vpx_idct4x4_1_add_dspr2>, TX_4X4, 1, 8, 1)
 };

 INSTANTIATE_TEST_CASE_P(DSPR2, PartialIDctTest,
                        ::testing::ValuesIn(dspr2_partial_idct_tests));
-#endif  // HAVE_DSPR2 && !CONFIG_EMULATE_HARDWARE && !CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_DSPR2 && !CONFIG_VP9_HIGHBITDEPTH

-#if HAVE_MSA && !CONFIG_EMULATE_HARDWARE && !CONFIG_VP9_HIGHBITDEPTH
+#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH
 // 32x32_135_ is implemented using the 1024 version.
 const PartialInvTxfmParam msa_partial_idct_tests[] = {
  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
             &wrapper<vpx_idct32x32_1024_add_msa>, TX_32X32, 1024, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
-             &wrapper<vpx_idct32x32_1024_add_msa>, TX_32X32, 135, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_34_add_c>,
             &wrapper<vpx_idct32x32_34_add_msa>, TX_32X32, 34, 8, 1),
-  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1024_add_c>,
+  make_tuple(&vpx_fdct32x32_c, &wrapper<vpx_idct32x32_1_add_c>,
             &wrapper<vpx_idct32x32_1_add_msa>, TX_32X32, 1, 8, 1),
  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
             &wrapper<vpx_idct16x16_256_add_msa>, TX_16X16, 256, 8, 1),
-  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_10_add_c>,
             &wrapper<vpx_idct16x16_10_add_msa>, TX_16X16, 10, 8, 1),
-  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_256_add_c>,
+  make_tuple(&vpx_fdct16x16_c, &wrapper<vpx_idct16x16_1_add_c>,
             &wrapper<vpx_idct16x16_1_add_msa>, TX_16X16, 1, 8, 1),
  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
             &wrapper<vpx_idct8x8_64_add_msa>, TX_8X8, 64, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
+  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_12_add_c>,
             &wrapper<vpx_idct8x8_12_add_msa>, TX_8X8, 12, 8, 1),
-  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_64_add_c>,
+  make_tuple(&vpx_fdct8x8_c, &wrapper<vpx_idct8x8_1_add_c>,
             &wrapper<vpx_idct8x8_1_add_msa>, TX_8X8, 1, 8, 1),
  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_16_add_c>,
             &wrapper<vpx_idct4x4_16_add_msa>, TX_4X4, 16, 8, 1),
-  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_16_add_c>,
+  make_tuple(&vpx_fdct4x4_c, &wrapper<vpx_idct4x4_1_add_c>,
             &wrapper<vpx_idct4x4_1_add_msa>, TX_4X4, 1, 8, 1)
 };

 INSTANTIATE_TEST_CASE_P(MSA, PartialIDctTest,
                        ::testing::ValuesIn(msa_partial_idct_tests));
-#endif  // HAVE_MSA && !CONFIG_EMULATE_HARDWARE && !CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH
+
+#endif  // !CONFIG_EMULATE_HARDWARE

 }  // namespace
--- a/test/pp_filter_test.cc
+++ b/test/pp_filter_test.cc
@@ -11,6 +11,7 @@
 #include "./vpx_config.h"
 #include "./vpx_dsp_rtcd.h"
 #include "test/acm_random.h"
+#include "test/buffer.h"
 #include "test/clear_system_state.h"
 #include "test/register_state_check.h"
 #include "third_party/googletest/src/include/gtest/gtest.h"
@@ -18,6 +19,7 @@
 #include "vpx_mem/vpx_mem.h"

 using libvpx_test::ACMRandom;
+using libvpx_test::Buffer;

 typedef void (*VpxPostProcDownAndAcrossMbRowFunc)(
    unsigned char *src_ptr, unsigned char *dst_ptr, int src_pixels_per_line,
@@ -54,31 +56,16 @@ TEST_P(VpxPostProcDownAndAcrossMbRowTest, CheckFilterOutput) {
  const int block_height = 16;

  // 5-tap filter needs 2 padding rows above and below the block in the input.
-  const int input_width = block_width;
-  const int input_height = block_height + 4;
-  const int input_stride = input_width;
-  const int input_size = input_width * input_height;
+  Buffer<uint8_t> src_image = Buffer<uint8_t>(block_width, block_height, 2);
+  ASSERT_TRUE(src_image.Init());

  // Filter extends output block by 8 samples at left and right edges.
-  const int output_width = block_width + 16;
-  const int output_height = block_height;
-  const int output_stride = output_width;
-  const int output_size = output_width * output_height;
-
-  uint8_t *const src_image = new uint8_t[input_size];
-  ASSERT_TRUE(src_image != NULL);
-
  // Though the left padding is only 8 bytes, the assembly code tries to
  // read 16 bytes before the pointer.
-  uint8_t *const dst_image = new uint8_t[output_size + 8];
-  ASSERT_TRUE(dst_image != NULL);
+  Buffer<uint8_t> dst_image =
+      Buffer<uint8_t>(block_width, block_height, 8, 16, 8, 8);
+  ASSERT_TRUE(dst_image.Init());

-  // Pointers to top-left pixel of block in the input and output images.
-  uint8_t *const src_image_ptr = src_image + (input_stride << 1);
-
-  // The assembly works in increments of 16. The first read may be offset by
-  // this amount.
-  uint8_t *const dst_image_ptr = dst_image + 16;
  uint8_t *const flimits =
      reinterpret_cast<uint8_t *>(vpx_memalign(16, block_width));
  (void)memset(flimits, 255, block_width);
@@ -86,37 +73,29 @@ TEST_P(VpxPostProcDownAndAcrossMbRowTest, CheckFilterOutput) {
  // Initialize pixels in the input:
  //   block pixels to value 1,
  //   border pixels to value 10.
-  (void)memset(src_image, 10, input_size);
-  uint8_t *pixel_ptr = src_image_ptr;
-  for (int i = 0; i < block_height; ++i) {
-    for (int j = 0; j < block_width; ++j) {
-      pixel_ptr[j] = 1;
-    }
-    pixel_ptr += input_stride;
-  }
+  src_image.SetPadding(10);
+  src_image.Set(1);

  // Initialize pixels in the output to 99.
-  (void)memset(dst_image, 99, output_size);
+  dst_image.Set(99);

-  ASM_REGISTER_STATE_CHECK(GetParam()(src_image_ptr, dst_image_ptr,
-                                      input_stride, output_stride, block_width,
-                                      flimits, 16));
+  ASM_REGISTER_STATE_CHECK(GetParam()(
+      src_image.TopLeftPixel(), dst_image.TopLeftPixel(), src_image.stride(),
+      dst_image.stride(), block_width, flimits, 16));

  static const uint8_t kExpectedOutput[block_height] = {
    4, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 4
  };

-  pixel_ptr = dst_image_ptr;
+  uint8_t *pixel_ptr = dst_image.TopLeftPixel();
  for (int i = 0; i < block_height; ++i) {
    for (int j = 0; j < block_width; ++j) {
-      ASSERT_EQ(kExpectedOutput[i], pixel_ptr[j]) << "at (" << i << ", " << j
-                                                  << ")";
+      ASSERT_EQ(kExpectedOutput[i], pixel_ptr[j])
+          << "at (" << i << ", " << j << ")";
    }
-    pixel_ptr += output_stride;
+    pixel_ptr += dst_image.stride();
  }

-  delete[] src_image;
-  delete[] dst_image;
  vpx_free(flimits);
 };

@@ -129,35 +108,20 @@ TEST_P(VpxPostProcDownAndAcrossMbRowTest, CheckCvsAssembly) {

  // 5-tap filter needs 2 padding rows above and below the block in the input.
  // SSE2 reads in blocks of 16. Pad an extra 8 in case the width is not %16.
-  const int input_width = block_width;
-  const int input_height = block_height + 4 + 8;
-  const int input_stride = input_width;
-  const int input_size = input_stride * input_height;
+  Buffer<uint8_t> src_image =
+      Buffer<uint8_t>(block_width, block_height, 2, 2, 10, 2);
+  ASSERT_TRUE(src_image.Init());

  // Filter extends output block by 8 samples at left and right edges.
+  // Though the left padding is only 8 bytes, there is 'above' padding as well
+  // so when the assembly code tries to read 16 bytes before the pointer it is
+  // not a problem.
  // SSE2 reads in blocks of 16. Pad an extra 8 in case the width is not %16.
-  const int output_width = block_width + 24;
-  const int output_height = block_height;
-  const int output_stride = output_width;
-  const int output_size = output_stride * output_height;
-
-  uint8_t *const src_image = new uint8_t[input_size];
-  ASSERT_TRUE(src_image != NULL);
-
-  // Though the left padding is only 8 bytes, the assembly code tries to
-  // read 16 bytes before the pointer.
-  uint8_t *const dst_image = new uint8_t[output_size + 8];
-  ASSERT_TRUE(dst_image != NULL);
-  uint8_t *const dst_image_ref = new uint8_t[output_size + 8];
-  ASSERT_TRUE(dst_image_ref != NULL);
-
-  // Pointers to top-left pixel of block in the input and output images.
-  uint8_t *const src_image_ptr = src_image + (input_stride << 1);
-
-  // The assembly works in increments of 16. The first read may be offset by
-  // this amount.
-  uint8_t *const dst_image_ptr = dst_image + 16;
-  uint8_t *const dst_image_ref_ptr = dst_image + 16;
+  Buffer<uint8_t> dst_image =
+      Buffer<uint8_t>(block_width, block_height, 8, 8, 16, 8);
+  ASSERT_TRUE(dst_image.Init());
+  Buffer<uint8_t> dst_image_ref = Buffer<uint8_t>(block_width, block_height, 8);
+  ASSERT_TRUE(dst_image_ref.Init());

  // Filter values are set in blocks of 16 for Y and 8 for U/V. Each macroblock
  // can have a different filter. SSE2 assembly reads flimits in blocks of 16 so
@@ -171,14 +135,8 @@ TEST_P(VpxPostProcDownAndAcrossMbRowTest, CheckCvsAssembly) {
  // Initialize pixels in the input:
  //   block pixels to random values.
  //   border pixels to value 10.
-  (void)memset(src_image, 10, input_size);
-  uint8_t *pixel_ptr = src_image_ptr;
-  for (int i = 0; i < block_height; ++i) {
-    for (int j = 0; j < block_width; ++j) {
-      pixel_ptr[j] = rnd.Rand8();
-    }
-    pixel_ptr += input_stride;
-  }
+  src_image.SetPadding(10);
+  src_image.Set(&rnd, &ACMRandom::Rand8);

  for (int blocks = 0; blocks < block_width; blocks += 8) {
    (void)memset(flimits, 0, sizeof(*flimits) * flimits_width);
@@ -186,29 +144,22 @@ TEST_P(VpxPostProcDownAndAcrossMbRowTest, CheckCvsAssembly) {
    for (int f = 0; f < 255; f++) {
      (void)memset(flimits + blocks, f, sizeof(*flimits) * 8);

-      (void)memset(dst_image, 0, output_size);
-      (void)memset(dst_image_ref, 0, output_size);
+      dst_image.Set(0);
+      dst_image_ref.Set(0);

      vpx_post_proc_down_and_across_mb_row_c(
-          src_image_ptr, dst_image_ref_ptr, input_stride, output_stride,
-          block_width, flimits, block_height);
-      ASM_REGISTER_STATE_CHECK(GetParam()(src_image_ptr, dst_image_ptr,
-                                          input_stride, output_stride,
-                                          block_width, flimits, 16));
+          src_image.TopLeftPixel(), dst_image_ref.TopLeftPixel(),
+          src_image.stride(), dst_image_ref.stride(), block_width, flimits,
+          block_height);
+      ASM_REGISTER_STATE_CHECK(
+          GetParam()(src_image.TopLeftPixel(), dst_image.TopLeftPixel(),
+                     src_image.stride(), dst_image.stride(), block_width,
+                     flimits, block_height));

-      for (int i = 0; i < block_height; ++i) {
-        for (int j = 0; j < block_width; ++j) {
-          ASSERT_EQ(dst_image_ref_ptr[j + i * output_stride],
-                    dst_image_ptr[j + i * output_stride])
-              << "at (" << i << ", " << j << ")";
-        }
-      }
+      ASSERT_TRUE(dst_image.CheckValues(dst_image_ref));
    }
  }

-  delete[] src_image;
-  delete[] dst_image;
-  delete[] dst_image_ref;
  vpx_free(flimits);
 }

@@ -231,8 +182,8 @@ class VpxMbPostProcAcrossIpTest
                     int rows, int cols, int src_pitch) {
    for (int r = 0; r < rows; r++) {
      for (int c = 0; c < cols; c++) {
-        ASSERT_EQ(expected_output[c], src_c[c]) << "at (" << r << ", " << c
-                                                << ")";
+        ASSERT_EQ(expected_output[c], src_c[c])
+            << "at (" << r << ", " << c << ")";
      }
      src_c += src_pitch;
    }
@@ -249,108 +200,83 @@ class VpxMbPostProcAcrossIpTest
 TEST_P(VpxMbPostProcAcrossIpTest, CheckLowFilterOutput) {
  const int rows = 16;
  const int cols = 16;
-  const int src_left_padding = 8;
-  const int src_right_padding = 17;
-  const int src_width = cols + src_left_padding + src_right_padding;
-  const int src_size = rows * src_width;

-  unsigned char *const src = new unsigned char[src_size];
-  ASSERT_TRUE(src != NULL);
-  memset(src, 10, src_size);
-  unsigned char *const s = src + src_left_padding;
-  SetCols(s, rows, cols, src_width);
+  Buffer<uint8_t> src = Buffer<uint8_t>(cols, rows, 8, 8, 17, 8);
+  ASSERT_TRUE(src.Init());
+  src.SetPadding(10);
+  SetCols(src.TopLeftPixel(), rows, cols, src.stride());

-  unsigned char *expected_output = new unsigned char[rows * cols];
-  ASSERT_TRUE(expected_output != NULL);
-  SetCols(expected_output, rows, cols, cols);
+  Buffer<uint8_t> expected_output = Buffer<uint8_t>(cols, rows, 0);
+  ASSERT_TRUE(expected_output.Init());
+  SetCols(expected_output.TopLeftPixel(), rows, cols, expected_output.stride());

-  RunFilterLevel(s, rows, cols, src_width, q2mbl(0), expected_output);
-  delete[] src;
-  delete[] expected_output;
+  RunFilterLevel(src.TopLeftPixel(), rows, cols, src.stride(), q2mbl(0),
+                 expected_output.TopLeftPixel());
 }

 TEST_P(VpxMbPostProcAcrossIpTest, CheckMediumFilterOutput) {
  const int rows = 16;
  const int cols = 16;
-  const int src_left_padding = 8;
-  const int src_right_padding = 17;
-  const int src_width = cols + src_left_padding + src_right_padding;
-  const int src_size = rows * src_width;

-  unsigned char *const src = new unsigned char[src_size];
-  ASSERT_TRUE(src != NULL);
-  memset(src, 10, src_size);
-  unsigned char *const s = src + src_left_padding;
+  Buffer<uint8_t> src = Buffer<uint8_t>(cols, rows, 8, 8, 17, 8);
+  ASSERT_TRUE(src.Init());
+  src.SetPadding(10);
+  SetCols(src.TopLeftPixel(), rows, cols, src.stride());

-  SetCols(s, rows, cols, src_width);
  static const unsigned char kExpectedOutput[cols] = {
    2, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 13
  };

-  RunFilterLevel(s, rows, cols, src_width, q2mbl(70), kExpectedOutput);
-
-  delete[] src;
+  RunFilterLevel(src.TopLeftPixel(), rows, cols, src.stride(), q2mbl(70),
+                 kExpectedOutput);
 }

 TEST_P(VpxMbPostProcAcrossIpTest, CheckHighFilterOutput) {
  const int rows = 16;
  const int cols = 16;
-  const int src_left_padding = 8;
-  const int src_right_padding = 17;
-  const int src_width = cols + src_left_padding + src_right_padding;
-  const int src_size = rows * src_width;

-  unsigned char *const src = new unsigned char[src_size];
-  ASSERT_TRUE(src != NULL);
-  unsigned char *const s = src + src_left_padding;
+  Buffer<uint8_t> src = Buffer<uint8_t>(cols, rows, 8, 8, 17, 8);
+  ASSERT_TRUE(src.Init());
+  src.SetPadding(10);
+  SetCols(src.TopLeftPixel(), rows, cols, src.stride());

-  memset(src, 10, src_size);
-  SetCols(s, rows, cols, src_width);
  static const unsigned char kExpectedOutput[cols] = {
    2, 2, 3, 4, 4, 5, 6, 7, 8, 9, 10, 11, 11, 12, 13, 13
  };

-  RunFilterLevel(s, rows, cols, src_width, INT_MAX, kExpectedOutput);
+  RunFilterLevel(src.TopLeftPixel(), rows, cols, src.stride(), INT_MAX,
+                 kExpectedOutput);

-  memset(src, 10, src_size);
-  SetCols(s, rows, cols, src_width);
-  RunFilterLevel(s, rows, cols, src_width, q2mbl(100), kExpectedOutput);
+  SetCols(src.TopLeftPixel(), rows, cols, src.stride());

-  delete[] src;
+  RunFilterLevel(src.TopLeftPixel(), rows, cols, src.stride(), q2mbl(100),
+                 kExpectedOutput);
 }

 TEST_P(VpxMbPostProcAcrossIpTest, CheckCvsAssembly) {
  const int rows = 16;
  const int cols = 16;
-  const int src_left_padding = 8;
-  const int src_right_padding = 17;
-  const int src_width = cols + src_left_padding + src_right_padding;
-  const int src_size = rows * src_width;

-  unsigned char *const c_mem = new unsigned char[src_size];
-  unsigned char *const asm_mem = new unsigned char[src_size];
-  ASSERT_TRUE(c_mem != NULL);
-  ASSERT_TRUE(asm_mem != NULL);
-  unsigned char *const src_c = c_mem + src_left_padding;
-  unsigned char *const src_asm = asm_mem + src_left_padding;
+  Buffer<uint8_t> c_mem = Buffer<uint8_t>(cols, rows, 8, 8, 17, 8);
+  ASSERT_TRUE(c_mem.Init());
+  Buffer<uint8_t> asm_mem = Buffer<uint8_t>(cols, rows, 8, 8, 17, 8);
+  ASSERT_TRUE(asm_mem.Init());

  // When level >= 100, the filter behaves the same as the level = INT_MAX
  // When level < 20, it behaves the same as the level = 0
  for (int level = 0; level < 100; level++) {
-    memset(c_mem, 10, src_size);
-    memset(asm_mem, 10, src_size);
-    SetCols(src_c, rows, cols, src_width);
-    SetCols(src_asm, rows, cols, src_width);
+    c_mem.SetPadding(10);
+    asm_mem.SetPadding(10);
+    SetCols(c_mem.TopLeftPixel(), rows, cols, c_mem.stride());
+    SetCols(asm_mem.TopLeftPixel(), rows, cols, asm_mem.stride());

-    vpx_mbpost_proc_across_ip_c(src_c, src_width, rows, cols, q2mbl(level));
-    ASM_REGISTER_STATE_CHECK(
-        GetParam()(src_asm, src_width, rows, cols, q2mbl(level)));
+    vpx_mbpost_proc_across_ip_c(c_mem.TopLeftPixel(), c_mem.stride(), rows,
+                                cols, q2mbl(level));
+    ASM_REGISTER_STATE_CHECK(GetParam()(
+        asm_mem.TopLeftPixel(), asm_mem.stride(), rows, cols, q2mbl(level)));

-    RunComparison(src_c, src_asm, rows, cols, src_width);
+    ASSERT_TRUE(asm_mem.CheckValues(c_mem));
  }
-
-  delete[] c_mem;
-  delete[] asm_mem;
 }

 class VpxMbPostProcDownTest
@@ -359,44 +285,10 @@ class VpxMbPostProcDownTest
  virtual void TearDown() { libvpx_test::ClearSystemState(); }

 protected:
-  void SetRows(unsigned char *src_c, int rows, int cols) {
+  void SetRows(unsigned char *src_c, int rows, int cols, int src_width) {
    for (int r = 0; r < rows; r++) {
      memset(src_c, r, cols);
-      src_c += cols;
-    }
-  }
-
-  void SetRandom(unsigned char *src_c, unsigned char *src_asm, int rows,
-                 int cols, int src_pitch) {
-    ACMRandom rnd;
-    rnd.Reset(ACMRandom::DeterministicSeed());
-
-    // Add some random noise to the input
-    for (int r = 0; r < rows; r++) {
-      for (int c = 0; c < cols; c++) {
-        const int noise = rnd(4);
-        src_c[c] = r + noise;
-        src_asm[c] = r + noise;
-      }
-      src_c += src_pitch;
-      src_asm += src_pitch;
-    }
-  }
-
-  void SetRandomSaturation(unsigned char *src_c, unsigned char *src_asm,
-                           int rows, int cols, int src_pitch) {
-    ACMRandom rnd;
-    rnd.Reset(ACMRandom::DeterministicSeed());
-
-    // Add some random noise to the input
-    for (int r = 0; r < rows; r++) {
-      for (int c = 0; c < cols; c++) {
-        const int noise = 3 * rnd(2);
-        src_c[c] = r + noise;
-        src_asm[c] = r + noise;
-      }
-      src_c += src_pitch;
-      src_asm += src_pitch;
+      src_c += src_width;
    }
  }

@@ -404,24 +296,13 @@ class VpxMbPostProcDownTest
                     int rows, int cols, int src_pitch) {
    for (int r = 0; r < rows; r++) {
      for (int c = 0; c < cols; c++) {
-        ASSERT_EQ(expected_output[r * rows + c], src_c[c]) << "at (" << r
-                                                           << ", " << c << ")";
+        ASSERT_EQ(expected_output[r * rows + c], src_c[c])
+            << "at (" << r << ", " << c << ")";
      }
      src_c += src_pitch;
    }
  }

-  void RunComparison(unsigned char *src_c, unsigned char *src_asm, int rows,
-                     int cols, int src_pitch) {
-    for (int r = 0; r < rows; r++) {
-      for (int c = 0; c < cols; c++) {
-        ASSERT_EQ(src_c[c], src_asm[c]) << "at (" << r << ", " << c << ")";
-      }
-      src_c += src_pitch;
-      src_asm += src_pitch;
-    }
-  }
-
  void RunFilterLevel(unsigned char *s, int rows, int cols, int src_width,
                      int filter_level, const unsigned char *expected_output) {
    ASM_REGISTER_STATE_CHECK(
@@ -433,17 +314,12 @@ class VpxMbPostProcDownTest
 TEST_P(VpxMbPostProcDownTest, CheckHighFilterOutput) {
  const int rows = 16;
  const int cols = 16;
-  const int src_pitch = cols;
-  const int src_top_padding = 8;
-  const int src_bottom_padding = 17;

-  const int src_size = cols * (rows + src_top_padding + src_bottom_padding);
-  unsigned char *const c_mem = new unsigned char[src_size];
-  ASSERT_TRUE(c_mem != NULL);
-  memset(c_mem, 10, src_size);
-  unsigned char *const src_c = c_mem + src_top_padding * src_pitch;
+  Buffer<uint8_t> src_c = Buffer<uint8_t>(cols, rows, 8, 8, 8, 17);
+  ASSERT_TRUE(src_c.Init());
+  src_c.SetPadding(10);

-  SetRows(src_c, rows, cols);
+  SetRows(src_c.TopLeftPixel(), rows, cols, src_c.stride());

  static const unsigned char kExpectedOutput[rows * cols] = {
    2,  2,  1,  1,  2,  2,  2,  2,  2,  2,  1,  1,  2,  2,  2,  2,  2,  2,  2,
@@ -462,29 +338,24 @@ TEST_P(VpxMbPostProcDownTest, CheckHighFilterOutput) {
    13, 13, 13, 13, 14, 13, 13, 13, 13
  };

-  RunFilterLevel(src_c, rows, cols, src_pitch, INT_MAX, kExpectedOutput);
+  RunFilterLevel(src_c.TopLeftPixel(), rows, cols, src_c.stride(), INT_MAX,
+                 kExpectedOutput);

-  memset(c_mem, 10, src_size);
-  SetRows(src_c, rows, cols);
-  RunFilterLevel(src_c, rows, cols, src_pitch, q2mbl(100), kExpectedOutput);
-
-  delete[] c_mem;
+  src_c.SetPadding(10);
+  SetRows(src_c.TopLeftPixel(), rows, cols, src_c.stride());
+  RunFilterLevel(src_c.TopLeftPixel(), rows, cols, src_c.stride(), q2mbl(100),
+                 kExpectedOutput);
 }

 TEST_P(VpxMbPostProcDownTest, CheckMediumFilterOutput) {
  const int rows = 16;
  const int cols = 16;
-  const int src_pitch = cols;
-  const int src_top_padding = 8;
-  const int src_bottom_padding = 17;

-  const int src_size = cols * (rows + src_top_padding + src_bottom_padding);
-  unsigned char *const c_mem = new unsigned char[src_size];
-  ASSERT_TRUE(c_mem != NULL);
-  memset(c_mem, 10, src_size);
-  unsigned char *const src_c = c_mem + src_top_padding * src_pitch;
+  Buffer<uint8_t> src_c = Buffer<uint8_t>(cols, rows, 8, 8, 8, 17);
+  ASSERT_TRUE(src_c.Init());
+  src_c.SetPadding(10);

-  SetRows(src_c, rows, cols);
+  SetRows(src_c.TopLeftPixel(), rows, cols, src_c.stride());

  static const unsigned char kExpectedOutput[rows * cols] = {
    2,  2,  1,  1,  2,  2,  2,  2,  2,  2,  1,  1,  2,  2,  2,  2,  2,  2,  2,
@@ -503,70 +374,65 @@ TEST_P(VpxMbPostProcDownTest, CheckMediumFilterOutput) {
    13, 13, 13, 13, 14, 13, 13, 13, 13
  };

-  RunFilterLevel(src_c, rows, cols, src_pitch, q2mbl(70), kExpectedOutput);
-
-  delete[] c_mem;
+  RunFilterLevel(src_c.TopLeftPixel(), rows, cols, src_c.stride(), q2mbl(70),
+                 kExpectedOutput);
 }

 TEST_P(VpxMbPostProcDownTest, CheckLowFilterOutput) {
  const int rows = 16;
  const int cols = 16;
-  const int src_pitch = cols;
-  const int src_top_padding = 8;
-  const int src_bottom_padding = 17;

-  const int src_size = cols * (rows + src_top_padding + src_bottom_padding);
-  unsigned char *const c_mem = new unsigned char[src_size];
-  ASSERT_TRUE(c_mem != NULL);
-  memset(c_mem, 10, src_size);
-  unsigned char *const src_c = c_mem + src_top_padding * src_pitch;
+  Buffer<uint8_t> src_c = Buffer<uint8_t>(cols, rows, 8, 8, 8, 17);
+  ASSERT_TRUE(src_c.Init());
+  src_c.SetPadding(10);

-  SetRows(src_c, rows, cols);
+  SetRows(src_c.TopLeftPixel(), rows, cols, src_c.stride());

  unsigned char *expected_output = new unsigned char[rows * cols];
  ASSERT_TRUE(expected_output != NULL);
-  SetRows(expected_output, rows, cols);
+  SetRows(expected_output, rows, cols, cols);

-  RunFilterLevel(src_c, rows, cols, src_pitch, q2mbl(0), expected_output);
+  RunFilterLevel(src_c.TopLeftPixel(), rows, cols, src_c.stride(), q2mbl(0),
+                 expected_output);

-  delete[] c_mem;
  delete[] expected_output;
 }

 TEST_P(VpxMbPostProcDownTest, CheckCvsAssembly) {
  const int rows = 16;
  const int cols = 16;
-  const int src_pitch = cols;
-  const int src_top_padding = 8;
-  const int src_bottom_padding = 17;
-  const int src_size = cols * (rows + src_top_padding + src_bottom_padding);
-  unsigned char *const c_mem = new unsigned char[src_size];
-  unsigned char *const asm_mem = new unsigned char[src_size];
-  ASSERT_TRUE(c_mem != NULL);
-  ASSERT_TRUE(asm_mem != NULL);
-  unsigned char *const src_c = c_mem + src_top_padding * src_pitch;
-  unsigned char *const src_asm = asm_mem + src_top_padding * src_pitch;
+
+  ACMRandom rnd;
+  rnd.Reset(ACMRandom::DeterministicSeed());
+
+  Buffer<uint8_t> src_c = Buffer<uint8_t>(cols, rows, 8, 8, 8, 17);
+  ASSERT_TRUE(src_c.Init());
+  Buffer<uint8_t> src_asm = Buffer<uint8_t>(cols, rows, 8, 8, 8, 17);
+  ASSERT_TRUE(src_asm.Init());

  for (int level = 0; level < 100; level++) {
-    memset(c_mem, 10, src_size);
-    memset(asm_mem, 10, src_size);
-    SetRandom(src_c, src_asm, rows, cols, src_pitch);
-    vpx_mbpost_proc_down_c(src_c, src_pitch, rows, cols, q2mbl(level));
-    ASM_REGISTER_STATE_CHECK(
-        GetParam()(src_asm, src_pitch, rows, cols, q2mbl(level)));
-    RunComparison(src_c, src_asm, rows, cols, src_pitch);
+    src_c.SetPadding(10);
+    src_asm.SetPadding(10);
+    src_c.Set(&rnd, &ACMRandom::Rand8);
+    src_asm.CopyFrom(src_c);

-    memset(c_mem, 10, src_size);
-    memset(asm_mem, 10, src_size);
-    SetRandomSaturation(src_c, src_asm, rows, cols, src_pitch);
-    vpx_mbpost_proc_down_c(src_c, src_pitch, rows, cols, q2mbl(level));
-    ASM_REGISTER_STATE_CHECK(
-        GetParam()(src_asm, src_pitch, rows, cols, q2mbl(level)));
-    RunComparison(src_c, src_asm, rows, cols, src_pitch);
+    vpx_mbpost_proc_down_c(src_c.TopLeftPixel(), src_c.stride(), rows, cols,
+                           q2mbl(level));
+    ASM_REGISTER_STATE_CHECK(GetParam()(
+        src_asm.TopLeftPixel(), src_asm.stride(), rows, cols, q2mbl(level)));
+    ASSERT_TRUE(src_asm.CheckValues(src_c));
+
+    src_c.SetPadding(10);
+    src_asm.SetPadding(10);
+    src_c.Set(&rnd, &ACMRandom::Rand8Extremes);
+    src_asm.CopyFrom(src_c);
+
+    vpx_mbpost_proc_down_c(src_c.TopLeftPixel(), src_c.stride(), rows, cols,
+                           q2mbl(level));
+    ASM_REGISTER_STATE_CHECK(GetParam()(
+        src_asm.TopLeftPixel(), src_asm.stride(), rows, cols, q2mbl(level)));
+    ASSERT_TRUE(src_asm.CheckValues(src_c));
  }
-
-  delete[] c_mem;
-  delete[] asm_mem;
 }

 INSTANTIATE_TEST_CASE_P(
@@ -598,6 +464,9 @@ INSTANTIATE_TEST_CASE_P(

 INSTANTIATE_TEST_CASE_P(NEON, VpxMbPostProcAcrossIpTest,
                        ::testing::Values(vpx_mbpost_proc_across_ip_neon));
+
+INSTANTIATE_TEST_CASE_P(NEON, VpxMbPostProcDownTest,
+                        ::testing::Values(vpx_mbpost_proc_down_neon));
 #endif  // HAVE_NEON

 #if HAVE_MSA
--- a/test/predict_test.cc
+++ b/test/predict_test.cc
@@ -324,6 +324,15 @@ INSTANTIATE_TEST_CASE_P(
                      make_tuple(4, 4, &vp8_sixtap_predict4x4_msa)));
 #endif

+#if HAVE_MMI
+INSTANTIATE_TEST_CASE_P(
+    MMI, SixtapPredictTest,
+    ::testing::Values(make_tuple(16, 16, &vp8_sixtap_predict16x16_mmi),
+                      make_tuple(8, 8, &vp8_sixtap_predict8x8_mmi),
+                      make_tuple(8, 4, &vp8_sixtap_predict8x4_mmi),
+                      make_tuple(4, 4, &vp8_sixtap_predict4x4_mmi)));
+#endif
+
 class BilinearPredictTest : public PredictTestBase {};

 TEST_P(BilinearPredictTest, TestWithRandomData) {
--- a/test/quantize_test.cc
+++ b/test/quantize_test.cc
@@ -200,4 +200,12 @@ INSTANTIATE_TEST_CASE_P(
        make_tuple(&vp8_fast_quantize_b_msa, &vp8_fast_quantize_b_c),
        make_tuple(&vp8_regular_quantize_b_msa, &vp8_regular_quantize_b_c)));
 #endif  // HAVE_MSA
+
+#if HAVE_MMI
+INSTANTIATE_TEST_CASE_P(
+    MMI, QuantizeTest,
+    ::testing::Values(
+        make_tuple(&vp8_fast_quantize_b_mmi, &vp8_fast_quantize_b_c),
+        make_tuple(&vp8_regular_quantize_b_mmi, &vp8_regular_quantize_b_c)));
+#endif  // HAVE_MMI
 }  // namespace
--- a/test/register_state_check.h
+++ b/test/register_state_check.h
@@ -113,8 +113,8 @@ class RegisterStateCheck {
    int64_t post_store[8];
    vpx_push_neon(post_store);
    for (int i = 0; i < 8; ++i) {
-      EXPECT_EQ(pre_store_[i], post_store[i]) << "d" << i + 8
-                                              << " has been modified";
+      EXPECT_EQ(pre_store_[i], post_store[i])
+          << "d" << i + 8 << " has been modified";
    }
  }

--- a/test/resize_test.cc
+++ b/test/resize_test.cc
@@ -298,10 +298,10 @@ TEST_P(ResizeTest, TestExternalResizeWorks) {
    unsigned int expected_h;
    ScaleForFrameNumber(frame, kInitialWidth, kInitialHeight, &expected_w,
                        &expected_h, 0);
-    EXPECT_EQ(expected_w, info->w) << "Frame " << frame
-                                   << " had unexpected width";
-    EXPECT_EQ(expected_h, info->h) << "Frame " << frame
-                                   << " had unexpected height";
+    EXPECT_EQ(expected_w, info->w)
+        << "Frame " << frame << " had unexpected width";
+    EXPECT_EQ(expected_h, info->h)
+        << "Frame " << frame << " had unexpected height";
  }
 }

@@ -513,10 +513,10 @@ TEST_P(ResizeRealtimeTest, TestExternalResizeWorks) {
    unsigned int expected_h;
    ScaleForFrameNumber(frame, kInitialWidth, kInitialHeight, &expected_w,
                        &expected_h, 1);
-    EXPECT_EQ(expected_w, info->w) << "Frame " << frame
-                                   << " had unexpected width";
-    EXPECT_EQ(expected_h, info->h) << "Frame " << frame
-                                   << " had unexpected height";
+    EXPECT_EQ(expected_w, info->w)
+        << "Frame " << frame << " had unexpected width";
+    EXPECT_EQ(expected_h, info->h)
+        << "Frame " << frame << " had unexpected height";
    EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
  }
 }
--- a/test/sad_test.cc
+++ b/test/sad_test.cc
@@ -644,19 +644,50 @@ INSTANTIATE_TEST_CASE_P(C, SADx4Test, ::testing::ValuesIn(x4d_c_tests));
 #if HAVE_NEON
 const SadMxNParam neon_tests[] = {
  SadMxNParam(64, 64, &vpx_sad64x64_neon),
+  SadMxNParam(64, 32, &vpx_sad64x32_neon),
  SadMxNParam(32, 32, &vpx_sad32x32_neon),
+  SadMxNParam(16, 32, &vpx_sad16x32_neon),
  SadMxNParam(16, 16, &vpx_sad16x16_neon),
  SadMxNParam(16, 8, &vpx_sad16x8_neon),
  SadMxNParam(8, 16, &vpx_sad8x16_neon),
  SadMxNParam(8, 8, &vpx_sad8x8_neon),
+  SadMxNParam(8, 4, &vpx_sad8x4_neon),
+  SadMxNParam(4, 8, &vpx_sad4x8_neon),
  SadMxNParam(4, 4, &vpx_sad4x4_neon),
 };
 INSTANTIATE_TEST_CASE_P(NEON, SADTest, ::testing::ValuesIn(neon_tests));

+const SadMxNAvgParam avg_neon_tests[] = {
+  SadMxNAvgParam(64, 64, &vpx_sad64x64_avg_neon),
+  SadMxNAvgParam(64, 32, &vpx_sad64x32_avg_neon),
+  SadMxNAvgParam(32, 64, &vpx_sad32x64_avg_neon),
+  SadMxNAvgParam(32, 32, &vpx_sad32x32_avg_neon),
+  SadMxNAvgParam(32, 16, &vpx_sad32x16_avg_neon),
+  SadMxNAvgParam(16, 32, &vpx_sad16x32_avg_neon),
+  SadMxNAvgParam(16, 16, &vpx_sad16x16_avg_neon),
+  SadMxNAvgParam(16, 8, &vpx_sad16x8_avg_neon),
+  SadMxNAvgParam(8, 16, &vpx_sad8x16_avg_neon),
+  SadMxNAvgParam(8, 8, &vpx_sad8x8_avg_neon),
+  SadMxNAvgParam(8, 4, &vpx_sad8x4_avg_neon),
+  SadMxNAvgParam(4, 8, &vpx_sad4x8_avg_neon),
+  SadMxNAvgParam(4, 4, &vpx_sad4x4_avg_neon),
+};
+INSTANTIATE_TEST_CASE_P(NEON, SADavgTest, ::testing::ValuesIn(avg_neon_tests));
+
 const SadMxNx4Param x4d_neon_tests[] = {
  SadMxNx4Param(64, 64, &vpx_sad64x64x4d_neon),
+  SadMxNx4Param(64, 32, &vpx_sad64x32x4d_neon),
+  SadMxNx4Param(32, 64, &vpx_sad32x64x4d_neon),
  SadMxNx4Param(32, 32, &vpx_sad32x32x4d_neon),
+  SadMxNx4Param(32, 16, &vpx_sad32x16x4d_neon),
+  SadMxNx4Param(16, 32, &vpx_sad16x32x4d_neon),
  SadMxNx4Param(16, 16, &vpx_sad16x16x4d_neon),
+  SadMxNx4Param(16, 8, &vpx_sad16x8x4d_neon),
+  SadMxNx4Param(8, 16, &vpx_sad8x16x4d_neon),
+  SadMxNx4Param(8, 8, &vpx_sad8x8x4d_neon),
+  SadMxNx4Param(8, 4, &vpx_sad8x4x4d_neon),
+  SadMxNx4Param(4, 8, &vpx_sad4x8x4d_neon),
+  SadMxNx4Param(4, 4, &vpx_sad4x4x4d_neon),
 };
 INSTANTIATE_TEST_CASE_P(NEON, SADx4Test, ::testing::ValuesIn(x4d_neon_tests));
 #endif  // HAVE_NEON
@@ -865,6 +896,14 @@ const SadMxNx4Param x4d_avx2_tests[] = {
 INSTANTIATE_TEST_CASE_P(AVX2, SADx4Test, ::testing::ValuesIn(x4d_avx2_tests));
 #endif  // HAVE_AVX2

+#if HAVE_AVX512
+const SadMxNx4Param x4d_avx512_tests[] = {
+  SadMxNx4Param(64, 64, &vpx_sad64x64x4d_avx512),
+};
+INSTANTIATE_TEST_CASE_P(AVX512, SADx4Test,
+                        ::testing::ValuesIn(x4d_avx512_tests));
+#endif  // HAVE_AVX512
+
 //------------------------------------------------------------------------------
 // MIPS functions
 #if HAVE_MSA
@@ -920,4 +959,98 @@ const SadMxNx4Param x4d_msa_tests[] = {
 INSTANTIATE_TEST_CASE_P(MSA, SADx4Test, ::testing::ValuesIn(x4d_msa_tests));
 #endif  // HAVE_MSA

+//------------------------------------------------------------------------------
+// VSX functions
+#if HAVE_VSX
+const SadMxNParam vsx_tests[] = {
+  SadMxNParam(64, 64, &vpx_sad64x64_vsx),
+  SadMxNParam(64, 32, &vpx_sad64x32_vsx),
+  SadMxNParam(32, 64, &vpx_sad32x64_vsx),
+  SadMxNParam(32, 32, &vpx_sad32x32_vsx),
+  SadMxNParam(32, 16, &vpx_sad32x16_vsx),
+  SadMxNParam(16, 32, &vpx_sad16x32_vsx),
+  SadMxNParam(16, 16, &vpx_sad16x16_vsx),
+  SadMxNParam(16, 8, &vpx_sad16x8_vsx),
+};
+INSTANTIATE_TEST_CASE_P(VSX, SADTest, ::testing::ValuesIn(vsx_tests));
+
+const SadMxNAvgParam avg_vsx_tests[] = {
+  SadMxNAvgParam(64, 64, &vpx_sad64x64_avg_vsx),
+  SadMxNAvgParam(64, 32, &vpx_sad64x32_avg_vsx),
+  SadMxNAvgParam(32, 64, &vpx_sad32x64_avg_vsx),
+  SadMxNAvgParam(32, 32, &vpx_sad32x32_avg_vsx),
+  SadMxNAvgParam(32, 16, &vpx_sad32x16_avg_vsx),
+  SadMxNAvgParam(16, 32, &vpx_sad16x32_avg_vsx),
+  SadMxNAvgParam(16, 16, &vpx_sad16x16_avg_vsx),
+  SadMxNAvgParam(16, 8, &vpx_sad16x8_avg_vsx),
+};
+INSTANTIATE_TEST_CASE_P(VSX, SADavgTest, ::testing::ValuesIn(avg_vsx_tests));
+
+const SadMxNx4Param x4d_vsx_tests[] = {
+  SadMxNx4Param(64, 64, &vpx_sad64x64x4d_vsx),
+  SadMxNx4Param(64, 32, &vpx_sad64x32x4d_vsx),
+  SadMxNx4Param(32, 64, &vpx_sad32x64x4d_vsx),
+  SadMxNx4Param(32, 32, &vpx_sad32x32x4d_vsx),
+  SadMxNx4Param(32, 16, &vpx_sad32x16x4d_vsx),
+  SadMxNx4Param(16, 32, &vpx_sad16x32x4d_vsx),
+  SadMxNx4Param(16, 16, &vpx_sad16x16x4d_vsx),
+  SadMxNx4Param(16, 8, &vpx_sad16x8x4d_vsx),
+};
+INSTANTIATE_TEST_CASE_P(VSX, SADx4Test, ::testing::ValuesIn(x4d_vsx_tests));
+#endif  // HAVE_VSX
+
+//------------------------------------------------------------------------------
+// Loongson functions
+#if HAVE_MMI
+const SadMxNParam mmi_tests[] = {
+  SadMxNParam(64, 64, &vpx_sad64x64_mmi),
+  SadMxNParam(64, 32, &vpx_sad64x32_mmi),
+  SadMxNParam(32, 64, &vpx_sad32x64_mmi),
+  SadMxNParam(32, 32, &vpx_sad32x32_mmi),
+  SadMxNParam(32, 16, &vpx_sad32x16_mmi),
+  SadMxNParam(16, 32, &vpx_sad16x32_mmi),
+  SadMxNParam(16, 16, &vpx_sad16x16_mmi),
+  SadMxNParam(16, 8, &vpx_sad16x8_mmi),
+  SadMxNParam(8, 16, &vpx_sad8x16_mmi),
+  SadMxNParam(8, 8, &vpx_sad8x8_mmi),
+  SadMxNParam(8, 4, &vpx_sad8x4_mmi),
+  SadMxNParam(4, 8, &vpx_sad4x8_mmi),
+  SadMxNParam(4, 4, &vpx_sad4x4_mmi),
+};
+INSTANTIATE_TEST_CASE_P(MMI, SADTest, ::testing::ValuesIn(mmi_tests));
+
+const SadMxNAvgParam avg_mmi_tests[] = {
+  SadMxNAvgParam(64, 64, &vpx_sad64x64_avg_mmi),
+  SadMxNAvgParam(64, 32, &vpx_sad64x32_avg_mmi),
+  SadMxNAvgParam(32, 64, &vpx_sad32x64_avg_mmi),
+  SadMxNAvgParam(32, 32, &vpx_sad32x32_avg_mmi),
+  SadMxNAvgParam(32, 16, &vpx_sad32x16_avg_mmi),
+  SadMxNAvgParam(16, 32, &vpx_sad16x32_avg_mmi),
+  SadMxNAvgParam(16, 16, &vpx_sad16x16_avg_mmi),
+  SadMxNAvgParam(16, 8, &vpx_sad16x8_avg_mmi),
+  SadMxNAvgParam(8, 16, &vpx_sad8x16_avg_mmi),
+  SadMxNAvgParam(8, 8, &vpx_sad8x8_avg_mmi),
+  SadMxNAvgParam(8, 4, &vpx_sad8x4_avg_mmi),
+  SadMxNAvgParam(4, 8, &vpx_sad4x8_avg_mmi),
+  SadMxNAvgParam(4, 4, &vpx_sad4x4_avg_mmi),
+};
+INSTANTIATE_TEST_CASE_P(MMI, SADavgTest, ::testing::ValuesIn(avg_mmi_tests));
+
+const SadMxNx4Param x4d_mmi_tests[] = {
+  SadMxNx4Param(64, 64, &vpx_sad64x64x4d_mmi),
+  SadMxNx4Param(64, 32, &vpx_sad64x32x4d_mmi),
+  SadMxNx4Param(32, 64, &vpx_sad32x64x4d_mmi),
+  SadMxNx4Param(32, 32, &vpx_sad32x32x4d_mmi),
+  SadMxNx4Param(32, 16, &vpx_sad32x16x4d_mmi),
+  SadMxNx4Param(16, 32, &vpx_sad16x32x4d_mmi),
+  SadMxNx4Param(16, 16, &vpx_sad16x16x4d_mmi),
+  SadMxNx4Param(16, 8, &vpx_sad16x8x4d_mmi),
+  SadMxNx4Param(8, 16, &vpx_sad8x16x4d_mmi),
+  SadMxNx4Param(8, 8, &vpx_sad8x8x4d_mmi),
+  SadMxNx4Param(8, 4, &vpx_sad8x4x4d_mmi),
+  SadMxNx4Param(4, 8, &vpx_sad4x8x4d_mmi),
+  SadMxNx4Param(4, 4, &vpx_sad4x4x4d_mmi),
+};
+INSTANTIATE_TEST_CASE_P(MMI, SADx4Test, ::testing::ValuesIn(x4d_mmi_tests));
+#endif  // HAVE_MMI
 }  // namespace
--- a/test/set_roi.cc
+++ b/test/set_roi.cc
@@ -146,14 +146,6 @@ TEST(VP8RoiMapTest, ParameterCheck) {
      if (deltas_valid != roi_retval) break;
    }

-    // Test that we report and error if cyclic refresh is enabled.
-    cpi.cyclic_refresh_mode_enabled = 1;
-    roi_retval =
-        vp8_set_roimap(&cpi, roi_map, cpi.common.mb_rows, cpi.common.mb_cols,
-                       delta_q, delta_lf, threshold);
-    EXPECT_EQ(-1, roi_retval) << "cyclic refresh check error";
-    cpi.cyclic_refresh_mode_enabled = 0;
-
    // Test invalid number of rows or colums.
    roi_retval =
        vp8_set_roimap(&cpi, roi_map, cpi.common.mb_rows + 1,
--- a/test/stress.sh
+++ b/test/stress.sh
@@ -8,8 +8,10 @@
 ##  in the file PATENTS.  All contributing project authors may
 ##  be found in the AUTHORS file in the root of the source tree.
 ##
-##  This file performs a stress test. It runs 5 encodes and 30 decodes in
-##  parallel.
+##  This file performs a stress test. It runs (STRESS_ONEPASS_MAX_JOBS,
+##  default=5) one, (STRESS_TWOPASS_MAX_JOBS, default=5) two pass &
+##  (STRESS_RT_MAX_JOBS, default=5) encodes and (STRESS_<codec>_DECODE_MAX_JOBS,
+##  default=30) decodes in parallel.

 . $(dirname $0)/tools_common.sh

@@ -75,20 +77,33 @@ stress() {
  local readonly codec="$1"
  local readonly webm="$2"
  local readonly decode_count="$3"
+  local readonly threads="$4"
+  local readonly enc_args="$5"
  local pids=""
  local rt_max_jobs=${STRESS_RT_MAX_JOBS:-5}
+  local onepass_max_jobs=${STRESS_ONEPASS_MAX_JOBS:-5}
  local twopass_max_jobs=${STRESS_TWOPASS_MAX_JOBS:-5}

  # Enable job control, so we can run multiple processes.
  set -m

+  # Start $onepass_max_jobs encode jobs in parallel.
+  for i in $(seq ${onepass_max_jobs}); do
+    bitrate=$(($i * 20 + 300))
+    eval "${VPX_TEST_PREFIX}" "${encoder}" "--codec=${codec} -w 1280 -h 720" \
+      "${YUV}" "-t ${threads} --limit=150 --test-decode=fatal --passes=1" \
+      "--target-bitrate=${bitrate} -o ${VPX_TEST_OUTPUT_DIR}/${i}.1pass.webm" \
+      "${enc_args}" ${devnull} &
+    pids="${pids} $!"
+  done
+
  # Start $twopass_max_jobs encode jobs in parallel.
  for i in $(seq ${twopass_max_jobs}); do
    bitrate=$(($i * 20 + 300))
    eval "${VPX_TEST_PREFIX}" "${encoder}" "--codec=${codec} -w 1280 -h 720" \
-      "${YUV}" "-t 4 --limit=150 --test-decode=fatal " \
-      "--target-bitrate=${bitrate} -o ${VPX_TEST_OUTPUT_DIR}/${i}.webm" \
-      ${devnull} &
+      "${YUV}" "-t ${threads} --limit=150 --test-decode=fatal --passes=2" \
+      "--target-bitrate=${bitrate} -o ${VPX_TEST_OUTPUT_DIR}/${i}.2pass.webm" \
+      "${enc_args}" ${devnull} &
    pids="${pids} $!"
  done

@@ -96,7 +111,7 @@ stress() {
  for i in $(seq ${rt_max_jobs}); do
    bitrate=$(($i * 20 + 300))
    eval "${VPX_TEST_PREFIX}" "${encoder}" "--codec=${codec} -w 1280 -h 720" \
-      "${YUV}" "-t 4 --limit=150 --test-decode=fatal " \
+      "${YUV}" "-t ${threads} --limit=150 --test-decode=fatal " \
      "--target-bitrate=${bitrate} --lag-in-frames=0 --error-resilient=1" \
      "--kf-min-dist=3000 --kf-max-dist=3000 --cpu-used=-6 --static-thresh=1" \
      "--end-usage=cbr --min-q=2 --max-q=56 --undershoot-pct=100" \
@@ -109,7 +124,7 @@ stress() {

  # Start $decode_count decode jobs in parallel.
  for i in $(seq "${decode_count}"); do
-    eval "${decoder}" "-t 4" "${webm}" "--noblit" ${devnull} &
+    eval "${decoder}" "-t ${threads}" "${webm}" "--noblit" ${devnull} &
    pids="${pids} $!"
  done

@@ -125,17 +140,30 @@ vp8_stress_test() {
  local vp8_max_jobs=${STRESS_VP8_DECODE_MAX_JOBS:-40}
  if [ "$(vp8_decode_available)" = "yes" -a \
       "$(vp8_encode_available)" = "yes" ]; then
-    stress vp8 "${VP8}" "${vp8_max_jobs}"
+    stress vp8 "${VP8}" "${vp8_max_jobs}" 4
  fi
 }

-vp9_stress_test() {
+vp9_stress() {
  local vp9_max_jobs=${STRESS_VP9_DECODE_MAX_JOBS:-25}

  if [ "$(vp9_decode_available)" = "yes" -a \
       "$(vp9_encode_available)" = "yes" ]; then
-    stress vp9 "${VP9}" "${vp9_max_jobs}"
+    stress vp9 "${VP9}" "${vp9_max_jobs}" "$@"
  fi
 }

-run_tests stress_verify_environment "vp8_stress_test vp9_stress_test"
+vp9_stress_test() {
+  for threads in 4 8 100; do
+    vp9_stress "$threads" "--row-mt=0"
+  done
+}
+
+vp9_stress_test_row_mt() {
+  for threads in 4 8 100; do
+    vp9_stress "$threads" "--row-mt=1"
+  done
+}
+
+run_tests stress_verify_environment \
+  "vp8_stress_test vp9_stress_test vp9_stress_test_row_mt"
--- a/test/sum_squares_test.cc
+++ b/test/sum_squares_test.cc
@@ -110,4 +110,10 @@ INSTANTIATE_TEST_CASE_P(
    ::testing::Values(make_tuple(&vpx_sum_squares_2d_i16_c,
                                 &vpx_sum_squares_2d_i16_sse2)));
 #endif  // HAVE_SSE2
+
+#if HAVE_MSA
+INSTANTIATE_TEST_CASE_P(MSA, SumSquaresTest, ::testing::Values(make_tuple(
+                                                 &vpx_sum_squares_2d_i16_c,
+                                                 &vpx_sum_squares_2d_i16_msa)));
+#endif  // HAVE_MSA
 }  // namespace
--- a/test/temporal_filter_test.cc
+++ b/test/temporal_filter_test.cc
@@ -0,0 +1,277 @@
+/*
+ *  Copyright (c) 2016 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 <limits>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vp9_rtcd.h"
+#include "test/acm_random.h"
+#include "test/buffer.h"
+#include "test/register_state_check.h"
+#include "vpx_ports/vpx_timer.h"
+
+namespace {
+
+using ::libvpx_test::ACMRandom;
+using ::libvpx_test::Buffer;
+
+typedef void (*TemporalFilterFunc)(const uint8_t *a, unsigned int stride,
+                                   const uint8_t *b, unsigned int w,
+                                   unsigned int h, int filter_strength,
+                                   int filter_weight, unsigned int *accumulator,
+                                   uint16_t *count);
+
+// Calculate the difference between 'a' and 'b', sum in blocks of 9, and apply
+// filter based on strength and weight. Store the resulting filter amount in
+// 'count' and apply it to 'b' and store it in 'accumulator'.
+void reference_filter(const Buffer<uint8_t> &a, const Buffer<uint8_t> &b, int w,
+                      int h, int filter_strength, int filter_weight,
+                      Buffer<unsigned int> *accumulator,
+                      Buffer<uint16_t> *count) {
+  Buffer<int> diff_sq = Buffer<int>(w, h, 0);
+  ASSERT_TRUE(diff_sq.Init());
+  diff_sq.Set(0);
+
+  int rounding = 0;
+  if (filter_strength > 0) {
+    rounding = 1 << (filter_strength - 1);
+  }
+
+  // Calculate all the differences. Avoids re-calculating a bunch of extra
+  // values.
+  for (int height = 0; height < h; ++height) {
+    for (int width = 0; width < w; ++width) {
+      int diff = a.TopLeftPixel()[height * a.stride() + width] -
+                 b.TopLeftPixel()[height * b.stride() + width];
+      diff_sq.TopLeftPixel()[height * diff_sq.stride() + width] = diff * diff;
+    }
+  }
+
+  // For any given point, sum the neighboring values and calculate the
+  // modifier.
+  for (int height = 0; height < h; ++height) {
+    for (int width = 0; width < w; ++width) {
+      // Determine how many values are being summed.
+      int summed_values = 9;
+
+      if (height == 0 || height == (h - 1)) {
+        summed_values -= 3;
+      }
+
+      if (width == 0 || width == (w - 1)) {
+        if (summed_values == 6) {  // corner
+          summed_values -= 2;
+        } else {
+          summed_values -= 3;
+        }
+      }
+
+      // Sum the diff_sq of the surrounding values.
+      int sum = 0;
+      for (int idy = -1; idy <= 1; ++idy) {
+        for (int idx = -1; idx <= 1; ++idx) {
+          const int y = height + idy;
+          const int x = width + idx;
+
+          // If inside the border.
+          if (y >= 0 && y < h && x >= 0 && x < w) {
+            sum += diff_sq.TopLeftPixel()[y * diff_sq.stride() + x];
+          }
+        }
+      }
+
+      sum *= 3;
+      sum /= summed_values;
+      sum += rounding;
+      sum >>= filter_strength;
+
+      // Clamp the value and invert it.
+      if (sum > 16) sum = 16;
+      sum = 16 - sum;
+
+      sum *= filter_weight;
+
+      count->TopLeftPixel()[height * count->stride() + width] += sum;
+      accumulator->TopLeftPixel()[height * accumulator->stride() + width] +=
+          sum * b.TopLeftPixel()[height * b.stride() + width];
+    }
+  }
+}
+
+class TemporalFilterTest : public ::testing::TestWithParam<TemporalFilterFunc> {
+ public:
+  virtual void SetUp() {
+    filter_func_ = GetParam();
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+  }
+
+ protected:
+  TemporalFilterFunc filter_func_;
+  ACMRandom rnd_;
+};
+
+TEST_P(TemporalFilterTest, SizeCombinations) {
+  // Depending on subsampling this function may be called with values of 8 or 16
+  // for width and height, in any combination.
+  Buffer<uint8_t> a = Buffer<uint8_t>(16, 16, 8);
+  ASSERT_TRUE(a.Init());
+
+  const int filter_weight = 2;
+  const int filter_strength = 6;
+
+  for (int width = 8; width <= 16; width += 8) {
+    for (int height = 8; height <= 16; height += 8) {
+      // The second buffer must not have any border.
+      Buffer<uint8_t> b = Buffer<uint8_t>(width, height, 0);
+      ASSERT_TRUE(b.Init());
+      Buffer<unsigned int> accum_ref = Buffer<unsigned int>(width, height, 0);
+      ASSERT_TRUE(accum_ref.Init());
+      Buffer<unsigned int> accum_chk = Buffer<unsigned int>(width, height, 0);
+      ASSERT_TRUE(accum_chk.Init());
+      Buffer<uint16_t> count_ref = Buffer<uint16_t>(width, height, 0);
+      ASSERT_TRUE(count_ref.Init());
+      Buffer<uint16_t> count_chk = Buffer<uint16_t>(width, height, 0);
+      ASSERT_TRUE(count_chk.Init());
+
+      // The difference between the buffers must be small to pass the threshold
+      // to apply the filter.
+      a.Set(&rnd_, 0, 7);
+      b.Set(&rnd_, 0, 7);
+
+      accum_ref.Set(rnd_.Rand8());
+      accum_chk.CopyFrom(accum_ref);
+      count_ref.Set(rnd_.Rand8());
+      count_chk.CopyFrom(count_ref);
+      reference_filter(a, b, width, height, filter_strength, filter_weight,
+                       &accum_ref, &count_ref);
+      ASM_REGISTER_STATE_CHECK(
+          filter_func_(a.TopLeftPixel(), a.stride(), b.TopLeftPixel(), width,
+                       height, filter_strength, filter_weight,
+                       accum_chk.TopLeftPixel(), count_chk.TopLeftPixel()));
+      EXPECT_TRUE(accum_chk.CheckValues(accum_ref));
+      EXPECT_TRUE(count_chk.CheckValues(count_ref));
+      if (HasFailure()) {
+        printf("Width: %d Height: %d\n", width, height);
+        count_chk.PrintDifference(count_ref);
+        accum_chk.PrintDifference(accum_ref);
+        return;
+      }
+    }
+  }
+}
+
+TEST_P(TemporalFilterTest, CompareReferenceRandom) {
+  for (int width = 8; width <= 16; width += 8) {
+    for (int height = 8; height <= 16; height += 8) {
+      Buffer<uint8_t> a = Buffer<uint8_t>(width, height, 8);
+      ASSERT_TRUE(a.Init());
+      // The second buffer must not have any border.
+      Buffer<uint8_t> b = Buffer<uint8_t>(width, height, 0);
+      ASSERT_TRUE(b.Init());
+      Buffer<unsigned int> accum_ref = Buffer<unsigned int>(width, height, 0);
+      ASSERT_TRUE(accum_ref.Init());
+      Buffer<unsigned int> accum_chk = Buffer<unsigned int>(width, height, 0);
+      ASSERT_TRUE(accum_chk.Init());
+      Buffer<uint16_t> count_ref = Buffer<uint16_t>(width, height, 0);
+      ASSERT_TRUE(count_ref.Init());
+      Buffer<uint16_t> count_chk = Buffer<uint16_t>(width, height, 0);
+      ASSERT_TRUE(count_chk.Init());
+
+      for (int filter_strength = 0; filter_strength <= 6; ++filter_strength) {
+        for (int filter_weight = 0; filter_weight <= 2; ++filter_weight) {
+          for (int repeat = 0; repeat < 100; ++repeat) {
+            if (repeat < 50) {
+              a.Set(&rnd_, 0, 7);
+              b.Set(&rnd_, 0, 7);
+            } else {
+              // Check large (but close) values as well.
+              a.Set(&rnd_, std::numeric_limits<uint8_t>::max() - 7,
+                    std::numeric_limits<uint8_t>::max());
+              b.Set(&rnd_, std::numeric_limits<uint8_t>::max() - 7,
+                    std::numeric_limits<uint8_t>::max());
+            }
+
+            accum_ref.Set(rnd_.Rand8());
+            accum_chk.CopyFrom(accum_ref);
+            count_ref.Set(rnd_.Rand8());
+            count_chk.CopyFrom(count_ref);
+            reference_filter(a, b, width, height, filter_strength,
+                             filter_weight, &accum_ref, &count_ref);
+            ASM_REGISTER_STATE_CHECK(filter_func_(
+                a.TopLeftPixel(), a.stride(), b.TopLeftPixel(), width, height,
+                filter_strength, filter_weight, accum_chk.TopLeftPixel(),
+                count_chk.TopLeftPixel()));
+            EXPECT_TRUE(accum_chk.CheckValues(accum_ref));
+            EXPECT_TRUE(count_chk.CheckValues(count_ref));
+            if (HasFailure()) {
+              printf("Weight: %d Strength: %d\n", filter_weight,
+                     filter_strength);
+              count_chk.PrintDifference(count_ref);
+              accum_chk.PrintDifference(accum_ref);
+              return;
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+TEST_P(TemporalFilterTest, DISABLED_Speed) {
+  Buffer<uint8_t> a = Buffer<uint8_t>(16, 16, 8);
+  ASSERT_TRUE(a.Init());
+
+  const int filter_weight = 2;
+  const int filter_strength = 6;
+
+  for (int width = 8; width <= 16; width += 8) {
+    for (int height = 8; height <= 16; height += 8) {
+      // The second buffer must not have any border.
+      Buffer<uint8_t> b = Buffer<uint8_t>(width, height, 0);
+      ASSERT_TRUE(b.Init());
+      Buffer<unsigned int> accum_ref = Buffer<unsigned int>(width, height, 0);
+      ASSERT_TRUE(accum_ref.Init());
+      Buffer<unsigned int> accum_chk = Buffer<unsigned int>(width, height, 0);
+      ASSERT_TRUE(accum_chk.Init());
+      Buffer<uint16_t> count_ref = Buffer<uint16_t>(width, height, 0);
+      ASSERT_TRUE(count_ref.Init());
+      Buffer<uint16_t> count_chk = Buffer<uint16_t>(width, height, 0);
+      ASSERT_TRUE(count_chk.Init());
+
+      a.Set(&rnd_, 0, 7);
+      b.Set(&rnd_, 0, 7);
+
+      accum_chk.Set(0);
+      count_chk.Set(0);
+
+      vpx_usec_timer timer;
+      vpx_usec_timer_start(&timer);
+      for (int i = 0; i < 10000; ++i) {
+        filter_func_(a.TopLeftPixel(), a.stride(), b.TopLeftPixel(), width,
+                     height, filter_strength, filter_weight,
+                     accum_chk.TopLeftPixel(), count_chk.TopLeftPixel());
+      }
+      vpx_usec_timer_mark(&timer);
+      const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+      printf("Temporal filter %dx%d time: %5d us\n", width, height,
+             elapsed_time);
+    }
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(C, TemporalFilterTest,
+                        ::testing::Values(&vp9_temporal_filter_apply_c));
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, TemporalFilterTest,
+                        ::testing::Values(&vp9_temporal_filter_apply_sse4_1));
+#endif  // HAVE_SSE4_1
+}  // namespace
--- a/test/test-data.mk
+++ b/test/test-data.mk
@@ -23,6 +23,7 @@ LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += niklas_1280_720_30.y4m
 LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += noisy_clip_640_360.y4m
 LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += rush_hour_444.y4m
 LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += screendata.y4m
+LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += niklas_640_480_30.yuv

 # Test vectors
 LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-00-comprehensive-001.ivf
@@ -731,6 +732,8 @@ LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp93-2-20-12bit-yuv444.webm.md5
 endif  # CONFIG_VP9_HIGHBITDEPTH

 # Invalid files for testing libvpx error checking.
+LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += invalid-bug-1443.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += invalid-bug-1443.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += invalid-vp80-00-comprehensive-018.ivf.2kf_0x6.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += invalid-vp80-00-comprehensive-018.ivf.2kf_0x6.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-01-v3.webm
@@ -771,6 +774,8 @@ LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s367
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp91-2-mixedrefcsp-444to420.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp91-2-mixedrefcsp-444to420.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-07-frame_parallel-1.webm
@@ -814,7 +819,6 @@ LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += kirland_640_480_30.yuv
 LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += macmarcomoving_640_480_30.yuv
 LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += macmarcostationary_640_480_30.yuv
 LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += niklas_1280_720_30.yuv
-LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += niklas_640_480_30.yuv
 LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += tacomanarrows_640_480_30.yuv
 LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += tacomasmallcameramovement_640_480_30.yuv
 LIBVPX_TEST_DATA-$(CONFIG_VP9_ENCODER) += thaloundeskmtg_640_480_30.yuv
@@ -874,3 +878,5 @@ LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-21-resize_inter_1920x1080_7_3-4
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-21-resize_inter_1920x1080_7_3-4.webm.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-22-svc_1280x720_3.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-22-svc_1280x720_3.ivf.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-22-svc_1280x720_1.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-22-svc_1280x720_1.webm.md5
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -6,6 +6,8 @@ b87815bf86020c592ccc7a846ba2e28ec8043902 *hantro_odd.yuv
 456d1493e52d32a5c30edf44a27debc1fa6b253a *invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-.ivf.res
 c123d1f9f02fb4143abb5e271916e3a3080de8f6 *invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-z.ivf
 456d1493e52d32a5c30edf44a27debc1fa6b253a *invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-z.ivf.res
+efafb92b7567bc04c3f1432ea6c268c1c31affd5 *invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-.ivf
+5d9474c0309b7ca09a182d888f73b37a8fe1362c *invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-.ivf.res
 fe346136b9b8c1e6f6084cc106485706915795e4 *invalid-vp90-01-v3.webm
 5d9474c0309b7ca09a182d888f73b37a8fe1362c *invalid-vp90-01-v3.webm.res
 d78e2fceba5ac942246503ec8366f879c4775ca5 *invalid-vp90-02-v2.webm
@@ -848,3 +850,7 @@ a000d568431d07379dd5a8ec066061c07e560b47 *invalid-vp90-2-00-quantizer-63.ivf.kf_
 6fa3d3ac306a3d9ce1d610b78441dc00d2c2d4b9 *tos_vp8.webm
 e402cbbf9e550ae017a1e9f1f73931c1d18474e8 *invalid-crbug-667044.webm
 d3964f9dad9f60363c81b688324d95b4ec7c8038 *invalid-crbug-667044.webm.res
+fd9df7f3f6992af1d7a9dde975c9a0d6f28c053d *invalid-bug-1443.ivf
+fd3020fa6e9ca5966206738654c97dec313b0a95 *invalid-bug-1443.ivf.res
+17696cd21e875f1d6e5d418cbf89feab02c8850a *vp90-2-22-svc_1280x720_1.webm
+e2f9e1e47a791b4e939a9bdc50bf7a25b3761f77 *vp90-2-22-svc_1280x720_1.webm.md5
--- a/test/test.mk
+++ b/test/test.mk
@@ -1,4 +1,5 @@
 LIBVPX_TEST_SRCS-yes += acm_random.h
+LIBVPX_TEST_SRCS-yes += buffer.h
 LIBVPX_TEST_SRCS-yes += clear_system_state.h
 LIBVPX_TEST_SRCS-yes += codec_factory.h
 LIBVPX_TEST_SRCS-yes += md5_helper.h
@@ -38,7 +39,6 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += byte_alignment_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += decode_svc_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += external_frame_buffer_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += user_priv_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += vp9_frame_parallel_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += active_map_refresh_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += active_map_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += borders_test.cc
@@ -47,6 +47,7 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += frame_size_tests.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_lossless_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_end_to_end_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_ethread_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_motion_vector_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += level_test.cc

 LIBVPX_TEST_SRCS-yes                   += decode_test_driver.cc
@@ -122,6 +123,7 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += vp8_fdct4x4_test.cc
 LIBVPX_TEST_SRCS-yes                   += idct_test.cc
 LIBVPX_TEST_SRCS-yes                   += predict_test.cc
 LIBVPX_TEST_SRCS-yes                   += vpx_scale_test.cc
+LIBVPX_TEST_SRCS-yes                   += vpx_scale_test.h

 ifeq ($(CONFIG_VP8_ENCODER)$(CONFIG_TEMPORAL_DENOISING),yesyes)
 LIBVPX_TEST_SRCS-$(HAVE_SSE2) += vp8_denoiser_sse2_test.cc
@@ -149,14 +151,20 @@ LIBVPX_TEST_SRCS-yes                   += vp9_intrapred_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += vp9_decrypt_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += vp9_thread_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += avg_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += comp_avg_pred_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct16x16_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct32x32_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct4x4_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct_partial_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct8x8_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += hadamard_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += minmax_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_scale_test.cc
+ifneq ($(CONFIG_REALTIME_ONLY),yes)
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += temporal_filter_test.cc
+endif
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_error_block_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_block_error_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_quantize_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_subtract_test.cc

@@ -167,7 +175,7 @@ LIBVPX_TEST_SRCS-$(CONFIG_INTERNAL_STATS) += consistency_test.cc
 endif

 ifeq ($(CONFIG_VP9_ENCODER)$(CONFIG_VP9_TEMPORAL_DENOISING),yesyes)
-LIBVPX_TEST_SRCS-$(HAVE_SSE2) += vp9_denoiser_sse2_test.cc
+LIBVPX_TEST_SRCS-yes += vp9_denoiser_test.cc
 endif
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_arf_freq_test.cc

--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -312,6 +312,31 @@ INTRA_PRED_TEST(MSA, TestIntraPred32, vpx_dc_predictor_32x32_msa,
                vpx_tm_predictor_32x32_msa)
 #endif  // HAVE_MSA

+#if HAVE_VSX
+INTRA_PRED_TEST(VSX, TestIntraPred4, NULL, NULL, NULL, NULL, NULL,
+                vpx_h_predictor_4x4_vsx, NULL, NULL, NULL, NULL, NULL, NULL,
+                vpx_tm_predictor_4x4_vsx)
+
+INTRA_PRED_TEST(VSX, TestIntraPred8, vpx_dc_predictor_8x8_vsx, NULL, NULL, NULL,
+                NULL, vpx_h_predictor_8x8_vsx, vpx_d45_predictor_8x8_vsx, NULL,
+                NULL, NULL, NULL, vpx_d63_predictor_8x8_vsx,
+                vpx_tm_predictor_8x8_vsx)
+
+INTRA_PRED_TEST(VSX, TestIntraPred16, vpx_dc_predictor_16x16_vsx,
+                vpx_dc_left_predictor_16x16_vsx, vpx_dc_top_predictor_16x16_vsx,
+                vpx_dc_128_predictor_16x16_vsx, vpx_v_predictor_16x16_vsx,
+                vpx_h_predictor_16x16_vsx, vpx_d45_predictor_16x16_vsx, NULL,
+                NULL, NULL, NULL, vpx_d63_predictor_16x16_vsx,
+                vpx_tm_predictor_16x16_vsx)
+
+INTRA_PRED_TEST(VSX, TestIntraPred32, vpx_dc_predictor_32x32_vsx,
+                vpx_dc_left_predictor_32x32_vsx, vpx_dc_top_predictor_32x32_vsx,
+                vpx_dc_128_predictor_32x32_vsx, vpx_v_predictor_32x32_vsx,
+                vpx_h_predictor_32x32_vsx, vpx_d45_predictor_32x32_vsx, NULL,
+                NULL, NULL, NULL, vpx_d63_predictor_32x32_vsx,
+                vpx_tm_predictor_32x32_vsx)
+#endif  // HAVE_VSX
+
 // -----------------------------------------------------------------------------

 #if CONFIG_VP9_HIGHBITDEPTH
@@ -455,29 +480,70 @@ HIGHBD_INTRA_PRED_TEST(
    vpx_highbd_d63_predictor_32x32_c, vpx_highbd_tm_predictor_32x32_c)

 #if HAVE_SSE2
-HIGHBD_INTRA_PRED_TEST(SSE2, TestHighbdIntraPred4,
-                       vpx_highbd_dc_predictor_4x4_sse2, NULL, NULL, NULL,
-                       vpx_highbd_v_predictor_4x4_sse2, NULL, NULL, NULL, NULL,
-                       NULL, NULL, NULL, vpx_highbd_tm_predictor_4x4_c)
+HIGHBD_INTRA_PRED_TEST(
+    SSE2, TestHighbdIntraPred4, vpx_highbd_dc_predictor_4x4_sse2,
+    vpx_highbd_dc_left_predictor_4x4_sse2, vpx_highbd_dc_top_predictor_4x4_sse2,
+    vpx_highbd_dc_128_predictor_4x4_sse2, vpx_highbd_v_predictor_4x4_sse2,
+    vpx_highbd_h_predictor_4x4_sse2, NULL, vpx_highbd_d135_predictor_4x4_sse2,
+    vpx_highbd_d117_predictor_4x4_sse2, vpx_highbd_d153_predictor_4x4_sse2,
+    vpx_highbd_d207_predictor_4x4_sse2, vpx_highbd_d63_predictor_4x4_sse2,
+    vpx_highbd_tm_predictor_4x4_c)

 HIGHBD_INTRA_PRED_TEST(SSE2, TestHighbdIntraPred8,
-                       vpx_highbd_dc_predictor_8x8_sse2, NULL, NULL, NULL,
-                       vpx_highbd_v_predictor_8x8_sse2, NULL, NULL, NULL, NULL,
-                       NULL, NULL, NULL, vpx_highbd_tm_predictor_8x8_sse2)
+                       vpx_highbd_dc_predictor_8x8_sse2,
+                       vpx_highbd_dc_left_predictor_8x8_sse2,
+                       vpx_highbd_dc_top_predictor_8x8_sse2,
+                       vpx_highbd_dc_128_predictor_8x8_sse2,
+                       vpx_highbd_v_predictor_8x8_sse2,
+                       vpx_highbd_h_predictor_8x8_sse2, NULL, NULL, NULL, NULL,
+                       NULL, NULL, vpx_highbd_tm_predictor_8x8_sse2)

 HIGHBD_INTRA_PRED_TEST(SSE2, TestHighbdIntraPred16,
-                       vpx_highbd_dc_predictor_16x16_sse2, NULL, NULL, NULL,
-                       vpx_highbd_v_predictor_16x16_sse2, NULL, NULL, NULL,
-                       NULL, NULL, NULL, NULL,
-                       vpx_highbd_tm_predictor_16x16_sse2)
+                       vpx_highbd_dc_predictor_16x16_sse2,
+                       vpx_highbd_dc_left_predictor_16x16_sse2,
+                       vpx_highbd_dc_top_predictor_16x16_sse2,
+                       vpx_highbd_dc_128_predictor_16x16_sse2,
+                       vpx_highbd_v_predictor_16x16_sse2,
+                       vpx_highbd_h_predictor_16x16_sse2, NULL, NULL, NULL,
+                       NULL, NULL, NULL, vpx_highbd_tm_predictor_16x16_sse2)

 HIGHBD_INTRA_PRED_TEST(SSE2, TestHighbdIntraPred32,
-                       vpx_highbd_dc_predictor_32x32_sse2, NULL, NULL, NULL,
-                       vpx_highbd_v_predictor_32x32_sse2, NULL, NULL, NULL,
-                       NULL, NULL, NULL, NULL,
-                       vpx_highbd_tm_predictor_32x32_sse2)
+                       vpx_highbd_dc_predictor_32x32_sse2,
+                       vpx_highbd_dc_left_predictor_32x32_sse2,
+                       vpx_highbd_dc_top_predictor_32x32_sse2,
+                       vpx_highbd_dc_128_predictor_32x32_sse2,
+                       vpx_highbd_v_predictor_32x32_sse2,
+                       vpx_highbd_h_predictor_32x32_sse2, NULL, NULL, NULL,
+                       NULL, NULL, NULL, vpx_highbd_tm_predictor_32x32_sse2)
 #endif  // HAVE_SSE2

+#if HAVE_SSSE3
+HIGHBD_INTRA_PRED_TEST(SSSE3, TestHighbdIntraPred4, NULL, NULL, NULL, NULL,
+                       NULL, NULL, vpx_highbd_d45_predictor_4x4_ssse3, NULL,
+                       NULL, NULL, NULL, NULL, NULL)
+HIGHBD_INTRA_PRED_TEST(SSSE3, TestHighbdIntraPred8, NULL, NULL, NULL, NULL,
+                       NULL, NULL, vpx_highbd_d45_predictor_8x8_ssse3,
+                       vpx_highbd_d135_predictor_8x8_ssse3,
+                       vpx_highbd_d117_predictor_8x8_ssse3,
+                       vpx_highbd_d153_predictor_8x8_ssse3,
+                       vpx_highbd_d207_predictor_8x8_ssse3,
+                       vpx_highbd_d63_predictor_8x8_ssse3, NULL)
+HIGHBD_INTRA_PRED_TEST(SSSE3, TestHighbdIntraPred16, NULL, NULL, NULL, NULL,
+                       NULL, NULL, vpx_highbd_d45_predictor_16x16_ssse3,
+                       vpx_highbd_d135_predictor_16x16_ssse3,
+                       vpx_highbd_d117_predictor_16x16_ssse3,
+                       vpx_highbd_d153_predictor_16x16_ssse3,
+                       vpx_highbd_d207_predictor_16x16_ssse3,
+                       vpx_highbd_d63_predictor_16x16_ssse3, NULL)
+HIGHBD_INTRA_PRED_TEST(SSSE3, TestHighbdIntraPred32, NULL, NULL, NULL, NULL,
+                       NULL, NULL, vpx_highbd_d45_predictor_32x32_ssse3,
+                       vpx_highbd_d135_predictor_32x32_ssse3,
+                       vpx_highbd_d117_predictor_32x32_ssse3,
+                       vpx_highbd_d153_predictor_32x32_ssse3,
+                       vpx_highbd_d207_predictor_32x32_ssse3,
+                       vpx_highbd_d63_predictor_32x32_ssse3, NULL)
+#endif  // HAVE_SSSE3
+
 #if HAVE_NEON
 HIGHBD_INTRA_PRED_TEST(
    NEON, TestHighbdIntraPred4, vpx_highbd_dc_predictor_4x4_neon,
--- a/test/test_libvpx.cc
+++ b/test/test_libvpx.cc
@@ -53,6 +53,9 @@ int main(int argc, char **argv) {
  }
  if (!(simd_caps & HAS_AVX)) append_negative_gtest_filter(":AVX.*:AVX/*");
  if (!(simd_caps & HAS_AVX2)) append_negative_gtest_filter(":AVX2.*:AVX2/*");
+  if (!(simd_caps & HAS_AVX512)) {
+    append_negative_gtest_filter(":AVX512.*:AVX512/*");
+  }
 #endif  // ARCH_X86 || ARCH_X86_64

 #if !CONFIG_SHARED
--- a/test/test_vector_test.cc
+++ b/test/test_vector_test.cc
@@ -28,13 +28,10 @@

 namespace {

-enum DecodeMode { kSerialMode, kFrameParallelMode };
+const int kThreads = 0;
+const int kFileName = 1;

-const int kDecodeMode = 0;
-const int kThreads = 1;
-const int kFileName = 2;
-
-typedef std::tr1::tuple<int, int, const char *> DecodeParam;
+typedef std::tr1::tuple<int, const char *> DecodeParam;

 class TestVectorTest : public ::libvpx_test::DecoderTest,
                       public ::libvpx_test::CodecTestWithParam<DecodeParam> {
@@ -53,8 +50,8 @@ class TestVectorTest : public ::libvpx_test::DecoderTest,

  void OpenMD5File(const std::string &md5_file_name_) {
    md5_file_ = libvpx_test::OpenTestDataFile(md5_file_name_);
-    ASSERT_TRUE(md5_file_ != NULL) << "Md5 file open failed. Filename: "
-                                   << md5_file_name_;
+    ASSERT_TRUE(md5_file_ != NULL)
+        << "Md5 file open failed. Filename: " << md5_file_name_;
  }

  virtual void DecompressedFrameHook(const vpx_image_t &img,
@@ -92,29 +89,14 @@ class TestVectorTest : public ::libvpx_test::DecoderTest,
 TEST_P(TestVectorTest, MD5Match) {
  const DecodeParam input = GET_PARAM(1);
  const std::string filename = std::tr1::get<kFileName>(input);
-  const int threads = std::tr1::get<kThreads>(input);
-  const int mode = std::tr1::get<kDecodeMode>(input);
  vpx_codec_flags_t flags = 0;
  vpx_codec_dec_cfg_t cfg = vpx_codec_dec_cfg_t();
  char str[256];

-  if (mode == kFrameParallelMode) {
-    flags |= VPX_CODEC_USE_FRAME_THREADING;
-#if CONFIG_VP9_DECODER
-    // TODO(hkuang): Fix frame parallel decode bug. See issue 1086.
-    if (resize_clips_.find(filename) != resize_clips_.end()) {
-      printf("Skipping the test file: %s, due to frame parallel decode bug.\n",
-             filename.c_str());
-      return;
-    }
-#endif
-  }
+  cfg.threads = std::tr1::get<kThreads>(input);

-  cfg.threads = threads;
-
-  snprintf(str, sizeof(str) / sizeof(str[0]) - 1,
-           "file: %s  mode: %s threads: %d", filename.c_str(),
-           mode == 0 ? "Serial" : "Parallel", threads);
+  snprintf(str, sizeof(str) / sizeof(str[0]) - 1, "file: %s threads: %d",
+           filename.c_str(), cfg.threads);
  SCOPED_TRACE(str);

  // Open compressed video file.
@@ -145,13 +127,10 @@ TEST_P(TestVectorTest, MD5Match) {
  ASSERT_NO_FATAL_FAILURE(RunLoop(video.get(), cfg));
 }

-// Test VP8 decode in serial mode with single thread.
-// NOTE: VP8 only support serial mode.
 #if CONFIG_VP8_DECODER
 VP8_INSTANTIATE_TEST_CASE(
    TestVectorTest,
    ::testing::Combine(
-        ::testing::Values(0),  // Serial Mode.
        ::testing::Values(1),  // Single thread.
        ::testing::ValuesIn(libvpx_test::kVP8TestVectors,
                            libvpx_test::kVP8TestVectors +
@@ -164,33 +143,28 @@ INSTANTIATE_TEST_CASE_P(
        ::testing::Values(
            static_cast<const libvpx_test::CodecFactory *>(&libvpx_test::kVP8)),
        ::testing::Combine(
-            ::testing::Values(0),    // Serial Mode.
-            ::testing::Range(1, 8),  // With 1 ~ 8 threads.
+            ::testing::Range(2, 9),  // With 2 ~ 8 threads.
            ::testing::ValuesIn(libvpx_test::kVP8TestVectors,
                                libvpx_test::kVP8TestVectors +
                                    libvpx_test::kNumVP8TestVectors))));

 #endif  // CONFIG_VP8_DECODER

-// Test VP9 decode in serial mode with single thread.
 #if CONFIG_VP9_DECODER
 VP9_INSTANTIATE_TEST_CASE(
    TestVectorTest,
    ::testing::Combine(
-        ::testing::Values(0),  // Serial Mode.
        ::testing::Values(1),  // Single thread.
        ::testing::ValuesIn(libvpx_test::kVP9TestVectors,
                            libvpx_test::kVP9TestVectors +
                                libvpx_test::kNumVP9TestVectors)));

-// Test VP9 decode in frame parallel mode with different number of threads.
 INSTANTIATE_TEST_CASE_P(
-    VP9MultiThreadedFrameParallel, TestVectorTest,
+    VP9MultiThreaded, TestVectorTest,
    ::testing::Combine(
        ::testing::Values(
            static_cast<const libvpx_test::CodecFactory *>(&libvpx_test::kVP9)),
        ::testing::Combine(
-            ::testing::Values(1),    // Frame Parallel mode.
            ::testing::Range(2, 9),  // With 2 ~ 8 threads.
            ::testing::ValuesIn(libvpx_test::kVP9TestVectors,
                                libvpx_test::kVP9TestVectors +
--- a/test/test_vectors.cc
+++ b/test/test_vectors.cc
@@ -371,6 +371,7 @@ const char *const kVP9TestVectors[] = {
 #endif  // CONFIG_VP9_HIGHBITDEPTH
  "vp90-2-20-big_superframe-01.webm",
  "vp90-2-20-big_superframe-02.webm",
+  "vp90-2-22-svc_1280x720_1.webm",
  RESIZE_TEST_VECTORS
 };
 const char *const kVP9TestVectorsSvc[] = { "vp90-2-22-svc_1280x720_3.ivf" };
--- a/test/twopass_encoder.sh
+++ b/test/twopass_encoder.sh
@@ -54,7 +54,10 @@ twopass_encoder_vp9() {
  fi
 }

-twopass_encoder_tests="twopass_encoder_vp8
-                       twopass_encoder_vp9"

-run_tests twopass_encoder_verify_environment "${twopass_encoder_tests}"
+if [ "$(vpx_config_option_enabled CONFIG_REALTIME_ONLY)" != "yes" ]; then
+  twopass_encoder_tests="twopass_encoder_vp8
+                         twopass_encoder_vp9"
+
+  run_tests twopass_encoder_verify_environment "${twopass_encoder_tests}"
+fi
--- a/test/variance_test.cc
+++ b/test/variance_test.cc
--- a/test/vp8_fdct4x4_test.cc
+++ b/test/vp8_fdct4x4_test.cc
@@ -17,12 +17,16 @@

 #include "third_party/googletest/src/include/gtest/gtest.h"

+#include "./vpx_config.h"
 #include "./vp8_rtcd.h"
 #include "test/acm_random.h"
 #include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"

 namespace {

+typedef void (*FdctFunc)(int16_t *a, int16_t *b, int a_stride);
+
 const int cospi8sqrt2minus1 = 20091;
 const int sinpi8sqrt2 = 35468;

@@ -68,10 +72,21 @@ void reference_idct4x4(const int16_t *input, int16_t *output) {

 using libvpx_test::ACMRandom;

-TEST(VP8FdctTest, SignBiasCheck) {
-  ACMRandom rnd(ACMRandom::DeterministicSeed());
+class FdctTest : public ::testing::TestWithParam<FdctFunc> {
+ public:
+  virtual void SetUp() {
+    fdct_func_ = GetParam();
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+  }
+
+ protected:
+  FdctFunc fdct_func_;
+  ACMRandom rnd_;
+};
+
+TEST_P(FdctTest, SignBiasCheck) {
  int16_t test_input_block[16];
-  int16_t test_output_block[16];
+  DECLARE_ALIGNED(16, int16_t, test_output_block[16]);
  const int pitch = 8;
  int count_sign_block[16][2];
  const int count_test_block = 1000000;
@@ -81,10 +96,10 @@ TEST(VP8FdctTest, SignBiasCheck) {
  for (int i = 0; i < count_test_block; ++i) {
    // Initialize a test block with input range [-255, 255].
    for (int j = 0; j < 16; ++j) {
-      test_input_block[j] = rnd.Rand8() - rnd.Rand8();
+      test_input_block[j] = rnd_.Rand8() - rnd_.Rand8();
    }

-    vp8_short_fdct4x4_c(test_input_block, test_output_block, pitch);
+    fdct_func_(test_input_block, test_output_block, pitch);

    for (int j = 0; j < 16; ++j) {
      if (test_output_block[j] < 0) {
@@ -110,10 +125,10 @@ TEST(VP8FdctTest, SignBiasCheck) {
  for (int i = 0; i < count_test_block; ++i) {
    // Initialize a test block with input range [-15, 15].
    for (int j = 0; j < 16; ++j) {
-      test_input_block[j] = (rnd.Rand8() >> 4) - (rnd.Rand8() >> 4);
+      test_input_block[j] = (rnd_.Rand8() >> 4) - (rnd_.Rand8() >> 4);
    }

-    vp8_short_fdct4x4_c(test_input_block, test_output_block, pitch);
+    fdct_func_(test_input_block, test_output_block, pitch);

    for (int j = 0; j < 16; ++j) {
      if (test_output_block[j] < 0) {
@@ -135,23 +150,22 @@ TEST(VP8FdctTest, SignBiasCheck) {
      << "Error: 4x4 FDCT has a sign bias > 10% for input range [-15, 15]";
 };

-TEST(VP8FdctTest, RoundTripErrorCheck) {
-  ACMRandom rnd(ACMRandom::DeterministicSeed());
+TEST_P(FdctTest, RoundTripErrorCheck) {
  int max_error = 0;
  double total_error = 0;
  const int count_test_block = 1000000;
  for (int i = 0; i < count_test_block; ++i) {
    int16_t test_input_block[16];
-    int16_t test_temp_block[16];
    int16_t test_output_block[16];
+    DECLARE_ALIGNED(16, int16_t, test_temp_block[16]);

    // Initialize a test block with input range [-255, 255].
    for (int j = 0; j < 16; ++j) {
-      test_input_block[j] = rnd.Rand8() - rnd.Rand8();
+      test_input_block[j] = rnd_.Rand8() - rnd_.Rand8();
    }

    const int pitch = 8;
-    vp8_short_fdct4x4_c(test_input_block, test_temp_block, pitch);
+    fdct_func_(test_input_block, test_temp_block, pitch);
    reference_idct4x4(test_temp_block, test_output_block);

    for (int j = 0; j < 16; ++j) {
@@ -169,4 +183,24 @@ TEST(VP8FdctTest, RoundTripErrorCheck) {
      << "Error: FDCT/IDCT has average roundtrip error > 1 per block";
 };

+INSTANTIATE_TEST_CASE_P(C, FdctTest, ::testing::Values(vp8_short_fdct4x4_c));
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, FdctTest,
+                        ::testing::Values(vp8_short_fdct4x4_neon));
+#endif  // HAVE_NEON
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, FdctTest,
+                        ::testing::Values(vp8_short_fdct4x4_sse2));
+#endif  // HAVE_SSE2
+
+#if HAVE_MSA
+INSTANTIATE_TEST_CASE_P(MSA, FdctTest,
+                        ::testing::Values(vp8_short_fdct4x4_msa));
+#endif  // HAVE_MSA
+#if HAVE_MMI
+INSTANTIATE_TEST_CASE_P(MMI, FdctTest,
+                        ::testing::Values(vp8_short_fdct4x4_mmi));
+#endif  // HAVE_MMI
 }  // namespace
--- a/test/vp9_block_error_test.cc
+++ b/test/vp9_block_error_test.cc
@@ -23,36 +23,36 @@
 #include "vp9/common/vp9_entropy.h"
 #include "vpx/vpx_codec.h"
 #include "vpx/vpx_integer.h"
+#include "vpx_dsp/vpx_dsp_common.h"

 using libvpx_test::ACMRandom;

 namespace {
-#if CONFIG_VP9_HIGHBITDEPTH
 const int kNumIterations = 1000;

-typedef int64_t (*ErrorBlockFunc)(const tran_low_t *coeff,
+typedef int64_t (*HBDBlockErrorFunc)(const tran_low_t *coeff,
+                                     const tran_low_t *dqcoeff,
+                                     intptr_t block_size, int64_t *ssz,
+                                     int bps);
+
+typedef std::tr1::tuple<HBDBlockErrorFunc, HBDBlockErrorFunc, vpx_bit_depth_t>
+    BlockErrorParam;
+
+typedef int64_t (*BlockErrorFunc)(const tran_low_t *coeff,
                                  const tran_low_t *dqcoeff,
-                                  intptr_t block_size, int64_t *ssz, int bps);
+                                  intptr_t block_size, int64_t *ssz);

-typedef std::tr1::tuple<ErrorBlockFunc, ErrorBlockFunc, vpx_bit_depth_t>
-    ErrorBlockParam;
-
-// wrapper for 8-bit block error functions without a 'bps' param.
-typedef int64_t (*HighBdBlockError8bit)(const tran_low_t *coeff,
-                                        const tran_low_t *dqcoeff,
-                                        intptr_t block_size, int64_t *ssz);
-template <HighBdBlockError8bit fn>
-int64_t HighBdBlockError8bitWrapper(const tran_low_t *coeff,
-                                    const tran_low_t *dqcoeff,
-                                    intptr_t block_size, int64_t *ssz,
-                                    int bps) {
-  EXPECT_EQ(8, bps);
+template <BlockErrorFunc fn>
+int64_t BlockError8BitWrapper(const tran_low_t *coeff,
+                              const tran_low_t *dqcoeff, intptr_t block_size,
+                              int64_t *ssz, int bps) {
+  EXPECT_EQ(bps, 8);
  return fn(coeff, dqcoeff, block_size, ssz);
 }

-class ErrorBlockTest : public ::testing::TestWithParam<ErrorBlockParam> {
+class BlockErrorTest : public ::testing::TestWithParam<BlockErrorParam> {
 public:
-  virtual ~ErrorBlockTest() {}
+  virtual ~BlockErrorTest() {}
  virtual void SetUp() {
    error_block_op_ = GET_PARAM(0);
    ref_error_block_op_ = GET_PARAM(1);
@@ -63,11 +63,11 @@ class ErrorBlockTest : public ::testing::TestWithParam<ErrorBlockParam> {

 protected:
  vpx_bit_depth_t bit_depth_;
-  ErrorBlockFunc error_block_op_;
-  ErrorBlockFunc ref_error_block_op_;
+  HBDBlockErrorFunc error_block_op_;
+  HBDBlockErrorFunc ref_error_block_op_;
 };

-TEST_P(ErrorBlockTest, OperationCheck) {
+TEST_P(BlockErrorTest, OperationCheck) {
  ACMRandom rnd(ACMRandom::DeterministicSeed());
  DECLARE_ALIGNED(16, tran_low_t, coeff[4096]);
  DECLARE_ALIGNED(16, tran_low_t, dqcoeff[4096]);
@@ -110,7 +110,7 @@ TEST_P(ErrorBlockTest, OperationCheck) {
      << "First failed at test case " << first_failure;
 }

-TEST_P(ErrorBlockTest, ExtremeValues) {
+TEST_P(BlockErrorTest, ExtremeValues) {
  ACMRandom rnd(ACMRandom::DeterministicSeed());
  DECLARE_ALIGNED(16, tran_low_t, coeff[4096]);
  DECLARE_ALIGNED(16, tran_low_t, dqcoeff[4096]);
@@ -171,29 +171,28 @@ TEST_P(ErrorBlockTest, ExtremeValues) {
 using std::tr1::make_tuple;

 #if HAVE_SSE2
-INSTANTIATE_TEST_CASE_P(
-    SSE2, ErrorBlockTest,
-    ::testing::Values(
-        make_tuple(&vp9_highbd_block_error_sse2, &vp9_highbd_block_error_c,
-                   VPX_BITS_10),
-        make_tuple(&vp9_highbd_block_error_sse2, &vp9_highbd_block_error_c,
-                   VPX_BITS_12),
-        make_tuple(&vp9_highbd_block_error_sse2, &vp9_highbd_block_error_c,
-                   VPX_BITS_8),
-        make_tuple(
-            &HighBdBlockError8bitWrapper<vp9_highbd_block_error_8bit_sse2>,
-            &HighBdBlockError8bitWrapper<vp9_highbd_block_error_8bit_c>,
-            VPX_BITS_8)));
+const BlockErrorParam sse2_block_error_tests[] = {
+#if CONFIG_VP9_HIGHBITDEPTH
+  make_tuple(&vp9_highbd_block_error_sse2, &vp9_highbd_block_error_c,
+             VPX_BITS_10),
+  make_tuple(&vp9_highbd_block_error_sse2, &vp9_highbd_block_error_c,
+             VPX_BITS_12),
+  make_tuple(&vp9_highbd_block_error_sse2, &vp9_highbd_block_error_c,
+             VPX_BITS_8),
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+  make_tuple(&BlockError8BitWrapper<vp9_block_error_sse2>,
+             &BlockError8BitWrapper<vp9_block_error_c>, VPX_BITS_8)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, BlockErrorTest,
+                        ::testing::ValuesIn(sse2_block_error_tests));
 #endif  // HAVE_SSE2

-#if HAVE_AVX
+#if HAVE_AVX2
 INSTANTIATE_TEST_CASE_P(
-    AVX, ErrorBlockTest,
-    ::testing::Values(make_tuple(
-        &HighBdBlockError8bitWrapper<vp9_highbd_block_error_8bit_avx>,
-        &HighBdBlockError8bitWrapper<vp9_highbd_block_error_8bit_c>,
-        VPX_BITS_8)));
-#endif  // HAVE_AVX
-
-#endif  // CONFIG_VP9_HIGHBITDEPTH
+    AVX2, BlockErrorTest,
+    ::testing::Values(make_tuple(&BlockError8BitWrapper<vp9_block_error_avx2>,
+                                 &BlockError8BitWrapper<vp9_block_error_c>,
+                                 VPX_BITS_8)));
+#endif  // HAVE_AVX2
 }  // namespace
--- a/test/vp9_denoiser_sse2_test.cc
+++ b/test/vp9_denoiser_sse2_test.cc
@@ -29,11 +29,21 @@ using libvpx_test::ACMRandom;
 namespace {

 const int kNumPixels = 64 * 64;
-class VP9DenoiserTest : public ::testing::TestWithParam<BLOCK_SIZE> {
+
+typedef int (*Vp9DenoiserFilterFunc)(const uint8_t *sig, int sig_stride,
+                                     const uint8_t *mc_avg, int mc_avg_stride,
+                                     uint8_t *avg, int avg_stride,
+                                     int increase_denoising, BLOCK_SIZE bs,
+                                     int motion_magnitude);
+typedef std::tr1::tuple<Vp9DenoiserFilterFunc, BLOCK_SIZE> VP9DenoiserTestParam;
+
+class VP9DenoiserTest
+    : public ::testing::Test,
+      public ::testing::WithParamInterface<VP9DenoiserTestParam> {
 public:
  virtual ~VP9DenoiserTest() {}

-  virtual void SetUp() { bs_ = GetParam(); }
+  virtual void SetUp() { bs_ = GET_PARAM(1); }

  virtual void TearDown() { libvpx_test::ClearSystemState(); }

@@ -76,9 +86,9 @@ TEST_P(VP9DenoiserTest, BitexactCheck) {
                                                   64, avg_block_c, 64, 0, bs_,
                                                   motion_magnitude_random));

-    ASM_REGISTER_STATE_CHECK(vp9_denoiser_filter_sse2(
-        sig_block, 64, mc_avg_block, 64, avg_block_sse2, 64, 0, bs_,
-        motion_magnitude_random));
+    ASM_REGISTER_STATE_CHECK(GET_PARAM(0)(sig_block, 64, mc_avg_block, 64,
+                                          avg_block_sse2, 64, 0, bs_,
+                                          motion_magnitude_random));

    // Test bitexactness.
    for (int h = 0; h < (4 << b_height_log2_lookup[bs_]); ++h) {
@@ -89,10 +99,36 @@ TEST_P(VP9DenoiserTest, BitexactCheck) {
  }
 }

+using std::tr1::make_tuple;
+
 // Test for all block size.
-INSTANTIATE_TEST_CASE_P(SSE2, VP9DenoiserTest,
-                        ::testing::Values(BLOCK_8X8, BLOCK_8X16, BLOCK_16X8,
-                                          BLOCK_16X16, BLOCK_16X32, BLOCK_32X16,
-                                          BLOCK_32X32, BLOCK_32X64, BLOCK_64X32,
-                                          BLOCK_64X64));
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, VP9DenoiserTest,
+    ::testing::Values(make_tuple(&vp9_denoiser_filter_sse2, BLOCK_8X8),
+                      make_tuple(&vp9_denoiser_filter_sse2, BLOCK_8X16),
+                      make_tuple(&vp9_denoiser_filter_sse2, BLOCK_16X8),
+                      make_tuple(&vp9_denoiser_filter_sse2, BLOCK_16X16),
+                      make_tuple(&vp9_denoiser_filter_sse2, BLOCK_16X32),
+                      make_tuple(&vp9_denoiser_filter_sse2, BLOCK_32X16),
+                      make_tuple(&vp9_denoiser_filter_sse2, BLOCK_32X32),
+                      make_tuple(&vp9_denoiser_filter_sse2, BLOCK_32X64),
+                      make_tuple(&vp9_denoiser_filter_sse2, BLOCK_64X32),
+                      make_tuple(&vp9_denoiser_filter_sse2, BLOCK_64X64)));
+#endif  // HAVE_SSE2
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(
+    NEON, VP9DenoiserTest,
+    ::testing::Values(make_tuple(&vp9_denoiser_filter_neon, BLOCK_8X8),
+                      make_tuple(&vp9_denoiser_filter_neon, BLOCK_8X16),
+                      make_tuple(&vp9_denoiser_filter_neon, BLOCK_16X8),
+                      make_tuple(&vp9_denoiser_filter_neon, BLOCK_16X16),
+                      make_tuple(&vp9_denoiser_filter_neon, BLOCK_16X32),
+                      make_tuple(&vp9_denoiser_filter_neon, BLOCK_32X16),
+                      make_tuple(&vp9_denoiser_filter_neon, BLOCK_32X32),
+                      make_tuple(&vp9_denoiser_filter_neon, BLOCK_32X64),
+                      make_tuple(&vp9_denoiser_filter_neon, BLOCK_64X32),
+                      make_tuple(&vp9_denoiser_filter_neon, BLOCK_64X64)));
+#endif
 }  // namespace
--- a/test/vp9_encoder_parms_get_to_decoder.cc
+++ b/test/vp9_encoder_parms_get_to_decoder.cc
@@ -99,9 +99,7 @@ class VpxEncoderParmsGetToDecoder
    vpx_codec_ctx_t *const vp9_decoder = decoder->GetDecoder();
    vpx_codec_alg_priv_t *const priv =
        reinterpret_cast<vpx_codec_alg_priv_t *>(vp9_decoder->priv);
-    FrameWorkerData *const worker_data =
-        reinterpret_cast<FrameWorkerData *>(priv->frame_workers[0].data1);
-    VP9_COMMON *const common = &worker_data->pbi->common;
+    VP9_COMMON *const common = &priv->pbi->common;

    if (encode_parms.lossless) {
      EXPECT_EQ(0, common->base_qindex);
--- a/test/vp9_ethread_test.cc
+++ b/test/vp9_ethread_test.cc
@@ -16,8 +16,207 @@
 #include "test/md5_helper.h"
 #include "test/util.h"
 #include "test/y4m_video_source.h"
+#include "vp9/encoder/vp9_firstpass.h"

 namespace {
+// FIRSTPASS_STATS struct:
+// {
+//   25 double members;
+//   1 int64_t member;
+// }
+// Whenever FIRSTPASS_STATS struct is modified, the following constants need to
+// be revisited.
+const int kDbl = 25;
+const int kInt = 1;
+const size_t kFirstPassStatsSz = kDbl * sizeof(double) + kInt * sizeof(int64_t);
+
+class VPxFirstPassEncoderThreadTest
+    : public ::libvpx_test::EncoderTest,
+      public ::libvpx_test::CodecTestWith2Params<libvpx_test::TestMode, int> {
+ protected:
+  VPxFirstPassEncoderThreadTest()
+      : EncoderTest(GET_PARAM(0)), encoder_initialized_(false), tiles_(0),
+        encoding_mode_(GET_PARAM(1)), set_cpu_used_(GET_PARAM(2)) {
+    init_flags_ = VPX_CODEC_USE_PSNR;
+
+    row_mt_mode_ = 1;
+    first_pass_only_ = true;
+    firstpass_stats_.buf = NULL;
+    firstpass_stats_.sz = 0;
+  }
+  virtual ~VPxFirstPassEncoderThreadTest() { free(firstpass_stats_.buf); }
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+
+    cfg_.rc_end_usage = VPX_VBR;
+    cfg_.rc_2pass_vbr_minsection_pct = 5;
+    cfg_.rc_2pass_vbr_maxsection_pct = 2000;
+    cfg_.rc_max_quantizer = 56;
+    cfg_.rc_min_quantizer = 0;
+  }
+
+  virtual void BeginPassHook(unsigned int /*pass*/) {
+    encoder_initialized_ = false;
+    abort_ = false;
+  }
+
+  virtual void EndPassHook() {
+    // For first pass stats test, only run first pass encoder.
+    if (first_pass_only_ && cfg_.g_pass == VPX_RC_FIRST_PASS)
+      abort_ |= first_pass_only_;
+  }
+
+  virtual void PreEncodeFrameHook(::libvpx_test::VideoSource * /*video*/,
+                                  ::libvpx_test::Encoder *encoder) {
+    if (!encoder_initialized_) {
+      // Encode in 2-pass mode.
+      encoder->Control(VP9E_SET_TILE_COLUMNS, tiles_);
+      encoder->Control(VP8E_SET_CPUUSED, set_cpu_used_);
+      encoder->Control(VP8E_SET_ENABLEAUTOALTREF, 1);
+      encoder->Control(VP8E_SET_ARNR_MAXFRAMES, 7);
+      encoder->Control(VP8E_SET_ARNR_STRENGTH, 5);
+      encoder->Control(VP8E_SET_ARNR_TYPE, 3);
+      encoder->Control(VP9E_SET_FRAME_PARALLEL_DECODING, 0);
+
+      if (encoding_mode_ == ::libvpx_test::kTwoPassGood)
+        encoder->Control(VP9E_SET_ROW_MT, row_mt_mode_);
+
+      encoder_initialized_ = true;
+    }
+  }
+
+  virtual void StatsPktHook(const vpx_codec_cx_pkt_t *pkt) {
+    const uint8_t *const pkt_buf =
+        reinterpret_cast<uint8_t *>(pkt->data.twopass_stats.buf);
+    const size_t pkt_size = pkt->data.twopass_stats.sz;
+
+    // First pass stats size equals sizeof(FIRSTPASS_STATS)
+    EXPECT_EQ(pkt_size, kFirstPassStatsSz)
+        << "Error: First pass stats size doesn't equal kFirstPassStatsSz";
+
+    firstpass_stats_.buf =
+        realloc(firstpass_stats_.buf, firstpass_stats_.sz + pkt_size);
+    memcpy((uint8_t *)firstpass_stats_.buf + firstpass_stats_.sz, pkt_buf,
+           pkt_size);
+    firstpass_stats_.sz += pkt_size;
+  }
+
+  bool encoder_initialized_;
+  int tiles_;
+  ::libvpx_test::TestMode encoding_mode_;
+  int set_cpu_used_;
+  int row_mt_mode_;
+  bool first_pass_only_;
+  vpx_fixed_buf_t firstpass_stats_;
+};
+
+static void compare_fp_stats(vpx_fixed_buf_t *fp_stats, double factor) {
+  // fp_stats consists of 2 set of first pass encoding stats. These 2 set of
+  // stats are compared to check if the stats match or at least are very close.
+  FIRSTPASS_STATS *stats1 = reinterpret_cast<FIRSTPASS_STATS *>(fp_stats->buf);
+  int nframes_ = (int)(fp_stats->sz / sizeof(FIRSTPASS_STATS));
+  FIRSTPASS_STATS *stats2 = stats1 + nframes_ / 2;
+  int i, j;
+
+  // The total stats are also output and included in the first pass stats. Here
+  // ignore that in the comparison.
+  for (i = 0; i < (nframes_ / 2 - 1); ++i) {
+    const double *frame_stats1 = reinterpret_cast<double *>(stats1);
+    const double *frame_stats2 = reinterpret_cast<double *>(stats2);
+
+    for (j = 0; j < kDbl; ++j) {
+      ASSERT_LE(fabs(*frame_stats1 - *frame_stats2),
+                fabs(*frame_stats1) / factor)
+          << "First failure @ frame #" << i << " stat #" << j << " ("
+          << *frame_stats1 << " vs. " << *frame_stats2 << ")";
+      frame_stats1++;
+      frame_stats2++;
+    }
+
+    stats1++;
+    stats2++;
+  }
+
+  // Reset firstpass_stats_ to 0.
+  memset((uint8_t *)fp_stats->buf, 0, fp_stats->sz);
+  fp_stats->sz = 0;
+}
+
+static void compare_fp_stats_md5(vpx_fixed_buf_t *fp_stats) {
+  // fp_stats consists of 2 set of first pass encoding stats. These 2 set of
+  // stats are compared to check if the stats match.
+  uint8_t *stats1 = reinterpret_cast<uint8_t *>(fp_stats->buf);
+  uint8_t *stats2 = stats1 + fp_stats->sz / 2;
+  ::libvpx_test::MD5 md5_row_mt_0, md5_row_mt_1;
+
+  md5_row_mt_0.Add(stats1, fp_stats->sz / 2);
+  const char *md5_row_mt_0_str = md5_row_mt_0.Get();
+
+  md5_row_mt_1.Add(stats2, fp_stats->sz / 2);
+  const char *md5_row_mt_1_str = md5_row_mt_1.Get();
+
+  // Check md5 match.
+  ASSERT_STREQ(md5_row_mt_0_str, md5_row_mt_1_str)
+      << "MD5 checksums don't match";
+
+  // Reset firstpass_stats_ to 0.
+  memset((uint8_t *)fp_stats->buf, 0, fp_stats->sz);
+  fp_stats->sz = 0;
+}
+
+TEST_P(VPxFirstPassEncoderThreadTest, FirstPassStatsTest) {
+  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 0, 60);
+
+  first_pass_only_ = true;
+  cfg_.rc_target_bitrate = 1000;
+
+  // Test row_mt_mode: 0 vs 1 at single thread case(threads = 1, tiles_ = 0)
+  tiles_ = 0;
+  cfg_.g_threads = 1;
+
+  row_mt_mode_ = 0;
+  init_flags_ = VPX_CODEC_USE_PSNR;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  row_mt_mode_ = 1;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  // Compare to check if using or not using row-mt generates close stats.
+  ASSERT_NO_FATAL_FAILURE(compare_fp_stats(&firstpass_stats_, 1000.0));
+
+  // Test single thread vs multiple threads
+  row_mt_mode_ = 1;
+  tiles_ = 0;
+
+  cfg_.g_threads = 1;
+  init_flags_ = VPX_CODEC_USE_PSNR;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  cfg_.g_threads = 4;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  // Compare to check if single-thread and multi-thread stats are close enough.
+  ASSERT_NO_FATAL_FAILURE(compare_fp_stats(&firstpass_stats_, 1000.0));
+
+  // Bit exact test in row_mt mode.
+  // When row_mt_mode_=1 and using >1 threads, the encoder generates bit exact
+  // result.
+  row_mt_mode_ = 1;
+  tiles_ = 2;
+
+  cfg_.g_threads = 2;
+  init_flags_ = VPX_CODEC_USE_PSNR;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  cfg_.g_threads = 8;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  // Compare to check if stats match with row-mt=0/1.
+  compare_fp_stats_md5(&firstpass_stats_);
+}
+
 class VPxEncoderThreadTest
    : public ::libvpx_test::EncoderTest,
      public ::libvpx_test::CodecTestWith4Params<libvpx_test::TestMode, int,
@@ -29,6 +228,9 @@ class VPxEncoderThreadTest
        encoding_mode_(GET_PARAM(1)), set_cpu_used_(GET_PARAM(2)) {
    init_flags_ = VPX_CODEC_USE_PSNR;
    md5_.clear();
+    row_mt_mode_ = 1;
+    psnr_ = 0.0;
+    nframes_ = 0;
  }
  virtual ~VPxEncoderThreadTest() {}

@@ -37,7 +239,6 @@ class VPxEncoderThreadTest
    SetMode(encoding_mode_);

    if (encoding_mode_ != ::libvpx_test::kRealTime) {
-      cfg_.g_lag_in_frames = 3;
      cfg_.rc_end_usage = VPX_VBR;
      cfg_.rc_2pass_vbr_minsection_pct = 5;
      cfg_.rc_2pass_vbr_maxsection_pct = 2000;
@@ -52,6 +253,8 @@ class VPxEncoderThreadTest

  virtual void BeginPassHook(unsigned int /*pass*/) {
    encoder_initialized_ = false;
+    psnr_ = 0.0;
+    nframes_ = 0;
  }

  virtual void PreEncodeFrameHook(::libvpx_test::VideoSource * /*video*/,
@@ -70,10 +273,17 @@ class VPxEncoderThreadTest
        encoder->Control(VP8E_SET_ENABLEAUTOALTREF, 0);
        encoder->Control(VP9E_SET_AQ_MODE, 3);
      }
+      encoder->Control(VP9E_SET_ROW_MT, row_mt_mode_);
+
      encoder_initialized_ = true;
    }
  }

+  virtual void PSNRPktHook(const vpx_codec_cx_pkt_t *pkt) {
+    psnr_ += pkt->data.psnr.psnr[0];
+    nframes_++;
+  }
+
  virtual void DecompressedFrameHook(const vpx_image_t &img,
                                     vpx_codec_pts_t /*pts*/) {
    ::libvpx_test::MD5 md5_res;
@@ -92,38 +302,102 @@ class VPxEncoderThreadTest
    return true;
  }

+  double GetAveragePsnr() const { return nframes_ ? (psnr_ / nframes_) : 0.0; }
+
  bool encoder_initialized_;
  int tiles_;
  int threads_;
  ::libvpx_test::TestMode encoding_mode_;
  int set_cpu_used_;
+  int row_mt_mode_;
+  double psnr_;
+  unsigned int nframes_;
  std::vector<std::string> md5_;
 };

 TEST_P(VPxEncoderThreadTest, EncoderResultTest) {
-  std::vector<std::string> single_thr_md5, multi_thr_md5;
-
  ::libvpx_test::Y4mVideoSource video("niklas_1280_720_30.y4m", 15, 20);
-
  cfg_.rc_target_bitrate = 1000;

+  // Part 1: Bit exact test for row_mt_mode_ = 0.
+  // This part keeps original unit tests done before row-mt code is checked in.
+  row_mt_mode_ = 0;
+
  // Encode using single thread.
  cfg_.g_threads = 1;
  init_flags_ = VPX_CODEC_USE_PSNR;
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  single_thr_md5 = md5_;
+  const std::vector<std::string> single_thr_md5 = md5_;
  md5_.clear();

  // Encode using multiple threads.
  cfg_.g_threads = threads_;
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-  multi_thr_md5 = md5_;
+  const std::vector<std::string> multi_thr_md5 = md5_;
  md5_.clear();

  // Compare to check if two vectors are equal.
  ASSERT_EQ(single_thr_md5, multi_thr_md5);
+
+  // Part 2: row_mt_mode_ = 0 vs row_mt_mode_ = 1 single thread bit exact test.
+  row_mt_mode_ = 1;
+
+  // Encode using single thread
+  cfg_.g_threads = 1;
+  init_flags_ = VPX_CODEC_USE_PSNR;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  std::vector<std::string> row_mt_single_thr_md5 = md5_;
+  md5_.clear();
+
+  ASSERT_EQ(single_thr_md5, row_mt_single_thr_md5);
+
+  // Part 3: Bit exact test with row-mt on
+  // When row_mt_mode_=1 and using >1 threads, the encoder generates bit exact
+  // result.
+  row_mt_mode_ = 1;
+  row_mt_single_thr_md5.clear();
+
+  // Encode using 2 threads.
+  cfg_.g_threads = 2;
+  init_flags_ = VPX_CODEC_USE_PSNR;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  row_mt_single_thr_md5 = md5_;
+  md5_.clear();
+
+  // Encode using multiple threads.
+  cfg_.g_threads = threads_;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  const std::vector<std::string> row_mt_multi_thr_md5 = md5_;
+  md5_.clear();
+
+  // Compare to check if two vectors are equal.
+  ASSERT_EQ(row_mt_single_thr_md5, row_mt_multi_thr_md5);
+
+  // Part 4: PSNR test with bit_match_mode_ = 0
+  row_mt_mode_ = 1;
+
+  // Encode using single thread.
+  cfg_.g_threads = 1;
+  init_flags_ = VPX_CODEC_USE_PSNR;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  const double single_thr_psnr = GetAveragePsnr();
+
+  // Encode using multiple threads.
+  cfg_.g_threads = threads_;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  const double multi_thr_psnr = GetAveragePsnr();
+
+  EXPECT_NEAR(single_thr_psnr, multi_thr_psnr, 0.1);
 }

+INSTANTIATE_TEST_CASE_P(
+    VP9, VPxFirstPassEncoderThreadTest,
+    ::testing::Combine(
+        ::testing::Values(
+            static_cast<const libvpx_test::CodecFactory *>(&libvpx_test::kVP9)),
+        ::testing::Values(::libvpx_test::kTwoPassGood),
+        ::testing::Range(0, 4)));  // cpu_used
+
 // Split this into two instantiations so that we can distinguish
 // between very slow runs ( ie cpu_speed 0 ) vs ones that can be
 // run nightly by adding Large to the title.
@@ -135,7 +409,7 @@ INSTANTIATE_TEST_CASE_P(
        ::testing::Values(::libvpx_test::kTwoPassGood,
                          ::libvpx_test::kOnePassGood,
                          ::libvpx_test::kRealTime),
-        ::testing::Range(2, 9),    // cpu_used
+        ::testing::Range(3, 9),    // cpu_used
        ::testing::Range(0, 3),    // tile_columns
        ::testing::Range(2, 5)));  // threads

@@ -147,7 +421,7 @@ INSTANTIATE_TEST_CASE_P(
        ::testing::Values(::libvpx_test::kTwoPassGood,
                          ::libvpx_test::kOnePassGood,
                          ::libvpx_test::kRealTime),
-        ::testing::Range(0, 2),    // cpu_used
+        ::testing::Range(0, 3),    // cpu_used
        ::testing::Range(0, 3),    // tile_columns
        ::testing::Range(2, 5)));  // threads

--- a/test/vp9_frame_parallel_test.cc
+++ b/test/vp9_frame_parallel_test.cc
@@ -1,217 +0,0 @@
-/*
- *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <cstdio>
-#include <cstdlib>
-#include <string>
-#include "third_party/googletest/src/include/gtest/gtest.h"
-#include "./vpx_config.h"
-#include "test/codec_factory.h"
-#include "test/decode_test_driver.h"
-#include "test/ivf_video_source.h"
-#include "test/md5_helper.h"
-#include "test/util.h"
-#if CONFIG_WEBM_IO
-#include "test/webm_video_source.h"
-#endif
-#include "vpx_mem/vpx_mem.h"
-
-namespace {
-
-using std::string;
-
-#if CONFIG_WEBM_IO
-
-struct PauseFileList {
-  const char *name;
-  // md5 sum for decoded frames which does not include skipped frames.
-  const char *expected_md5;
-  const int pause_frame_num;
-};
-
-// Decodes |filename| with |num_threads|. Pause at the specified frame_num,
-// seek to next key frame and then continue decoding until the end. Return
-// the md5 of the decoded frames which does not include skipped frames.
-string DecodeFileWithPause(const string &filename, int num_threads,
-                           int pause_num) {
-  libvpx_test::WebMVideoSource video(filename);
-  video.Init();
-  int in_frames = 0;
-  int out_frames = 0;
-
-  vpx_codec_dec_cfg_t cfg = vpx_codec_dec_cfg_t();
-  cfg.threads = num_threads;
-  vpx_codec_flags_t flags = 0;
-  flags |= VPX_CODEC_USE_FRAME_THREADING;
-  libvpx_test::VP9Decoder decoder(cfg, flags);
-
-  libvpx_test::MD5 md5;
-  video.Begin();
-
-  do {
-    ++in_frames;
-    const vpx_codec_err_t res =
-        decoder.DecodeFrame(video.cxdata(), video.frame_size());
-    if (res != VPX_CODEC_OK) {
-      EXPECT_EQ(VPX_CODEC_OK, res) << decoder.DecodeError();
-      break;
-    }
-
-    // Pause at specified frame number.
-    if (in_frames == pause_num) {
-      // Flush the decoder and then seek to next key frame.
-      decoder.DecodeFrame(NULL, 0);
-      video.SeekToNextKeyFrame();
-    } else {
-      video.Next();
-    }
-
-    // Flush the decoder at the end of the video.
-    if (!video.cxdata()) decoder.DecodeFrame(NULL, 0);
-
-    libvpx_test::DxDataIterator dec_iter = decoder.GetDxData();
-    const vpx_image_t *img;
-
-    // Get decompressed data
-    while ((img = dec_iter.Next())) {
-      ++out_frames;
-      md5.Add(img);
-    }
-  } while (video.cxdata() != NULL);
-
-  EXPECT_EQ(in_frames, out_frames)
-      << "Input frame count does not match output frame count";
-
-  return string(md5.Get());
-}
-
-void DecodeFilesWithPause(const PauseFileList files[]) {
-  for (const PauseFileList *iter = files; iter->name != NULL; ++iter) {
-    SCOPED_TRACE(iter->name);
-    for (int t = 2; t <= 8; ++t) {
-      EXPECT_EQ(iter->expected_md5,
-                DecodeFileWithPause(iter->name, t, iter->pause_frame_num))
-          << "threads = " << t;
-    }
-  }
-}
-
-TEST(VP9MultiThreadedFrameParallel, PauseSeekResume) {
-  // vp90-2-07-frame_parallel-1.webm is a 40 frame video file with
-  // one key frame for every ten frames.
-  static const PauseFileList files[] = {
-    { "vp90-2-07-frame_parallel-1.webm", "6ea7c3875d67252e7caf2bc6e75b36b1",
-      6 },
-    { "vp90-2-07-frame_parallel-1.webm", "4bb634160c7356a8d7d4299b6dc83a45",
-      12 },
-    { "vp90-2-07-frame_parallel-1.webm", "89772591e6ef461f9fa754f916c78ed8",
-      26 },
-    { NULL, NULL, 0 },
-  };
-  DecodeFilesWithPause(files);
-}
-
-struct FileList {
-  const char *name;
-  // md5 sum for decoded frames which does not include corrupted frames.
-  const char *expected_md5;
-  // Expected number of decoded frames which does not include corrupted frames.
-  const int expected_frame_count;
-};
-
-// Decodes |filename| with |num_threads|. Return the md5 of the decoded
-// frames which does not include corrupted frames.
-string DecodeFile(const string &filename, int num_threads,
-                  int expected_frame_count) {
-  libvpx_test::WebMVideoSource video(filename);
-  video.Init();
-
-  vpx_codec_dec_cfg_t cfg = vpx_codec_dec_cfg_t();
-  cfg.threads = num_threads;
-  const vpx_codec_flags_t flags = VPX_CODEC_USE_FRAME_THREADING;
-  libvpx_test::VP9Decoder decoder(cfg, flags);
-
-  libvpx_test::MD5 md5;
-  video.Begin();
-
-  int out_frames = 0;
-  do {
-    const vpx_codec_err_t res =
-        decoder.DecodeFrame(video.cxdata(), video.frame_size());
-    // TODO(hkuang): frame parallel mode should return an error on corruption.
-    if (res != VPX_CODEC_OK) {
-      EXPECT_EQ(VPX_CODEC_OK, res) << decoder.DecodeError();
-      break;
-    }
-
-    video.Next();
-
-    // Flush the decoder at the end of the video.
-    if (!video.cxdata()) decoder.DecodeFrame(NULL, 0);
-
-    libvpx_test::DxDataIterator dec_iter = decoder.GetDxData();
-    const vpx_image_t *img;
-
-    // Get decompressed data
-    while ((img = dec_iter.Next())) {
-      ++out_frames;
-      md5.Add(img);
-    }
-  } while (video.cxdata() != NULL);
-
-  EXPECT_EQ(expected_frame_count, out_frames)
-      << "Input frame count does not match expected output frame count";
-
-  return string(md5.Get());
-}
-
-void DecodeFiles(const FileList files[]) {
-  for (const FileList *iter = files; iter->name != NULL; ++iter) {
-    SCOPED_TRACE(iter->name);
-    for (int t = 2; t <= 8; ++t) {
-      EXPECT_EQ(iter->expected_md5,
-                DecodeFile(iter->name, t, iter->expected_frame_count))
-          << "threads = " << t;
-    }
-  }
-}
-
-TEST(VP9MultiThreadedFrameParallel, InvalidFileTest) {
-  static const FileList files[] = {
-    // invalid-vp90-2-07-frame_parallel-1.webm is a 40 frame video file with
-    // one key frame for every ten frames. The 11th frame has corrupted data.
-    { "invalid-vp90-2-07-frame_parallel-1.webm",
-      "0549d0f45f60deaef8eb708e6c0eb6cb", 30 },
-    // invalid-vp90-2-07-frame_parallel-2.webm is a 40 frame video file with
-    // one key frame for every ten frames. The 1st and 31st frames have
-    // corrupted data.
-    { "invalid-vp90-2-07-frame_parallel-2.webm",
-      "6a1f3cf6f9e7a364212fadb9580d525e", 20 },
-    // invalid-vp90-2-07-frame_parallel-3.webm is a 40 frame video file with
-    // one key frame for every ten frames. The 5th and 13th frames have
-    // corrupted data.
-    { "invalid-vp90-2-07-frame_parallel-3.webm",
-      "8256544308de926b0681e04685b98677", 27 },
-    { NULL, NULL, 0 },
-  };
-  DecodeFiles(files);
-}
-
-TEST(VP9MultiThreadedFrameParallel, ValidFileTest) {
-  static const FileList files[] = {
-#if CONFIG_VP9_HIGHBITDEPTH
-    { "vp92-2-20-10bit-yuv420.webm", "a16b99df180c584e8db2ffeda987d293", 10 },
-#endif
-    { NULL, NULL, 0 },
-  };
-  DecodeFiles(files);
-}
-#endif  // CONFIG_WEBM_IO
-}  // namespace
--- a/test/vp9_intrapred_test.cc
+++ b/test/vp9_intrapred_test.cc
@@ -378,6 +378,60 @@ INSTANTIATE_TEST_CASE_P(
                       8)));
 #endif  // HAVE_MSA

+#if HAVE_VSX
+INSTANTIATE_TEST_CASE_P(
+    VSX, VP9IntraPredTest,
+    ::testing::Values(
+        IntraPredParam(&vpx_d45_predictor_8x8_vsx, &vpx_d45_predictor_8x8_c, 8,
+                       8),
+        IntraPredParam(&vpx_d45_predictor_16x16_vsx, &vpx_d45_predictor_16x16_c,
+                       16, 8),
+        IntraPredParam(&vpx_d45_predictor_32x32_vsx, &vpx_d45_predictor_32x32_c,
+                       32, 8),
+        IntraPredParam(&vpx_d63_predictor_8x8_vsx, &vpx_d63_predictor_8x8_c, 8,
+                       8),
+        IntraPredParam(&vpx_d63_predictor_16x16_vsx, &vpx_d63_predictor_16x16_c,
+                       16, 8),
+        IntraPredParam(&vpx_d63_predictor_32x32_vsx, &vpx_d63_predictor_32x32_c,
+                       32, 8),
+        IntraPredParam(&vpx_dc_128_predictor_16x16_vsx,
+                       &vpx_dc_128_predictor_16x16_c, 16, 8),
+        IntraPredParam(&vpx_dc_128_predictor_32x32_vsx,
+                       &vpx_dc_128_predictor_32x32_c, 32, 8),
+        IntraPredParam(&vpx_dc_left_predictor_16x16_vsx,
+                       &vpx_dc_left_predictor_16x16_c, 16, 8),
+        IntraPredParam(&vpx_dc_left_predictor_32x32_vsx,
+                       &vpx_dc_left_predictor_32x32_c, 32, 8),
+        IntraPredParam(&vpx_dc_predictor_8x8_vsx, &vpx_dc_predictor_8x8_c, 8,
+                       8),
+        IntraPredParam(&vpx_dc_predictor_16x16_vsx, &vpx_dc_predictor_16x16_c,
+                       16, 8),
+        IntraPredParam(&vpx_dc_predictor_32x32_vsx, &vpx_dc_predictor_32x32_c,
+                       32, 8),
+        IntraPredParam(&vpx_dc_top_predictor_16x16_vsx,
+                       &vpx_dc_top_predictor_16x16_c, 16, 8),
+        IntraPredParam(&vpx_dc_top_predictor_32x32_vsx,
+                       &vpx_dc_top_predictor_32x32_c, 32, 8),
+        IntraPredParam(&vpx_h_predictor_4x4_vsx, &vpx_h_predictor_4x4_c, 4, 8),
+        IntraPredParam(&vpx_h_predictor_8x8_vsx, &vpx_h_predictor_8x8_c, 8, 8),
+        IntraPredParam(&vpx_h_predictor_16x16_vsx, &vpx_h_predictor_16x16_c, 16,
+                       8),
+        IntraPredParam(&vpx_h_predictor_32x32_vsx, &vpx_h_predictor_32x32_c, 32,
+                       8),
+        IntraPredParam(&vpx_tm_predictor_4x4_vsx, &vpx_tm_predictor_4x4_c, 4,
+                       8),
+        IntraPredParam(&vpx_tm_predictor_8x8_vsx, &vpx_tm_predictor_8x8_c, 8,
+                       8),
+        IntraPredParam(&vpx_tm_predictor_16x16_vsx, &vpx_tm_predictor_16x16_c,
+                       16, 8),
+        IntraPredParam(&vpx_tm_predictor_32x32_vsx, &vpx_tm_predictor_32x32_c,
+                       32, 8),
+        IntraPredParam(&vpx_v_predictor_16x16_vsx, &vpx_v_predictor_16x16_c, 16,
+                       8),
+        IntraPredParam(&vpx_v_predictor_32x32_vsx, &vpx_v_predictor_32x32_c, 32,
+                       8)));
+#endif  // HAVE_VSX
+
 #if CONFIG_VP9_HIGHBITDEPTH
 typedef void (*HighbdIntraPred)(uint16_t *dst, ptrdiff_t stride,
                                const uint16_t *above, const uint16_t *left,
@@ -413,10 +467,164 @@ TEST_P(VP9HighbdIntraPredTest, HighbdIntraPredTests) {
  RunTest(left_col, above_data, dst, ref_dst);
 }

+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(
+    SSSE3_TO_C_8, VP9HighbdIntraPredTest,
+    ::testing::Values(
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_4x4_ssse3,
+                             &vpx_highbd_d45_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_8x8_ssse3,
+                             &vpx_highbd_d45_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_16x16_ssse3,
+                             &vpx_highbd_d45_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_32x32_ssse3,
+                             &vpx_highbd_d45_predictor_32x32_c, 32, 8),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_8x8_ssse3,
+                             &vpx_highbd_d63_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_16x16_ssse3,
+                             &vpx_highbd_d63_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_32x32_c,
+                             &vpx_highbd_d63_predictor_32x32_ssse3, 32, 8),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_8x8_ssse3,
+                             &vpx_highbd_d117_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_16x16_ssse3,
+                             &vpx_highbd_d117_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_32x32_c,
+                             &vpx_highbd_d117_predictor_32x32_ssse3, 32, 8),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_8x8_ssse3,
+                             &vpx_highbd_d135_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_16x16_ssse3,
+                             &vpx_highbd_d135_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_32x32_ssse3,
+                             &vpx_highbd_d135_predictor_32x32_c, 32, 8),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_8x8_ssse3,
+                             &vpx_highbd_d153_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_16x16_ssse3,
+                             &vpx_highbd_d153_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_32x32_ssse3,
+                             &vpx_highbd_d153_predictor_32x32_c, 32, 8),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_8x8_ssse3,
+                             &vpx_highbd_d207_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_16x16_ssse3,
+                             &vpx_highbd_d207_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_32x32_ssse3,
+                             &vpx_highbd_d207_predictor_32x32_c, 32, 8)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSSE3_TO_C_10, VP9HighbdIntraPredTest,
+    ::testing::Values(
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_4x4_ssse3,
+                             &vpx_highbd_d45_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_8x8_ssse3,
+                             &vpx_highbd_d45_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_16x16_ssse3,
+                             &vpx_highbd_d45_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_32x32_ssse3,
+                             &vpx_highbd_d45_predictor_32x32_c, 32, 10),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_8x8_ssse3,
+                             &vpx_highbd_d63_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_16x16_ssse3,
+                             &vpx_highbd_d63_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_32x32_c,
+                             &vpx_highbd_d63_predictor_32x32_ssse3, 32, 10),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_8x8_ssse3,
+                             &vpx_highbd_d117_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_16x16_ssse3,
+                             &vpx_highbd_d117_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_32x32_c,
+                             &vpx_highbd_d117_predictor_32x32_ssse3, 32, 10),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_8x8_ssse3,
+                             &vpx_highbd_d135_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_16x16_ssse3,
+                             &vpx_highbd_d135_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_32x32_ssse3,
+                             &vpx_highbd_d135_predictor_32x32_c, 32, 10),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_8x8_ssse3,
+                             &vpx_highbd_d153_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_16x16_ssse3,
+                             &vpx_highbd_d153_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_32x32_ssse3,
+                             &vpx_highbd_d153_predictor_32x32_c, 32, 10),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_8x8_ssse3,
+                             &vpx_highbd_d207_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_16x16_ssse3,
+                             &vpx_highbd_d207_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_32x32_ssse3,
+                             &vpx_highbd_d207_predictor_32x32_c, 32, 10)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSSE3_TO_C_12, VP9HighbdIntraPredTest,
+    ::testing::Values(
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_4x4_ssse3,
+                             &vpx_highbd_d45_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_8x8_ssse3,
+                             &vpx_highbd_d45_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_16x16_ssse3,
+                             &vpx_highbd_d45_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_d45_predictor_32x32_ssse3,
+                             &vpx_highbd_d45_predictor_32x32_c, 32, 12),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_8x8_ssse3,
+                             &vpx_highbd_d63_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_16x16_ssse3,
+                             &vpx_highbd_d63_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_32x32_c,
+                             &vpx_highbd_d63_predictor_32x32_ssse3, 32, 12),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_8x8_ssse3,
+                             &vpx_highbd_d117_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_16x16_ssse3,
+                             &vpx_highbd_d117_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_32x32_c,
+                             &vpx_highbd_d117_predictor_32x32_ssse3, 32, 12),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_8x8_ssse3,
+                             &vpx_highbd_d135_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_16x16_ssse3,
+                             &vpx_highbd_d135_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_32x32_ssse3,
+                             &vpx_highbd_d135_predictor_32x32_c, 32, 12),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_8x8_ssse3,
+                             &vpx_highbd_d153_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_16x16_ssse3,
+                             &vpx_highbd_d153_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_32x32_ssse3,
+                             &vpx_highbd_d153_predictor_32x32_c, 32, 12),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_8x8_ssse3,
+                             &vpx_highbd_d207_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_16x16_ssse3,
+                             &vpx_highbd_d207_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_32x32_ssse3,
+                             &vpx_highbd_d207_predictor_32x32_c, 32, 12)));
+#endif  // HAVE_SSSE3
+
 #if HAVE_SSE2
 INSTANTIATE_TEST_CASE_P(
    SSE2_TO_C_8, VP9HighbdIntraPredTest,
    ::testing::Values(
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_4x4_sse2,
+                             &vpx_highbd_dc_128_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_8x8_sse2,
+                             &vpx_highbd_dc_128_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_16x16_sse2,
+                             &vpx_highbd_dc_128_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_32x32_sse2,
+                             &vpx_highbd_dc_128_predictor_32x32_c, 32, 8),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_4x4_sse2,
+                             &vpx_highbd_d63_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_4x4_sse2,
+                             &vpx_highbd_d117_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_4x4_sse2,
+                             &vpx_highbd_d135_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_4x4_sse2,
+                             &vpx_highbd_d153_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_4x4_sse2,
+                             &vpx_highbd_d207_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_4x4_sse2,
+                             &vpx_highbd_dc_left_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_8x8_sse2,
+                             &vpx_highbd_dc_left_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_16x16_sse2,
+                             &vpx_highbd_dc_left_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_32x32_sse2,
+                             &vpx_highbd_dc_left_predictor_32x32_c, 32, 8),
        HighbdIntraPredParam(&vpx_highbd_dc_predictor_4x4_sse2,
                             &vpx_highbd_dc_predictor_4x4_c, 4, 8),
        HighbdIntraPredParam(&vpx_highbd_dc_predictor_8x8_sse2,
@@ -425,6 +633,14 @@ INSTANTIATE_TEST_CASE_P(
                             &vpx_highbd_dc_predictor_16x16_c, 16, 8),
        HighbdIntraPredParam(&vpx_highbd_dc_predictor_32x32_sse2,
                             &vpx_highbd_dc_predictor_32x32_c, 32, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_4x4_sse2,
+                             &vpx_highbd_dc_top_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_8x8_sse2,
+                             &vpx_highbd_dc_top_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_16x16_sse2,
+                             &vpx_highbd_dc_top_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_32x32_sse2,
+                             &vpx_highbd_dc_top_predictor_32x32_c, 32, 8),
        HighbdIntraPredParam(&vpx_highbd_tm_predictor_4x4_sse2,
                             &vpx_highbd_tm_predictor_4x4_c, 4, 8),
        HighbdIntraPredParam(&vpx_highbd_tm_predictor_8x8_sse2,
@@ -433,6 +649,14 @@ INSTANTIATE_TEST_CASE_P(
                             &vpx_highbd_tm_predictor_16x16_c, 16, 8),
        HighbdIntraPredParam(&vpx_highbd_tm_predictor_32x32_sse2,
                             &vpx_highbd_tm_predictor_32x32_c, 32, 8),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_4x4_sse2,
+                             &vpx_highbd_h_predictor_4x4_c, 4, 8),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_8x8_sse2,
+                             &vpx_highbd_h_predictor_8x8_c, 8, 8),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_16x16_sse2,
+                             &vpx_highbd_h_predictor_16x16_c, 16, 8),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_32x32_sse2,
+                             &vpx_highbd_h_predictor_32x32_c, 32, 8),
        HighbdIntraPredParam(&vpx_highbd_v_predictor_4x4_sse2,
                             &vpx_highbd_v_predictor_4x4_c, 4, 8),
        HighbdIntraPredParam(&vpx_highbd_v_predictor_8x8_sse2,
@@ -445,6 +669,32 @@ INSTANTIATE_TEST_CASE_P(
 INSTANTIATE_TEST_CASE_P(
    SSE2_TO_C_10, VP9HighbdIntraPredTest,
    ::testing::Values(
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_4x4_sse2,
+                             &vpx_highbd_dc_128_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_8x8_sse2,
+                             &vpx_highbd_dc_128_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_16x16_sse2,
+                             &vpx_highbd_dc_128_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_32x32_sse2,
+                             &vpx_highbd_dc_128_predictor_32x32_c, 32, 10),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_4x4_sse2,
+                             &vpx_highbd_d63_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_4x4_sse2,
+                             &vpx_highbd_d117_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_4x4_sse2,
+                             &vpx_highbd_d135_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_4x4_sse2,
+                             &vpx_highbd_d153_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_4x4_sse2,
+                             &vpx_highbd_d207_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_4x4_sse2,
+                             &vpx_highbd_dc_left_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_8x8_sse2,
+                             &vpx_highbd_dc_left_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_16x16_sse2,
+                             &vpx_highbd_dc_left_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_32x32_sse2,
+                             &vpx_highbd_dc_left_predictor_32x32_c, 32, 10),
        HighbdIntraPredParam(&vpx_highbd_dc_predictor_4x4_sse2,
                             &vpx_highbd_dc_predictor_4x4_c, 4, 10),
        HighbdIntraPredParam(&vpx_highbd_dc_predictor_8x8_sse2,
@@ -453,6 +703,14 @@ INSTANTIATE_TEST_CASE_P(
                             &vpx_highbd_dc_predictor_16x16_c, 16, 10),
        HighbdIntraPredParam(&vpx_highbd_dc_predictor_32x32_sse2,
                             &vpx_highbd_dc_predictor_32x32_c, 32, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_4x4_sse2,
+                             &vpx_highbd_dc_top_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_8x8_sse2,
+                             &vpx_highbd_dc_top_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_16x16_sse2,
+                             &vpx_highbd_dc_top_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_32x32_sse2,
+                             &vpx_highbd_dc_top_predictor_32x32_c, 32, 10),
        HighbdIntraPredParam(&vpx_highbd_tm_predictor_4x4_sse2,
                             &vpx_highbd_tm_predictor_4x4_c, 4, 10),
        HighbdIntraPredParam(&vpx_highbd_tm_predictor_8x8_sse2,
@@ -461,6 +719,14 @@ INSTANTIATE_TEST_CASE_P(
                             &vpx_highbd_tm_predictor_16x16_c, 16, 10),
        HighbdIntraPredParam(&vpx_highbd_tm_predictor_32x32_sse2,
                             &vpx_highbd_tm_predictor_32x32_c, 32, 10),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_4x4_sse2,
+                             &vpx_highbd_h_predictor_4x4_c, 4, 10),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_8x8_sse2,
+                             &vpx_highbd_h_predictor_8x8_c, 8, 10),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_16x16_sse2,
+                             &vpx_highbd_h_predictor_16x16_c, 16, 10),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_32x32_sse2,
+                             &vpx_highbd_h_predictor_32x32_c, 32, 10),
        HighbdIntraPredParam(&vpx_highbd_v_predictor_4x4_sse2,
                             &vpx_highbd_v_predictor_4x4_c, 4, 10),
        HighbdIntraPredParam(&vpx_highbd_v_predictor_8x8_sse2,
@@ -473,6 +739,32 @@ INSTANTIATE_TEST_CASE_P(
 INSTANTIATE_TEST_CASE_P(
    SSE2_TO_C_12, VP9HighbdIntraPredTest,
    ::testing::Values(
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_4x4_sse2,
+                             &vpx_highbd_dc_128_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_8x8_sse2,
+                             &vpx_highbd_dc_128_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_16x16_sse2,
+                             &vpx_highbd_dc_128_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_32x32_sse2,
+                             &vpx_highbd_dc_128_predictor_32x32_c, 32, 12),
+        HighbdIntraPredParam(&vpx_highbd_d63_predictor_4x4_sse2,
+                             &vpx_highbd_d63_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_d117_predictor_4x4_sse2,
+                             &vpx_highbd_d117_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_d135_predictor_4x4_sse2,
+                             &vpx_highbd_d135_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_d153_predictor_4x4_sse2,
+                             &vpx_highbd_d153_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_d207_predictor_4x4_sse2,
+                             &vpx_highbd_d207_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_4x4_sse2,
+                             &vpx_highbd_dc_left_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_8x8_sse2,
+                             &vpx_highbd_dc_left_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_16x16_sse2,
+                             &vpx_highbd_dc_left_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_32x32_sse2,
+                             &vpx_highbd_dc_left_predictor_32x32_c, 32, 12),
        HighbdIntraPredParam(&vpx_highbd_dc_predictor_4x4_sse2,
                             &vpx_highbd_dc_predictor_4x4_c, 4, 12),
        HighbdIntraPredParam(&vpx_highbd_dc_predictor_8x8_sse2,
@@ -481,6 +773,14 @@ INSTANTIATE_TEST_CASE_P(
                             &vpx_highbd_dc_predictor_16x16_c, 16, 12),
        HighbdIntraPredParam(&vpx_highbd_dc_predictor_32x32_sse2,
                             &vpx_highbd_dc_predictor_32x32_c, 32, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_4x4_sse2,
+                             &vpx_highbd_dc_top_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_8x8_sse2,
+                             &vpx_highbd_dc_top_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_16x16_sse2,
+                             &vpx_highbd_dc_top_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_32x32_sse2,
+                             &vpx_highbd_dc_top_predictor_32x32_c, 32, 12),
        HighbdIntraPredParam(&vpx_highbd_tm_predictor_4x4_sse2,
                             &vpx_highbd_tm_predictor_4x4_c, 4, 12),
        HighbdIntraPredParam(&vpx_highbd_tm_predictor_8x8_sse2,
@@ -489,6 +789,14 @@ INSTANTIATE_TEST_CASE_P(
                             &vpx_highbd_tm_predictor_16x16_c, 16, 12),
        HighbdIntraPredParam(&vpx_highbd_tm_predictor_32x32_sse2,
                             &vpx_highbd_tm_predictor_32x32_c, 32, 12),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_4x4_sse2,
+                             &vpx_highbd_h_predictor_4x4_c, 4, 12),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_8x8_sse2,
+                             &vpx_highbd_h_predictor_8x8_c, 8, 12),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_16x16_sse2,
+                             &vpx_highbd_h_predictor_16x16_c, 16, 12),
+        HighbdIntraPredParam(&vpx_highbd_h_predictor_32x32_sse2,
+                             &vpx_highbd_h_predictor_32x32_c, 32, 12),
        HighbdIntraPredParam(&vpx_highbd_v_predictor_4x4_sse2,
                             &vpx_highbd_v_predictor_4x4_c, 4, 12),
        HighbdIntraPredParam(&vpx_highbd_v_predictor_8x8_sse2,
--- a/test/vp9_motion_vector_test.cc
+++ b/test/vp9_motion_vector_test.cc
@@ -0,0 +1,97 @@
+/*
+ *  Copyright (c) 2017 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 "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/util.h"
+#include "test/yuv_video_source.h"
+
+namespace {
+#define MAX_EXTREME_MV 1
+#define MIN_EXTREME_MV 2
+
+// Encoding modes
+const libvpx_test::TestMode kEncodingModeVectors[] = {
+  ::libvpx_test::kTwoPassGood, ::libvpx_test::kOnePassGood,
+  ::libvpx_test::kRealTime,
+};
+
+// Encoding speeds
+const int kCpuUsedVectors[] = { 0, 1, 2, 3, 4, 5, 6 };
+
+// MV test modes: 1 - always use maximum MV; 2 - always use minimum MV.
+const int kMVTestModes[] = { MAX_EXTREME_MV, MIN_EXTREME_MV };
+
+class MotionVectorTestLarge
+    : public ::libvpx_test::EncoderTest,
+      public ::libvpx_test::CodecTestWith3Params<libvpx_test::TestMode, int,
+                                                 int> {
+ protected:
+  MotionVectorTestLarge()
+      : EncoderTest(GET_PARAM(0)), encoding_mode_(GET_PARAM(1)),
+        cpu_used_(GET_PARAM(2)), mv_test_mode_(GET_PARAM(3)) {}
+
+  virtual ~MotionVectorTestLarge() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+    if (encoding_mode_ != ::libvpx_test::kRealTime) {
+      cfg_.g_lag_in_frames = 3;
+      cfg_.rc_end_usage = VPX_VBR;
+    } else {
+      cfg_.g_lag_in_frames = 0;
+      cfg_.rc_end_usage = VPX_CBR;
+      cfg_.rc_buf_sz = 1000;
+      cfg_.rc_buf_initial_sz = 500;
+      cfg_.rc_buf_optimal_sz = 600;
+    }
+  }
+
+  virtual void PreEncodeFrameHook(::libvpx_test::VideoSource *video,
+                                  ::libvpx_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(VP8E_SET_CPUUSED, cpu_used_);
+      encoder->Control(VP9E_ENABLE_MOTION_VECTOR_UNIT_TEST, mv_test_mode_);
+      if (encoding_mode_ != ::libvpx_test::kRealTime) {
+        encoder->Control(VP8E_SET_ENABLEAUTOALTREF, 1);
+        encoder->Control(VP8E_SET_ARNR_MAXFRAMES, 7);
+        encoder->Control(VP8E_SET_ARNR_STRENGTH, 5);
+        encoder->Control(VP8E_SET_ARNR_TYPE, 3);
+      }
+    }
+  }
+
+  libvpx_test::TestMode encoding_mode_;
+  int cpu_used_;
+  int mv_test_mode_;
+};
+
+TEST_P(MotionVectorTestLarge, OverallTest) {
+  cfg_.rc_target_bitrate = 24000;
+  cfg_.g_profile = 0;
+  init_flags_ = VPX_CODEC_USE_PSNR;
+
+  testing::internal::scoped_ptr<libvpx_test::VideoSource> video;
+  video.reset(new libvpx_test::YUVVideoSource(
+      "niklas_640_480_30.yuv", VPX_IMG_FMT_I420, 3840, 2160,  // 2048, 1080,
+      30, 1, 0, 5));
+
+  ASSERT_TRUE(video.get() != NULL);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(video.get()));
+}
+
+VP9_INSTANTIATE_TEST_CASE(MotionVectorTestLarge,
+                          ::testing::ValuesIn(kEncodingModeVectors),
+                          ::testing::ValuesIn(kCpuUsedVectors),
+                          ::testing::ValuesIn(kMVTestModes));
+}  // namespace
--- a/test/vp9_quantize_test.cc
+++ b/test/vp9_quantize_test.cc
@@ -14,9 +14,11 @@

 #include "third_party/googletest/src/include/gtest/gtest.h"

+#include "./vp9_rtcd.h"
 #include "./vpx_config.h"
 #include "./vpx_dsp_rtcd.h"
 #include "test/acm_random.h"
+#include "test/buffer.h"
 #include "test/clear_system_state.h"
 #include "test/register_state_check.h"
 #include "test/util.h"
@@ -24,11 +26,12 @@
 #include "vp9/common/vp9_scan.h"
 #include "vpx/vpx_codec.h"
 #include "vpx/vpx_integer.h"
+#include "vpx_ports/vpx_timer.h"

 using libvpx_test::ACMRandom;
+using libvpx_test::Buffer;

 namespace {
-#if CONFIG_VP9_HIGHBITDEPTH
 const int number_of_iterations = 100;

 typedef void (*QuantizeFunc)(const tran_low_t *coeff, intptr_t count,
@@ -38,307 +41,494 @@ typedef void (*QuantizeFunc)(const tran_low_t *coeff, intptr_t count,
                             tran_low_t *dqcoeff, const int16_t *dequant,
                             uint16_t *eob, const int16_t *scan,
                             const int16_t *iscan);
-typedef std::tr1::tuple<QuantizeFunc, QuantizeFunc, vpx_bit_depth_t>
+typedef std::tr1::tuple<QuantizeFunc, QuantizeFunc, vpx_bit_depth_t,
+                        int /*max_size*/, bool /*is_fp*/>
    QuantizeParam;

-class VP9QuantizeTest : public ::testing::TestWithParam<QuantizeParam> {
+// Wrapper for FP version which does not use zbin or quant_shift.
+typedef void (*QuantizeFPFunc)(const tran_low_t *coeff, intptr_t count,
+                               int skip_block, const int16_t *round,
+                               const int16_t *quant, tran_low_t *qcoeff,
+                               tran_low_t *dqcoeff, const int16_t *dequant,
+                               uint16_t *eob, const int16_t *scan,
+                               const int16_t *iscan);
+
+template <QuantizeFPFunc fn>
+void QuantFPWrapper(const tran_low_t *coeff, intptr_t count, int skip_block,
+                    const int16_t *zbin, const int16_t *round,
+                    const int16_t *quant, const int16_t *quant_shift,
+                    tran_low_t *qcoeff, tran_low_t *dqcoeff,
+                    const int16_t *dequant, uint16_t *eob, const int16_t *scan,
+                    const int16_t *iscan) {
+  (void)zbin;
+  (void)quant_shift;
+
+  fn(coeff, count, skip_block, round, quant, qcoeff, dqcoeff, dequant, eob,
+     scan, iscan);
+}
+
+class VP9QuantizeBase {
 public:
-  virtual ~VP9QuantizeTest() {}
-  virtual void SetUp() {
-    quantize_op_ = GET_PARAM(0);
-    ref_quantize_op_ = GET_PARAM(1);
-    bit_depth_ = GET_PARAM(2);
-    mask_ = (1 << bit_depth_) - 1;
+  VP9QuantizeBase(vpx_bit_depth_t bit_depth, int max_size, bool is_fp)
+      : bit_depth_(bit_depth), max_size_(max_size), is_fp_(is_fp) {
+    max_value_ = (1 << bit_depth_) - 1;
+    zbin_ptr_ =
+        reinterpret_cast<int16_t *>(vpx_memalign(16, 8 * sizeof(*zbin_ptr_)));
+    round_fp_ptr_ = reinterpret_cast<int16_t *>(
+        vpx_memalign(16, 8 * sizeof(*round_fp_ptr_)));
+    quant_fp_ptr_ = reinterpret_cast<int16_t *>(
+        vpx_memalign(16, 8 * sizeof(*quant_fp_ptr_)));
+    round_ptr_ =
+        reinterpret_cast<int16_t *>(vpx_memalign(16, 8 * sizeof(*round_ptr_)));
+    quant_ptr_ =
+        reinterpret_cast<int16_t *>(vpx_memalign(16, 8 * sizeof(*quant_ptr_)));
+    quant_shift_ptr_ = reinterpret_cast<int16_t *>(
+        vpx_memalign(16, 8 * sizeof(*quant_shift_ptr_)));
+    dequant_ptr_ = reinterpret_cast<int16_t *>(
+        vpx_memalign(16, 8 * sizeof(*dequant_ptr_)));
  }

-  virtual void TearDown() { libvpx_test::ClearSystemState(); }
-
- protected:
-  vpx_bit_depth_t bit_depth_;
-  int mask_;
-  QuantizeFunc quantize_op_;
-  QuantizeFunc ref_quantize_op_;
-};
-
-class VP9Quantize32Test : public ::testing::TestWithParam<QuantizeParam> {
- public:
-  virtual ~VP9Quantize32Test() {}
-  virtual void SetUp() {
-    quantize_op_ = GET_PARAM(0);
-    ref_quantize_op_ = GET_PARAM(1);
-    bit_depth_ = GET_PARAM(2);
-    mask_ = (1 << bit_depth_) - 1;
+  ~VP9QuantizeBase() {
+    vpx_free(zbin_ptr_);
+    vpx_free(round_fp_ptr_);
+    vpx_free(quant_fp_ptr_);
+    vpx_free(round_ptr_);
+    vpx_free(quant_ptr_);
+    vpx_free(quant_shift_ptr_);
+    vpx_free(dequant_ptr_);
+    zbin_ptr_ = NULL;
+    round_fp_ptr_ = NULL;
+    quant_fp_ptr_ = NULL;
+    round_ptr_ = NULL;
+    quant_ptr_ = NULL;
+    quant_shift_ptr_ = NULL;
+    dequant_ptr_ = NULL;
+    libvpx_test::ClearSystemState();
  }

-  virtual void TearDown() { libvpx_test::ClearSystemState(); }
+ protected:
+  int16_t *zbin_ptr_;
+  int16_t *round_fp_ptr_;
+  int16_t *quant_fp_ptr_;
+  int16_t *round_ptr_;
+  int16_t *quant_ptr_;
+  int16_t *quant_shift_ptr_;
+  int16_t *dequant_ptr_;
+  const vpx_bit_depth_t bit_depth_;
+  int max_value_;
+  const int max_size_;
+  const bool is_fp_;
+};
+
+class VP9QuantizeTest : public VP9QuantizeBase,
+                        public ::testing::TestWithParam<QuantizeParam> {
+ public:
+  VP9QuantizeTest()
+      : VP9QuantizeBase(GET_PARAM(2), GET_PARAM(3), GET_PARAM(4)),
+        quantize_op_(GET_PARAM(0)), ref_quantize_op_(GET_PARAM(1)) {}

 protected:
-  vpx_bit_depth_t bit_depth_;
-  int mask_;
-  QuantizeFunc quantize_op_;
-  QuantizeFunc ref_quantize_op_;
+  const QuantizeFunc quantize_op_;
+  const QuantizeFunc ref_quantize_op_;
 };

+// This quantizer compares the AC coefficients to the quantization step size to
+// determine if further multiplication operations are needed.
+// Based on vp9_quantize_fp_sse2().
+void quantize_fp_nz_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
+                      int skip_block, const int16_t *round_ptr,
+                      const int16_t *quant_ptr, tran_low_t *qcoeff_ptr,
+                      tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr,
+                      uint16_t *eob_ptr, const int16_t *scan,
+                      const int16_t *iscan) {
+  int i, eob = -1;
+  const int thr = dequant_ptr[1] >> 1;
+  (void)iscan;
+  (void)skip_block;
+  assert(!skip_block);
+
+  // Quantization pass: All coefficients with index >= zero_flag are
+  // skippable. Note: zero_flag can be zero.
+  for (i = 0; i < n_coeffs; i += 16) {
+    int y;
+    int nzflag_cnt = 0;
+    int abs_coeff[16];
+    int coeff_sign[16];
+
+    // count nzflag for each row (16 tran_low_t)
+    for (y = 0; y < 16; ++y) {
+      const int rc = i + y;
+      const int coeff = coeff_ptr[rc];
+      coeff_sign[y] = (coeff >> 31);
+      abs_coeff[y] = (coeff ^ coeff_sign[y]) - coeff_sign[y];
+      // The first 16 are skipped in the sse2 code.  Do the same here to match.
+      if (i >= 16 && (abs_coeff[y] <= thr)) {
+        nzflag_cnt++;
+      }
+    }
+
+    for (y = 0; y < 16; ++y) {
+      const int rc = i + y;
+      // If all of the AC coeffs in a row has magnitude less than the
+      // quantization step_size/2, quantize to zero.
+      if (nzflag_cnt < 16) {
+        int tmp =
+            clamp(abs_coeff[y] + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
+        tmp = (tmp * quant_ptr[rc != 0]) >> 16;
+        qcoeff_ptr[rc] = (tmp ^ coeff_sign[y]) - coeff_sign[y];
+        dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
+      } else {
+        qcoeff_ptr[rc] = 0;
+        dqcoeff_ptr[rc] = 0;
+      }
+    }
+  }
+
+  // Scan for eob.
+  for (i = 0; i < n_coeffs; i++) {
+    // Use the scan order to find the correct eob.
+    const int rc = scan[i];
+    if (qcoeff_ptr[rc]) {
+      eob = i;
+    }
+  }
+  *eob_ptr = eob + 1;
+}
+
+void GenerateHelperArrays(ACMRandom *rnd, int16_t *zbin, int16_t *round,
+                          int16_t *quant, int16_t *quant_shift,
+                          int16_t *dequant, int16_t *round_fp,
+                          int16_t *quant_fp) {
+  // Max when q == 0.  Otherwise, it is 48 for Y and 42 for U/V.
+  const int max_qrounding_factor_fp = 64;
+
+  for (int j = 0; j < 2; j++) {
+    // The range is 4 to 1828 in the VP9 tables.
+    const int qlookup = rnd->RandRange(1825) + 4;
+    round_fp[j] = (max_qrounding_factor_fp * qlookup) >> 7;
+    quant_fp[j] = (1 << 16) / qlookup;
+
+    // Values determined by deconstructing vp9_init_quantizer().
+    // zbin may be up to 1143 for 8 and 10 bit Y values, or 1200 for 12 bit Y
+    // values or U/V values of any bit depth. This is because y_delta is not
+    // factored into the vp9_ac_quant() call.
+    zbin[j] = rnd->RandRange(1200);
+
+    // round may be up to 685 for Y values or 914 for U/V.
+    round[j] = rnd->RandRange(914);
+    // quant ranges from 1 to -32703
+    quant[j] = static_cast<int>(rnd->RandRange(32704)) - 32703;
+    // quant_shift goes up to 1 << 16.
+    quant_shift[j] = rnd->RandRange(16384);
+    // dequant maxes out at 1828 for all cases.
+    dequant[j] = rnd->RandRange(1828);
+  }
+  for (int j = 2; j < 8; j++) {
+    zbin[j] = zbin[1];
+    round_fp[j] = round_fp[1];
+    quant_fp[j] = quant_fp[1];
+    round[j] = round[1];
+    quant[j] = quant[1];
+    quant_shift[j] = quant_shift[1];
+    dequant[j] = dequant[1];
+  }
+}
+
 TEST_P(VP9QuantizeTest, OperationCheck) {
  ACMRandom rnd(ACMRandom::DeterministicSeed());
-  DECLARE_ALIGNED(16, tran_low_t, coeff_ptr[256]);
-  DECLARE_ALIGNED(16, int16_t, zbin_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, round_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, quant_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, quant_shift_ptr[2]);
-  DECLARE_ALIGNED(16, tran_low_t, qcoeff_ptr[256]);
-  DECLARE_ALIGNED(16, tran_low_t, dqcoeff_ptr[256]);
-  DECLARE_ALIGNED(16, tran_low_t, ref_qcoeff_ptr[256]);
-  DECLARE_ALIGNED(16, tran_low_t, ref_dqcoeff_ptr[256]);
-  DECLARE_ALIGNED(16, int16_t, dequant_ptr[2]);
-  DECLARE_ALIGNED(16, uint16_t, eob_ptr[1]);
-  DECLARE_ALIGNED(16, uint16_t, ref_eob_ptr[1]);
-  int err_count_total = 0;
-  int first_failure = -1;
-  for (int i = 0; i < number_of_iterations; ++i) {
-    const int skip_block = i == 0;
-    const TX_SIZE sz = (TX_SIZE)(i % 3);  // TX_4X4, TX_8X8 TX_16X16
-    const TX_TYPE tx_type = (TX_TYPE)((i >> 2) % 3);
-    const scan_order *scan_order = &vp9_scan_orders[sz][tx_type];
-    const int count = (4 << sz) * (4 << sz);  // 16, 64, 256
-    int err_count = 0;
-    *eob_ptr = rnd.Rand16();
-    *ref_eob_ptr = *eob_ptr;
-    for (int j = 0; j < count; j++) {
-      coeff_ptr[j] = rnd.Rand16() & mask_;
-    }
-    for (int j = 0; j < 2; j++) {
-      zbin_ptr[j] = rnd.Rand16() & mask_;
-      round_ptr[j] = rnd.Rand16();
-      quant_ptr[j] = rnd.Rand16();
-      quant_shift_ptr[j] = rnd.Rand16();
-      dequant_ptr[j] = rnd.Rand16();
-    }
-    ref_quantize_op_(coeff_ptr, count, skip_block, zbin_ptr, round_ptr,
-                     quant_ptr, quant_shift_ptr, ref_qcoeff_ptr,
-                     ref_dqcoeff_ptr, dequant_ptr, ref_eob_ptr,
-                     scan_order->scan, scan_order->iscan);
-    ASM_REGISTER_STATE_CHECK(quantize_op_(
-        coeff_ptr, count, skip_block, zbin_ptr, round_ptr, quant_ptr,
-        quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr, eob_ptr,
-        scan_order->scan, scan_order->iscan));
-    for (int j = 0; j < sz; ++j) {
-      err_count += (ref_qcoeff_ptr[j] != qcoeff_ptr[j]) |
-                   (ref_dqcoeff_ptr[j] != dqcoeff_ptr[j]);
-    }
-    err_count += (*ref_eob_ptr != *eob_ptr);
-    if (err_count && !err_count_total) {
-      first_failure = i;
-    }
-    err_count_total += err_count;
-  }
-  EXPECT_EQ(0, err_count_total)
-      << "Error: Quantization Test, C output doesn't match SSE2 output. "
-      << "First failed at test case " << first_failure;
-}
+  Buffer<tran_low_t> coeff = Buffer<tran_low_t>(max_size_, max_size_, 0, 16);
+  ASSERT_TRUE(coeff.Init());
+  Buffer<tran_low_t> qcoeff = Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(qcoeff.Init());
+  Buffer<tran_low_t> dqcoeff = Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(dqcoeff.Init());
+  Buffer<tran_low_t> ref_qcoeff =
+      Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(ref_qcoeff.Init());
+  Buffer<tran_low_t> ref_dqcoeff =
+      Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(ref_dqcoeff.Init());
+  uint16_t eob, ref_eob;

-TEST_P(VP9Quantize32Test, OperationCheck) {
-  ACMRandom rnd(ACMRandom::DeterministicSeed());
-  DECLARE_ALIGNED(16, tran_low_t, coeff_ptr[1024]);
-  DECLARE_ALIGNED(16, int16_t, zbin_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, round_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, quant_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, quant_shift_ptr[2]);
-  DECLARE_ALIGNED(16, tran_low_t, qcoeff_ptr[1024]);
-  DECLARE_ALIGNED(16, tran_low_t, dqcoeff_ptr[1024]);
-  DECLARE_ALIGNED(16, tran_low_t, ref_qcoeff_ptr[1024]);
-  DECLARE_ALIGNED(16, tran_low_t, ref_dqcoeff_ptr[1024]);
-  DECLARE_ALIGNED(16, int16_t, dequant_ptr[2]);
-  DECLARE_ALIGNED(16, uint16_t, eob_ptr[1]);
-  DECLARE_ALIGNED(16, uint16_t, ref_eob_ptr[1]);
-  int err_count_total = 0;
-  int first_failure = -1;
  for (int i = 0; i < number_of_iterations; ++i) {
-    const int skip_block = i == 0;
-    const TX_SIZE sz = TX_32X32;
-    const TX_TYPE tx_type = (TX_TYPE)(i % 4);
+    // Test skip block for the first three iterations to catch all the different
+    // sizes.
+    const int skip_block = 0;
+    TX_SIZE sz;
+    if (max_size_ == 16) {
+      sz = static_cast<TX_SIZE>(i % 3);  // TX_4X4, TX_8X8 TX_16X16
+    } else {
+      sz = TX_32X32;
+    }
+    const TX_TYPE tx_type = static_cast<TX_TYPE>((i >> 2) % 3);
    const scan_order *scan_order = &vp9_scan_orders[sz][tx_type];
-    const int count = (4 << sz) * (4 << sz);  // 1024
-    int err_count = 0;
-    *eob_ptr = rnd.Rand16();
-    *ref_eob_ptr = *eob_ptr;
-    for (int j = 0; j < count; j++) {
-      coeff_ptr[j] = rnd.Rand16() & mask_;
-    }
-    for (int j = 0; j < 2; j++) {
-      zbin_ptr[j] = rnd.Rand16() & mask_;
-      round_ptr[j] = rnd.Rand16();
-      quant_ptr[j] = rnd.Rand16();
-      quant_shift_ptr[j] = rnd.Rand16();
-      dequant_ptr[j] = rnd.Rand16();
-    }
-    ref_quantize_op_(coeff_ptr, count, skip_block, zbin_ptr, round_ptr,
-                     quant_ptr, quant_shift_ptr, ref_qcoeff_ptr,
-                     ref_dqcoeff_ptr, dequant_ptr, ref_eob_ptr,
+    const int count = (4 << sz) * (4 << sz);
+    coeff.Set(&rnd, -max_value_, max_value_);
+    GenerateHelperArrays(&rnd, zbin_ptr_, round_ptr_, quant_ptr_,
+                         quant_shift_ptr_, dequant_ptr_, round_fp_ptr_,
+                         quant_fp_ptr_);
+    int16_t *r_ptr = (is_fp_) ? round_fp_ptr_ : round_ptr_;
+    int16_t *q_ptr = (is_fp_) ? quant_fp_ptr_ : quant_ptr_;
+    ref_quantize_op_(coeff.TopLeftPixel(), count, skip_block, zbin_ptr_, r_ptr,
+                     q_ptr, quant_shift_ptr_, ref_qcoeff.TopLeftPixel(),
+                     ref_dqcoeff.TopLeftPixel(), dequant_ptr_, &ref_eob,
                     scan_order->scan, scan_order->iscan);
+
    ASM_REGISTER_STATE_CHECK(quantize_op_(
-        coeff_ptr, count, skip_block, zbin_ptr, round_ptr, quant_ptr,
-        quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr, eob_ptr,
-        scan_order->scan, scan_order->iscan));
-    for (int j = 0; j < sz; ++j) {
-      err_count += (ref_qcoeff_ptr[j] != qcoeff_ptr[j]) |
-                   (ref_dqcoeff_ptr[j] != dqcoeff_ptr[j]);
+        coeff.TopLeftPixel(), count, skip_block, zbin_ptr_, r_ptr, q_ptr,
+        quant_shift_ptr_, qcoeff.TopLeftPixel(), dqcoeff.TopLeftPixel(),
+        dequant_ptr_, &eob, scan_order->scan, scan_order->iscan));
+
+    EXPECT_TRUE(qcoeff.CheckValues(ref_qcoeff));
+    EXPECT_TRUE(dqcoeff.CheckValues(ref_dqcoeff));
+
+    EXPECT_EQ(eob, ref_eob);
+
+    if (HasFailure()) {
+      printf("Failure on iteration %d.\n", i);
+      qcoeff.PrintDifference(ref_qcoeff);
+      dqcoeff.PrintDifference(ref_dqcoeff);
+      return;
    }
-    err_count += (*ref_eob_ptr != *eob_ptr);
-    if (err_count && !err_count_total) {
-      first_failure = i;
-    }
-    err_count_total += err_count;
  }
-  EXPECT_EQ(0, err_count_total)
-      << "Error: Quantization Test, C output doesn't match SSE2 output. "
-      << "First failed at test case " << first_failure;
 }

 TEST_P(VP9QuantizeTest, EOBCheck) {
  ACMRandom rnd(ACMRandom::DeterministicSeed());
-  DECLARE_ALIGNED(16, tran_low_t, coeff_ptr[256]);
-  DECLARE_ALIGNED(16, int16_t, zbin_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, round_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, quant_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, quant_shift_ptr[2]);
-  DECLARE_ALIGNED(16, tran_low_t, qcoeff_ptr[256]);
-  DECLARE_ALIGNED(16, tran_low_t, dqcoeff_ptr[256]);
-  DECLARE_ALIGNED(16, tran_low_t, ref_qcoeff_ptr[256]);
-  DECLARE_ALIGNED(16, tran_low_t, ref_dqcoeff_ptr[256]);
-  DECLARE_ALIGNED(16, int16_t, dequant_ptr[2]);
-  DECLARE_ALIGNED(16, uint16_t, eob_ptr[1]);
-  DECLARE_ALIGNED(16, uint16_t, ref_eob_ptr[1]);
-  int err_count_total = 0;
-  int first_failure = -1;
+  Buffer<tran_low_t> coeff = Buffer<tran_low_t>(max_size_, max_size_, 0, 16);
+  ASSERT_TRUE(coeff.Init());
+  Buffer<tran_low_t> qcoeff = Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(qcoeff.Init());
+  Buffer<tran_low_t> dqcoeff = Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(dqcoeff.Init());
+  Buffer<tran_low_t> ref_qcoeff =
+      Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(ref_qcoeff.Init());
+  Buffer<tran_low_t> ref_dqcoeff =
+      Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(ref_dqcoeff.Init());
+  uint16_t eob, ref_eob;
+
  for (int i = 0; i < number_of_iterations; ++i) {
-    int skip_block = i == 0;
-    TX_SIZE sz = (TX_SIZE)(i % 3);  // TX_4X4, TX_8X8 TX_16X16
-    TX_TYPE tx_type = (TX_TYPE)((i >> 2) % 3);
+    const int skip_block = 0;
+    TX_SIZE sz;
+    if (max_size_ == 16) {
+      sz = static_cast<TX_SIZE>(i % 3);  // TX_4X4, TX_8X8 TX_16X16
+    } else {
+      sz = TX_32X32;
+    }
+    const TX_TYPE tx_type = static_cast<TX_TYPE>((i >> 2) % 3);
    const scan_order *scan_order = &vp9_scan_orders[sz][tx_type];
-    int count = (4 << sz) * (4 << sz);  // 16, 64, 256
-    int err_count = 0;
-    *eob_ptr = rnd.Rand16();
-    *ref_eob_ptr = *eob_ptr;
+    int count = (4 << sz) * (4 << sz);
    // Two random entries
-    for (int j = 0; j < count; j++) {
-      coeff_ptr[j] = 0;
-    }
-    coeff_ptr[rnd(count)] = rnd.Rand16() & mask_;
-    coeff_ptr[rnd(count)] = rnd.Rand16() & mask_;
-    for (int j = 0; j < 2; j++) {
-      zbin_ptr[j] = rnd.Rand16() & mask_;
-      round_ptr[j] = rnd.Rand16();
-      quant_ptr[j] = rnd.Rand16();
-      quant_shift_ptr[j] = rnd.Rand16();
-      dequant_ptr[j] = rnd.Rand16();
-    }
-
-    ref_quantize_op_(coeff_ptr, count, skip_block, zbin_ptr, round_ptr,
-                     quant_ptr, quant_shift_ptr, ref_qcoeff_ptr,
-                     ref_dqcoeff_ptr, dequant_ptr, ref_eob_ptr,
+    coeff.Set(0);
+    coeff.TopLeftPixel()[rnd(count)] =
+        static_cast<int>(rnd.RandRange(max_value_ * 2)) - max_value_;
+    coeff.TopLeftPixel()[rnd(count)] =
+        static_cast<int>(rnd.RandRange(max_value_ * 2)) - max_value_;
+    GenerateHelperArrays(&rnd, zbin_ptr_, round_ptr_, quant_ptr_,
+                         quant_shift_ptr_, dequant_ptr_, round_fp_ptr_,
+                         quant_fp_ptr_);
+    int16_t *r_ptr = (is_fp_) ? round_fp_ptr_ : round_ptr_;
+    int16_t *q_ptr = (is_fp_) ? quant_fp_ptr_ : quant_ptr_;
+    ref_quantize_op_(coeff.TopLeftPixel(), count, skip_block, zbin_ptr_, r_ptr,
+                     q_ptr, quant_shift_ptr_, ref_qcoeff.TopLeftPixel(),
+                     ref_dqcoeff.TopLeftPixel(), dequant_ptr_, &ref_eob,
                     scan_order->scan, scan_order->iscan);
-    ASM_REGISTER_STATE_CHECK(quantize_op_(
-        coeff_ptr, count, skip_block, zbin_ptr, round_ptr, quant_ptr,
-        quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr, eob_ptr,
-        scan_order->scan, scan_order->iscan));

-    for (int j = 0; j < sz; ++j) {
-      err_count += (ref_qcoeff_ptr[j] != qcoeff_ptr[j]) |
-                   (ref_dqcoeff_ptr[j] != dqcoeff_ptr[j]);
+    ASM_REGISTER_STATE_CHECK(quantize_op_(
+        coeff.TopLeftPixel(), count, skip_block, zbin_ptr_, r_ptr, q_ptr,
+        quant_shift_ptr_, qcoeff.TopLeftPixel(), dqcoeff.TopLeftPixel(),
+        dequant_ptr_, &eob, scan_order->scan, scan_order->iscan));
+
+    EXPECT_TRUE(qcoeff.CheckValues(ref_qcoeff));
+    EXPECT_TRUE(dqcoeff.CheckValues(ref_dqcoeff));
+
+    EXPECT_EQ(eob, ref_eob);
+
+    if (HasFailure()) {
+      printf("Failure on iteration %d.\n", i);
+      qcoeff.PrintDifference(ref_qcoeff);
+      dqcoeff.PrintDifference(ref_dqcoeff);
+      return;
    }
-    err_count += (*ref_eob_ptr != *eob_ptr);
-    if (err_count && !err_count_total) {
-      first_failure = i;
-    }
-    err_count_total += err_count;
  }
-  EXPECT_EQ(0, err_count_total)
-      << "Error: Quantization Test, C output doesn't match SSE2 output. "
-      << "First failed at test case " << first_failure;
 }

-TEST_P(VP9Quantize32Test, EOBCheck) {
+TEST_P(VP9QuantizeTest, DISABLED_Speed) {
  ACMRandom rnd(ACMRandom::DeterministicSeed());
-  DECLARE_ALIGNED(16, tran_low_t, coeff_ptr[1024]);
-  DECLARE_ALIGNED(16, int16_t, zbin_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, round_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, quant_ptr[2]);
-  DECLARE_ALIGNED(16, int16_t, quant_shift_ptr[2]);
-  DECLARE_ALIGNED(16, tran_low_t, qcoeff_ptr[1024]);
-  DECLARE_ALIGNED(16, tran_low_t, dqcoeff_ptr[1024]);
-  DECLARE_ALIGNED(16, tran_low_t, ref_qcoeff_ptr[1024]);
-  DECLARE_ALIGNED(16, tran_low_t, ref_dqcoeff_ptr[1024]);
-  DECLARE_ALIGNED(16, int16_t, dequant_ptr[2]);
-  DECLARE_ALIGNED(16, uint16_t, eob_ptr[1]);
-  DECLARE_ALIGNED(16, uint16_t, ref_eob_ptr[1]);
-  int err_count_total = 0;
-  int first_failure = -1;
-  for (int i = 0; i < number_of_iterations; ++i) {
-    int skip_block = i == 0;
-    TX_SIZE sz = TX_32X32;
-    TX_TYPE tx_type = (TX_TYPE)(i % 4);
-    const scan_order *scan_order = &vp9_scan_orders[sz][tx_type];
-    int count = (4 << sz) * (4 << sz);  // 1024
-    int err_count = 0;
-    *eob_ptr = rnd.Rand16();
-    *ref_eob_ptr = *eob_ptr;
-    for (int j = 0; j < count; j++) {
-      coeff_ptr[j] = 0;
-    }
-    // Two random entries
-    coeff_ptr[rnd(count)] = rnd.Rand16() & mask_;
-    coeff_ptr[rnd(count)] = rnd.Rand16() & mask_;
-    for (int j = 0; j < 2; j++) {
-      zbin_ptr[j] = rnd.Rand16() & mask_;
-      round_ptr[j] = rnd.Rand16();
-      quant_ptr[j] = rnd.Rand16();
-      quant_shift_ptr[j] = rnd.Rand16();
-      dequant_ptr[j] = rnd.Rand16();
-    }
+  Buffer<tran_low_t> coeff = Buffer<tran_low_t>(max_size_, max_size_, 0, 16);
+  ASSERT_TRUE(coeff.Init());
+  Buffer<tran_low_t> qcoeff = Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(qcoeff.Init());
+  Buffer<tran_low_t> dqcoeff = Buffer<tran_low_t>(max_size_, max_size_, 0, 32);
+  ASSERT_TRUE(dqcoeff.Init());
+  uint16_t eob;
+  TX_SIZE starting_sz, ending_sz;

-    ref_quantize_op_(coeff_ptr, count, skip_block, zbin_ptr, round_ptr,
-                     quant_ptr, quant_shift_ptr, ref_qcoeff_ptr,
-                     ref_dqcoeff_ptr, dequant_ptr, ref_eob_ptr,
-                     scan_order->scan, scan_order->iscan);
-    ASM_REGISTER_STATE_CHECK(quantize_op_(
-        coeff_ptr, count, skip_block, zbin_ptr, round_ptr, quant_ptr,
-        quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr, eob_ptr,
-        scan_order->scan, scan_order->iscan));
-
-    for (int j = 0; j < sz; ++j) {
-      err_count += (ref_qcoeff_ptr[j] != qcoeff_ptr[j]) |
-                   (ref_dqcoeff_ptr[j] != dqcoeff_ptr[j]);
-    }
-    err_count += (*ref_eob_ptr != *eob_ptr);
-    if (err_count && !err_count_total) {
-      first_failure = i;
-    }
-    err_count_total += err_count;
+  if (max_size_ == 16) {
+    starting_sz = TX_4X4;
+    ending_sz = TX_16X16;
+  } else {
+    starting_sz = TX_32X32;
+    ending_sz = TX_32X32;
+  }
+
+  for (TX_SIZE sz = starting_sz; sz <= ending_sz; ++sz) {
+    // zbin > coeff, zbin < coeff.
+    for (int i = 0; i < 2; ++i) {
+      const int skip_block = 0;
+      // TX_TYPE defines the scan order. That is not relevant to the speed test.
+      // Pick the first one.
+      const TX_TYPE tx_type = DCT_DCT;
+      const scan_order *scan_order = &vp9_scan_orders[sz][tx_type];
+      const int count = (4 << sz) * (4 << sz);
+
+      GenerateHelperArrays(&rnd, zbin_ptr_, round_ptr_, quant_ptr_,
+                           quant_shift_ptr_, dequant_ptr_, round_fp_ptr_,
+                           quant_fp_ptr_);
+      int16_t *r_ptr = (is_fp_) ? round_fp_ptr_ : round_ptr_;
+      int16_t *q_ptr = (is_fp_) ? quant_fp_ptr_ : quant_ptr_;
+
+      if (i == 0) {
+        // When |coeff values| are less than zbin the results are 0.
+        int threshold = 100;
+        if (max_size_ == 32) {
+          // For 32x32, the threshold is halved. Double it to keep the values
+          // from clearing it.
+          threshold = 200;
+        }
+        for (int j = 0; j < 8; ++j) zbin_ptr_[j] = threshold;
+        coeff.Set(&rnd, -99, 99);
+      } else if (i == 1) {
+        for (int j = 0; j < 8; ++j) zbin_ptr_[j] = 50;
+        coeff.Set(&rnd, -500, 500);
+      }
+
+      vpx_usec_timer timer;
+      vpx_usec_timer_start(&timer);
+      for (int j = 0; j < 100000000 / count; ++j) {
+        quantize_op_(coeff.TopLeftPixel(), count, skip_block, zbin_ptr_, r_ptr,
+                     q_ptr, quant_shift_ptr_, qcoeff.TopLeftPixel(),
+                     dqcoeff.TopLeftPixel(), dequant_ptr_, &eob,
+                     scan_order->scan, scan_order->iscan);
+      }
+      vpx_usec_timer_mark(&timer);
+      const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
+      if (i == 0) printf("Bypass calculations.\n");
+      if (i == 1) printf("Full calculations.\n");
+      printf("Quantize %dx%d time: %5d ms\n", 4 << sz, 4 << sz,
+             elapsed_time / 1000);
+    }
+    printf("\n");
  }
-  EXPECT_EQ(0, err_count_total)
-      << "Error: Quantization Test, C output doesn't match SSE2 output. "
-      << "First failed at test case " << first_failure;
 }
+
 using std::tr1::make_tuple;

 #if HAVE_SSE2
+#if CONFIG_VP9_HIGHBITDEPTH
+// TODO(johannkoenig): Fix vpx_quantize_b_sse2 in highbitdepth builds.
+// make_tuple(&vpx_quantize_b_sse2, &vpx_highbd_quantize_b_c, VPX_BITS_8),
 INSTANTIATE_TEST_CASE_P(
    SSE2, VP9QuantizeTest,
-    ::testing::Values(make_tuple(&vpx_highbd_quantize_b_sse2,
-                                 &vpx_highbd_quantize_b_c, VPX_BITS_8),
-                      make_tuple(&vpx_highbd_quantize_b_sse2,
-                                 &vpx_highbd_quantize_b_c, VPX_BITS_10),
-                      make_tuple(&vpx_highbd_quantize_b_sse2,
-                                 &vpx_highbd_quantize_b_c, VPX_BITS_12)));
+    ::testing::Values(
+        make_tuple(&vpx_highbd_quantize_b_sse2, &vpx_highbd_quantize_b_c,
+                   VPX_BITS_8, 16, false),
+        make_tuple(&vpx_highbd_quantize_b_sse2, &vpx_highbd_quantize_b_c,
+                   VPX_BITS_10, 16, false),
+        make_tuple(&vpx_highbd_quantize_b_sse2, &vpx_highbd_quantize_b_c,
+                   VPX_BITS_12, 16, false),
+        make_tuple(&vpx_highbd_quantize_b_32x32_sse2,
+                   &vpx_highbd_quantize_b_32x32_c, VPX_BITS_8, 32, false),
+        make_tuple(&vpx_highbd_quantize_b_32x32_sse2,
+                   &vpx_highbd_quantize_b_32x32_c, VPX_BITS_10, 32, false),
+        make_tuple(&vpx_highbd_quantize_b_32x32_sse2,
+                   &vpx_highbd_quantize_b_32x32_c, VPX_BITS_12, 32, false)));
+
+#else
 INSTANTIATE_TEST_CASE_P(
-    SSE2, VP9Quantize32Test,
-    ::testing::Values(make_tuple(&vpx_highbd_quantize_b_32x32_sse2,
-                                 &vpx_highbd_quantize_b_32x32_c, VPX_BITS_8),
-                      make_tuple(&vpx_highbd_quantize_b_32x32_sse2,
-                                 &vpx_highbd_quantize_b_32x32_c, VPX_BITS_10),
-                      make_tuple(&vpx_highbd_quantize_b_32x32_sse2,
-                                 &vpx_highbd_quantize_b_32x32_c, VPX_BITS_12)));
-#endif  // HAVE_SSE2
+    SSE2, VP9QuantizeTest,
+    ::testing::Values(make_tuple(&vpx_quantize_b_sse2, &vpx_quantize_b_c,
+                                 VPX_BITS_8, 16, false),
+                      make_tuple(&QuantFPWrapper<vp9_quantize_fp_sse2>,
+                                 &QuantFPWrapper<quantize_fp_nz_c>, VPX_BITS_8,
+                                 16, true)));
 #endif  // CONFIG_VP9_HIGHBITDEPTH
+#endif  // HAVE_SSE2
+
+#if HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, VP9QuantizeTest,
+    ::testing::Values(make_tuple(&vpx_quantize_b_ssse3, &vpx_quantize_b_c,
+                                 VPX_BITS_8, 16, false),
+                      make_tuple(&QuantFPWrapper<vp9_quantize_fp_ssse3>,
+                                 &QuantFPWrapper<quantize_fp_nz_c>, VPX_BITS_8,
+                                 16, true)));
+#else
+INSTANTIATE_TEST_CASE_P(SSSE3, VP9QuantizeTest,
+                        ::testing::Values(make_tuple(&vpx_quantize_b_ssse3,
+                                                     &vpx_quantize_b_c,
+                                                     VPX_BITS_8, 16, false)));
+#endif
+
+#if ARCH_X86_64
+// TODO(johannkoenig): SSSE3 optimizations do not yet pass this test.
+INSTANTIATE_TEST_CASE_P(
+    DISABLED_SSSE3, VP9QuantizeTest,
+    ::testing::Values(make_tuple(&vpx_quantize_b_32x32_ssse3,
+                                 &vpx_quantize_b_32x32_c, VPX_BITS_8, 32,
+                                 false),
+                      make_tuple(&QuantFPWrapper<vp9_quantize_fp_32x32_ssse3>,
+                                 &QuantFPWrapper<vp9_quantize_fp_32x32_c>,
+                                 VPX_BITS_8, 32, true)));
+#endif  // ARCH_X86_64
+#endif  // HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH
+
+// TODO(johannkoenig): AVX optimizations do not yet pass the 32x32 test or
+// highbitdepth configurations.
+#if HAVE_AVX && !CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    AVX, VP9QuantizeTest,
+    ::testing::Values(make_tuple(&vpx_quantize_b_avx, &vpx_quantize_b_c,
+                                 VPX_BITS_8, 16, false),
+                      // Even though SSSE3 and AVX do not match the reference
+                      // code, we can keep them in sync with each other.
+                      make_tuple(&vpx_quantize_b_32x32_avx,
+                                 &vpx_quantize_b_32x32_ssse3, VPX_BITS_8, 32,
+                                 false)));
+#endif  // HAVE_AVX && !CONFIG_VP9_HIGHBITDEPTH
+
+// TODO(webm:1448): dqcoeff is not handled correctly in HBD builds.
+#if HAVE_NEON && !CONFIG_VP9_HIGHBITDEPTH
+INSTANTIATE_TEST_CASE_P(
+    NEON, VP9QuantizeTest,
+    ::testing::Values(make_tuple(&vpx_quantize_b_neon, &vpx_quantize_b_c,
+                                 VPX_BITS_8, 16, false),
+                      make_tuple(&vpx_quantize_b_32x32_neon,
+                                 &vpx_quantize_b_32x32_c, VPX_BITS_8, 32,
+                                 false),
+                      make_tuple(&QuantFPWrapper<vp9_quantize_fp_neon>,
+                                 &QuantFPWrapper<vp9_quantize_fp_c>, VPX_BITS_8,
+                                 16, true),
+                      make_tuple(&QuantFPWrapper<vp9_quantize_fp_32x32_neon>,
+                                 &QuantFPWrapper<vp9_quantize_fp_32x32_c>,
+                                 VPX_BITS_8, 32, true)));
+#endif  // HAVE_NEON && !CONFIG_VP9_HIGHBITDEPTH
+
+// Only useful to compare "Speed" test results.
+INSTANTIATE_TEST_CASE_P(
+    DISABLED_C, VP9QuantizeTest,
+    ::testing::Values(
+        make_tuple(&vpx_quantize_b_c, &vpx_quantize_b_c, VPX_BITS_8, 16, false),
+        make_tuple(&vpx_quantize_b_32x32_c, &vpx_quantize_b_32x32_c, VPX_BITS_8,
+                   32, false),
+        make_tuple(&QuantFPWrapper<vp9_quantize_fp_c>,
+                   &QuantFPWrapper<vp9_quantize_fp_c>, VPX_BITS_8, 16, true),
+        make_tuple(&QuantFPWrapper<quantize_fp_nz_c>,
+                   &QuantFPWrapper<quantize_fp_nz_c>, VPX_BITS_8, 16, true),
+        make_tuple(&QuantFPWrapper<vp9_quantize_fp_32x32_c>,
+                   &QuantFPWrapper<vp9_quantize_fp_32x32_c>, VPX_BITS_8, 32,
+                   true)));
 }  // namespace
--- a/test/vp9_scale_test.cc
+++ b/test/vp9_scale_test.cc
@@ -0,0 +1,214 @@
+/*
+ *  Copyright (c) 2017 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vp9_rtcd.h"
+#include "./vpx_config.h"
+#include "./vpx_scale_rtcd.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/vpx_scale_test.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_ports/vpx_timer.h"
+#include "vpx_scale/yv12config.h"
+
+namespace libvpx_test {
+
+typedef void (*ScaleFrameFunc)(const YV12_BUFFER_CONFIG *src,
+                               YV12_BUFFER_CONFIG *dst,
+                               INTERP_FILTER filter_type, int phase_scaler);
+
+class ScaleTest : public VpxScaleBase,
+                  public ::testing::TestWithParam<ScaleFrameFunc> {
+ public:
+  virtual ~ScaleTest() {}
+
+ protected:
+  virtual void SetUp() { scale_fn_ = GetParam(); }
+
+  void ReferenceScaleFrame(INTERP_FILTER filter_type, int phase_scaler) {
+    vp9_scale_and_extend_frame_c(&img_, &ref_img_, filter_type, phase_scaler);
+  }
+
+  void ScaleFrame(INTERP_FILTER filter_type, int phase_scaler) {
+    ASM_REGISTER_STATE_CHECK(
+        scale_fn_(&img_, &dst_img_, filter_type, phase_scaler));
+  }
+
+  void RunTest() {
+    static const int kNumSizesToTest = 20;
+    static const int kNumScaleFactorsToTest = 4;
+    static const int kSizesToTest[] = {
+      2,  4,  6,  8,  10, 12, 14, 16, 18,  20,
+      22, 24, 26, 28, 30, 32, 34, 68, 128, 134
+    };
+    static const int kScaleFactors[] = { 1, 2, 3, 4 };
+    for (INTERP_FILTER filter_type = 0; filter_type < 4; ++filter_type) {
+      for (int phase_scaler = 0; phase_scaler < 16; ++phase_scaler) {
+        for (int h = 0; h < kNumSizesToTest; ++h) {
+          const int src_height = kSizesToTest[h];
+          for (int w = 0; w < kNumSizesToTest; ++w) {
+            const int src_width = kSizesToTest[w];
+            for (int sf_up_idx = 0; sf_up_idx < kNumScaleFactorsToTest;
+                 ++sf_up_idx) {
+              const int sf_up = kScaleFactors[sf_up_idx];
+              for (int sf_down_idx = 0; sf_down_idx < kNumScaleFactorsToTest;
+                   ++sf_down_idx) {
+                const int sf_down = kScaleFactors[sf_down_idx];
+                const int dst_width = src_width * sf_up / sf_down;
+                const int dst_height = src_height * sf_up / sf_down;
+                if (sf_up == sf_down && sf_up != 1) {
+                  continue;
+                }
+                // I420 frame width and height must be even.
+                if (!dst_width || !dst_height || dst_width & 1 ||
+                    dst_height & 1) {
+                  continue;
+                }
+                // vpx_convolve8_c() has restriction on the step which cannot
+                // exceed 64 (ratio 1 to 4).
+                if (src_width > 4 * dst_width || src_height > 4 * dst_height) {
+                  continue;
+                }
+                ASSERT_NO_FATAL_FAILURE(ResetScaleImages(
+                    src_width, src_height, dst_width, dst_height));
+                ReferenceScaleFrame(filter_type, phase_scaler);
+                ScaleFrame(filter_type, phase_scaler);
+                if (memcmp(dst_img_.buffer_alloc, ref_img_.buffer_alloc,
+                           ref_img_.frame_size)) {
+                  printf(
+                      "filter_type = %d, phase_scaler = %d, src_width = %4d, "
+                      "src_height = %4d, dst_width = %4d, dst_height = %4d, "
+                      "scale factor = %d:%d\n",
+                      filter_type, phase_scaler, src_width, src_height,
+                      dst_width, dst_height, sf_down, sf_up);
+                  PrintDiff();
+                }
+                CompareImages(dst_img_);
+                DeallocScaleImages();
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+
+  void PrintDiffComponent(const uint8_t *const ref, const uint8_t *const opt,
+                          const int stride, const int width, const int height,
+                          const int plane_idx) const {
+    for (int y = 0; y < height; y++) {
+      for (int x = 0; x < width; x++) {
+        if (ref[y * stride + x] != opt[y * stride + x]) {
+          printf("Plane %d pixel[%d][%d] diff:%6d (ref),%6d (opt)\n", plane_idx,
+                 y, x, ref[y * stride + x], opt[y * stride + x]);
+          break;
+        }
+      }
+    }
+  }
+
+  void PrintDiff() const {
+    assert(ref_img_.y_stride == dst_img_.y_stride);
+    assert(ref_img_.y_width == dst_img_.y_width);
+    assert(ref_img_.y_height == dst_img_.y_height);
+    assert(ref_img_.uv_stride == dst_img_.uv_stride);
+    assert(ref_img_.uv_width == dst_img_.uv_width);
+    assert(ref_img_.uv_height == dst_img_.uv_height);
+
+    if (memcmp(dst_img_.buffer_alloc, ref_img_.buffer_alloc,
+               ref_img_.frame_size)) {
+      PrintDiffComponent(ref_img_.y_buffer, dst_img_.y_buffer,
+                         ref_img_.y_stride, ref_img_.y_width, ref_img_.y_height,
+                         0);
+      PrintDiffComponent(ref_img_.u_buffer, dst_img_.u_buffer,
+                         ref_img_.uv_stride, ref_img_.uv_width,
+                         ref_img_.uv_height, 1);
+      PrintDiffComponent(ref_img_.v_buffer, dst_img_.v_buffer,
+                         ref_img_.uv_stride, ref_img_.uv_width,
+                         ref_img_.uv_height, 2);
+    }
+  }
+
+  ScaleFrameFunc scale_fn_;
+};
+
+TEST_P(ScaleTest, ScaleFrame) { ASSERT_NO_FATAL_FAILURE(RunTest()); }
+
+TEST_P(ScaleTest, DISABLED_Speed) {
+  static const int kCountSpeedTestBlock = 100;
+  static const int kNumScaleFactorsToTest = 4;
+  static const int kScaleFactors[] = { 1, 2, 3, 4 };
+  const int src_width = 1280;
+  const int src_height = 720;
+  for (INTERP_FILTER filter_type = 2; filter_type < 4; ++filter_type) {
+    for (int phase_scaler = 0; phase_scaler < 2; ++phase_scaler) {
+      for (int sf_up_idx = 0; sf_up_idx < kNumScaleFactorsToTest; ++sf_up_idx) {
+        const int sf_up = kScaleFactors[sf_up_idx];
+        for (int sf_down_idx = 0; sf_down_idx < kNumScaleFactorsToTest;
+             ++sf_down_idx) {
+          const int sf_down = kScaleFactors[sf_down_idx];
+          const int dst_width = src_width * sf_up / sf_down;
+          const int dst_height = src_height * sf_up / sf_down;
+          if (sf_up == sf_down && sf_up != 1) {
+            continue;
+          }
+          // I420 frame width and height must be even.
+          if (dst_width & 1 || dst_height & 1) {
+            continue;
+          }
+          ASSERT_NO_FATAL_FAILURE(
+              ResetScaleImages(src_width, src_height, dst_width, dst_height));
+          ASM_REGISTER_STATE_CHECK(
+              ReferenceScaleFrame(filter_type, phase_scaler));
+
+          vpx_usec_timer timer;
+          vpx_usec_timer_start(&timer);
+          for (int i = 0; i < kCountSpeedTestBlock; ++i) {
+            ScaleFrame(filter_type, phase_scaler);
+          }
+          libvpx_test::ClearSystemState();
+          vpx_usec_timer_mark(&timer);
+          const int elapsed_time =
+              static_cast<int>(vpx_usec_timer_elapsed(&timer) / 1000);
+          CompareImages(dst_img_);
+          DeallocScaleImages();
+
+          printf(
+              "filter_type = %d, phase_scaler = %d, src_width = %4d, "
+              "src_height = %4d, dst_width = %4d, dst_height = %4d, "
+              "scale factor = %d:%d, scale time: %5d ms\n",
+              filter_type, phase_scaler, src_width, src_height, dst_width,
+              dst_height, sf_down, sf_up, elapsed_time);
+        }
+      }
+    }
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(C, ScaleTest,
+                        ::testing::Values(vp9_scale_and_extend_frame_c));
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(SSSE3, ScaleTest,
+                        ::testing::Values(vp9_scale_and_extend_frame_ssse3));
+#endif  // HAVE_SSSE3
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, ScaleTest,
+                        ::testing::Values(vp9_scale_and_extend_frame_neon));
+#endif  // HAVE_NEON
+
+}  // namespace libvpx_test
--- a/test/vp9_skip_loopfilter_test.cc
+++ b/test/vp9_skip_loopfilter_test.cc
@@ -85,8 +85,8 @@ class SkipLoopFilterTest {
  // TODO(fgalligan): Move the MD5 testing code into another class.
  void OpenMd5File(const std::string &md5_file_name) {
    md5_file_ = libvpx_test::OpenTestDataFile(md5_file_name);
-    ASSERT_TRUE(md5_file_ != NULL) << "MD5 file open failed. Filename: "
-                                   << md5_file_name;
+    ASSERT_TRUE(md5_file_ != NULL)
+        << "MD5 file open failed. Filename: " << md5_file_name;
  }

  // Reads the next line of the MD5 file.
--- a/test/vp9_subtract_test.cc
+++ b/test/vp9_subtract_test.cc
@@ -101,4 +101,9 @@ INSTANTIATE_TEST_CASE_P(MSA, VP9SubtractBlockTest,
                        ::testing::Values(vpx_subtract_block_msa));
 #endif

+#if HAVE_MMI
+INSTANTIATE_TEST_CASE_P(MMI, VP9SubtractBlockTest,
+                        ::testing::Values(vpx_subtract_block_mmi));
+#endif
+
 }  // namespace vp9
--- a/test/vp9_thread_test.cc
+++ b/test/vp9_thread_test.cc
@@ -187,8 +187,8 @@ void DecodeFiles(const FileList files[]) {
  for (const FileList *iter = files; iter->name != NULL; ++iter) {
    SCOPED_TRACE(iter->name);
    for (int t = 1; t <= 8; ++t) {
-      EXPECT_EQ(iter->expected_md5, DecodeFile(iter->name, t)) << "threads = "
-                                                               << t;
+      EXPECT_EQ(iter->expected_md5, DecodeFile(iter->name, t))
+          << "threads = " << t;
    }
  }
 }
--- a/test/vpx_scale_test.cc
+++ b/test/vpx_scale_test.cc
@@ -14,149 +14,17 @@
 #include "./vpx_scale_rtcd.h"
 #include "test/clear_system_state.h"
 #include "test/register_state_check.h"
+#include "test/vpx_scale_test.h"
 #include "vpx_mem/vpx_mem.h"
+#include "vpx_ports/vpx_timer.h"
 #include "vpx_scale/yv12config.h"

-namespace {
+namespace libvpx_test {

 typedef void (*ExtendFrameBorderFunc)(YV12_BUFFER_CONFIG *ybf);
 typedef void (*CopyFrameFunc)(const YV12_BUFFER_CONFIG *src_ybf,
                              YV12_BUFFER_CONFIG *dst_ybf);

-class VpxScaleBase {
- public:
-  virtual ~VpxScaleBase() { libvpx_test::ClearSystemState(); }
-
-  void ResetImage(int width, int height) {
-    width_ = width;
-    height_ = height;
-    memset(&img_, 0, sizeof(img_));
-    ASSERT_EQ(0, vp8_yv12_alloc_frame_buffer(&img_, width_, height_,
-                                             VP8BORDERINPIXELS));
-    memset(img_.buffer_alloc, kBufFiller, img_.frame_size);
-    FillPlane(img_.y_buffer, img_.y_crop_width, img_.y_crop_height,
-              img_.y_stride);
-    FillPlane(img_.u_buffer, img_.uv_crop_width, img_.uv_crop_height,
-              img_.uv_stride);
-    FillPlane(img_.v_buffer, img_.uv_crop_width, img_.uv_crop_height,
-              img_.uv_stride);
-
-    memset(&ref_img_, 0, sizeof(ref_img_));
-    ASSERT_EQ(0, vp8_yv12_alloc_frame_buffer(&ref_img_, width_, height_,
-                                             VP8BORDERINPIXELS));
-    memset(ref_img_.buffer_alloc, kBufFiller, ref_img_.frame_size);
-
-    memset(&cpy_img_, 0, sizeof(cpy_img_));
-    ASSERT_EQ(0, vp8_yv12_alloc_frame_buffer(&cpy_img_, width_, height_,
-                                             VP8BORDERINPIXELS));
-    memset(cpy_img_.buffer_alloc, kBufFiller, cpy_img_.frame_size);
-    ReferenceCopyFrame();
-  }
-
-  void DeallocImage() {
-    vp8_yv12_de_alloc_frame_buffer(&img_);
-    vp8_yv12_de_alloc_frame_buffer(&ref_img_);
-    vp8_yv12_de_alloc_frame_buffer(&cpy_img_);
-  }
-
- protected:
-  static const int kBufFiller = 123;
-  static const int kBufMax = kBufFiller - 1;
-
-  static void FillPlane(uint8_t *buf, int width, int height, int stride) {
-    for (int y = 0; y < height; ++y) {
-      for (int x = 0; x < width; ++x) {
-        buf[x + (y * stride)] = (x + (width * y)) % kBufMax;
-      }
-    }
-  }
-
-  static void ExtendPlane(uint8_t *buf, int crop_width, int crop_height,
-                          int width, int height, int stride, int padding) {
-    // Copy the outermost visible pixel to a distance of at least 'padding.'
-    // The buffers are allocated such that there may be excess space outside the
-    // padding. As long as the minimum amount of padding is achieved it is not
-    // necessary to fill this space as well.
-    uint8_t *left = buf - padding;
-    uint8_t *right = buf + crop_width;
-    const int right_extend = padding + (width - crop_width);
-    const int bottom_extend = padding + (height - crop_height);
-
-    // Fill the border pixels from the nearest image pixel.
-    for (int y = 0; y < crop_height; ++y) {
-      memset(left, left[padding], padding);
-      memset(right, right[-1], right_extend);
-      left += stride;
-      right += stride;
-    }
-
-    left = buf - padding;
-    uint8_t *top = left - (stride * padding);
-    // The buffer does not always extend as far as the stride.
-    // Equivalent to padding + width + padding.
-    const int extend_width = padding + crop_width + right_extend;
-
-    // The first row was already extended to the left and right. Copy it up.
-    for (int y = 0; y < padding; ++y) {
-      memcpy(top, left, extend_width);
-      top += stride;
-    }
-
-    uint8_t *bottom = left + (crop_height * stride);
-    for (int y = 0; y < bottom_extend; ++y) {
-      memcpy(bottom, left + (crop_height - 1) * stride, extend_width);
-      bottom += stride;
-    }
-  }
-
-  void ReferenceExtendBorder() {
-    ExtendPlane(ref_img_.y_buffer, ref_img_.y_crop_width,
-                ref_img_.y_crop_height, ref_img_.y_width, ref_img_.y_height,
-                ref_img_.y_stride, ref_img_.border);
-    ExtendPlane(ref_img_.u_buffer, ref_img_.uv_crop_width,
-                ref_img_.uv_crop_height, ref_img_.uv_width, ref_img_.uv_height,
-                ref_img_.uv_stride, ref_img_.border / 2);
-    ExtendPlane(ref_img_.v_buffer, ref_img_.uv_crop_width,
-                ref_img_.uv_crop_height, ref_img_.uv_width, ref_img_.uv_height,
-                ref_img_.uv_stride, ref_img_.border / 2);
-  }
-
-  void ReferenceCopyFrame() {
-    // Copy img_ to ref_img_ and extend frame borders. This will be used for
-    // verifying extend_fn_ as well as copy_frame_fn_.
-    EXPECT_EQ(ref_img_.frame_size, img_.frame_size);
-    for (int y = 0; y < img_.y_crop_height; ++y) {
-      for (int x = 0; x < img_.y_crop_width; ++x) {
-        ref_img_.y_buffer[x + y * ref_img_.y_stride] =
-            img_.y_buffer[x + y * img_.y_stride];
-      }
-    }
-
-    for (int y = 0; y < img_.uv_crop_height; ++y) {
-      for (int x = 0; x < img_.uv_crop_width; ++x) {
-        ref_img_.u_buffer[x + y * ref_img_.uv_stride] =
-            img_.u_buffer[x + y * img_.uv_stride];
-        ref_img_.v_buffer[x + y * ref_img_.uv_stride] =
-            img_.v_buffer[x + y * img_.uv_stride];
-      }
-    }
-
-    ReferenceExtendBorder();
-  }
-
-  void CompareImages(const YV12_BUFFER_CONFIG actual) {
-    EXPECT_EQ(ref_img_.frame_size, actual.frame_size);
-    EXPECT_EQ(0, memcmp(ref_img_.buffer_alloc, actual.buffer_alloc,
-                        ref_img_.frame_size));
-  }
-
-  YV12_BUFFER_CONFIG img_;
-  YV12_BUFFER_CONFIG ref_img_;
-  YV12_BUFFER_CONFIG cpy_img_;
-  int width_;
-  int height_;
-};
-
 class ExtendBorderTest
    : public VpxScaleBase,
      public ::testing::TestWithParam<ExtendFrameBorderFunc> {
@@ -178,11 +46,11 @@ class ExtendBorderTest
    static const int kSizesToTest[] = { 1, 15, 33, 145, 512, 1025, 16383 };
    for (int h = 0; h < kNumSizesToTest; ++h) {
      for (int w = 0; w < kNumSizesToTest; ++w) {
-        ResetImage(kSizesToTest[w], kSizesToTest[h]);
+        ASSERT_NO_FATAL_FAILURE(ResetImages(kSizesToTest[w], kSizesToTest[h]));
+        ReferenceCopyFrame();
        ExtendBorder();
-        ReferenceExtendBorder();
        CompareImages(img_);
-        DeallocImage();
+        DeallocImages();
      }
    }
  }
@@ -204,7 +72,7 @@ class CopyFrameTest : public VpxScaleBase,
  virtual void SetUp() { copy_frame_fn_ = GetParam(); }

  void CopyFrame() {
-    ASM_REGISTER_STATE_CHECK(copy_frame_fn_(&img_, &cpy_img_));
+    ASM_REGISTER_STATE_CHECK(copy_frame_fn_(&img_, &dst_img_));
  }

  void RunTest() {
@@ -217,11 +85,11 @@ class CopyFrameTest : public VpxScaleBase,
    static const int kSizesToTest[] = { 1, 15, 33, 145, 512, 1025, 16383 };
    for (int h = 0; h < kNumSizesToTest; ++h) {
      for (int w = 0; w < kNumSizesToTest; ++w) {
-        ResetImage(kSizesToTest[w], kSizesToTest[h]);
+        ASSERT_NO_FATAL_FAILURE(ResetImages(kSizesToTest[w], kSizesToTest[h]));
        ReferenceCopyFrame();
        CopyFrame();
-        CompareImages(cpy_img_);
-        DeallocImage();
+        CompareImages(dst_img_);
+        DeallocImages();
      }
    }
  }
@@ -233,4 +101,5 @@ TEST_P(CopyFrameTest, CopyFrame) { ASSERT_NO_FATAL_FAILURE(RunTest()); }

 INSTANTIATE_TEST_CASE_P(C, CopyFrameTest,
                        ::testing::Values(vp8_yv12_copy_frame_c));
-}  // namespace
+
+}  // namespace libvpx_test
--- a/test/vpx_scale_test.h
+++ b/test/vpx_scale_test.h
@@ -0,0 +1,200 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef TEST_VPX_SCALE_TEST_H_
+#define TEST_VPX_SCALE_TEST_H_
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vpx_config.h"
+#include "./vpx_scale_rtcd.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_scale/yv12config.h"
+
+using libvpx_test::ACMRandom;
+
+namespace libvpx_test {
+
+class VpxScaleBase {
+ public:
+  virtual ~VpxScaleBase() { libvpx_test::ClearSystemState(); }
+
+  void ResetImage(YV12_BUFFER_CONFIG *const img, const int width,
+                  const int height) {
+    memset(img, 0, sizeof(*img));
+    ASSERT_EQ(
+        0, vp8_yv12_alloc_frame_buffer(img, width, height, VP8BORDERINPIXELS));
+    memset(img->buffer_alloc, kBufFiller, img->frame_size);
+  }
+
+  void ResetImages(const int width, const int height) {
+    ResetImage(&img_, width, height);
+    ResetImage(&ref_img_, width, height);
+    ResetImage(&dst_img_, width, height);
+
+    FillPlane(img_.y_buffer, img_.y_crop_width, img_.y_crop_height,
+              img_.y_stride);
+    FillPlane(img_.u_buffer, img_.uv_crop_width, img_.uv_crop_height,
+              img_.uv_stride);
+    FillPlane(img_.v_buffer, img_.uv_crop_width, img_.uv_crop_height,
+              img_.uv_stride);
+  }
+
+  void ResetScaleImage(YV12_BUFFER_CONFIG *const img, const int width,
+                       const int height) {
+    memset(img, 0, sizeof(*img));
+#if CONFIG_VP9_HIGHBITDEPTH
+    ASSERT_EQ(0, vpx_alloc_frame_buffer(img, width, height, 1, 1, 0,
+                                        VP9_ENC_BORDER_IN_PIXELS, 0));
+#else
+    ASSERT_EQ(0, vpx_alloc_frame_buffer(img, width, height, 1, 1,
+                                        VP9_ENC_BORDER_IN_PIXELS, 0));
+#endif
+    memset(img->buffer_alloc, kBufFiller, img->frame_size);
+  }
+
+  void ResetScaleImages(const int src_width, const int src_height,
+                        const int dst_width, const int dst_height) {
+    ResetScaleImage(&img_, src_width, src_height);
+    ResetScaleImage(&ref_img_, dst_width, dst_height);
+    ResetScaleImage(&dst_img_, dst_width, dst_height);
+    FillPlaneExtreme(img_.y_buffer, img_.y_crop_width, img_.y_crop_height,
+                     img_.y_stride);
+    FillPlaneExtreme(img_.u_buffer, img_.uv_crop_width, img_.uv_crop_height,
+                     img_.uv_stride);
+    FillPlaneExtreme(img_.v_buffer, img_.uv_crop_width, img_.uv_crop_height,
+                     img_.uv_stride);
+  }
+
+  void DeallocImages() {
+    vp8_yv12_de_alloc_frame_buffer(&img_);
+    vp8_yv12_de_alloc_frame_buffer(&ref_img_);
+    vp8_yv12_de_alloc_frame_buffer(&dst_img_);
+  }
+
+  void DeallocScaleImages() {
+    vpx_free_frame_buffer(&img_);
+    vpx_free_frame_buffer(&ref_img_);
+    vpx_free_frame_buffer(&dst_img_);
+  }
+
+ protected:
+  static const int kBufFiller = 123;
+  static const int kBufMax = kBufFiller - 1;
+
+  static void FillPlane(uint8_t *const buf, const int width, const int height,
+                        const int stride) {
+    for (int y = 0; y < height; ++y) {
+      for (int x = 0; x < width; ++x) {
+        buf[x + (y * stride)] = (x + (width * y)) % kBufMax;
+      }
+    }
+  }
+
+  static void FillPlaneExtreme(uint8_t *const buf, const int width,
+                               const int height, const int stride) {
+    ACMRandom rnd;
+    for (int y = 0; y < height; ++y) {
+      for (int x = 0; x < width; ++x) {
+        buf[x + (y * stride)] = rnd.Rand8() % 2 ? 255 : 0;
+      }
+    }
+  }
+
+  static void ExtendPlane(uint8_t *buf, int crop_width, int crop_height,
+                          int width, int height, int stride, int padding) {
+    // Copy the outermost visible pixel to a distance of at least 'padding.'
+    // The buffers are allocated such that there may be excess space outside the
+    // padding. As long as the minimum amount of padding is achieved it is not
+    // necessary to fill this space as well.
+    uint8_t *left = buf - padding;
+    uint8_t *right = buf + crop_width;
+    const int right_extend = padding + (width - crop_width);
+    const int bottom_extend = padding + (height - crop_height);
+
+    // Fill the border pixels from the nearest image pixel.
+    for (int y = 0; y < crop_height; ++y) {
+      memset(left, left[padding], padding);
+      memset(right, right[-1], right_extend);
+      left += stride;
+      right += stride;
+    }
+
+    left = buf - padding;
+    uint8_t *top = left - (stride * padding);
+    // The buffer does not always extend as far as the stride.
+    // Equivalent to padding + width + padding.
+    const int extend_width = padding + crop_width + right_extend;
+
+    // The first row was already extended to the left and right. Copy it up.
+    for (int y = 0; y < padding; ++y) {
+      memcpy(top, left, extend_width);
+      top += stride;
+    }
+
+    uint8_t *bottom = left + (crop_height * stride);
+    for (int y = 0; y < bottom_extend; ++y) {
+      memcpy(bottom, left + (crop_height - 1) * stride, extend_width);
+      bottom += stride;
+    }
+  }
+
+  void ReferenceExtendBorder() {
+    ExtendPlane(ref_img_.y_buffer, ref_img_.y_crop_width,
+                ref_img_.y_crop_height, ref_img_.y_width, ref_img_.y_height,
+                ref_img_.y_stride, ref_img_.border);
+    ExtendPlane(ref_img_.u_buffer, ref_img_.uv_crop_width,
+                ref_img_.uv_crop_height, ref_img_.uv_width, ref_img_.uv_height,
+                ref_img_.uv_stride, ref_img_.border / 2);
+    ExtendPlane(ref_img_.v_buffer, ref_img_.uv_crop_width,
+                ref_img_.uv_crop_height, ref_img_.uv_width, ref_img_.uv_height,
+                ref_img_.uv_stride, ref_img_.border / 2);
+  }
+
+  void ReferenceCopyFrame() {
+    // Copy img_ to ref_img_ and extend frame borders. This will be used for
+    // verifying extend_fn_ as well as copy_frame_fn_.
+    EXPECT_EQ(ref_img_.frame_size, img_.frame_size);
+    for (int y = 0; y < img_.y_crop_height; ++y) {
+      for (int x = 0; x < img_.y_crop_width; ++x) {
+        ref_img_.y_buffer[x + y * ref_img_.y_stride] =
+            img_.y_buffer[x + y * img_.y_stride];
+      }
+    }
+
+    for (int y = 0; y < img_.uv_crop_height; ++y) {
+      for (int x = 0; x < img_.uv_crop_width; ++x) {
+        ref_img_.u_buffer[x + y * ref_img_.uv_stride] =
+            img_.u_buffer[x + y * img_.uv_stride];
+        ref_img_.v_buffer[x + y * ref_img_.uv_stride] =
+            img_.v_buffer[x + y * img_.uv_stride];
+      }
+    }
+
+    ReferenceExtendBorder();
+  }
+
+  void CompareImages(const YV12_BUFFER_CONFIG actual) {
+    EXPECT_EQ(ref_img_.frame_size, actual.frame_size);
+    EXPECT_EQ(0, memcmp(ref_img_.buffer_alloc, actual.buffer_alloc,
+                        ref_img_.frame_size));
+  }
+
+  YV12_BUFFER_CONFIG img_;
+  YV12_BUFFER_CONFIG ref_img_;
+  YV12_BUFFER_CONFIG dst_img_;
+};
+
+}  // namespace libvpx_test
+
+#endif  // TEST_VPX_SCALE_TEST_H_
--- a/test/vpx_temporal_svc_encoder.sh
+++ b/test/vpx_temporal_svc_encoder.sh
@@ -41,6 +41,7 @@ vpx_tsvc_encoder() {
  local speed="6"
  local frame_drop_thresh="30"
  local max_threads="4"
+  local error_resilient="1"

  shift 2

@@ -51,11 +52,19 @@ vpx_tsvc_encoder() {

  # TODO(tomfinegan): Verify file output for all thread runs.
  for threads in $(seq $max_threads); do
-    eval "${VPX_TEST_PREFIX}" "${encoder}" "${YUV_RAW_INPUT}" "${output_file}" \
-        "${codec}" "${YUV_RAW_INPUT_WIDTH}" "${YUV_RAW_INPUT_HEIGHT}" \
-        "${timebase_num}" "${timebase_den}" "${speed}" "${frame_drop_thresh}" \
-        "${threads}" "$@" \
-        ${devnull}
+    if [ "$(vpx_config_option_enabled CONFIG_VP9_HIGHBITDEPTH)" != "yes" ]; then
+      eval "${VPX_TEST_PREFIX}" "${encoder}" "${YUV_RAW_INPUT}" \
+        "${output_file}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
+        "${YUV_RAW_INPUT_HEIGHT}" "${timebase_num}" "${timebase_den}" \
+        "${speed}" "${frame_drop_thresh}" "${error_resilient}" "${threads}" \
+        "$@" ${devnull}
+    else
+      eval "${VPX_TEST_PREFIX}" "${encoder}" "${YUV_RAW_INPUT}" \
+        "${output_file}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
+        "${YUV_RAW_INPUT_HEIGHT}" "${timebase_num}" "${timebase_den}" \
+        "${speed}" "${frame_drop_thresh}" "${error_resilient}" "${threads}" \
+        "$@" "8" ${devnull}
+    fi
  done
 }

@@ -76,193 +85,217 @@ files_exist() {

 vpx_tsvc_encoder_vp8_mode_0() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 0 200 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_0"
+    vpx_tsvc_encoder vp8 "${output_basename}" 0 200 || return 1
    # Mode 0 produces 1 stream
-    files_exist "${FUNCNAME}" 1 || return 1
+    files_exist "${output_basename}" 1 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_1() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 1 200 400 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_1"
+    vpx_tsvc_encoder vp8 "${output_basename}" 1 200 400 || return 1
    # Mode 1 produces 2 streams
-    files_exist "${FUNCNAME}" 2 || return 1
+    files_exist "${output_basename}" 2 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_2() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 2 200 400 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_2"
+    vpx_tsvc_encoder vp8 "${output_basename}" 2 200 400 || return 1
    # Mode 2 produces 2 streams
-    files_exist "${FUNCNAME}" 2 || return 1
+    files_exist "${output_basename}" 2 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_3() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 3 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_3"
+    vpx_tsvc_encoder vp8 "${output_basename}" 3 200 400 600 || return 1
    # Mode 3 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_4() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 4 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_4"
+    vpx_tsvc_encoder vp8 "${output_basename}" 4 200 400 600 || return 1
    # Mode 4 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_5() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 5 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_5"
+    vpx_tsvc_encoder vp8 "${output_basename}" 5 200 400 600 || return 1
    # Mode 5 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_6() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 6 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_6"
+    vpx_tsvc_encoder vp8 "${output_basename}" 6 200 400 600 || return 1
    # Mode 6 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_7() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 7 200 400 600 800 1000 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_7"
+    vpx_tsvc_encoder vp8 "${output_basename}" 7 200 400 600 800 1000 || return 1
    # Mode 7 produces 5 streams
-    files_exist "${FUNCNAME}" 5 || return 1
+    files_exist "${output_basename}" 5 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_8() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 8 200 400 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_8"
+    vpx_tsvc_encoder vp8 "${output_basename}" 8 200 400 || return 1
    # Mode 8 produces 2 streams
-    files_exist "${FUNCNAME}" 2 || return 1
+    files_exist "${output_basename}" 2 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_9() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 9 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_9"
+    vpx_tsvc_encoder vp8 "${output_basename}" 9 200 400 600 || return 1
    # Mode 9 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_10() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 10 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_10"
+    vpx_tsvc_encoder vp8 "${output_basename}" 10 200 400 600 || return 1
    # Mode 10 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp8_mode_11() {
  if [ "$(vp8_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp8 "${FUNCNAME}" 11 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp8_mode_11"
+    vpx_tsvc_encoder vp8 "${output_basename}" 11 200 400 600 || return 1
    # Mode 11 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_0() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 0 200 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_0"
+    vpx_tsvc_encoder vp9 "${output_basename}" 0 200 || return 1
    # Mode 0 produces 1 stream
-    files_exist "${FUNCNAME}" 1 || return 1
+    files_exist "${output_basename}" 1 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_1() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 1 200 400 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_1"
+    vpx_tsvc_encoder vp9 "${output_basename}" 1 200 400 || return 1
    # Mode 1 produces 2 streams
-    files_exist "${FUNCNAME}" 2 || return 1
+    files_exist "${output_basename}" 2 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_2() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 2 200 400 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_2"
+    vpx_tsvc_encoder vp9 "${output_basename}" 2 200 400 || return 1
    # Mode 2 produces 2 streams
-    files_exist "${FUNCNAME}" 2 || return 1
+    files_exist "${output_basename}" 2 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_3() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 3 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_3"
+    vpx_tsvc_encoder vp9 "${output_basename}" 3 200 400 600 || return 1
    # Mode 3 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_4() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 4 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_4"
+    vpx_tsvc_encoder vp9 "${output_basename}" 4 200 400 600 || return 1
    # Mode 4 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_5() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 5 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_5"
+    vpx_tsvc_encoder vp9 "${output_basename}" 5 200 400 600 || return 1
    # Mode 5 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_6() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 6 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_6"
+    vpx_tsvc_encoder vp9 "${output_basename}" 6 200 400 600 || return 1
    # Mode 6 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_7() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 7 200 400 600 800 1000 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_7"
+    vpx_tsvc_encoder vp9 "${output_basename}" 7 200 400 600 800 1000 || return 1
    # Mode 7 produces 5 streams
-    files_exist "${FUNCNAME}" 5 || return 1
+    files_exist "${output_basename}" 5 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_8() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 8 200 400 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_8"
+    vpx_tsvc_encoder vp9 "${output_basename}" 8 200 400 || return 1
    # Mode 8 produces 2 streams
-    files_exist "${FUNCNAME}" 2 || return 1
+    files_exist "${output_basename}" 2 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_9() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 9 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_9"
+    vpx_tsvc_encoder vp9 "${output_basename}" 9 200 400 600 || return 1
    # Mode 9 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_10() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 10 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_10"
+    vpx_tsvc_encoder vp9 "${output_basename}" 10 200 400 600 || return 1
    # Mode 10 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

 vpx_tsvc_encoder_vp9_mode_11() {
  if [ "$(vp9_encode_available)" = "yes" ]; then
-    vpx_tsvc_encoder vp9 "${FUNCNAME}" 11 200 400 600 || return 1
+    local readonly output_basename="vpx_tsvc_encoder_vp9_mode_11"
+    vpx_tsvc_encoder vp9 "${output_basename}" 11 200 400 600 || return 1
    # Mode 11 produces 3 streams
-    files_exist "${FUNCNAME}" 3 || return 1
+    files_exist "${output_basename}" 3 || return 1
  fi
 }

--- a/test/vpxenc.sh
+++ b/test/vpxenc.sh
@@ -90,6 +90,15 @@ vpxenc_rt_params() {
    --undershoot-pct=50"
 }

+# Forces --passes to 1 with CONFIG_REALTIME_ONLY.
+vpxenc_passes_param() {
+  if [ "$(vpx_config_option_enabled CONFIG_REALTIME_ONLY)" = "yes" ]; then
+    echo "--passes=1"
+  else
+    echo "--passes=2"
+  fi
+}
+
 # Wrapper function for running vpxenc with pipe input. Requires that
 # LIBVPX_BIN_PATH points to the directory containing vpxenc. $1 is used as the
 # input file path and shifted away. All remaining parameters are passed through
@@ -218,9 +227,11 @@ vpxenc_vp8_ivf_piped_input() {
 vpxenc_vp9_ivf() {
  if [ "$(vpxenc_can_encode_vp9)" = "yes" ]; then
    local readonly output="${VPX_TEST_OUTPUT_DIR}/vp9.ivf"
+    local readonly passes=$(vpxenc_passes_param)
    vpxenc $(yuv_input_hantro_collage) \
      --codec=vp9 \
      --limit="${TEST_FRAMES}" \
+      "${passes}" \
      --ivf \
      --output="${output}"

@@ -235,9 +246,11 @@ vpxenc_vp9_webm() {
  if [ "$(vpxenc_can_encode_vp9)" = "yes" ] && \
     [ "$(webm_io_available)" = "yes" ]; then
    local readonly output="${VPX_TEST_OUTPUT_DIR}/vp9.webm"
+    local readonly passes=$(vpxenc_passes_param)
    vpxenc $(yuv_input_hantro_collage) \
      --codec=vp9 \
      --limit="${TEST_FRAMES}" \
+      "${passes}" \
      --output="${output}"

    if [ ! -e "${output}" ]; then
@@ -339,11 +352,13 @@ vpxenc_vp9_webm_2pass() {
 vpxenc_vp9_ivf_lossless() {
  if [ "$(vpxenc_can_encode_vp9)" = "yes" ]; then
    local readonly output="${VPX_TEST_OUTPUT_DIR}/vp9_lossless.ivf"
+    local readonly passes=$(vpxenc_passes_param)
    vpxenc $(yuv_input_hantro_collage) \
      --codec=vp9 \
      --limit="${TEST_FRAMES}" \
      --ivf \
      --output="${output}" \
+      "${passes}" \
      --lossless=1

    if [ ! -e "${output}" ]; then
@@ -356,11 +371,13 @@ vpxenc_vp9_ivf_lossless() {
 vpxenc_vp9_ivf_minq0_maxq0() {
  if [ "$(vpxenc_can_encode_vp9)" = "yes" ]; then
    local readonly output="${VPX_TEST_OUTPUT_DIR}/vp9_lossless_minq0_maxq0.ivf"
+    local readonly passes=$(vpxenc_passes_param)
    vpxenc $(yuv_input_hantro_collage) \
      --codec=vp9 \
      --limit="${TEST_FRAMES}" \
      --ivf \
      --output="${output}" \
+      "${passes}" \
      --min-q=0 \
      --max-q=0

@@ -377,12 +394,13 @@ vpxenc_vp9_webm_lag10_frames20() {
    local readonly lag_total_frames=20
    local readonly lag_frames=10
    local readonly output="${VPX_TEST_OUTPUT_DIR}/vp9_lag10_frames20.webm"
+    local readonly passes=$(vpxenc_passes_param)
    vpxenc $(yuv_input_hantro_collage) \
      --codec=vp9 \
      --limit="${lag_total_frames}" \
      --lag-in-frames="${lag_frames}" \
      --output="${output}" \
-      --passes=2 \
+      "${passes}" \
      --auto-alt-ref=1

    if [ ! -e "${output}" ]; then
@@ -397,9 +415,11 @@ vpxenc_vp9_webm_non_square_par() {
  if [ "$(vpxenc_can_encode_vp9)" = "yes" ] && \
     [ "$(webm_io_available)" = "yes" ]; then
    local readonly output="${VPX_TEST_OUTPUT_DIR}/vp9_non_square_par.webm"
+    local readonly passes=$(vpxenc_passes_param)
    vpxenc $(y4m_input_non_square_par) \
      --codec=vp9 \
      --limit="${TEST_FRAMES}" \
+      "${passes}" \
      --output="${output}"

    if [ ! -e "${output}" ]; then
@@ -412,18 +432,21 @@ vpxenc_vp9_webm_non_square_par() {
 vpxenc_tests="vpxenc_vp8_ivf
              vpxenc_vp8_webm
              vpxenc_vp8_webm_rt
-              vpxenc_vp8_webm_2pass
-              vpxenc_vp8_webm_lag10_frames20
              vpxenc_vp8_ivf_piped_input
              vpxenc_vp9_ivf
              vpxenc_vp9_webm
              vpxenc_vp9_webm_rt
              vpxenc_vp9_webm_rt_multithread_tiled
              vpxenc_vp9_webm_rt_multithread_tiled_frameparallel
-              vpxenc_vp9_webm_2pass
              vpxenc_vp9_ivf_lossless
              vpxenc_vp9_ivf_minq0_maxq0
              vpxenc_vp9_webm_lag10_frames20
              vpxenc_vp9_webm_non_square_par"
+if [ "$(vpx_config_option_enabled CONFIG_REALTIME_ONLY)" != "yes" ]; then
+  vpxenc_tests="$vpxenc_tests
+                vpxenc_vp8_webm_2pass
+                vpxenc_vp8_webm_lag10_frames20
+                vpxenc_vp9_webm_2pass"
+fi

 run_tests vpxenc_verify_environment "${vpxenc_tests}"
--- a/test/webm_video_source.h
+++ b/test/webm_video_source.h
@@ -40,8 +40,8 @@ class WebMVideoSource : public CompressedVideoSource {

  virtual void Begin() {
    vpx_ctx_->file = OpenTestDataFile(file_name_);
-    ASSERT_TRUE(vpx_ctx_->file != NULL) << "Input file open failed. Filename: "
-                                        << file_name_;
+    ASSERT_TRUE(vpx_ctx_->file != NULL)
+        << "Input file open failed. Filename: " << file_name_;

    ASSERT_EQ(file_is_webm(webm_ctx_, vpx_ctx_), 1) << "file is not WebM";

--- a/test/y4m_video_source.h
+++ b/test/y4m_video_source.h
@@ -34,8 +34,8 @@ class Y4mVideoSource : public VideoSource {
  virtual void OpenSource() {
    CloseSource();
    input_file_ = OpenTestDataFile(file_name_);
-    ASSERT_TRUE(input_file_ != NULL) << "Input file open failed. Filename: "
-                                     << file_name_;
+    ASSERT_TRUE(input_file_ != NULL)
+        << "Input file open failed. Filename: " << file_name_;
  }

  virtual void ReadSourceToStart() {
--- a/test/yuv_video_source.h
+++ b/test/yuv_video_source.h
@@ -43,8 +43,8 @@ class YUVVideoSource : public VideoSource {
  virtual void Begin() {
    if (input_file_) fclose(input_file_);
    input_file_ = OpenTestDataFile(file_name_);
-    ASSERT_TRUE(input_file_ != NULL) << "Input file open failed. Filename: "
-                                     << file_name_;
+    ASSERT_TRUE(input_file_ != NULL)
+        << "Input file open failed. Filename: " << file_name_;
    if (start_) {
      fseek(input_file_, static_cast<unsigned>(raw_size_) * start_, SEEK_SET);
    }
--- a/third_party/googletest/README.libvpx
+++ b/third_party/googletest/README.libvpx
@@ -1,7 +1,7 @@
-URL: http://code.google.com/p/googletest/
-Version: 1.7.0
+URL: https://github.com/google/googletest
+Version: 1.8.0
 License: BSD
-License File: COPYING
+License File: LICENSE

 Description:
 Google's framework for writing C++ tests on a variety of platforms
@@ -12,10 +12,13 @@ failures, various options for running the tests, and XML test report
 generation.

 Local Modifications:
- Removed unused declarations of kPathSeparatorString to have warning
-  free build.
- Added GTEST_ATTRIBUTE_UNUSED_ to test registering dummies in TEST_P
-  and INSTANTIATE_TEST_CASE_P to remove warnings about unused variables
-  under GCC 5.
- Only define g_in_fast_death_test_child for non-Windows builds; quiets an
-  unused variable warning.
+- Remove everything but:
+  googletest-release-1.8.0/googletest/
+   CHANGES
+   CONTRIBUTORS
+   include
+   LICENSE
+   README.md
+   src
+- Suppress unsigned overflow instrumentation in the LCG
+  https://github.com/google/googletest/pull/1066
--- a/third_party/googletest/src/README
+++ b/third_party/googletest/src/README
@@ -1,435 +0,0 @@
-Google C++ Testing Framework
-============================
-
-http://code.google.com/p/googletest/
-
-Overview
--------
-
-Google's framework for writing C++ tests on a variety of platforms
-(Linux, Mac OS X, Windows, Windows CE, Symbian, etc).  Based on the
-xUnit architecture.  Supports automatic test discovery, a rich set of
-assertions, user-defined assertions, death tests, fatal and non-fatal
-failures, various options for running the tests, and XML test report
-generation.
-
-Please see the project page above for more information as well as the
-mailing list for questions, discussions, and development.  There is
-also an IRC channel on OFTC (irc.oftc.net) #gtest available.  Please
-join us!
-
-Requirements for End Users
--------------------------
-
-Google Test is designed to have fairly minimal requirements to build
-and use with your projects, but there are some.  Currently, we support
-Linux, Windows, Mac OS X, and Cygwin.  We will also make our best
-effort to support other platforms (e.g. Solaris, AIX, and z/OS).
-However, since core members of the Google Test project have no access
-to these platforms, Google Test may have outstanding issues there.  If
-you notice any problems on your platform, please notify
-googletestframework@googlegroups.com.  Patches for fixing them are
-even more welcome!
-
-### Linux Requirements ###
-
-These are the base requirements to build and use Google Test from a source
-package (as described below):
-  * GNU-compatible Make or gmake
-  * POSIX-standard shell
-  * POSIX(-2) Regular Expressions (regex.h)
-  * A C++98-standard-compliant compiler
-
-### Windows Requirements ###
-
-  * Microsoft Visual C++ 7.1 or newer
-
-### Cygwin Requirements ###
-
-  * Cygwin 1.5.25-14 or newer
-
-### Mac OS X Requirements ###
-
-  * Mac OS X 10.4 Tiger or newer
-  * Developer Tools Installed
-
-Also, you'll need CMake 2.6.4 or higher if you want to build the
-samples using the provided CMake script, regardless of the platform.
-
-Requirements for Contributors
-----------------------------
-
-We welcome patches.  If you plan to contribute a patch, you need to
-build Google Test and its own tests from an SVN checkout (described
-below), which has further requirements:
-
-  * Python version 2.3 or newer (for running some of the tests and
-    re-generating certain source files from templates)
-  * CMake 2.6.4 or newer
-
-Getting the Source
------------------
-
-There are two primary ways of getting Google Test's source code: you
-can download a stable source release in your preferred archive format,
-or directly check out the source from our Subversion (SVN) repositary.
-The SVN checkout requires a few extra steps and some extra software
-packages on your system, but lets you track the latest development and
-make patches much more easily, so we highly encourage it.
-
-### Source Package ###
-
-Google Test is released in versioned source packages which can be
-downloaded from the download page [1].  Several different archive
-formats are provided, but the only difference is the tools used to
-manipulate them, and the size of the resulting file.  Download
-whichever you are most comfortable with.
-
-  [1] http://code.google.com/p/googletest/downloads/list
-
-Once the package is downloaded, expand it using whichever tools you
-prefer for that type.  This will result in a new directory with the
-name "gtest-X.Y.Z" which contains all of the source code.  Here are
-some examples on Linux:
-
-  tar -xvzf gtest-X.Y.Z.tar.gz
-  tar -xvjf gtest-X.Y.Z.tar.bz2
-  unzip gtest-X.Y.Z.zip
-
-### SVN Checkout ###
-
-To check out the main branch (also known as the "trunk") of Google
-Test, run the following Subversion command:
-
-  svn checkout http://googletest.googlecode.com/svn/trunk/ gtest-svn
-
-Setting up the Build
--------------------
-
-To build Google Test and your tests that use it, you need to tell your
-build system where to find its headers and source files.  The exact
-way to do it depends on which build system you use, and is usually
-straightforward.
-
-### Generic Build Instructions ###
-
-Suppose you put Google Test in directory ${GTEST_DIR}.  To build it,
-create a library build target (or a project as called by Visual Studio
-and Xcode) to compile
-
-  ${GTEST_DIR}/src/gtest-all.cc
-
-with ${GTEST_DIR}/include in the system header search path and ${GTEST_DIR}
-in the normal header search path.  Assuming a Linux-like system and gcc,
-something like the following will do:
-
-  g++ -isystem ${GTEST_DIR}/include -I${GTEST_DIR} \
-      -pthread -c ${GTEST_DIR}/src/gtest-all.cc
-  ar -rv libgtest.a gtest-all.o
-
-(We need -pthread as Google Test uses threads.)
-
-Next, you should compile your test source file with
-${GTEST_DIR}/include in the system header search path, and link it
-with gtest and any other necessary libraries:
-
-  g++ -isystem ${GTEST_DIR}/include -pthread path/to/your_test.cc libgtest.a \
-      -o your_test
-
-As an example, the make/ directory contains a Makefile that you can
-use to build Google Test on systems where GNU make is available
-(e.g. Linux, Mac OS X, and Cygwin).  It doesn't try to build Google
-Test's own tests.  Instead, it just builds the Google Test library and
-a sample test.  You can use it as a starting point for your own build
-script.
-
-If the default settings are correct for your environment, the
-following commands should succeed:
-
-  cd ${GTEST_DIR}/make
-  make
-  ./sample1_unittest
-
-If you see errors, try to tweak the contents of make/Makefile to make
-them go away.  There are instructions in make/Makefile on how to do
-it.
-
-### Using CMake ###
-
-Google Test comes with a CMake build script (CMakeLists.txt) that can
-be used on a wide range of platforms ("C" stands for cross-platofrm.).
-If you don't have CMake installed already, you can download it for
-free from http://www.cmake.org/.
-
-CMake works by generating native makefiles or build projects that can
-be used in the compiler environment of your choice.  The typical
-workflow starts with:
-
-  mkdir mybuild       # Create a directory to hold the build output.
-  cd mybuild
-  cmake ${GTEST_DIR}  # Generate native build scripts.
-
-If you want to build Google Test's samples, you should replace the
-last command with
-
-  cmake -Dgtest_build_samples=ON ${GTEST_DIR}
-
-If you are on a *nix system, you should now see a Makefile in the
-current directory.  Just type 'make' to build gtest.
-
-If you use Windows and have Vistual Studio installed, a gtest.sln file
-and several .vcproj files will be created.  You can then build them
-using Visual Studio.
-
-On Mac OS X with Xcode installed, a .xcodeproj file will be generated.
-
-### Legacy Build Scripts ###
-
-Before settling on CMake, we have been providing hand-maintained build
-projects/scripts for Visual Studio, Xcode, and Autotools.  While we
-continue to provide them for convenience, they are not actively
-maintained any more.  We highly recommend that you follow the
-instructions in the previous two sections to integrate Google Test
-with your existing build system.
-
-If you still need to use the legacy build scripts, here's how:
-
-The msvc\ folder contains two solutions with Visual C++ projects.
-Open the gtest.sln or gtest-md.sln file using Visual Studio, and you
-are ready to build Google Test the same way you build any Visual
-Studio project.  Files that have names ending with -md use DLL
-versions of Microsoft runtime libraries (the /MD or the /MDd compiler
-option).  Files without that suffix use static versions of the runtime
-libraries (the /MT or the /MTd option).  Please note that one must use
-the same option to compile both gtest and the test code.  If you use
-Visual Studio 2005 or above, we recommend the -md version as /MD is
-the default for new projects in these versions of Visual Studio.
-
-On Mac OS X, open the gtest.xcodeproj in the xcode/ folder using
-Xcode.  Build the "gtest" target.  The universal binary framework will
-end up in your selected build directory (selected in the Xcode
-"Preferences..." -> "Building" pane and defaults to xcode/build).
-Alternatively, at the command line, enter:
-
-  xcodebuild
-
-This will build the "Release" configuration of gtest.framework in your
-default build location.  See the "xcodebuild" man page for more
-information about building different configurations and building in
-different locations.
-
-If you wish to use the Google Test Xcode project with Xcode 4.x and
-above, you need to either:
- * update the SDK configuration options in xcode/Config/General.xconfig.
-   Comment options SDKROOT, MACOS_DEPLOYMENT_TARGET, and GCC_VERSION. If
-   you choose this route you lose the ability to target earlier versions
-   of MacOS X.
- * Install an SDK for an earlier version. This doesn't appear to be
-   supported by Apple, but has been reported to work
-   (http://stackoverflow.com/questions/5378518).
-
-Tweaking Google Test
--------------------
-
-Google Test can be used in diverse environments.  The default
-configuration may not work (or may not work well) out of the box in
-some environments.  However, you can easily tweak Google Test by
-defining control macros on the compiler command line.  Generally,
-these macros are named like GTEST_XYZ and you define them to either 1
-or 0 to enable or disable a certain feature.
-
-We list the most frequently used macros below.  For a complete list,
-see file include/gtest/internal/gtest-port.h.
-
-### Choosing a TR1 Tuple Library ###
-
-Some Google Test features require the C++ Technical Report 1 (TR1)
-tuple library, which is not yet available with all compilers.  The
-good news is that Google Test implements a subset of TR1 tuple that's
-enough for its own need, and will automatically use this when the
-compiler doesn't provide TR1 tuple.
-
-Usually you don't need to care about which tuple library Google Test
-uses.  However, if your project already uses TR1 tuple, you need to
-tell Google Test to use the same TR1 tuple library the rest of your
-project uses, or the two tuple implementations will clash.  To do
-that, add
-
-  -DGTEST_USE_OWN_TR1_TUPLE=0
-
-to the compiler flags while compiling Google Test and your tests.  If
-you want to force Google Test to use its own tuple library, just add
-
-  -DGTEST_USE_OWN_TR1_TUPLE=1
-
-to the compiler flags instead.
-
-If you don't want Google Test to use tuple at all, add
-
-  -DGTEST_HAS_TR1_TUPLE=0
-
-and all features using tuple will be disabled.
-
-### Multi-threaded Tests ###
-
-Google Test is thread-safe where the pthread library is available.
-After #include "gtest/gtest.h", you can check the GTEST_IS_THREADSAFE
-macro to see whether this is the case (yes if the macro is #defined to
-1, no if it's undefined.).
-
-If Google Test doesn't correctly detect whether pthread is available
-in your environment, you can force it with
-
-  -DGTEST_HAS_PTHREAD=1
-
-or
-
-  -DGTEST_HAS_PTHREAD=0
-
-When Google Test uses pthread, you may need to add flags to your
-compiler and/or linker to select the pthread library, or you'll get
-link errors.  If you use the CMake script or the deprecated Autotools
-script, this is taken care of for you.  If you use your own build
-script, you'll need to read your compiler and linker's manual to
-figure out what flags to add.
-
-### As a Shared Library (DLL) ###
-
-Google Test is compact, so most users can build and link it as a
-static library for the simplicity.  You can choose to use Google Test
-as a shared library (known as a DLL on Windows) if you prefer.
-
-To compile *gtest* as a shared library, add
-
-  -DGTEST_CREATE_SHARED_LIBRARY=1
-
-to the compiler flags.  You'll also need to tell the linker to produce
-a shared library instead - consult your linker's manual for how to do
-it.
-
-To compile your *tests* that use the gtest shared library, add
-
-  -DGTEST_LINKED_AS_SHARED_LIBRARY=1
-
-to the compiler flags.
-
-Note: while the above steps aren't technically necessary today when
-using some compilers (e.g. GCC), they may become necessary in the
-future, if we decide to improve the speed of loading the library (see
-http://gcc.gnu.org/wiki/Visibility for details).  Therefore you are
-recommended to always add the above flags when using Google Test as a
-shared library.  Otherwise a future release of Google Test may break
-your build script.
-
-### Avoiding Macro Name Clashes ###
-
-In C++, macros don't obey namespaces.  Therefore two libraries that
-both define a macro of the same name will clash if you #include both
-definitions.  In case a Google Test macro clashes with another
-library, you can force Google Test to rename its macro to avoid the
-conflict.
-
-Specifically, if both Google Test and some other code define macro
-FOO, you can add
-
-  -DGTEST_DONT_DEFINE_FOO=1
-
-to the compiler flags to tell Google Test to change the macro's name
-from FOO to GTEST_FOO.  Currently FOO can be FAIL, SUCCEED, or TEST.
-For example, with -DGTEST_DONT_DEFINE_TEST=1, you'll need to write
-
-  GTEST_TEST(SomeTest, DoesThis) { ... }
-
-instead of
-
-  TEST(SomeTest, DoesThis) { ... }
-
-in order to define a test.
-
-Upgrating from an Earlier Version
---------------------------------
-
-We strive to keep Google Test releases backward compatible.
-Sometimes, though, we have to make some breaking changes for the
-users' long-term benefits.  This section describes what you'll need to
-do if you are upgrading from an earlier version of Google Test.
-
-### Upgrading from 1.3.0 or Earlier ###
-
-You may need to explicitly enable or disable Google Test's own TR1
-tuple library.  See the instructions in section "Choosing a TR1 Tuple
-Library".
-
-### Upgrading from 1.4.0 or Earlier ###
-
-The Autotools build script (configure + make) is no longer officially
-supportted.  You are encouraged to migrate to your own build system or
-use CMake.  If you still need to use Autotools, you can find
-instructions in the README file from Google Test 1.4.0.
-
-On platforms where the pthread library is available, Google Test uses
-it in order to be thread-safe.  See the "Multi-threaded Tests" section
-for what this means to your build script.
-
-If you use Microsoft Visual C++ 7.1 with exceptions disabled, Google
-Test will no longer compile.  This should affect very few people, as a
-large portion of STL (including <string>) doesn't compile in this mode
-anyway.  We decided to stop supporting it in order to greatly simplify
-Google Test's implementation.
-
-Developing Google Test
----------------------
-
-This section discusses how to make your own changes to Google Test.
-
-### Testing Google Test Itself ###
-
-To make sure your changes work as intended and don't break existing
-functionality, you'll want to compile and run Google Test's own tests.
-For that you can use CMake:
-
-  mkdir mybuild
-  cd mybuild
-  cmake -Dgtest_build_tests=ON ${GTEST_DIR}
-
-Make sure you have Python installed, as some of Google Test's tests
-are written in Python.  If the cmake command complains about not being
-able to find Python ("Could NOT find PythonInterp (missing:
-PYTHON_EXECUTABLE)"), try telling it explicitly where your Python
-executable can be found:
-
-  cmake -DPYTHON_EXECUTABLE=path/to/python -Dgtest_build_tests=ON ${GTEST_DIR}
-
-Next, you can build Google Test and all of its own tests.  On *nix,
-this is usually done by 'make'.  To run the tests, do
-
-  make test
-
-All tests should pass.
-
-### Regenerating Source Files ###
-
-Some of Google Test's source files are generated from templates (not
-in the C++ sense) using a script.  A template file is named FOO.pump,
-where FOO is the name of the file it will generate.  For example, the
-file include/gtest/internal/gtest-type-util.h.pump is used to generate
-gtest-type-util.h in the same directory.
-
-Normally you don't need to worry about regenerating the source files,
-unless you need to modify them.  In that case, you should modify the
-corresponding .pump files instead and run the pump.py Python script to
-regenerate them.  You can find pump.py in the scripts/ directory.
-Read the Pump manual [2] for how to use it.
-
-  [2] http://code.google.com/p/googletest/wiki/PumpManual
-
-### Contributing a Patch ###
-
-We welcome patches.  Please read the Google Test developer's guide [3]
-for how you can contribute.  In particular, make sure you have signed
-the Contributor License Agreement, or we won't be able to accept the
-patch.
-
-  [3] http://code.google.com/p/googletest/wiki/GoogleTestDevGuide
-
-Happy testing!
--- a/third_party/googletest/src/README.md
+++ b/third_party/googletest/src/README.md
@@ -0,0 +1,280 @@
+
+### Generic Build Instructions ###
+
+#### Setup ####
+
+To build Google Test and your tests that use it, you need to tell your
+build system where to find its headers and source files.  The exact
+way to do it depends on which build system you use, and is usually
+straightforward.
+
+#### Build ####
+
+Suppose you put Google Test in directory `${GTEST_DIR}`.  To build it,
+create a library build target (or a project as called by Visual Studio
+and Xcode) to compile
+
+    ${GTEST_DIR}/src/gtest-all.cc
+
+with `${GTEST_DIR}/include` in the system header search path and `${GTEST_DIR}`
+in the normal header search path.  Assuming a Linux-like system and gcc,
+something like the following will do:
+
+    g++ -isystem ${GTEST_DIR}/include -I${GTEST_DIR} \
+        -pthread -c ${GTEST_DIR}/src/gtest-all.cc
+    ar -rv libgtest.a gtest-all.o
+
+(We need `-pthread` as Google Test uses threads.)
+
+Next, you should compile your test source file with
+`${GTEST_DIR}/include` in the system header search path, and link it
+with gtest and any other necessary libraries:
+
+    g++ -isystem ${GTEST_DIR}/include -pthread path/to/your_test.cc libgtest.a \
+        -o your_test
+
+As an example, the make/ directory contains a Makefile that you can
+use to build Google Test on systems where GNU make is available
+(e.g. Linux, Mac OS X, and Cygwin).  It doesn't try to build Google
+Test's own tests.  Instead, it just builds the Google Test library and
+a sample test.  You can use it as a starting point for your own build
+script.
+
+If the default settings are correct for your environment, the
+following commands should succeed:
+
+    cd ${GTEST_DIR}/make
+    make
+    ./sample1_unittest
+
+If you see errors, try to tweak the contents of `make/Makefile` to make
+them go away.  There are instructions in `make/Makefile` on how to do
+it.
+
+### Using CMake ###
+
+Google Test comes with a CMake build script (
+[CMakeLists.txt](CMakeLists.txt)) that can be used on a wide range of platforms ("C" stands for
+cross-platform.). If you don't have CMake installed already, you can
+download it for free from <http://www.cmake.org/>.
+
+CMake works by generating native makefiles or build projects that can
+be used in the compiler environment of your choice.  The typical
+workflow starts with:
+
+    mkdir mybuild       # Create a directory to hold the build output.
+    cd mybuild
+    cmake ${GTEST_DIR}  # Generate native build scripts.
+
+If you want to build Google Test's samples, you should replace the
+last command with
+
+    cmake -Dgtest_build_samples=ON ${GTEST_DIR}
+
+If you are on a \*nix system, you should now see a Makefile in the
+current directory.  Just type 'make' to build gtest.
+
+If you use Windows and have Visual Studio installed, a `gtest.sln` file
+and several `.vcproj` files will be created.  You can then build them
+using Visual Studio.
+
+On Mac OS X with Xcode installed, a `.xcodeproj` file will be generated.
+
+### Legacy Build Scripts ###
+
+Before settling on CMake, we have been providing hand-maintained build
+projects/scripts for Visual Studio, Xcode, and Autotools.  While we
+continue to provide them for convenience, they are not actively
+maintained any more.  We highly recommend that you follow the
+instructions in the previous two sections to integrate Google Test
+with your existing build system.
+
+If you still need to use the legacy build scripts, here's how:
+
+The msvc\ folder contains two solutions with Visual C++ projects.
+Open the `gtest.sln` or `gtest-md.sln` file using Visual Studio, and you
+are ready to build Google Test the same way you build any Visual
+Studio project.  Files that have names ending with -md use DLL
+versions of Microsoft runtime libraries (the /MD or the /MDd compiler
+option).  Files without that suffix use static versions of the runtime
+libraries (the /MT or the /MTd option).  Please note that one must use
+the same option to compile both gtest and the test code.  If you use
+Visual Studio 2005 or above, we recommend the -md version as /MD is
+the default for new projects in these versions of Visual Studio.
+
+On Mac OS X, open the `gtest.xcodeproj` in the `xcode/` folder using
+Xcode.  Build the "gtest" target.  The universal binary framework will
+end up in your selected build directory (selected in the Xcode
+"Preferences..." -> "Building" pane and defaults to xcode/build).
+Alternatively, at the command line, enter:
+
+    xcodebuild
+
+This will build the "Release" configuration of gtest.framework in your
+default build location.  See the "xcodebuild" man page for more
+information about building different configurations and building in
+different locations.
+
+If you wish to use the Google Test Xcode project with Xcode 4.x and
+above, you need to either:
+
+ * update the SDK configuration options in xcode/Config/General.xconfig.
+   Comment options `SDKROOT`, `MACOS_DEPLOYMENT_TARGET`, and `GCC_VERSION`. If
+   you choose this route you lose the ability to target earlier versions
+   of MacOS X.
+ * Install an SDK for an earlier version. This doesn't appear to be
+   supported by Apple, but has been reported to work
+   (http://stackoverflow.com/questions/5378518).
+
+### Tweaking Google Test ###
+
+Google Test can be used in diverse environments.  The default
+configuration may not work (or may not work well) out of the box in
+some environments.  However, you can easily tweak Google Test by
+defining control macros on the compiler command line.  Generally,
+these macros are named like `GTEST_XYZ` and you define them to either 1
+or 0 to enable or disable a certain feature.
+
+We list the most frequently used macros below.  For a complete list,
+see file [include/gtest/internal/gtest-port.h](include/gtest/internal/gtest-port.h).
+
+### Choosing a TR1 Tuple Library ###
+
+Some Google Test features require the C++ Technical Report 1 (TR1)
+tuple library, which is not yet available with all compilers.  The
+good news is that Google Test implements a subset of TR1 tuple that's
+enough for its own need, and will automatically use this when the
+compiler doesn't provide TR1 tuple.
+
+Usually you don't need to care about which tuple library Google Test
+uses.  However, if your project already uses TR1 tuple, you need to
+tell Google Test to use the same TR1 tuple library the rest of your
+project uses, or the two tuple implementations will clash.  To do
+that, add
+
+    -DGTEST_USE_OWN_TR1_TUPLE=0
+
+to the compiler flags while compiling Google Test and your tests.  If
+you want to force Google Test to use its own tuple library, just add
+
+    -DGTEST_USE_OWN_TR1_TUPLE=1
+
+to the compiler flags instead.
+
+If you don't want Google Test to use tuple at all, add
+
+    -DGTEST_HAS_TR1_TUPLE=0
+
+and all features using tuple will be disabled.
+
+### Multi-threaded Tests ###
+
+Google Test is thread-safe where the pthread library is available.
+After `#include "gtest/gtest.h"`, you can check the `GTEST_IS_THREADSAFE`
+macro to see whether this is the case (yes if the macro is `#defined` to
+1, no if it's undefined.).
+
+If Google Test doesn't correctly detect whether pthread is available
+in your environment, you can force it with
+
+    -DGTEST_HAS_PTHREAD=1
+
+or
+
+    -DGTEST_HAS_PTHREAD=0
+
+When Google Test uses pthread, you may need to add flags to your
+compiler and/or linker to select the pthread library, or you'll get
+link errors.  If you use the CMake script or the deprecated Autotools
+script, this is taken care of for you.  If you use your own build
+script, you'll need to read your compiler and linker's manual to
+figure out what flags to add.
+
+### As a Shared Library (DLL) ###
+
+Google Test is compact, so most users can build and link it as a
+static library for the simplicity.  You can choose to use Google Test
+as a shared library (known as a DLL on Windows) if you prefer.
+
+To compile *gtest* as a shared library, add
+
+    -DGTEST_CREATE_SHARED_LIBRARY=1
+
+to the compiler flags.  You'll also need to tell the linker to produce
+a shared library instead - consult your linker's manual for how to do
+it.
+
+To compile your *tests* that use the gtest shared library, add
+
+    -DGTEST_LINKED_AS_SHARED_LIBRARY=1
+
+to the compiler flags.
+
+Note: while the above steps aren't technically necessary today when
+using some compilers (e.g. GCC), they may become necessary in the
+future, if we decide to improve the speed of loading the library (see
+<http://gcc.gnu.org/wiki/Visibility> for details).  Therefore you are
+recommended to always add the above flags when using Google Test as a
+shared library.  Otherwise a future release of Google Test may break
+your build script.
+
+### Avoiding Macro Name Clashes ###
+
+In C++, macros don't obey namespaces.  Therefore two libraries that
+both define a macro of the same name will clash if you `#include` both
+definitions.  In case a Google Test macro clashes with another
+library, you can force Google Test to rename its macro to avoid the
+conflict.
+
+Specifically, if both Google Test and some other code define macro
+FOO, you can add
+
+    -DGTEST_DONT_DEFINE_FOO=1
+
+to the compiler flags to tell Google Test to change the macro's name
+from `FOO` to `GTEST_FOO`.  Currently `FOO` can be `FAIL`, `SUCCEED`,
+or `TEST`.  For example, with `-DGTEST_DONT_DEFINE_TEST=1`, you'll
+need to write
+
+    GTEST_TEST(SomeTest, DoesThis) { ... }
+
+instead of
+
+    TEST(SomeTest, DoesThis) { ... }
+
+in order to define a test.
+
+## Developing Google Test ##
+
+This section discusses how to make your own changes to Google Test.
+
+### Testing Google Test Itself ###
+
+To make sure your changes work as intended and don't break existing
+functionality, you'll want to compile and run Google Test's own tests.
+For that you can use CMake:
+
+    mkdir mybuild
+    cd mybuild
+    cmake -Dgtest_build_tests=ON ${GTEST_DIR}
+
+Make sure you have Python installed, as some of Google Test's tests
+are written in Python.  If the cmake command complains about not being
+able to find Python (`Could NOT find PythonInterp (missing:
+PYTHON_EXECUTABLE)`), try telling it explicitly where your Python
+executable can be found:
+
+    cmake -DPYTHON_EXECUTABLE=path/to/python -Dgtest_build_tests=ON ${GTEST_DIR}
+
+Next, you can build Google Test and all of its own tests.  On \*nix,
+this is usually done by 'make'.  To run the tests, do
+
+    make test
+
+All tests should pass.
+
+Normally you don't need to worry about regenerating the source files,
+unless you need to modify them.  In that case, you should modify the
+corresponding .pump files instead and run the pump.py Python script to
+regenerate them.  You can find pump.py in the [scripts/](scripts/) directory.
+Read the [Pump manual](docs/PumpManual.md) for how to use it.
--- a/third_party/googletest/src/include/gtest/gtest-death-test.h
+++ b/third_party/googletest/src/include/gtest/gtest-death-test.h
@@ -0,0 +1,294 @@
+// Copyright 2005, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+// Author: wan@google.com (Zhanyong Wan)
+//
+// The Google C++ Testing Framework (Google Test)
+//
+// This header file defines the public API for death tests.  It is
+// #included by gtest.h so a user doesn't need to include this
+// directly.
+
+#ifndef GTEST_INCLUDE_GTEST_GTEST_DEATH_TEST_H_
+#define GTEST_INCLUDE_GTEST_GTEST_DEATH_TEST_H_
+
+#include "gtest/internal/gtest-death-test-internal.h"
+
+namespace testing {
+
+// This flag controls the style of death tests.  Valid values are "threadsafe",
+// meaning that the death test child process will re-execute the test binary
+// from the start, running only a single death test, or "fast",
+// meaning that the child process will execute the test logic immediately
+// after forking.
+GTEST_DECLARE_string_(death_test_style);
+
+#if GTEST_HAS_DEATH_TEST
+
+namespace internal {
+
+// Returns a Boolean value indicating whether the caller is currently
+// executing in the context of the death test child process.  Tools such as
+// Valgrind heap checkers may need this to modify their behavior in death
+// tests.  IMPORTANT: This is an internal utility.  Using it may break the
+// implementation of death tests.  User code MUST NOT use it.
+GTEST_API_ bool InDeathTestChild();
+
+}  // namespace internal
+
+// The following macros are useful for writing death tests.
+
+// Here's what happens when an ASSERT_DEATH* or EXPECT_DEATH* is
+// executed:
+//
+//   1. It generates a warning if there is more than one active
+//   thread.  This is because it's safe to fork() or clone() only
+//   when there is a single thread.
+//
+//   2. The parent process clone()s a sub-process and runs the death
+//   test in it; the sub-process exits with code 0 at the end of the
+//   death test, if it hasn't exited already.
+//
+//   3. The parent process waits for the sub-process to terminate.
+//
+//   4. The parent process checks the exit code and error message of
+//   the sub-process.
+//
+// Examples:
+//
+//   ASSERT_DEATH(server.SendMessage(56, "Hello"), "Invalid port number");
+//   for (int i = 0; i < 5; i++) {
+//     EXPECT_DEATH(server.ProcessRequest(i),
+//                  "Invalid request .* in ProcessRequest()")
+//                  << "Failed to die on request " << i;
+//   }
+//
+//   ASSERT_EXIT(server.ExitNow(), ::testing::ExitedWithCode(0), "Exiting");
+//
+//   bool KilledBySIGHUP(int exit_code) {
+//     return WIFSIGNALED(exit_code) && WTERMSIG(exit_code) == SIGHUP;
+//   }
+//
+//   ASSERT_EXIT(client.HangUpServer(), KilledBySIGHUP, "Hanging up!");
+//
+// On the regular expressions used in death tests:
+//
+//   On POSIX-compliant systems (*nix), we use the <regex.h> library,
+//   which uses the POSIX extended regex syntax.
+//
+//   On other platforms (e.g. Windows), we only support a simple regex
+//   syntax implemented as part of Google Test.  This limited
+//   implementation should be enough most of the time when writing
+//   death tests; though it lacks many features you can find in PCRE
+//   or POSIX extended regex syntax.  For example, we don't support
+//   union ("x|y"), grouping ("(xy)"), brackets ("[xy]"), and
+//   repetition count ("x{5,7}"), among others.
+//
+//   Below is the syntax that we do support.  We chose it to be a
+//   subset of both PCRE and POSIX extended regex, so it's easy to
+//   learn wherever you come from.  In the following: 'A' denotes a
+//   literal character, period (.), or a single \\ escape sequence;
+//   'x' and 'y' denote regular expressions; 'm' and 'n' are for
+//   natural numbers.
+//
+//     c     matches any literal character c
+//     \\d   matches any decimal digit
+//     \\D   matches any character that's not a decimal digit
+//     \\f   matches \f
+//     \\n   matches \n
+//     \\r   matches \r
+//     \\s   matches any ASCII whitespace, including \n
+//     \\S   matches any character that's not a whitespace
+//     \\t   matches \t
+//     \\v   matches \v
+//     \\w   matches any letter, _, or decimal digit
+//     \\W   matches any character that \\w doesn't match
+//     \\c   matches any literal character c, which must be a punctuation
+//     .     matches any single character except \n
+//     A?    matches 0 or 1 occurrences of A
+//     A*    matches 0 or many occurrences of A
+//     A+    matches 1 or many occurrences of A
+//     ^     matches the beginning of a string (not that of each line)
+//     $     matches the end of a string (not that of each line)
+//     xy    matches x followed by y
+//
+//   If you accidentally use PCRE or POSIX extended regex features
+//   not implemented by us, you will get a run-time failure.  In that
+//   case, please try to rewrite your regular expression within the
+//   above syntax.
+//
+//   This implementation is *not* meant to be as highly tuned or robust
+//   as a compiled regex library, but should perform well enough for a
+//   death test, which already incurs significant overhead by launching
+//   a child process.
+//
+// Known caveats:
+//
+//   A "threadsafe" style death test obtains the path to the test
+//   program from argv[0] and re-executes it in the sub-process.  For
+//   simplicity, the current implementation doesn't search the PATH
+//   when launching the sub-process.  This means that the user must
+//   invoke the test program via a path that contains at least one
+//   path separator (e.g. path/to/foo_test and
+//   /absolute/path/to/bar_test are fine, but foo_test is not).  This
+//   is rarely a problem as people usually don't put the test binary
+//   directory in PATH.
+//
+// TODO(wan@google.com): make thread-safe death tests search the PATH.
+
+// Asserts that a given statement causes the program to exit, with an
+// integer exit status that satisfies predicate, and emitting error output
+// that matches regex.
+# define ASSERT_EXIT(statement, predicate, regex) \
+    GTEST_DEATH_TEST_(statement, predicate, regex, GTEST_FATAL_FAILURE_)
+
+// Like ASSERT_EXIT, but continues on to successive tests in the
+// test case, if any:
+# define EXPECT_EXIT(statement, predicate, regex) \
+    GTEST_DEATH_TEST_(statement, predicate, regex, GTEST_NONFATAL_FAILURE_)
+
+// Asserts that a given statement causes the program to exit, either by
+// explicitly exiting with a nonzero exit code or being killed by a
+// signal, and emitting error output that matches regex.
+# define ASSERT_DEATH(statement, regex) \
+    ASSERT_EXIT(statement, ::testing::internal::ExitedUnsuccessfully, regex)
+
+// Like ASSERT_DEATH, but continues on to successive tests in the
+// test case, if any:
+# define EXPECT_DEATH(statement, regex) \
+    EXPECT_EXIT(statement, ::testing::internal::ExitedUnsuccessfully, regex)
+
+// Two predicate classes that can be used in {ASSERT,EXPECT}_EXIT*:
+
+// Tests that an exit code describes a normal exit with a given exit code.
+class GTEST_API_ ExitedWithCode {
+ public:
+  explicit ExitedWithCode(int exit_code);
+  bool operator()(int exit_status) const;
+ private:
+  // No implementation - assignment is unsupported.
+  void operator=(const ExitedWithCode& other);
+
+  const int exit_code_;
+};
+
+# if !GTEST_OS_WINDOWS
+// Tests that an exit code describes an exit due to termination by a
+// given signal.
+class GTEST_API_ KilledBySignal {
+ public:
+  explicit KilledBySignal(int signum);
+  bool operator()(int exit_status) const;
+ private:
+  const int signum_;
+};
+# endif  // !GTEST_OS_WINDOWS
+
+// EXPECT_DEBUG_DEATH asserts that the given statements die in debug mode.
+// The death testing framework causes this to have interesting semantics,
+// since the sideeffects of the call are only visible in opt mode, and not
+// in debug mode.
+//
+// In practice, this can be used to test functions that utilize the
+// LOG(DFATAL) macro using the following style:
+//
+// int DieInDebugOr12(int* sideeffect) {
+//   if (sideeffect) {
+//     *sideeffect = 12;
+//   }
+//   LOG(DFATAL) << "death";
+//   return 12;
+// }
+//
+// TEST(TestCase, TestDieOr12WorksInDgbAndOpt) {
+//   int sideeffect = 0;
+//   // Only asserts in dbg.
+//   EXPECT_DEBUG_DEATH(DieInDebugOr12(&sideeffect), "death");
+//
+// #ifdef NDEBUG
+//   // opt-mode has sideeffect visible.
+//   EXPECT_EQ(12, sideeffect);
+// #else
+//   // dbg-mode no visible sideeffect.
+//   EXPECT_EQ(0, sideeffect);
+// #endif
+// }
+//
+// This will assert that DieInDebugReturn12InOpt() crashes in debug
+// mode, usually due to a DCHECK or LOG(DFATAL), but returns the
+// appropriate fallback value (12 in this case) in opt mode. If you
+// need to test that a function has appropriate side-effects in opt
+// mode, include assertions against the side-effects.  A general
+// pattern for this is:
+//
+// EXPECT_DEBUG_DEATH({
+//   // Side-effects here will have an effect after this statement in
+//   // opt mode, but none in debug mode.
+//   EXPECT_EQ(12, DieInDebugOr12(&sideeffect));
+// }, "death");
+//
+# ifdef NDEBUG
+
+#  define EXPECT_DEBUG_DEATH(statement, regex) \
+  GTEST_EXECUTE_STATEMENT_(statement, regex)
+
+#  define ASSERT_DEBUG_DEATH(statement, regex) \
+  GTEST_EXECUTE_STATEMENT_(statement, regex)
+
+# else
+
+#  define EXPECT_DEBUG_DEATH(statement, regex) \
+  EXPECT_DEATH(statement, regex)
+
+#  define ASSERT_DEBUG_DEATH(statement, regex) \
+  ASSERT_DEATH(statement, regex)
+
+# endif  // NDEBUG for EXPECT_DEBUG_DEATH
+#endif  // GTEST_HAS_DEATH_TEST
+
+// EXPECT_DEATH_IF_SUPPORTED(statement, regex) and
+// ASSERT_DEATH_IF_SUPPORTED(statement, regex) expand to real death tests if
+// death tests are supported; otherwise they just issue a warning.  This is
+// useful when you are combining death test assertions with normal test
+// assertions in one test.
+#if GTEST_HAS_DEATH_TEST
+# define EXPECT_DEATH_IF_SUPPORTED(statement, regex) \
+    EXPECT_DEATH(statement, regex)
+# define ASSERT_DEATH_IF_SUPPORTED(statement, regex) \
+    ASSERT_DEATH(statement, regex)
+#else
+# define EXPECT_DEATH_IF_SUPPORTED(statement, regex) \
+    GTEST_UNSUPPORTED_DEATH_TEST_(statement, regex, )
+# define ASSERT_DEATH_IF_SUPPORTED(statement, regex) \
+    GTEST_UNSUPPORTED_DEATH_TEST_(statement, regex, return)
+#endif
+
+}  // namespace testing
+
+#endif  // GTEST_INCLUDE_GTEST_GTEST_DEATH_TEST_H_
--- a/third_party/googletest/src/include/gtest/gtest-message.h
+++ b/third_party/googletest/src/include/gtest/gtest-message.h
@@ -0,0 +1,250 @@
+// Copyright 2005, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+// Author: wan@google.com (Zhanyong Wan)
+//
+// The Google C++ Testing Framework (Google Test)
+//
+// This header file defines the Message class.
+//
+// IMPORTANT NOTE: Due to limitation of the C++ language, we have to
+// leave some internal implementation details in this header file.
+// They are clearly marked by comments like this:
+//
+//   // INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
+//
+// Such code is NOT meant to be used by a user directly, and is subject
+// to CHANGE WITHOUT NOTICE.  Therefore DO NOT DEPEND ON IT in a user
+// program!
+
+#ifndef GTEST_INCLUDE_GTEST_GTEST_MESSAGE_H_
+#define GTEST_INCLUDE_GTEST_GTEST_MESSAGE_H_
+
+#include <limits>
+
+#include "gtest/internal/gtest-port.h"
+
+// Ensures that there is at least one operator<< in the global namespace.
+// See Message& operator<<(...) below for why.
+void operator<<(const testing::internal::Secret&, int);
+
+namespace testing {
+
+// The Message class works like an ostream repeater.
+//
+// Typical usage:
+//
+//   1. You stream a bunch of values to a Message object.
+//      It will remember the text in a stringstream.
+//   2. Then you stream the Message object to an ostream.
+//      This causes the text in the Message to be streamed
+//      to the ostream.
+//
+// For example;
+//
+//   testing::Message foo;
+//   foo << 1 << " != " << 2;
+//   std::cout << foo;
+//
+// will print "1 != 2".
+//
+// Message is not intended to be inherited from.  In particular, its
+// destructor is not virtual.
+//
+// Note that stringstream behaves differently in gcc and in MSVC.  You
+// can stream a NULL char pointer to it in the former, but not in the
+// latter (it causes an access violation if you do).  The Message
+// class hides this difference by treating a NULL char pointer as
+// "(null)".
+class GTEST_API_ Message {
+ private:
+  // The type of basic IO manipulators (endl, ends, and flush) for
+  // narrow streams.
+  typedef std::ostream& (*BasicNarrowIoManip)(std::ostream&);
+
+ public:
+  // Constructs an empty Message.
+  Message();
+
+  // Copy constructor.
+  Message(const Message& msg) : ss_(new ::std::stringstream) {  // NOLINT
+    *ss_ << msg.GetString();
+  }
+
+  // Constructs a Message from a C-string.
+  explicit Message(const char* str) : ss_(new ::std::stringstream) {
+    *ss_ << str;
+  }
+
+#if GTEST_OS_SYMBIAN
+  // Streams a value (either a pointer or not) to this object.
+  template <typename T>
+  inline Message& operator <<(const T& value) {
+    StreamHelper(typename internal::is_pointer<T>::type(), value);
+    return *this;
+  }
+#else
+  // Streams a non-pointer value to this object.
+  template <typename T>
+  inline Message& operator <<(const T& val) {
+    // Some libraries overload << for STL containers.  These
+    // overloads are defined in the global namespace instead of ::std.
+    //
+    // C++'s symbol lookup rule (i.e. Koenig lookup) says that these
+    // overloads are visible in either the std namespace or the global
+    // namespace, but not other namespaces, including the testing
+    // namespace which Google Test's Message class is in.
+    //
+    // To allow STL containers (and other types that has a << operator
+    // defined in the global namespace) to be used in Google Test
+    // assertions, testing::Message must access the custom << operator
+    // from the global namespace.  With this using declaration,
+    // overloads of << defined in the global namespace and those
+    // visible via Koenig lookup are both exposed in this function.
+    using ::operator <<;
+    *ss_ << val;
+    return *this;
+  }
+
+  // Streams a pointer value to this object.
+  //
+  // This function is an overload of the previous one.  When you
+  // stream a pointer to a Message, this definition will be used as it
+  // is more specialized.  (The C++ Standard, section
+  // [temp.func.order].)  If you stream a non-pointer, then the
+  // previous definition will be used.
+  //
+  // The reason for this overload is that streaming a NULL pointer to
+  // ostream is undefined behavior.  Depending on the compiler, you
+  // may get "0", "(nil)", "(null)", or an access violation.  To
+  // ensure consistent result across compilers, we always treat NULL
+  // as "(null)".
+  template <typename T>
+  inline Message& operator <<(T* const& pointer) {  // NOLINT
+    if (pointer == NULL) {
+      *ss_ << "(null)";
+    } else {
+      *ss_ << pointer;
+    }
+    return *this;
+  }
+#endif  // GTEST_OS_SYMBIAN
+
+  // Since the basic IO manipulators are overloaded for both narrow
+  // and wide streams, we have to provide this specialized definition
+  // of operator <<, even though its body is the same as the
+  // templatized version above.  Without this definition, streaming
+  // endl or other basic IO manipulators to Message will confuse the
+  // compiler.
+  Message& operator <<(BasicNarrowIoManip val) {
+    *ss_ << val;
+    return *this;
+  }
+
+  // Instead of 1/0, we want to see true/false for bool values.
+  Message& operator <<(bool b) {
+    return *this << (b ? "true" : "false");
+  }
+
+  // These two overloads allow streaming a wide C string to a Message
+  // using the UTF-8 encoding.
+  Message& operator <<(const wchar_t* wide_c_str);
+  Message& operator <<(wchar_t* wide_c_str);
+
+#if GTEST_HAS_STD_WSTRING
+  // Converts the given wide string to a narrow string using the UTF-8
+  // encoding, and streams the result to this Message object.
+  Message& operator <<(const ::std::wstring& wstr);
+#endif  // GTEST_HAS_STD_WSTRING
+
+#if GTEST_HAS_GLOBAL_WSTRING
+  // Converts the given wide string to a narrow string using the UTF-8
+  // encoding, and streams the result to this Message object.
+  Message& operator <<(const ::wstring& wstr);
+#endif  // GTEST_HAS_GLOBAL_WSTRING
+
+  // Gets the text streamed to this object so far as an std::string.
+  // Each '\0' character in the buffer is replaced with "\\0".
+  //
+  // INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
+  std::string GetString() const;
+
+ private:
+
+#if GTEST_OS_SYMBIAN
+  // These are needed as the Nokia Symbian Compiler cannot decide between
+  // const T& and const T* in a function template. The Nokia compiler _can_
+  // decide between class template specializations for T and T*, so a
+  // tr1::type_traits-like is_pointer works, and we can overload on that.
+  template <typename T>
+  inline void StreamHelper(internal::true_type /*is_pointer*/, T* pointer) {
+    if (pointer == NULL) {
+      *ss_ << "(null)";
+    } else {
+      *ss_ << pointer;
+    }
+  }
+  template <typename T>
+  inline void StreamHelper(internal::false_type /*is_pointer*/,
+                           const T& value) {
+    // See the comments in Message& operator <<(const T&) above for why
+    // we need this using statement.
+    using ::operator <<;
+    *ss_ << value;
+  }
+#endif  // GTEST_OS_SYMBIAN
+
+  // We'll hold the text streamed to this object here.
+  const internal::scoped_ptr< ::std::stringstream> ss_;
+
+  // We declare (but don't implement) this to prevent the compiler
+  // from implementing the assignment operator.
+  void operator=(const Message&);
+};
+
+// Streams a Message to an ostream.
+inline std::ostream& operator <<(std::ostream& os, const Message& sb) {
+  return os << sb.GetString();
+}
+
+namespace internal {
+
+// Converts a streamable value to an std::string.  A NULL pointer is
+// converted to "(null)".  When the input value is a ::string,
+// ::std::string, ::wstring, or ::std::wstring object, each NUL
+// character in it is replaced with "\\0".
+template <typename T>
+std::string StreamableToString(const T& streamable) {
+  return (Message() << streamable).GetString();
+}
+
+}  // namespace internal
+}  // namespace testing
+
+#endif  // GTEST_INCLUDE_GTEST_GTEST_MESSAGE_H_
--- a/third_party/googletest/src/include/gtest/gtest-param-test.h
+++ b/third_party/googletest/src/include/gtest/gtest-param-test.h
--- a/third_party/googletest/src/include/gtest/gtest-param-test.h.pump
+++ b/third_party/googletest/src/include/gtest/gtest-param-test.h.pump
@@ -0,0 +1,510 @@
+$$ -*- mode: c++; -*-
+$var n = 50  $$ Maximum length of Values arguments we want to support.
+$var maxtuple = 10  $$ Maximum number of Combine arguments we want to support.
+// Copyright 2008, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+// Authors: vladl@google.com (Vlad Losev)
+//
+// Macros and functions for implementing parameterized tests
+// in Google C++ Testing Framework (Google Test)
+//
+// This file is generated by a SCRIPT.  DO NOT EDIT BY HAND!
+//
+#ifndef GTEST_INCLUDE_GTEST_GTEST_PARAM_TEST_H_
+#define GTEST_INCLUDE_GTEST_GTEST_PARAM_TEST_H_
+
+
+// Value-parameterized tests allow you to test your code with different
+// parameters without writing multiple copies of the same test.
+//
+// Here is how you use value-parameterized tests:
+
+#if 0
+
+// To write value-parameterized tests, first you should define a fixture
+// class. It is usually derived from testing::TestWithParam<T> (see below for
+// another inheritance scheme that's sometimes useful in more complicated
+// class hierarchies), where the type of your parameter values.
+// TestWithParam<T> is itself derived from testing::Test. T can be any
+// copyable type. If it's a raw pointer, you are responsible for managing the
+// lifespan of the pointed values.
+
+class FooTest : public ::testing::TestWithParam<const char*> {
+  // You can implement all the usual class fixture members here.
+};
+
+// Then, use the TEST_P macro to define as many parameterized tests
+// for this fixture as you want. The _P suffix is for "parameterized"
+// or "pattern", whichever you prefer to think.
+
+TEST_P(FooTest, DoesBlah) {
+  // Inside a test, access the test parameter with the GetParam() method
+  // of the TestWithParam<T> class:
+  EXPECT_TRUE(foo.Blah(GetParam()));
+  ...
+}
+
+TEST_P(FooTest, HasBlahBlah) {
+  ...
+}
+
+// Finally, you can use INSTANTIATE_TEST_CASE_P to instantiate the test
+// case with any set of parameters you want. Google Test defines a number
+// of functions for generating test parameters. They return what we call
+// (surprise!) parameter generators. Here is a  summary of them, which
+// are all in the testing namespace:
+//
+//
+//  Range(begin, end [, step]) - Yields values {begin, begin+step,
+//                               begin+step+step, ...}. The values do not
+//                               include end. step defaults to 1.
+//  Values(v1, v2, ..., vN)    - Yields values {v1, v2, ..., vN}.
+//  ValuesIn(container)        - Yields values from a C-style array, an STL
+//  ValuesIn(begin,end)          container, or an iterator range [begin, end).
+//  Bool()                     - Yields sequence {false, true}.
+//  Combine(g1, g2, ..., gN)   - Yields all combinations (the Cartesian product
+//                               for the math savvy) of the values generated
+//                               by the N generators.
+//
+// For more details, see comments at the definitions of these functions below
+// in this file.
+//
+// The following statement will instantiate tests from the FooTest test case
+// each with parameter values "meeny", "miny", and "moe".
+
+INSTANTIATE_TEST_CASE_P(InstantiationName,
+                        FooTest,
+                        Values("meeny", "miny", "moe"));
+
+// To distinguish different instances of the pattern, (yes, you
+// can instantiate it more then once) the first argument to the
+// INSTANTIATE_TEST_CASE_P macro is a prefix that will be added to the
+// actual test case name. Remember to pick unique prefixes for different
+// instantiations. The tests from the instantiation above will have
+// these names:
+//
+//    * InstantiationName/FooTest.DoesBlah/0 for "meeny"
+//    * InstantiationName/FooTest.DoesBlah/1 for "miny"
+//    * InstantiationName/FooTest.DoesBlah/2 for "moe"
+//    * InstantiationName/FooTest.HasBlahBlah/0 for "meeny"
+//    * InstantiationName/FooTest.HasBlahBlah/1 for "miny"
+//    * InstantiationName/FooTest.HasBlahBlah/2 for "moe"
+//
+// You can use these names in --gtest_filter.
+//
+// This statement will instantiate all tests from FooTest again, each
+// with parameter values "cat" and "dog":
+
+const char* pets[] = {"cat", "dog"};
+INSTANTIATE_TEST_CASE_P(AnotherInstantiationName, FooTest, ValuesIn(pets));
+
+// The tests from the instantiation above will have these names:
+//
+//    * AnotherInstantiationName/FooTest.DoesBlah/0 for "cat"
+//    * AnotherInstantiationName/FooTest.DoesBlah/1 for "dog"
+//    * AnotherInstantiationName/FooTest.HasBlahBlah/0 for "cat"
+//    * AnotherInstantiationName/FooTest.HasBlahBlah/1 for "dog"
+//
+// Please note that INSTANTIATE_TEST_CASE_P will instantiate all tests
+// in the given test case, whether their definitions come before or
+// AFTER the INSTANTIATE_TEST_CASE_P statement.
+//
+// Please also note that generator expressions (including parameters to the
+// generators) are evaluated in InitGoogleTest(), after main() has started.
+// This allows the user on one hand, to adjust generator parameters in order
+// to dynamically determine a set of tests to run and on the other hand,
+// give the user a chance to inspect the generated tests with Google Test
+// reflection API before RUN_ALL_TESTS() is executed.
+//
+// You can see samples/sample7_unittest.cc and samples/sample8_unittest.cc
+// for more examples.
+//
+// In the future, we plan to publish the API for defining new parameter
+// generators. But for now this interface remains part of the internal
+// implementation and is subject to change.
+//
+//
+// A parameterized test fixture must be derived from testing::Test and from
+// testing::WithParamInterface<T>, where T is the type of the parameter
+// values. Inheriting from TestWithParam<T> satisfies that requirement because
+// TestWithParam<T> inherits from both Test and WithParamInterface. In more
+// complicated hierarchies, however, it is occasionally useful to inherit
+// separately from Test and WithParamInterface. For example:
+
+class BaseTest : public ::testing::Test {
+  // You can inherit all the usual members for a non-parameterized test
+  // fixture here.
+};
+
+class DerivedTest : public BaseTest, public ::testing::WithParamInterface<int> {
+  // The usual test fixture members go here too.
+};
+
+TEST_F(BaseTest, HasFoo) {
+  // This is an ordinary non-parameterized test.
+}
+
+TEST_P(DerivedTest, DoesBlah) {
+  // GetParam works just the same here as if you inherit from TestWithParam.
+  EXPECT_TRUE(foo.Blah(GetParam()));
+}
+
+#endif  // 0
+
+#include "gtest/internal/gtest-port.h"
+
+#if !GTEST_OS_SYMBIAN
+# include <utility>
+#endif
+
+// scripts/fuse_gtest.py depends on gtest's own header being #included
+// *unconditionally*.  Therefore these #includes cannot be moved
+// inside #if GTEST_HAS_PARAM_TEST.
+#include "gtest/internal/gtest-internal.h"
+#include "gtest/internal/gtest-param-util.h"
+#include "gtest/internal/gtest-param-util-generated.h"
+
+#if GTEST_HAS_PARAM_TEST
+
+namespace testing {
+
+// Functions producing parameter generators.
+//
+// Google Test uses these generators to produce parameters for value-
+// parameterized tests. When a parameterized test case is instantiated
+// with a particular generator, Google Test creates and runs tests
+// for each element in the sequence produced by the generator.
+//
+// In the following sample, tests from test case FooTest are instantiated
+// each three times with parameter values 3, 5, and 8:
+//
+// class FooTest : public TestWithParam<int> { ... };
+//
+// TEST_P(FooTest, TestThis) {
+// }
+// TEST_P(FooTest, TestThat) {
+// }
+// INSTANTIATE_TEST_CASE_P(TestSequence, FooTest, Values(3, 5, 8));
+//
+
+// Range() returns generators providing sequences of values in a range.
+//
+// Synopsis:
+// Range(start, end)
+//   - returns a generator producing a sequence of values {start, start+1,
+//     start+2, ..., }.
+// Range(start, end, step)
+//   - returns a generator producing a sequence of values {start, start+step,
+//     start+step+step, ..., }.
+// Notes:
+//   * The generated sequences never include end. For example, Range(1, 5)
+//     returns a generator producing a sequence {1, 2, 3, 4}. Range(1, 9, 2)
+//     returns a generator producing {1, 3, 5, 7}.
+//   * start and end must have the same type. That type may be any integral or
+//     floating-point type or a user defined type satisfying these conditions:
+//     * It must be assignable (have operator=() defined).
+//     * It must have operator+() (operator+(int-compatible type) for
+//       two-operand version).
+//     * It must have operator<() defined.
+//     Elements in the resulting sequences will also have that type.
+//   * Condition start < end must be satisfied in order for resulting sequences
+//     to contain any elements.
+//
+template <typename T, typename IncrementT>
+internal::ParamGenerator<T> Range(T start, T end, IncrementT step) {
+  return internal::ParamGenerator<T>(
+      new internal::RangeGenerator<T, IncrementT>(start, end, step));
+}
+
+template <typename T>
+internal::ParamGenerator<T> Range(T start, T end) {
+  return Range(start, end, 1);
+}
+
+// ValuesIn() function allows generation of tests with parameters coming from
+// a container.
+//
+// Synopsis:
+// ValuesIn(const T (&array)[N])
+//   - returns a generator producing sequences with elements from
+//     a C-style array.
+// ValuesIn(const Container& container)
+//   - returns a generator producing sequences with elements from
+//     an STL-style container.
+// ValuesIn(Iterator begin, Iterator end)
+//   - returns a generator producing sequences with elements from
+//     a range [begin, end) defined by a pair of STL-style iterators. These
+//     iterators can also be plain C pointers.
+//
+// Please note that ValuesIn copies the values from the containers
+// passed in and keeps them to generate tests in RUN_ALL_TESTS().
+//
+// Examples:
+//
+// This instantiates tests from test case StringTest
+// each with C-string values of "foo", "bar", and "baz":
+//
+// const char* strings[] = {"foo", "bar", "baz"};
+// INSTANTIATE_TEST_CASE_P(StringSequence, SrtingTest, ValuesIn(strings));
+//
+// This instantiates tests from test case StlStringTest
+// each with STL strings with values "a" and "b":
+//
+// ::std::vector< ::std::string> GetParameterStrings() {
+//   ::std::vector< ::std::string> v;
+//   v.push_back("a");
+//   v.push_back("b");
+//   return v;
+// }
+//
+// INSTANTIATE_TEST_CASE_P(CharSequence,
+//                         StlStringTest,
+//                         ValuesIn(GetParameterStrings()));
+//
+//
+// This will also instantiate tests from CharTest
+// each with parameter values 'a' and 'b':
+//
+// ::std::list<char> GetParameterChars() {
+//   ::std::list<char> list;
+//   list.push_back('a');
+//   list.push_back('b');
+//   return list;
+// }
+// ::std::list<char> l = GetParameterChars();
+// INSTANTIATE_TEST_CASE_P(CharSequence2,
+//                         CharTest,
+//                         ValuesIn(l.begin(), l.end()));
+//
+template <typename ForwardIterator>
+internal::ParamGenerator<
+  typename ::testing::internal::IteratorTraits<ForwardIterator>::value_type>
+ValuesIn(ForwardIterator begin, ForwardIterator end) {
+  typedef typename ::testing::internal::IteratorTraits<ForwardIterator>
+      ::value_type ParamType;
+  return internal::ParamGenerator<ParamType>(
+      new internal::ValuesInIteratorRangeGenerator<ParamType>(begin, end));
+}
+
+template <typename T, size_t N>
+internal::ParamGenerator<T> ValuesIn(const T (&array)[N]) {
+  return ValuesIn(array, array + N);
+}
+
+template <class Container>
+internal::ParamGenerator<typename Container::value_type> ValuesIn(
+    const Container& container) {
+  return ValuesIn(container.begin(), container.end());
+}
+
+// Values() allows generating tests from explicitly specified list of
+// parameters.
+//
+// Synopsis:
+// Values(T v1, T v2, ..., T vN)
+//   - returns a generator producing sequences with elements v1, v2, ..., vN.
+//
+// For example, this instantiates tests from test case BarTest each
+// with values "one", "two", and "three":
+//
+// INSTANTIATE_TEST_CASE_P(NumSequence, BarTest, Values("one", "two", "three"));
+//
+// This instantiates tests from test case BazTest each with values 1, 2, 3.5.
+// The exact type of values will depend on the type of parameter in BazTest.
+//
+// INSTANTIATE_TEST_CASE_P(FloatingNumbers, BazTest, Values(1, 2, 3.5));
+//
+// Currently, Values() supports from 1 to $n parameters.
+//
+$range i 1..n
+$for i [[
+$range j 1..i
+
+template <$for j, [[typename T$j]]>
+internal::ValueArray$i<$for j, [[T$j]]> Values($for j, [[T$j v$j]]) {
+  return internal::ValueArray$i<$for j, [[T$j]]>($for j, [[v$j]]);
+}
+
+]]
+
+// Bool() allows generating tests with parameters in a set of (false, true).
+//
+// Synopsis:
+// Bool()
+//   - returns a generator producing sequences with elements {false, true}.
+//
+// It is useful when testing code that depends on Boolean flags. Combinations
+// of multiple flags can be tested when several Bool()'s are combined using
+// Combine() function.
+//
+// In the following example all tests in the test case FlagDependentTest
+// will be instantiated twice with parameters false and true.
+//
+// class FlagDependentTest : public testing::TestWithParam<bool> {
+//   virtual void SetUp() {
+//     external_flag = GetParam();
+//   }
+// }
+// INSTANTIATE_TEST_CASE_P(BoolSequence, FlagDependentTest, Bool());
+//
+inline internal::ParamGenerator<bool> Bool() {
+  return Values(false, true);
+}
+
+# if GTEST_HAS_COMBINE
+// Combine() allows the user to combine two or more sequences to produce
+// values of a Cartesian product of those sequences' elements.
+//
+// Synopsis:
+// Combine(gen1, gen2, ..., genN)
+//   - returns a generator producing sequences with elements coming from
+//     the Cartesian product of elements from the sequences generated by
+//     gen1, gen2, ..., genN. The sequence elements will have a type of
+//     tuple<T1, T2, ..., TN> where T1, T2, ..., TN are the types
+//     of elements from sequences produces by gen1, gen2, ..., genN.
+//
+// Combine can have up to $maxtuple arguments. This number is currently limited
+// by the maximum number of elements in the tuple implementation used by Google
+// Test.
+//
+// Example:
+//
+// This will instantiate tests in test case AnimalTest each one with
+// the parameter values tuple("cat", BLACK), tuple("cat", WHITE),
+// tuple("dog", BLACK), and tuple("dog", WHITE):
+//
+// enum Color { BLACK, GRAY, WHITE };
+// class AnimalTest
+//     : public testing::TestWithParam<tuple<const char*, Color> > {...};
+//
+// TEST_P(AnimalTest, AnimalLooksNice) {...}
+//
+// INSTANTIATE_TEST_CASE_P(AnimalVariations, AnimalTest,
+//                         Combine(Values("cat", "dog"),
+//                                 Values(BLACK, WHITE)));
+//
+// This will instantiate tests in FlagDependentTest with all variations of two
+// Boolean flags:
+//
+// class FlagDependentTest
+//     : public testing::TestWithParam<tuple<bool, bool> > {
+//   virtual void SetUp() {
+//     // Assigns external_flag_1 and external_flag_2 values from the tuple.
+//     tie(external_flag_1, external_flag_2) = GetParam();
+//   }
+// };
+//
+// TEST_P(FlagDependentTest, TestFeature1) {
+//   // Test your code using external_flag_1 and external_flag_2 here.
+// }
+// INSTANTIATE_TEST_CASE_P(TwoBoolSequence, FlagDependentTest,
+//                         Combine(Bool(), Bool()));
+//
+$range i 2..maxtuple
+$for i [[
+$range j 1..i
+
+template <$for j, [[typename Generator$j]]>
+internal::CartesianProductHolder$i<$for j, [[Generator$j]]> Combine(
+    $for j, [[const Generator$j& g$j]]) {
+  return internal::CartesianProductHolder$i<$for j, [[Generator$j]]>(
+      $for j, [[g$j]]);
+}
+
+]]
+# endif  // GTEST_HAS_COMBINE
+
+
+
+# define TEST_P(test_case_name, test_name) \
+  class GTEST_TEST_CLASS_NAME_(test_case_name, test_name) \
+      : public test_case_name { \
+   public: \
+    GTEST_TEST_CLASS_NAME_(test_case_name, test_name)() {} \
+    virtual void TestBody(); \
+   private: \
+    static int AddToRegistry() { \
+      ::testing::UnitTest::GetInstance()->parameterized_test_registry(). \
+          GetTestCasePatternHolder<test_case_name>(\
+              #test_case_name, \
+              ::testing::internal::CodeLocation(\
+                  __FILE__, __LINE__))->AddTestPattern(\
+                      #test_case_name, \
+                      #test_name, \
+                      new ::testing::internal::TestMetaFactory< \
+                          GTEST_TEST_CLASS_NAME_(\
+                              test_case_name, test_name)>()); \
+      return 0; \
+    } \
+    static int gtest_registering_dummy_ GTEST_ATTRIBUTE_UNUSED_; \
+    GTEST_DISALLOW_COPY_AND_ASSIGN_(\
+        GTEST_TEST_CLASS_NAME_(test_case_name, test_name)); \
+  }; \
+  int GTEST_TEST_CLASS_NAME_(test_case_name, \
+                             test_name)::gtest_registering_dummy_ = \
+      GTEST_TEST_CLASS_NAME_(test_case_name, test_name)::AddToRegistry(); \
+  void GTEST_TEST_CLASS_NAME_(test_case_name, test_name)::TestBody()
+
+// The optional last argument to INSTANTIATE_TEST_CASE_P allows the user
+// to specify a function or functor that generates custom test name suffixes
+// based on the test parameters. The function should accept one argument of
+// type testing::TestParamInfo<class ParamType>, and return std::string.
+//
+// testing::PrintToStringParamName is a builtin test suffix generator that
+// returns the value of testing::PrintToString(GetParam()).
+//
+// Note: test names must be non-empty, unique, and may only contain ASCII
+// alphanumeric characters or underscore. Because PrintToString adds quotes
+// to std::string and C strings, it won't work for these types.
+
+# define INSTANTIATE_TEST_CASE_P(prefix, test_case_name, generator, ...) \
+  ::testing::internal::ParamGenerator<test_case_name::ParamType> \
+      gtest_##prefix##test_case_name##_EvalGenerator_() { return generator; } \
+  ::std::string gtest_##prefix##test_case_name##_EvalGenerateName_( \
+      const ::testing::TestParamInfo<test_case_name::ParamType>& info) { \
+    return ::testing::internal::GetParamNameGen<test_case_name::ParamType> \
+        (__VA_ARGS__)(info); \
+  } \
+  int gtest_##prefix##test_case_name##_dummy_ GTEST_ATTRIBUTE_UNUSED_ = \
+      ::testing::UnitTest::GetInstance()->parameterized_test_registry(). \
+          GetTestCasePatternHolder<test_case_name>(\
+              #test_case_name, \
+              ::testing::internal::CodeLocation(\
+                  __FILE__, __LINE__))->AddTestCaseInstantiation(\
+                      #prefix, \
+                      &gtest_##prefix##test_case_name##_EvalGenerator_, \
+                      &gtest_##prefix##test_case_name##_EvalGenerateName_, \
+                      __FILE__, __LINE__)
+
+}  // namespace testing
+
+#endif  // GTEST_HAS_PARAM_TEST
+
+#endif  // GTEST_INCLUDE_GTEST_GTEST_PARAM_TEST_H_
--- a/third_party/googletest/src/include/gtest/gtest-printers.h
+++ b/third_party/googletest/src/include/gtest/gtest-printers.h
@@ -0,0 +1,993 @@
+// Copyright 2007, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+// Author: wan@google.com (Zhanyong Wan)
+
+// Google Test - The Google C++ Testing Framework
+//
+// This file implements a universal value printer that can print a
+// value of any type T:
+//
+//   void ::testing::internal::UniversalPrinter<T>::Print(value, ostream_ptr);
+//
+// A user can teach this function how to print a class type T by
+// defining either operator<<() or PrintTo() in the namespace that
+// defines T.  More specifically, the FIRST defined function in the
+// following list will be used (assuming T is defined in namespace
+// foo):
+//
+//   1. foo::PrintTo(const T&, ostream*)
+//   2. operator<<(ostream&, const T&) defined in either foo or the
+//      global namespace.
+//
+// If none of the above is defined, it will print the debug string of
+// the value if it is a protocol buffer, or print the raw bytes in the
+// value otherwise.
+//
+// To aid debugging: when T is a reference type, the address of the
+// value is also printed; when T is a (const) char pointer, both the
+// pointer value and the NUL-terminated string it points to are
+// printed.
+//
+// We also provide some convenient wrappers:
+//
+//   // Prints a value to a string.  For a (const or not) char
+//   // pointer, the NUL-terminated string (but not the pointer) is
+//   // printed.
+//   std::string ::testing::PrintToString(const T& value);
+//
+//   // Prints a value tersely: for a reference type, the referenced
+//   // value (but not the address) is printed; for a (const or not) char
+//   // pointer, the NUL-terminated string (but not the pointer) is
+//   // printed.
+//   void ::testing::internal::UniversalTersePrint(const T& value, ostream*);
+//
+//   // Prints value using the type inferred by the compiler.  The difference
+//   // from UniversalTersePrint() is that this function prints both the
+//   // pointer and the NUL-terminated string for a (const or not) char pointer.
+//   void ::testing::internal::UniversalPrint(const T& value, ostream*);
+//
+//   // Prints the fields of a tuple tersely to a string vector, one
+//   // element for each field. Tuple support must be enabled in
+//   // gtest-port.h.
+//   std::vector<string> UniversalTersePrintTupleFieldsToStrings(
+//       const Tuple& value);
+//
+// Known limitation:
+//
+// The print primitives print the elements of an STL-style container
+// using the compiler-inferred type of *iter where iter is a
+// const_iterator of the container.  When const_iterator is an input
+// iterator but not a forward iterator, this inferred type may not
+// match value_type, and the print output may be incorrect.  In
+// practice, this is rarely a problem as for most containers
+// const_iterator is a forward iterator.  We'll fix this if there's an
+// actual need for it.  Note that this fix cannot rely on value_type
+// being defined as many user-defined container types don't have
+// value_type.
+
+#ifndef GTEST_INCLUDE_GTEST_GTEST_PRINTERS_H_
+#define GTEST_INCLUDE_GTEST_GTEST_PRINTERS_H_
+
+#include <ostream>  // NOLINT
+#include <sstream>
+#include <string>
+#include <utility>
+#include <vector>
+#include "gtest/internal/gtest-port.h"
+#include "gtest/internal/gtest-internal.h"
+
+#if GTEST_HAS_STD_TUPLE_
+# include <tuple>
+#endif
+
+namespace testing {
+
+// Definitions in the 'internal' and 'internal2' name spaces are
+// subject to change without notice.  DO NOT USE THEM IN USER CODE!
+namespace internal2 {
+
+// Prints the given number of bytes in the given object to the given
+// ostream.
+GTEST_API_ void PrintBytesInObjectTo(const unsigned char* obj_bytes,
+                                     size_t count,
+                                     ::std::ostream* os);
+
+// For selecting which printer to use when a given type has neither <<
+// nor PrintTo().
+enum TypeKind {
+  kProtobuf,              // a protobuf type
+  kConvertibleToInteger,  // a type implicitly convertible to BiggestInt
+                          // (e.g. a named or unnamed enum type)
+  kOtherType              // anything else
+};
+
+// TypeWithoutFormatter<T, kTypeKind>::PrintValue(value, os) is called
+// by the universal printer to print a value of type T when neither
+// operator<< nor PrintTo() is defined for T, where kTypeKind is the
+// "kind" of T as defined by enum TypeKind.
+template <typename T, TypeKind kTypeKind>
+class TypeWithoutFormatter {
+ public:
+  // This default version is called when kTypeKind is kOtherType.
+  static void PrintValue(const T& value, ::std::ostream* os) {
+    PrintBytesInObjectTo(reinterpret_cast<const unsigned char*>(&value),
+                         sizeof(value), os);
+  }
+};
+
+// We print a protobuf using its ShortDebugString() when the string
+// doesn't exceed this many characters; otherwise we print it using
+// DebugString() for better readability.
+const size_t kProtobufOneLinerMaxLength = 50;
+
+template <typename T>
+class TypeWithoutFormatter<T, kProtobuf> {
+ public:
+  static void PrintValue(const T& value, ::std::ostream* os) {
+    const ::testing::internal::string short_str = value.ShortDebugString();
+    const ::testing::internal::string pretty_str =
+        short_str.length() <= kProtobufOneLinerMaxLength ?
+        short_str : ("\n" + value.DebugString());
+    *os << ("<" + pretty_str + ">");
+  }
+};
+
+template <typename T>
+class TypeWithoutFormatter<T, kConvertibleToInteger> {
+ public:
+  // Since T has no << operator or PrintTo() but can be implicitly
+  // converted to BiggestInt, we print it as a BiggestInt.
+  //
+  // Most likely T is an enum type (either named or unnamed), in which
+  // case printing it as an integer is the desired behavior.  In case
+  // T is not an enum, printing it as an integer is the best we can do
+  // given that it has no user-defined printer.
+  static void PrintValue(const T& value, ::std::ostream* os) {
+    const internal::BiggestInt kBigInt = value;
+    *os << kBigInt;
+  }
+};
+
+// Prints the given value to the given ostream.  If the value is a
+// protocol message, its debug string is printed; if it's an enum or
+// of a type implicitly convertible to BiggestInt, it's printed as an
+// integer; otherwise the bytes in the value are printed.  This is
+// what UniversalPrinter<T>::Print() does when it knows nothing about
+// type T and T has neither << operator nor PrintTo().
+//
+// A user can override this behavior for a class type Foo by defining
+// a << operator in the namespace where Foo is defined.
+//
+// We put this operator in namespace 'internal2' instead of 'internal'
+// to simplify the implementation, as much code in 'internal' needs to
+// use << in STL, which would conflict with our own << were it defined
+// in 'internal'.
+//
+// Note that this operator<< takes a generic std::basic_ostream<Char,
+// CharTraits> type instead of the more restricted std::ostream.  If
+// we define it to take an std::ostream instead, we'll get an
+// "ambiguous overloads" compiler error when trying to print a type
+// Foo that supports streaming to std::basic_ostream<Char,
+// CharTraits>, as the compiler cannot tell whether
+// operator<<(std::ostream&, const T&) or
+// operator<<(std::basic_stream<Char, CharTraits>, const Foo&) is more
+// specific.
+template <typename Char, typename CharTraits, typename T>
+::std::basic_ostream<Char, CharTraits>& operator<<(
+    ::std::basic_ostream<Char, CharTraits>& os, const T& x) {
+  TypeWithoutFormatter<T,
+      (internal::IsAProtocolMessage<T>::value ? kProtobuf :
+       internal::ImplicitlyConvertible<const T&, internal::BiggestInt>::value ?
+       kConvertibleToInteger : kOtherType)>::PrintValue(x, &os);
+  return os;
+}
+
+}  // namespace internal2
+}  // namespace testing
+
+// This namespace MUST NOT BE NESTED IN ::testing, or the name look-up
+// magic needed for implementing UniversalPrinter won't work.
+namespace testing_internal {
+
+// Used to print a value that is not an STL-style container when the
+// user doesn't define PrintTo() for it.
+template <typename T>
+void DefaultPrintNonContainerTo(const T& value, ::std::ostream* os) {
+  // With the following statement, during unqualified name lookup,
+  // testing::internal2::operator<< appears as if it was declared in
+  // the nearest enclosing namespace that contains both
+  // ::testing_internal and ::testing::internal2, i.e. the global
+  // namespace.  For more details, refer to the C++ Standard section
+  // 7.3.4-1 [namespace.udir].  This allows us to fall back onto
+  // testing::internal2::operator<< in case T doesn't come with a <<
+  // operator.
+  //
+  // We cannot write 'using ::testing::internal2::operator<<;', which
+  // gcc 3.3 fails to compile due to a compiler bug.
+  using namespace ::testing::internal2;  // NOLINT
+
+  // Assuming T is defined in namespace foo, in the next statement,
+  // the compiler will consider all of:
+  //
+  //   1. foo::operator<< (thanks to Koenig look-up),
+  //   2. ::operator<< (as the current namespace is enclosed in ::),
+  //   3. testing::internal2::operator<< (thanks to the using statement above).
+  //
+  // The operator<< whose type matches T best will be picked.
+  //
+  // We deliberately allow #2 to be a candidate, as sometimes it's
+  // impossible to define #1 (e.g. when foo is ::std, defining
+  // anything in it is undefined behavior unless you are a compiler
+  // vendor.).
+  *os << value;
+}
+
+}  // namespace testing_internal
+
+namespace testing {
+namespace internal {
+
+// FormatForComparison<ToPrint, OtherOperand>::Format(value) formats a
+// value of type ToPrint that is an operand of a comparison assertion
+// (e.g. ASSERT_EQ).  OtherOperand is the type of the other operand in
+// the comparison, and is used to help determine the best way to
+// format the value.  In particular, when the value is a C string
+// (char pointer) and the other operand is an STL string object, we
+// want to format the C string as a string, since we know it is
+// compared by value with the string object.  If the value is a char
+// pointer but the other operand is not an STL string object, we don't
+// know whether the pointer is supposed to point to a NUL-terminated
+// string, and thus want to print it as a pointer to be safe.
+//
+// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
+
+// The default case.
+template <typename ToPrint, typename OtherOperand>
+class FormatForComparison {
+ public:
+  static ::std::string Format(const ToPrint& value) {
+    return ::testing::PrintToString(value);
+  }
+};
+
+// Array.
+template <typename ToPrint, size_t N, typename OtherOperand>
+class FormatForComparison<ToPrint[N], OtherOperand> {
+ public:
+  static ::std::string Format(const ToPrint* value) {
+    return FormatForComparison<const ToPrint*, OtherOperand>::Format(value);
+  }
+};
+
+// By default, print C string as pointers to be safe, as we don't know
+// whether they actually point to a NUL-terminated string.
+
+#define GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(CharType)                \
+  template <typename OtherOperand>                                      \
+  class FormatForComparison<CharType*, OtherOperand> {                  \
+   public:                                                              \
+    static ::std::string Format(CharType* value) {                      \
+      return ::testing::PrintToString(static_cast<const void*>(value)); \
+    }                                                                   \
+  }
+
+GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(char);
+GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(const char);
+GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(wchar_t);
+GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_(const wchar_t);
+
+#undef GTEST_IMPL_FORMAT_C_STRING_AS_POINTER_
+
+// If a C string is compared with an STL string object, we know it's meant
+// to point to a NUL-terminated string, and thus can print it as a string.
+
+#define GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(CharType, OtherStringType) \
+  template <>                                                           \
+  class FormatForComparison<CharType*, OtherStringType> {               \
+   public:                                                              \
+    static ::std::string Format(CharType* value) {                      \
+      return ::testing::PrintToString(value);                           \
+    }                                                                   \
+  }
+
+GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(char, ::std::string);
+GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(const char, ::std::string);
+
+#if GTEST_HAS_GLOBAL_STRING
+GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(char, ::string);
+GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(const char, ::string);
+#endif
+
+#if GTEST_HAS_GLOBAL_WSTRING
+GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(wchar_t, ::wstring);
+GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(const wchar_t, ::wstring);
+#endif
+
+#if GTEST_HAS_STD_WSTRING
+GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(wchar_t, ::std::wstring);
+GTEST_IMPL_FORMAT_C_STRING_AS_STRING_(const wchar_t, ::std::wstring);
+#endif
+
+#undef GTEST_IMPL_FORMAT_C_STRING_AS_STRING_
+
+// Formats a comparison assertion (e.g. ASSERT_EQ, EXPECT_LT, and etc)
+// operand to be used in a failure message.  The type (but not value)
+// of the other operand may affect the format.  This allows us to
+// print a char* as a raw pointer when it is compared against another
+// char* or void*, and print it as a C string when it is compared
+// against an std::string object, for example.
+//
+// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
+template <typename T1, typename T2>
+std::string FormatForComparisonFailureMessage(
+    const T1& value, const T2& /* other_operand */) {
+  return FormatForComparison<T1, T2>::Format(value);
+}
+
+// UniversalPrinter<T>::Print(value, ostream_ptr) prints the given
+// value to the given ostream.  The caller must ensure that
+// 'ostream_ptr' is not NULL, or the behavior is undefined.
+//
+// We define UniversalPrinter as a class template (as opposed to a
+// function template), as we need to partially specialize it for
+// reference types, which cannot be done with function templates.
+template <typename T>
+class UniversalPrinter;
+
+template <typename T>
+void UniversalPrint(const T& value, ::std::ostream* os);
+
+// Used to print an STL-style container when the user doesn't define
+// a PrintTo() for it.
+template <typename C>
+void DefaultPrintTo(IsContainer /* dummy */,
+                    false_type /* is not a pointer */,
+                    const C& container, ::std::ostream* os) {
+  const size_t kMaxCount = 32;  // The maximum number of elements to print.
+  *os << '{';
+  size_t count = 0;
+  for (typename C::const_iterator it = container.begin();
+       it != container.end(); ++it, ++count) {
+    if (count > 0) {
+      *os << ',';
+      if (count == kMaxCount) {  // Enough has been printed.
+        *os << " ...";
+        break;
+      }
+    }
+    *os << ' ';
+    // We cannot call PrintTo(*it, os) here as PrintTo() doesn't
+    // handle *it being a native array.
+    internal::UniversalPrint(*it, os);
+  }
+
+  if (count > 0) {
+    *os << ' ';
+  }
+  *os << '}';
+}
+
+// Used to print a pointer that is neither a char pointer nor a member
+// pointer, when the user doesn't define PrintTo() for it.  (A member
+// variable pointer or member function pointer doesn't really point to
+// a location in the address space.  Their representation is
+// implementation-defined.  Therefore they will be printed as raw
+// bytes.)
+template <typename T>
+void DefaultPrintTo(IsNotContainer /* dummy */,
+                    true_type /* is a pointer */,
+                    T* p, ::std::ostream* os) {
+  if (p == NULL) {
+    *os << "NULL";
+  } else {
+    // C++ doesn't allow casting from a function pointer to any object
+    // pointer.
+    //
+    // IsTrue() silences warnings: "Condition is always true",
+    // "unreachable code".
+    if (IsTrue(ImplicitlyConvertible<T*, const void*>::value)) {
+      // T is not a function type.  We just call << to print p,
+      // relying on ADL to pick up user-defined << for their pointer
+      // types, if any.
+      *os << p;
+    } else {
+      // T is a function type, so '*os << p' doesn't do what we want
+      // (it just prints p as bool).  We want to print p as a const
+      // void*.  However, we cannot cast it to const void* directly,
+      // even using reinterpret_cast, as earlier versions of gcc
+      // (e.g. 3.4.5) cannot compile the cast when p is a function
+      // pointer.  Casting to UInt64 first solves the problem.
+      *os << reinterpret_cast<const void*>(
+          reinterpret_cast<internal::UInt64>(p));
+    }
+  }
+}
+
+// Used to print a non-container, non-pointer value when the user
+// doesn't define PrintTo() for it.
+template <typename T>
+void DefaultPrintTo(IsNotContainer /* dummy */,
+                    false_type /* is not a pointer */,
+                    const T& value, ::std::ostream* os) {
+  ::testing_internal::DefaultPrintNonContainerTo(value, os);
+}
+
+// Prints the given value using the << operator if it has one;
+// otherwise prints the bytes in it.  This is what
+// UniversalPrinter<T>::Print() does when PrintTo() is not specialized
+// or overloaded for type T.
+//
+// A user can override this behavior for a class type Foo by defining
+// an overload of PrintTo() in the namespace where Foo is defined.  We
+// give the user this option as sometimes defining a << operator for
+// Foo is not desirable (e.g. the coding style may prevent doing it,
+// or there is already a << operator but it doesn't do what the user
+// wants).
+template <typename T>
+void PrintTo(const T& value, ::std::ostream* os) {
+  // DefaultPrintTo() is overloaded.  The type of its first two
+  // arguments determine which version will be picked.  If T is an
+  // STL-style container, the version for container will be called; if
+  // T is a pointer, the pointer version will be called; otherwise the
+  // generic version will be called.
+  //
+  // Note that we check for container types here, prior to we check
+  // for protocol message types in our operator<<.  The rationale is:
+  //
+  // For protocol messages, we want to give people a chance to
+  // override Google Mock's format by defining a PrintTo() or
+  // operator<<.  For STL containers, other formats can be
+  // incompatible with Google Mock's format for the container
+  // elements; therefore we check for container types here to ensure
+  // that our format is used.
+  //
+  // The second argument of DefaultPrintTo() is needed to bypass a bug
+  // in Symbian's C++ compiler that prevents it from picking the right
+  // overload between:
+  //
+  //   PrintTo(const T& x, ...);
+  //   PrintTo(T* x, ...);
+  DefaultPrintTo(IsContainerTest<T>(0), is_pointer<T>(), value, os);
+}
+
+// The following list of PrintTo() overloads tells
+// UniversalPrinter<T>::Print() how to print standard types (built-in
+// types, strings, plain arrays, and pointers).
+
+// Overloads for various char types.
+GTEST_API_ void PrintTo(unsigned char c, ::std::ostream* os);
+GTEST_API_ void PrintTo(signed char c, ::std::ostream* os);
+inline void PrintTo(char c, ::std::ostream* os) {
+  // When printing a plain char, we always treat it as unsigned.  This
+  // way, the output won't be affected by whether the compiler thinks
+  // char is signed or not.
+  PrintTo(static_cast<unsigned char>(c), os);
+}
+
+// Overloads for other simple built-in types.
+inline void PrintTo(bool x, ::std::ostream* os) {
+  *os << (x ? "true" : "false");
+}
+
+// Overload for wchar_t type.
+// Prints a wchar_t as a symbol if it is printable or as its internal
+// code otherwise and also as its decimal code (except for L'\0').
+// The L'\0' char is printed as "L'\\0'". The decimal code is printed
+// as signed integer when wchar_t is implemented by the compiler
+// as a signed type and is printed as an unsigned integer when wchar_t
+// is implemented as an unsigned type.
+GTEST_API_ void PrintTo(wchar_t wc, ::std::ostream* os);
+
+// Overloads for C strings.
+GTEST_API_ void PrintTo(const char* s, ::std::ostream* os);
+inline void PrintTo(char* s, ::std::ostream* os) {
+  PrintTo(ImplicitCast_<const char*>(s), os);
+}
+
+// signed/unsigned char is often used for representing binary data, so
+// we print pointers to it as void* to be safe.
+inline void PrintTo(const signed char* s, ::std::ostream* os) {
+  PrintTo(ImplicitCast_<const void*>(s), os);
+}
+inline void PrintTo(signed char* s, ::std::ostream* os) {
+  PrintTo(ImplicitCast_<const void*>(s), os);
+}
+inline void PrintTo(const unsigned char* s, ::std::ostream* os) {
+  PrintTo(ImplicitCast_<const void*>(s), os);
+}
+inline void PrintTo(unsigned char* s, ::std::ostream* os) {
+  PrintTo(ImplicitCast_<const void*>(s), os);
+}
+
+// MSVC can be configured to define wchar_t as a typedef of unsigned
+// short.  It defines _NATIVE_WCHAR_T_DEFINED when wchar_t is a native
+// type.  When wchar_t is a typedef, defining an overload for const
+// wchar_t* would cause unsigned short* be printed as a wide string,
+// possibly causing invalid memory accesses.
+#if !defined(_MSC_VER) || defined(_NATIVE_WCHAR_T_DEFINED)
+// Overloads for wide C strings
+GTEST_API_ void PrintTo(const wchar_t* s, ::std::ostream* os);
+inline void PrintTo(wchar_t* s, ::std::ostream* os) {
+  PrintTo(ImplicitCast_<const wchar_t*>(s), os);
+}
+#endif
+
+// Overload for C arrays.  Multi-dimensional arrays are printed
+// properly.
+
+// Prints the given number of elements in an array, without printing
+// the curly braces.
+template <typename T>
+void PrintRawArrayTo(const T a[], size_t count, ::std::ostream* os) {
+  UniversalPrint(a[0], os);
+  for (size_t i = 1; i != count; i++) {
+    *os << ", ";
+    UniversalPrint(a[i], os);
+  }
+}
+
+// Overloads for ::string and ::std::string.
+#if GTEST_HAS_GLOBAL_STRING
+GTEST_API_ void PrintStringTo(const ::string&s, ::std::ostream* os);
+inline void PrintTo(const ::string& s, ::std::ostream* os) {
+  PrintStringTo(s, os);
+}
+#endif  // GTEST_HAS_GLOBAL_STRING
+
+GTEST_API_ void PrintStringTo(const ::std::string&s, ::std::ostream* os);
+inline void PrintTo(const ::std::string& s, ::std::ostream* os) {
+  PrintStringTo(s, os);
+}
+
+// Overloads for ::wstring and ::std::wstring.
+#if GTEST_HAS_GLOBAL_WSTRING
+GTEST_API_ void PrintWideStringTo(const ::wstring&s, ::std::ostream* os);
+inline void PrintTo(const ::wstring& s, ::std::ostream* os) {
+  PrintWideStringTo(s, os);
+}
+#endif  // GTEST_HAS_GLOBAL_WSTRING
+
+#if GTEST_HAS_STD_WSTRING
+GTEST_API_ void PrintWideStringTo(const ::std::wstring&s, ::std::ostream* os);
+inline void PrintTo(const ::std::wstring& s, ::std::ostream* os) {
+  PrintWideStringTo(s, os);
+}
+#endif  // GTEST_HAS_STD_WSTRING
+
+#if GTEST_HAS_TR1_TUPLE || GTEST_HAS_STD_TUPLE_
+// Helper function for printing a tuple.  T must be instantiated with
+// a tuple type.
+template <typename T>
+void PrintTupleTo(const T& t, ::std::ostream* os);
+#endif  // GTEST_HAS_TR1_TUPLE || GTEST_HAS_STD_TUPLE_
+
+#if GTEST_HAS_TR1_TUPLE
+// Overload for ::std::tr1::tuple.  Needed for printing function arguments,
+// which are packed as tuples.
+
+// Overloaded PrintTo() for tuples of various arities.  We support
+// tuples of up-to 10 fields.  The following implementation works
+// regardless of whether tr1::tuple is implemented using the
+// non-standard variadic template feature or not.
+
+inline void PrintTo(const ::std::tr1::tuple<>& t, ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1>
+void PrintTo(const ::std::tr1::tuple<T1>& t, ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1, typename T2>
+void PrintTo(const ::std::tr1::tuple<T1, T2>& t, ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1, typename T2, typename T3>
+void PrintTo(const ::std::tr1::tuple<T1, T2, T3>& t, ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1, typename T2, typename T3, typename T4>
+void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4>& t, ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1, typename T2, typename T3, typename T4, typename T5>
+void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5>& t,
+             ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1, typename T2, typename T3, typename T4, typename T5,
+          typename T6>
+void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6>& t,
+             ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1, typename T2, typename T3, typename T4, typename T5,
+          typename T6, typename T7>
+void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6, T7>& t,
+             ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1, typename T2, typename T3, typename T4, typename T5,
+          typename T6, typename T7, typename T8>
+void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6, T7, T8>& t,
+             ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1, typename T2, typename T3, typename T4, typename T5,
+          typename T6, typename T7, typename T8, typename T9>
+void PrintTo(const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6, T7, T8, T9>& t,
+             ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+
+template <typename T1, typename T2, typename T3, typename T4, typename T5,
+          typename T6, typename T7, typename T8, typename T9, typename T10>
+void PrintTo(
+    const ::std::tr1::tuple<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10>& t,
+    ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+#endif  // GTEST_HAS_TR1_TUPLE
+
+#if GTEST_HAS_STD_TUPLE_
+template <typename... Types>
+void PrintTo(const ::std::tuple<Types...>& t, ::std::ostream* os) {
+  PrintTupleTo(t, os);
+}
+#endif  // GTEST_HAS_STD_TUPLE_
+
+// Overload for std::pair.
+template <typename T1, typename T2>
+void PrintTo(const ::std::pair<T1, T2>& value, ::std::ostream* os) {
+  *os << '(';
+  // We cannot use UniversalPrint(value.first, os) here, as T1 may be
+  // a reference type.  The same for printing value.second.
+  UniversalPrinter<T1>::Print(value.first, os);
+  *os << ", ";
+  UniversalPrinter<T2>::Print(value.second, os);
+  *os << ')';
+}
+
+// Implements printing a non-reference type T by letting the compiler
+// pick the right overload of PrintTo() for T.
+template <typename T>
+class UniversalPrinter {
+ public:
+  // MSVC warns about adding const to a function type, so we want to
+  // disable the warning.
+  GTEST_DISABLE_MSC_WARNINGS_PUSH_(4180)
+
+  // Note: we deliberately don't call this PrintTo(), as that name
+  // conflicts with ::testing::internal::PrintTo in the body of the
+  // function.
+  static void Print(const T& value, ::std::ostream* os) {
+    // By default, ::testing::internal::PrintTo() is used for printing
+    // the value.
+    //
+    // Thanks to Koenig look-up, if T is a class and has its own
+    // PrintTo() function defined in its namespace, that function will
+    // be visible here.  Since it is more specific than the generic ones
+    // in ::testing::internal, it will be picked by the compiler in the
+    // following statement - exactly what we want.
+    PrintTo(value, os);
+  }
+
+  GTEST_DISABLE_MSC_WARNINGS_POP_()
+};
+
+// UniversalPrintArray(begin, len, os) prints an array of 'len'
+// elements, starting at address 'begin'.
+template <typename T>
+void UniversalPrintArray(const T* begin, size_t len, ::std::ostream* os) {
+  if (len == 0) {
+    *os << "{}";
+  } else {
+    *os << "{ ";
+    const size_t kThreshold = 18;
+    const size_t kChunkSize = 8;
+    // If the array has more than kThreshold elements, we'll have to
+    // omit some details by printing only the first and the last
+    // kChunkSize elements.
+    // TODO(wan@google.com): let the user control the threshold using a flag.
+    if (len <= kThreshold) {
+      PrintRawArrayTo(begin, len, os);
+    } else {
+      PrintRawArrayTo(begin, kChunkSize, os);
+      *os << ", ..., ";
+      PrintRawArrayTo(begin + len - kChunkSize, kChunkSize, os);
+    }
+    *os << " }";
+  }
+}
+// This overload prints a (const) char array compactly.
+GTEST_API_ void UniversalPrintArray(
+    const char* begin, size_t len, ::std::ostream* os);
+
+// This overload prints a (const) wchar_t array compactly.
+GTEST_API_ void UniversalPrintArray(
+    const wchar_t* begin, size_t len, ::std::ostream* os);
+
+// Implements printing an array type T[N].
+template <typename T, size_t N>
+class UniversalPrinter<T[N]> {
+ public:
+  // Prints the given array, omitting some elements when there are too
+  // many.
+  static void Print(const T (&a)[N], ::std::ostream* os) {
+    UniversalPrintArray(a, N, os);
+  }
+};
+
+// Implements printing a reference type T&.
+template <typename T>
+class UniversalPrinter<T&> {
+ public:
+  // MSVC warns about adding const to a function type, so we want to
+  // disable the warning.
+  GTEST_DISABLE_MSC_WARNINGS_PUSH_(4180)
+
+  static void Print(const T& value, ::std::ostream* os) {
+    // Prints the address of the value.  We use reinterpret_cast here
+    // as static_cast doesn't compile when T is a function type.
+    *os << "@" << reinterpret_cast<const void*>(&value) << " ";
+
+    // Then prints the value itself.
+    UniversalPrint(value, os);
+  }
+
+  GTEST_DISABLE_MSC_WARNINGS_POP_()
+};
+
+// Prints a value tersely: for a reference type, the referenced value
+// (but not the address) is printed; for a (const) char pointer, the
+// NUL-terminated string (but not the pointer) is printed.
+
+template <typename T>
+class UniversalTersePrinter {
+ public:
+  static void Print(const T& value, ::std::ostream* os) {
+    UniversalPrint(value, os);
+  }
+};
+template <typename T>
+class UniversalTersePrinter<T&> {
+ public:
+  static void Print(const T& value, ::std::ostream* os) {
+    UniversalPrint(value, os);
+  }
+};
+template <typename T, size_t N>
+class UniversalTersePrinter<T[N]> {
+ public:
+  static void Print(const T (&value)[N], ::std::ostream* os) {
+    UniversalPrinter<T[N]>::Print(value, os);
+  }
+};
+template <>
+class UniversalTersePrinter<const char*> {
+ public:
+  static void Print(const char* str, ::std::ostream* os) {
+    if (str == NULL) {
+      *os << "NULL";
+    } else {
+      UniversalPrint(string(str), os);
+    }
+  }
+};
+template <>
+class UniversalTersePrinter<char*> {
+ public:
+  static void Print(char* str, ::std::ostream* os) {
+    UniversalTersePrinter<const char*>::Print(str, os);
+  }
+};
+
+#if GTEST_HAS_STD_WSTRING
+template <>
+class UniversalTersePrinter<const wchar_t*> {
+ public:
+  static void Print(const wchar_t* str, ::std::ostream* os) {
+    if (str == NULL) {
+      *os << "NULL";
+    } else {
+      UniversalPrint(::std::wstring(str), os);
+    }
+  }
+};
+#endif
+
+template <>
+class UniversalTersePrinter<wchar_t*> {
+ public:
+  static void Print(wchar_t* str, ::std::ostream* os) {
+    UniversalTersePrinter<const wchar_t*>::Print(str, os);
+  }
+};
+
+template <typename T>
+void UniversalTersePrint(const T& value, ::std::ostream* os) {
+  UniversalTersePrinter<T>::Print(value, os);
+}
+
+// Prints a value using the type inferred by the compiler.  The
+// difference between this and UniversalTersePrint() is that for a
+// (const) char pointer, this prints both the pointer and the
+// NUL-terminated string.
+template <typename T>
+void UniversalPrint(const T& value, ::std::ostream* os) {
+  // A workarond for the bug in VC++ 7.1 that prevents us from instantiating
+  // UniversalPrinter with T directly.
+  typedef T T1;
+  UniversalPrinter<T1>::Print(value, os);
+}
+
+typedef ::std::vector<string> Strings;
+
+// TuplePolicy<TupleT> must provide:
+// - tuple_size
+//     size of tuple TupleT.
+// - get<size_t I>(const TupleT& t)
+//     static function extracting element I of tuple TupleT.
+// - tuple_element<size_t I>::type
+//     type of element I of tuple TupleT.
+template <typename TupleT>
+struct TuplePolicy;
+
+#if GTEST_HAS_TR1_TUPLE
+template <typename TupleT>
+struct TuplePolicy {
+  typedef TupleT Tuple;
+  static const size_t tuple_size = ::std::tr1::tuple_size<Tuple>::value;
+
+  template <size_t I>
+  struct tuple_element : ::std::tr1::tuple_element<I, Tuple> {};
+
+  template <size_t I>
+  static typename AddReference<
+      const typename ::std::tr1::tuple_element<I, Tuple>::type>::type get(
+      const Tuple& tuple) {
+    return ::std::tr1::get<I>(tuple);
+  }
+};
+template <typename TupleT>
+const size_t TuplePolicy<TupleT>::tuple_size;
+#endif  // GTEST_HAS_TR1_TUPLE
+
+#if GTEST_HAS_STD_TUPLE_
+template <typename... Types>
+struct TuplePolicy< ::std::tuple<Types...> > {
+  typedef ::std::tuple<Types...> Tuple;
+  static const size_t tuple_size = ::std::tuple_size<Tuple>::value;
+
+  template <size_t I>
+  struct tuple_element : ::std::tuple_element<I, Tuple> {};
+
+  template <size_t I>
+  static const typename ::std::tuple_element<I, Tuple>::type& get(
+      const Tuple& tuple) {
+    return ::std::get<I>(tuple);
+  }
+};
+template <typename... Types>
+const size_t TuplePolicy< ::std::tuple<Types...> >::tuple_size;
+#endif  // GTEST_HAS_STD_TUPLE_
+
+#if GTEST_HAS_TR1_TUPLE || GTEST_HAS_STD_TUPLE_
+// This helper template allows PrintTo() for tuples and
+// UniversalTersePrintTupleFieldsToStrings() to be defined by
+// induction on the number of tuple fields.  The idea is that
+// TuplePrefixPrinter<N>::PrintPrefixTo(t, os) prints the first N
+// fields in tuple t, and can be defined in terms of
+// TuplePrefixPrinter<N - 1>.
+//
+// The inductive case.
+template <size_t N>
+struct TuplePrefixPrinter {
+  // Prints the first N fields of a tuple.
+  template <typename Tuple>
+  static void PrintPrefixTo(const Tuple& t, ::std::ostream* os) {
+    TuplePrefixPrinter<N - 1>::PrintPrefixTo(t, os);
+    GTEST_INTENTIONAL_CONST_COND_PUSH_()
+    if (N > 1) {
+    GTEST_INTENTIONAL_CONST_COND_POP_()
+      *os << ", ";
+    }
+    UniversalPrinter<
+        typename TuplePolicy<Tuple>::template tuple_element<N - 1>::type>
+        ::Print(TuplePolicy<Tuple>::template get<N - 1>(t), os);
+  }
+
+  // Tersely prints the first N fields of a tuple to a string vector,
+  // one element for each field.
+  template <typename Tuple>
+  static void TersePrintPrefixToStrings(const Tuple& t, Strings* strings) {
+    TuplePrefixPrinter<N - 1>::TersePrintPrefixToStrings(t, strings);
+    ::std::stringstream ss;
+    UniversalTersePrint(TuplePolicy<Tuple>::template get<N - 1>(t), &ss);
+    strings->push_back(ss.str());
+  }
+};
+
+// Base case.
+template <>
+struct TuplePrefixPrinter<0> {
+  template <typename Tuple>
+  static void PrintPrefixTo(const Tuple&, ::std::ostream*) {}
+
+  template <typename Tuple>
+  static void TersePrintPrefixToStrings(const Tuple&, Strings*) {}
+};
+
+// Helper function for printing a tuple.
+// Tuple must be either std::tr1::tuple or std::tuple type.
+template <typename Tuple>
+void PrintTupleTo(const Tuple& t, ::std::ostream* os) {
+  *os << "(";
+  TuplePrefixPrinter<TuplePolicy<Tuple>::tuple_size>::PrintPrefixTo(t, os);
+  *os << ")";
+}
+
+// Prints the fields of a tuple tersely to a string vector, one
+// element for each field.  See the comment before
+// UniversalTersePrint() for how we define "tersely".
+template <typename Tuple>
+Strings UniversalTersePrintTupleFieldsToStrings(const Tuple& value) {
+  Strings result;
+  TuplePrefixPrinter<TuplePolicy<Tuple>::tuple_size>::
+      TersePrintPrefixToStrings(value, &result);
+  return result;
+}
+#endif  // GTEST_HAS_TR1_TUPLE || GTEST_HAS_STD_TUPLE_
+
+}  // namespace internal
+
+template <typename T>
+::std::string PrintToString(const T& value) {
+  ::std::stringstream ss;
+  internal::UniversalTersePrinter<T>::Print(value, &ss);
+  return ss.str();
+}
+
+}  // namespace testing
+
+// Include any custom printer added by the local installation.
+// We must include this header at the end to make sure it can use the
+// declarations from this file.
+#include "gtest/internal/custom/gtest-printers.h"
+
+#endif  // GTEST_INCLUDE_GTEST_GTEST_PRINTERS_H_
--- a/third_party/googletest/src/include/gtest/gtest-spi.h
+++ b/third_party/googletest/src/include/gtest/gtest-spi.h
@@ -0,0 +1,232 @@
+// Copyright 2007, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+// Author: wan@google.com (Zhanyong Wan)
+//
+// Utilities for testing Google Test itself and code that uses Google Test
+// (e.g. frameworks built on top of Google Test).
+
+#ifndef GTEST_INCLUDE_GTEST_GTEST_SPI_H_
+#define GTEST_INCLUDE_GTEST_GTEST_SPI_H_
+
+#include "gtest/gtest.h"
+
+namespace testing {
+
+// This helper class can be used to mock out Google Test failure reporting
+// so that we can test Google Test or code that builds on Google Test.
+//
+// An object of this class appends a TestPartResult object to the
+// TestPartResultArray object given in the constructor whenever a Google Test
+// failure is reported. It can either intercept only failures that are
+// generated in the same thread that created this object or it can intercept
+// all generated failures. The scope of this mock object can be controlled with
+// the second argument to the two arguments constructor.
+class GTEST_API_ ScopedFakeTestPartResultReporter
+    : public TestPartResultReporterInterface {
+ public:
+  // The two possible mocking modes of this object.
+  enum InterceptMode {
+    INTERCEPT_ONLY_CURRENT_THREAD,  // Intercepts only thread local failures.
+    INTERCEPT_ALL_THREADS           // Intercepts all failures.
+  };
+
+  // The c'tor sets this object as the test part result reporter used
+  // by Google Test.  The 'result' parameter specifies where to report the
+  // results. This reporter will only catch failures generated in the current
+  // thread. DEPRECATED
+  explicit ScopedFakeTestPartResultReporter(TestPartResultArray* result);
+
+  // Same as above, but you can choose the interception scope of this object.
+  ScopedFakeTestPartResultReporter(InterceptMode intercept_mode,
+                                   TestPartResultArray* result);
+
+  // The d'tor restores the previous test part result reporter.
+  virtual ~ScopedFakeTestPartResultReporter();
+
+  // Appends the TestPartResult object to the TestPartResultArray
+  // received in the constructor.
+  //
+  // This method is from the TestPartResultReporterInterface
+  // interface.
+  virtual void ReportTestPartResult(const TestPartResult& result);
+ private:
+  void Init();
+
+  const InterceptMode intercept_mode_;
+  TestPartResultReporterInterface* old_reporter_;
+  TestPartResultArray* const result_;
+
+  GTEST_DISALLOW_COPY_AND_ASSIGN_(ScopedFakeTestPartResultReporter);
+};
+
+namespace internal {
+
+// A helper class for implementing EXPECT_FATAL_FAILURE() and
+// EXPECT_NONFATAL_FAILURE().  Its destructor verifies that the given
+// TestPartResultArray contains exactly one failure that has the given
+// type and contains the given substring.  If that's not the case, a
+// non-fatal failure will be generated.
+class GTEST_API_ SingleFailureChecker {
+ public:
+  // The constructor remembers the arguments.
+  SingleFailureChecker(const TestPartResultArray* results,
+                       TestPartResult::Type type,
+                       const string& substr);
+  ~SingleFailureChecker();
+ private:
+  const TestPartResultArray* const results_;
+  const TestPartResult::Type type_;
+  const string substr_;
+
+  GTEST_DISALLOW_COPY_AND_ASSIGN_(SingleFailureChecker);
+};
+
+}  // namespace internal
+
+}  // namespace testing
+
+// A set of macros for testing Google Test assertions or code that's expected
+// to generate Google Test fatal failures.  It verifies that the given
+// statement will cause exactly one fatal Google Test failure with 'substr'
+// being part of the failure message.
+//
+// There are two different versions of this macro. EXPECT_FATAL_FAILURE only
+// affects and considers failures generated in the current thread and
+// EXPECT_FATAL_FAILURE_ON_ALL_THREADS does the same but for all threads.
+//
+// The verification of the assertion is done correctly even when the statement
+// throws an exception or aborts the current function.
+//
+// Known restrictions:
+//   - 'statement' cannot reference local non-static variables or
+//     non-static members of the current object.
+//   - 'statement' cannot return a value.
+//   - You cannot stream a failure message to this macro.
+//
+// Note that even though the implementations of the following two
+// macros are much alike, we cannot refactor them to use a common
+// helper macro, due to some peculiarity in how the preprocessor
+// works.  The AcceptsMacroThatExpandsToUnprotectedComma test in
+// gtest_unittest.cc will fail to compile if we do that.
+#define EXPECT_FATAL_FAILURE(statement, substr) \
+  do { \
+    class GTestExpectFatalFailureHelper {\
+     public:\
+      static void Execute() { statement; }\
+    };\
+    ::testing::TestPartResultArray gtest_failures;\
+    ::testing::internal::SingleFailureChecker gtest_checker(\
+        &gtest_failures, ::testing::TestPartResult::kFatalFailure, (substr));\
+    {\
+      ::testing::ScopedFakeTestPartResultReporter gtest_reporter(\
+          ::testing::ScopedFakeTestPartResultReporter:: \
+          INTERCEPT_ONLY_CURRENT_THREAD, &gtest_failures);\
+      GTestExpectFatalFailureHelper::Execute();\
+    }\
+  } while (::testing::internal::AlwaysFalse())
+
+#define EXPECT_FATAL_FAILURE_ON_ALL_THREADS(statement, substr) \
+  do { \
+    class GTestExpectFatalFailureHelper {\
+     public:\
+      static void Execute() { statement; }\
+    };\
+    ::testing::TestPartResultArray gtest_failures;\
+    ::testing::internal::SingleFailureChecker gtest_checker(\
+        &gtest_failures, ::testing::TestPartResult::kFatalFailure, (substr));\
+    {\
+      ::testing::ScopedFakeTestPartResultReporter gtest_reporter(\
+          ::testing::ScopedFakeTestPartResultReporter:: \
+          INTERCEPT_ALL_THREADS, &gtest_failures);\
+      GTestExpectFatalFailureHelper::Execute();\
+    }\
+  } while (::testing::internal::AlwaysFalse())
+
+// A macro for testing Google Test assertions or code that's expected to
+// generate Google Test non-fatal failures.  It asserts that the given
+// statement will cause exactly one non-fatal Google Test failure with 'substr'
+// being part of the failure message.
+//
+// There are two different versions of this macro. EXPECT_NONFATAL_FAILURE only
+// affects and considers failures generated in the current thread and
+// EXPECT_NONFATAL_FAILURE_ON_ALL_THREADS does the same but for all threads.
+//
+// 'statement' is allowed to reference local variables and members of
+// the current object.
+//
+// The verification of the assertion is done correctly even when the statement
+// throws an exception or aborts the current function.
+//
+// Known restrictions:
+//   - You cannot stream a failure message to this macro.
+//
+// Note that even though the implementations of the following two
+// macros are much alike, we cannot refactor them to use a common
+// helper macro, due to some peculiarity in how the preprocessor
+// works.  If we do that, the code won't compile when the user gives
+// EXPECT_NONFATAL_FAILURE() a statement that contains a macro that
+// expands to code containing an unprotected comma.  The
+// AcceptsMacroThatExpandsToUnprotectedComma test in gtest_unittest.cc
+// catches that.
+//
+// For the same reason, we have to write
+//   if (::testing::internal::AlwaysTrue()) { statement; }
+// instead of
+//   GTEST_SUPPRESS_UNREACHABLE_CODE_WARNING_BELOW_(statement)
+// to avoid an MSVC warning on unreachable code.
+#define EXPECT_NONFATAL_FAILURE(statement, substr) \
+  do {\
+    ::testing::TestPartResultArray gtest_failures;\
+    ::testing::internal::SingleFailureChecker gtest_checker(\
+        &gtest_failures, ::testing::TestPartResult::kNonFatalFailure, \
+        (substr));\
+    {\
+      ::testing::ScopedFakeTestPartResultReporter gtest_reporter(\
+          ::testing::ScopedFakeTestPartResultReporter:: \
+          INTERCEPT_ONLY_CURRENT_THREAD, &gtest_failures);\
+      if (::testing::internal::AlwaysTrue()) { statement; }\
+    }\
+  } while (::testing::internal::AlwaysFalse())
+
+#define EXPECT_NONFATAL_FAILURE_ON_ALL_THREADS(statement, substr) \
+  do {\
+    ::testing::TestPartResultArray gtest_failures;\
+    ::testing::internal::SingleFailureChecker gtest_checker(\
+        &gtest_failures, ::testing::TestPartResult::kNonFatalFailure, \
+        (substr));\
+    {\
+      ::testing::ScopedFakeTestPartResultReporter gtest_reporter(\
+          ::testing::ScopedFakeTestPartResultReporter::INTERCEPT_ALL_THREADS, \
+          &gtest_failures);\
+      if (::testing::internal::AlwaysTrue()) { statement; }\
+    }\
+  } while (::testing::internal::AlwaysFalse())
+
+#endif  // GTEST_INCLUDE_GTEST_GTEST_SPI_H_
--- a/third_party/googletest/src/include/gtest/gtest-test-part.h
+++ b/third_party/googletest/src/include/gtest/gtest-test-part.h
@@ -0,0 +1,179 @@
+// Copyright 2008, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+// Author: mheule@google.com (Markus Heule)
+//
+
+#ifndef GTEST_INCLUDE_GTEST_GTEST_TEST_PART_H_
+#define GTEST_INCLUDE_GTEST_GTEST_TEST_PART_H_
+
+#include <iosfwd>
+#include <vector>
+#include "gtest/internal/gtest-internal.h"
+#include "gtest/internal/gtest-string.h"
+
+namespace testing {
+
+// A copyable object representing the result of a test part (i.e. an
+// assertion or an explicit FAIL(), ADD_FAILURE(), or SUCCESS()).
+//
+// Don't inherit from TestPartResult as its destructor is not virtual.
+class GTEST_API_ TestPartResult {
+ public:
+  // The possible outcomes of a test part (i.e. an assertion or an
+  // explicit SUCCEED(), FAIL(), or ADD_FAILURE()).
+  enum Type {
+    kSuccess,          // Succeeded.
+    kNonFatalFailure,  // Failed but the test can continue.
+    kFatalFailure      // Failed and the test should be terminated.
+  };
+
+  // C'tor.  TestPartResult does NOT have a default constructor.
+  // Always use this constructor (with parameters) to create a
+  // TestPartResult object.
+  TestPartResult(Type a_type,
+                 const char* a_file_name,
+                 int a_line_number,
+                 const char* a_message)
+      : type_(a_type),
+        file_name_(a_file_name == NULL ? "" : a_file_name),
+        line_number_(a_line_number),
+        summary_(ExtractSummary(a_message)),
+        message_(a_message) {
+  }
+
+  // Gets the outcome of the test part.
+  Type type() const { return type_; }
+
+  // Gets the name of the source file where the test part took place, or
+  // NULL if it's unknown.
+  const char* file_name() const {
+    return file_name_.empty() ? NULL : file_name_.c_str();
+  }
+
+  // Gets the line in the source file where the test part took place,
+  // or -1 if it's unknown.
+  int line_number() const { return line_number_; }
+
+  // Gets the summary of the failure message.
+  const char* summary() const { return summary_.c_str(); }
+
+  // Gets the message associated with the test part.
+  const char* message() const { return message_.c_str(); }
+
+  // Returns true iff the test part passed.
+  bool passed() const { return type_ == kSuccess; }
+
+  // Returns true iff the test part failed.
+  bool failed() const { return type_ != kSuccess; }
+
+  // Returns true iff the test part non-fatally failed.
+  bool nonfatally_failed() const { return type_ == kNonFatalFailure; }
+
+  // Returns true iff the test part fatally failed.
+  bool fatally_failed() const { return type_ == kFatalFailure; }
+
+ private:
+  Type type_;
+
+  // Gets the summary of the failure message by omitting the stack
+  // trace in it.
+  static std::string ExtractSummary(const char* message);
+
+  // The name of the source file where the test part took place, or
+  // "" if the source file is unknown.
+  std::string file_name_;
+  // The line in the source file where the test part took place, or -1
+  // if the line number is unknown.
+  int line_number_;
+  std::string summary_;  // The test failure summary.
+  std::string message_;  // The test failure message.
+};
+
+// Prints a TestPartResult object.
+std::ostream& operator<<(std::ostream& os, const TestPartResult& result);
+
+// An array of TestPartResult objects.
+//
+// Don't inherit from TestPartResultArray as its destructor is not
+// virtual.
+class GTEST_API_ TestPartResultArray {
+ public:
+  TestPartResultArray() {}
+
+  // Appends the given TestPartResult to the array.
+  void Append(const TestPartResult& result);
+
+  // Returns the TestPartResult at the given index (0-based).
+  const TestPartResult& GetTestPartResult(int index) const;
+
+  // Returns the number of TestPartResult objects in the array.
+  int size() const;
+
+ private:
+  std::vector<TestPartResult> array_;
+
+  GTEST_DISALLOW_COPY_AND_ASSIGN_(TestPartResultArray);
+};
+
+// This interface knows how to report a test part result.
+class TestPartResultReporterInterface {
+ public:
+  virtual ~TestPartResultReporterInterface() {}
+
+  virtual void ReportTestPartResult(const TestPartResult& result) = 0;
+};
+
+namespace internal {
+
+// This helper class is used by {ASSERT|EXPECT}_NO_FATAL_FAILURE to check if a
+// statement generates new fatal failures. To do so it registers itself as the
+// current test part result reporter. Besides checking if fatal failures were
+// reported, it only delegates the reporting to the former result reporter.
+// The original result reporter is restored in the destructor.
+// INTERNAL IMPLEMENTATION - DO NOT USE IN A USER PROGRAM.
+class GTEST_API_ HasNewFatalFailureHelper
+    : public TestPartResultReporterInterface {
+ public:
+  HasNewFatalFailureHelper();
+  virtual ~HasNewFatalFailureHelper();
+  virtual void ReportTestPartResult(const TestPartResult& result);
+  bool has_new_fatal_failure() const { return has_new_fatal_failure_; }
+ private:
+  bool has_new_fatal_failure_;
+  TestPartResultReporterInterface* original_reporter_;
+
+  GTEST_DISALLOW_COPY_AND_ASSIGN_(HasNewFatalFailureHelper);
+};
+
+}  // namespace internal
+
+}  // namespace testing
+
+#endif  // GTEST_INCLUDE_GTEST_GTEST_TEST_PART_H_
--- a/third_party/googletest/src/include/gtest/gtest-typed-test.h
+++ b/third_party/googletest/src/include/gtest/gtest-typed-test.h
@@ -0,0 +1,263 @@
+// Copyright 2008 Google Inc.
+// All Rights Reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+// Author: wan@google.com (Zhanyong Wan)
+
+#ifndef GTEST_INCLUDE_GTEST_GTEST_TYPED_TEST_H_
+#define GTEST_INCLUDE_GTEST_GTEST_TYPED_TEST_H_
+
+// This header implements typed tests and type-parameterized tests.
+
+// Typed (aka type-driven) tests repeat the same test for types in a
+// list.  You must know which types you want to test with when writing
+// typed tests. Here's how you do it:
+
+#if 0
+
+// First, define a fixture class template.  It should be parameterized
+// by a type.  Remember to derive it from testing::Test.
+template <typename T>
+class FooTest : public testing::Test {
+ public:
+  ...
+  typedef std::list<T> List;
+  static T shared_;
+  T value_;
+};
+
+// Next, associate a list of types with the test case, which will be
+// repeated for each type in the list.  The typedef is necessary for
+// the macro to parse correctly.
+typedef testing::Types<char, int, unsigned int> MyTypes;
+TYPED_TEST_CASE(FooTest, MyTypes);
+
+// If the type list contains only one type, you can write that type
+// directly without Types<...>:
+//   TYPED_TEST_CASE(FooTest, int);
+
+// Then, use TYPED_TEST() instead of TEST_F() to define as many typed
+// tests for this test case as you want.
+TYPED_TEST(FooTest, DoesBlah) {
+  // Inside a test, refer to TypeParam to get the type parameter.
+  // Since we are inside a derived class template, C++ requires use to
+  // visit the members of FooTest via 'this'.
+  TypeParam n = this->value_;
+
+  // To visit static members of the fixture, add the TestFixture::
+  // prefix.
+  n += TestFixture::shared_;
+
+  // To refer to typedefs in the fixture, add the "typename
+  // TestFixture::" prefix.
+  typename TestFixture::List values;
+  values.push_back(n);
+  ...
+}
+
+TYPED_TEST(FooTest, HasPropertyA) { ... }
+
+#endif  // 0
+
+// Type-parameterized tests are abstract test patterns parameterized
+// by a type.  Compared with typed tests, type-parameterized tests
+// allow you to define the test pattern without knowing what the type
+// parameters are.  The defined pattern can be instantiated with
+// different types any number of times, in any number of translation
+// units.
+//
+// If you are designing an interface or concept, you can define a
+// suite of type-parameterized tests to verify properties that any
+// valid implementation of the interface/concept should have.  Then,
+// each implementation can easily instantiate the test suite to verify
+// that it conforms to the requirements, without having to write
+// similar tests repeatedly.  Here's an example:
+
+#if 0
+
+// First, define a fixture class template.  It should be parameterized
+// by a type.  Remember to derive it from testing::Test.
+template <typename T>
+class FooTest : public testing::Test {
+  ...
+};
+
+// Next, declare that you will define a type-parameterized test case
+// (the _P suffix is for "parameterized" or "pattern", whichever you
+// prefer):
+TYPED_TEST_CASE_P(FooTest);
+
+// Then, use TYPED_TEST_P() to define as many type-parameterized tests
+// for this type-parameterized test case as you want.
+TYPED_TEST_P(FooTest, DoesBlah) {
+  // Inside a test, refer to TypeParam to get the type parameter.
+  TypeParam n = 0;
+  ...
+}
+
+TYPED_TEST_P(FooTest, HasPropertyA) { ... }
+
+// Now the tricky part: you need to register all test patterns before
+// you can instantiate them.  The first argument of the macro is the
+// test case name; the rest are the names of the tests in this test
+// case.
+REGISTER_TYPED_TEST_CASE_P(FooTest,
+                           DoesBlah, HasPropertyA);
+
+// Finally, you are free to instantiate the pattern with the types you
+// want.  If you put the above code in a header file, you can #include
+// it in multiple C++ source files and instantiate it multiple times.
+//
+// To distinguish different instances of the pattern, the first
+// argument to the INSTANTIATE_* macro is a prefix that will be added
+// to the actual test case name.  Remember to pick unique prefixes for
+// different instances.
+typedef testing::Types<char, int, unsigned int> MyTypes;
+INSTANTIATE_TYPED_TEST_CASE_P(My, FooTest, MyTypes);
+
+// If the type list contains only one type, you can write that type
+// directly without Types<...>:
+//   INSTANTIATE_TYPED_TEST_CASE_P(My, FooTest, int);
+
+#endif  // 0
+
+#include "gtest/internal/gtest-port.h"
+#include "gtest/internal/gtest-type-util.h"
+
+// Implements typed tests.
+
+#if GTEST_HAS_TYPED_TEST
+
+// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
+//
+// Expands to the name of the typedef for the type parameters of the
+// given test case.
+# define GTEST_TYPE_PARAMS_(TestCaseName) gtest_type_params_##TestCaseName##_
+
+// The 'Types' template argument below must have spaces around it
+// since some compilers may choke on '>>' when passing a template
+// instance (e.g. Types<int>)
+# define TYPED_TEST_CASE(CaseName, Types) \
+  typedef ::testing::internal::TypeList< Types >::type \
+      GTEST_TYPE_PARAMS_(CaseName)
+
+# define TYPED_TEST(CaseName, TestName) \
+  template <typename gtest_TypeParam_> \
+  class GTEST_TEST_CLASS_NAME_(CaseName, TestName) \
+      : public CaseName<gtest_TypeParam_> { \
+   private: \
+    typedef CaseName<gtest_TypeParam_> TestFixture; \
+    typedef gtest_TypeParam_ TypeParam; \
+    virtual void TestBody(); \
+  }; \
+  bool gtest_##CaseName##_##TestName##_registered_ GTEST_ATTRIBUTE_UNUSED_ = \
+      ::testing::internal::TypeParameterizedTest< \
+          CaseName, \
+          ::testing::internal::TemplateSel< \
+              GTEST_TEST_CLASS_NAME_(CaseName, TestName)>, \
+          GTEST_TYPE_PARAMS_(CaseName)>::Register(\
+              "", ::testing::internal::CodeLocation(__FILE__, __LINE__), \
+              #CaseName, #TestName, 0); \
+  template <typename gtest_TypeParam_> \
+  void GTEST_TEST_CLASS_NAME_(CaseName, TestName)<gtest_TypeParam_>::TestBody()
+
+#endif  // GTEST_HAS_TYPED_TEST
+
+// Implements type-parameterized tests.
+
+#if GTEST_HAS_TYPED_TEST_P
+
+// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
+//
+// Expands to the namespace name that the type-parameterized tests for
+// the given type-parameterized test case are defined in.  The exact
+// name of the namespace is subject to change without notice.
+# define GTEST_CASE_NAMESPACE_(TestCaseName) \
+  gtest_case_##TestCaseName##_
+
+// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE.
+//
+// Expands to the name of the variable used to remember the names of
+// the defined tests in the given test case.
+# define GTEST_TYPED_TEST_CASE_P_STATE_(TestCaseName) \
+  gtest_typed_test_case_p_state_##TestCaseName##_
+
+// INTERNAL IMPLEMENTATION - DO NOT USE IN USER CODE DIRECTLY.
+//
+// Expands to the name of the variable used to remember the names of
+// the registered tests in the given test case.
+# define GTEST_REGISTERED_TEST_NAMES_(TestCaseName) \
+  gtest_registered_test_names_##TestCaseName##_
+
+// The variables defined in the type-parameterized test macros are
+// static as typically these macros are used in a .h file that can be
+// #included in multiple translation units linked together.
+# define TYPED_TEST_CASE_P(CaseName) \
+  static ::testing::internal::TypedTestCasePState \
+      GTEST_TYPED_TEST_CASE_P_STATE_(CaseName)
+
+# define TYPED_TEST_P(CaseName, TestName) \
+  namespace GTEST_CASE_NAMESPACE_(CaseName) { \
+  template <typename gtest_TypeParam_> \
+  class TestName : public CaseName<gtest_TypeParam_> { \
+   private: \
+    typedef CaseName<gtest_TypeParam_> TestFixture; \
+    typedef gtest_TypeParam_ TypeParam; \
+    virtual void TestBody(); \
+  }; \
+  static bool gtest_##TestName##_defined_ GTEST_ATTRIBUTE_UNUSED_ = \
+      GTEST_TYPED_TEST_CASE_P_STATE_(CaseName).AddTestName(\
+          __FILE__, __LINE__, #CaseName, #TestName); \
+  } \
+  template <typename gtest_TypeParam_> \
+  void GTEST_CASE_NAMESPACE_(CaseName)::TestName<gtest_TypeParam_>::TestBody()
+
+# define REGISTER_TYPED_TEST_CASE_P(CaseName, ...) \
+  namespace GTEST_CASE_NAMESPACE_(CaseName) { \
+  typedef ::testing::internal::Templates<__VA_ARGS__>::type gtest_AllTests_; \
+  } \
+  static const char* const GTEST_REGISTERED_TEST_NAMES_(CaseName) = \
+      GTEST_TYPED_TEST_CASE_P_STATE_(CaseName).VerifyRegisteredTestNames(\
+          __FILE__, __LINE__, #__VA_ARGS__)
+
+// The 'Types' template argument below must have spaces around it
+// since some compilers may choke on '>>' when passing a template
+// instance (e.g. Types<int>)
+# define INSTANTIATE_TYPED_TEST_CASE_P(Prefix, CaseName, Types) \
+  bool gtest_##Prefix##_##CaseName GTEST_ATTRIBUTE_UNUSED_ = \
+      ::testing::internal::TypeParameterizedTestCase<CaseName, \
+          GTEST_CASE_NAMESPACE_(CaseName)::gtest_AllTests_, \
+          ::testing::internal::TypeList< Types >::type>::Register(\
+              #Prefix, \
+              ::testing::internal::CodeLocation(__FILE__, __LINE__), \
+              &GTEST_TYPED_TEST_CASE_P_STATE_(CaseName), \
+              #CaseName, GTEST_REGISTERED_TEST_NAMES_(CaseName))
+
+#endif  // GTEST_HAS_TYPED_TEST_P
+
+#endif  // GTEST_INCLUDE_GTEST_GTEST_TYPED_TEST_H_
--- a/third_party/googletest/src/include/gtest/gtest.h
+++ b/third_party/googletest/src/include/gtest/gtest.h
--- a/third_party/googletest/src/include/gtest/gtest_pred_impl.h
+++ b/third_party/googletest/src/include/gtest/gtest_pred_impl.h
@@ -0,0 +1,358 @@
+// Copyright 2006, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// This file is AUTOMATICALLY GENERATED on 10/31/2011 by command
+// 'gen_gtest_pred_impl.py 5'.  DO NOT EDIT BY HAND!
+//
+// Implements a family of generic predicate assertion macros.
+
+#ifndef GTEST_INCLUDE_GTEST_GTEST_PRED_IMPL_H_
+#define GTEST_INCLUDE_GTEST_GTEST_PRED_IMPL_H_
+
+// Makes sure this header is not included before gtest.h.
+#ifndef GTEST_INCLUDE_GTEST_GTEST_H_
+# error Do not include gtest_pred_impl.h directly.  Include gtest.h instead.
+#endif  // GTEST_INCLUDE_GTEST_GTEST_H_
+
+// This header implements a family of generic predicate assertion
+// macros:
+//
+//   ASSERT_PRED_FORMAT1(pred_format, v1)
+//   ASSERT_PRED_FORMAT2(pred_format, v1, v2)
+//   ...
+//
+// where pred_format is a function or functor that takes n (in the
+// case of ASSERT_PRED_FORMATn) values and their source expression
+// text, and returns a testing::AssertionResult.  See the definition
+// of ASSERT_EQ in gtest.h for an example.
+//
+// If you don't care about formatting, you can use the more
+// restrictive version:
+//
+//   ASSERT_PRED1(pred, v1)
+//   ASSERT_PRED2(pred, v1, v2)
+//   ...
+//
+// where pred is an n-ary function or functor that returns bool,
+// and the values v1, v2, ..., must support the << operator for
+// streaming to std::ostream.
+//
+// We also define the EXPECT_* variations.
+//
+// For now we only support predicates whose arity is at most 5.
+// Please email googletestframework@googlegroups.com if you need
+// support for higher arities.
+
+// GTEST_ASSERT_ is the basic statement to which all of the assertions
+// in this file reduce.  Don't use this in your code.
+
+#define GTEST_ASSERT_(expression, on_failure) \
+  GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
+  if (const ::testing::AssertionResult gtest_ar = (expression)) \
+    ; \
+  else \
+    on_failure(gtest_ar.failure_message())
+
+
+// Helper function for implementing {EXPECT|ASSERT}_PRED1.  Don't use
+// this in your code.
+template <typename Pred,
+          typename T1>
+AssertionResult AssertPred1Helper(const char* pred_text,
+                                  const char* e1,
+                                  Pred pred,
+                                  const T1& v1) {
+  if (pred(v1)) return AssertionSuccess();
+
+  return AssertionFailure() << pred_text << "("
+                            << e1 << ") evaluates to false, where"
+                            << "\n" << e1 << " evaluates to " << v1;
+}
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT1.
+// Don't use this in your code.
+#define GTEST_PRED_FORMAT1_(pred_format, v1, on_failure)\
+  GTEST_ASSERT_(pred_format(#v1, v1), \
+                on_failure)
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED1.  Don't use
+// this in your code.
+#define GTEST_PRED1_(pred, v1, on_failure)\
+  GTEST_ASSERT_(::testing::AssertPred1Helper(#pred, \
+                                             #v1, \
+                                             pred, \
+                                             v1), on_failure)
+
+// Unary predicate assertion macros.
+#define EXPECT_PRED_FORMAT1(pred_format, v1) \
+  GTEST_PRED_FORMAT1_(pred_format, v1, GTEST_NONFATAL_FAILURE_)
+#define EXPECT_PRED1(pred, v1) \
+  GTEST_PRED1_(pred, v1, GTEST_NONFATAL_FAILURE_)
+#define ASSERT_PRED_FORMAT1(pred_format, v1) \
+  GTEST_PRED_FORMAT1_(pred_format, v1, GTEST_FATAL_FAILURE_)
+#define ASSERT_PRED1(pred, v1) \
+  GTEST_PRED1_(pred, v1, GTEST_FATAL_FAILURE_)
+
+
+
+// Helper function for implementing {EXPECT|ASSERT}_PRED2.  Don't use
+// this in your code.
+template <typename Pred,
+          typename T1,
+          typename T2>
+AssertionResult AssertPred2Helper(const char* pred_text,
+                                  const char* e1,
+                                  const char* e2,
+                                  Pred pred,
+                                  const T1& v1,
+                                  const T2& v2) {
+  if (pred(v1, v2)) return AssertionSuccess();
+
+  return AssertionFailure() << pred_text << "("
+                            << e1 << ", "
+                            << e2 << ") evaluates to false, where"
+                            << "\n" << e1 << " evaluates to " << v1
+                            << "\n" << e2 << " evaluates to " << v2;
+}
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT2.
+// Don't use this in your code.
+#define GTEST_PRED_FORMAT2_(pred_format, v1, v2, on_failure)\
+  GTEST_ASSERT_(pred_format(#v1, #v2, v1, v2), \
+                on_failure)
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED2.  Don't use
+// this in your code.
+#define GTEST_PRED2_(pred, v1, v2, on_failure)\
+  GTEST_ASSERT_(::testing::AssertPred2Helper(#pred, \
+                                             #v1, \
+                                             #v2, \
+                                             pred, \
+                                             v1, \
+                                             v2), on_failure)
+
+// Binary predicate assertion macros.
+#define EXPECT_PRED_FORMAT2(pred_format, v1, v2) \
+  GTEST_PRED_FORMAT2_(pred_format, v1, v2, GTEST_NONFATAL_FAILURE_)
+#define EXPECT_PRED2(pred, v1, v2) \
+  GTEST_PRED2_(pred, v1, v2, GTEST_NONFATAL_FAILURE_)
+#define ASSERT_PRED_FORMAT2(pred_format, v1, v2) \
+  GTEST_PRED_FORMAT2_(pred_format, v1, v2, GTEST_FATAL_FAILURE_)
+#define ASSERT_PRED2(pred, v1, v2) \
+  GTEST_PRED2_(pred, v1, v2, GTEST_FATAL_FAILURE_)
+
+
+
+// Helper function for implementing {EXPECT|ASSERT}_PRED3.  Don't use
+// this in your code.
+template <typename Pred,
+          typename T1,
+          typename T2,
+          typename T3>
+AssertionResult AssertPred3Helper(const char* pred_text,
+                                  const char* e1,
+                                  const char* e2,
+                                  const char* e3,
+                                  Pred pred,
+                                  const T1& v1,
+                                  const T2& v2,
+                                  const T3& v3) {
+  if (pred(v1, v2, v3)) return AssertionSuccess();
+
+  return AssertionFailure() << pred_text << "("
+                            << e1 << ", "
+                            << e2 << ", "
+                            << e3 << ") evaluates to false, where"
+                            << "\n" << e1 << " evaluates to " << v1
+                            << "\n" << e2 << " evaluates to " << v2
+                            << "\n" << e3 << " evaluates to " << v3;
+}
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT3.
+// Don't use this in your code.
+#define GTEST_PRED_FORMAT3_(pred_format, v1, v2, v3, on_failure)\
+  GTEST_ASSERT_(pred_format(#v1, #v2, #v3, v1, v2, v3), \
+                on_failure)
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED3.  Don't use
+// this in your code.
+#define GTEST_PRED3_(pred, v1, v2, v3, on_failure)\
+  GTEST_ASSERT_(::testing::AssertPred3Helper(#pred, \
+                                             #v1, \
+                                             #v2, \
+                                             #v3, \
+                                             pred, \
+                                             v1, \
+                                             v2, \
+                                             v3), on_failure)
+
+// Ternary predicate assertion macros.
+#define EXPECT_PRED_FORMAT3(pred_format, v1, v2, v3) \
+  GTEST_PRED_FORMAT3_(pred_format, v1, v2, v3, GTEST_NONFATAL_FAILURE_)
+#define EXPECT_PRED3(pred, v1, v2, v3) \
+  GTEST_PRED3_(pred, v1, v2, v3, GTEST_NONFATAL_FAILURE_)
+#define ASSERT_PRED_FORMAT3(pred_format, v1, v2, v3) \
+  GTEST_PRED_FORMAT3_(pred_format, v1, v2, v3, GTEST_FATAL_FAILURE_)
+#define ASSERT_PRED3(pred, v1, v2, v3) \
+  GTEST_PRED3_(pred, v1, v2, v3, GTEST_FATAL_FAILURE_)
+
+
+
+// Helper function for implementing {EXPECT|ASSERT}_PRED4.  Don't use
+// this in your code.
+template <typename Pred,
+          typename T1,
+          typename T2,
+          typename T3,
+          typename T4>
+AssertionResult AssertPred4Helper(const char* pred_text,
+                                  const char* e1,
+                                  const char* e2,
+                                  const char* e3,
+                                  const char* e4,
+                                  Pred pred,
+                                  const T1& v1,
+                                  const T2& v2,
+                                  const T3& v3,
+                                  const T4& v4) {
+  if (pred(v1, v2, v3, v4)) return AssertionSuccess();
+
+  return AssertionFailure() << pred_text << "("
+                            << e1 << ", "
+                            << e2 << ", "
+                            << e3 << ", "
+                            << e4 << ") evaluates to false, where"
+                            << "\n" << e1 << " evaluates to " << v1
+                            << "\n" << e2 << " evaluates to " << v2
+                            << "\n" << e3 << " evaluates to " << v3
+                            << "\n" << e4 << " evaluates to " << v4;
+}
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT4.
+// Don't use this in your code.
+#define GTEST_PRED_FORMAT4_(pred_format, v1, v2, v3, v4, on_failure)\
+  GTEST_ASSERT_(pred_format(#v1, #v2, #v3, #v4, v1, v2, v3, v4), \
+                on_failure)
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED4.  Don't use
+// this in your code.
+#define GTEST_PRED4_(pred, v1, v2, v3, v4, on_failure)\
+  GTEST_ASSERT_(::testing::AssertPred4Helper(#pred, \
+                                             #v1, \
+                                             #v2, \
+                                             #v3, \
+                                             #v4, \
+                                             pred, \
+                                             v1, \
+                                             v2, \
+                                             v3, \
+                                             v4), on_failure)
+
+// 4-ary predicate assertion macros.
+#define EXPECT_PRED_FORMAT4(pred_format, v1, v2, v3, v4) \
+  GTEST_PRED_FORMAT4_(pred_format, v1, v2, v3, v4, GTEST_NONFATAL_FAILURE_)
+#define EXPECT_PRED4(pred, v1, v2, v3, v4) \
+  GTEST_PRED4_(pred, v1, v2, v3, v4, GTEST_NONFATAL_FAILURE_)
+#define ASSERT_PRED_FORMAT4(pred_format, v1, v2, v3, v4) \
+  GTEST_PRED_FORMAT4_(pred_format, v1, v2, v3, v4, GTEST_FATAL_FAILURE_)
+#define ASSERT_PRED4(pred, v1, v2, v3, v4) \
+  GTEST_PRED4_(pred, v1, v2, v3, v4, GTEST_FATAL_FAILURE_)
+
+
+
+// Helper function for implementing {EXPECT|ASSERT}_PRED5.  Don't use
+// this in your code.
+template <typename Pred,
+          typename T1,
+          typename T2,
+          typename T3,
+          typename T4,
+          typename T5>
+AssertionResult AssertPred5Helper(const char* pred_text,
+                                  const char* e1,
+                                  const char* e2,
+                                  const char* e3,
+                                  const char* e4,
+                                  const char* e5,
+                                  Pred pred,
+                                  const T1& v1,
+                                  const T2& v2,
+                                  const T3& v3,
+                                  const T4& v4,
+                                  const T5& v5) {
+  if (pred(v1, v2, v3, v4, v5)) return AssertionSuccess();
+
+  return AssertionFailure() << pred_text << "("
+                            << e1 << ", "
+                            << e2 << ", "
+                            << e3 << ", "
+                            << e4 << ", "
+                            << e5 << ") evaluates to false, where"
+                            << "\n" << e1 << " evaluates to " << v1
+                            << "\n" << e2 << " evaluates to " << v2
+                            << "\n" << e3 << " evaluates to " << v3
+                            << "\n" << e4 << " evaluates to " << v4
+                            << "\n" << e5 << " evaluates to " << v5;
+}
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED_FORMAT5.
+// Don't use this in your code.
+#define GTEST_PRED_FORMAT5_(pred_format, v1, v2, v3, v4, v5, on_failure)\
+  GTEST_ASSERT_(pred_format(#v1, #v2, #v3, #v4, #v5, v1, v2, v3, v4, v5), \
+                on_failure)
+
+// Internal macro for implementing {EXPECT|ASSERT}_PRED5.  Don't use
+// this in your code.
+#define GTEST_PRED5_(pred, v1, v2, v3, v4, v5, on_failure)\
+  GTEST_ASSERT_(::testing::AssertPred5Helper(#pred, \
+                                             #v1, \
+                                             #v2, \
+                                             #v3, \
+                                             #v4, \
+                                             #v5, \
+                                             pred, \
+                                             v1, \
+                                             v2, \
+                                             v3, \
+                                             v4, \
+                                             v5), on_failure)
+
+// 5-ary predicate assertion macros.
+#define EXPECT_PRED_FORMAT5(pred_format, v1, v2, v3, v4, v5) \
+  GTEST_PRED_FORMAT5_(pred_format, v1, v2, v3, v4, v5, GTEST_NONFATAL_FAILURE_)
+#define EXPECT_PRED5(pred, v1, v2, v3, v4, v5) \
+  GTEST_PRED5_(pred, v1, v2, v3, v4, v5, GTEST_NONFATAL_FAILURE_)
+#define ASSERT_PRED_FORMAT5(pred_format, v1, v2, v3, v4, v5) \
+  GTEST_PRED_FORMAT5_(pred_format, v1, v2, v3, v4, v5, GTEST_FATAL_FAILURE_)
+#define ASSERT_PRED5(pred, v1, v2, v3, v4, v5) \
+  GTEST_PRED5_(pred, v1, v2, v3, v4, v5, GTEST_FATAL_FAILURE_)
+
+
+
+#endif  // GTEST_INCLUDE_GTEST_GTEST_PRED_IMPL_H_
--- a/third_party/googletest/src/include/gtest/gtest_prod.h
+++ b/third_party/googletest/src/include/gtest/gtest_prod.h
@@ -0,0 +1,58 @@
+// Copyright 2006, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+// Author: wan@google.com (Zhanyong Wan)
+//
+// Google C++ Testing Framework definitions useful in production code.
+
+#ifndef GTEST_INCLUDE_GTEST_GTEST_PROD_H_
+#define GTEST_INCLUDE_GTEST_GTEST_PROD_H_
+
+// When you need to test the private or protected members of a class,
+// use the FRIEND_TEST macro to declare your tests as friends of the
+// class.  For example:
+//
+// class MyClass {
+//  private:
+//   void MyMethod();
+//   FRIEND_TEST(MyClassTest, MyMethod);
+// };
+//
+// class MyClassTest : public testing::Test {
+//   // ...
+// };
+//
+// TEST_F(MyClassTest, MyMethod) {
+//   // Can call MyClass::MyMethod() here.
+// }
+
+#define FRIEND_TEST(test_case_name, test_name)\
+friend class test_case_name##_##test_name##_Test
+
+#endif  // GTEST_INCLUDE_GTEST_GTEST_PROD_H_
--- a/Show More
+++ b/Show More